- Email: [email protected]
The Reconfigurable Maze Provides Flexible, Scalable, Reproducible, and Repeatable Tests
Journal Pre-proof The reconfigurable maze provides flexible, scalable, reproducible and repeatable tests Satoshi Hoshino, Riku Takahashi, Kana Mieno, Yuta Tamatsu, Hirotsugu Azechi, Kaoru Ide, Susumu Takahashi PII:
S2589-0042(19)30532-2
DOI:
https://doi.org/10.1016/j.isci.2019.100787
Reference:
ISCI 100787
To appear in:
ISCIENCE
Received Date: 16 October 2019 Revised Date:
25 November 2019
Accepted Date: 13 December 2019
Please cite this article as: Hoshino, S., Takahashi, R., Mieno, K., Tamatsu, Y., Azechi, H., Ide, K., Takahashi, S., The reconfigurable maze provides flexible, scalable, reproducible and repeatable tests, ISCIENCE (2020), doi: https://doi.org/10.1016/j.isci.2019.100787. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 The Author(s).
Scalable LEGO-like modular system Parts of the reconfigurable maze Treadmill
Feeder
Movable wall
Runway
Runway
Tower with baseplate
Flexible interlocking mechanism
Protrusions
Repeatable tests
Floor-based breadboard
Reproducible tests Square
Cruciform
Assembly time < 4 min
T-maze
W-maze
Radial maze
1
Title:
2
The reconfigurable maze provides flexible, scalable, reproducible and repeatable tests
3
4
Author: Satoshi Hoshino1, Riku Takahashi1, Kana Mieno1, Yuta Tamatsu1, Hirotsugu
5
Azechi1, Kaoru Ide1, Susumu Takahashi1,2,*
6
7
8
9 10
Affiliations: 1.
Laboratory of Cognitive and Behavioral Neuroscience, Graduate School of
Brain Science, Doshisha University, Kyotanabe city, Kyoto, 610-0394, Japan. 2.
Lead Contact
11 12
*Correspondence to:
13
Susumu Takahashi
14
E-mail: [email protected]
15 16
17
1 Lead Contact: Susumu Takahashi
18
Summary
19
Multiple mazes are routinely used to test the performance of animals because each has
20
disadvantages inherent to its shape. However, the maze shape cannot be flexibly and
21
rapidly reproduced in a repeatable and scalable way in a single environment. Here, to
22
overcome the lack of flexibility, scalability, reproducibility and repeatability, we
23
develop a reconfigurable maze system that consists of interlocking runways and an array
24
of accompanying parts. It allows experimenters to rapidly and flexibly configure a variety
25
of maze structures along the grid pattern in a repeatable and scalable manner. Spatial
26
navigational behavior and hippocampal place coding were not impaired by the
27
interlocking mechanism. As a proof-of-principle demonstration, we demonstrate that the
28
maze morphing induces location remapping of the spatial receptive field. The
29
reconfigurable
30
reproducibility, therefore facilitating consistent investigation into the neuronal substrates
31
for learning and memory and allowing screening for behavioral phenotypes.
maze
thus
provides
flexibility,
32
2 Lead Contact: Susumu Takahashi
scalability,
repeatability,
and
33
Introduction
34
Several shapes of mazes such as the T-maze (Small, 1901; Dudchenko, 2001), plus
35
maze (Olton and Feustle, 1981), radial arm maze (Olton and Samuelson, 1976; Olton,
36
Walker and Gage, 1978), and figure-8 maze (Wood et al., 2000) have been designed as
37
behavioral tests to assess the performance of working (Dudchenko, 2004), reference
38
(Olton and Paras, 1979; Xu et al., 2019) and episodic-like memory (Babb and Crystal,
39
2005), and spatial navigation (O’Keefe and Nadel, 1978), as well as for studying
40
anxiety (Walf and Frye, 2007). As individual tests of learning and memory, all have pros
41
and cons (Andersen et al., 2009). To compensate for the disadvantages, evidence from a
42
battery of maze tests is usually accumulated to understand specific learning and memory
43
(Tecott and Nestler, 2004). However, each test is often conducted in a different real or
44
virtual experimental room because conventional mazes cannot be easily rebuilt in a
45
systematically arranged manner. Whereas the details of the maze structure, including
46
shape, coordination, and dimensions, are crucial aspects of the test results, the structures
47
themselves cannot be precisely reproduced in different laboratories in a repeatable way.
48
Thus, conventional maze tests lack reproducibility and repeatability.
49
Place cells found in the hippocampus encode an animal’s location and are deeply
50
involved in spatial navigation ability (O’Keefe and Nadel, 1978). Place coding responds
3 Lead Contact: Susumu Takahashi
51
dramatically to changes in the surrounding environment (Muller and Kubie, 1987),
52
suggesting that even if an animal experiences a single maze set inside different rooms,
53
the hippocampus may generate different cognitive maps for each room, despite the
54
mazes having identical shapes. On the other hand, conventional visual-based virtual
55
reality (VR) systems can realize unlimited shapes of mazes in visual space. However,
56
the lack of non-visual sensory feedback in passive VR results in altered neuronal
57
responses. Furthermore, it forces animals to be partially or fully immobilized because
58
one of the best applications of VR is to monitor neuronal activity from two-photon
59
calcium imaging during the spatial navigation of head-fixed animals. For instance, a
60
fraction of place cells reduced their firings in the spatial receptive field apparently
61
because of the loss of vestibular feedback in the VR (Aronov and Tank, 2014). These
62
altered neuronal responses impede detailed interpretations of the learning and memory
63
mechanisms involved in maze tests.
64
Here we develop a maze system that overcomes most of these limitations by allowing
65
the reconfiguration of the shape of the maze in a single physical environment. We
66
demonstrate the use of the reconfigurable maze system to replicate existing mazes,
67
including the T-maze, plus maze, W-maze, figure-8 maze, and radial arm maze. We also
68
investigate the spatial navigational behavior and coding of the hippocampal place cells
4 Lead Contact: Susumu Takahashi
69
of rats to address concerns over possible distortions caused by the interlocking
70
mechanisms, and perform a novel experiment that morphs the shape of the maze.
71
Together, these contribute to deciphering the neuronal underpinning of spatial
72
navigation.
5 Lead Contact: Susumu Takahashi
73
Results
74
Implementation of the reconfigurable maze
75
We developed a maze system that enables the reconfiguration of the shape of the
76
maze in a single physical environment using interlocking runways. Figure 1
77
demonstrates the existing standard mazes configured by our maze system in an
78
enclosure: the T-maze, W-maze, figure-8 maze, plus maze, and radial arm maze. The
79
runways are each placed atop towers with baseplates (Table S1); these baseplates have
80
protrusions that connect to a grid of holes in a floor-based breadboard (Figure 1f). The
81
insertion of the protrusions into the breadboard connection enables coordination with
82
the grid pattern and minimizes the swinging of the runways resulting from the
83
movement of the animal (Figure 1g). Similarly, an array of accompanying parts,
84
including feeders, movable walls, and treadmills placed atop towers of their own, can
85
also be attached on the breadboards. Thus, the feeder can be placed by the side of any
86
runway to change the reward location (Figure 1h, arrows). The movable wall can be
87
placed at any of the interlocking gaps between runways as a dead end to dynamically
88
control possible running paths (Figure 1h, arrowheads), and any runway can be replaced
89
by the treadmill (Figure 1h, double arrowheads). Moreover, the shut-off sensor can be
90
attached alongside any runway to signal a triggering event to the feeders and treadmills
6 Lead Contact: Susumu Takahashi
91
(Figure 1h, dashed arrows). Placing these interlocking parts onto the grid pattern
92
enables the experimenters, even if they are not familiar with the maze, to precisely
93
reproduce a variety of coordinated patterns of mazes for a brief period in a repeatable
94
manner in a single physical environment. The maze also provides a scalable
95
experimental setup because the complexity and the area of the maze is incrementally
96
expandable by adding extra parts. The time it took beginners to assemble a maze was
97
not significantly different from the assembly time of experts who had used the
98
reconfigurable maze daily over three months when the morphing of a maze from square
99
to cruciform was timed (3 experts: 131 ± 9 sec; 3 beginners: 158 ± 12 sec
100
(mean ± SEM); Figure 1j, Video S1–S6). Runway sections with tall sidewalls (35 cm
101
height) for reducing fear can be placed to test anxiety (Figure S1). Instead of runways,
102
open platforms can also be embedded into the maze, which enables the plus maze and
103
radial arm maze to be configured (Figure 1d–e). Furthermore, the scheduler for the
104
timing of sensors and actuators installed in the accompanying parts allows the
105
experimenter to design several behavioral tasks, even within a single maze shape. Such
106
flexible maze design also allows experimenters to rapidly select the tasks best matched
107
to the needs of a particular experiment during preliminary studies. In accordance with
108
this flexible design principle, we also developed a small version of the reconfigurable
7 Lead Contact: Susumu Takahashi
109
maze system for mice (Figure 1i, Table S2). Using the mouse version, we could also
110
configure existing standard mazes: the T-maze, W-maze, and figure-8 maze (Figure S2).
111
The assembly time for morphing a maze shape from square to T between experts and
112
beginners was not significantly different (3 experts: 204 ± 15 sec; 3 beginners: 186 ± 11
113
sec (mean ± SEM); Figure 1k), further demonstrating the usability of our system for
114
screening behavioral phenotypes in mice.
115
116
Interlocking gaps do not alter navigational behavior
117
To manufacture the interlocking parts at a reasonable cost or by hand, a short gap
118
between runways (~1 cm) (Figure 2a) is a prerequisite margin for the parts to interlock.
119
Ideally, perfectly manufactured parts with a precision under 1 mm would not require
120
these gaps as margins; however, such high-precision manufacturing is expensive and
121
does not permit the use of low-precision handcrafted parts. When the gaps are not
122
included as a margin, runways manufactured with lower precision (>1 cm) will overlap
123
if the location of the runway is only slightly shifted, as shown in Figure S3.
124
We next answered the question of whether this gap affects animal behavior. Four rats
125
were trained to smoothly run along a square-shaped maze in a clockwise direction. The
126
running speed at the gaps did not abruptly change from the running speed between gaps
8 Lead Contact: Susumu Takahashi
127
(Figure 2b). Incidences of abrupt changes in the head direction were significantly lower
128
at the gaps than between gaps (Figure 2c). These results suggest that the gaps did not
129
distort normal rats’ behaviors. To corroborate the result, we prepared two gapless
130
runways the total lengths (149 cm long) of which were the same as those of the three
131
interlocking runways, including the gaps between the sections. We then replaced three
132
runway sections at the top or bottom of the square-shaped maze with a single runway
133
without gaps (Figure S4a). Five rats were trained to run on the mazes, with or without
134
gaps, over a 20-minute interval. Both running speed and head direction on the gaps
135
were not significantly different from those on the corresponding portions of the single
136
runway without gaps (Figure S4b–c).
137
As shown from the trajectory (Figure 2a) and abrupt changes in the head direction
138
(Figure 2c), the rats appeared to frequently exhibit exploratory behaviors at preferred
139
locations (Hu and Amsel, 1995). To examine whether such exploration is preferentially
140
observed at gap locations, four rats were trained to run in a clockwise direction on a
141
square-shaped maze configuration. All rats sometimes paused and exhibited exploratory
142
behaviors around food dispensers or corners. Indeed, the occupancy time when the rats
143
visited the gap location around a food dispenser (Gap #7) was significantly longer than
144
the occupancy time on two other gaps (Gap #3 and #10) (Figure 2e), suggesting that
9 Lead Contact: Susumu Takahashi
145
unidentified factors, excluding the gap presence, might also affect rats’ behaviors
146
around food dispensers or corners. In addition, it is well known that such exploratory
147
behaviors modulate hippocampal neuronal activity (Gauthier and Tank, 2018). Thus, to
148
exclusively examine the gap effects, we focused on the gaps (#3, #4, #9, and #10)
149
between the three interlocking runways arranged in a straight line on the top and bottom
150
sides of the square-shaped maze in the subsequent analyses.
151
152
Interlocking gaps do not alter hippocampal place coding
153
To determine whether the gaps distort neuronal activity in terms of spatial navigation,
154
we monitored the activity of 236 neurons from the hippocampal CA1 of both
155
hemispheres of four rats running on the square-shaped maze in a clockwise direction
156
(Figure 3a, Table S3). The average width of place fields of 62 place cells on the gaps
157
settled between the 5th and 95th percentiles of distribution for shuffled data from the
158
entire set of place fields (Figure 3b). Moreover, the number of place fields from the 62
159
place cells and the firing rate of multi-unit activity in the hippocampal CA1 on the gaps
160
did not change when compared with recordings taken between the gaps (Figure 3c–d).
161
At the ensemble level, the trajectory decoded from the activity of simultaneously
162
monitored place cells using a memoryless Bayesian decoder (Zhang et al., 1998)
10 Lead Contact: Susumu Takahashi
163
depicted a seamless path even on the gap locations (Figure 3e–f). Thus, both behavioral
164
and neuronal responses at the gaps support the view that rats behave as if they perceive
165
the interlocking gap and gapless portions on the runways similarly.
166
167
Rats and mice can learn spatial alternation tasks in the reconfigured maze
168
After the five rats became familiar with the square-shaped maze, the maze shape was
169
morphed from square to figure-8 using the reconfigurable maze system. The rats were
170
trained to alternate between left and right at the branch point of the central stem for one
171
hour per day over five days (Figure 4a). All rats gradually learned the spatial alternation
172
task (Figure 4c). They achieved a mean score of greater than 80% correct choices over
173
50 consecutive trials on the second day. To demonstrate the usability of the attachable
174
walls and treadmills, three rats were then tested on a delayed version of the spatial
175
alternation task by partially reconfiguring the maze: a runway in the center of the stem
176
in the figure-8 shaped maze was replaced by a treadmill (Figure 4b). To force the rats to
177
run on the treadmill for a 7 sec delay period, movable walls were set at the gaps in front
178
and behind the treadmill. The rats achieved a mean score of greater than 70% correct
179
choices over 50 consecutive trials at the beginning, and their performance improved
180
after 10 days of testing (Figure 4d).
11 Lead Contact: Susumu Takahashi
181
After three mice were familiar with the linear track and rectangular maze in the
182
mouse version of the reconfigurable maze, the maze shape was morphed to a T-maze
183
(Figure 4e–f). The mice were trained to alternate between left and right at two branch
184
points for one hour per day over five days. All mice gradually learned the spatial
185
alternation task (Figure 4g). They achieved a mean score of greater than 75% correct
186
choices over 20 consecutive trials on the second day. These results and maze
187
reconfigurations demonstrate that the reconfigurable maze system can consistently
188
replicate the existing standard mazes and serve as tests of spatial learning and memory.
189
190
Place field remapping during the maze morphing experiment
191
Several lines of evidence suggest that the hippocampal place coding partially or
192
completely changes in response to external or internal cues (Muller and Kubie, 1987;
193
Leutgeb et al., 2005). This phenomenon is called the remapping of place fields. To
194
demonstrate the repeatability and reproducibility of our reconfigurable maze system
195
over existing mazes, we provide here a novel morphing experiment to examine the
196
remapping of place fields between different shapes of a maze with an identical path
197
length when the maze shape morphs from square to cruciform to square. As the
198
cruciform maze is an internal structure of the square maze, the total path length does not
12 Lead Contact: Susumu Takahashi
199
change during morphing. Moreover, the four outermost runways of the cruciform maze
200
physically overlap with the path of the square-shaped maze. As our maze realizes the
201
morphing in one enclosure, it enables the identification of factors influencing
202
hippocampal place coding: path integration and spatial reference frame (Figure 5a).
203
To examine the remapping of the place fields, we examined 129 place cells
204
monitored from the dorsal hippocampal CA1 of four rats. Whereas the total path length
205
of the runway and available external landmarks in the mazes were nearly identical
206
across maze shape morphs, the spatial correlation of the firing rate map of place cells
207
within the overlapping runways in the square-shaped maze between the first and second
208
exposures was significantly larger than that between the square-shaped maze and the
209
cruciform maze (Figure 5b–c). This suggests that the maze shape can be a cue capable
210
of inducing remapping of the location of place fields as a spatial reference frame.
211
Moreover, the difference in the maximum firing rate of place fields within the
212
overlapping runways in the square-shaped maze between the first and second exposures
213
was similar to that between the square-shaped maze and the cruciform maze
214
(Figure 5b, d). These results suggest that the maze morphing from square to cruciform is
215
represented by a difference in place field locations without a change in place field firing
216
rate.
13 Lead Contact: Susumu Takahashi
217
218
Discussion
219
We demonstrate our reconfigurable maze system’s ability to provide flexible, scalable,
220
repeatable and reproducible tests by configuring standard existing mazes in a single real
221
environment and by examining spatial navigational behavior and neuronal activity
222
during learning and memory performances in the mazes. Our maze system has the
223
potential to become an invaluable, scalable tool for the study of learning and memory:
224
including working and reference memory, spatial navigation, and decision making, as
225
well as anxiety. Using the reconfigurable maze, any experimenter can consistently
226
replicate a variety of mazes in their laboratory, and the position of the maze parts can be
227
reconfigured for a brief period in a repeatable manner to test several task performances
228
in a single physical environment.
229
Currently, a variety of maze tests are routinely used to gain a greater understanding of
230
learning and memory in animals and for screening behavioral phenotypes in animal
231
models of diseases. Although evidence is being accumulated, it is not necessarily
232
available to contribute to our understanding of learning and memory mechanisms in the
233
brain, because the underlying neuronal signatures are not necessarily similar, even when
234
identical maze shapes are used across different conditions. For instance, if a behavioral
14 Lead Contact: Susumu Takahashi
235
phenotype for spatial memory deficits emerged in two maze tests, each conducted in a
236
different room, the tests may only specifically show the remapping of hippocampal
237
place coding across the different rooms, as demonstrated in the maze morphing
238
experiment of the present study. Indeed, recent studies suggest that place cell activity
239
sequences are linked to future planning (Pfeiffer and Foster, 2013) and episodic-like
240
memory retrieval (Takahashi, 2015), as well as neurodegenerative diseases (Cheng and
241
Ji, 2013). Therefore, such neuronal dissociations should be minimized in a comparative
242
study across maze tests. Place cells show an irregular firing pattern in passive VR
243
experiments (Aronov and Tank, 2014). Therefore, we designed our maze based on the
244
concept that various shapes can be constructed in a single real environment, where
245
several types of learning and memory tests can be performed. Recently developed active
246
VR can markedly minimize such dissociations, specifically for proximal cues (Aronov
247
and Tank, 2014; Stowers et al., 2017). While our maze system and the active VR are not
248
exclusive, the combination will enable the production of unlimited experimental
249
situations without the neuronal distortions arising from both proximal and distal cues in
250
the near future.
251
Our reconfigurable maze system is compatible with techniques for monitoring
252
neuronal activity such as extracellular multiple single-unit recording (Wilson and
15 Lead Contact: Susumu Takahashi
253
McNaughton, 1993) and single-photon calcium imaging (Ziv et al., 2013) for freely
254
behaving rats or mice. As demonstrated by the remapping of place fields across
255
morphed mazes, the combination must be accelerated to allow understanding of the
256
neuronal underpinning of learning and memory. Furthermore, our maze enables rapid
257
prototyping of tasks by reconfiguring the coordination of the parts, including runways,
258
rewards, and obstacles, within a few minutes, accelerating studies on learning and
259
memory.
260
The reconfigurable maze is thus a promising maze platform for testing the
261
performance of learning and memory, including working, reference and episodic
262
memory, decision-making, spatial navigation, and anxiety, as well as for screening
263
behavioral phenotypes of mice, including transgenic models of disease.
264
265
Limitations of the study
266
The initial setup cost and laboratory space for introducing the reconfigurable maze
267
may be barriers to widespread distribution. However, the scalable features minimize
268
these barriers. For instance, a minimal setup realized by removing optional modules
269
from the full configuration (demonstrated by a square-shaped maze without movable
270
walls or shut-off sensors) will reduce the cost. Additionally, the modular architecture
16 Lead Contact: Susumu Takahashi
271
allows the configuration of small mazes to fit limited laboratory space. While the
272
minimal setup may not be a compelling reason for laboratories to invest in the
273
reconfigurable maze, the complexity can be expanded incrementally by adding
274
supplementary parts. Moreover, the set up and validation for novel configured mazes
275
within a laboratory may be time-consuming. Provided that the reconfigurable maze will
276
be widely distributed and standardized, the overall cost will be lowered and information
277
resources describing how to set up new configurations of the maze and validate their
278
performance for specific aims will be shared between researchers. In this way,
279
implementation of the reconfigurable maze will ensure reproducibility and repeatability.
280
Despite its flexibility, reproducibility, repeatability, and scalability, our maze has a
281
few limitations. Whereas oblique runways at an angle of 45 degrees could be embedded
282
into our maze as demonstrated by the configured radial arm maze (Figure 1e), the angle
283
cannot be flexibly modified because the coordinated grid assignment of the parts
284
requires the runways to be placed at a fixed angle. Unlike the recently developed
285
honeycomb maze (Wood et al., 2018), the tower supporting the runway cannot be raised
286
and lowered automatically. Whereas adjustable mechanisms are capable of overcoming
287
those limitations, they increase the damage rate of the physical system during
288
behavioral and neuronal recordings, reducing the major advantage of reproducibility
17 Lead Contact: Susumu Takahashi
289
and repeatability.
290
18 Lead Contact: Susumu Takahashi
291
292
Acknowledgments
293
This work is supported by the JSPS KAKENHI (16H06543, 16H02840, and 19H01131)
294
to S.T..
295
296
Data and code availability
297
The
298
(https://github.com/TakahashiLab/ReconfigurableMazeParts). The controlling software
299
is also freely available (https://github.com/TakahashiLab/ReconfigurableMazeGUI).
detail
of
the
maze
parts
is
freely
available
300
301
Author Contributions
302
S.T. conceived the project. S.H. developed control software. S.H., K.M., R.T., and
303
Y.T. performed behavioral experiments. H.A. and K.I. performed the histological
304
verification. S.H., H.A., and K.I. made apparatus. S.H. performed electrophysiological
305
experiments. S.H. and S.T. performed surgery, data analyses, and wrote the manuscript.
306
307
Declaration of Interests
19 Lead Contact: Susumu Takahashi
308
S.T. is an inventor on Japanese unexamined patent application publication (No.
309
P2019-115315A, applicant: Doshisha University) pertaining to the reconfigurable maze.
310
All remaining authors declare no conflict of interest.
311 312
20 Lead Contact: Susumu Takahashi
313
Reference
314
Andersen, P. et al. (eds) (2009) The Hippocampus Book. Oxford University Press. doi:
315
10.1093/acprof:oso/9780195100273.001.0001.
316
Aronov, D. and Tank, D. W. (2014) ‘Engagement of Neural Circuits Underlying 2D
317
Spatial Navigation in a Rodent Virtual Reality System’, Neuron. Elsevier Inc., 84(2), pp.
318
442–456. doi: 10.1016/j.neuron.2014.08.042.
319
Babb, S. J. and Crystal, J. D. (2005) ‘Discrimination of what, when, and where:
320
Implications for episodic-like memory in rats’, Learning and Motivation. doi:
321
10.1016/j.lmot.2005.02.009.
322
Cheng, J. and Ji, D. (2013) ‘Rigid firing sequences undermine spatial memory codes in
323
a neurodegenerative mouse model’, eLife, 2013(2), pp. 1–26. doi: 10.7554/eLife.00647.
324
Dudchenko, P. A. (2001) ‘How do animals actually solve the T maze?’, Behavioral
325
Neuroscience, 115(4), pp. 850–860. doi: 10.1037/0735-7044.115.4.850.
326
Dudchenko, P. A. (2004) ‘An overview of the tasks used to test working memory in
327
rodents’, in Neuroscience and Biobehavioral Reviews, pp. 699–709. doi:
328
10.1016/j.neubiorev.2004.09.002.
329
Gauthier, J. L. and Tank, D. W. (2018) ‘A Dedicated Population for Reward Coding in
330
the Hippocampus’, Neuron, 99(1), pp. 179-193.e7. doi: 10.1016/j.neuron.2018.06.008.
331
Hu, D. and Amsel, A. (1995) ‘A simple test of the vicarious trial-and-error hypothesis 21 Lead Contact: Susumu Takahashi
332
of hippocampal function.’, Proceedings of the National Academy of Sciences of the
333
United States of America, 92, pp. 5506–5509. doi: 10.1073/pnas.92.12.5506.
334
Leutgeb, S. et al. (2005) ‘Independent codes for spatial and episodic memory in
335
hippocampal neuronal ensembles’, Science, 309(5734), pp. 619–623.
336
Muller, R. U. and Kubie, J. L. (1987) ‘The Effects of Changes in the Environment on
337
the Spatial Firing of Hippocampal Complex-Spike Cells’, Journal of Neuroscience, 7(7),
338
pp. 1951–1968.
339
O’Keefe, J. and Nadel, L. (1978) The Hippocampus as a Cognitive Map. Oxford
340
University Press. doi: 10.1017/CBO9781107415324.004.
341
Olton, D. S. and Feustle, W. A. (1981) ‘Hippocampal function required for nonspatial
342
working memory’, Experimental Brain Research, 41(3–4), pp. 380–389. doi:
343
10.1007/BF00238896.
344
Olton, D. S. and Paras, B. C. (1979) ‘Spatial memory and hippocampal function’,
345
Neuropsychologia. doi: 10.1016/0028-3932(79)90042-3.
346
Olton, D. S. and Samuelson, R. J. (1976) ‘Remembrance of places passed: Spatial
347
memory in rats’, Journal of Experimental Psychology: Animal Behavior Processes. doi:
348
10.1037/0097-7403.2.2.97.
349
Olton, D. S., Walker, J. A. and Gage, F. H. (1978) ‘Hippocampal connections and
22 Lead Contact: Susumu Takahashi
350
spatial discrimination’, Brain Research. doi: 10.1016/0006-8993(78)90930-7.
351
Pfeiffer, B. E. and Foster, D. J. (2013) ‘Hippocampal place-cell sequences depict future
352
paths to remembered goals.’, Nature, 497, pp. 74–9. doi: 10.1038/nature12112.
353
Small, W. S. (1901) ‘Experimental Study of the Mental Processes of the Rat. II’, The
354
American Journal of Psychology. doi: 10.2307/1412534.
355
Stowers, J. R. et al. (2017) ‘Virtual reality for freely moving animals’, Nature Methods,
356
14(10), pp. 995–1002. doi: 10.1038/nmeth.4399.
357
Takahashi, S. (2015) ‘Episodic-like memory trace in awake replay of hippocampal
358
place cell activity sequences’, eLife, 4. doi: 10.7554/eLife.08105.
359
Tecott, L. H. and Nestler, E. J. (2004) ‘Neurobehavioral assessment in the information
360
age’, Nature Neuroscience, pp. 462–466. doi: 10.1038/nn1225.
361
Walf, A. A. and Frye, C. A. (2007) ‘The use of the elevated plus maze as an assay of
362
anxiety-related behavior in rodents’, Nature Protocols, 2(2), pp. 322–328. doi:
363
10.1038/nprot.2007.44.
364
Wilson, M. A. and McNaughton, B. L. (1993) ‘Dynamics of the hippocampal ensemble
365
code for space.’, Science, 261(5124), pp. 1055–1058. doi: 10.1126/science.8351520.
366
Wood, E. R. et al. (2000) ‘Hippocampal neurons encode information about different
367
types of memory episodes occurring in the same location’, Neuron, 27(3), pp. 623–633.
23 Lead Contact: Susumu Takahashi
368
Wood, R. A. et al. (2018) ‘The honeycomb maze provides a novel test to study
369
hippocampal-dependent spatial navigation’, Nature, 554(7690), pp. 102–105. doi:
370
10.1038/nature25433.
371
Xu, H. et al. (2019) ‘Assembly Responses of Hippocampal CA1 Place Cells Predict
372
Learned Behavior in Goal-Directed Spatial Tasks on the Radial Eight-Arm Maze’,
373
Neuron, 101(1), pp. 119-132.e4. doi: 10.1016/j.neuron.2018.11.015.
374
Zhang, K. et al. (1998) ‘Interpreting neuronal population activity by reconstruction:
375
unified framework with application to hippocampal place cells.’, Journal of
376
neurophysiology, 79, pp. 1017–1044.
377
Ziv, Y. et al. (2013) ‘Long-term dynamics of CA1 hippocampal place codes.’, Nature
378
neuroscience. Nature Publishing Group, 16(3), pp. 264–6. doi: 10.1038/nn.3329.
379 380
24 Lead Contact: Susumu Takahashi
381
Figure Legends
382 383
Figure 1. The reconfigurable maze. (a–e) The maze can be configured to form T (a),
384
W (b), figure-8 (c), plus (d), and radial arm (e) mazes in a single enclosure. (f) Runway
385
sections are placed atop towers. (g) The baseplate of each tower has four protrusions
386
that coordinate the placement of the section on the breadboard in a flexible, repeatable
387
way. (h) Configured plus maze for rats with two feeders (arrows), two movable walls
388
(arrowheads), one treadmill (double arrowhead), and two shut-off sensors (dashed
389
arrows). (i) Small version of the reconfigurable maze for mice, configured as a figure-8
390
maze with two feeders (arrows), two movable walls (arrowheads), one treadmill (double
391
arrowhead), and two shut-off sensors (dashed arrows). (j) Assembly time of the rat
392
version of the reconfigurable maze for morphing the shape from square to cruciform.
393
Performance improved with consecutive trials in a day (performance vs. trial: F2,8 =
394
7.453, P = 0.0149), but not with expertise (expert/beginner) (performance vs. expertise:
395
F1,4 = 5.654, P = 0.0762). There was no significant interaction between expertise and
396
experience in a day (expertise vs. trial: F2,8 = 0.320, P = 0.735). Two-way mixed
397
ANOVA was used.
398
(k) Assembly time of the mouse version of the reconfigurable maze for morphing the
399
shape from a rectangle to a T. There was no significant difference between performance
25 Lead Contact: Susumu Takahashi
400
and consecutive trials in a day (performance vs. trial: F2,8 = 3.997, P = 0.0625) or
401
expertise (expert/beginner) (performance vs. expertise: F1,4 = 0.351, P = 0.58). There
402
was no significant interaction between expertise and experience in a day (expertise vs.
403
trial: F2,8 = 0.658, P = 0.543). Two-way mixed ANOVA was used.
404
405
Figure 2. Interlocking gaps do not alter navigational behavior. (a) Schematics of the
406
square-shaped maze test. An example running trajectory of a rat is superimposed on the
407
maze. The numbers indicate gap locations. R1 and R2 indicate food dispensers.
408
Linearized gap locations are illustrated at the bottom. (b–c) Left, the average running
409
speed and head direction of four rats as a function of the linearized location. The dots
410
indicate the location of outliers for each lap. Right, the average number of outliers on
411
the regions around the gap (gray shaded areas) and others (running speed: t = −0.19, df
412
= 6, P = 0.86; head direction: t = −5.53, df = 6, P = 0.0015, Two-tailed paired t-test) **:
413
P < 0.01, n.s.: P > 0.05. Error bars indicate SEM. (d) Left, the percentage of the
414
occupancy time over the entire maze as a function of the linearized location of four rats.
415
Right, the median percentage of the occupancy time at gap locations (median, first and
416
third quartiles, minimum, and maximum indicated). The rats preferentially slow down at
417
gap #7 as compared with gaps #3 and #10, which are located at the top and bottom of
26 Lead Contact: Susumu Takahashi
418
the maze (F1,3 = 152.69, P = 0.0011; gap #7 vs. #3: P = 0.049; gap #7 vs. #10: P = 0.030,
419
One-way repeated-measures ANOVA with post hoc Tukey’s HSD test).
420
421
Figure 3. Interlocking gaps do not alter hippocampal place coding. (a) Left,
422
normalized firing rate map of two representative place cells that have place fields
423
around gap locations in the hippocampal CA1 on the square-shaped maze configuration.
424
Right, normalized firing rate maps of 65 hippocampal place cells simultaneously
425
monitored from a rat (ζA) ordered by the latency of their peak firing rates. Each line is a
426
single unit. Gap locations are indicated at the top. Red indicates maximum firing rates,
427
and blue indicates silent. (b) Distribution of the average width of place fields randomly
428
shuffled from the original sample covering the gaps (3,000 shuffles). Red line indicates
429
the average width of the place fields on the gaps. Dotted lines indicate 5th and 95th
430
percentiles for the shuffled data. (c–d) Left, the number of place fields (c) and the firing
431
rate of multi-unit activity (MUA) in the hippocampal CA1 (d) as a function of
432
linearized location on the square-shaped maze. Right, the average number of place
433
fields (c) (z = 0.14, P = 0.89) and the average firing rate of MUA (z = −1.51, P = 0.13)
434
on the gaps (#3, 4, 9, 10) (green shaded area) and between them (orange shaded area).
435
(e) Representative posterior probability of locations decoded by the Bayesian decoder
27 Lead Contact: Susumu Takahashi
436
from 89 simultaneously monitored cells. Values are indicated by the color bar (right). (f)
437
Left, the difference between actual and decoded locations as a function of linearized
438
location on the square-shaped maze. Right, the average difference between actual and
439
decoded locations on the gaps (#3, 4, 9, 10) (green shaded area) and between them
440
(orange shaded area) (z = 1.34, P = 0.18). Bin width is set at 3 cm. All error bars
441
indicate SEM. All were from two-tailed Wilcoxon rank-sum test, n.s.: P > 0.05.
442
443
Figure 4. Rats and mice can learn to navigate the configured mazes. (a) Schematic
444
of the figure-8 shaped maze created using the rat version of the reconfigurable maze.
445
Example animal trajectories are superimposed on top of the maze. The red circle
446
indicates a branch point where the rat must decide which direction to turn. (b) Same as
447
in (a) but with a treadmill (green). Red dashed lines indicate the gap locations where
448
movable walls are placed to force the rats to run on the treadmill for a delay period. (c)
449
Spatial alternation task performance improved with experience (performance vs. testing
450
day: F1, 4 = 795.9, P < 10−5). (d) Performance of the delayed version of the spatial
451
alternation task of rats improved with experience (performance vs. testing day: F1,2 =
452
531.8, P = 0.0019). All were from one-way repeated-measures ANOVA. Error bars
453
indicate SEM. (e) Top view of the configured double T-maze using the mouse version of
28 Lead Contact: Susumu Takahashi
454
the reconfigurable maze. (f) Schematics of the double T-maze. Example mouse
455
trajectories are superimposed on the maze. Arrows mark the running directions from the
456
food dispensers (R1/R2). The red circles indicate branch points where the mice must
457
decide which direction to turn. Red dotted lines indicate the locations where the
458
movable walls are placed to prevent a reverse run. (g) Spatial alternation task
459
performance of mice improved with experience (performance vs. testing day: F1,2 =
460
7320, P = 0.000136). One-way repeated-measures ANOVA was used. Error bars
461
indicate SEM.
462
463
Figure 5. Place field remapping during the maze morphing experiment. (a)
464
Schematics of square and cruciform mazes configured by our system (top) and the
465
representative running trajectory of a rat (bottom). The runways enclosed by red circles
466
were not moved during the morphing experiment. (b) Representative change of the
467
place field of a place cell recorded from a rat (ζA) during maze morphing. The
468
maximum firing rate is displayed above the top left corner. (c–d) Violin plots of spatial
469
similarity as the spatial correlation of the firing rate map (c) (z = −3.98, P = 10−4) and
470
rate similarity as firing rate difference within place fields (d) (z = −0.60, P = 0.55)
471
between the square maze and cruciform maze, and between the first and second
29 Lead Contact: Susumu Takahashi
472
exposures of the square maze (left). The bar indicates the median. Surrounding each
473
side is a rotated kernel density plot. Their cumulative frequency is graphed to the right.
474
All were from two-tailed Wilcoxon signed-rank test. ***: P < 0.001, n.s.: P > 0.05.
475
30 Lead Contact: Susumu Takahashi
Highlights:
The reconfigurable maze enables flexibly in building a variety of maze paths.
The maze ensures reproducibility, repeatability, and scalability.
The maze does not impair spatial navigational behavior or neuronal activity.
Various tests of learning and memory can be conducted in a single environment.
S2589-0042(19)30532-2
DOI:
https://doi.org/10.1016/j.isci.2019.100787
Reference:
ISCI 100787
To appear in:
ISCIENCE
Received Date: 16 October 2019 Revised Date:
25 November 2019
Accepted Date: 13 December 2019
Please cite this article as: Hoshino, S., Takahashi, R., Mieno, K., Tamatsu, Y., Azechi, H., Ide, K., Takahashi, S., The reconfigurable maze provides flexible, scalable, reproducible and repeatable tests, ISCIENCE (2020), doi: https://doi.org/10.1016/j.isci.2019.100787. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 The Author(s).
Scalable LEGO-like modular system Parts of the reconfigurable maze Treadmill
Feeder
Movable wall
Runway
Runway
Tower with baseplate
Flexible interlocking mechanism
Protrusions
Repeatable tests
Floor-based breadboard
Reproducible tests Square
Cruciform
Assembly time < 4 min
T-maze
W-maze
Radial maze
1
Title:
2
The reconfigurable maze provides flexible, scalable, reproducible and repeatable tests
3
4
Author: Satoshi Hoshino1, Riku Takahashi1, Kana Mieno1, Yuta Tamatsu1, Hirotsugu
5
Azechi1, Kaoru Ide1, Susumu Takahashi1,2,*
6
7
8
9 10
Affiliations: 1.
Laboratory of Cognitive and Behavioral Neuroscience, Graduate School of
Brain Science, Doshisha University, Kyotanabe city, Kyoto, 610-0394, Japan. 2.
Lead Contact
11 12
*Correspondence to:
13
Susumu Takahashi
14
E-mail: [email protected]
15 16
17
1 Lead Contact: Susumu Takahashi
18
Summary
19
Multiple mazes are routinely used to test the performance of animals because each has
20
disadvantages inherent to its shape. However, the maze shape cannot be flexibly and
21
rapidly reproduced in a repeatable and scalable way in a single environment. Here, to
22
overcome the lack of flexibility, scalability, reproducibility and repeatability, we
23
develop a reconfigurable maze system that consists of interlocking runways and an array
24
of accompanying parts. It allows experimenters to rapidly and flexibly configure a variety
25
of maze structures along the grid pattern in a repeatable and scalable manner. Spatial
26
navigational behavior and hippocampal place coding were not impaired by the
27
interlocking mechanism. As a proof-of-principle demonstration, we demonstrate that the
28
maze morphing induces location remapping of the spatial receptive field. The
29
reconfigurable
30
reproducibility, therefore facilitating consistent investigation into the neuronal substrates
31
for learning and memory and allowing screening for behavioral phenotypes.
maze
thus
provides
flexibility,
32
2 Lead Contact: Susumu Takahashi
scalability,
repeatability,
and
33
Introduction
34
Several shapes of mazes such as the T-maze (Small, 1901; Dudchenko, 2001), plus
35
maze (Olton and Feustle, 1981), radial arm maze (Olton and Samuelson, 1976; Olton,
36
Walker and Gage, 1978), and figure-8 maze (Wood et al., 2000) have been designed as
37
behavioral tests to assess the performance of working (Dudchenko, 2004), reference
38
(Olton and Paras, 1979; Xu et al., 2019) and episodic-like memory (Babb and Crystal,
39
2005), and spatial navigation (O’Keefe and Nadel, 1978), as well as for studying
40
anxiety (Walf and Frye, 2007). As individual tests of learning and memory, all have pros
41
and cons (Andersen et al., 2009). To compensate for the disadvantages, evidence from a
42
battery of maze tests is usually accumulated to understand specific learning and memory
43
(Tecott and Nestler, 2004). However, each test is often conducted in a different real or
44
virtual experimental room because conventional mazes cannot be easily rebuilt in a
45
systematically arranged manner. Whereas the details of the maze structure, including
46
shape, coordination, and dimensions, are crucial aspects of the test results, the structures
47
themselves cannot be precisely reproduced in different laboratories in a repeatable way.
48
Thus, conventional maze tests lack reproducibility and repeatability.
49
Place cells found in the hippocampus encode an animal’s location and are deeply
50
involved in spatial navigation ability (O’Keefe and Nadel, 1978). Place coding responds
3 Lead Contact: Susumu Takahashi
51
dramatically to changes in the surrounding environment (Muller and Kubie, 1987),
52
suggesting that even if an animal experiences a single maze set inside different rooms,
53
the hippocampus may generate different cognitive maps for each room, despite the
54
mazes having identical shapes. On the other hand, conventional visual-based virtual
55
reality (VR) systems can realize unlimited shapes of mazes in visual space. However,
56
the lack of non-visual sensory feedback in passive VR results in altered neuronal
57
responses. Furthermore, it forces animals to be partially or fully immobilized because
58
one of the best applications of VR is to monitor neuronal activity from two-photon
59
calcium imaging during the spatial navigation of head-fixed animals. For instance, a
60
fraction of place cells reduced their firings in the spatial receptive field apparently
61
because of the loss of vestibular feedback in the VR (Aronov and Tank, 2014). These
62
altered neuronal responses impede detailed interpretations of the learning and memory
63
mechanisms involved in maze tests.
64
Here we develop a maze system that overcomes most of these limitations by allowing
65
the reconfiguration of the shape of the maze in a single physical environment. We
66
demonstrate the use of the reconfigurable maze system to replicate existing mazes,
67
including the T-maze, plus maze, W-maze, figure-8 maze, and radial arm maze. We also
68
investigate the spatial navigational behavior and coding of the hippocampal place cells
4 Lead Contact: Susumu Takahashi
69
of rats to address concerns over possible distortions caused by the interlocking
70
mechanisms, and perform a novel experiment that morphs the shape of the maze.
71
Together, these contribute to deciphering the neuronal underpinning of spatial
72
navigation.
5 Lead Contact: Susumu Takahashi
73
Results
74
Implementation of the reconfigurable maze
75
We developed a maze system that enables the reconfiguration of the shape of the
76
maze in a single physical environment using interlocking runways. Figure 1
77
demonstrates the existing standard mazes configured by our maze system in an
78
enclosure: the T-maze, W-maze, figure-8 maze, plus maze, and radial arm maze. The
79
runways are each placed atop towers with baseplates (Table S1); these baseplates have
80
protrusions that connect to a grid of holes in a floor-based breadboard (Figure 1f). The
81
insertion of the protrusions into the breadboard connection enables coordination with
82
the grid pattern and minimizes the swinging of the runways resulting from the
83
movement of the animal (Figure 1g). Similarly, an array of accompanying parts,
84
including feeders, movable walls, and treadmills placed atop towers of their own, can
85
also be attached on the breadboards. Thus, the feeder can be placed by the side of any
86
runway to change the reward location (Figure 1h, arrows). The movable wall can be
87
placed at any of the interlocking gaps between runways as a dead end to dynamically
88
control possible running paths (Figure 1h, arrowheads), and any runway can be replaced
89
by the treadmill (Figure 1h, double arrowheads). Moreover, the shut-off sensor can be
90
attached alongside any runway to signal a triggering event to the feeders and treadmills
6 Lead Contact: Susumu Takahashi
91
(Figure 1h, dashed arrows). Placing these interlocking parts onto the grid pattern
92
enables the experimenters, even if they are not familiar with the maze, to precisely
93
reproduce a variety of coordinated patterns of mazes for a brief period in a repeatable
94
manner in a single physical environment. The maze also provides a scalable
95
experimental setup because the complexity and the area of the maze is incrementally
96
expandable by adding extra parts. The time it took beginners to assemble a maze was
97
not significantly different from the assembly time of experts who had used the
98
reconfigurable maze daily over three months when the morphing of a maze from square
99
to cruciform was timed (3 experts: 131 ± 9 sec; 3 beginners: 158 ± 12 sec
100
(mean ± SEM); Figure 1j, Video S1–S6). Runway sections with tall sidewalls (35 cm
101
height) for reducing fear can be placed to test anxiety (Figure S1). Instead of runways,
102
open platforms can also be embedded into the maze, which enables the plus maze and
103
radial arm maze to be configured (Figure 1d–e). Furthermore, the scheduler for the
104
timing of sensors and actuators installed in the accompanying parts allows the
105
experimenter to design several behavioral tasks, even within a single maze shape. Such
106
flexible maze design also allows experimenters to rapidly select the tasks best matched
107
to the needs of a particular experiment during preliminary studies. In accordance with
108
this flexible design principle, we also developed a small version of the reconfigurable
7 Lead Contact: Susumu Takahashi
109
maze system for mice (Figure 1i, Table S2). Using the mouse version, we could also
110
configure existing standard mazes: the T-maze, W-maze, and figure-8 maze (Figure S2).
111
The assembly time for morphing a maze shape from square to T between experts and
112
beginners was not significantly different (3 experts: 204 ± 15 sec; 3 beginners: 186 ± 11
113
sec (mean ± SEM); Figure 1k), further demonstrating the usability of our system for
114
screening behavioral phenotypes in mice.
115
116
Interlocking gaps do not alter navigational behavior
117
To manufacture the interlocking parts at a reasonable cost or by hand, a short gap
118
between runways (~1 cm) (Figure 2a) is a prerequisite margin for the parts to interlock.
119
Ideally, perfectly manufactured parts with a precision under 1 mm would not require
120
these gaps as margins; however, such high-precision manufacturing is expensive and
121
does not permit the use of low-precision handcrafted parts. When the gaps are not
122
included as a margin, runways manufactured with lower precision (>1 cm) will overlap
123
if the location of the runway is only slightly shifted, as shown in Figure S3.
124
We next answered the question of whether this gap affects animal behavior. Four rats
125
were trained to smoothly run along a square-shaped maze in a clockwise direction. The
126
running speed at the gaps did not abruptly change from the running speed between gaps
8 Lead Contact: Susumu Takahashi
127
(Figure 2b). Incidences of abrupt changes in the head direction were significantly lower
128
at the gaps than between gaps (Figure 2c). These results suggest that the gaps did not
129
distort normal rats’ behaviors. To corroborate the result, we prepared two gapless
130
runways the total lengths (149 cm long) of which were the same as those of the three
131
interlocking runways, including the gaps between the sections. We then replaced three
132
runway sections at the top or bottom of the square-shaped maze with a single runway
133
without gaps (Figure S4a). Five rats were trained to run on the mazes, with or without
134
gaps, over a 20-minute interval. Both running speed and head direction on the gaps
135
were not significantly different from those on the corresponding portions of the single
136
runway without gaps (Figure S4b–c).
137
As shown from the trajectory (Figure 2a) and abrupt changes in the head direction
138
(Figure 2c), the rats appeared to frequently exhibit exploratory behaviors at preferred
139
locations (Hu and Amsel, 1995). To examine whether such exploration is preferentially
140
observed at gap locations, four rats were trained to run in a clockwise direction on a
141
square-shaped maze configuration. All rats sometimes paused and exhibited exploratory
142
behaviors around food dispensers or corners. Indeed, the occupancy time when the rats
143
visited the gap location around a food dispenser (Gap #7) was significantly longer than
144
the occupancy time on two other gaps (Gap #3 and #10) (Figure 2e), suggesting that
9 Lead Contact: Susumu Takahashi
145
unidentified factors, excluding the gap presence, might also affect rats’ behaviors
146
around food dispensers or corners. In addition, it is well known that such exploratory
147
behaviors modulate hippocampal neuronal activity (Gauthier and Tank, 2018). Thus, to
148
exclusively examine the gap effects, we focused on the gaps (#3, #4, #9, and #10)
149
between the three interlocking runways arranged in a straight line on the top and bottom
150
sides of the square-shaped maze in the subsequent analyses.
151
152
Interlocking gaps do not alter hippocampal place coding
153
To determine whether the gaps distort neuronal activity in terms of spatial navigation,
154
we monitored the activity of 236 neurons from the hippocampal CA1 of both
155
hemispheres of four rats running on the square-shaped maze in a clockwise direction
156
(Figure 3a, Table S3). The average width of place fields of 62 place cells on the gaps
157
settled between the 5th and 95th percentiles of distribution for shuffled data from the
158
entire set of place fields (Figure 3b). Moreover, the number of place fields from the 62
159
place cells and the firing rate of multi-unit activity in the hippocampal CA1 on the gaps
160
did not change when compared with recordings taken between the gaps (Figure 3c–d).
161
At the ensemble level, the trajectory decoded from the activity of simultaneously
162
monitored place cells using a memoryless Bayesian decoder (Zhang et al., 1998)
10 Lead Contact: Susumu Takahashi
163
depicted a seamless path even on the gap locations (Figure 3e–f). Thus, both behavioral
164
and neuronal responses at the gaps support the view that rats behave as if they perceive
165
the interlocking gap and gapless portions on the runways similarly.
166
167
Rats and mice can learn spatial alternation tasks in the reconfigured maze
168
After the five rats became familiar with the square-shaped maze, the maze shape was
169
morphed from square to figure-8 using the reconfigurable maze system. The rats were
170
trained to alternate between left and right at the branch point of the central stem for one
171
hour per day over five days (Figure 4a). All rats gradually learned the spatial alternation
172
task (Figure 4c). They achieved a mean score of greater than 80% correct choices over
173
50 consecutive trials on the second day. To demonstrate the usability of the attachable
174
walls and treadmills, three rats were then tested on a delayed version of the spatial
175
alternation task by partially reconfiguring the maze: a runway in the center of the stem
176
in the figure-8 shaped maze was replaced by a treadmill (Figure 4b). To force the rats to
177
run on the treadmill for a 7 sec delay period, movable walls were set at the gaps in front
178
and behind the treadmill. The rats achieved a mean score of greater than 70% correct
179
choices over 50 consecutive trials at the beginning, and their performance improved
180
after 10 days of testing (Figure 4d).
11 Lead Contact: Susumu Takahashi
181
After three mice were familiar with the linear track and rectangular maze in the
182
mouse version of the reconfigurable maze, the maze shape was morphed to a T-maze
183
(Figure 4e–f). The mice were trained to alternate between left and right at two branch
184
points for one hour per day over five days. All mice gradually learned the spatial
185
alternation task (Figure 4g). They achieved a mean score of greater than 75% correct
186
choices over 20 consecutive trials on the second day. These results and maze
187
reconfigurations demonstrate that the reconfigurable maze system can consistently
188
replicate the existing standard mazes and serve as tests of spatial learning and memory.
189
190
Place field remapping during the maze morphing experiment
191
Several lines of evidence suggest that the hippocampal place coding partially or
192
completely changes in response to external or internal cues (Muller and Kubie, 1987;
193
Leutgeb et al., 2005). This phenomenon is called the remapping of place fields. To
194
demonstrate the repeatability and reproducibility of our reconfigurable maze system
195
over existing mazes, we provide here a novel morphing experiment to examine the
196
remapping of place fields between different shapes of a maze with an identical path
197
length when the maze shape morphs from square to cruciform to square. As the
198
cruciform maze is an internal structure of the square maze, the total path length does not
12 Lead Contact: Susumu Takahashi
199
change during morphing. Moreover, the four outermost runways of the cruciform maze
200
physically overlap with the path of the square-shaped maze. As our maze realizes the
201
morphing in one enclosure, it enables the identification of factors influencing
202
hippocampal place coding: path integration and spatial reference frame (Figure 5a).
203
To examine the remapping of the place fields, we examined 129 place cells
204
monitored from the dorsal hippocampal CA1 of four rats. Whereas the total path length
205
of the runway and available external landmarks in the mazes were nearly identical
206
across maze shape morphs, the spatial correlation of the firing rate map of place cells
207
within the overlapping runways in the square-shaped maze between the first and second
208
exposures was significantly larger than that between the square-shaped maze and the
209
cruciform maze (Figure 5b–c). This suggests that the maze shape can be a cue capable
210
of inducing remapping of the location of place fields as a spatial reference frame.
211
Moreover, the difference in the maximum firing rate of place fields within the
212
overlapping runways in the square-shaped maze between the first and second exposures
213
was similar to that between the square-shaped maze and the cruciform maze
214
(Figure 5b, d). These results suggest that the maze morphing from square to cruciform is
215
represented by a difference in place field locations without a change in place field firing
216
rate.
13 Lead Contact: Susumu Takahashi
217
218
Discussion
219
We demonstrate our reconfigurable maze system’s ability to provide flexible, scalable,
220
repeatable and reproducible tests by configuring standard existing mazes in a single real
221
environment and by examining spatial navigational behavior and neuronal activity
222
during learning and memory performances in the mazes. Our maze system has the
223
potential to become an invaluable, scalable tool for the study of learning and memory:
224
including working and reference memory, spatial navigation, and decision making, as
225
well as anxiety. Using the reconfigurable maze, any experimenter can consistently
226
replicate a variety of mazes in their laboratory, and the position of the maze parts can be
227
reconfigured for a brief period in a repeatable manner to test several task performances
228
in a single physical environment.
229
Currently, a variety of maze tests are routinely used to gain a greater understanding of
230
learning and memory in animals and for screening behavioral phenotypes in animal
231
models of diseases. Although evidence is being accumulated, it is not necessarily
232
available to contribute to our understanding of learning and memory mechanisms in the
233
brain, because the underlying neuronal signatures are not necessarily similar, even when
234
identical maze shapes are used across different conditions. For instance, if a behavioral
14 Lead Contact: Susumu Takahashi
235
phenotype for spatial memory deficits emerged in two maze tests, each conducted in a
236
different room, the tests may only specifically show the remapping of hippocampal
237
place coding across the different rooms, as demonstrated in the maze morphing
238
experiment of the present study. Indeed, recent studies suggest that place cell activity
239
sequences are linked to future planning (Pfeiffer and Foster, 2013) and episodic-like
240
memory retrieval (Takahashi, 2015), as well as neurodegenerative diseases (Cheng and
241
Ji, 2013). Therefore, such neuronal dissociations should be minimized in a comparative
242
study across maze tests. Place cells show an irregular firing pattern in passive VR
243
experiments (Aronov and Tank, 2014). Therefore, we designed our maze based on the
244
concept that various shapes can be constructed in a single real environment, where
245
several types of learning and memory tests can be performed. Recently developed active
246
VR can markedly minimize such dissociations, specifically for proximal cues (Aronov
247
and Tank, 2014; Stowers et al., 2017). While our maze system and the active VR are not
248
exclusive, the combination will enable the production of unlimited experimental
249
situations without the neuronal distortions arising from both proximal and distal cues in
250
the near future.
251
Our reconfigurable maze system is compatible with techniques for monitoring
252
neuronal activity such as extracellular multiple single-unit recording (Wilson and
15 Lead Contact: Susumu Takahashi
253
McNaughton, 1993) and single-photon calcium imaging (Ziv et al., 2013) for freely
254
behaving rats or mice. As demonstrated by the remapping of place fields across
255
morphed mazes, the combination must be accelerated to allow understanding of the
256
neuronal underpinning of learning and memory. Furthermore, our maze enables rapid
257
prototyping of tasks by reconfiguring the coordination of the parts, including runways,
258
rewards, and obstacles, within a few minutes, accelerating studies on learning and
259
memory.
260
The reconfigurable maze is thus a promising maze platform for testing the
261
performance of learning and memory, including working, reference and episodic
262
memory, decision-making, spatial navigation, and anxiety, as well as for screening
263
behavioral phenotypes of mice, including transgenic models of disease.
264
265
Limitations of the study
266
The initial setup cost and laboratory space for introducing the reconfigurable maze
267
may be barriers to widespread distribution. However, the scalable features minimize
268
these barriers. For instance, a minimal setup realized by removing optional modules
269
from the full configuration (demonstrated by a square-shaped maze without movable
270
walls or shut-off sensors) will reduce the cost. Additionally, the modular architecture
16 Lead Contact: Susumu Takahashi
271
allows the configuration of small mazes to fit limited laboratory space. While the
272
minimal setup may not be a compelling reason for laboratories to invest in the
273
reconfigurable maze, the complexity can be expanded incrementally by adding
274
supplementary parts. Moreover, the set up and validation for novel configured mazes
275
within a laboratory may be time-consuming. Provided that the reconfigurable maze will
276
be widely distributed and standardized, the overall cost will be lowered and information
277
resources describing how to set up new configurations of the maze and validate their
278
performance for specific aims will be shared between researchers. In this way,
279
implementation of the reconfigurable maze will ensure reproducibility and repeatability.
280
Despite its flexibility, reproducibility, repeatability, and scalability, our maze has a
281
few limitations. Whereas oblique runways at an angle of 45 degrees could be embedded
282
into our maze as demonstrated by the configured radial arm maze (Figure 1e), the angle
283
cannot be flexibly modified because the coordinated grid assignment of the parts
284
requires the runways to be placed at a fixed angle. Unlike the recently developed
285
honeycomb maze (Wood et al., 2018), the tower supporting the runway cannot be raised
286
and lowered automatically. Whereas adjustable mechanisms are capable of overcoming
287
those limitations, they increase the damage rate of the physical system during
288
behavioral and neuronal recordings, reducing the major advantage of reproducibility
17 Lead Contact: Susumu Takahashi
289
and repeatability.
290
18 Lead Contact: Susumu Takahashi
291
292
Acknowledgments
293
This work is supported by the JSPS KAKENHI (16H06543, 16H02840, and 19H01131)
294
to S.T..
295
296
Data and code availability
297
The
298
(https://github.com/TakahashiLab/ReconfigurableMazeParts). The controlling software
299
is also freely available (https://github.com/TakahashiLab/ReconfigurableMazeGUI).
detail
of
the
maze
parts
is
freely
available
300
301
Author Contributions
302
S.T. conceived the project. S.H. developed control software. S.H., K.M., R.T., and
303
Y.T. performed behavioral experiments. H.A. and K.I. performed the histological
304
verification. S.H., H.A., and K.I. made apparatus. S.H. performed electrophysiological
305
experiments. S.H. and S.T. performed surgery, data analyses, and wrote the manuscript.
306
307
Declaration of Interests
19 Lead Contact: Susumu Takahashi
308
S.T. is an inventor on Japanese unexamined patent application publication (No.
309
P2019-115315A, applicant: Doshisha University) pertaining to the reconfigurable maze.
310
All remaining authors declare no conflict of interest.
311 312
20 Lead Contact: Susumu Takahashi
313
Reference
314
Andersen, P. et al. (eds) (2009) The Hippocampus Book. Oxford University Press. doi:
315
10.1093/acprof:oso/9780195100273.001.0001.
316
Aronov, D. and Tank, D. W. (2014) ‘Engagement of Neural Circuits Underlying 2D
317
Spatial Navigation in a Rodent Virtual Reality System’, Neuron. Elsevier Inc., 84(2), pp.
318
442–456. doi: 10.1016/j.neuron.2014.08.042.
319
Babb, S. J. and Crystal, J. D. (2005) ‘Discrimination of what, when, and where:
320
Implications for episodic-like memory in rats’, Learning and Motivation. doi:
321
10.1016/j.lmot.2005.02.009.
322
Cheng, J. and Ji, D. (2013) ‘Rigid firing sequences undermine spatial memory codes in
323
a neurodegenerative mouse model’, eLife, 2013(2), pp. 1–26. doi: 10.7554/eLife.00647.
324
Dudchenko, P. A. (2001) ‘How do animals actually solve the T maze?’, Behavioral
325
Neuroscience, 115(4), pp. 850–860. doi: 10.1037/0735-7044.115.4.850.
326
Dudchenko, P. A. (2004) ‘An overview of the tasks used to test working memory in
327
rodents’, in Neuroscience and Biobehavioral Reviews, pp. 699–709. doi:
328
10.1016/j.neubiorev.2004.09.002.
329
Gauthier, J. L. and Tank, D. W. (2018) ‘A Dedicated Population for Reward Coding in
330
the Hippocampus’, Neuron, 99(1), pp. 179-193.e7. doi: 10.1016/j.neuron.2018.06.008.
331
Hu, D. and Amsel, A. (1995) ‘A simple test of the vicarious trial-and-error hypothesis 21 Lead Contact: Susumu Takahashi
332
of hippocampal function.’, Proceedings of the National Academy of Sciences of the
333
United States of America, 92, pp. 5506–5509. doi: 10.1073/pnas.92.12.5506.
334
Leutgeb, S. et al. (2005) ‘Independent codes for spatial and episodic memory in
335
hippocampal neuronal ensembles’, Science, 309(5734), pp. 619–623.
336
Muller, R. U. and Kubie, J. L. (1987) ‘The Effects of Changes in the Environment on
337
the Spatial Firing of Hippocampal Complex-Spike Cells’, Journal of Neuroscience, 7(7),
338
pp. 1951–1968.
339
O’Keefe, J. and Nadel, L. (1978) The Hippocampus as a Cognitive Map. Oxford
340
University Press. doi: 10.1017/CBO9781107415324.004.
341
Olton, D. S. and Feustle, W. A. (1981) ‘Hippocampal function required for nonspatial
342
working memory’, Experimental Brain Research, 41(3–4), pp. 380–389. doi:
343
10.1007/BF00238896.
344
Olton, D. S. and Paras, B. C. (1979) ‘Spatial memory and hippocampal function’,
345
Neuropsychologia. doi: 10.1016/0028-3932(79)90042-3.
346
Olton, D. S. and Samuelson, R. J. (1976) ‘Remembrance of places passed: Spatial
347
memory in rats’, Journal of Experimental Psychology: Animal Behavior Processes. doi:
348
10.1037/0097-7403.2.2.97.
349
Olton, D. S., Walker, J. A. and Gage, F. H. (1978) ‘Hippocampal connections and
22 Lead Contact: Susumu Takahashi
350
spatial discrimination’, Brain Research. doi: 10.1016/0006-8993(78)90930-7.
351
Pfeiffer, B. E. and Foster, D. J. (2013) ‘Hippocampal place-cell sequences depict future
352
paths to remembered goals.’, Nature, 497, pp. 74–9. doi: 10.1038/nature12112.
353
Small, W. S. (1901) ‘Experimental Study of the Mental Processes of the Rat. II’, The
354
American Journal of Psychology. doi: 10.2307/1412534.
355
Stowers, J. R. et al. (2017) ‘Virtual reality for freely moving animals’, Nature Methods,
356
14(10), pp. 995–1002. doi: 10.1038/nmeth.4399.
357
Takahashi, S. (2015) ‘Episodic-like memory trace in awake replay of hippocampal
358
place cell activity sequences’, eLife, 4. doi: 10.7554/eLife.08105.
359
Tecott, L. H. and Nestler, E. J. (2004) ‘Neurobehavioral assessment in the information
360
age’, Nature Neuroscience, pp. 462–466. doi: 10.1038/nn1225.
361
Walf, A. A. and Frye, C. A. (2007) ‘The use of the elevated plus maze as an assay of
362
anxiety-related behavior in rodents’, Nature Protocols, 2(2), pp. 322–328. doi:
363
10.1038/nprot.2007.44.
364
Wilson, M. A. and McNaughton, B. L. (1993) ‘Dynamics of the hippocampal ensemble
365
code for space.’, Science, 261(5124), pp. 1055–1058. doi: 10.1126/science.8351520.
366
Wood, E. R. et al. (2000) ‘Hippocampal neurons encode information about different
367
types of memory episodes occurring in the same location’, Neuron, 27(3), pp. 623–633.
23 Lead Contact: Susumu Takahashi
368
Wood, R. A. et al. (2018) ‘The honeycomb maze provides a novel test to study
369
hippocampal-dependent spatial navigation’, Nature, 554(7690), pp. 102–105. doi:
370
10.1038/nature25433.
371
Xu, H. et al. (2019) ‘Assembly Responses of Hippocampal CA1 Place Cells Predict
372
Learned Behavior in Goal-Directed Spatial Tasks on the Radial Eight-Arm Maze’,
373
Neuron, 101(1), pp. 119-132.e4. doi: 10.1016/j.neuron.2018.11.015.
374
Zhang, K. et al. (1998) ‘Interpreting neuronal population activity by reconstruction:
375
unified framework with application to hippocampal place cells.’, Journal of
376
neurophysiology, 79, pp. 1017–1044.
377
Ziv, Y. et al. (2013) ‘Long-term dynamics of CA1 hippocampal place codes.’, Nature
378
neuroscience. Nature Publishing Group, 16(3), pp. 264–6. doi: 10.1038/nn.3329.
379 380
24 Lead Contact: Susumu Takahashi
381
Figure Legends
382 383
Figure 1. The reconfigurable maze. (a–e) The maze can be configured to form T (a),
384
W (b), figure-8 (c), plus (d), and radial arm (e) mazes in a single enclosure. (f) Runway
385
sections are placed atop towers. (g) The baseplate of each tower has four protrusions
386
that coordinate the placement of the section on the breadboard in a flexible, repeatable
387
way. (h) Configured plus maze for rats with two feeders (arrows), two movable walls
388
(arrowheads), one treadmill (double arrowhead), and two shut-off sensors (dashed
389
arrows). (i) Small version of the reconfigurable maze for mice, configured as a figure-8
390
maze with two feeders (arrows), two movable walls (arrowheads), one treadmill (double
391
arrowhead), and two shut-off sensors (dashed arrows). (j) Assembly time of the rat
392
version of the reconfigurable maze for morphing the shape from square to cruciform.
393
Performance improved with consecutive trials in a day (performance vs. trial: F2,8 =
394
7.453, P = 0.0149), but not with expertise (expert/beginner) (performance vs. expertise:
395
F1,4 = 5.654, P = 0.0762). There was no significant interaction between expertise and
396
experience in a day (expertise vs. trial: F2,8 = 0.320, P = 0.735). Two-way mixed
397
ANOVA was used.
398
(k) Assembly time of the mouse version of the reconfigurable maze for morphing the
399
shape from a rectangle to a T. There was no significant difference between performance
25 Lead Contact: Susumu Takahashi
400
and consecutive trials in a day (performance vs. trial: F2,8 = 3.997, P = 0.0625) or
401
expertise (expert/beginner) (performance vs. expertise: F1,4 = 0.351, P = 0.58). There
402
was no significant interaction between expertise and experience in a day (expertise vs.
403
trial: F2,8 = 0.658, P = 0.543). Two-way mixed ANOVA was used.
404
405
Figure 2. Interlocking gaps do not alter navigational behavior. (a) Schematics of the
406
square-shaped maze test. An example running trajectory of a rat is superimposed on the
407
maze. The numbers indicate gap locations. R1 and R2 indicate food dispensers.
408
Linearized gap locations are illustrated at the bottom. (b–c) Left, the average running
409
speed and head direction of four rats as a function of the linearized location. The dots
410
indicate the location of outliers for each lap. Right, the average number of outliers on
411
the regions around the gap (gray shaded areas) and others (running speed: t = −0.19, df
412
= 6, P = 0.86; head direction: t = −5.53, df = 6, P = 0.0015, Two-tailed paired t-test) **:
413
P < 0.01, n.s.: P > 0.05. Error bars indicate SEM. (d) Left, the percentage of the
414
occupancy time over the entire maze as a function of the linearized location of four rats.
415
Right, the median percentage of the occupancy time at gap locations (median, first and
416
third quartiles, minimum, and maximum indicated). The rats preferentially slow down at
417
gap #7 as compared with gaps #3 and #10, which are located at the top and bottom of
26 Lead Contact: Susumu Takahashi
418
the maze (F1,3 = 152.69, P = 0.0011; gap #7 vs. #3: P = 0.049; gap #7 vs. #10: P = 0.030,
419
One-way repeated-measures ANOVA with post hoc Tukey’s HSD test).
420
421
Figure 3. Interlocking gaps do not alter hippocampal place coding. (a) Left,
422
normalized firing rate map of two representative place cells that have place fields
423
around gap locations in the hippocampal CA1 on the square-shaped maze configuration.
424
Right, normalized firing rate maps of 65 hippocampal place cells simultaneously
425
monitored from a rat (ζA) ordered by the latency of their peak firing rates. Each line is a
426
single unit. Gap locations are indicated at the top. Red indicates maximum firing rates,
427
and blue indicates silent. (b) Distribution of the average width of place fields randomly
428
shuffled from the original sample covering the gaps (3,000 shuffles). Red line indicates
429
the average width of the place fields on the gaps. Dotted lines indicate 5th and 95th
430
percentiles for the shuffled data. (c–d) Left, the number of place fields (c) and the firing
431
rate of multi-unit activity (MUA) in the hippocampal CA1 (d) as a function of
432
linearized location on the square-shaped maze. Right, the average number of place
433
fields (c) (z = 0.14, P = 0.89) and the average firing rate of MUA (z = −1.51, P = 0.13)
434
on the gaps (#3, 4, 9, 10) (green shaded area) and between them (orange shaded area).
435
(e) Representative posterior probability of locations decoded by the Bayesian decoder
27 Lead Contact: Susumu Takahashi
436
from 89 simultaneously monitored cells. Values are indicated by the color bar (right). (f)
437
Left, the difference between actual and decoded locations as a function of linearized
438
location on the square-shaped maze. Right, the average difference between actual and
439
decoded locations on the gaps (#3, 4, 9, 10) (green shaded area) and between them
440
(orange shaded area) (z = 1.34, P = 0.18). Bin width is set at 3 cm. All error bars
441
indicate SEM. All were from two-tailed Wilcoxon rank-sum test, n.s.: P > 0.05.
442
443
Figure 4. Rats and mice can learn to navigate the configured mazes. (a) Schematic
444
of the figure-8 shaped maze created using the rat version of the reconfigurable maze.
445
Example animal trajectories are superimposed on top of the maze. The red circle
446
indicates a branch point where the rat must decide which direction to turn. (b) Same as
447
in (a) but with a treadmill (green). Red dashed lines indicate the gap locations where
448
movable walls are placed to force the rats to run on the treadmill for a delay period. (c)
449
Spatial alternation task performance improved with experience (performance vs. testing
450
day: F1, 4 = 795.9, P < 10−5). (d) Performance of the delayed version of the spatial
451
alternation task of rats improved with experience (performance vs. testing day: F1,2 =
452
531.8, P = 0.0019). All were from one-way repeated-measures ANOVA. Error bars
453
indicate SEM. (e) Top view of the configured double T-maze using the mouse version of
28 Lead Contact: Susumu Takahashi
454
the reconfigurable maze. (f) Schematics of the double T-maze. Example mouse
455
trajectories are superimposed on the maze. Arrows mark the running directions from the
456
food dispensers (R1/R2). The red circles indicate branch points where the mice must
457
decide which direction to turn. Red dotted lines indicate the locations where the
458
movable walls are placed to prevent a reverse run. (g) Spatial alternation task
459
performance of mice improved with experience (performance vs. testing day: F1,2 =
460
7320, P = 0.000136). One-way repeated-measures ANOVA was used. Error bars
461
indicate SEM.
462
463
Figure 5. Place field remapping during the maze morphing experiment. (a)
464
Schematics of square and cruciform mazes configured by our system (top) and the
465
representative running trajectory of a rat (bottom). The runways enclosed by red circles
466
were not moved during the morphing experiment. (b) Representative change of the
467
place field of a place cell recorded from a rat (ζA) during maze morphing. The
468
maximum firing rate is displayed above the top left corner. (c–d) Violin plots of spatial
469
similarity as the spatial correlation of the firing rate map (c) (z = −3.98, P = 10−4) and
470
rate similarity as firing rate difference within place fields (d) (z = −0.60, P = 0.55)
471
between the square maze and cruciform maze, and between the first and second
29 Lead Contact: Susumu Takahashi
472
exposures of the square maze (left). The bar indicates the median. Surrounding each
473
side is a rotated kernel density plot. Their cumulative frequency is graphed to the right.
474
All were from two-tailed Wilcoxon signed-rank test. ***: P < 0.001, n.s.: P > 0.05.
475
30 Lead Contact: Susumu Takahashi
Highlights:
The reconfigurable maze enables flexibly in building a variety of maze paths.
The maze ensures reproducibility, repeatability, and scalability.
The maze does not impair spatial navigational behavior or neuronal activity.
Various tests of learning and memory can be conducted in a single environment.