Content uploaded by Joshua B Julian
Author content
All content in this area was uploaded by Joshua B Julian on May 22, 2015
Content may be subject to copyright.
Place recognition and heading retrieval are mediated
by dissociable cognitive systems in mice
Joshua B. Julian, Alexander T. Keinath, Isabel A. Muzzio, and Russell A. Epstein
1
Department of Psychology, University of Pennsylvania, Philadelphia, PA 19104
Edited by Charles R. Gallistel, Rutgers, The State University of New Jersey, Piscataway, NJ, and approved April 8, 2015 (received for review December 18, 2014)
A lost navigator must identify its current location and recover its
facing direction to restore its bearings. We tested the idea that
these two tasks—place recognition and heading retrieval—might
be mediated by distinct cognitive systems in mice. Previous work
has shown that numerous species, including young children and
rodents, use the geometric shape of local space to regain their
sense of direction after disorientation, often ignoring nongeomet-
ric cues even when they are informative. Notably, these experi-
ments have almost always been performed in single-chamber
environments in which there is no ambiguity about place identity.
We examined the navigational behavior of mice in a two-chamber
paradigm in which animals had to both recognize the chamber in
which they were located (place recognition) and recover their fac-
ing direction within that chamber (heading retrieval). In two exper-
iments, we found that mice used nongeometric features for place
recognition, but simultaneously failed to use these same features
for heading retrieval, instead relying exclusively on spatial geom-
etry. These results suggest the existence of separate systems for
place recognition and heading retrieval in mice that are differ-
entially sensitive to geometric and nongeometric cues. We spec-
ulate that a similar cognitive architecture may underlie human
navigational behavior.
navigation
|
scene perception
|
spatial representation
|
geometry processing
|
neural specialization
Anavigator who becomes lost must solve two tasks to regain
her bearings. First, she must identify her current location, a
process we term place recognition. Second, she must identify her
current facing direction, a process we term heading retrieval.
These two tasks are logically dissociable from each other: A “you
are here”map identifies location without revealing heading,
whereas a compass reveals heading without identifying location.
Neurophysiological work on rodents suggests that the outputs of
these two processes are represented by distinct neural populations:
Location is coded in the hippocampus, in both general terms
(different environments elicit different hippocampal maps) and
specific terms (place cells fire at specific coordinates within an
environment), whereas heading is encoded by head direction
(HD) cells in several structures including the postsubiculum,
thalamus, and retrosplenial cortex (1–3). However, little is known
about the systems that determine these quantities from perceptual
inputs. In particular, it is not known whether place recognition
and heading retrieval are mediated by the same or different
processing streams.
Here, we use a novel behavioral paradigm to test the hy-
pothesis that the mechanisms that mediate place recognition at
the coarse level (i.e., identification of the current environment) in
mice are dissociable from the mechanisms that mediate heading
retrieval. We use a variant of a spatial reorientation task that has
been used extensively to study navigation behavior in a variety of
species, including rodents and human children (4–7). In the
standard version of the task, the animal (or human) navigator is
first familiarized with a rectangular chamber with a hidden reward
in one of the corners. Once it learns the location of the reward, the
navigator is then removed from the chamber, disoriented, and
placed back into the center of the chamber by facing a randomly
chosen direction. By observing which corner the navigator
chooses when searching for the reward, it is possible to determine
which cues it uses to orient itself in space. Many studies using this
task have demonstrated that geometric cues (i.e., the shape of the
chamber) exert strong control over behavior, often to the exclusion
of other cues. For example, rats trained in this task will search
equally often in the correct corner and in the corner that is di-
agonally opposite. This pattern of behavior indicates that the ani-
mal is using geometry as a cue, because these two corners have the
same spatial relationship to the chamber geometry. The animals
will often ignore other orienting cues such as odors, visual patterns,
and wall color, even when these cues provide polarizing in-
formation that could potentially resolve the geometric ambiguity
(6, 8). Although the exclusive reliance on geometric cues is not
found under all circumstances (9, 10), it has been observed in a
large number of studies.
An important aspect of this classical paradigm, which to our
knowledge has not been previously commented on, is the fact
that there is no ambiguity about the identity of the environment,
because the experiment is typically performed within a single
chamber (although, see refs. 11 and 12). Thus, in the standard
version of the task, the navigator needs only to reestablish his or
her heading direction to find the reward. Therefore, to examine
place recognition and heading retrieval simultaneously, we used
a novel version of the task in which there were two chambers,
each with unique identifiable features and a different reward
location. We first taught the mice the locations of the reward in
each chamber, and then tested them while alternating between
the two chambers on different test trials. To find the reward in
this case, the mouse must both identify the chamber and de-
termine which direction it is facing within the chamber—in
other words, it must perform both place recognition and heading
retrieval.
Significance
The ability to recover one’s bearings when lost is critical for
successful navigation. To accomplish this feat, a navigator must
identify its current location (place recognition), and it must also
recover its facing direction (heading retrieval). Using a novel
behavioral paradigm, we demonstrate that mice use one set of
cues to determine their location and then ignore these same cues
when determining their heading, although the cues are infor-
mative in both cases. These results suggest that place recogni-
tion and heading retrieval are mediated by different processing
systems that operate in partial independence of each other. This
finding has important implications for understanding the cog-
nitive architecture underlying spatial navigation.
Author contributions: J.B.J., A.T.K., I.A.M., and R.A.E. designed research; J.B.J. and A.T.K.
performed research; J.B.J. and A.T.K. analyzed data; and J.B.J. and R.A.E. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
1
To whom correspondence should be addressed. Email: epstein@psych.upenn.edu.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1424194112/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1424194112 PNAS Early Edition
|
1of6
PSYCHOLOGICAL AND
COGNITIVE SCIENCES
We hypothesized that these two processes—place recognition
and heading retrieval—would be differentially controlled by geo-
metric and nongeometric cues. To test this hypothesis, we used
two geometrically identical rectangular chambers that contained
unique features that allowed them to be discriminated. In Ex-
periment 1, each chamber contained a striped feature attached to
the short wall, which was vertically aligned in one chamber but
horizontally aligned in the other. In Experiment 2, each chamber
contained a vertically striped feature, which was attached to the
short wall in one chamber but attached to the long wall in the
other. Critically, in both cases, the feature was potentially infor-
mative about both the identity of the chamber and heading within
the chamber. To anticipate, we found that mice used the features
to disambiguate the chambers but not to disambiguate headings
within the chambers. In other words, they used features for place
recognition but not heading retrieval, thus demonstrating a dis-
sociation between these two processes.
Results
We first set out to show that mice trained in a classical single-
chamber reorientation paradigm use geometric cues to reorient
themselves while ignoring nongeometric cues—a pattern often
found in other species. Previous work has demonstrated that
mice use geometry for reorientation, but the effect of polarizing
nongeometric cues in the presence of orienting geometry has not
been tested (13). We trained 16 disoriented mice to locate a
reward in the corner of single rectangular (20 ×30 ×25 cm)
chamber with a polarizing cue along one short wall (Fig. 1A). Fig.
1Bpresents the average proportion of trials that mice searched
in each of the four chamber corners over 16 total test trials. Mice
searched for the reward more often in the two corners that were
geometrically appropriate (C and R in Fig. 1A) than in the two
corners that were geometrically inappropriate (F and N), repli-
cating the previous finding of sensitivity to geometry [Cohen’s
d=1.30, t(15) =5.20, P<0.001]. Moreover, they failed to use
the orienting feature to distinguish the correct corner (C) from
the geometrically equivalent corner that was diagonally opposite
(R), thus showing the same insensitivity to nongeometric cues
when determining facing direction often found in other species
[Cohen’sd=0.39, t(15) =1.57, P=0.14].
We then examined the navigational behavior of 16 disoriented
mice in a two-chamber paradigm. The animals were presented
alternately with two rectangular chambers that were geometrically
identical (20 ×30 ×25 cm) but distinguishable by stripes along
one short wall. The stripes were vertical in one chamber and
horizontal in the other (Fig. 2A). Because this feature both dif-
ferentiated between the chambers and acted as a polarizing cue, it
could be used for both place recognition and heading retrieval. In
one chamber, mice were rewarded when they searched in the left
corner nearest the striped wall, and in the other, when they
searched in the right corner nearest the striped wall. We pre-
dicted that mice would use the stripes to identify the chamber in
which they were located, but would not use the stripes to dis-
ambiguate between geometrically equivalent headings.
The results upheld our predictions. Fig. 2Bpresents the av-
erage proportion of trials that mice searched in each of the four
corners in each of the two chambers (16 total test trials per
chamber). Typical animal behavior in this task can be viewed in
Movie S1. In neither chamber did the distribution of search
frequencies across all corners (C, R, N, and F in Fig. 2A) differ
significantly from those of the control animals trained in the
classical single-chamber paradigm [both X
2
(3) <6.05, P>0.11]. In
both chambers, the animals searched more often in the geo-
metrically appropriate corners (C and R in Fig. 2A; bolded in Fig.
2B) than the geometrically inappropriate corners (N and F in Fig.
2A) [horizontally striped chamber: Cohen’sd=0.65, t(15) =2.59,
P=0.02; vertically striped chamber: Cohen’sd=0.68, t(15) =2.73,
P=0.02; Fig. 3A]. This observation was confirmed by a 2 (absolute
corner location: long wall left or right) ×2 (chamber: vertically
striped or horizontally striped) repeated-measures ANOVA, which
revealed a significant interaction between absolute corner location
and chamber [F(1,15) =22.72, P<0.001, η
p
2
=0.60; Fig. 2C]. Be-
cause the geometrically appropriate corners differed between the
two chambers, this pattern of performance indicates that the mice
must have distinguished between the chambers. Given that the
identity of the feature (horizontal vs. vertical) was the only thing
that differed between the two chambers, these findings strongly
suggest that the animals used the feature for chamber discrimina-
tion (i.e., place recognition).
We then performed an additional statistical test to see whether
the animals used this feature to distinguish between geometrically
equivalent headings within each chamber. The classic finding with
rectangular chambers is that animals do not distinguish the re-
warded location from the diagonally opposite location, even in the
presence of a nongeometric polarizing cue. We replicate the
classic finding here: In neither context did animals search more at
the correct location than the diagonally opposite corner [both
Cohen’sd<0.14, t(15) <0.57, P>0.58]. Thus, the mice used the
striped feature to distinguish between the chambers, but simulta-
neously failed to use this potentially informative feature to dis-
ambiguate between headings. That is, for heading retrieval, the
mice solely relied on geometry.
In Experiment 2, we further explored the range of features
that are used for place recognition. In particular, we asked
whether the mice could discriminate between the chambers based
on the spatial location of a feature relative to chamber geometry.
The paradigm was similar to Experiment 1. A new group of dis-
oriented mice (n=16) were trained to locate rewards in the
corners of two rectangular chambers, with different reward loca-
tions in each chamber. In this case, the same vertical striped
feature was present in both environments, but in different loca-
tions: In one chamber, the feature was on a short wall, whereas in
the other chamber, it was in the center of a long wall (Fig. 3A).
Thus, to disambiguate the chambers, the animals had to process
the location of the feature relative to the chamber geometry. They
could not distinguish the chambers on the basis of feature
identity alone.
A
C
R
N
F
38%
29%
16%
17%
(4.1)
(2.4)(2.2)
(2.1)
B
Fig. 1. Design and results for the preliminary experiment, which used the
classical one-chamber reorientation paradigm. (A) Disoriented mice were
trained to locate a reward in a single rectangular chamber with a visual
feature along one short wall. C, R, N, and F denote the four cups in the
corners of the chamber, where C denotes the correct corner (i.e., the corner
with the hidden reward), R the rotationally equivalent corner (i.e., the cor-
ner geometrically equivalent to C), N the near corner (i.e., the corner that is
closest to C), and F the far corner (i.e., the nonrotationally equivalent corner
farthest from C). (B) Percentage of first digs in each of the four corners of the
chamber (and SEMs). The star denotes the rewarded location. Mice searched
significantly more often at C and R (bolded) than N and F, but there was no
significant difference between the percentage of digs at C and R. This pat-
tern reprises the classical results.
2of6
|
www.pnas.org/cgi/doi/10.1073/pnas.1424194112 Julian et al.
Fig. 3Bshows the average proportion that mice searched in
each of the four corners, separately for each context. In neither
chamber did the distribution of search frequencies across all
corners differ significantly from those of the control animals
trained to locate a reward in the classical single-chamber para-
digm [both X
2
(3) <2.61, P>0.46]. We again found that in both
chambers, mice searched more often in the geometrically ap-
propriate corners than in the geometrically inappropriate cor-
ners [long-wall chamber: Cohen’sd=1.31, t(15) =5.58, P<
0.0001; short-wall chamber: Cohen’sd=0.92, t(15) =3.93, P=
0.001; Fig. 3C]. Confirming this finding, a 2 (absolute corner
location: long wall left or right) ×2 (chamber: long wall feature
or short wall feature) repeated-measures ANOVA revealed a
significant interaction between absolute corner location and
chamber [F(1,15) =54.578, P<0.001, η
p
2
=0.78; Fig. 3C].
Moreover, we once again observed that animals searched in the
correct corner and the geometrically equivalent corner with
equal frequency [both Cohen’sd<0.39, t(15) <1.59, P>0.14].
These results replicate the pattern of findings from Experi-
ment 1. Once again, mice used a cue to distinguish between the
chambers and then ignored the same cue when determining their
facing direction. In this case, the cue in question was the location
of the striped feature relative to the geometry. These results
suggest that the place recognition system can use a variety of
cues, including both spatial and nonspatial features, and that
information about the location of a cue relative to chamber ge-
ometry can be incorporated into its calculations. The heading
retrieval system, however, seems to rely solely on geometry (at
least in our experiments; see Discussion).
A possible alternative account of the results in Experiment 2 is
that the animals did not, in fact, distinguish between the cham-
bers, but rather treated the two contexts as identical and used the
feature to specify a principal orientational axis for the environ-
ment (14). We think such an account is unlikely, because it would
require the animals to ignore the geometry of the room when
determining heading; moreover, it would require them to use the
feature as an axis-defining cue (North–South vs. East–West) but
not as a polarizing cue (North vs. South). Nevertheless, to test
this possibility, we ran 15 of the 16 animals in Experiment 2 in
two square chambers (one large and one small) following the last
day of testing. Each of these chambers had the vertical striped
feature along one wall (Fig. S1). We reasoned that if mice were
using the feature to define the principal axis while ignoring ge-
ometry, then they should continue to use this strategy in the
square chamber. In this case, they should search in the location
on the left side of the feature and also in the diagonally opposite
corner. However, this is not what we observed. Instead, the mice
searched no more often at left-of-feature corner and the corner
diagonally opposite than they did at right-of-feature corner and
the corner diagonally opposite [small-square chamber: Cohen’s
d=0, t(14) =0.00, P=1.0; large-square chamber: Cohen’sd=
0.23, t(14) =0.90, P=0.38; Fig. S1]. Thus, the mice did not use
strategy of going to the corners on the left-of-feature diagonal
during the main part of Experiment 2, but rather used the lo-
cation of the feature to distinguish the chambers and the ge-
ometry of the chamber to determine their heading.
Taken together, the results from Experiments 1 and 2 indicate
that when disoriented mice were faced with a situation in which
B
28%
30%
22%
20%
(2.8)
(2.9)(2.0)
(3.3)
30%
29%
22%
19%
(2.4)
(1.7)(3.4)
(2.4)
A
CN
RF
C
R
N
F
C
Vertical
Chamber
Horizontal
Chamber
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Proportion First Dig
*
*
***
Long Wall Left
Long Wall Righ
t
Absolute Corner
Location
Fig. 2. Design and results for Experiment 1. (A) Mice were trained to locate a hidden reward in two rectangular chambers that had identical geometry but were
distinguishable by the orientation of stripes (vertical vs. horizontal) along a single short wall. C, R, N, and F denote the four cups in the corners of the chambers,
where C denotes the correct corner, R the geometrically equivalent corner, N the near corner, and F the far corner. Note that the location of the rewardedcup
differed between the two chambers. (B) Shows the average percentage of first digs (and SEMs) in each corner of the two chambers. Stars denote the rewarded
locations; bolded numbers indicate digs in geometrically appropriate corners. (C) The bar chart shows the same data as in B, but averaged over geometrically
equivalent corn ers. Error bars denote ±1 SEM. Mice dug more often in the corners that were geometrically appropriate for each chamber, thus indicating that
they distinguished between the chambers. Moreover, they did not distinguish between geometrically appropriate corners. *P<0.05, ***P<0.001.
BAC
C
RF
NC
NR
F32%
31%
20%
17%
(2.3)
(3.0)(1.9)
(2.8)
36%
29%
20%
15%
(2.0)
(2.1)(2.6)
(3.0)
Long Wall Left
Long Wall Righ
t
Absolute Corner
Location
Short Wall
Chamber
Long Wall
Chamber
***
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Proportion First Dig
**
***
Fig. 3. Design and results for Experiment 2. (A) The design was the same as Experiment 1, but in this case, the two chambers were distinguished by the location of
vertical stripes either on the short wall or the long wall. (B) Shows the average percentage of first digs (and SEMs) in each corner of the two chambers. Stars denote
the rewarded locations; bolded numbers indicated digs in geometrically appropriate corners. (C) The bar chart shows the same data as in B, but averaged over
geometrically equivalent corners. Error bars denote ±1 SEM. Mice dug more often in the corners that were geometrically appropriate for each chamber, thus
indicating that they distinguished between the chambers. Moreover, they did not distinguish between geometrically appropriate corners.**P<0.01, ***P<0.001.
Julian et al. PNAS Early Edition
|
3of6
PSYCHOLOGICAL AND
COGNITIVE SCIENCES
they had to both identify their environment and also reestablish
a sense of direction within that environment, they used both
nongeometric (Exp. 1) and geometric (Exp. 2) information to
identify their environment, but only geometric information to
reestablish their sense of direction. To further test this account,
we calculated the Bayes factor (15, 16) comparing the alternative
hypothesis that the geometrically appropriate corners were
chosen more often than the inappropriate corners to the null
hypothesis that the geometrically appropriate and inappropriate
corners were chosen equally often. Combining data from both
experiments, this analysis revealed an average Bayes factor of
3.81 in favor of the alternative hypothesis that animals used the
feature to discriminate chambers, a magnitude that is considered
to provide “substantial”evidence (17). To verify that the same
cues that were used for place recognition were ignored for
heading retrieval, we computed the Bayes factor comparing the
alternative hypothesis that animals searched more at the correct
corner than the diagonally opposite corner to the null hypothesis
that the proportion of searches at both geometrically appropriate
corners was equal. In this case, the average Bayes factor was
1.79 ×10
−6
, which provides evidence in favor of the null hypothesis.
Discussion
Using a two-chamber spatial reorientation paradigm, we found a
dissociation between two fundamental components of spatial
navigation: place recognition and heading retrieval. When dis-
oriented mice were faced with a situation in which they had to
both identify their environment and also reestablish a sense of
direction within that environment, they used both geometric and
nongeometric information to identify their environment, but
relied solely on spatial geometry to retrieve their heading. Crit-
ically, the same cue that was used for place recognition was ig-
nored for heading retrieval, although it was highly informative
in both cases. Thus, our results cannot be explained by unequal
salience of cues.
We demonstrated this dissociation between place recognition
and heading retrieval in two experiments. In Experiment 1, the
animals searched for hidden rewards in two geometrically iden-
tical rectangular chambers, each of which had a distinguishing
feature (horizontal vs. vertical stripes) along one of the short
walls. In Experiment 2, the chambers were also geometrically
identical rectangles, but in this case, the distinguishing feature
was the location of a vertically striped feature relative to the
chamber geometry (along short wall vs. along long wall). In both
experiments, we reprised the classic results from the literature by
showing that the animals searched for the reward more often in the
two corners that were geometrically appropriate for each chamber
than in the corners that were geometrically inappropriate; fur-
thermore, they did not distinguish between the two geometrically
appropriate corners (i.e., the correct corner and its rotational op-
posite). The fact that the animals chose the corners that were
geometrically appropriate for each chamber indicates that they
must have used the identity (Exp. 1) or location (Exp. 2) of the
striped feature to distinguish between the chambers, because these
were the only disambiguating cues. However, the fact that they did
not distinguish between the two geometrically appropriate corners
indicates that they did not use the striped features to distinguish
between headings, although these features clearly polarized the
environment. These results demonstrate a functional dissociation
between place recognition and heading retrieval: The striped fea-
ture acts as a treatment that selectively affects one process (place
recognition) but does not affect the other (heading retrieval). (See
SI Discussion 1 for further consideration of this point.)
To our knowledge, this is the first demonstration of this disso-
ciation. A previous reorientation study by Horne et al. reported
that rats could discriminate between a rectangular chamber with
all black walls and a rectangular chamber with all white walls (12).
This result is consistent with ours insofar as it indicates that the
animals can use nongeometric cues for place recognition. How-
ever, because the wall colors in the Horne study did not specify a
unique heading within the chambers, their design did not allow
them to dissociate between place recognition and heading re-
trieval as we do here.
Why might heading retrieval and place recognition rely on
distinct cognitive systems? One possibility is that solving these two
tasks requires different computations. Place recognition likely in-
volves identification of scenes or landmarks that a navigator can
use to determine her general environmental context. Identification
might be achieved by matching the contents of the current view
with the contents of a previously stored view consisting of a
combination of geometric and nongeometric information (18). In
this account, place recognition would be akin to object recogni-
tion, but performed on navigationally relevant stimuli. By contrast,
heading retrieval might involve interpreting the environment in
terms of a spatial reference system from which orientational axes
can be recovered (14). Although the precise computations un-
derlying heading retrieval are unknown, previous work suggests
that at least for humans, heading retrieval is not performed by
view matching (4, 19, 20). (See SI Discussion 2 for further con-
sideration of the implications of the present experiments for view-
matching theories of reorientation.)
Notably, previous work has identified a possible neuroanatomi-
cal basis for this behavioral dissociation. In humans, neuroimaging
and neuropsychological work suggests that place recognition is
primarily mediated by the parahippocampal place area (PPA), a
region of medial occipitotemporal cortex that responds strongly
when subjects view environmental scenes or landmark objects
(21–23), whereas heading retrieval is primarily mediated by a
system centered around the retrosplenial complex (RSC) in the
medial parietal lobe (24–28). Analogous to the current findings,
the PPA appears to be sensitive to both geometric and non-
geometric information (22, 29–33), whereas RSC appears to be
especially sensitive to geometry when people retrieve spatial
information from memory (28). In rodents, the homologous re-
gions are postrhinal cortex (34), which has been shown to be
important for place recognition (35), and retrosplenial cortex,
which has been shown to be important for deriving directional
information from environmental cues (36). Retrosplenial cortex
contains head direction (HD) cells, which discharge selectively
when the head of an animal is oriented in a particular facing
direction (37), and a previous report demonstrated that these
cells are primarily sensitive to environmental geometry rather
than nongeometric features after disorientation (38). In addition,
neurons that code allocentric locations relative to geometric
boundaries have been identified in the entorhinal cortex (39) and
subiculum (40) of the rodent, and these cells might be important
for retrieving the location of the reward within the chamber after
chamber identity and heading have been reestablished.
A possible caveat concerning our interpretation of the present
experiments in terms of separable systems for place recognition
and heading retrieval is that, as with any behavioral dissociation,
we cannot know for certain the identity of the processes that we
have dissociated. Although we think that place recognition and
heading retrieval provide the most parsimonious descriptions of
these processes, other accounts may also explain the data. For
example, rather than distinguishing between the chambers as
distinct environments, the animals might be distinguishing be-
tween two different situations that occur in the same environ-
ment, just as a person might distinguish between a wedding and
a funeral that both occur in the same building. Although the
spatial environment of both events is the same in this example,
the contextual features surrounding each situation and the ap-
propriate behaviors are different. Relatedly, we cannot know for
certain that the second system supports retrieval of heading. An
alternative possibility is that it codes egocentric locations relative
to geometric boundaries, and that the animals choose their dig
4of6
|
www.pnas.org/cgi/doi/10.1073/pnas.1424194112 Julian et al.
locations based on a strategy of approaching a corner with a
particular local geometric configuration (e.g., short-wall left),
without recovering heading at all. In addition, we emphasize once
again that our results pertain to the mechanisms that allow the
animal to recover its bearings after disorientation and do not
necessarily provide insight into the mechanisms that allow the
animal to maintain its bearings when oriented.
Finally, it is worth considering the implications of our findings
for the ongoing debate about the nature of the cognitive mecha-
nisms underlying spatial reorientation. Two theories are most
prominent. The first theory builds on the classic results by arguing
that reorientation is mediated by an encapsulated cognitive
module (41) that specifies a navigator’s position and orientation
relative to the geometric structure of the environment but is
insensitive to nongeometric features (5–7). The second theory ar-
gues that a range of environmental cues, including both geometry
and nongeometric features, can guide spatial reorientation (9, 10,
42) and that the combination of cues used in any given situation
can vary depending on their salience and reliability. Although our
results might seem at first glance to fit more closely with the first
view insofar as we postulate the operation of independent mech-
anisms, one of which is especially sensitive to geometry, it is im-
portant to note that our argument does not require that these two
mechanisms be modular. More specifically, the dissociability of the
place recognition and heading retrieval systems that we demon-
strate here does not require heading retrieval to be impervious to
nongeometric information under all circumstances. The key
point is that we have found one set of circumstances in which
nongeometric information is used for one function but not the
other, thus establishing the independent operation of the two
mechanisms. That said, if our conclusion that there are separate
cognitive systems for place recognition and heading retrieval is
correct, it may affect the interpretation of cue competition ef-
fects that have been taken as evidence in favor of nonmodular
theories (43–48). In particular, some cue competition studies
have observed that when animals learn to find a goal in a
chamber containing both featural and geometric cues, and the
featural cues are then altered or removed, then the animals are
impaired at finding the goal. These findings have been interpreted
as indicating that the learning of locations relative to featural cues
can overshadow the learning of locations relative to geometric
cues, in contradiction to the predictions of the modular theory.
However, our results suggest an alternative account: When the
featural cues are changed, animals may believe that they are in a
different place for which they do not know the location of the
reward. Thus, some cue competition effects may be explained by
the existence of a place recognition system that is sensitive to
nongeometric features. Conversely, under this interpretation, the
failure of a feature to interfere with learning based on environ-
mental geometry (49–52) may indicate that the feature did not
form an integral part of the representation of that place.
In sum, our experiments demonstrate a dissociation between
place recognition and heading retrieval in mice. Whereas place
recognition is sensitive to both featural and geometric information,
heading retrieval is primarily guided by spatial geometry. These
findings indicate that place recognition and heading retrieval are
mediated by different cognitive systems that operate with some
degree of independence from each other. For a lost navigator to
regain her bearings, she must solve not one but two problems, and
both systems must work in concert to get her on her way.
Methods
Subjects. Distinct groups of 16 male C57BL/6 mice, 2–5 mo old (Jackson
Laboratory), participated in the classical single-chamber paradigm, Experi-
ment 1, and Experiment 2 (48 animals total). Mice were housed individually
and kept on a 12-h light/dark cycle for at least 2 wk before the beginning of
the experiments. They had access to water ad libitum, but to increase mo-
tivation to participate in the task, they were maintained at 85–90% of their
free-feed weight. Starting 4 d before the experiment, animals were shaped
to dig in a medicine cup for a food reward (Kellogg’s Cocoa Krispies) in their
home cage by providing them once daily with the reward gradually buried
deeper under scented bedding. Animal living conditions were consistent
with the standards set forth by the Association for Assessment and Ac-
creditation of Laboratory Animal Care. All experiments were approved by
the Institution of Animal Care and Use Committee of the University of
Pennsylvania and were conducted in accordance with NIH guidelines.
Apparatus. The classical single-chamber experiment was conducted in a
rectangular (20 ×30 ×25 cm) chamber. Exp. 1 and 2 were both conducted
in two geometrically identical rectangular (20 ×30 ×25 cm) chambers.
The walls and floor of all chambers were covered in white laminate. In the
single-chamber experiment, there were three black stripes (either vertical
or horizontal, balanced across animals) along one short wall (Fig. 1A). In
Exp. 1, the two chambers were distinguished by three black stripes along
the short wall, which were vertical in one chamber and horizontal in the
other (Fig. 2A). In Exp. 2, chambers were distinguished by the location of
three vertical black stripes, which were placed along the short wall in one
chamber and in the center of the long wall in the other (Fig. 2B). In all
experiments, stripes were 4 cm in width. Testing in all chambers occurred
in the same location in the experimental room. The chambers were sur-
rounded by a square black curtain with rounded corners, were uniformly
lit from overhead, and a white noise generator was hung centrally above
the chamber to ensure that animals could not use extraneous sounds as
beacons. Cups were embedded in each of the four corners of the chamber
floors. The cups contained odor-masked bedding, consisting of 1 g of
odor mask (either ground cumin or ginger) for every 100 g of bedding.
Mouse behavior was recorded by using LimeLight video tracking system
(Coulbourn Instruments) via an overhead, centrally located camera.
Design and Procedure. A pilot experiment showed that mice could discriminate
the horizontal and vertical stripes to a performance criterion of 75% correct after
eight training trials. Thus, all experiments began with a training phase consisting
of four training trials per chamber per day for 2 d, with successive trials alter-
nated across chambers (8 trials total in the one chamber experiment; 16 trials
total in the two chamber experiments). During this training, mice were taught
to search for a reward, which was visible for the first two training trials per
chamber and buried in the remaining training trials. In the single-chamber
experiment, the reward was always located in one of the two corners nearest
the striped feature. In Exp. 1 and 2, the reward was always located in one of
these two corners in one chamber, and inthe other feature-adjoining corner in
the other chamber. These locations were counterbalanced across animals;
however, forall analyses and figures, the percentage of searchesat each corner
are reflected such that correct corner is the same for all animals.
Animals were disoriented before the start of every trial. To disorient an
animal, it was placed in a PVC cylinder with a detachable base and lid. The
experimenter slowly rotated the cylinder on a turntable roughly four full
clockwise then four full counterclockwise revolutions. The cylinder was then
carried to the chamber, and the base was slid out from underneath the animal.
The cylinder was lifted to start a trial. To ensure that the animals could not use
any room cues that were not completely eliminated by use of a surrounding
curtain and a white-noise generator, chambers were rotated 90° or 180° before
each trial, counterbalanced so that all orientations relative to the room were
experienced equally often. The chambers were cleaned with ethanol at the end
of each trial to remove odor trails. The intertrial interval was 3–5min.
Following training, animals were tested in one session per day for 4 d. In
the single chamber experiment, testing sessions consisted of two rewarded
and two unrewarded trials (interleaved). In Exp. 1 and 2, testing sessions
consisted of two rewarded and two unrewarded trials per chamber. Thus, in
all cases, there were a total of 16 test trials per chamber. In Exp. 1 and 2,
chambers were tested in an interleaved fashion, as were the rewarded and
unrewarded trials. So, a sequence for one session might be: Chamber 1
(rewarded), Chamber 2 (rewarded), Chamber 1 (unrewarded), Chamber 2 (un-
rewarded), etc. The order in which chambers were tested was counterbalanced
across sessions. Duri ng reward trials, mice were remove d from the apparatus
after theyhad found the reward. During unrewarded trials,they were removed
after their first dig, or after 45 s (whichever came later). Digs were counted
whenever an animal removed bedding from a cup by using one or both paws.
Unrewardedtrials were included to train the mice to concentrate their first dig
at the reward location and as a control for the possibility that mice could smell
the reward during rewarded trials. However, there was no difference in the
distribut ion of first digs across corners between rewarded and u nrewarded
trials in the classical single-chamber experiment [X
2
(3) =4.67, P=0.20], Exp. 1
Julian et al. PNAS Early Edition
|
5of6
PSYCHOLOGICAL AND
COGNITIVE SCIENCES
[X
2
(3) =0.70, P=0.87], or Exp. 2 [X
2
(3) =5.44, P=0.14]. Therefore, we col-
lapsed across rewarded and unrewarded trials for all analyses.
Dig locations were coded following testing by an experimenter blind to
condition. The dependent measure was the first corner in which the animal
dug. Paired sample ttests were used to assess whether the proportion of
digs were distributed in the chambers according to the geometry. For Exp. 1
and 2, repeated measures ANOVA with absolute corner location (long wall
left or right) and chambers as within-subjects factors were used to compare
the search behavior across chambers. All reported statistics are based on
two-tailed significance tests.
The day following t he final Exp. 2 testing session, 15 of th e 16 animals that
participated in Exp. 2 were then run in a control experiment in which the
animal’s search behavior was observed in two square chambers, one large
(30 ×30 ×25 cm) and one small (20 ×20 ×25 cm). Cups were embedded in
each of the four corners of the chamber floors. Both chambers had the same
vertical stripe feature along one wall that used in Exp. 2. There were four
interleaved probe trials per square. Every trial per chamber, the chambers were
rotated 90° or 180°. Animals were disoriented before the start of each trial.
ACKNOWLEDGMENTS. We thank Carina Zhang and Darby Marx for help
with data collection, and Nora Newcombe and Steven Marchette for
comments on the manuscript. This work was supported by US National
Institutes of Health Grant EY022350 and National Science Foundation (NSF)
Grants SBE-0541957 and SBE-1041707 (to R.A.E.), NSF CAREER Award
1256941 (to I.A.M.), NSF Integrative Graduate Education and Research
Traineeship (to A.T.K. and J.B.J.), and NSF Graduate Research Fellowship
(to J.B.J.).
1. Taube JS (2007) The head direction signal: Origins and sensory-motor integration.
Annu Rev Neurosci 30:181–207.
2. Colgin LL, Moser EI, Moser M-B (2008) Understanding memory through hippocampal
remapping. Trends Neurosci 31(9):469–477.
3. O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evi-
dence from unit activity in the freely-moving rat. Brain Res 34(1):171–175.
4. Cheng K, Huttenlocher J, Newcombe NS (2013) 25 years of research on the use of
geometry in spatial reorientation: A current theoretical perspective. Psychon Bull Rev
20(6):1033–1054.
5. Gallistel CR (1990) The Organization of Learning (MIT Press, Cambridge, MA).
6. Cheng K (1986) A purely geometric module in the rat’s spatial representation. Cog-
nition 23(2):149–178.
7. Hermer L, Spelke ES (1994) A geometric process for spatial reorientation in young
children. Nature 370(6484):57–59.
8. Margules J, Gallistel CR (1988) Heading in the rat: Determination by environmental
shape. Anim Learn Behav 16(4):404–410.
9. Learmonth AE, Nadel L, Newcombe NS (2002) Children’s use of landmarks: Implica-
tions for modularity theory. Psychol Sci 13(4):337–341.
10. Cheng K, Newcombe NS (2005) Is there a geometric module for spatial orientation?
Squaring theory and evidence. Psychon Bull Rev 12(1):1–23.
11. Graham M, Good MA, McGregor A, Pearce JM (2006) Spatial learning based on the
shape of the environment is influenced by properties of the objects forming the
shape. J Exp Psychol Anim Behav Process 32(1):44–59.
12. Horne MR, et al. (2010) Lesions of the perirhinal cortex do not impair integration of
visual and geometric information in rats. Behav Neurosci 124(3):311–320.
13. Twyman AD, Newcombe NS, Gould TJ (2009) Of mice (Mus musculus) and toddlers
(Homo sapiens): Evidence for species-general spatial reorientation. J Comp Psychol
123(3):342–345.
14. Cheng K, Gallistel C (2005) Shape parameters explain data from spatial trans-
formations: Comment on Pearce et al. (2004) and Tommasi & Polli (2004) J Exp Psychol
Anim Behav Process 31(2)254–259, discussion 260–261.
15. Dienes Z (2011) Bayesian versus orthodox statistics: Which side are you on? Perspect
Psychol Sci 6(3):274–290.
16. Gallistel CR (2009) The importance of proving the null. Psychol Rev 116(2):439–453.
17. Jeffreys H (1998) The Theory of Probability (Oxford Univ Press, Oxford).
18. Wang R, Spelke E (2002) Human spatial representation: Insights from animals. Trends
Cogn Sci 6(9):376–382.
19. Nardini M, Thomas RL, Knowland VC, Braddick OJ, Atkinson J (2009) A viewpoint-
independent process for spatial reorientation. Cognition 112(2):241–248.
20. Lee SA, Spelke ES (2011) Young children reorient by computing layout geometry, not
by matching images of the environment. Psychon Bull Rev 18(1):192–198.
21. Aguirre GK, Zarahn E, D’Esposito M (1998) An area within human ventral cortex
sensitive to “building”stimuli: Evidence and implications. Neuron 21(2):373–383.
22. Epstein R, Kanwisher N (1998) A cortical representation of the local visual environ-
ment. Nature 392(6676):598–601.
23. Epstein RA, Vass LK (2014) Neural systems for landmark-based wayfinding in humans.
Philos Trans R Soc Lond B Biol Sci 369(1635):20120533.
24. Epstein RA (2008) Parahippocampal and retrosplenial contributions to human spatial
navigation. Trends Cogn Sci 12(10):388–396.
25. Vann SD, Aggleton JP, Maguire EA (2009) What does the retrosplenial cortex do? Nat
Rev Neurosci 10(11):792–802.
26. Baumann O, Mattingley JB (2010) Medial parietal cortex encodes perceived heading
direction in humans. J Neurosci 30(39):12897–12901.
27. Vass LK, Epstein RA (2013) Abstract representations of location and facing direction in
the human brain. J Neurosci 33(14):6133–6142.
28. Marchette SA, Vass LK, Ryan J, Epstein RA (2014) Anchoring the neural compass:
Coding of local spatial reference frames in human medial parietal lobe. Nat Neurosci
17(11):1598–1606.
29. Dilks DD, Julian JB, Kubilius J, SpelkeES, Kanwisher N (2011) Mirror-imagesensitivity and
invariance in object and scene processing pathways. J Neurosci 31(31):11305–11312.
30. Wolbers T, Klatzky RL, Loomis JM, Wutte MG, Giudice NA (2011) Modality-indepen-
dent coding of spatial layout in the human brain. Curr Biol 21(11):984–989.
31. Kravitz DJ, Peng CS, Baker CI (2011) Real-world scene representations in high-level
visual cortex: It’s the spaces more than the places. J Neurosci 31(20):7322–7333.
32. Cant JS, Xu Y (2012) Object ensemble processing in human anterior-medial ventral
visual cortex. J Neurosci 32(22):7685–7700.
33. Park S, Brady TF, Greene MR, Oliva A (2011) Disentangling scene content from spatial
boundary: Complementary roles for the parahippocampal place area and lateral oc-
cipital complex in representing real-world scenes. J Neurosci 31(4):1333–1340.
34. Burwell RD, Witter MP, Amaral DG (1995) Perirhinal and postrhinal cortices of the rat:
A review of the neuroanatomical literature and comparison with findings from the
monkey brain. Hippocampus 5(5):390–408.
35. Norman G, Eacott MJ (2005) Dissociable effects of lesions to the perirhinal cortex and
the postrhinal cortex on memory for context and objects in rats. Behav Neurosci
119(2):557–566.
36. Pothuizen HH, Aggleton JP, Vann SD (2008) Do rats with retrosplenial cortex lesions
lack direction? Eur J Neurosci 28(12):2486–2498.
37. Taube JS (1998) Head direction cells and the neurophysiological basis for a sense of
direction. Prog Neurobiol 55(3):225–256.
38. Knight R, Hayman R, Lin Ginzberg L, Jeffery K (2011) Geometric cues influence head
direction cells only weakly in nondisoriented rats. J Neurosci 31(44):15681–15692.
39. Solstad T, Boccara CN, Kropff E, Moser M-B, Moser EI (2008) Representation of geo-
metric borders in the entorhinal cortex. Science 322(5909):1865–1868.
40. Lever C, Burton S, Jeewajee A, O’Keefe J, Burgess N (2009) Boundary vector cells in the
subiculum of the hippocampal formation. J Neurosci 29(31):9771–9777.
41. Fodor JA (1983) The Modularity of Mind: An Essay on Faculty Psychology (MIT Press,
Cambridge, MA).
42. Newcombe NS, Ratliff KR (2007) Explaining the development of spatial reorientation:
Modularity-plus-language versus the emergence of adaptive combination. The
Emerging Spatial Mind, eds Plumert JM, Spencer JP (Oxford Univ Press, New York),
53–76.
43. Horne MR, Pearce JM (2009) A landmark blocks searching for a hidden platform in an
environment with a distinctive shape after extended pretraining. Learn Behav 37(2):
167–178.
44. Horne MR, Iordanova MD, Pearce JM (2010) Spatial learning based on boundaries in
rats is hippocampus-dependent and prone to overshadowing. Behav Neurosci 124(5):
623–632.
45. Gray ER, Bloomfield LL, Ferrey A, Spetch ML, Sturdy CB (2005) Spatial encoding in
mountain chickadees: Features overshadow geometry. Biol Lett 1(3):314–317.
46. Pearce JM, Graham M, Good MA, Jones PM, McGregor A (2006) Potentiation, over-
shadowing, and blocking of spatial learning based on the shape of the environment.
J Exp Psychol Anim Behav Process 32(3):201–214.
47. Kosaki Y, Austen JM, McGregor A (2013) Overshadowing of geometry learning by
discrete landmarks in the water maze: Effects of relative salience and relative validity
of competing cues. J Exp Psychol Anim Behav Process 39(2):126–139.
48. Wilson PN, Alexander T (2008) Blocking of spatial learning between enclosure ge-
ometry and a local landmark. J Exp Psychol Learn Mem Cogn 34(6):1369–1376.
49. Hayward A, McGregor A, Good MA, Pearce JM (2003) Absence of overshadowing and
blocking between landmarks and the geometric cues provided by the shape of a test
arena. Q J Exp Psychol B 56(1):114–126.
50. Hayward A, Good MA, Pearce JM (2004) Failure of a landmark to restrict spatial
learning based on the shape of the environment. Q J Exp Psychol B 57(4):289–314.
51. McGregor A, Horne MR, Esber GR, Pearce JM (2009) Absence of overshadowing be-
tween a landmark and geometric cues in a distinctively shaped environment: A test of
Miller and Shettleworth (2007). J Exp Psychol Anim Behav Process 35(3):357–370.
52. Doeller CF, Burgess N (2008) Distinct error-correcting and incidental learning of
location relative to landmarks and boundaries. Proc Natl Acad Sci USA 105(15):
5909–5914.
6of6
|
www.pnas.org/cgi/doi/10.1073/pnas.1424194112 Julian et al.
Supporting Information
Julian et al. 10.1073/pnas.1424194112
SI Discussion 1
Two mental processes are considered dissociable if they are
separately modifiable. Separate modifiability is established by
showing that there is an experimental manipulation or treatment
that affects the operation of one process but not the other. In the
current experiment, the treatment is the featural cue, which has
no effect on heading retrieval, but is essential for place recog-
nition. A possible counterargument to this dissociation logic is
that the feature cue might be alternatively conceptualized not as a
single cue, but as two separate cues. In Experiment 1, for example,
the orientation of the stripes is potentially informative about
place recognition, whereas the location of the stripes is potentially
informative about heading retrieval. Thus, the results in this case
could possibly be explained by the operation of a single mech-
anism mediating both place recognition and heading retrieval that
happens to be sensitive to the visual appearance of features (and
also chamber geometry) but insensitive to the spatial locations of
features. In this view, the feature is not a single treatment that
affects one process but not the other (thus demonstrating separate
modifiability), but two different treatments that have unequal ef-
fects on a single underlying process. This alternative account seems
less likely to explain the results of Experiment 2, because the
feature cue here provides information for both tasks in virtue of its
spatial location. However, even in this case, the alternative account
cannot be entirely dismissed, because it is possible that the mice
might interpret the stripes on the long wall as being perceptually
different from the stripes along the short wall. If so, then the results
could be explained by a single mechanism that is sensitive to the
visual appearance of features but insensitive to their spatial lo-
cations. Despite these caveats, we nevertheless believe that the
most parsimonious interpretation of our results is that there are two
separate systems used for reorientation when lost, one for place
recognition and one for heading retrieval. Note that even stronger
evidence for this claim would come from a double dissociation in
which one treatment affects place recognition but not heading
retrieval while a different treatment affects heading retrieval but
not place recognition, but we do not demonstrate a double dis-
sociation here.
SI Discussion 2
Consistent with the standard view (1), we posit that reorientation
involves the retrieval of spatial heading, in addition to identifi-
cation of the environment. An alternative theory proposes that
reorientation involves view matching rather than heading re-
trieval. In this account, the animal finds the goal location by
attempting to match the current visual input to a stored repre-
sentation of the visual input previously experienced at the goal
location. Computational view-matching models have been shown
to accurately describe reorientation behavior in single chamber
environments (2–4), indicating that standard reorientation be-
havior could be based on view matching. It is important to note,
however, that these studies do not provide evidence against the
idea that reorientation is based on spatial geometry—they only
show that, absent other data, both theories are equally likely.
More recent tests in children (5, 6) and chicks (7) that have pitted
view matching against reorientation by geometry have found re-
sults that are difficult to explain in terms of view matching. In
particular, these studies have found that navigators are able to
solve reorientation tasks that cannot, in principle, be solved by
view matching (6) and they fail to solve some reorientation tasks
that could be solved by view matching (5, 7). Some of these
findings are discussed in depth in a recent review from Cheng
et al. (8), in which the authors argue against view-matching as a
possible account of vertebrate reorientation behavior. Never-
theless, whether rodents reorient by spatial retrieval or view
matching remains an open question.
The present studies may speak to this ongoing debate in two
ways. First, in the present experiments, the same cues used for
place recognition were ignored for heading retrieval. Thus, at a
minimum, view-matching theories would need to be revised to
accommodate two sets of views, one that allows the animal to
determine the identity of the chamber, and another that guides
the animal to the correct location within the chamber. One
possibility is that these two view systems might store views with
different spatial frequency content, thus leading to differential
sensitivity to the featural cue. Second, the present experiments
may provide evidence against some view-matching theories in
which the stored views consist of edges (4). The nongeometric
features used in the present experiments, including in the stan-
dard one-chamber reorientation task, were horizontal or vertical
black stripes containing highly salient edge information. Given
that the mice ignored the visual stripes when determining the goal
location within each chamber, it seems unlikely that the animals
navigated within the chamber by matching edges in the current
view with those in a previously experienced view.
1. Cheng K (1986) A purely geometric module in the rat’s spatial representation. Cog-
nition 23(2):149–178.
2. Stürzl W, Cheung A, Cheng K, Zeil J (2008) The information content of panoramic
images I: The rotational errors and the similarity of views in rectangular experimental
arenas. J Exp Psychol Anim Behav Process 34(1):1–14.
3. Wystrach A, Cheng K, Sosa S, Beugnon G (2011) Geometry, features, and panoramic
views: Ants in rectangular arenas. J Exp Psychol Anim Behav Process 37(4):420–435.
4. Sheynikhovich D, Chavarriaga R, Strösslin T, Arleo A, Gerstner W (2009) Is there a
geometric module for spatial orientation? Insights from a rodent navigation model.
Psychol Rev 116(3):540–566.
5. Lee SA, Spelke ES (2011) Young children reorient by computing layout geometry, not
by matching images of the environment. Psychon Bull Rev 18(1):192–198.
6. Nardini M, Thomas RL, Knowland VC, Braddick OJ, Atkinson J (2009) A viewpoint-
independent process for spatial reorientation. Cognition 112(2):241–248.
7. Lee SA, Spelke ES, Vallortigara G (2012) Chicks, like children, spontaneously reorient by
three-dimensional environmental geometry, not by image matching. Biol Lett 8(4):
492–494.
8. Cheng K, Huttenlocher J, Newcombe NS (2013) 25 years of research on the use of
geometry in spatial reorientation: A current theoretical perspective. Psychon Bull Rev
20(6):1033–1054.
Julian et al. www.pnas.org/cgi/content/short/1424194112 1of2
30%
20%
25%
25%
(5.6)
(4.4)(6.0)
(4.9)
33%
22%
30%
15%
(5.8)
(5.4)(4.1)
(4.4)
Fig. S1. Results from the square chambers. Following the last day of testing in Experiment 2, disoriented animals were tested in two square chambers with a
vertical striped feature along one wall. The average percentage of first digs (and SEMs) at each corner in both size square chambers is shown. The star denotes
the location that was correct relative to the feature in the rectangular chambers in Experiment 2. The animals did not go to the “correct”cup, indicating that
they did not use a response-based strategy of choosing the cup on the appropriate side of the feature irrespective of context.
Movie S1. Movie depicting typical animal behavior in Experiment 1. The movie shows a single testing session (eight trials) from a single animal. Test chambers
(horizontal and vertical) were interleaved. For each chamber, C, R, N, and F denote the four cups in the corners of the chamber, where C denotes the correct
corner (i.e., the corner with the hidden reward), R the geometrically equivalent corner (i.e., the corner rotationally equivalent to C), N the near corner (i.e., the
corner that is closest to C), and F the far corner (i.e., the nonrotationally equivalent corner farthest from C). The horizontal chamber was rewarded at the long
wall left location nearest the feature, and the vertical chamber was rewarded at the long wall right location nearest the feature. When the animal digs,the
letter denoting the corner at which the animal dug turns red.
Movie S1
Julian et al. www.pnas.org/cgi/content/short/1424194112 2of2