Edinburgh Research Explorer
A network linking scene perception and spatial memory systems
in posterior cerebral cortex
Citation for published version:
Steel, A, Billings, M, Silson, EH & Robertson, CE 2021, 'A network linking scene perception and spatial
memory systems in posterior cerebral cortex', Nature Communications, vol. 12, 2632 .
https://doi.org/10.1038/s41467-021-22848-z
Digital Object Identifier (DOI):
10.1038/s41467-021-22848-z
Link:
Link to publication record in Edinburgh Research Explorer
Document Version:
Publisher's PDF, also known as Version of record
Published In:
Nature Communications
General rights
Copyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s)
and / or other copyright owners and it is a condition of accessing these publications that users recognise and
abide by the legal requirements associated with these rights.
Take down policy
The University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorer
content complies with UK legislation. If you believe that the public display of this file breaches copyright please
contact openaccess@ed.ac.uk providing details, and we will remove access to the work immediately and
investigate your claim.
Download date: 10. Aug. 2022
ARTICLE
A network linking scene perception and spatial
memory systems in posterior cerebral cortex
Adam Steel
1
✉
, Madeleine M. Billings
1
, Edward H. Silson
2
& Caroline E. Robertson
1
The neural systems supporting scene-perception and spatial-memory systems of the human
brain are well-described. But how do these neural systems interact? Here, using fine-grained
individual-subject fMRI, we report three cortical areas of the human brain, each lying
immediately anterior to a region of the scene perception network in posterior cerebral cortex,
that selectively activate when recalling familiar real-world locations. Despite their close
proximity to the scene-perception areas, network analyses show that these regions constitute
a distinct functional network that interfaces with spatial memory systems during naturalistic
scene understanding. These “place-memory areas” offer a new framework for understanding
how the brain implements memory-guided visual behaviors, including navigation.
https://doi.org/10.1038/s41467-021-22848-z
OPEN
1
Department of Psychology and Brain Sciences, Dartmouth College, Hanover, NH, USA.
2
Psychology, School of Philosophy, Psychology, and Language
Sciences, University of Edinburgh, Edinburgh EH8 9JZ, UK.
✉
email: adam.steel@dartmouth.edu
NATURE COMMUNICATIONS | (2021) 12:2632 | https://doi.org/10.1038/s41467-021-22848-z | www.nature.com/naturecommunications 1
1234567890():,;
A
s we navigate through our world, the visual scene in front
of us is seamlessly integrated with our memory of the
broader spatial environment. The neural systems sup-
porting visual scene processing in the posterior cerebral cortex
1–9
and spatial memory in the hippocampus and medial temporal
lobe
10–18
are well described. But how do visual and spatio-
mnemonic systems interface in the brain to give rise to memory-
guided visual experience?
Two disparate lines of inquiry yield different hypotheses. On the
one hand, mechanistic accounts of memory often posit that explicit
recall of visual stimuli reinstates perceptual representations in visual
areas
19–23
, including the three areas of the scene-perception network
(parahippocampal place area (PPA
2
), occipital place area (OPA; also
referred to as the transverse occipital sulcus
3,24–27
), and medial place
area (MPA; also referred to as retrosplenial complex
8,9,19,28–31
).
However, recent studies question the co-localization of perceptual
and memory-based representations and instead suggest a perception
to memory transition moving anteriorly from areas of the scene-
perception network
1,32–36
.Theselaterfindings are consistent with
neuropsychological observations, where patients with intact per-
ception but disrupted mental imagery, and vice versa, have been
described
37–39
. Resolving this discrepancy is critical to under-
standing how contextual information from memory is brought to
bear on visual representations in the brain. In short, do scene per-
ception and memory share common neural substrates?
Here, we sought to describe the neural basis of perceptually- and
mnemonically-driven representations of real-world scenes in the
humanbrain.Todothis,weconducted a series of experiments using
fine-grained, individual-subjectfMRItoassesswhetherthese
responses are co-localized in the brain (Experiment 1). Surprisingly,
this experiment revealed strong evidence to the contrary: in all
subjects, visually recalling real-world locations evoked activity in
three previously undescribed areas in the posterior cerebral cortex,
each lying immediately anterior to one of the three regions of the
scene-perception network. We next characterized the functional and
network properties of these “place-memo ry areas” (Experiments 2–3)
and explicitly tested the wide-spread view that the neural substrates
of perception and memory recall (i.e., mental imagery) activate
shared regions of the human brain (Experiment 4). Together, our
results reveal a network of brain areas that collectively bridge the
scene perception and spatial memory systems of the human brain
and may facilitate the integration of the local visual environment with
spatial memory representations.
Results
Distinct topography of scene-perception and place-memory
activity. In Experiment 1, we mapped the topography of perceptual
(henceforth “scene-perception”) and mnemonic (henceforth “place-
memory”) activity related to real-world scene processing in 14 adult
participants. We first independently localized the three regions of
the scene-perception network
1
by comparing activation when
participants viewed images of scenes, faces, and objects (Methods).
Then, in the same individuals, we localized areas that showed
preferential BOLD activation when participants recalled personally
familiar real-world locations (e.g. their house) versus faces (e.g. their
mother)
33
. We subsequently compared the topography (i.e. anato-
mical location and spatial extent) of the scene-perception and place-
memory preferring areas.
Comparison of scene-perception and place-memory activation
revealed a striking topographical pattern: in all participants, we
found three clusters of place-memory activity in the posterior
cerebral cortex, each paired with one of the three scene-
perception regions (Fig. 1). We henceforth refer to these clusters
as ‘place-memory areas’ for brevity. On the lateral, ventral, and
medial cortical surfaces, the pairs of place-memory and scene-
perception areas exhibited a systematic topographical relation-
ship: the center-of-mass of each place-memory area was located
consistently anterior to the center-of-mass of its corresponding
scene-perception area in every individual participant (Lateral
pairs: t(13) = 16.41, p < 0.0001, d = 4.39; Ventral pairs: t(13) =
PPA
OPA
MPA
Legend
Dor
Ant
Vent
Post
30
15
Post
Dor
Ant
Vent
30
15
Med
Ant
Lat
Post
30
15
Med
Ant
Lat
Post
30
15
Medial
64%
34%
2%
27%
48%
24%
Lateral
35%
16%
49%
Ventral
63%
11%
26%
Ventral
(faces)
Proportion of significant nodes
(Perception and Memory)
Overlap
Memory
Perception
Lateral
Medial
Ventral
Faces
Memory for places vs. people
t > 3.3, p < 0.001
< 20-20 > t-statistic
Perception
Center of mass
Distance (mm)
Memory
Center of mass
Fig. 1 Distinct topography of place-memory and scene-perception activity in posterior cerebral cortex. In all participants, three place-memory areas
were observed, each located significantly anterior to one region of the scene-perception network. One example participant in Experiment 1 is shown (See
Supplementary Fig. 1 and Supplementary Video 1 for thresholded and unthresholded activation maps for all participants (n = 14)). The participant’s scene
perception ROIs are outlined in white, and place-memory activity is shown in warm colors. The scene-perception network (parahippocampal place area
[PPA], occipital place area [OPA], and medial place area [MPA]) was localized by comparing the BOLD response when participants viewed images of
scenes versus with faces (outlined in white, thresholded at vertex-wise p < 0.001). Place-memory areas on each surface were localized in separate fMRI
runs by comparing the BOLD response when participants recalled personally familiar places versus people (warm colors, thresholded at vertex-wise p <
0.001). Polar plots: for each cortical surface, the center of mass of place-memory activation was significantly anterior to the center of mass of scene-
perception activation in all participants (all ts > 5, p < 0.001). In contrast, face memory activation was spatially co-localized with the face-selective fusiform
face area (FFA) on the ventral surface, and no anterior shift was observed (cool colors, t
9
= 0.1211, p = 0.906). Statistical analyses revealed no difference
between the hemispheres, so, for clarity, only right hemisphere is shown. Inset: while the activation during place memory was systematically anterior to
activation during scene perception, the spatial overlap between perception and memory activation varied across the cortical surfaces. Note that the axes
(posterior-anterior) of each polar plot are aligned to its associated cortical surface. The data in the polar plot reflects distance in millimeters.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22848-z
2 NATURE COMMUNICATIONS | (2021) 12:2632 | https://doi.org/10.1038/s41467-021-22848-z | www.nature.com/naturecommunications
12.115, p < 0.0001, d = 3.24; Medial pairs: t(13) = 5.99, p < 0.0001,
d = 1.6; Supplementary Fig. 1a; Supplementary Video 1). Impor-
tantly, control analyses using (1) a more conservative contrast
(scenes > faces + objects) to define scene-perception areas
40,41
and (2) a probabilistic ROI-based definition of PPA
41
confirmed
that the anterior shift of memory relative to perception was not
due to our definition of the scene-perception areas (Supplemen-
tary Figs. 2, 3).
Across the cortical surfaces, the degree of overlap between the
scene-perception and place-memory areas varied systematically
(Fig. 1, inset). On the ventral surface, place-memory activation
emerged rapidly at the anterior 60–70% of the scene-perceptio n area
(PPA) and continued anteriorly beyond the scene-perc eptio n area’s
anterior extent (Supplementary Fig. 4). In contrast, on the medial
surface, the scene-perception area (MPA) was encompassed within
the most posterior portion of the place-memory responsive region.
On the lateral surface, the place-memory and scene-perception area
(OPA) partially overlapped, but to a lesser extent than on the other
cortical surfaces. This systematic variation across the cortical
surfaces was present in all participants (Supplementary Figure 1a,
Supplementary Video 1).
We performed two additional analyses to characterize the
cortical locations of the place-memory areas. First, we asked: do
the locations of the place-memory areas fall in retinotopic, or
non-retinotopic cortex? We hypothesized that, unlike scene
perception areas, which process visual input, place-memory areas
would lie in non-retinotopic cortex. To test this, we performed a
group analysis of the place-memory localizer and compared the
peak activity for each area to retinotopic maps based on a
probabilistic atlas
42
. Strikingly, on all cortical surfaces, place-
memory areas were consistently located anterior to the cortical
retinotopic maps (Fig. 2a) with minimal overlap (Supplementary
Fig. 5). This location outside of retinotopic cortex is consistent
with their role as preferentially mnemonic, rather than visual,
areas. Second, we asked: how do the locations of the place-
memory areas relate to known anatomical landmarks in
the brain? To address this question, we compared the results of
the group place-memory localizer to anatomical parcels from the
Glasser atlas
43
. This revealed a remarkable pattern: the peaks of
the place-memory preferring activity fell at the intersection of
areas known to be important for visual and spatial processing.
Medially, the peak fell within the parietal occipital sulcus area 1
near the retrosplenial cortex; ventrally, the peak fell between
parahippocampal area 1 and the presubiculum, and laterally the
peak fell within area PGp and PGi (Fig. 2b). Thus, the place-
memory areas straddle visual and spatio-mnemonic functional
areas, and, based on this localization, these areas are poised to
play a role in bridging these processes. In contrast, the scene-
perception areas fell in retinotopic cortex and areas associated
with perception (Supplementary Fig. 6).
In contrast to the striking distinction between scene-perception
and place-memory, memory-related activation for faces was
centered on the perceptually-driven FFA on the ventral surface
(Difference of centers-of-mass: t(9) = 0.1211, p = 0.906, d =
0.17), and we did not observe consistent memory-driven
activation near the occipital face area on the lateral surface. The
anterior shift for places was greater than that for faces on all
surfaces (Main effect ROI: F(3,85) = 14.842, p < 0.001, t vs faces:
all ps<0.001, all ds > 2.26). Importantly, face and place memories
are complex and multifaceted, and factors such as vividness and
memory age can in fluence the activity elicited by memory
recall
44–46
. However, the difference between categories was not
due to subjective differences in recall ability, as participants
reported equal subjective vividness for place and face memory
recall, although vividness of recollection was not related to
activation magnitude in place-memory or scene-perception areas
(Supplementary Fig. 7). We reasoned that one potential
explanation for the anterior distinction of memory versus
perception for scenes/places but not faces might be the BOLD
signal-dropout artifact affecting lateral ventral temporal cortex.
However, we ruled out this explanation by replicating Experiment
1 using an advanced multi-echo fMRI sequence that improved
signal from the lateral and anterior temporal lobe and obtaining
similar results: an anterior shift for place-memory relative to
scene-perception, but no distinction between the location of face
memory relative to face-perception activity (Supplementary
Fig. 1b). All in all, the consistent topographical relationship
between perception and memory observed for scenes was not
observed for faces.
These findings show that place-memory recall engages three
distinct brain areas that are systematically paired with the scene-
perception areas. We next sought to characterize the functional
and network properties of these areas, specifically to test whether
scene-perception and place-memory areas were functionally, as
well as anatomically, dissociable. To ensure that we evaluated
functionally homogenous regions and to control for differences in
ROI sizes across regions, we constrained the scene-perception
and place-memory areas to the unique members of the top 300
most scene-perception/place-memory preferring vertices for all
subsequent experiments. Importantly, on all cortical surfaces, the
constrained place-memory area were anterior to the scene-
perception area (Supplementary Fig. 8).
Place-memory areas respond preferentially to familiar places.
In Experiment 1, we identified three place-memory areas that
showed category-selectivity for remembered places as compared
with perceived scenes. But are these areas only driven by top-
down, constructive memory recall tasks, or more broadly driven
by memory tasks, including tasks that rely less on top-down
signals, such as recognition memory? We examined this question
in Experiment 2. In this experiment, participants provided a list
of real-world locations that were familiar to them. Then, in the
scanner, participants performed a covert recognition task,
wherein they were shown panning videos of their personally
familiar locations or unfamiliar locations taken from another
participant’s familiar locations (created using Google StreetView;
Supplementary Video 2-5; see Materials and Methods). We then
compared the activity of the place-memory areas with the scene-
perception areas across the familiarity conditions. We hypothe-
sized that if the place-memory areas were preferentially driven by
mnemonic tasks as compared with scene areas, the place-memory
network would respond more strongly than the scene-perception
network when viewing videos of familiar versus unfamiliar
locations.
As predicted, we found that the place-memory areas were
preferentially driven by videos of familiar compared to unfamiliar
locations, and, critically, this difference was significantly greater
than we observed in the scene-perception areas (Region ×
Familiarity interaction – Lateral: F(1,91) = 20.98, p < 0.001,
t(13) = 6.40, p < 0.001, d = 2.26; Ventral: F(1,91) = 7.00, p =
0.01, t(13) = 4.443, p < 0.001, d = 1.55; Medial: F(1,91)
= 7.321,
p = 0.008, t(13) = 7.19, d = 1.05; Fig. 3a,b). The scene-perception
areas were also driven more by familiar compared to unfamiliar
videos (OPA: t(13) = 4.38, p = 0.0001, d = 1.171; PPA: t(13) =
4.46, p = 0.0001, d = 1.192; MPA: t(13) = 7.231, p < 0.0001, d =
1.932) but to a significantly lower extent than the place-memory
areas, as noted above. Importantly, the familiarity effect was not
observed in control regions, the amygdala (Supplementary Fig. 9)
or early visual cortex (Supplementary Fig. 10), arguing against a
purely attention-related account of this effect. However, con-
sistent with its role in recognition memory, the hippocampus
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22848-z ARTICLE
NATURE COMMUNICATIONS | (2021) 12:2632 | https://doi.org/10.1038/s41467-021-22848-z | www.nature.com/naturecommunications 3
Memory for places vs. people
t > 7.3, p = 1e-6, q = 1.5e-4
< 20-20 > t-statistic
Retintopic atlas (Wang et al. 2015)
Anatomical
Parcellation (Glasser et al. 2016)
a
b
IPS3
IPS2
IPS1
V3A
LO1
LO2
MST
hMT
IPS0
V3B
PHC2PHC1
VO2
VO1
MIP
IP1
PGs
PGi
TPO3
IPS1
V3B
V3C/D
LO1
LO3
IP0
PGp
POS1
POS2
RSC
pSub
dTVA
Prostriate
PH1
PH2
PH3
ERC
PRC
Fig. 2 Location of place-memory activity in relation to known functional and anatomical landmarks. a The place-memory areas are anterior to and have
minimal overlap with retinotopic cortex. We conducted a group analysis of the place-memory localizer (place > people memory recall; thresholded at
vertex-wise t > 7.3, p = 1e−6) and compared the resulting activation to the most probable location of the cortical retinotopic maps using the atlas
(overlaid) from Wang et al. (2015)
42
. The peak of place-memory activity was anterior toretinotopic maps on the lateral and ventral surfaces and there was
very little overlap between place-memory activity and retinotopic maps. b The place-memory areas fall at the intersection between anatomical parcels
known to be involved in visual processing and those associated with spatial memory. Comparing the peaks of place-memory activity with parcels from
Glasser et al.
43
(overlaid) revealed that the place-memory areas fell at the intersection of parcels associated with visual and spatial processing. Activation
maps are replotted in panels a and b to allow comparison between parcellations.
Viewing familiar versus unfamiliar places
t > 3.3, p < 0.001
< 20-20 >
Activation when viewing
familiar vs unfamiliar places (t-stat)
Medial
MPA MPMA
Ventral
PPA VPMA
-1
2
5
14
Lateral
OPA LPMA
8
11
t-statistic
LPMA
MPMA
VPMA
OPA
MPA
PPA
ab
*** * **
Fig. 3 The place-memory areas respond preferentially to familiar stimuli . a, b The place-memory areas preferentially activate familiar stimuli. a.
Experiment 3. Participants viewed viewing panning movies of personally familiar places versus unfamiliar places, tailored to each participant using Google
StreetView (see Supplementary Videos 2–5). The cortical surface depicts the contrast of BOLD activity for a single participant, thresholded at vertex-wise
p < 0.001. Only significant vertices within the scene perception (white) and place-memory (burgundy) areas are shown. b Average t-statistic of vertices in
the scene-perception (open bars) and place-memory areas (filled bars) when viewing videos of personally familiar places compared to unfamiliar places.
On each cortical surface, the place-memory areas showed an enhanced response to familiar stimuli compared to the scene-perception areas (all ts > 2.6,
ps < 0.01). Connected points depict individual participants. The hippocampus also showed a preferential response to familiar compared to unfamiliar place
movies (Supplementary Fig. 9a). The amygdala (Supplementary Fig. 9b) and early visual cortex (Supplementary Fig. 10) did not show a preferential
response to familiar place movies, arguing against a purely attentional account of this effect. OPA—occipital place area, LPMA—lateral place-memory area,
PPA—parahippocampal place area, VPMA—ventral place-memory area, MPA—medial place area, MPMA—medial place-memory area.
ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22848-z
4 NATURE COMMUNICATIONS | (2021) 12:2632 | https://doi.org/10.1038/s41467-021-22848-z | www.nature.com/naturecommunications
![Fig. 1 Distinct topography of place-memory and scene-perception activity in posterior cerebral cortex. In all participants, three place-memory areas were observed, each located significantly anterior to one region of the scene-perception network. One example participant in Experiment 1 is shown (See Supplementary Fig. 1 and Supplementary Video 1 for thresholded and unthresholded activation maps for all participants (n= 14)). The participant’s scene perception ROIs are outlined in white, and place-memory activity is shown in warm colors. The scene-perception network (parahippocampal place area [PPA], occipital place area [OPA], and medial place area [MPA]) was localized by comparing the BOLD response when participants viewed images of scenes versus with faces (outlined in white, thresholded at vertex-wise p < 0.001). Place-memory areas on each surface were localized in separate fMRI runs by comparing the BOLD response when participants recalled personally familiar places versus people (warm colors, thresholded at vertex-wise p < 0.001). Polar plots: for each cortical surface, the center of mass of place-memory activation was significantly anterior to the center of mass of scene-](/figures/fig-1-distinct-topography-of-place-memory-and-scene-2nfxljwn.webp)
