Conserved Circuits for Direction Selectivity in the Primate Retina
1
Sara S. Patterson
1,2,*
, Briyana N. Bembry
2
, Marcus A. Mazzeferri
2
, Maureen Neitz
2
, Fred Rieke
3
,
2
Robijanto Soetedjo
3,4
, Jay Neitz
2,*
3
4
1
Center for Visual Science, University of Rochester, Rochester, NY, 14620
5
2
Department of Ophthalmology, University of Washington, Seattle, WA, 98109
6
3
Department of Physiology and Biophysics, University of Washington, Seattle, WA, 98195
7
4
Washington National Primate Research Center, University of Washington, Seattle, WA, 98195
8
* Correspondence: spatte16@ur.rochester.edu, jneitz@uw.edu
9
10
The detection of motion direction is a fundamental visual function and a classic model for
11
neural computation
1,2
. In the non-primate mammalian retina, direction selectivity arises in
12
starburst amacrine cell (SAC) dendrites, which provide selective inhibition to ON and ON-
13
OFF direction selective retinal ganglion cells (dsRGCs)
3,4
. While SACs are present in
14
primates
5
, their connectivity is unknown and the existence of primate dsRGCs remains an
15
open question. Here we present a connectomic reconstruction of the primate ON SAC
16
circuit from a serial electron microscopy volume of macaque central retina. We show that
17
the structural basis for the SAC’s ability to compute and confer directional selectivity on
18
post-synaptic RGCs
6
is conserved in primates and that SACs selectively target a single
19
ganglion cell type, a candidate homolog to the mammalian ON-sustained dsRGCs that
20
project to the accessory optic system and contribute to gaze-stabilizing reflexes
7,8
. These
21
results indicate that the capacity to compute motion direction is present in the retina, far
22
earlier in the primate visual system than classically thought, and they shed light on the
23
distinguishing features of primate motion processing by revealing the extent to which
24
ancestral motion circuits are conserved.
25
26
Neurons responding preferentially to motion in specific directions are found across species and
27
throughout the visual system
1,2
. The underlying mechanisms have been extensively studied in
28
ON and ON-OFF dsRGCs of the non-primate mammalian retina
9
. Each consists of multiple
29
subtypes preferring motion in different directions
10
. Their direction selectivity begins with SACs,
30
radially-symmetric interneurons present in every mammalian retina studied to date
5,11
. SAC
31
dendrites operate independently, computing outward motion from the soma
12
and providing
32
selective GABAergic inhibition to dsRGC subtypes preferring motion in the opposite direction
3,4
.
33
The intensive study of direction selective retinal circuits has yielded significant insight into
34
the general principles of neuronal computation
1
, yet the direct applications to primate vision are
35
unclear. Despite being a standard feature in the early visual systems of other species, direction
36
selectivity has yet to be demonstrated in the primate retina. Several lines of evidence indicate
37
some retinal capacity to compute motion direction may be conserved
13–15
, yet primate dsRGCs
38
remain elusive. As such, a classic interpretation is that the expanded primate cortex replaced the
39
need for retinal direction selectivity and the other highly-specialized computations found in non-
40
primate retinal ganglion cells (RGCs)
16,17
. Alternatively, the absence of primate dsRGCs from the
41
literature could reflect a sampling bias. The primate retina is dominated by three RGC types and
42
the rarity of the other ~15 anatomically-defined types severely limits the possibility of identifying
43
dsRGCs with the electrophysiology approaches used in other species
9,18
. Thus, the underlying
44
question is not only whether primate dsRGCs exist, but also how to find them.
45
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted July 22, 2021. ; https://doi.org/10.1101/2021.07.21.453225doi: bioRxiv preprint
An alternative strategy is to identify candidate dsRGCs from the neurons post-synaptic to
46
SACs. However, this approach is currently limited by a lack of information on primate SAC
47
circuitry. Here we use serial block-face scanning electron microscopy and connectomics
19
to fill
48
this gap in knowledge and determine the extent to which the mammalian retinal direction
49
selectivity circuitry is conserved in primates.
50
51
Connectomic Reconstruction of ON SACs
52
We reconstructed a population of 8 ON SACs from a 220 x 220 x 170 μm volume of macaque
53
central retina (~1.5 mm inferior to the fovea) sectioned horizontally at 90 nm and imaged at 7.5
54
nm/pixel (see Methods)
20
. The volume spanned from the photoreceptor outer segments to the
55
ganglion cell layer, enabling 3D reconstruction of complete retinal circuits while maintaining the
56
resolution necessary to identify synapses. Each reconstructed ON SAC exhibited the stereotyped
57
morphological features and stratification described across species (Fig. 1)
5
. We focused on the
58
ON SACs because primate OFF SACs are nearly absent from the fovea and outnumbered 10:1
59
by ON SACs in the periphery
5,21
. Consistent with previous reports of primate SACs, the dendritic
60
fields in Fig. 1 are sparse compared to non-primate SACs, both in branching density and coverage
61
factor
5
.
62
Direction selectivity in dsRGCs depends critically on SACs
11
. To determine whether
63
primate ON SACs are well-suited to play a similar role, we investigated the structural basis for
64
two features essential to this role: centrifugal motion tuning and asymmetric inhibition of post-
65
synaptic neurons
6
. SACs are ideal for connectomic analysis as decades of careful study in other
66
species has detailed the incredibly precise relationships between their anatomy, physiology and
67
function
6
. We began with the SACs’ selectivity for outward motion, which is supported by cell-
68
intrinsic mechanisms (morphology and radial synapse distribution
22,23
) and amplified by circuit-
69
level mechanisms (temporally-diverse excitatory bipolar cell input
24,25
and SAC-SAC lateral
70
inhibition)
26,27
. While the relative contributions of these and other mechanisms remains an active
71
area of investigation
2,6
, together they provide a solid blueprint to begin our connectomic
72
investigation. If primate SACs confer direction selectivity on downstream RGCs, we expect to find
73
evidence for these mechanisms.
74
We first asked whether the proximal-distal synaptic gradient supporting direction
75
selectivity in mammalian SAC dendrites is also present in primates. As expected, the SACs’
76
output synapses were confined to varicosities on the distal dendrites while bipolar cell input was
77
located closer to the soma, often at the small spines extending from the thin proximal dendrites
78
(Fig. 1d-f). This distribution of excitatory input and synaptic output, combined with the SAC’s
79
characteristic morphology, is critical for centrifugal motion tuning
22,23
.
80
SACs receive the majority of their inhibitory input from neighboring SACs
22
and the
81
resulting lateral inhibition reduces sensitivity to inward motion
26,27
. We were able to reliably
82
classified the pre-synaptic amacrine cells as SACs or non-SACs for 54 of 63 inhibitory synaptic
83
inputs to the SAC in Fig. 1e and confirm that 60.38% came from other ON SACs. Interestingly,
84
the exact location of inhibitory input along each SAC dendrite varies between species and is
85
hypothesized to scale with eye size to preserve angular velocity tuning
22
. SACs in species with
86
larger eyes have greater inter-soma distances and receive inhibition from other SACs more
87
distally. The macaque eye size of 200 μm/degree of visual angle
28
is consistent with the low SAC
88
coverage factor
5
and our observed bias of inhibitory synaptic input to the distal dendrites (Fig. 1a,
89
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted July 22, 2021. ; https://doi.org/10.1101/2021.07.21.453225doi: bioRxiv preprint
1e-f). These results demonstrate SAC-SAC lateral inhibition is present in primates and positioned
90
to compute a behaviorally-relevant measure of motion direction.
91
Lastly, we investigated the “space-time wiring” hypothesis
24
, which proposes that
92
sustained bipolar cell input is located closer to the soma than transient bipolar cell input. The
93
resulting temporally-diverse excitation sums for centrifugal, but not centripetal, motion
24,25
;
94
although this mechanism remains controversial
6,22,23
. We reconstructed the axon terminals of
95
presynaptic bipolar cells and classified each as either ON midget, DB4 diffuse, or DB5 diffuse
96
(Fig. 2a, 2d-f). The “space-time wiring” hypothesis predicts that midget bipolar cells will be closest
97
to the soma as their responses are more sustained than diffuse bipolar cells
29
. To test this
98
prediction, we calculated the radial distance from the SAC’s soma to each bipolar cell synaptic
99
input. Indeed, ON midget bipolar cells were significantly closer to the soma than the diffuse bipolar
100
cells (Fig. 2b-c). However, the underlying distributions show substantial overlap between the two
101
groups. Thus, our results mirror those in other species – while we find evidence for space-time
102
wiring, the overall efficacy could be reduced by the lack of spatial segregation between bipolar
103
cell types
22
.
104
Taken together, our investigation of the structure and synaptic input to ON SACs revealed
105
that multiple mechanisms contributing to centrifugal motion tuning in mammalian SACs are
106
conserved in primates. However, the SAC’s ability to confer direction selectivity depends not only
107
on their responses to outward motion, but also their selective wiring with specific dsRGC
108
subtypes. To address this, we next asked which RGCs were post-synaptic to the ON SACs and
109
whether they received asymmetric SAC inhibition.
110
111
SAC Synaptic Output to RGCs
112
SAC output synapses have a highly distinctive ultrastructure with large synaptic contacts where
113
the SAC’s processes completely engulf the postsynaptic dendrite
3,21,30,31
. We frequently observed
114
these stereotyped ”wrap-around” synapses at the varicosities of distal SAC dendrites (Fig. 3b).
115
We mapped each output synapse, then reconstructed and classified the postsynaptic neurons
116
(Fig. 3a). As expected, the SACs’ output targeted other SACs and non-SAC amacrine cells, but
117
rarely bipolar cells. Crucially, RGCs received the majority of the SACs’ output.
118
Amazingly, we found the ON SAC’s synaptic output targeted a single RGC type with
119
striking selectivity (Fig. 3f). The RGC’s morphology matches the recursive monostratified RGC
120
(rmRGC), a rarely-encountered neuron described only in the largest anatomical surveys of
121
primate RGCs
18,32
. Although the rmRGC’s physiology and connectivity are unknown, a role in
122
direction selectivity has been proposed before on the basis of their strong resemblance to the
123
highly stereotyped morphology of ON-sustained dsRGCs in other vertebrates
33–35
. The recursive,
124
looping branching pattern that ON dsRGCs share with rmRGCs is attributed to their co-
125
fasciculation with the SAC plexus
36,37
. We observed similar co-stratification and co-fasciculation
126
between the rmRGC and ON SACs (Fig. 3d-e).
127
Because each SAC dendrite is independently tuned to outward motion, the directionality
128
of SAC inhibition can be predicted from the angle of the vector between the soma and each output
129
synapse
3,31,38,39
. To estimate the rmRGC's direction selectivity, we calculated this for 135 of the
130
rmRGC’s conventional synaptic inputs where a presynaptic SAC could be sufficiently
131
reconstructed (Fig. 3g). Because the SAC’s output is inhibitory, the rmRGC is predicted to prefer
132
motion in the opposite direction of the strong bias shown in Fig. 3h. The distribution of dendritic
133
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted July 22, 2021. ; https://doi.org/10.1101/2021.07.21.453225doi: bioRxiv preprint
angles was highly non-uniform (p < 0.0001; Rayleigh test), indicating the asymmetric SAC
134
inhibition necessary for direction selectivity is present.
135
To confirm our findings, we next searched for additional rmRGCs. The dendritic fields of
136
ON dsRGCs tuned to the same direction tile while those tuned to different directions overlap
39,40
,
137
predicting additional rmRGCs should be present within the first rmRGC’s dendritic field. Indeed,
138
we found additional rmRGCs, both in the first volume taken from the inferior retina and in a
139
separate volume of central nasal retina, each exhibiting the same characteristic morphology,
140
stratification and stereotyped SAC input (Fig. 3i, S1). The dendrites of overlapping rmRGCs were
141
often directly adjacent, consistent with mammalian ON dsRGCs’ cofasciculation with each other
142
and the ON SAC plexus
39,40
(Fig. 3i-j). Moreover, puncta adherens were frequently observed
143
between adjacent rmRGC dendrites, a unique feature previously reported in rabbit ON dsRGCs
30
144
(Fig. S2).
145
While the other rmRGCs’ proximity to the volume’s edge prevented an unbiased dendritic
146
angle analysis, we did observe a striking trend: the SAC dendrites providing input to the first
147
rmRGC rarely, if ever, synapsed on the second rmRGC in Fig. 3j, despite frequently being in
148
close proximity. For example, the SAC in Fig. 3e synapsed on the first rmRGC 40 times, but only
149
three times on the second rmRGC (Fig. S3). These results are consistent with the hypothesis that
150
the overlapping rmRGCs prefer movement in different directions and supports asymmetric
151
inhibition from SACs as an underlying mechanism for their distinct directional tuning. Taken
152
together, the rmRGC’s structure and circuitry is consistent with ON dsRGCs in other species.
153
154
Retinal Input to the AOS
155
Like SACs, ON dsRGCs play a fundamental and highly conserved role in vision
41
. From mice and
156
rabbits to turtles and birds, ON dsRGCs share not only the characteristic morphology shown in
157
Fig. 3c, but also projections to the AOS, which coordinates with the vestibular system to stabilize
158
gaze with the optokinetic reflex
34,35,42,43
. As in other species, neurons in the primate AOS exhibit
159
similar response properties to ON dsRGCs
42,44
and receive direct retinal input
45
, yet the specific
160
RGC types are unknown. Previous retrograde tracer injections to the nucleus of the optic tract
161
and dorsal terminal nucleus (NOT-DTN) complex of the AOS revealed input from two RGC types,
162
one resembling the rmRGC
46
; however, incompletely-filled dendritic fields prevented
163
unambiguous classification. We repeated this experiment with the goal of targeting individual
164
RGCs for detailed morphological characterization and comparison to our reconstructions.
165
We injected the retrograde tracer rhodamine dextran into the NOT-DTN
47
, which was
166
located by identifying neurons with characteristic response properties, including direction
167
selectivity during horizontal smooth pursuit (Fig. 4a, S4a). Further confirmation was obtained with
168
post-mortem histology (Fig. S4b). Retrogradely-labeled RGCs were identified in an ex vivo
169
flatmount preparation by clumps of rhodamine fluorescence within their soma (Fig. 4b) and filled
170
with fluorescent dye to reveal their dendritic field structure (Fig. 4c-d, S5). All were rare wide-field
171
RGCs and a subset exhibited the characteristic curving dendrites that are hallmarks of both
172
rmRGCs and ON dsRGCs, confirming the NOT-DTN of the AOS receives direct rmRGC input.
173
174
Discussion
175
Here, we identified a likely homolog to the ON dsRGC, establishing an anatomically-conserved
176
circuit from SACs to the AOS for the gaze-stabilizing, compensatory eye movements essential for
177
visually-guided navigation
42
. Past research on the primate AOS has largely focused on the role
178
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted July 22, 2021. ; https://doi.org/10.1101/2021.07.21.453225doi: bioRxiv preprint
of cortical feedback; however, recent studies on human congenital nystagmus suggest a direct
179
retinal contribution involving GABAergic signaling by SACs
13,14,48
. Until now, the RGCs linking
180
SACs to the AOS were unknown.
181
We estimate that rmRGCs make up ~1% of all RGCs in the central retina. This rarity likely
182
reflects their large dendritic fields (Fig. S6) rather than their importance in vision. The low density
183
of rmRGCs undoubtedly creates a challenging sampling bias that could explain their absence in
184
surveys recording from single RGCs with microelectrodes or relatively small patches of peripheral
185
retina with multielectrode arrays. Although follow-up physiological studies guided by our results
186
will be important, waiting for these experiments to become feasible delays essential insights into
187
primate vision. Here we demonstrate that connectomics – a technique best known for large-scale
188
dense reconstructions
24,39,49
– can also be utilized for focused, hypothesis-driven questions about
189
otherwise intractable neural circuits
19
with implications for human health and disease
13,14,48
.
190
Interestingly, we did not observe SAC input to a ON-OFF dsRGC homolog, although we
191
cannot rule out the possibility that they are present in the peripheral retina where OFF SAC density
192
increases. However, the central retina mediates most conscious vision, indicating any ON-OFF
193
dsRGC homolog confined to the periphery would be best suited for non-image-forming vision and
194
independent of the cortical direction selectivity underlying motion perception
50
.
195
This work underscores the benefits of bridging primate and non-primate research
16
. The
196
SAC and rmRGC join a growing list of primate retinal neurons, like the intrinsically photosensitive
197
RGCs, that are understood only through prior research in non-primate species. The strong
198
correspondence between rmRGCs and ON dsRGCs raises the intriguing possibility that other
199
rare primate RGCs with unknown functions may have well-studied non-primate counterparts.
200
More generally, this challenges a widely-held view that the computational goals of the primate
201
retina are unique from other species and maximize information transmission to the cortex rather
202
than extraction of meaningful visual features, such as motion direction.
203
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted July 22, 2021. ; https://doi.org/10.1101/2021.07.21.453225doi: bioRxiv preprint
