Article

Artificial Synaptic Rewiring Demonstrates that

Distinct Neural Circuit Configurations Underlie
Homologous Behaviors
Graphical Abstract Authors
Akira Sakurai, Paul S. Katz

Correspondence
akira@gsu.edu

In Brief
As with anatomical traits, behaviors can
be homologous. Sakurai and Katz
addresses differences in neural circuits
for homologous behaviors of two
species. Different network configurations
produce rhythmic activity in one species
as well as across species. Neural circuits
may have undergone hidden divergence
while retaining behavioral similarity.

Highlights
d Two nudibranchs have homologous behaviors produced by
distinct network mechanisms

d Homologous neurons have different synaptic connectivity

d Artificial synapses replaced those blocked by curare to

restore motor pattern

d Artificial rewiring using other species’ connections restored

an impaired circuit

Sakurai & Katz, 2017, Current Biology 27, 1–14

June 19, 2017 ª 2017 Elsevier Ltd.
http://dx.doi.org/10.1016/j.cub.2017.05.016
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016
Current Biology
Article
Artificial Synaptic Rewiring Demonstrates
that Distinct Neural Circuit Configurations
Underlie Homologous Behaviors
Akira Sakurai1,2,3,* and Paul S. Katz1
1Neuroscience Institute, Georgia State University, Atlanta, GA 30302-5030, USA
2Twitter: @akirasakurai
3Lead Contact
*Correspondence: akira@gsu.edu
http://dx.doi.org/10.1016/j.cub.2017.05.016
SUMMARY
Behavioral homology is often assumed to involve
similarity in underlying neuronal mechanisms. Here,
we provide a counterexample where homologous be-
haviors are produced by neurons with different syn-
aptic connectivity. The nudibranch molluscs Melibe
leonina and Dendronotus iris exhibit homologous
swimming behaviors, consisting of alternating left
and right body flexions. The swim central pattern gen-
erators (CPGs) in both species are composed of bilat-
erally symmetric interneurons, which are individually
identified and reciprocally inhibit their contralateral
counterparts, contributing to left-right burst alterna-
tion in the swim motor patterns. In Melibe, the swim
CPG contains two parts that interact to produce sta-
ble rhythmic bursting; one part is the primary half-
center kernel, and the other part, which consists of
a bilateral pair of neurons called Si3, regulates period
length. The Dendronotus swim CPG is simpler, with
Si3 being part of the primary half-center oscillator.
Application of curare (d-tubocurarine) selectively
blocked the Si3 synapses in both species. In Melibe,
curare application caused the burst duration of the
swim motor pattern to lengthen, whereas in Dendro-
notus, curare halted bursting altogether. In both spe-
cies, replacing the curare-blocked Si3 synapses with
artificial synapses using dynamic clamp restored the
original rhythmic bursting, thereby affirming the roles
of those synapses. The curare-impaired bursting in
Dendronotus was also restored by rewiring the ho-
mologous neurons into a Melibe-like primary half-
center oscillator configuration, indicating that the
connectivity itself could account for species differ-
ences in circuit responses to curare. The results sug-
gest that synaptic connectivity diverged during evo-
lution while behavior was conserved.
INTRODUCTION
Behaviors that are homologous and similar in form would natu-
rally be assumed to be produced by similar neural mechanisms.
This assumption has been challenged by several authors who
proposed that neural mechanism and behavior represent distinct
levels of biological hierarchical organization and could therefore
undergo separate evolutionary trajectories [1–4]. This implies
that homologous behaviors may be produced by divergent neu-
ral mechanisms. However, there are few, if any, examples in the
literature demonstrating such hidden divergence in neural mech-
anism underlying homologous behaviors in different species. In
this study, we demonstrate that motor patterns underlying ho-
mologous swimming behaviors are produced by distinct circuit
configurations of neurons in two sea slug species. Furthermore,
we show that the circuit in one species can be rewired into the
configuration of the other to produce a similar activity pattern.
Melibe leonina and Dendronotus iris (Mollusca, Gastropoda,
Nudibranchia) both swim by flattening their bodies in the sagittal
plane and repeatedly flexing from side to side for a period of about
2–3 s [5, 6]. This left-right (LR) swimming behavior is exhibited by
species in every genus in the clade that includes Melibe and Den-
dronotus (Figure 1A, bracket) [7, 8]. Furthermore, many species
belonging to the genera of Melibe and Dendronotus exhibit LR
swimming (Table S1). Although, some species of Melibe and Den-
dronotus have not been observed to swim, the lack of information
on these rarely sighted species does not necessarily indicate an
absence of the behavior. The predominance of LR swimming
within this clade indicates that the most recent common ancestor
also produced LR swimming and that LR swimming is therefore
homologous among the members of the clade.
In addition to the behaviors being homologous, individual
identified neurons also have been found to be homologous
across nudibranchs [6, 13, 14]. Neurons are individually identi-
fied in animals within a species based on a set of neuroanatom-
ical criteria, and those criteria are applied across species to
determine whether the neurons are homologous [15]. The central
pattern generator (CPG) circuits underlying the swimming be-
haviors in Melibe and Dendronotus contain homologous identi-
fied neurons [6, 11, 12]. However, the synaptic connectivity of
the swim CPGs differs.
In Melibe, the swim CPG consists of just four bilaterally sym-
metric swim interneurons (Si1–4), each of which reciprocally in-
hibits its contralateral counterpart (Figure 1B, Si4 is not shown
for simplicity), contributing to LR burst alternation (Figure 1C;
see Movie S1) [11]. The ipsilateral Si1, Si2, and contralateral
Si4 are electrically coupled and form the primary half-center
kernel, with the two halves firing bursts in alternation. Si3 forms
inhibitory synapses onto the contralateral Si1 and Si2 (Figure 1B)
Current Biology 27, 1–14, June 19, 2017 ª 2017 Elsevier Ltd. 1
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016
A Figure 1. Abbreviated Phylogenetic Tree of
B D
C E
and fires after the contralateral Si1/Si2 group (Figure 1C). Si3 is
activated by a slow excitatory synapse from the contralateral
Si2 [11]. Thus, feedforward excitation from Si2 to the contralat-
eral Si3 leads to the delay in the onset of Si3 firing. Si3 then feeds
back inhibition to Si1/Si2 and terminates the burst, allowing the
2 Current Biology 27, 1–14, June 19, 2017
Nudibranchia and the Swim CPGs in Melibe
and Dendronotus
(A) Melibe and Dendronotus belong to a clade
(indicated by bracket) that contains only genera
that swim with left-right (LR) body flexions. This
suggests that their most recent common ancestor
(black dot) shared this behavior, and thus the
behavior is homologous. Tree is based on [7–10].
NS indicates a non-swimmer.
(B) A schematic diagram of the Melibe swim CPG,
which consists of Swim interneurons (Si1, Si2, and
Si3) in left (L) and right (R) sides of the brain as
indicated by numbered circles. Si4 is omitted for
simplicity. Each neuron forms reciprocal inhibitory
synapses with its contralateral counterpart. Si2
makes a slow excitatory synapse onto the
contralateral Si3, which returns inhibitory synap-
ses onto the contralateral Si1 and Si2. Based on
Sakurai et al. [11].
(C) The swim motor pattern recorded intracellu-
larly from six swim interneurons. The gray shading
indicates the duration of a right Si3 burst. All
neurons burst in LR alternation. The Si3 bursts are
phase-delayed from the contralateral Si1 and Si2
bursts. The maximal membrane potentials during
the swim motor pattern for Si1, Si2, and Si3 were

58.3 ± 4.7 (n = 22),
59.5 ± 4.9 (n = 16), and

53.3 ± 4.4 (n = 20) mV, respectively (mean ± SD).
(D) The Dendronotus swim interneurons form a
circuit that is distinct from that of Melibe. The Si1
pair does not have reciprocally inhibitory synap-
ses but is electrically connected and with the
Si2 pair [6]. Si2 and Si3 form reciprocal inhibition
with their contralateral counterparts. Si3 makes
an excitatory synapse onto the contralateral Si2;
they are also electrically coupled. Based on Sa-
kurai et al. [6] and Sakurai and Katz [12].
(E) The Dendronotus Si1 fires irregularly, not in
bursts; only Si2 and Si3 exhibit alternating bursts.
The gray shading indicates the duration of a left
Si2 burst. The Si3 bursts lead Si2 bursts by a few
spikes. The gray shading indicates the duration of
a right Si3 burst. The maximal membrane poten-
tials during the swim motor pattern for Si1, Si2,
and Si3 were
39.8 ± 6.3 (n = 17),
50.4 ± 5.3
(n = 26), and
44.4 ± 3.7 (n = 10) mV, respectively
(mean ± SD).
In (B) and (D), lines terminating in filled circles
indicate inhibitory synapses. Triangles indicate
excitatory synapses; the long triangles indicate
slow excitatory synapses (Si2-to-Si3 in Melibe),
whereas the short triangles fast excitatory syn-
apses (Si3-to-Si2 in Dendronotus). Resistor sym-
bols indicate electrical connections. The thick-
ness of the resistor line indicates the strength of
connection [6, 11, 12]. Open circuits in (B) and (D)
and boxes in (C) and (E) indicate neurons located
in the left half of the brain; filled circuits and boxes
indicate neurons in the right half of the brain. See
also Table S1 and Movies S1 and S2.
other Si1 and Si2 to fire a burst [11]. Si4, which bursts in phase
with the contralateral Si1/Si2, contributes to terminating contra-
lateral Si3 bursts [11].
The Dendronotus swim CPG is simpler than that of Melibe; it is
composed of only two pairs of reciprocally inhibitory neurons,
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016
A left and right Si2 and Si3 (Figure 1D). In Dendronotus, Si1 does
B half-center kernel. This suggests that, although the neural mech-
C tials became undetectable (Figures 2Aiv and 2Biv). The only
Figure 2. Curare Blocks Si3 Synapses in Both Species
The schematics show the Si3 synaptic connections onto the neurons in (A) and
(B). See also Figure S1 and Data S1.
(A) Action potentials evoked in Si3 produced discrete, one-for-one IPSPs with
a constant latency in the contralateral Si1, Si2, and Si3 (i). The latencies to the
IPSPs in Si1, Si2, and Si3 were 15, 21, and 11 ms. The IPSPs decreased in
amplitude with increasing doses of curare (ii–iv). Simultaneous recordings from
four neurons were made in the presence of Hi-Di saline. A small amount of
constant current (<0.5 nA) was injected into L-Si3 via bridge-balanced
not have reciprocal inhibition across the midline. Instead, Si1 is
electrically coupled to its contralateral counterpart and bilaterally
with Si2 [6]. Si1 is not rhythmically active but fires irregularly
throughout the swim motor pattern (Figure 1E [6]; see Movie
S2). Unlike in Melibe, Si3 makes an excitatory synapse onto
the contralateral Si2 to which it is also electrically connected
[12]. As a result, Si3 fires nearly synchronously with the contralat-
eral Si2 instead of being delayed (Figure 1E). Unlike in Melibe, Si3
is responsible for initiating bursts in the contralateral Si2, not
terminating them. The electrical coupling between Si2 and
Si3 creates a positive feedback loop that maintains excitation
until the other Si2/Si3 pair either is released or escapes from
inhibition [12].
Here, we probe the circuit further and demonstrate that the
species difference in synaptic connectivity leads to distinct
neural mechanisms that produce homologous LR swimming be-
haviors. Furthermore, we demonstrate that Dendronotus neu-
rons can also produce rhythmic bursting activity when they are
artificially rewired to have the configuration of Melibe primary
anisms for swimming diverged over the course evolution, inter-
mediate forms of the circuit might still have been functional.
RESULTS
Si3 Synapses Are Curare Sensitive in Both Species
Curare selectively blocked all Si3 chemical synapses in both
species. In Melibe, spiking in Si3 produced one-for-one, con-
stant-latency inhibitory postsynaptic potentials (IPSPs) in the
contralateral Si1, Si2, and Si3 (Figure 2Ai). In Dendronotus,
spikes in Si3 produced one-for-one, constant-latency excitatory
postsynaptic potentials (EPSPs) in the contralateral Si2 and
IPSPs in the contralateral Si3 (Figure 2Bi). In both species,
bath application of curare reduced the amplitudes of the
postsynaptic potentials in a dose-dependent manner (Figures
2Aii–iv and 2Bii–iv). At 10
4 M curare, the unitary synaptic poten-
electrode to induce action potentials. Five to six traces were overlaid. In this
example, the Si3-evoked IPSPs appeared as monophasic hyperpolarizing
responses. However, in 56% preparations (n = 10 of 18), the Si3-evoked IPSPs
in Si2 were biphasic with initial depolarization and subsequent hyperpolar-
ization [11]. In such cases, 10
4 M curare blocked both synaptic components
in all preparations examined. Traces were triggered at the peak of action po-
tentials. The dotted lines indicate
54 mV (R-Si1),
50 mV (R-Si2), and

50 mV (R-Si3).
(B) In Dendronotus, the Si3 spikes produced one-to-one, constant latency
EPSPs in the contralateral Si2 and IPSPs in the contralateral Si3 (i). The latency
from the peak of Si3 spike to the IPSPs in Si3 was 19 ms and to the EPSP in Si2
11 ms. Both EPSPs and IPSPs decreased in amplitude with increasing doses
of curare (ii–iv). Dotted lines indicate
40 mV. L-Si3 was depolarized by in-
jecting 0.5–1 nA current through the recording electrode.
(C) The relationship between the dose (M) of curare and the strength of Si3
synapse. Normalized amplitude of the Si3-evoked IPSPs in the contralateral
Si1–3 in Melibe (filled circles; n = 8 synapses in four animals) or the Si3-evoked
EPSPs in the contralateral Si2 in Dendronotus (open squares; n = 6 synapses in
six animals) were plotted against the concentration of curare in Hi-Di saline and
fitted to a standard Hill model. The Hill slope and logEC50 for Melibe Si3-
evoked IPSP are
1.1 and
5.5, whereas those for Dendronotus Si3-evoked
EPSP are
1.0 and
6.3, respectively. Data are represented as mean ± SD.
Current Biology 27, 1–14, June 19, 2017 3
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

B C D

E F

Figure 3. In Melibe, Curare Slows the Swim Motor Pattern

(A) Simultaneous intracellular recordings from R-Si1, L-Si2, and R-Si3 show the motor pattern in normal saline (left). Bath application of 0.1 mM curare (gray bar,
right) slowed down the rhythm by extending the duration of bursts and causing them to become irregular. Schematics above the traces indicate the neural circuit.
The blocked Si3 synapses are shown in gray (right).
(B) Curare (0.1 mM) significantly decreased the burst frequency (inverse burst period). In normal saline, the burst frequency was 0.26 ± 0.09 (n = 16 animals) Hz,
which decreased to 0.06 ± 0.02 Hz (n = 16 animals) in curare. The burst frequency recovered to 0.19 ± 0.08 Hz (n = 8 animals) after more than 30 min washout of
curare. There was a statistically significant effect of curare on the burst frequency (F(2,22) = 47.9, p < 0.001 by one-way repeated-measures [RM] ANOVA).
(C) Curare (0.1 mM) significantly increased the coefficient of variation (CoV) in the burst period. CoV of the burst period during the swim motor pattern was
measured from the right Si1 or Si2 in each individual in the duration of at least 40 s. In normal saline (white), the CoV was 0.04 ± 0.02 (n = 16 animals), which
increased to 0.21 ± 0.12 (n = 16 animals) in curare (dark gray). The CoV decreased to 0.08 ± 0.03 (n = 8 animals) after washout of curare (light gray). There was a
statistically significant effect of curare on the CoV (F(2,22) = 20.8, p < 0.001 by one-way RM ANOVA).
(D) Curare had no effect on the intraburst spike frequencies of the swim interneurons. The bar graph shows averaged intraburst spike frequencies (Hz) measured
from Si1, Si2, and Si3, in the normal saline (white), in curare (dark gray), and after washout (light gray). Curare had no significant effect on intraburst firing
(legend continued on next page)
4 Current Biology 27, 1–14, June 19, 2017
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

exception is a small depolarizing potential in the Dendronotus
contralateral Si2 that is likely due the electrical coupling between
Si3 and Si2 [12]. Curare had no effect on any of the other CPG
synapses in either species (Figure S1).
The excitatory and inhibitory synaptic responses had different
sensitivities to curare. The Si3-to-Si2 EPSPs in Dendronotus
were greatly reduced in amplitude in response to 10
6 M curare
(Figure 2Bii). However, this concentration caused only a small
change in the Si3-to-Si3 IPSPs, which were recorded simulta-
neously in the same animal (Figure 2Bii). In Melibe, the Si3-
evoked IPSPs showed a similar insensitivity to 10
6 M curare
(Figure 2Aii). The EC50 for the IPSPs was about ten times higher
than for the EPSPs (Figure 2C).
Blocking Si3 Synapses in Melibe Slows the Swim Motor
Pattern
In Melibe, removal of Si3 synapses with curare significantly
extended the burst duration and decreased the cycle frequency
(Figures 3A and 3B). Curare also caused the burst period to fluc-
tuate; the coefficient of variation (CoV) of the burst period
increased significantly (Figure 3C). Nevertheless, the LR alterna-
tion of bursts persisted in the presence of curare (Figure 3A), and
the intraburst spiking of Si1, Si2, and Si3 was not affected
(Figure 3D).
Long burst period with extended burst duration, as
observed during curare application, could also be produced
by suppressing Si3 firing using bilateral hyperpolarizing current
injection (Figure 3E). In contrast, the Si3 pair alone cannot pro-
duce rhythmic bursts without the excitatory drive from the pri-
mary kernel; bilateral hyperpolarization of the Si1/Si2 pair
caused both Si3s to fire tonically with intermittent halts (Fig-
ure 3F). These results support our previous conclusion that
the Melibe swim CPG consists of two-interconnected parts
[11]. The primary half-center kernel of Si1, Si2, and Si4 is
capable of oscillating without the Si3 pair. In contrast, the
Si3 pair cannot operate as a half-center oscillator despite its
reciprocal inhibition. The results also support the hypothesis
that the Si3 synaptic inhibition regulates the burst duration
by acting as inhibitory feedback to terminate each Si1/2 burst
through strong inhibitory synapses without affecting their intra-
burst spiking [11].
Restoration of the Melibe Swim Motor Pattern Using
Artificial Si3 Synapses
To test the role of Si3 synapses in the generation of the Melibe
swim motor pattern, artificial synaptic connections were created
using the dynamic-clamp technique [16–18] to replace the Si3-
to-Si1 inhibition that was blocked by curare (Figures 4A and
S2A). Although curare blocks all synapses from Si3 and poten-
tially other curare-sensitive synapses from other neurons, this
experimental manipulation replaces just the specific synapses
from Si3 to Si1, allowing us to be confident of the role played
by this particular connection.
While the dynamic clamp was engaged, the bursts in Si1
became shorter again and the rhythm accelerated (Figure 4A).
Similar results were obtained when Si3-to-Si2 synapses were ar-
tificially replaced (n = 3, data not shown). Increasing the conduc-
tance of the artificial synapses shortened burst period (increased
the burst frequency, Figure 4B). The maximal recovery of the
burst frequency varied among preparations ranging from 41%
to 77% of the original burst frequency. In the presence of curare,
restoration of only the Si3-Si3 connection had no noticeable ef-
fect on bursting activity (n = 5; Figure S3A). Restoration of the Si3
synapses onto either Si1 or Si2 also regularized the burst period,
causing the CoV of the burst period to drop (Figure 4C). How-
ever, dynamic clamping caused no significant change in the in-
traburst spike frequency of Si1 or Si2 (Figure 4D). These results
indicate that the inhibitory synapses from Si3 to Si1/2 play a
crucial role in sustaining the burst frequency and the stability of
the motor pattern but have no role in providing the excitatory
drive onto Si1/2.
Blocking Si3 Synapses in Dendronotus Halts the Swim
Motor Pattern
Curare had a different effect on the swim motor pattern in Den-
dronotus compared to Melibe; instead of extending the bursts,
it stopped them altogether (Figures 5A and 5B). Unlike in Melibe,
spike frequency declined significantly in both Si2 and Si3 in the
presence of curare (Figure 5C). The curare-induced blockade
of the swim motor pattern could be replicated by suppressing
spiking in both Si3s with hyperpolarizing current injection (Fig-
ure 5D). This demonstrates that the CPG requires the excitatory
synaptic drive from Si3 to Si2 to produce its rhythmic output and
maintain a high level of spiking. Similarly, the Si3 pair alone could
not produce rhythmic bursting when spiking in both Si2s was
suppressed (Figure 5E). The results are consistent with our pre-
vious findings that the swim motor pattern was blocked when the
Si3-to-Si2 EPSPs were canceled out using inverse dynamic-
clamp synapses [12].
Restoration of the Dendronotus Swim Motor Pattern
Using Artificial Si3 Synapses
In Dendronotus, the swim motor pattern was restored in the
presence of curare by replacing the blocked Si3-to-Si2 synapses
with artificial synapses generated by the dynamic clamp (Fig-
ure 6A). Immediately upon introduction of the artificial excitatory
synapses, all four neurons exhibited rhythmic bursting until dy-
namic clamping was turned off. In eight of 12 preparations, the
dynamic-clamp replacement of blocked Si3-to-Si2 synapses
successfully induced bursting activity. In four preparations,

(Si1, F(2,14) = 1.44, p = 0.27 by one-way RM ANOVA, n = 14 cells; Si2, F(2,14) = 0.92, p = 0.42 by one-way RM ANOVA, n = 12 cells; Si3, F(2,16) = 2.27, p = 0.14 by
one-way ANOVA, n = 13 cells).
In (B)–(D), data are represented as mean ± SD. Bracket with asterisks indicating significant differences by all pairwise multiple comparison procedures
(Holm-Sidak method, two asterisks for p < 0.001 and one asterisk for p = 0.011). Individual data points are shown on each bar as open circles.
(E) Suppression of firing in Si3s by hyperpolarizing current injection (
4 nA) had similar effect as curare application by extending the burst duration of Si1 and Si2.
Schematic above the traces indicate both Si3s were simultaneously hyperpolarized by injecting a negative current step.
(F) The Si3 pair do not act as a half-center oscillator when not receiving periodic synaptic input from Si2. Suppression of firing in both Si1 and Si2 by hyper-
polarizing current injection (
4 nA) made Si3 fire tonically with no LR alternation.
In (A), (E), and (F), dotted lines indicate 0 mV. See also Data S2.

Current Biology 27, 1–14, June 19, 2017 5

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

B C D

Figure 4. Dynamic-Clamp Replacement of Curare-Blocked Synapses Restores Periodicity in Melibe

(A) Simultaneous intracellular microelectrode recordings of the left and right Si1s and Si3s. In normal saline (left), the motor pattern progressed for a period of less
than 5 s. In the presence of curare (0.1 mM), the period increased and became irregular (right, gray bar). The burst period shortened and became regular when the
Si3-to-Si1 synapses that were blocked by curare were replaced with artificial synapses (300 nS) generated by the dynamic clamp (indicated in red). The Isyn traces
indicate the artificial synaptic current injected when the dynamic clamp was on (red bar). When the clamp was turned off, the burst period lengthened and became
irregular again. Dotted lines indicate 0 mV.
(B) The burst frequency (Hz) increased with increased strength of the Si3 synapse. The increase in the conductance of Si3-to-Si1/2 synapse in dynamic clamping
caused significant increase in the burst frequency (F(5,18) = 8.3, p < 0.001 by one-way RM ANOVA, n = 6 animals). All pairwise multiple comparisons (Holm-Sidak
method) revealed significant differences between 300 and 0 nS (p < 0.001), 200 and 0 nS (p = 0.003), and 400 and 0 nS (p = 0.018).
(C) The CoV of the burst periods decreased significantly when Si3-to-Si1 synapse was introduced by the dynamic clamp (F(5,18) = 4.3, p = 0.009 by one-way RM
ANOVA, n = 6 animals). All pairwise multiple comparisons (Holm-Sidak method) revealed significant differences between 200 and 0 nS (p = 0.025), 300 and 0 nS
(p = 0.034), and 100 and 0 nS (p = 0.033).
(D) The intraburst spike frequency of Si1 or Si2 showed no change after replacement with artificial dynamic-clamp synapses (F(5,18) = 0.35, p = 0.88 by one-way
RM ANOVA).
In (B)–(D), experiments were obtained by using either Si1 (n = 3 preparations) or Si2 (n = 3 preparations). In each graph, the white bar shows the control value in
normal saline (N.S.) before curare, a gray bar shows the value in 0.1 mM curare, and red bars show the values under the dynamic clamp with different
conductance (from 50 to 400 nS) of the artificial synapses. Symbols represent individual preparations. Due to technical difficulty, bilateral dynamic clamping as
shown in (A) was performed in three preparations (two with the Si3-to-Si1 synapses and one with the Si3-to-Si2 synapses). In other preparations, only one Si3-to-
Si1 synapse (n = 1 preparation) or one Si3-to-Si2 synapse (n = 2 preparations) was replaced by the dynamic clamp. Data are represented as mean ± SD. See also
Figures S2A and S3A and Data S3.

increasing the conductance up to 120 nS failed to induce the
swim-like bursting activity. In preparations that exhibited rhyth-
mic bursting, both the burst frequency and the intraburst spike
frequency increased with increased artificial synaptic conduc-
tance (Figures 6B and 6C). During dynamic clamping, more
spikes were induced in Si2 because Si3 spikes more reliably
drove Si2 (Figure S2B). In contrast, replacement of the mutually
inhibitory connection between the Si3 pair alone did not induce
rhythmic bursting (n = 3; Figure S3B). These results support
the conclusion of our previous study that the excitatory drive
from Si3 is necessary for this CPG to generate alternating
bursting pattern [12].

6 Current Biology 27, 1–14, June 19, 2017

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

A Dendronotus
2 2 2 2

3 3 3 3

Normal saline Curare

L-Si2

R-Si2

L-Si3

R-Si3 50
mV
10 s
B C
∗∗ ∗∗ ∗∗
Intraburst spikes (Hz)

16
0.3 ∗∗
Burst freq. (Hz)

14
12 Normal saline
∗∗ ∗
0.2 10 Curare
8 Wash
6
0.1
4
2
0.0 0
Si2 Si3
C rm.
W re
h
as
a
o
ur
N

D E
2 2 2 2

3 3 3 3

-5 nA
L-Si2 L-Si2

-5 nA
R-Si2 R-Si2

-2 nA
L-Si3 L-Si3

-2 nA
R-Si3 50 R-Si3 50
mV mV
10 s 10 s

Figure 5. In Dendronotus, Curare Eliminates Bursting Activity in Both Si2 and Si3
(A) Simultaneous intracellular microelectrode recordings from left and right Si2s and Si3s show a typical bursting pattern in normal saline (left). In the presence of
0.1 mM curare (gray bar, right), both Si2 and Si3 ceased bursting and showed irregular spiking.
(B) Curare (0.1 mM) halted bursting activity in most of the preparations (n = 30 of 35). In the normal saline, the burst frequency was 0.21 ± 0.05 Hz (n = 35 animals),
which decreased to 0.02 ± 0.05 Hz in curare (n = 35 animals). The burst frequency recovered to 0.25 ± 0.06 Hz after washout of curare (n = 8 animals). There was a
statistically significant effect of curare on the burst frequency (F(2,41) = 122.4, p < 0.001 by one-way RM ANOVA). The burst frequency (Hz) was measured from the
right Si2.
(C) Curare significantly reduced the intraburst spike frequencies of the swim interneurons. The bar graph shows averaged intraburst spike frequencies (Hz)
measured from Si2, and Si3, in the normal saline (white), in curare (dark gray), and after washout (light gray). In both neurons, the spike frequency dropped
significantly when curare was applied (Si2, F(2,41) = 62.9, p < 0.001 by one-way RM ANOVA, n = 35 cells; Si3, F(2,21) = 16.3, p < 0.001 by one-way RM ANOVA,
n = 27 cells). When bursting in LR alternation was not detected in the presence of curare, then average spike frequency over a time period of >30 s was used as the
intraburst spike frequency.
(legend continued on next page)

Current Biology 27, 1–14, June 19, 2017 7

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

B C

Figure 6. In Dendronotus, Bursting Activity Is Recovered by Replacing the Blocked Synapses with Artificial Dynamic-Clamp Synapses
(A) In normal saline (left), both Si2s and Si3s exhibited normal bursting activity. In the presence of 0.1 mM curare (right, gray bar), Si3 synapses were blocked and
regular bursting ceased. When just the Si3-to-Si2 synapses were replaced by artificial dynamic synapses (80 nS, red bar), all four neurons immediately resumed
alternating bursts. When the dynamic clamp was turned off, the bursting immediately ceased again. Isyn shows the current injected by the dynamic clamp. Dotted
lines indicate 0 mV.
(B) In preparations that exhibited rhythmic bursting (n = 8 of 12 preparations), the burst frequency increased with increased artificial synaptic conductance from
Si3 to Si2 (F(5,21) = 7.95, p < 0.001 by one-way RM ANOVA, n = 8 animals). All pairwise multiple comparison (Holm-Sidak method) revealed significant differences
between 120 and 0 nS (p < 0.001), 80 and 0 nS (p = 0.01), 120 and 40 nS (p = 0.02), and 160 and 0 nS (p = 0.02).
(C) The intraburst spike frequency of Si2 significantly increased with increasing artificial synaptic conductance (F(5,21) = 6.4, p < 0.001 by one-way RM ANOVA,
n = 8 animals). All pairwise multiple comparison (Holm-Sidak method) showed that adding a synaptic conductance from 80 to 120 nS caused a significant in-
creases in the spike frequency (p = 0.008). When bursting in LR alternation was not detected in the presence of curare, then average spike frequency of the time
period of >30 s was used as the intraburst spike frequency.
In (B) and (C), a white bar shows the control value in normal saline (N.S.) before curare, a gray bar shows the value in 0.1 mM curare, and red bars show the values
under the dynamic clamp with different conductance (from 40 to 200 nS) of the artificial Si3-to-Si2 synapses. Symbols represent individual preparations that
exhibited rhythmic bursting. The graphs do not contain data from four preparations, in which dynamic clamp failed to induce rhythmic bursting. Data are rep-
resented as mean ± SD. See also Figures S2B and S3B and Data S5.

Rewiring the Dendronotus Circuit to a Melibe-like a circuit configuration like the primary half-center kernel of
Configuration Induces Bursting Melibe would also allow the Dendronotus neurons to burst in
In view of the fact that Si1 and Si2 in Melibe continued to burst the presence of curare. We added artificial electrical connections
when Si3 synapses were blocked by curare, we tested whether between the ipsilateral Si1 and Si2 and inhibitory synapses

In (B) and (C), brackets with an asterisk indicate significant difference by all pairwise multiple comparison procedures (Holm-Sidak method, two asterisks for
p < 0.001 and one asterisk for p = 0.02). Data are represented as mean ± SD. Individual data points are shown on each bar as open circles.
(D) Injection of hyperpolarizing current into both Si3s (at down arrows) halted bursting in the Si2s, which went silent or spiked sporadically.
(E) Injection of hyperpolarizing current into both Si2s (down arrows) caused the both Si3s to fire tonically. Large Si3-evoked EPSPs can be seen in the two Si2s. In
(A)–(E), dotted lines indicate 0 mV.
See also Data S4.

8 Current Biology 27, 1–14, June 19, 2017

Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

B C

D E

F G

(legend on next page)

Current Biology 27, 1–14, June 19, 2017 9

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

between the contralateral neurons. This network connectivity
caused Si1 to fire together with ipsilateral Si2 spikes, while the
contralateral Si1 was mostly inhibited by the artificial IPSPs (Fig-
ure S4). Under such conditions, the network produced bursts in
LR alternation in all preparations examined (n = 10; Figure 7A).
During dynamic clamp, there were significant increases in the
burst frequency (Figure 7B) and the intraburst spike frequency
(Figure 7C). Increasing the conductance of electrical connection
above 20 nS did not further increase the burst frequency
(Figure 7B), but it did increase the intraburst spike frequency
(Figure 7C). Addition of only the inhibitory synapses induced little
regular bursting activity (n = 8, Figure 7E). In contrast, adding
only the electrical connection induced bursting in eight of 13
preparations, but a higher conductance was needed (Figure 7E).
In preparations that exhibited rhythmic bursting, increasing the
conductance of the electrical connection increased the burst fre-
quency and the intraburst spike frequency (Figures 7F and 7G).
The apparent gradual increase in the burst frequency was likely
because the majority of preparations failed to show bursting at
lower conductances (20 and 40 nS).
Thus, without the Si3 pair, the Dendronotus Si2 pair can still
generate rhythmic bursting when connected to the Si1 pair like
the primary half-center kernel of the Melibe swim CPG. The elec-
trical connection between the ipsilateral Si1 and Si2 appeared to
have a more crucial role in restoration of the bursting activity.
This bursting activity included Si1, which never bursts during
normal Dendronotus swim motor programs [6].
DISCUSSION
Species Differences in Circuit Connectivity
This and previous studies showed that the connectivity of the
neural circuits of Melibe and Dendronotus differ substantially
from each other despite the presence of homologous neurons
and homologous behaviors [6, 11, 12]. We previously posited
that Si1 and Si2 are homologous in the two species because of
their neuroanatomical and neurochemical features [6]. Si3 is
also likely to be homologous in both species because it has a
similar neuroanatomy and all of its synaptic actions are blocked
by curare, suggesting that it may be cholinergic in both species
(see below). In contrast, curare did not block the synapses of the
other CPG neurons.
Just as with Si1, there are substantial species differences in
the synaptic connections made by Si3. In particular, in Dendro-
notus, Si3 evokes a very large EPSP in the contralateral Si2
that is missing or extremely reduced in Melibe. The Melibe
swim CPG, on the other hand, contains more contralateral inhi-
bition than does the Dendronotus swim CPG. These results sug-
gest that synaptic connections might be more phylogenetically
labile than the neurotransmitter phenotype of a neuron.
Species differences in specific synaptic connectivity among
homologous identified neurons have been reported across inver-
tebrate phyla. In flies, there are phylogenetic differences in con-
nectivity and structure of the photoreceptor synapses in retina
[19]. In the stomatogastric nervous system of decapod crusta-
ceans, there are differences between crab and spiny lobster in
electrical connections among homologous motor neurons; how-
ever, the pyloric motor pattern is highly conserved [20, 21]. In
leeches, differences in the polarity of synaptic connections among
the mechanosensory P cells underlie the difference in behavioral
expressions in two species [22]. Differences in synaptic connec-
tivity among homologous neurons have also been reported in
two species of nematode with different feeding behaviors [23].
In vertebrates, there are just a few examples of microcircuitry
among specific cell types reported to exhibit species differ-
ences. One example is that the organization of inhibitory inputs

Figure 7. Bursting Activity in Dendronotus Is Recovered through Cross-Species Rewiring

(A) In normal saline (left), the two SI1s fire irregularly while the two Si2s burst in alternation. In the presence of 0.1 mM curare (right, gray bar), all bursting ceased.
The Melibe circuit configuration was mimicked by addition of electrical coupling (80 nS) between Si1 and Si2 ipsilaterally and creation of inhibitory synapses
(100 nS) between Si1s and from Si2 to Si1 as shown by red synapses in circuit diagram. When the dynamic clamp was turned on (red bar), both Si1s and Si2s
immediately burst in LR alternation. They ceased bursting immediately after turning off of the dynamic clamp. Isyn shows the current injected by the dynamic
clamp. Dotted lines indicate 0 mV.
(B) The addition of electrical coupling together with the inhibitory synapses caused a significant increase in the Si2 burst frequency (F(5,23) = 12.73, p < 0.001 by
one-way RM ANOVA, n = 12 animals). In all 12 preparations rhythmic bursting was observed during dynamic clamping. All pairwise multiple comparison (Holm-
Sidak method) revealed significant differences between 40 and 0 nS (p < 0.001), 60 and 0 nS (p < 0.001), 80 and 0 nS (p < 0.001), 100 and 0 nS (p = 0.002), and 20
and 0 nS (p = 0.003). While changing the conductance of electrical coupling, the strength of the IPSPs was fixed at 50 or 100 nS.
(C) With increased electrical coupling, the intraburst spike frequency of Si2 increased significantly (F(5,23) = 24.02, p < 0.001 by one-way RM ANOVA, n = 12
animals). All pairwise multiple comparison (Holm-Sidak method) revealed significant differences between 40 and 0 nS (p < 0.001), 60 and 0 nS (p < 0.001), 80 and
0 nS (p < 0.001), 100 and 0 nS (0 < 0.001), 100 and 20 nS (p = 0.002), 80 and 20 nS (p = 0.002), and 60 and 20 nS (p = 0.016).
(D) Creating redundant reciprocal inhibition was not sufficient to cause stable rhythmic bursting. Adding only inhibitory synapses (100 nS, purple bar) between
Si1s and from Si2 to Si1 failed to produce bursting in any of the preparations examined (n = 5).
(E) Adding only the electrical connection between ipsilateral Si1 and Si2 induced rhythmic bursting in Dendronotus. In the presence of curare (0.1 mM), electrical
coupling (100 nS) was added between Si1 and Si2 ipsilaterally by the dynamic clamp (orange bar). Although the bursting is not as clear as that shown in Figure 6,
both Si1 and Si2 exhibited bursting in LR alternation in nine of 12 preparations. Dotted lines indicate 0 mV.
(F) The addition of electrical connections alone significantly increased the burst frequency (F(5,20) = 11.2, p < 0.001 by one-way RM ANOVA, n = 9 animals). All
pairwise multiple comparison (Holm-Sidak method) revealed significant differences as follows: 80 versus 0 nS, p < 0.001; 100 versus 0 nS, p < 0.001; 80 versus
20 nS, p = 0.025; 100 versus 20 nS, p = 0.025; 80 versus 40 nS, p = 0.032; 100 versus 40 nS, p = 0.033; 60 versus 0 nS, p < 0.032. Rhythmic bursting was observed
in nine of 12 preparations by adding artificial electrical connection of up to 100 nS. The graphs do not contain data from three preparations, in which the dynamic
clamp failed to induce rhythmic bursting.
(G) Artificial electrical connections increased the intraburst spike frequency in Si2 (F(5,20) = 5.22, p = 0.003 by one-way RM ANOVA, n = 9 animals). All pairwise
multiple comparison (Holm-Sidak method) revealed significant differences in 80 versus 0 nS (p = 0.036) and 60 versus 0 nS (p = 0.049). The graphs do not contain
data from three preparations, in which the dynamic clamp failed to induce rhythmic bursting. In each graph, data are represented as mean ± SD. Symbols
represent individual preparations.
See also Figure S4 and Data S6.

10 Current Biology 27, 1–14, June 19, 2017

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016
to Mauthner neurons differs across fish species [24]. The micro-
circuitry of starburst amacrine cells differs between rabbits and
mice to compensate for differences in eye size [25]. The CA3
pyramidal neurons of macaques and rodents also show some
differences in projection topography within the hippocampus
[26–28]. Thus, the evolution of microcircuitry could play a role
in evolution of behavior.
Excitatory and Inhibitory Putative Cholinergic Synapses
Si3 evokes either fast EPSPs or fast IPSPs, which have different
reversal potentials but are both blocked by curare. The EC50 for
curare-sensitive EPSPs and IPSPs differed by a factor of ten
regardless of whether comparing within a species (Dendronotus
Si3-to-Si2 versus Si3-to-Si3) or between species (Dendronotus
Si3-to-Si2 versus Melibe Si3-to-Si1/2), suggesting that they are
mediated by different receptors. It was previously shown that
curare blocks both EPSPs and IPSPS evoked by cholinergic
neurons in Lymnaea [29]. Curare also blocks both excitatory
and inhibitory responses of Aplysia neurons to acetylcholine
[30]. Furthermore, molecular expression experiments demon-
strate that molluscan nicotinic acetylcholine receptors (nAChRs)
include subunits that are anion selective as well as others that
are cation selective [31, 32]. Therefore, it is possible that the
Si3-evoked synaptic potentials in Si2 in Melibe and Dendronotus
are both mediated by nAChRs. However, because curare is not a
specific antagonist for the nAChRs, but rather a blocker for a
class of receptor-activated Na+ and Cl– responses [33], we
cannot be certain that Si3 releases the same neurotransmitter
in both species. In any case, the difference in the sign of the
Si3 synapses is one of the fundamental contrasts in the architec-
ture of the two CPGs. If Si3 is cholinergic in both species, then it
suggests a genetic mechanism underlying the species differ-
ences in connectivity. We predict that the Si2 homologs differ
in their expression of nAChR subunits; in Dendronotus, Si2
should express cation-selective subunits, whereas Si1 and Si2
in Melibe should express anion-selective subunits. Furthermore,
Si3 in both species should express anion-selective subunits.
Different Neural Mechanisms for a Similar Motor
Behavior
The swimming behaviors of Melibe and Dendronotus are similar
in form and are homologous based on phylogeny. Yet, the neural
mechanisms that produce these behaviors are distinct. In Me-
libe, there are two parts of the CPG, the primary half-center
kernel, which is made up of Si1, Si2, and Si4 pairs, and the Si3
pair. The two parts form complex synaptic interactions with
each other: the excitatory synapses from Si2 onto the contralat-
eral Si3, two types of biphasic synapses from Si1 onto left and
right Si3, a slow inhibitory synapse from Si4 onto the contralat-
eral Si3, and inhibitory synapses from Si3 onto the contralateral
Si1 and Si2 [11]. With the Si3 synapses blocked by curare, the
primary kernel can still generate alternating bursts alone, which
also keep driving the Si3 bursts, but the rhythm is slower and
more irregular. The Si3 pair alone cannot produce rhythmic
bursting by itself; it requires the rhythmic excitatory drive from
Si2 [11]. Replacement of the Si3-to-Si1/Si2 synapse with artificial
synapses restored the swim motor pattern by increasing the fre-
quency and the stability, suggesting that Si3 regulates the peri-
odicity of the motor pattern.
In Dendronotus, the swim CPG is a simple half-center oscil-
lator composed of Si2 and Si3. The excitatory synapse from
Si3 to Si2 is necessary for bursting [12]; when it is blocked by
curare, bursting ceases altogether. Again, the swim motor
pattern was recovered by the replacement of Si3-to-Si2 synap-
ses using the dynamic clamp. Previously, we performed a similar
experiment in a non-swimming preparation, in which we found
that boosting Si3-evoked EPSPs in Si2-induced rhythmic
bursting of Si2. Under such conditions, Si3 is pulled into oscilla-
tion through the electrical connection [12]. Thus, in Dendronotus,
the synaptic drive from Si3 to Si2 plays an essential role in burst
generation, whereas, in Melibe, Si3 plays a role in regulating
burst frequency by providing inhibitory feedback to the primary
kernel, not in generating bursts.
Dendronotus and Melibe differ in the synaptic connectivity of
Si1 and Si2 [6]. Only in Melibe do Si1 and Si2 exhibit strong ipsi-
lateral electrical coupling and contralateral reciprocal inhibition.
When the Si3 synapses were blocked by curare in Dendronotus,
artificial electrical coupling between ipsilateral Si1 and Si2 by the
dynamic clamp caused them to burst. This demonstrates that
the synaptic connectivity itself, not species-specific properties
of the neurons, is responsible for the different network responses
in curare. Moreover, the excitatory coupling within each half-
center had a stronger effect in producing rhythmic bursting
than did the redundant reciprocal inhibition. This may give a
hint as to why the electrically coupled Si1/2/4 in Melibe can
generate alternating bursts in curare, but not Si2 alone in Den-
dronotus. In a number of systems, electrical connections among
inhibitory interneurons contribute to synchronization of activity
[34–39]. Furthermore, modeling studies have pointed to complex
roles of electrical coupling in rhythmogenesis [40, 41]. The re-
sults may also indicate similar endogenous membrane proper-
ties of Si1 and Si2 in two species. The contributions of endoge-
nous membrane properties to the production of robust bursting
activity in these CPGs need to be examined further.
The results in the Dendronotus dynamic-clamping experi-
ments also indicate that the artificial electrical coupling between
Si1 and Si2 played a similar role as the excitatory synapse from
Si3 to Si2 by providing the excitatory drive onto each of the recip-
rocally inhibitory neurons. This indicates that, regardless of
whether it is from the Si3 synapse or the Si1 electrical connection,
the tonic excitatory drive onto Si2 is necessary to turn the circuit
into a half-center oscillator. In early hypothetical models of half-
center oscillators, it was proposed that the tonic excitatory drive
was required to activate the half-center kernel [42, 43]. In our
previous study, we gave the tonic excitatory drive by applying a
depolarizing current step of up to 3 nA to both of Si2s, which
induced slow bursting about 25% of the normal swim motor
pattern burst frequency. In this study, dynamic clamping created
the excitatory drive that was incorporated into the oscillator and
produced bursting at frequencies as high as the normal swim
motor pattern. A similar excitatory drive plays important roles in
providing rhythmic excitation to the mutually inhibitory neurons
in in vertebrate locomotor networks [44–50]. Thus, it may be a
general rule that rhythmic circuits contain a neural element that
provides the excitatory drive from within the circuit.
The species differences in the swim CPGs of Melibe and Den-
dronotus represent a real-life instantiation of theoretical studies,
which propose that there are multiple ways to connect neurons
Current Biology 27, 1–14, June 19, 2017 11
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016
to get similar rhythmic outputs [51, 52]. Other studies have
shown that synaptic strength can vary substantially between in-
dividuals of the same species, with little effect on the motor
output [53–56]. In the crab stomatogastric ganglion, different
neuromodulatory mechanisms in the same animal produce
similar motor patterns by involving distinct sets of neurons
[57, 58]. This study, however, shows that it is neither the strength
of connections that differed between the two CPGs nor recruiting
different neurons for a similar function; rather, they employ totally
different synaptic strategies, using curare-sensitive synapses in
opposite ways, to produce the same behavioral outputs.
Homologous Behaviors Have a Different Neural Basis
In general, swimming is rare among nudibranchs. Of the more
than 2,000 species, only about 50 have been observed to
swim with LR body flexion [8]. A recent molecular phylogeny re-
vealed that Melibe and Dendronotus belong to a monophyletic
clade [7]. All of the genera in this clade have species that swim
with LR alternation [8]. This suggests that LR swimming existed
in the most recent common ancestor of this clade and therefore
is homologous. Thus, the neural mechanisms for this behavior
diverged, while the behavior itself was conserved.
One can only speculate about the reasons why neural mecha-
nisms for homologous behaviors might have diverged. Perhaps
some of the neurons in one or the other species have taken on
additional functions that provide selective pressure to alter the
ancestral connectivity. Perhaps there was no selective advan-
tage to either CPG conformation and they both function well
enough, permitting the circuitry merely to ‘‘drift’’ away from their
ancestral state and become stabilized in their current species-
typical configurations. This ‘‘neural drift’’ might be analogous
to genetic drift wherein mutations accumulate over time hidden
from phenotypic change until they are fixed by selection [59].
The results support the concept that neural mechanism and
behavior represent separate levels of the biological hierarchy
[1–4]. This has important implications for extrapolating neural
mechanisms by simply comparing behaviors of species that
share homologous behaviors; underlying neural mechanisms
and microcircuitry can diverge while retaining behavioral function.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
d METHOD DETAILS
B Preparation
B Electrophysiology
B Dynamic clamp
d QUANTIFICATION AND STATISTICAL ANALYSIS
SUPPLEMENTAL INFORMATION
Supplemental Information includes five figures, one table, six data files, and
two movies and can be found with this article online at http://dx.doi.org/10.
1016/j.cub.2017.05.016.
12 Current Biology 27, 1–14, June 19, 2017
AUTHOR CONTRIBUTIONS
A.S. conducted the experiments and analyzed the results. A.S. and P.S.K. de-
signed the experiments and wrote the manuscript.
ACKNOWLEDGMENTS
We thank Dr. Andrey Shilnikov for discussions and for critically reading the
manuscript and Dr. Farzan Nadim for helpful advice on setting up the dy-
namic-clamp system in an earlier stage of this study. We also thank the anon-
ymous referees for helpful comments. This work was supported by National
Science Foundation grant IOS-1455527.
Received: February 5, 2017
Revised: April 6, 2017
Accepted: May 5, 2017
Published: June 1, 2017
REFERENCES
1. Sommer, R.J. (2008). Homology and the hierarchy of biological systems.
BioEssays 30, 653–658.
2. Striedter, G.F., and Northcutt, R.G. (1991). Biological hierarchies and the
concept of homology. Brain Behav. Evol. 38, 177–189.
3. Lauder, G.V. (1994). Homology, form, and function. In Homology: The
Hierarchical Basis of Comparative Biology, B.K. Hall, ed. (Academic
Press), pp. 152–197.
4. Katz, P.S. (2016). Evolution of central pattern generators and rhythmic be-
haviours. Philos. Trans. R. Soc. Lond. B Biol. Sci. 371, 20150057.
5. Lawrence, K.A., and Watson, W.H., 3rd. (2002). Swimming behavior of the
nudibranch Melibe leonina. Biol. Bull. 203, 144–151.
6. Sakurai, A., Newcomb, J.M., Lillvis, J.L., and Katz, P.S. (2011). Different
roles for homologous interneurons in species exhibiting similar rhythmic
behaviors. Curr. Biol. 21, 1036–1043.
7. Goodheart, J.A., Bazinet, A.L., Collins, A.G., and Cummings, M.P. (2015).
Relationships within Cladobranchia (Gastropoda: Nudibranchia) based on
RNA-seq data: An initial investigation. R. Soc. Open Sci. 2, 150196.
8. Newcomb, J.M., Sakurai, A., Lillvis, J.L., Gunaratne, C.A., and Katz, P.S.
(2012). Homology and homoplasy of swimming behaviors and neural cir-
cuits in the Nudipleura (Mollusca, Gastropoda, Opisthobranchia). Proc.
Natl. Acad. Sci. USA 109 (Suppl 1 ), 10669–10676.
9. Gosliner, T.M., and Smith, V.G. (2003). Systematic review and phylo-
genetic analysis of the nudibranch genus Melibe (Opisthobranchia:
Dendronotacea) with descriptions of three new species. Proc. Calif.
Acad. Sci. 54, 302–355.
10. Stout, C.C., Pola, M., and Valdes, A. (2010). Phylogenetic analysis of
Dendronotus nudibranchs with emphasis on northeastern Pacific species.
J. Molluscan Stud. 76, 367–375.
11. Sakurai, A., Gunaratne, C.A., and Katz, P.S. (2014). Two interconnected
kernels of reciprocally inhibitory interneurons underlie alternating left-right
swim motor pattern generation in the mollusk Melibe leonina.
J. Neurophysiol. 112, 1317–1328.
12. Sakurai, A., and Katz, P.S. (2016). The central pattern generator underlying
swimming in Dendronotus iris: A simple half-center network oscillator with
a twist. J. Neurophysiol. 116, 1728–1742.
13. Newcomb, J.M., and Katz, P.S. (2007). Homologues of serotonergic cen-
tral pattern generator neurons in related nudibranch molluscs with diver-
gent behaviors. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav.
Physiol. 193, 425–443.
14. Lillvis, J.L., Gunaratne, C.A., and Katz, P.S. (2012). Neurochemical and
neuroanatomical identification of central pattern generator neuron homo-
logues in Nudipleura molluscs. PLoS ONE 7, e31737.
15. Croll, R.P. (1987). Identified neurons and cellular homologies. In Nervous
Systems in Invertebrates, M.A. Ali, ed. (Plenum Publishing), pp. 41–59.
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016
16. Sharp, A.A., Abbott, L.F., and Marder, E. (1992). Artificial electrical synap-
ses in oscillatory networks. J. Neurophysiol. 67, 1691–1694.
17. Sharp, A.A., O’Neil, M.B., Abbott, L.F., and Marder, E. (1993). The dynamic
clamp: Artificial conductances in biological neurons. Trends Neurosci. 16,
389–394.
18. Sharp, A.A., O’Neil, M.B., Abbott, L.F., and Marder, E. (1993). Dynamic
clamp: Computer-generated conductances in real neurons.
J. Neurophysiol. 69, 992–995.
19. Shaw, S.R., and Meinertzhagen, I.A. (1986). Evolutionary progression at
synaptic connections made by identified homologous neurones. Proc.
Natl. Acad. Sci. USA 83, 7961–7965.
20. Katz, P.S., and Tazaki, K. (1992). Comparative and evolutionary aspects of
the crustacean stomatogastric system. In Dynamic Biological Networks:
The Stomatogastric Nervous System, R.M. Harris-Warrick, E. Marder,
A.I. Selverston, and M. Moulins, eds. (MIT Press), pp. 221–261.
21. Weimann, J.M., Meyrand, P., and Marder, E. (1991). Neurons that form
multiple pattern generators: Identification and multiple activity patterns
of gastric/pyloric neurons in the crab stomatogastric system.
J. Neurophysiol. 65, 111–122.
22. Baltzley, M.J., Gaudry, Q., and Kristan, W.B., Jr. (2010). Species-specific
behavioral patterns correlate with differences in synaptic connections
between homologous mechanosensory neurons. J. Comp. Physiol.
A Neuroethol. Sens. Neural Behav. Physiol. 196, 181–197.
23. Bumbarger, D.J., Riebesell, M., Rödelsperger, C., and Sommer, R.J.
(2013). System-wide rewiring underlies behavioral differences in predatory
and bacterial-feeding nematodes. Cell 152, 109–119.
24. Bierman, H.S., Zottoli, S.J., and Hale, M.E. (2009). Evolution of the
Mauthner axon cap. Brain Behav. Evol. 73, 174–187.
25. Ding, H., Smith, R.G., Poleg-Polsky, A., Diamond, J.S., and Briggman, K.L.
(2016). Species-specific wiring for direction selectivity in the mammalian
retina. Nature 535, 105–110.
26. Kondo, H., Lavenex, P., and Amaral, D.G. (2009). Intrinsic connections of
the macaque monkey hippocampal formation: II. CA3 connections.
J. Comp. Neurol. 515, 349–377.
27. Amaral, D.G., Insausti, R., and Cowan, W.M. (1984). The commissural con-
nections of the monkey hippocampal formation. J. Comp. Neurol. 224,
307–336.
28. Swanson, L.W., Wyss, J.M., and Cowan, W.M. (1978). An autoradio-
graphic study of the organization of intrahippocampal association
pathways in the rat. J. Comp. Neurol. 181, 681–715.
29. Elliott, C.J., Stow, R.A., and Hastwell, C. (1992). Cholinergic interneurons
in the feeding system of the pond snail Lymnaea stagnalis. I. Cholinergic
receptors on feeding neurons. Philos. Trans. R. Soc. Lond. B Biol. Sci.
336, 157–166.
30. Kehoe, J. (1972). Three acetylcholine receptors in Aplysia neurones.
J. Physiol. 225, 115–146.
31. van Nierop, P., Keramidas, A., Bertrand, S., van Minnen, J., Gouwenberg,
Y., Bertrand, D., and Smit, A.B. (2005). Identification of molluscan nicotinic
acetylcholine receptor (nAChR) subunits involved in formation of cation-
and anion-selective nAChRs. J. Neurosci. 25, 10617–10626.
32. van Nierop, P., Bertrand, S., Munno, D.W., Gouwenberg, Y., van Minnen,
J., Spafford, J.D., Syed, N.I., Bertrand, D., and Smit, A.B. (2006).
Identification and functional expression of a family of nicotinic acetylcho-
line receptor subunits in the central nervous system of the mollusc
Lymnaea stagnalis. J. Biol. Chem. 281, 1680–1691.
33. Carpenter, D.O., Swann, J.W., and Yarowsky, P.J. (1977). Effect of curare
on responses to different putative neurotransmitters in Aplysia neurons.
J. Neurobiol. 8, 119–132.
34. Arshavsky YuI, Beloozerova, I.N., Orlovsky, G.N., Panchin YuV, and
Pavlova, G.A. (1985). Control of locomotion in marine mollusc Clione lima-
cina. II. Rhythmic neurons of pedal ganglia. Exp. Brain Res. 58, 263–272.
35. Kiehn, O., and Tresch, M.C. (2002). Gap junctions and motor behavior.
Trends Neurosci. 25, 108–115.
36. Bennett, M.V., and Zukin, R.S. (2004). Electrical coupling and neuronal
synchronization in the Mammalian brain. Neuron 41, 495–511.
37. Beierlein, M., Gibson, J.R., and Connors, B.W. (2000). A network of elec-
trically coupled interneurons drives synchronized inhibition in neocortex.
Nat. Neurosci. 3, 904–910.
38. Marder, E., and Eisen, J.S. (1984). Electrically coupled pacemaker neu-
rons respond differently to same physiological inputs and neurotransmit-
ters. J. Neurophysiol. 51, 1362–1374.
39. Mouser, C., Bose, A., and Nadim, F. (2016). The role of electrical coupling
in generating and modulating oscillations in a neuronal network. Math.
Biosci. 278, 11–21.
40. Kepler, T.B., Marder, E., and Abbott, L.F. (1990). The effect of electrical
coupling on the frequency of model neuronal oscillators. Science 248,
83–85.
41. Kopell, N., and Ermentrout, B. (2004). Chemical and electrical synapses
perform complementary roles in the synchronization of interneuronal net-
works. Proc. Natl. Acad. Sci. USA 101, 15482–15487.
42. Friesen, W.O., and Stent, G.S. (1977). Generation of a locomotory rhythm by
a neural network with reccurrent cyclic inhibition. Biol. Cybern. 28, 27–40.
43. Friesen, W.O. (1994). Reciprocal inhibition: A mechanism underlying oscil-
latory animal movements. Neurosci. Biobehav. Rev. 18, 547–553.
44. Grillner, S. (2006). Biological pattern generation: The cellular and compu-
tational logic of networks in motion. Neuron 52, 751–766.
45. Grillner, S. (2003). The motor infrastructure: From ion channels to neuronal
networks. Nat. Rev. Neurosci. 4, 573–586.
46. Grillner, S., and El Manira, A. (2015). The intrinsic operation of the networks
that make us locomote. Curr. Opin. Neurobiol. 31, 244–249.
47. Ha€gglund, M., Borgius, L., Dougherty, K.J., and Kiehn, O. (2010).
Activation of groups of excitatory neurons in the mammalian spinal cord
or hindbrain evokes locomotion. Nat. Neurosci. 13, 246–252.
48. Kiehn, O. (2016). Decoding the organization of spinal circuits that control
locomotion. Nat. Rev. Neurosci. 17, 224–238.
49. Roberts, A., Soffe, S.R., Wolf, E.S., Yoshida, M., and Zhao, F.Y. (1998).
Central circuits controlling locomotion in young frog tadpoles. Ann.
N Y Acad. Sci. 860, 19–34.
50. Roberts, A., Li, W.C., Soffe, S.R., and Wolf, E. (2008). Origin of excitatory
drive to a spinal locomotor network. Brain Res. Brain Res. Rev. 57, 22–28.
51. Marder, E. (2011). Variability, compensation, and modulation in neurons
and circuits. Proc. Natl. Acad. Sci. USA 108 (Suppl 3 ), 15542–15548.
52. Prinz, A.A., Bucher, D., and Marder, E. (2004). Similar network activity from
disparate circuit parameters. Nat. Neurosci. 7, 1345–1352.
53. Sakurai, A., Tamvacakis, A.N., and Katz, P.S. (2014). Hidden synaptic dif-
ferences in a neural circuit underlie differential behavioral susceptibility to
a neural injury. eLife 3, 02598.
54. Roffman, R.C., Norris, B.J., and Calabrese, R.L. (2012). Animal-to-animal
variability of connection strength in the leech heartbeat central pattern
generator. J. Neurophysiol. 107, 1681–1693.
55. Norris, B.J., Wenning, A., Wright, T.M., and Calabrese, R.L. (2011).
Constancy and variability in the output of a central pattern generator.
J. Neurosci. 31, 4663–4674.
56. Goaillard, J.M., Taylor, A.L., Schulz, D.J., and Marder, E. (2009).
Functional consequences of animal-to-animal variation in circuit parame-
ters. Nat. Neurosci. 12, 1424–1430.
57. Rodriguez, J.C., Blitz, D.M., and Nusbaum, M.P. (2013). Convergent
rhythm generation from divergent cellular mechanisms. J. Neurosci. 33,
18047–18064.
58. Saideman, S.R., Blitz, D.M., and Nusbaum, M.P. (2007). Convergent motor
patterns from divergent circuits. J. Neurosci. 27, 6664–6674.
59. Wright, S. (1955). Classification of the factors of evolution. Cold Spring
Harb. Symp. Quant. Biol. 20, 16–24D.
60. Kemenes, I., Marra, V., Crossley, M., Samu, D., Staras, K., Kemenes, G.,
and Nowotny, T. (2011). Dynamic clamp with StdpC software. Nat.
Protoc. 6, 405–417.
Current Biology 27, 1–14, June 19, 2017 13
 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

61. Sakurai, A., and Katz, P.S. (2003). Spike timing-dependent serotonergic
neuromodulation of synaptic strength intrinsic to a central pattern gener-
ator circuit. J. Neurosci. 23, 10745–10755.
62. Katz, P.S., Getting, P.A., and Frost, W.N. (1994). Dynamic neuromodula-
tion of synaptic strength intrinsic to a central pattern generator circuit.
Nature 367, 729–731.
63. Liao, X., and Walters, E.T. (2002). The use of elevated divalent cation so-
lutions to isolate monosynaptic components of sensorimotor connections
in Aplysia. J. Neurosci. Methods 120, 45–54.
64. Byrne, J.H., Castellucci, V.F., and Kandel, E.R. (1978). Contribution of in-
dividual mechanoreceptor sensory neurons to defensive gill-withdrawal
reflex in Aplysia. J. Neurophysiol. 41, 418–431.
65. Destexhe, A., Mainen, Z.F., and Sejnowski, T.J. (1994). An efficient method
for computing synaptic conductances based on a kinetic model of recep-
tor binding. Neural Comput. 6, 14–18.
66. Sharp, A.A., Skinner, F.K., and Marder, E. (1996). Mechanisms of oscilla-
tion in dynamic clamp constructed two-cell half-center circuits.
J. Neurophysiol. 76, 867–883.

14 Current Biology 27, 1–14, June 19, 2017

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Chemicals, Peptides, and Recombinant Proteins
(+)-Tubocurarine chloride pentahydrate Sigma-Aldrich 93750
Experimental Models: Organisms/Strains
Wild animal specimen: Melibe leonina Monterey Abalone Company; Marinus Scientific; Living Elements N/A
Wild animal specimen: Dendronotus iris Monterey Abalone Company; Marinus Scientific; Living Elements N/A
Software and Algorithms
SigmaPlot Jandel Scientific N/A
Spike2 Cambridge Electronic Design N/A
StdpC Kemenes et al., 2011 [60] N/A
Other
Artificial sea salt Instant Ocean N/A
Sylgard 184 Dow Corning 0006228027
AxoClamp 2B Molecular Devices N/A
1401Plus Cambridge Electronic Design N/A
Micro1401 Cambridge Electronic Design N/A
I/O PCI card National Instruments PCI-6229
BNC connector blocks National Instruments BNC2110
Micropipette puller Sutter Instrument Model P-97
Glass capillaries Sutter Instrument BF100-78-10

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Akira
Sakurai (akira@gsu.edu)

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Adult specimens of Melibe leonina (see Movie S1) and Dendronotus iris (see Movie S2) (Nudibranchia, Gastropoda, Mollusca;
2-15 cm and 6 to 20 cm in body length, respectively) were collected by Monterey Abalone Company (Monterey, CA), Marinus Sci-
entific, LLC (Long Beach, CA), and Living Elements Ltd (Delta, BC, Canada). These species are obligate hermaphrodites. Animals
were kept in artificial seawater tanks at 10 to 12 C with a 12:12 light/dark cycle for up to 2 weeks before use. > 19 Melibe specimens
and > 39 Dendronotus specimens were used in this study.

METHOD DETAILS

Preparation
For all experiments, the brain, which consists of the fused cerebropleural and pedal ganglia, was isolated. Animals were first anes-
thetized by injection of 0.33 M MgCl2 solution into the body cavity. An incision was made in the left body wall near the esophagus. The
brain was taken out together with a portion of the esophagus and transferred to a petri dish lined with Sylgard 184 (Dow corning,
Midland, MI). Artificial seawater (Instant Ocean, Mentor, OH) was used as normal saline. The brain was superfused with artificial
seawater at a rate of 0.5-1.0 mL/min. After pinning the brain in the dish with the dorsal side up, the esophagus was removed, and
the pedal ganglia and pedal commissures were positioned anterior to the cerebral ganglia. Connective tissue surrounding the brain
was manually removed with forceps and fine scissors while keeping the brain at 4 C to reduce neuronal activity. The temperature was
raised to 10 C before starting electrophysiological experiments.

Electrophysiology
Suction electrodes made from polyethylene tubing were placed on Pedal Nerves 3 (PdN3) for stimulation. To evoke the swim motor
program, either the left or the right PdN3 was stimulated with a train of voltage pulses (5–15 V, 1 ms) at 5 Hz for 1-3 s. Intracellular

Current Biology 27, 1–14.e1–e3, June 19, 2017 e1

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

recordings were obtained using 15-35 MU glass microelectrodes filled with 3 M potassium chloride or mixture of 2 M potassium ac-
etate and 0.1 M potassium chloride. The microelectrodes were made from borosilicate glass capillaries (O.D.: 1.0 mm, I.D.: 0.78 mm)
with a micropipette puller (Model P-97, Sutter Instrument, Novato, CA). Each electrode was connected to an Axoclamp 2B amplifier
(Molecular Devices, Sunnyvale, CA). The membrane potential waveforms were digitized (> 2 kHz) with a 1401Plus or Micro1401 A/D
converter from Cambridge Electronic Design (Cambridge, UK). Data acquisition and analysis were performed with Spike2 software
(Cambridge Electric Design, Cambridge, UK), and SigmaPlot (Jandel Scientific, San Rafael, CA).
To test for monosynaptic connectivity and direct electrical coupling between the swim interneurons, high divalent cation (Hi-Di)
saline was used to raise the threshold for spiking and reduce spontaneous neural firing [6, 61–64]. The composition of the Hi-Di saline
was (in mM): 285 NaCl, 10 KCl, 25 CaCl2, 125 MgCl2, 11 D-glucose, and 10 HEPES, pH 7.6. A train of action potentials was evoked in a
neuron by injecting a current step (0.5 to 2 nA) through a bridge-balanced electrode.
To distinguish bursting in left-right alternation from irregular spiking, we performed a cross-correlation analysis using Spike2 soft-
ware (CED, Cambridge, UK) as described previously [12]. A train of Si2 spikes was counted as a burst if the spike interval was less
than 0.5 s and the duration of the spike train was more than 0.5 s. The burst frequency was measured by the interval of median spike in
each burst of Si1 or Si2 and averaged over 40 s. When bursting in left-right alternation was not detected, then average spike fre-
quency in the duration of > 30 s was used as the intraburst spike frequency.
Curare [(+)-Tubocurarine chloride pentahydrate, Sigma-Aldrich, St. Louis, MO] was dissolved in either artificial seawater or Hi-Di
saline just before use to attain the final concentrations (0.1 mM to 100 mM). Curare was bath-applied by switching the superfusion
paths (Melibe, n = 16 animals; Dendronotus n = 35 animals). Missing data were due to the loss of intracellular recording from the
neurons.

Dynamic clamp
Dynamic-clamp experiments were performed in 6 Melibe specimens and 24 Dendronotus specimens. To perform dynamic-clamp
experiments, we used a separate computer equipped with an I/O PCI card (PCI-6229), and two BNC connector blocks
(BNC2110; National Instruments, Austin, TX). Dynamic-clamp software StdpC [60] was used to create up to six artificial chemical
synapses or electrical coupling simultaneously in four neurons at maximum. To perform dynamic clamping, neurons were impaled
with a single electrode under discontinuous current clamp mode or with two microelectrodes, one for membrane potential recording
and the other for current injection. The current injected into the postsynaptic neuron, Isyn, was calculated in each dynamic-clamp
cycle using a first order kinetics model of the release of neurotransmitter [60, 65, 66]:


Isyn = gsyn SðtÞ Vsyn
Vpost ðtÞ ; (eq. 1)

where S(t) is the instantaneous synaptic activation, gsyn is the maximum synaptic conductance, Vsyn is the reversal potential of the
synapse. In Melibe, we previously showed by current clamp recording that the reversal potential of Si3-to-Si3 synapse was relatively
close to the resting potential of the postsynaptic neuron whereas that of Si2-to-Si2 synapse was at more hyperpolarized level
because we could not reverse the polarity of IPSP by injecting large hyperpolarizing current [11]. We made a few successful record-
ings of synaptic current by voltage-clamping the swim interneurons in the two species (Figure S5). In Melibe, the Si3-evoked IPSC/Ps
recorded in Si3 had a mean reversal potential of
57.6 ± 4.5 mV, ranging from 54.1 to 62.8 mV (n = 3); the Si3-evoked IPSC/Ps in Si1
had mean reversal potentials of
70.7 ± 9.2 mV, ranging from
57.5 to
78.6 mV (n = 4). We could not detect reliable signals for Si2-
evoked IPSCs in Si2 when voltage-clamped. Considering the long distance from the Si1 cell body to the pedal ganglion where Si3
dendrites are located, it is likely that the reversal potential of IPSCs in Si1 may contain a large error due to poor space clamp. In Den-
dronotus, the mean reversal potential of the Si2-evoked IPSC/Ps in Si2 was
73.9 ± 8.1 mV, ranging from
65.0 to
85.3 mV (n = 4);
that of Si3-evoked IPSC/Ps in Si3 was
46.4 ± 3.9 mV, ranging from
43.5 to
51.9 mV (n = 4); and that of Si3-evoked EPSC/Ps in Si2
was
6.3 ± 4.3 mV, ranging from
0.2 to
12.3 mV (n = 5). Again, such a large range of values may suggest that these synapses were
all electrotonically distant from the somatic recording sites. Because of its technical difficulty, we did not attempt to obtain exact
values for each experiment; we set Vsyn to
60 mV for Melibe Si3 synapses, 0 mV for Dendronotus Si3-to-Si2 synapses. Since
the reversal potential of the Si2-evoked IPSP/C in both species showed more hyperpolarized levels than the Si3-evoked IPSP/C,
Vsyn was set to
80 mV for Dendronotus Si2-to-Si1 and Si1-to-Si1 synapses.
The instantaneous activation, S(t) is given by the differential equation:
dS
ð1
SN ðVpre ÞÞt syn = ðSN ðVpre Þ
SðtÞÞ; (eq. 2)
dt
where
8  
< tanh Vpre
Vthresh
>
if Vpre > Vthresh
SN ðVpre Þ = Vslope (eq. 3)
>
:
0 otherwise

SN is the steady-state synaptic activation and tsyn is the time constant for synaptic decay. In Melibe, Si3-evoked IPSCs recorded in a
post-synaptic cells in the pedal ganglia showed a decay time constant of 35 to 46 ms (n = 3). In Dendronotus, the Si3-evoked EPSCs
had a decay time constant of approximately 20 ms [12], whereas the Si2-evoked IPSCs ranged from 63 ms to 163 ms (n = 3). For

e2 Current Biology 27, 1–14.e1–e3, June 19, 2017

 Please cite this article in press as: Sakurai and Katz, Artificial Synaptic Rewiring Demonstrates that Distinct Neural Circuit Configurations Underlie
Homologous Behaviors, Current Biology (2017), http://dx.doi.org/10.1016/j.cub.2017.05.016

Dendronotus Si3 excitatory synapses, tsyn was set to 40 ms to make the injected current more effective without injecting large ampli-
tude currents that often became harmful and caused recording conditions to deteriorate. For other inhibitory synapses, tsyn was fixed
at 80 ms so that we could promptly perform dynamic clamping while retaining intracellular recording from multiple neurons. Vpre is the
presynaptic membrane potential of Si3 and Vthresh is the threshold potential for the release of neurotransmitter, which was set to 50%
of the amplitude of the smallest Si3 action potentials. Since all synaptic transmissions were spike-mediated, the synaptic slope
parameter of the activation curve (Vslope) was unchanged from the default value (25 mV). gsyn was varied from 40 to 200 nS for the
Dendronotus Si3-to-Si2 synapse and 50 to 400 nS for other inhibitory synapses.
For electrical synapses, the currents Ies1 and Ies2 to be injected into two neurons are calculated according to:
Ies1 ðtÞ = gsyn ½Ves2 ðtÞ
Ves1 ðtÞ; (eq. 4)
and
Ies2 ðtÞ =
Ies1 ðtÞ; (eq. 5)
where Ves1(t) and Ves2(t) are the membrane potential of the two cells. gsyn was varied from 20 to 100 nS.
Dynamic clamping was performed with the interval of more than 40 s with no defined order of conductance. Specimens were
excluded from analysis when Si2 showed no alternating bursting with all conductance values examined.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical comparisons were performed using SigmaPlot ver. 12.5 (Jandel Scientific, San Rafael, CA) for one-way repeated-mea-
sures (RM) ANOVA (Figures 3B and 3C, n = 16 animals; Figure 3D, Si1, n = 14 cells, Si2, n = 12 cells, Si3, n = 13 cells; Figures
4B–4D, n = 6 animals; Figures 5B, n = 35 animals, 5C, Si2, n = 35 cells, Si3, n = 27 cells; Figures 6B and 6C, n = 8 animals; Figures
7B and 7C, n = 12 animals, 7F and 7G, n = 9 animals) followed by all pairwise multiple comparison procedures (Holm-Sidak method).
Missing data were due to the loss of intracellular recording from the neurons. p values are stated in the figure legends. In all tests,
Shapiro-Wilk test was used to assume normality of data structure; it showed p > 0.05 in all statistical analyses in this study. Results
are expressed as the mean ± SD in Figures 2C, 3B–3D, 4B–4D, 5B, 5C, 6B, 6C, 7B, 7C, 7F, and 7G.
The dose-response curves in Figure 2C were created with SigmaPlot with an equation:
1
min
f = min + (eq. 6)
1 + 10ðlogEC50
xÞHillslope
where x is the log dose and min is the minimum amplitude of the synaptic potentials normalized to the control value.

Current Biology 27, 1–14.e1–e3, June 19, 2017 e3