Peripheral multidendritic sensory neurons
are necessary for rhythmic locomotion
behavior in Drosophila larvae
Wei Song, Maika Onishi, Lily Yeh Jan, and Yuh Nung Jan*
Departments of Physiology and Biochemistry, Howard Hughes Medical Institute, University of California, San Francisco, CA 94143
Contributed by Yuh Nung Jan, February 1, 2007 (sent for review January 3, 2007)
From breathing to walking, rhythmic movements encompass phys-
iological processes important across the entire animal kingdom. It
is thought by many that the generation of rhythmic behavior is
peripheral sensory input. Sensory feedback is, however, required
to modify or coordinate the motor activity in response to the
circumstances of actual movement. In contrast to this notion, we
report here that sensory input is necessary for the generation of
Drosophila larval locomotion, a form of rhythmic behavior. Block-
age of all peripheral sensory inputs resulted in cessation of larval
crawling. By conditionally silencing various subsets of larval pe-
ripheral sensory neurons, we identified the multiple dendritic (MD)
neurons as the neurons essential for the generation of rhythmic
peristaltic locomotion. By recording the locomotive motor activi-
ties, we further demonstrate that removal of MD neuron input
disrupted rhythmic motor firing pattern in a way that prolonged
the stereotyped segmental motor firing duration and prevented
the propagation of posterior to anterior segmental motor firing.
These findings reveal that MD sensory neuron input is a necessary
rhythmic behavior ? sensory feedback ? locomotion generation ?
motor pattern ? central pattern generator
behavior is that all rhythmic movements are generated by
specialized circuits within the CNS called central pattern gen-
erators (CPGs), which can operate without sensory inputs (3–5).
Nevertheless, a functional motor program also requires sensory
inputs reporting peripheral feedback due to the actual move-
ments to produce the correct behavior (6, 7). This model draws
support from reports of sensory feedback affecting the timing
and magnitude of motor activity generated by CPGs and from
studies based on surgical removal of afferent input (1, 4, 6, 7).
Whereas there are electrophysiological demonstrations of the
influence of sensory feedback on the output of CPGs, how
interactions between sensory neurons in the peripheral and
neurons in the CNS contribute to rhythmic behavior in an intact
organism is not known. It is therefore important to ask whether
sensory inputs are essential for rhythmic behavior and how
sensory inputs contribute to rhythmic behavior.
Drosophila is ideal for investigations of neural circuits under-
lying rhythmic behavior due to its amenability to genetic ma-
nipulation. Larval crawling, used for travel across various types
of terrain, is accomplished by repeated peristaltic wave-like
strides (8) and involves molecules mediating neuronal signaling
(9–12). Whereas one study suggests that embryonic assembly of
CPG for larval locomotion does not require sensory inputs, the
same study reports that interference with the sensory function
during embryogenesis severely disrupts the actual patterns of
locomotion (13). To explore how sensory inputs contribute to
larval locomotion, we conducted a Gal4/upstream activating
sequence (UAS)-based screen to identify components of the
hythmic behaviors comprise a cyclic, repetitive set of move-
ments, such as locomotion, respiration, and mastication (1,
neural circuitry for larval locomotion, revealing that sensory
neuron function is necessary for larval locomotion.
Results and Discussion
Drosophila Larval Sensory Neurons Are Necessary for Locomotion
Behavior. Drosophila larval crawling entails repeated, rhythmic,
peristaltic contraction. During each peristaltic stride, muscle
contractions are propagated from one end of the body to the
other end, passing through all 11 segments one by one [support-
ing information (SI) Fig. 6 and Movie 1]. To identify sensory
neurons important for larval locomotion, we crossed ?1,000
Gal4 driver lines to UAS-shibirets(UAS-shits) (‘‘ts’’ indicates
temperature sensitive) to transiently reduce the synaptic func-
tion of various subsets of Gal4 expressing neurons at the third
larval instar and looked for gross locomotion defects in peri-
stalsis rhythmicity and speed. Because shits, a temperature-
sensitive mutant form of dynamin, blocks synaptic vesicle recy-
cling and disrupts synaptic transmission only at restrictive
temperatures (?29°C) (14, 15), we could raise these larvae at
18°C and ensure normal development of the larval crawling
circuitry. We then subjected third instar foraging larvae to
restrictive temperature (37°C) for 15 min to silence neurons
expressing shitsand observed the locomotion behavior.
For the majority of the analyzed Gal4 lines, incubation at
restrictive temperature resulted in enhanced locomotion speeds
as observed in wild-type larvae (Fig. 1a); faster crawling and
increased activity might be a natural avoidance response to high
temperatures (16, 17). We isolated 130 Gal4 lines that give rise
to different types of defects in larval locomotion (see Methods).
Among those 130 lines, UAS-shitsexpression driven by 10 Gal4
lines resulted in drastically disrupted peristaltic rhythm and
reduced locomotion speed. Using these 10 Gal4 lines to drive the
expression of UAS–mCD8–GFP to identify the neurons that
were silenced in the larvae with defective rhythmic locomotion
(18), we found that all 10 Gal4 lines showed expression in some
or all peripheral sensory neurons. In particular, the 5-40-Gal4
line drives gene expression in all larval peripheral sensory
neurons, but not in CNS neurons (Fig. 2). This Gal4 line
(SN-Gal4) allowed us to study the role of peripheral sensory
neurons in larval locomotion.
First, we subjected SN-Gal4/UAS-shitslarvae to restrictive
temperature (37°C) for different durations to examine more
M.O. analyzed data; and W.S., M.O., L.Y.J., and Y.N.J. wrote the paper.
The authors declare no conflict of interest.
Freely available online through the PNAS open access option.
Abbreviations: CPG, central pattern generator; Kir, inwardly rectifying K?channel; MD,
multiple dendritic; shits, shibirets; TNT, tetanus toxin; ts, temperature sensitive; UAS,
upstream activating sequence.
*To whom correspondence should be addressed. E-mail: email@example.com.
This article contains supporting information online at www.pnas.org/cgi/content/full/
© 2007 by The National Academy of Sciences of the USA
www.pnas.org?cgi?doi?10.1073?pnas.0700895104 PNAS ?
March 20, 2007 ?
vol. 104 ?
no. 12 ?
closely the impact of reducing sensory inputs as a consequence
of impaired synaptic vesicle recycling in sensory neurons.
Whereas wild-type larvae showed either no change or an accel-
eration of crawling with increases in incubation duration, SN-
Gal4/UAS-shitslarvae exhibited a progressive decrease in linear
locomotion speed and became immobilized after 20 min at 37°C
(Fig. 1a and SI Movies 2 and 3). The percentage of immobilized
SN-Gal4/UAS-shitslarvae also grew with prolonged exposure to
37°C, reaching 100% after 20 min (Fig. 1b), similar to our
observations of late embryo and first instar larval locomotion
(data not shown). The effect of sensory deprivation at restrictive
temperature is reversible; shifting larvae from 37°C to 18°C
restored crawling behavior (data not shown). Whereas sensory
neuron function was essential for the rhythmic crawling, larvae
immobilized by exposure to 37°C still exhibited movements of
the mouth hook (SI Movie 3), indicating that some other type of
movement remains. To corroborate these results, we used
Gal80ts/UAS-inwardly rectifying K?channel (Kir) as an alter-
native means to silence all sensory neurons conditionally at third
larval instar by acutely overexpressing Kir so that the neurons
would be much less excitable (19). We found that larvae carrying
both SN-Gal4 and Gal80ts/UAS-Kir became immobilized after
exposure to restrictive temperature for 30 min (data not shown).
In a previous study, Suster and Bate (13) reported that they used
neurons throughout development without abolishing the ability of
the CPG to generate peristaltic waves in the embryo, albeit the
waves were abnormal. How might this study be reconciled with our
results? One possibility is that the Gal4 drivers used in these two
studies have different strengths. Another possibility might be that
homeostatic processes could make CPGs sensitive to acute with-
drawal of a sensory input in our study but resistant to chronic
withdrawal in the study of Suster and Bate (13, 20). We therefore
did a side-by-side comparison of the driver (PO163-Gal4) used by
Suster and Bate and our driver (SN-Gal4). Compared with GFP
expression driven by SN-Gal4, GFP expression driven by PO-163-
Gal4 was weaker and further diminished during development.
Moreover, TNT or Kir expression by PO-163-Gal4 allowed larvae
to the results reported by Suster and Bate. In contrast, TNT or Kir
0 2.5 5 7.5 10 15 20 (min)
Duration at restrictive temperature
Percentage of immobile larvae (%)
0 2.5 5 7.5 10 15 20 (min)
Duration at restrictive temperature
Larval locomotion speed (mm/s)
% of immobile larvae
Larval locomotion speed (mm/s)
- + + + +
- - SN Type II Type I
restrictive temperature (37°C) until they ceased crawling. n ? 20 per column. (b) Percentage of SN-Gal4/UAS-shitslarvae immobilized after different incubation
durations at restrictive temperature, reaching ?100% in 20 min. n ? 20 per column. No wild-type larvae were found to be immobile after the same incubation
treatment. (c) Larval locomotion speed after inactivation of different types of sensory neurons. All experiments were done after a 10-min exposure to restrictive
temperature (37°C). n ? 15 per column.*, P ? 0.001 versus wild-type. (d) Percentage of larvae immobilized after 20 min at restrictive temperature. n ? 15 per
column. All error bars indicate SE.
MD sensory neurons’ function is essential for larval crawling. (a) Locomotion speed of SN-Gal4/UAS-shitslarvae decreased with longer exposure to
www.pnas.org?cgi?doi?10.1073?pnas.0700895104 Song et al.
expression driven by SN-Gal4 rendered the embryos immobilized
and unable to hatch (n ? 28). It thus seems likely that the
PO-163-Gal4 driver is not sufficiently strong to produce enough
TNT light chain in the sensory neurons to completely inactivate all
of the sensory neurons, hence producing the weaker effect (13).
Larval Locomotion Requires Multiple Dendritic (MD) Sensory Neurons.
The Drosophila larval peripheral nervous system is composed of
segmentally repeated sensory neurons, which are further divided
into type I and type II neurons (21). Type I neurons have ciliated
monopolar dendrites, whereas wild type II neurons (also known
as MD neurons) have MD projections (Fig. 2b). To determine
which sensory neurons are necessary for rhythmic crawling
behavior, we selectively expressed shitsin either or both subtypes
of peripheral sensory neurons by using Gal4 lines, including
dcirl-Gal4/UAS-shits(Type I-Gal4/UAS-shits) for shitsexpres-
sion in type I sensory neurons (chordotonal and external sensory
organs) and 21-7-Gal4/UAS-shits(MD-Gal4/UAS-shits) for shits
expression only in type II or MD neurons. Whereas depriving
inputs from type I sensory neurons had no obvious effect on
crawling, reducing pan-MD or pan-sensory input drastically
decreased the crawling speed (0.094 ? 0.022 mm/s, and 0.095 ?
0.033 mm/s, respectively), about six times slower than control
values (Fig. 1c and SI Movie 4). After 20 min at restrictive
temperature, MD-Gal4/UAS-shitslarvae were immobile (Fig.
1d), implicating MD neurons as the type of sensory neurons that
relay sensory inputs essential for generating larval locomotion.
No locomotion defects were observed when all those Gal4 lines
larvae were exposed to restrictive temperature (data not shown).
Sensory Inputs from MD Neurons Is Required for Normal Peristaltic
Contraction Duration. How might impaired MD neuron function
have reduced the speed of larval crawling? Larvae move forward
by making rhythmic peristaltic contractions. A reduction of
speed could arise from either a delay in finishing each peristaltic
wave, resulting in prolonged stride duration and reduced fre-
MD neurons may regulate crawling, we used high-magnification
video recordings to analyze the phenotype of larvae with par-
tially inactivated sensory neurons (those that had been exposed
to the restrictive temperature for durations ?20 min). We
focused on studying the peristaltic movement during each con-
traction stride and assessed the peristalsis wave duration and
stride length (Fig. 3a and data not shown). Over time, at 37°C,
the duration of each peristalsis wave of motion remained ?1 sec
for wild type larvae. In contrast, the peristalsis wave duration
increased progressively as MD-Gal4/UAS-shitsand SN-Gal4/
UAS-shitslarvae remained at 37°C, and the muscle contractions
became more arrhythmic and sustained for longer periods of
time (SI Movies 2 and 4). There was a corresponding decrease
in peristalsis wave frequency (Fig. 3b) but no significant reduc-
tion in stride length (data not shown). Thus, reduction of MD
sensory neuron function results in a dramatic delay for each
system sensory neuron. (a) SN-Gal4 drives mCD8::GFP expression in all sensory
ing of different types of sensory neurons. (c) Axonal projection of all sensory
neurons (green) in larval ventral nerve cord (VNC) and brain (br) (red). Neuronal
staining by anti-elav antibody was at a 1:50 dilution. (Scales bar: 100 ?m.)
SN-Gal4 drives mCD8::GFP expression specifically in peripheral nervous
Duration at restrictive temperature
05 10 15 20 (min)
Peristalsis stride frequency (Hz)
Duration at restrictive temperature
05 1015 20 (min)
peristalsis. (a) Peristalsis duration increased with the duration of exposing
MD-Gal4/UAS-shitslarvae to 37°C. Peristalsis duration is not shown for the
20-min time point, because the larvae are immobile by then. n ? 20 per
exposed to 37°C. n ? 20 per symbol.
MD neuron function is required for the completion of locomotive
Song et al.
March 20, 2007 ?
vol. 104 ?
no. 12 ?
peristalsis wave and eventually a complete arrest of peristalsis
activity, implying that the circuitry that generates larval loco-
motion depends on MD neuron inputs.
MD Neuron Function Is Required for the Formation of Rhythmic
Locomotive Motor Pattern. The larval peristaltic crawling behavior
is accompanied by rhythmic firing of motoneuron axons in the
segmental nerves, which exhibit intersegmental coordination of
neuronal activity as revealed by an en passant recording from larval
segmental nerves in semiintact larvae displaying rhythmic peristal-
tic contraction (22) (Fig. 4a). In wild-type larvae with smooth and
coordinated peristaltic muscle contraction, the concurrent rhyth-
interval (Fig. 4 c and d). With increasing exposure to 37°C, burst
frequency increased, whereas burst duration remained unchanged
(Fig. 4 c and d). When MD-Gal4/UAS-shitslarvae were brought to
37°C, however, there was progressively more arrhythmic and irreg-
irregular with longer exposure to the restrictive temperature (Fig.
4 e and f), accounting for the dramatic delay in peristalsis stride
duration observed in intact animals (Fig. 3a). These results clearly
indicate that neural circuits within the larval CNS cannot maintain
a basic fictive firing pattern without sensory feedback involving the
MD neuron inputs.
Larval locomotion is a multisegmented movement that re-
quires intersegmental coordination of muscle contraction, just as
the walking of human beings requires coordination among limb
to-anterior rhythmic peristaltic contraction is associated with a
transition of motor bursting from one to the next segment (22).
5a). However, in MD-Gal4/UAS-shitslarvae after 15 min at the
restrictive temperature, this phase transition is lost: The burst
duration in the most posterior (L7) segment was prolonged, and,
during this time, the immediately anterior (L6) segment re-
mained silent (Fig. 5b). Accordingly, we observed that the L7
muscle kept on contracting, and the contraction could not be
propagated from L7 to L6. The failure of the propagation of
segmental nerve firing was observed during anterior-to-
posterior propagation as well (data not shown). Consistent with
the finding that the bursting pattern from isolated ventral nerve
ganglia was arrhythmic and uncoordinated among segments
(22), unlike the rhythmic firing detected by an en passant
recording of segmented nerves with preserved sensory inputs
(Fig. 5a), this result explains how a lack of sensory inputs finally
abolished rhythmic peristaltic contraction.
By using genetic manipulations in an intact animal, this study
demonstrates that the generation of larval locomotion requires
sensory inputs. Using 5-40-Gal4, a pan sensory neuron driver, to
block all larval sensory inputs, we show that sensory inputs are
essential for larval locomotion behavior. Similar results obtained
with 21-7-Gal4, which is active in all MD neurons, implicate MD
neurons as the type of sensory neurons that are required for
of segmental nerves. (a) Diagram of segmental nerve and body wall muscles
(b) Segmental nerve burst duration increased with prolonged inactivation of
sensory neuron (n ? 10). (c and d) Wild-type larvae retained rhythmic motor
firing at 37°C in both segment L6 (c) and L7 (d). (e and f) Prolonged burst
bars indicate SE.*, P ? 0.001 versus wild type.
tal nerve firing from L7 to L6. (a) Coordinated phase shift of segmental nerve
firing in wild-type larvae after 15 min at 37°C. (b) Firing of L7, but not L6,
segmental nerves in MDGal4/UAS-shitslarvae after 15 min at 37°C. (c) Hypo-
thetical model for the role of MD neuron in generating the intersegmentally
coordinated motor neuron (MN) firing for forward peristaltic locomotion. In
restrict the firing duration of motor neurons in the same segment (MN7) by
inhibiting the CPG circuit activity and to activate motor neurons in the
preceding segment (MN6) by exciting the CPG circuit activity in a process that
likely involves the CPG but critically depends on MD sensory neuron activity.
Blocking MD sensory neuron input prevents propagation of segmen-
www.pnas.org?cgi?doi?10.1073?pnas.0700895104Song et al.
peristaltic locomotion. Moreover, by recording the motor firing
pattern underlying locomotive peristalsis, we found that reduc-
tion of MD neuron function drastically prolonged motor burst
duration and eliminated the propagation of motor burst from
one segmental nerve to the next. These electrophysiology data
suggest that MD neuron activities are required for terminating
the motor firing in one segment and initiating another motor
burst in the next segment to produce the larval locomotion,
probably through interaction between MD neurons and CPG
circuits (see the model in Fig. 5c).
By using both approaches for acute and chronic withdrawal of
sensory inputs, we have shown that a block of sensory inputs
leads to the arrest of Drosophila larval crawling. To reconcile our
findings with a previous study in which locomotion persisted
despite the overexpression of TNT light chain by another
Drosophila pan-sensory Gal4, PO163-Gal4, notwithstanding a
reduction in peristalsis contraction frequency in embryo and
crawling speed in first instar larvae (13), we compared PO163-
Gal4 with the two Gal4 drivers used in this study, SN-Gal4 and
MD-Gal4, and found that the expression of UAS-TNT or
UAS-Kir driven by PO163-Gal4 but not SN-Gal4 or MD-Gal4
yielded mobile third instar larvae. It thus appears that PO163-
Gal4 does not produce enough TNT light chain in the sensory
neurons to completely block sensory inputs. Another method
used by Suster and Bate to remove sensory input involves a
senseless mutant. However, they pointed out that a significant
number of MD-type sensory neurons persisted in the senseless
mutant (13). Therefore, this experiment suffers from the same
caveat, i.e., the failure to completely remove sensory inputs. The
importance of the sensory input in the generation of larval
locomotion may not be fully appreciated in studies involving
incomplete removal of sensory inputs.
Historically, there have been two main hypotheses about the
trol and central control (1). The first hypothesis emphasizes that
rhythmic patterns are achieved through the use of sensory inputs
from the moving parts of the body and that loss of the normal
theory holds that CPGs within the CNS are intrinsically capable of
providing the proper timing of muscle activation without sensory
inputs. It seems likely that both central and sensory contributions
are important for the production of normal rhythmic behavior (4,
a modulating or coordinating role (13). This conclusion mainly
relies on two arguments: (i) There is no unequivocal experimental
data to support the hypothesis that peripheral sensory feedback is
necessary for the generation of properly timed rhythmic motor
outputs. (ii) Isolated, deafferented nervous systems are still able to
generate properly timed rhythmic output in the absence of periph-
eral feedbacks. However, both of those arguments have their
limitations when considering whole-animal behavior in an intact
Because the role of sensory inputs in rhythmic behavior has
often been assessed based on surgical operations to remove
sensory inputs, it raises such concerns about whether all sensory
inputs are blocked or whether motor outputs remain intact after
surgical operation (1), which might be the reason why different
laboratories achieved different results with the same protocols
(24–27). It is also difficult to study animal behavior following the
surgical operation. By using a genetics approach in an intact
animal, we can overcome this uncertainty and show that sensory
inputs are indispensable in generating rhythmic behavior.
The CPG was originally defined as a neural circuit that can
produce a repetitive, rhythmic neural output (1, 28, 29) (this is
the definition we used in this study), but it has been widely used
to refer to a neural circuit that is capable of generating a
rhythmic behavior. In reality, the majority of rhythmic behaviors
are multisegmented movements, such as walking, typing, and in
a simpler organism, larval crawling, all of which require a well
coordinated spatiotemporal sequence of muscle activations in
different body segments. For example, walking is based on the
coordinated movement sequences of all muscles on the legs, and
it has been proposed to be comprised of multiple CPGs, each of
which controls a single muscle of the leg (6). Therefore, for
multisegmented behavior, coordinating the temporal sequence
of individual CPGs should be necessary for the generation of
Our findings identify an essential role for MD neurons of the
Drosophila larval peripheral nervous system in rhythmic loco-
motion. Whereas the physiological function of MD neurons is
poorly understood, it is clear that MD neurons spread their
dendrites along the epidermis and tile the internal epithelial
surface of the entire larval body wall (30). A previous study has
demonstrated that the MD neuron counterparts in Manduca
larvae respond to the depression and stretch of the body wall
(31). In adult tsetse flies, the ventral cluster of MD neurons is
sensitive to body wall distention during blood feeding (32).
Similar to Drosophila larvae, the leech also contains segmentally
repeated MD neurons innervating its body wall. Interestingly,
these neurons hyperpolarize in response to body wall stretching
and depolarize upon the release of body wall stretching (33). The
implication that MD neurons in these animals serve a proprio-
ceptive function to sense the stretch of the animal body wall is
consistent with our hypothesis that MD neuron in Drosophila
larvae could sense the contraction and the release of the
contraction of each segment, and this information is subse-
quently sent back to the CNS and used to construct a chain of
reflex action that underlies peristaltic locomotion.
In vertebrates, human patients who lose somesthetic afferent
input as a result of acquired large-fiber sensory neuropathy are
consequently unable to control their movement and posture (34,
35). However, it is still unclear how loss of sensory inputs deprive
movement ability. Here, by using a genetic approach to silence
sensory neurons in intact animals, we created a similar dysfunc-
tional movement phenotype in Drosophila larvae. This and
similar combinations of behavior, electrophysiology, and genetic
approaches should make it feasible to better decipher the role of
sensory feedback in the generation of rhythmic behavior.
Materials and Methods
Fly Stocks. Flies were kept on standard media at 25°C. Gal4
drivers used for screening were provided by U. Heberlein
(University of California, San Francisco). The UAS-shitsflies
were acquired from T. Kitamoto (University of Iowa, Iowa City,
IA) (14). 21-7-Gal4 was provided by H. Lee and Y. Xiang (both
at the University of California, San Francisco). At third instar,
21-7-Gal4 drives gene expression exclusively in all MD sensory
neurons. dcirl-Gal4 was generated by cloning 1–1,018 base pairs
upstream of the dcirl translational start into pPTGAL vector. At
the third instar, dcirl-Gal4 drives gene expression in all type I
sensory neurons and in some interneurons in the ventral nerve
Larval Locomotion Analysis. Third-instar foraging stage larvae
were plated on grape agar to allow larvae to crawl out of food
medium. Clean larvae were then selected and placed on 2% agar
plates. These larvae were placed in a 37°C incubator for a given
time period (VWR Scientific, West Chester, PA). Their loco-
motion behavior was subsequently videotaped under high-
magnification by using a digital camcorder (3CCD Handycam;
Sony, Tokyo, Japan) attached to a microscope (MZ FLIII; Leica,
maintained for heat-shocked larvae by placing the agar plates on
a 37°C heat block (Sybron, Portsmouth, NH). Larvae were
videotaped for up to 60 sec at one frame per 0.033 sec.
Song et al.
March 20, 2007 ?
vol. 104 ?
no. 12 ?