direct observation of on and off pathways in the drosophila visual system

Please cite this article in press as: Strother et al., Direct Observation of ON and OFF Pathways in the Drosophila Visual System, Cur-rent Biology (2014), http://dx.doi.org/10.1016/j.cub.2014.03.017

Direct Observation of ON an

Current Biology 24, 1–8, May 5, 2014 ª2014 Elsevier Ltd All rights reserved http://dx.doi.org/10.1016/j.cub.2014.03.017

Reportd OFF

Pathways in the Drosophila Visual System

James A. Strother,1 Aljoscha Nern,1 and Michael B. Reiser1,*1Janelia Farm Research Campus, Howard Hughes MedicalInstitute, 19700 Helix Drive, Ashburn, VA 20147-2408, USA

Summary

Visual motion perception is critical to many animal behav-

iors, and flies have emerged as a powerful model system forexploring this fundamental neural computation. Although

numerous studies have suggested that fly motion visionis governed by a simple neural circuit [1–3], the implemen-

tation of this circuit has remained mysterious for decades.Connectomics and neurogenetics have produced a surge

in recent progress, and several studies have shown selec-

tivity for light increments (ON) or decrements (OFF) in keyelements associated with this circuit [4–7]. However, related

studies have reached disparate conclusions about wherethis selectivity emerges and whether it plays a major role

in motion vision [8–13]. To address these questions, weexamined activity in the neuropil thought to be responsible

for visual motion detection, the medulla, of Drosophilamelanogaster in response to a range of visual stimuli using

two-photon calcium imaging. We confirmed that the inputneurons of the medulla, the LMCs, are not responsible

for light-on and light-off selectivity. We then examined thepan-neural response of medulla neurons and found promi-

nent selectivity for light-on and light-off in layers of the me-dulla associated with two anatomically derived pathways

(L1/L2 associated) [14, 15]. We next examined the activityof prominent interneurons within each pathway (Mi1 and

Tm1) and found that these neurons have correspondingselectivity for light-on or light-off. These results provide

direct evidence that motion is computed in parallel light-on and light-off pathways, demonstrate that this selectivity

emerges in neurons immediately downstream of the LMCs,and specify where crucial elements of motion computation

occur.

Results and Discussion

We used two-photon imaging of a fluorescent calcium indica-tor to examine the activity of neurons within the fly visual sys-tem (Figure 1A). Flies have proven to be an excellent systemfor studying visual processing [1–3, 16–18], and broad conser-vation in the neuroanatomy of the visual system across adiversity of arthropods [19–21] suggests that a deeper under-standing of fly vision will produce widely generalizable con-clusions. Beneath the fly retina, the visual system consistsof four ganglia called the lamina, medulla, lobula, and lobulaplate (Figure 1B). The lamina is the first layer of the visual sys-tem and contains the primary synaptic targets of the photore-ceptors. Motion detection is presumed to occur within themedulla, because output neurons of the medulla show motionsensitivity [7]. We measured the calcium activity of the neu-rons that relay information from the lamina into the medulla,

*Correspondence: [email protected]

the lamina monopolar cells (LMCs; L1, L2, L3, and L4), andcompared this to the cumulative activity of all neurons withineach layer of the medulla neuropil. We made use of the GAL4/UAS system to express the genetically encoded calcium indi-cator GCaMP5G [22] in either the LMCs L1–L4 or pan neuro-nally (details of the fly genotypes used are given in Table S1available online, images of the driver lines are in Figure S1,and further details are in Supplemental Experimental Proce-dures). In vivo two-photon imaging of indicator fluorescenceprovided a relatively noninvasive measure of population-levelneuronal activity in response to visual stimuli presented on anovel projector-based display (Figure 1A). The recorded cal-cium response to the appearance of a small black disc revealsthree critical features of our results: (1) the responses arelocalized to a few medulla columns, which is consistent withthe retinotopic organization of the medulla; (2) the layeredstructure of the medulla is readily discernible in these re-sponses; and (3) light-on and light-off produce characteristicresponses that vary between layers of the medulla (Figures 1Cand 1D; Movie S1).

Pan-Neuronal Calcium Imaging Reveals that Light-On and

Light-Off Responses Are Segregated by Medulla LayerTo examine light-on and light-off responses within the me-dulla, we presented flies with a visual stimulus that consistedof a disc that alternated between bright and dark on an inter-mediate-intensity background (Figure 2A). A brief flickeringprotocol was first used to survey the medulla for a localizedcalcium response, which determined the imaging regionused for subsequent experiments. In all layers of the medullathat contain LMC processes, the LMCs showed decreasedactivity in response to light increments and increased activityin response to light decrements (Figures 2B, 2C, and S2A).Although the calcium indicator was simultaneously expressedin LMCs L1–L4, the spatial separation of the LMC arborizations(Figure S1B) made it possible to assign the activity measuredin each layer to specific cell types (Figure 2C; M1: L1; M2:L2+L4; M3: L3; M4: L4; and M5: L1). In all five layers, activitywas dominated by a tonic decrease in response to light-onand a tonic increase in response to light-off. However, a phasicresponse to light decrements but not light increments wasobserved in layer M2 (likely from L2 given the absence of asimilar response inM4), andmodest but statistically significantdifferences in the magnitude of the response to light incre-ments and decrements could be discerned in all layers, exceptin layer M5 (Figure S2A). These results are consistent with pre-vious calcium imaging studies in which L1, L2, L3, and L4 wereall found to show increased activity in response to light decre-ments and decreased activity in response to light increments,with slight-to-moderate selectivity depending on the specificstimulus conditions [5, 8, 10, 12, 23].In a parallel set of experiments, we examined the summed

response of all (or nearly all) medulla neurons to an identicalstimulus using a pan-neuronal driver (R57C10) [24]. This driverwas previously shown to drive expression in all knowncolumnar medulla neuron types using stochastic single-celllabeling experiments [15] (A.N., unpublished data), and doublelabeling with an antibody [25] for an established pan-neuronalmarker [26] suggested that it is expressed in all columnar and

http://dx.doi.org/10.1016/j.cub.2014.03.017


mailto:[email protected]

Retina

Lamina

Medulla

Lobula

Lobula Plate

Lamina monopolar cells

Medulla

Raw Fluorescence Δ Fluorescence

M1M2M3M4M5 10μm

Distal

Proximal

Medulla

Pan-neuronal

Raw Fluorescence

M1M2M3M4M5M6M8M9M10

Δ Fluorescence

10μmProximal

Distal

A B C

D

L1 L2 L3 L4

Figure 1. The Layered Structure of the Drosophila Medulla Is Revealed by Two-Photon Calcium Imaging

(A) Medulla neurons expressing the fluorescent calcium indicator GCaMP5G were imaged in head-fixed flies using two-photon microscopy. Visual stimuli

were presented using a high-speed projector system.

(B) In this schematic of the fly visual system, representative columnar neuron types are drawn with the position of the cell body indicated by a disc and the

main arborizations indicated by short bars (photoreceptors in red, lamina monopolar cells in blue, medulla intrinsic neurons and transmedullary neurons in

green, and T cells in orange; depicted neurons are R1–R6, R7, L1, L2, L3, L4, Mi1, Mi4, Tm1, Tm3, T3, T4, and T5). Each of these cell types is present within

every column of the fly visual system.

(C) Responses to visual stimuli were recorded in the medulla of flies that simultaneously express the calcium indicator in the lamina monopolar cells L1, L2,

L3, and L4 (detailed in Figures S1A and S1B). The medulla has a layered structure, and the schematic on the left indicates the approximate position of each

layer and the layers in which each neuron type arborizes. The calcium response to a 5� disc changing from a high to low light intensity is displayed in the

center image, with white arrows indicating the position of each layer. The retinotopic nature of the response to the visual stimulus is accentuated by

subtracting the response just prior to the change in the visual stimulus and is displayed in the right image (mean response of 1 s after light change minus

mean response of 1 s before change).

(D) Responses to visual stimuli were also recorded in themedulla of flies that simultaneously express the calcium indicator in all neurons (detailed in Figures

S1C and S1D). All results are displayed as in (C), although here all layers of the medulla are shown rather than just the first five.

See also Figure S1 and Movie S1.

Current Biology Vol 24 No 92


tangential medulla neurons (Figure S1D). In contrast to theLMCs, the pan-neuronal activity displayed a complex spatio-temporal structure in response to the flickering disc stimulus(Figures 2B, 2C, and S2B). The pan-neuronal responses inmost layers were strongly rectified, that is, the response tolight-on did not resemble the inverse of the response tolight-off. This rectification is indicative of nonlinear processingthat would result in selectivity for either light-on or light-off sig-nals. Increased activity in response to light increments wasobserved in layers M1, M6, M8/9, and M10. Increased activityin response to light decrements was observed in M1, M2, M3,and M8/9 (Figure S2B). For pan-neuronal images, layers M4/5and M8/9 could not be readily distinguished based on eitheranatomy or activity and were treated as aggregates. In alllayers, the magnitude of the response to light incrementsand decrements differed (Figure S2B). Finally, the LMCresponse was found to be significantly less rectified than thepan-neuronal response in each layer (Figure S2C). Althoughthe pan-neuronal response represents the summed activityof many cell types, this alone cannot explain the rectification,because the linear sum of multiple unrectified responsesremains unrectified.

The activity of the LMCs is dominated by a tonic, unrectifiedresponse to light changes (Figures 2B and 2C, top), whereasthe pan-neuronal activity shows substantial phasic, rectifiedresponses to both light increments and decrements (Figures2B and 2C, bottom). This difference, summarized in the statis-tical results of Figures S2A–S2C, strongly suggests that therectification of medulla responses to light-on and light-off

stimuli occurs downstream of the LMCs. The conclusion thatthe rectification occurs postsynaptic to the LMCs, and is nota difference between the intrinsic activity of L1 and L2, is sup-ported by previous studies in which L1 and L2 were found torespond to both light increments and decrements [5, 8] (butsee [10], which reaches a different conclusion).Can the complex spatiotemporal structure of the pan-

neuronal response be decomposed into simpler, constituentresponses, or is this complexity a necessary consequence ofmeasuring the summed activity of a diverse neuronal popu-lation? To further explore this question, we performed aprincipal component analysis (PCA) of the responses to theflickering disc stimulus. The PCAwas performed using individ-ual pixels from both LMC and pan-neuronal data sets as vari-ates and stimulus-synchronized time points as observations.Consequently, the obtained principal components utilizedall available spatial (and temporal) information and simulta-neously described both the LMC and pan-neuronal datasets. The first three principal components were found to cap-ture a substantial fraction of the observed dynamics (Fig-ure 2D; LMC, 60% and pan-neuronal, 61% of variance ofeach pixel at each time point across all animals). The firstprincipal component captures a tonic, unrectified responseto changes in light intensity; the second component describesa phasic, rectified response to light increments (an onresponse); and the third component corresponds to a phasic,rectified response to light decrements (an off response).Composing a prediction of the pan-neuronal responsebased on the extracted principal components produces a

ON/OFF Pathways in the Drosophila Visual System3


spatiotemporal pattern (Figure 2E) that qualitatively capturesall of the features of the original averaged data set (Figure 2B).The spatial distribution of the signals in the second and thirdPCA components are strikingly similar to anatomical pathwaysthat were suggested by a study of quantitative correlations inthe neuronal arborization patterns of Golgi-impregnated me-dulla columnar neurons [14]. We have reproduced the sche-matic view of these pathways from this classic study andcompared them with our PCA results (Figures 2F and 2G).The signal energy in PC2 is very similar to the arborizationdensity for members of the ‘‘L1 pathway,’’ whereas the signalenergy in PC3 is well aligned with the arborization density formembers of the ‘‘L2 pathway.’’ One confound to interpretingthese responses as evidence of actual neuronal pathways isit assumes the stimulus we selected reflects generalizablephenomena. We have examined the response to movingedges of different polarities and have found that the temporalresponse is similar to the response observed for a flickeringdisc, suggesting that our primary stimulus is highly relevantfor examining components of the motion detection circuitry(Figures S2D–S2I).

The alignment of the spatial distribution of the observed re-sponses with the classic neuroanatomical evidence suggeststhat the light-on response reflects activity in members of the‘‘L1 pathway’’ and the light-off response reflects activity inmembers of the ‘‘L2 pathway.’’ This result agreeswith previousstudies that found that L1 and L2 blockade eliminates re-sponses to light-on and light-off motion, respectively, in lobulaplate tangential cells [4, 6] and walking behavioral assays [8,12]. This result is also in agreement with measurements ofselectivity for light-on motion by T4 neurons and selectivityfor light-off motion by T5 neurons [7], which have been impli-cated as the targets of the L1 and L2 pathways, respectively,based on connectivity [15]. There is broad agreement that L1and L2 are together required for motion detection [4, 8, 11,13]; however, several behavioral studies identified other spe-cific phenotypes upon inactivation. In one study, blocking L1reduced the response to back-to-front motion, and blockingL2 reduced the response to front-to-back motion [11]. Inanother study, separately blocking L1 or L2 had no effect onthe response to either light-on or light-off moving edges [13].Although anatomical evidence suggests that the L1 and L2pathways are quite dissimilar, there are prominent intercon-nections between these pathways, notably through the feed-forward neuron L5 and the feedback neurons C2 and C3 [15].Because of the presence of these interactions between thepathways, as well as clear evidence for at least one additionalpathway (associated with the LMC L3 [15]), there is no simpleprediction for how complete L1 or L2 blockage would beexpected to affect behavior. In contrast to these behavioralmethods that can only provide an indirect measure of neuronalfunction, our imaging results directly confirm that L1 and L2are the inputs to pathways with light-on and light-off selec-tivity, and we show that this selectivity emerges downstreamof the LMCs.

Proximal Layers of the Medulla Are Low-Pass FilteredRelative to Distal Layers

The classic model for fly motion detection, the Hassenstein-Reichardt correlator (HRC) [27, 28], posits that motion iscomputed from the coincident arrival of delayed and nonde-layed signals from neighboring locations on the fly eye. Neu-rons conveying the delayed signals are expected to respondslowly to changes in light intensity, whereas neurons carrying

the nondelayed signals are expected to respond more rapidly.Because the pan-neuronal responses reciprocated the selec-tivity for light-on and light-off motion proposed in severalmore recent models of motion computation [6, 29], we nextprobed the layer-specific responses to temporally varyingvisual stimuli. Flies were presented with a flickering disccentered over the receptive field of the imaged neurons, andthe disc was flickered at frequencies between 0.33 and243 Hz. In a related set of trials, we also presented flies withgrating patterns that moved at several different speeds (Fig-ures S3A–S3C). Because the kinetics of the calcium indicatorlimit our ability to observe the instantaneous fluctuationsin calcium concentration that might occur at higher flickerfrequencies [22], we used the time-averaged fluorescenceto characterize how the time-averaged neuronal activity isaffected by stimulus frequency. This approach was success-fully used in previous studies to describe the temporal filteringof lobula plate tangential cells [30] and is supported bymodelsof the indicator dynamics (Figures S3D–S3H).In the first five layers of the medulla, the time-averaged

LMC activity decreased gradually with flicker frequency andconverged to a minimum value for frequencies greater than81 Hz. At these high frequencies (well above the flicker fusionrate of Drosophila photoreceptors [31]), the time-averaged ac-tivity wemeasure is expected to converge to the response to afull-field intermediate-intensity stimulus. The LMC activity ineach of the M1–M4 layers was significantly affected by fre-quency (p < 0.001, one-way ANOVA, Bonferroni correction),whereas activity in M5 was not significantly affected by fre-quency (p > 0.05), and only layers M2 and M5 differed fromeach other (p < 0.01, two-way ANOVA, Bonferroni correction).The responses of L1 in M1 and M5 were not significantlydifferent (p > 0.05, two-way ANOVA, Bonferroni correction),and the additional variability in the response of M5 was attrib-uted to the weaker responses in that layer, as was alsoobserved in a previous study [8]. The LMC responses arelargely consistent with the results of previous electrophysio-logical studies that showed that fly LMCs rapidly hyperpolarizein response to light increments and depolarize in response tolight decrements [17] and remain responsive to oscillationsin light intensity at frequencies up to w30 Hz (3 dB corner fre-quency [16, 32, 33]). In contrast to previous electrophysiolog-ical studies [16, 32, 33], we did not observe any high-passfiltering in the LMC responses. The absence of high-passfiltering is consistent with the sustained responses observedfor slowly flickering discs (Figures 2A–2C) and similar toprevious calcium imaging experiments [5, 8], suggesting adifference between the intracellular calcium dynamics andmembrane potential dynamics.The time-averaged pan-neuronal activity in distal layers

M1–M6 was similar to that observed in the LMCs, and itdecreased gradually with flicker frequency and converged toa minimum level for frequencies greater than 81 Hz. However,the pan-neuronal activity in the proximal layers M8–M10 wasmarkedly different. The pan-neuronal activity in layersM8–M10 decayed rapidly for frequencies above 1 Hz andconverged to a minimum value for frequencies greater than9 Hz. No differences were observed between the pan-neuronal responses of M1–M6 (p > 0.05, two-way ANOVA,layer-frequency interaction), and no differences wereobserved between the responses of M8/9 and M10 (p >0.05), but the pooled responses of M1–M6 were significantlydifferent from the pooled responses of M8–M10 (p < 0.001).These conclusions are not limited to flickering stimuli because

10μm Raw Fluorescence

LMCs

Pan-

neur

onal

M1M2

M3M4M5

M1M2

M3M4M5M6

M8M9

M10

Distal

Proximal

Distal

150 %0

10s 10sTime

M1

M2

M3

M4

M5

M1

M2

M3

M8/9

M6

M10

M4/5

100%

5sTime

(L1)

LMC type

(L2+L4)

(L3)

(L1)

(L4)

10s 10sTime

10s 10sTime

10s 10sTime

10s 10sTime

PC1 PC2 PC3 Predicted Response

LMCs

Pan-

neur

onal

M1M2

M3M4M5

M1M2

M3M4M5

M6

M8M9

M10

M1M2

M3M4M5

M1M2

M3M4M5M6

M8M9

M10

Distal

Distal

Proximal

150 %0

+50 %-50%88% of LMCs45% of PanN

Percent ofExplainable

Variance LMCs > PanN* LMCs < PanN** LMCs < PanNns

5% of LMCs42% of PanN

7% of LMCs13% of PanN

D E

M1M2M3M4M5M6M8M9

M10Medulla Medulla

M1M2M3M4M5M6M8M9

M10

L1 L2 Tm1Mi1 L1 Mi1 L2Anatomical “L1 Pathway” Anatomical “L2 Pathway”PC2

EnergyPC3

Energy Tm1

F G

CBA

Figure 2. The Lamina Monopolar Cell and Pan-Neuronal Calcium Responses Reveal Anatomically Segregated Pathways for Processing of Light-On and

Light-Off

(A) Representative images of the LMC and pan-neuronal calcium responses to a flickering disc stimulus. The flickering disc was 30� in diameter, flickered

between high and low light levels at a frequency of 0.05 Hz, and was centered over the receptive field of the imaged neurons. The response to a light incre-

ment is shown in the images on the left, and the response to a light decrement is shown in the images on the right. The proximal and distal borders of the

(legend continued on next page)





we observe similar temporal filtering properties within bothLMC and pan-neuronal responses to moving square-wavegratings (Figures S3A–S3C).

The pan-neuronal response to flickering discs of differentfrequencies suggests that neurons with dense arborizationsin the distal layers of the medulla respond strongly to high-fre-quency stimuli, whereas neurons with dense arborizations inthe proximal layers of the medulla respond weakly to thesame high-frequency stimuli. Anatomical evidence indicatesthat many medulla columnar neurons receive inputs in thedistal layers and have outputs in the proximal layers [14].Together this suggests that substantial temporal filteringoccurs within the medulla columnar neurons, and responsesto high-frequency stimuli are severely attenuated at their out-puts. These results may shed additional light on the neuralcircuits that implement motion computation. In the simplestimplementation of the HRC, the temporal delay is a low-pass filter and correlation is realized as a multiplicationof the two signals (Figure 3E) [27]. Thus, this model predictsthat neurons carrying nondelayed signals will respond tostimuli over a wide range of frequencies, whereas neuronscarrying delayed signals will only respond to lower frequencystimuli, and that a multiplicative interaction will occur in aregion where they overlap. If we interpret our results in thiscontext, it would suggest that the multiplicative interactionis in the distal layers of the medulla, in areas that receiveboth delayed and nondelayed signals. However, if this werethe case then we would also expect to observe motion selec-tivity in either or both the distal or proximal layers, which wedid not (Figures S2D–S2I and S3A–S3C; although spatial aver-aging across columns may obscure this motion selectivity). Itis likely that a literal application of the simple HRC model tothe medulla is not possible; both the temporal filtering ofthe nondelayed and delayed arms, as well as the nonlinearitymay not be as simple as we have assumed. Nevertheless, ourresults do show that the temporal filtering that is critical formotion detection does occur between the input and outputlayers of the medulla.

medulla are indicated by dashed white lines, and the position of each layer of

layer).

(B) Time history of calcium responses to the same flickering disc stimulus. Coll

the mean response of each row of pixels, for each fly, was examined as a func

normalized by the average fluorescence over the experimental protocol (motiv

Results shown are the median values from multiple individuals (LMC, n = 8; pan

and correspond approximately to the positions of (A), whereas time points are

axis indicate the periods during which the disc was on and off, respectively.

(C) Time series for calcium responses at selected positions within the medulla.

lines below each curve represent a F=F value of zero. The white and gray back

tively. For LMC images, each layer was readily identifiable and is shown separ

distinguished based on either anatomy or activity and consequently are labele

LMC responses are less rectified than pan-neuronal responses (see Figure S2

(D) Principal components of LMC and pan-neuronal calcium responses to a flic

pixels of all trials as variates and each time point as an observation, such that th

neuronal data sets. The contribution of each principal component to the predic

within the medulla are given on the vertical axis (e.g., M1 is the first layer), where

on the horizontal axis indicate the periods during which the disc was on and o

each principal component is presented below each panel, and significant d

sets are indicated with asterisks (two-sided Mann-Whitney U test, *p < 0.05, *

(E) Predicted response to flickering disc stimulus based on first three principa

(F) Schematic of the ‘‘L1 pathway.’’ Several processing pathways in the optic lob

correlated arborization patterns. The layers involved in the ‘‘L1 pathway’’ propo

schematic. The representative neurons of Figure 1B are reproduced here, and t

schematic and shown as Golgi stains. The mean signal energy in PC2 is show

(G) Schematic of ‘‘L2 pathway.’’ The displayed schematic indicates the layer

includes the L2 and Tm1 neurons. The mean signal energy in PC3 is shown to

See also Figure S2.

Responses of Individual Neuron Types Central to the L1and L2 Pathways Reciprocate Pan-Neuronal Results

Do the principal components of Figure 2 represent the activityof actual neurons? Because the spatial distribution of the pan-neuronal responses closely matched the anatomical L1 and L2pathways, we next examined the activity of key neurontypes within each of these pathways. We selected the medullacolumnar interneuron Mi1 as a representative of the L1pathway and the transmedullary columnar neuron Tm1 as arepresentative of the L2 pathway. Both of these cell typeshave archetypal morphology [14] (Figures 2F and 2G), andeach receives the largest number of synapses from theirrespective LMCs [15]. As with the LMC and pan-neuronalstudies, we expressed the fluorescent calcium indicatorGCaMP5G using the GAL4/UAS expression system andimaged the calcium responses using two-photon microscopy(see Supplemental Experimental Procedures).The Mi1 cell type is characterized by dense arborizations

in medulla layers M1, M5, and M9–M10 ([18], Figures 4A andS4A). When presented with a flickering disc stimulus, theresponse of Mi1 neurons in layer M10 closely mirrored thepan-neuronal response in M10 (Figure 4B). Activity in Mi1increased in response to light increments, decayed to baselinelevels after several seconds, and showed a minimal responseto light decrements. Similarly, the response of Tm1 to a flick-ering disc stimulus closely matched the pan-neuronal light-off response in M8/9 (Tm1 has dense arborizations in M2–M3and M9 [18]; Figures 4A and S4B). Activity in Tm1 increasedin response to light decrements, decayed to baseline levelsafter several seconds, and showed no response to light incre-ments (Figure 4B). The responses of Mi1 and Tm1 appearto be primarily driven by the luminance change, since the re-sponses to a set of motion stimuli all show similar responses(Figures S4C–S4F). The activity of Mi1 and Tm1 in the moreproximal layers of the medulla was comparable to the re-sponses in the deeper layers; however, the responses in layersM9 and M10 were often larger and so were used for thisanalysis.

the medulla is indicated along the left edge the images (e.g., M1 is the first

ected images were rotated and scaled to a common coordinate system, and

tion of time. Activity is presented as F=F, or the instantaneous fluorescence

ation for this metric is provided in Supplemental Experimental Procedures).

-neuronal, n = 7). Positions within the medulla are given on the vertical axis

displayed on the horizontal axis. The white and black bars on the horizontal

Results are calculated as in (B), error bars indicate the SEM, and solid black

grounds indicate the periods during which the disc was on and off, respec-

ately. For pan-neuronal images, layers M4/5 and M8/9 could not be readily

d as aggregates. Statistical analysis of these time series demonstrates that

C).

kering disc stimulus. The principal components analysis used the individual

e calculated components simultaneously represent both the LMC and pan-

ted F=F is shown as a function of position in the medulla and time. Positions

as time points are displayed on the horizontal axis. The white and black bars

ff, respectively. The percentage of the explainable variance contributed by

ifferences in these percentages between the LMC and pan-neuronal data

*p < 0.01; ns, not significant).

l components. Compare to (B).

e have been previously proposed based on the identification of neuronswith

sed by Bausenwein et al. [14] are indicated with gray shading in the medulla

wo central neurons belonging to the pathway, L1 andMi1, are marked on the

n to the right of the schematic as a function of position.

s involved in the ‘‘L2 pathway’’ proposed in Bausenwein et al. [14], which

the right of the schematic as a function of position.

0.33 Hz 9 Hz

Raw Fluorescence10μm

LMCs

Pan-

neur

onal

M1M2

M3M4M5

M1M2

M3M4M5M6

M8M9

M10

Flicker Frequency (Hz)

Distal

Proximal

Distal

0.33 1 3 9 27 81 243Flicker Frequency (Hz)

150 %0

0.33 Hz

243 Hzto

0.33 Hz

243 Hzto

M1

M2

M3

M4

M5

M1

M2

M3

M8/9

M6

M10

M4/5

150%0%

Scaled

f (Hz) f (Hz)

A B C D E

Predelay Postdelay

Figure 3. The Calcium Responses in Output Layers of the Medulla Are Filtered Relative to Responses in the Input Layers

(A) Representative images of the LMC and pan-neuronal calcium responses to a disc flickering at slow (0.33 Hz) and fast (9 Hz) frequencies. The flickering

disc was 30� in diameter, flickered between high and low light levels, and was centered over the receptive field of the imaged neurons. The displayed image

is themean fluorescence during the stimulus, averaged over multiple flicker periods. The proximal and distal borders of themedulla are indicated by dashed

white lines, and the position of each layer of the medulla is indicated along the left edge the images.

(B) Calcium responses to the same flickering disc stimulus over a range of frequencies. Displayed values are mean F=F during the stimulus, averaged over

multiple flicker periods, and represent the median value frommultiple individuals (LMC, n = 8; pan-neuronal, n = 7). Positions within the medulla are given on

the vertical axis and correspond approximately to the positions of (A).

(C) Frequency response of F=F for selected positions within the medulla. Results are calculated as in (B), error bars indicate the SEM, and solid black lines

below each curve represent a F=F value of zero. As in previous figures, for the pan-neuronal data set M4/5 and M8/9 are labeled as aggregates.

(D) Scaled frequency response of calcium responses for selected positions. To enable comparisons of the shape of the frequency response curves, results

from (C) were scaled so that the maximum and minimum values have the same vertical positions on all subpanels.

(E) Schematic of Hassenstein-Reichardt correlator (HRC). The cups represent photoreceptor inputs. Red lines indicate signals prior to the temporal delay

element, and blue lines indicate signals after the delay element. The frequency response of the predelay signals extends to high frequencies (where it is

limited by phototransduction and LMCprocessing), whereas the frequency response of the postdelay signals is low-pass filtered, as indicated by schematic

frequency response curves below the diagram.

See also Figure S3 and Movie S2.



The response of Mi1 to flickering discs of different fre-quencies closely followed the pan-neuronal response inM10 (Figure 3; reproduced in Figure 4). Activity in Mi1decreased rapidly for flicker frequencies greater than 1 Hzand converged to a minimum value for frequencies above9 Hz. The response of Mi1 at different flicker frequencieswas not significantly different from the pooled pan-neuronalM8–M10 responses (two-way ANOVA, interaction effect, p >0.05) but was significantly different from the pooled M1–M6responses (p < 0.001). The Tm1 response to flickering discsalso resembled the filtered pan-neuronal response in layerM8/9 (Figure 4). Activity in Tm1 decreased rapidly for flickerfrequencies greater than 1 Hz and converged to a minimumvalue for frequencies above 9 Hz. However, the response ofTm1 to different flicker frequencies was significantly differentfrom both the pooled pan-neuronal M8–M10 responses(two-way ANOVA, interaction effect, p < 0.001) and the pooledM1–M6 responses (p < 0.001). Similar temporal filtering wasobserved in the responses to moving square-wave gratings,although these responses showed somewhat greater vari-ability (Figures S4G and S4H). We attribute the smaller re-sponses measured from Tm1 (Figures 4B and 4C) to stimulussize selectivity we observe in this cell type (but is apparently

absent in Mi1; Figure S4I). The response of Tm1 to a 30�

disc (used to enable comparison to the pan-neuronal dataset) is substantially reduced from the response to a smallerdisc, suggesting a prominent contribution from an inhibitorysurround (Figure S4J).These results support the conclusions drawn from the

pan-neuronal activity. We observed selectivity for light-on ina prominent interneuron of the L1 pathway and selectivity forlight-off in a prominent interneuron of the L2 pathway. Eachof these interneurons is predominantly postsynaptic to theirrespective LMC, suggesting that this selectivity arises immedi-ately downstream of the LMCs. The similarity between theresponses of these two interneurons and the pan-neuronalresponses could be interpreted as evidence that they alonedominate the pan-neuronal activity. However, it is more likelythat these neurons are only examples of a broader functionalsegregation and that each pathway contains multiple neurontypes with similar properties. This interpretation is consistentwith the recent identification of selectivity for light-off in theTm2 neuron [23], which has connectivity similar to Tm1 andis associated with the L2 pathway [15, 23]. Rectified responsesin other cell types may explain some differences between thearborization density of these neurons and the anatomical

10μm Raw Fluorescence

Pan-

neur

onal

Tm1

Mi1

M1

M2

M5

M9M8

M2M5

M1

M9

M10

M9M10

M8/9

M10

Mi1

100%

0%

Tm1

TemporalResponse

10s 10sTime

150%

0%

FrequencyResponse

Max

Min

ScaledFrequencyResponse

0.33Frequency (Hz)

243 0.33Frequency (Hz)

243

A B C D

Figure 4. Individual Neuron Types within the L1 and L2 Pathways Show

Rectification and Temporal Filtering

(A) Representative images of calcium responses to a flickering disc stim-

ulus. The response to a flickering disc stimulus was examined in flies that

specifically express a calcium indicator in either Mi1 or Tm1 neurons (see

Figures S4A and S4B). The stimulus protocol and analysis of images was

identical to that performed for LMC and pan-neuronal imaging (see Figures

2 and 3). Images represent the response of Mi1 to a light decrement and of

Tm1 to a light increment. The pan-neuronal response to a light increment is

reproduced from Figure 2A for comparison.

(B) Time history of Mi1, Tm1, and pan-neuronal calcium responses to the

same flickering disc stimulus. Results shown are the median values from

multiple individuals (Mi1, n = 8; Tm1, n = 7, pan-neuronal reproduced from

Figure 2C), error bars indicate the SEM, and solid black lines below each

curve represent a F=F value of zero. The white and black bars on the

horizontal axis indicate the periods during which the disc was on and off,

respectively.

(C) Frequency response of Mi1, Tm1, and pan-neuronal F=F to flickering

disc stimulus. Results shown are the mean F=F during the stimulus period,

averaged over multiple flicker periods. Presented values represent the

median values from multiple individuals (Mi1, n = 8; Tm1, n = 7), and error

bars indicate the SEM.

(D) Scaled frequency response of Mi1, Tm1, and pan-neuronal calcium

responses. To enable comparisons of the shape of the frequency response

curves, results from (C) were scaled so that the minimum and maximum

values have the same vertical positions on all subpanels.

See also Figure S4.



organization of the pathways extracted from the pan-neuronalresponses (Figure 2G, notably in layer M3).

Conclusions

We imaged neural activity in the fly medulla in response to avariety of visual stimuli. We found that selectivity for light-onor light-off was not present in the inputs to the medulla (theLMCs) but was present in the pan-neuronal activity of the me-dulla. The spatial distribution of the pan-neuronal light-on andlight-off responses closely coincided with the L1 and L2 path-ways, respectively, that were previously derived from anatom-ical studies. We then imaged prominent interneurons withinthe L1 and L2 pathways and found corresponding selectivity.These results are strong evidence for the emerging hypothe-ses that the light-on and light-off responses are mediatedby the L1 and L2 pathways [4, 7, 8, 15], respectively, andthat the rectification that characterizes these responsesemerges downstream of the LMCs. In addition, we found

that a remarkably large fraction of the total activity in the me-dulla could be attributed to activity in just these two pathways.If the L1 and L2 pathways are principally involved in motioncomputation, then this result suggests that motion computa-tion accounts for a large fraction of the total activity in themedulla as evoked by our stimuli. An alternative interpretationis that the L1 and L2 pathways represent a more generalfunctional segregation for light-on and light-off selectivitywithin the medulla, of which motion computation is a specificexample. Furthermore, we found that the input layers of themedulla were significantly more responsive to higher flicker(and motion) frequencies than the output layers, indicatingthat high-frequency temporal information is lost by processingwithin the medulla. This transformation may reflect the steppreceding motion detection, or it might be a signature ofthe very heart of motion computation. Finally, this study usespan-neuronal imaging within a large region of the fly brain tocharacterize stimulus feature selectivity (Figures 2A–2C), andit uses a combination of functional and anatomical insightsto identify specific cell types associated with this selectivity(Figures 2D–2G and 4). We expect that this approach willbecome increasingly powerful as advances in electron micro-scopy methods continue to reveal the connectivity of the flybrain and intersectional genetic techniques increasinglyenable targeting of any cell type of interest.

Supplemental Information

Supplemental Information includes Supplemental Experimental Proce-

dures, four figures, one table, and two movies and can be found with this

article online at http://dx.doi.org/10.1016/j.cub.2014.03.017.

Acknowledgments

We thank the Janelia Fly Light Project Team for some confocal images, the

Developmental Studies Hybridoma Bank for antibodies, and Emily Willis for

assistance with artwork. A.N. thanks Gerry Rubin, in whose lab he per-

formed this work, for his support and encouragement. We also thank Gerry

Rubin, Vivek Jayaraman, Stephen Huston, and three anonymous reviewers

for comments on the manuscript. This project was supported by HHMI.

Received: February 3, 2014

Revised: February 28, 2014

Accepted: March 6, 2014

Published: April 3, 2014

References

1. Hassenstein, V.B., and Reichardt, W. (1956). Systemtheoretische

Analyse der Zeit-, Reihenfolgen- und Vorzeichenauswertung bei der

Bewegungsperzeption des Russelkafers Chlorophanus. Z. Naturforsch.

B 11, 513–524.

2. Egelhaaf, M., and Borst, A. (1989). Transient and steady-state response

properties of movement detectors. J. Opt. Soc. Am. A 6, 116–127.

3. Gotz, K.G. (1964). Optomotorische Untersuchung des visuellen Systems

einiger Augenmutanten der Fruchtfliege Drosophila [Optomoter studies

of the visual system of several eye mutants of the fruit fly Drosophila].

Kybernetik 2, 77–92.

4. Joesch, M., Schnell, B., Raghu, S.V., Reiff, D.F., and Borst, A. (2010). ON

and OFF pathways in Drosophila motion vision. Nature 468, 300–304.

5. Freifeld, L., Clark, D.A., Schnitzer, M.J., Horowitz, M.A., and Clandinin,

T.R. (2013). GABAergic lateral interactions tune the early stages of visual

processing in Drosophila. Neuron 78, 1075–1089.

6. Joesch, M., Weber, F., Eichner, H., and Borst, A. (2013). Functional

specialization of parallel motion detection circuits in the fly. J. Neurosci.

33, 902–905.

7. Maisak, M.S., Haag, J., Ammer, G., Serbe, E., Meier, M., Leonhardt, A.,

Schilling, T., Bahl, A., Rubin, G.M., Nern, A., et al. (2013). A directional

tuning map of Drosophila elementary motion detectors. Nature 500,

212–216.




8. Clark, D.A., Bursztyn, L., Horowitz, M.A., Schnitzer, M.J., and Clandinin,

T.R. (2011). Defining the computational structure of the motion detector

in Drosophila. Neuron 70, 1165–1177.

9. Egelhaaf, M., and Borst, A. (1992). Are there separate ON and OFF

channels in fly motion vision? Vis. Neurosci. 8, 151–164.

10. Reiff, D.F., Plett, J., Mank, M., Griesbeck, O., and Borst, A. (2010).

Visualizing retinotopic half-wave rectified input to the motion detection

circuitry of Drosophila. Nat. Neurosci. 13, 973–978.

11. Rister, J., Pauls, D., Schnell, B., Ting, C.Y., Lee, C.H., Sinakevitch, I.,

Morante, J., Strausfeld, N.J., Ito, K., and Heisenberg, M. (2007).

Dissection of the peripheral motion channel in the visual system of

Drosophila melanogaster. Neuron 56, 155–170.

12. Silies,M., Gohl, D.M., Fisher, Y.E., Freifeld, L., Clark, D.A., andClandinin,

T.R. (2013). Modular use of peripheral input channels tunes motion-

detecting circuitry. Neuron 79, 111–127.

13. Tuthill, J.C., Nern, A., Holtz, S.L., Rubin, G.M., and Reiser, M.B. (2013).

Contributions of the 12 neuron classes in the fly lamina to motion vision.

Neuron 79, 128–140.

14. Bausenwein, B., Dittrich, A.P., and Fischbach, K.F. (1992). The optic

lobe of Drosophila melanogaster. II. Sorting of retinotopic pathways in

the medulla. Cell Tissue Res. 267, 17–28.

15. Takemura, S.Y., Bharioke, A., Lu, Z., Nern, A., Vitaladevuni, S., Rivlin,

P.K., Katz, W.T., Olbris, D.J., Plaza, S.M., Winston, P., et al. (2013). A

visual motion detection circuit suggested by Drosophila connectomics.

Nature 500, 175–181.

16. Jarvilehto, M., and Zettler, F. (1973). Electrophysiological-histological

studies on some functional properties of visual cells and second order

neurons of an insect retina. Z. Zellforsch. Mikrosk. Anat. 136, 291–306.

17. Laughlin, S.B., and Hardie, R.C. (1978). Common strategies for light

adaptation in the peripheral visual systems of fly and dragonfly.

J. Comp. Physiol. 128, 319–340.

18. Fischbach, K.-F., and Dittrich, A. (1989). The optic lobe of Drosophila

melanogaster. I. A Golgi analysis of wild-type structure. Cell Tissue

Res. 258, 441–475.

19. Osorio, D., and Bacon, J.P. (1994). A good eye for arthropod evolution.

Bioessays 16, 419–424.

20. Strausfeld, N.J., and Nassel, D.R. (1981). Neuroarchitecture of brain

regions that subserve the compound eyes of Crustacea and Insects. In

Handbook of Sensory Physiology, Volume VII/6B, H. Autrum, ed.

(New York: Springer-Verlag).

21. Beron de Astrada, M., and Tomsic, D. (2002). Physiology and

morphology of visual movement detector neurons in a crab

(Decapoda: Brachyura). J. Comp. Physiol. A Neuroethol. Sens. Neural

Behav. Physiol. 188, 539–551.

22. Akerboom, J., Chen, T.W., Wardill, T.J., Tian, L., Marvin, J.S., Mutlu, S.,

Calderon, N.C., Esposti, F., Borghuis, B.G., Sun, X.R., et al. (2012).

Optimization of a GCaMP calcium indicator for neural activity imaging.

J. Neurosci. 32, 13819–13840.

23. Meier, M., Serbe, E., Maisak, M.S., Haag, J., Dickson, B.J., and Borst, A.

(2014). Neural Circuit Components of the Drosophila OFF Motion Vision

Pathway. Curr. Biol. 24, 385–392.

24. Jenett, A., Rubin, G.M., Ngo, T.T., Shepherd, D., Murphy, C., Dionne, H.,

Pfeiffer, B.D., Cavallaro, A., Hall, D., Jeter, J., et al. (2012). A GAL4-driver

line resource for Drosophila neurobiology. Cell Rep. 2, 991–1001.

25. O’Neill, E.M., Rebay, I., Tjian, R., andRubin, G.M. (1994). The activities of

two Ets-related transcription factors required for Drosophila eye devel-

opment are modulated by the Ras/MAPK pathway. Cell 78, 137–147.

26. Campos, A.R., Rosen, D.R., Robinow, S.N., and White, K. (1987).

Molecular analysis of the locus elav in Drosophila melanogaster:

a gene whose embryonic expression is neural specific. EMBO J. 6,

425–431.

27. Reichardt, W. (1961). Autocorrelation, a principle for the evaluation

of sensory information by the central nervous system. In Sensory

Communication, W.A. Rosenblith, ed. (Cambridge: MIT Press).

28. Borst, A., and Egelhaaf, M. (1989). Principles of visual motion detection.

Trends Neurosci. 12, 297–306.

29. Eichner, H., Joesch, M., Schnell, B., Reiff, D.F., and Borst, A. (2011).

Internal structure of the fly elementary motion detector. Neuron 70,

1155–1164.

30. Tuthill, J.C., Chiappe, M.E., and Reiser, M.B. (2011). Neural correlates of

illusory motion perception inDrosophila. Proc. Natl. Acad. Sci. USA 108,

9685–9690.

31. Miall, R.C. (1978). The flicker fusion frequencies of six laboratory

insects, and the response of the compound eye to mains fluorescent

‘ripple’. Physiol. Entomol. 3, 99–106.

32. French, A.S., and Jarvilehto, M. (1978). The transmission of information

by first and second order neurons in the fly visual system. J. Comp.

Physiol. 126, 87–96.

33. Coombe, P.E., Srinivasan, M.V., and Guy, R.G. (1989). Are the large

monopolar cells of the insect lamina on the optomotor pathway?

J. Comp. Physiol. A 166, 23–35.

direct observation of on and off pathways in the drosophila visual system

Documents