Selective integration of diverse taste inputs within a single taste modality

  1. Julia U Deere
  2. Arvin A Sarkissian
  3. Meifeng Yang
  4. Hannah A Uttley
  5. Nicole Martinez Santana
  6. Lam Nguyen
  7. Kaushiki Ravi
  8. Anita V Devineni  Is a corresponding author
  1. Zuckerman Mind Brain Behavior Institute, Columbia University, United States
  2. Neuroscience Graduate Program, Emory University, United States
  3. Department of Biology, Emory University, United States

Abstract

A fundamental question in sensory processing is how different channels of sensory input are processed to regulate behavior. Different input channels may converge onto common downstream pathways to drive the same behaviors, or they may activate separate pathways to regulate distinct behaviors. We investigated this question in the Drosophila bitter taste system, which contains diverse bitter-sensing cells residing in different taste organs. First, we optogenetically activated subsets of bitter neurons within each organ. These subsets elicited broad and highly overlapping behavioral effects, suggesting that they converge onto common downstream pathways, but we also observed behavioral differences that argue for biased convergence. Consistent with these results, transsynaptic tracing revealed that bitter neurons in different organs connect to overlapping downstream pathways with biased connectivity. We investigated taste processing in one type of downstream bitter neuron that projects to the higher brain. These neurons integrate input from multiple organs and regulate specific taste-related behaviors. We then traced downstream circuits, providing the first glimpse into taste processing in the higher brain. Together, these results reveal that different bitter inputs are selectively integrated early in the circuit, enabling the pooling of information, while the circuit then diverges into multiple pathways that may have different roles.

Editor's evaluation

This important manuscript addresses the complexity of processing and representation within bitter taste perception using the Drosophila model. The authors provide convincing experimental support for distinct anatomical pathways that process bitter tastes and converge on joint downstream neurons to elicit avoidance responses. The combination of behavioral assays, in vivo physiology, optogenetic manipulation, and connectomics leads the authors to a compelling model of bitter taste processing.

https://doi.org/10.7554/eLife.84856.sa0

Introduction

Within a sensory system, sensory neurons are often specialized to detect different types of stimuli. For instance, different mechanoreceptor neurons detect different types of touch (Abraira and Ginty, 2013) and different types of cones detect distinct wavelengths of light (Rushton, 1972). Different sensory neurons may also detect stimuli in distinct locations, such as touch receptors in different parts of the body or photoreceptors in different parts of the retina. A fundamental question thus arises in each sensory system (Estebanez et al., 2018; Rompani et al., 2017; King et al., 2018; Haverkamp et al., 2018; Bates et al., 2020; Jin et al., 2021): when are distinct channels of sensory input merged and when are they kept separate in order to guide behavior? If they are merged, how and where does this integration occur?

Like other sensory systems, the gustatory system contains diverse types of sensory cells. Most studies classify taste cells based on the taste modality that they detect, such as sweet or bitter, which are often studied as homogeneous populations (Liman et al., 2014). However, each taste modality contains a repertoire of functionally diverse sensory cells. This diversity has been well characterized in Drosophila melanogaster. Flies contain sweet- and bitter-sensing cells in multiple organs, including external organs such as the legs and labellum (the distal segment of the proboscis, the fly’s feeding organ) and internal organs such as the pharynx (Scott, 2018; Figure 1A). Sweet- and bitter-sensing neurons within a single organ also exhibit functional diversity (Scott, 2018; Weiss et al., 2011; Ling et al., 2014; Fujii et al., 2015; Chen and Dahanukar, 2017). For example, the labellum contains four classes of bitter-sensing neurons (S-a, S-b, I-a, and I-b), defined by the gustatory receptors(Grs) that they express, and these classes show different response profiles to bitter tastants (Weiss et al., 2011). Flies also contain taste cells to detect other modalities, such as salt, sour (acid), and fat, but their functional diversity is not as well established (Jaeger et al., 2018; Ahn et al., 2017; Tauber et al., 2017; Mi et al., 2021; Ganguly et al., 2021).

Models for bitter taste processing and Gal4 lines to target bitter neuron subsets.

(A) Schematic depicting three major taste organs in the fly. (B) Models for how different subsets of bitter-sensing neurons could be processed in the brain. (C–D) Expression patterns of Gal4 lines used to target subsets of bitter-sensing neurons. (C) Expression of each Gal4 line in the foreleg, labellum, and pharynx. Endogenous expression of the TdT marker was imaged in flies carrying each Gal4 along with UAS-Chrim-TdT. Autofluorescence that does not represent labeled cells is visible in some pictures, such as at leg joints. (D) Table summarizing each expression pattern, based on previous studies cited in the text. Numbers refer to the number of cells per side (left or right) that each line is expressed in relative to the total number of bitter-sensing cells in that organ (leg refers to all three legs). Source data is provided for the table in D.

Figure 1—source data 1

Expression pattern of Gal4 lines used to target bitter neuron subsets.

https://cdn.elifesciences.org/articles/84856/elife-84856-fig1-data1-v1.xlsx

Thus, a taste modality such as ‘sweet’ or ‘bitter’ does not represent a single pathway, but instead comprises multiple channels of sensory input within and across organs. How are these diverse channels of taste input translated into behavioral output? One possibility is that these input channels converge onto common downstream pathways to regulate the same set of behaviors, representing an ‘integrative model’ (Figure 1B, left). Alternatively, different input channels may be processed separately to regulate distinct behaviors, a ‘segregated processing model’ (Figure 1B, right). This would allow the fly to execute specific behavioral responses depending on the compound that is detected or the organ that detects it.

Previous work, focusing mainly on appetitive tastes, suggests that taste neurons in different organs have different behavioral roles (Thoma et al., 2016; Murata et al., 2017; Chen et al., 2021; Chen et al., 2022). Appetitive tastes such as sugar elicit a sequence of behaviors, with specific organs regulating each step (Dethier, 1976; Scott, 2018; Mahishi and Huetteroth, 2019). Flies first detect the presence of sugar with their legs. Depending on the sugar concentration and the fly’s hunger state, taste detection by the leg may cause the fly to stop walking and extend its proboscis to contact the food. Labellar taste stimulation may then trigger the initiation of food consumption. Food consumption leads to stimulation of pharyngeal taste neurons, which regulate the duration of food ingestion (LeDue et al., 2015; Yapici et al., 2016; Joseph et al., 2017) and elicit local search behavior to find more food (Murata et al., 2017). Thus, sugar-sensing inputs from different organs appear to be processed separately to regulate different aspects of behavior.

However, it is not clear whether the same principle applies to aversive tastes such as bitter. Some studies suggest that different types of bitter neurons have different behavioral roles (Joseph and Heberlein, 2012; Chen et al., 2019), but this has not been widely examined. The anatomical organization of sensory projections also suggests the possibility of segregated processing: bitter-sensing neurons in different organs project axons to different regions of the subesophageal zone (SEZ), the primary taste area of the brain, suggesting potential connections to distinct downstream pathways (Wang et al., 2004; Kwon et al., 2014). Finally, the observation that bitter-sensing neurons in different organs show different response dynamics suggests that these neurons may have distinct functional roles (Devineni et al., 2021).

In this study, we leverage a wide array of genetic tools and behavioral assays to systematically address this longstanding question: are different channels of bitter input integrated or processed separately to regulate behavior? First, we optogenetically activated subsets of bitter neurons in different organs. Bitter neuron subsets elicited a broad and highly overlapping set of behaviors, suggesting that these inputs converge onto common downstream pathways, but we also observed differences that argue for biased convergence. Next, we used transsynaptic tracing to determine whether different types of bitter neurons connect to common or distinct pathways. Consistent with our behavioral results, bitter neurons in different organs connect to overlapping second-order pathways with biased connectivity. Finally, we studied a novel type of putative second-order neuron that projects to the higher brain. We show that it integrates bitter input from multiple organs, and we characterized its response properties, behavioral role, and downstream connectivity. Together, these results suggest that bitter inputs from different organs are selectively integrated at the first synapse and drive largely overlapping behaviors, contrasting with the prevailing view that different taste organs have different functions. Our studies also provide insight into the functional role of downstream bitter neurons and the organization of taste pathways in the higher brain.

Results

Subsets of bitter neurons in different organs act in parallel to regulate feeding

The goal of this study was to uncover the logic and circuit architecture underlying how different channels of bitter input are processed in the brain. Different bitter inputs, including input from different taste organs or neuronal classes, could be integrated in the brain to regulate the same behaviors (Figure 1B, left) or processed separately to regulate distinct behaviors (Figure 1B, right). We first aimed to distinguish between these models by optogenetically activating subtypes of bitter neurons within specific organs and examining their effect on a wide range of behaviors. We chose to perform activation experiments rather than neuronal silencing because we wanted to determine which behaviors each set of neurons is capable of driving, which implies connectivity to downstream behavioral circuits. Neuronal silencing may not reveal the same effects due to redundant functions of bitter neurons, especially because we are only targeting a subset of bitter neurons within each organ (see below). Moreover, optogenetic activation allows precise control over the timing and duration of taste stimulation. As a control, for each experiment we activated the entire population of bitter-sensing neurons and ensured that the resulting behavioral effects are consistent with known effects of bitter taste.

We used previously characterized Gal4 lines (Weiss et al., 2011; Ling et al., 2014; Kwon et al., 2014; Chen and Dahanukar, 2017) to target all bitter-sensing neurons across the body (Gr33a-Gal4) as well as four subsets of bitter neurons within specific organs (Figure 1C–D): (1) bitter neurons in the leg (Gr58c-Gal4); (2) bitter neurons in the labellum belonging to classes S-a and I-a (Gr59c-Gal4); (3) bitter neurons in the labellum belonging to class S-b (Gr22f-Gal4); and (4) a pharyngeal bitter neuron termed V6 (Gr9a-Gal4). The expression patterns of these four Gal4 lines have been carefully validated by previous studies, showing that each line is expressed in only one of the three taste organs under study (leg, labellum, or pharynx) (Weiss et al., 2011; Ling et al., 2014; Kwon et al., 2014; Chen and Dahanukar, 2017). We also imaged these organs to confirm the organ-specific expression pattern of each line (Figure 1C).

Because these Gal4 lines label different numbers of neurons (Figure 1D) and may also vary in their expression strength, we focused on analyzing whether the activation of each neuronal subset could drive any significant change in each behavior rather than comparing the strength of behavioral effects across lines. We used the light-activated channel Chrimson to drive neuronal activation (Klapoetke et al., 2014). Chrimson has been previously used to activate taste sensory neurons (Klapoetke et al., 2014; Musso et al., 2019; Moreira et al., 2019), and bitter-sensing neurons expressing Chrimson are effectively activated by red light (Devineni et al., 2021).

We began by examining the effect of bitter neuron activation on feeding responses. We first tested the proboscis extension response (PER), an appetitive response that represents the initiation of feeding and is suppressed by bitter compounds (Dethier, 1976; Scott, 2018; Wang et al., 2004). As expected, activating the entire population of bitter-sensing neurons strongly suppressed PER to sugar (Figure 2A). Interestingly, activating any of the four bitter neuron subsets also suppressed PER to a similar extent (Figure 2A). Thus, subsets of bitter neurons within any of the three organs – the leg, labellum, or pharynx – can suppress the initiation of feeding.

Figure 2 with 1 supplement see all
Bitter neuron subsets in different organs act in parallel to regulate feeding responses.

(A) Effect of activating bitter neuron subsets on proboscis extension response (PER) to 100 mM sucrose (n=5–8 sets of flies). Bar graphs for each neuronal subset (left) represent the fraction of flies showing PER with and without light stimulation. Bar graph on the right shows the degree of PER suppression elicited by neuronal activation, quantified as: 1 – (PER with light/PER without light). (B, E) Schematic of optoPAD setup used to test how neuronal activation (B) or silencing (E) affects feeding. Food sources contained 100 mM sucrose for activation experiments (B) and 50 mM sucrose +1 mM quinine for silencing experiments (E). (C, D, F, G) Effects of bitter neuron activation (C–D; n=40–48 flies) or silencing (F–G; n=40–62 flies) on feeding for 1 hr. (C, F) Cumulative feeding duration on the control and opto stim food sources for experimental flies. Values for the last time point (1 hr) were compared for control versus opto stim using paired t-tests. (D, G) Bars represent the difference in each feeding parameter between the control and opto stim food for each genotype. For each feeding measure, experimental flies were compared to both controls, and effects not labeled with an asterisk are not significant. In all panels, Gal4 lines used were Gr33a-Gal4 (all bitter neurons), Gr58c-Gal4 (leg), Gr59c-Gal4 (labellum, S-a + I-a), Gr22f-Gal4 (labellum, S-b), and Gr9a-Gal4 (pharynx). For all figures: *p<0.05, **p<0.01, ***p<0.001, ns = not significant (p>0.05). Unless otherwise specified, two groups were compared using unpaired t-tests and more than two groups were compared using one-way ANOVA followed by Dunnett’s test comparing experimental flies to each control.

PER is an immediate response that is tested in immobilized flies. To examine more naturalistic feeding behavior in freely moving flies over a longer time period, we used the optoPAD assay (Moreira et al., 2019). The optoPAD uses a capacitance sensor to detect feeding events, which consist of individual ‘sips’ clustered into feeding bursts and feeding bouts (Figure 2—figure supplement 1A; Itskov et al., 2014). We used closed-loop light stimulation to optogenetically activate bitter neurons whenever the fly interacted with one food source (100 mM sucrose), while a second identical food source not linked to optogenetic stimulation served as the control (Figure 2B). Activating all bitter-sensing neurons or any of the four subsets caused a near-complete suppression of feeding (Figure 2C–D). This suppression was evident in many different feeding-related parameters. All four bitter neuron subsets suppressed the total feeding duration, number of feeding bouts, sip number, and sip duration (Figure 2D). The number of feeding bursts was suppressed by all subsets except for leg neurons, although they showed the same trend (Figure 2—figure supplement 1B). The duration of individual feeding bouts was suppressed by only one subset, the pharyngeal neurons, suggesting that this specific parameter may be primarily regulated by the pharynx (Figure 2—figure supplement 1B).

We next asked whether silencing these bitter neuron subsets prevents flies from suppressing feeding in the presence of natural bitter taste. We used the light-gated chloride channel GTACR to optogenetically silence neuronal activity (Mohammad et al., 2017). We presented flies with two bitter-containing food sources (50 mM sucrose + 1 mM quinine), and we used closed-loop stimulation to silence bitter neurons whenever the fly interacted with one of the two food sources (Figure 2E). As expected, silencing all bitter neurons across the body strongly increased feeding (Figure 2F–G). We did not expect that silencing bitter neuron subsets would have a strong effect because each subset alone robustly suppresses feeding (Figure 2C–D), suggesting that they act redundantly. Silencing two of the four subsets (neurons in the leg and pharynx) increased feeding in experimental flies, with pharyngeal neurons showing a stronger effect (Figure 2F). However, the effects of light stimulation generally did not reach statistical significance when compared with genetic controls (Figure 2G and Figure 2—figure supplement 1C). Thus, activating each bitter neuron subset had strong effects on feeding whereas silencing had weaker or null effects. These results suggest that different bitter neuron subsets act in a parallel and partially redundant manner to regulate feeding, implying that these input channels are integrated by downstream feeding circuits.

Bitter neuron subsets in different organs elicit similar effects on locomotion

Although different bitter neuron subsets elicited similar effects on feeding behavior, they may exert different effects on other taste-related behaviors. We tested their effect on locomotion, which is known to be affected by taste (Flood et al., 2013; Thoma et al., 2016; Tadres et al., 2020). To monitor locomotor behavior, we filmed flies and tracked their movement in a circular arena (Figure 3A–B; Aso et al., 2014b). To activate bitter neurons, we delivered 5 s light stimulation at three different intensities (‘low’, ‘medium’, and ‘high’, corresponding to 4, 20, and 35 µW/mm2). Activating the entire population of bitter neurons strongly stimulated locomotion at all light intensities. Light onset elicited a transient increase in turning and a more sustained increase in forward velocity and the fraction of flies moving (Figure 3C–D and Figure 3—figure supplement 1A). When light stimulation was turned off, flies immediately stopped moving, and this locomotor suppression persisted for nearly a minute (Figure 3C and Figure 3—figure supplement 1A–B). We observed the same behavioral effects with continuous light stimulation (Figure 3C–D) and 50 Hz pulsed light (Figure 3—figure supplement 1C–D). Together, these results suggest that when a fly senses a bitter stimulus, it displays a sequence of stereotyped behaviors mediating aversion: (1) it first turns in an attempt to orient away from the stimulus, (2) it increases forward velocity in an attempt to run away from the stimulus, and (3) when the stimulus disappears, it freezes because it may perceive its current position to be a safe (bitter-free) location.

Figure 3 with 2 supplements see all
Bitter neuron subsets in three different organs elicit similar effects on locomotion.

(A) Arena used for tracking flies during optogenetic activation. (B) Schematic of protocol to test the effect of bitter neuron activation on locomotion. 5 s light stimulation was used at three different light intensities (low, medium, high). (C–D) Locomotor effects elicited by activating all bitter-sensing neurons using Gr33a-Gal4 (n=11–12 trials). Left graphs show forward (C) and angular (D) velocity over time for medium light intensity (shading indicates light on). Bar graphs quantify the change in these parameters at light onset. (E–F) Locomotor effects elicited by activating bitter neuron subsets with medium intensity light (n=10–12 trials). Effects of other light intensities are shown in Figure 3—figure supplement 2. Traces show forward (E) or angular (F) velocity over time (shading indicates light on). Bar graphs quantify the change in these parameters at light onset. Gal4 lines used were Gr33a-Gal4 (all bitter neurons), Gr58c-Gal4 (leg), Gr59c-Gal4 (labellum, S-a + I-a), Gr22f-Gal4 (labellum, S-b), and Gr9a-Gal4 (pharynx). Experimental effects not labeled with an asterisk are not significant.

We then tested whether these effects could be evoked by subsets of bitter neurons. We hypothesized that locomotor changes may be regulated by bitter neurons in the leg, analogous to the role of tarsal sweet-sensing neurons in mediating locomotor responses to sugar (Thoma et al., 2016). Instead, activating any of the four bitter neuron subsets enhanced locomotion for at least one of the three light intensities (Figure 3E and Figure 3—figure supplement 2A). Three of the four subsets (all except labellar S-b neurons) also elicited an increase in turning at light onset as well as locomotor suppression after light was turned off, although the latter effect was only significant for two lines (Figure 3E–F and Figure 3—figure supplement 2). These results show that bitter neurons in three different organs can drive locomotor changes typical of aversion, again suggesting that different bitter inputs converge onto common behavioral circuits.

Bitter neuron subsets in multiple organs can elicit innate and learned aversion

We next tested the effect of bitter neuron activation on innate and learned aversion. Flies avoid residing in areas containing bitter compounds (Marella et al., 2006; Joseph and Heberlein, 2012). This positional aversion may correlate with feeding aversion, but it can be displayed in the absence of feeding and is mediated by different motor circuits (i.e., motor neurons controlling the legs versus the proboscis). Positional aversion may also be related to locomotor responses (Figure 3), as stimulation of speed and turning would increase the likelihood that flies leave a bitter substrate. However, it is not obvious that activating each bitter neuron subset should have an identical effect in both behavioral assays, as aversion is measured over a longer timescale and active choice processes may contribute.

To measure innate aversion, we presented light in two opposing quadrants of the arena and quantified the flies’ preference for residing in the light quadrants, where they experience bitter neuron activation (Figure 4A). Expressing Chrimson in all bitter neurons elicited strong positional aversion at all light intensities (Figure 4B and Figure 4—figure supplement 1A). The effect was similar for continuous or 50 Hz pulsed light stimulation (Figure 4—figure supplement 1B). We then tested the effect of each bitter neuron subset. Activating three of the four bitter neuron subsets elicited innate aversion (Figure 4C and Figure 4—figure supplement 2). This included neuronal subsets in the leg, labellum, and pharynx, with only labellar S-b neurons failing to elicit an effect.

Figure 4 with 2 supplements see all
Subsets of bitter neurons in multiple organs elicit innate and learned aversion.

(A) Protocol to test the effect of bitter neuron activation on innate positional aversion. (B–C) Effect of activating all bitter-sensing neurons (B) or subsets of bitter neurons (C) on innate aversion (n=20–24 trials, 10–12 sets of flies). Light preference is quantified by the preference index (PI): (# flies in light quadrants – # flies in non-light quadrants)/total # flies. Negative values indicate aversion. Line graphs show the PI over 30 s of low intensity light stimulation (shading indicates light on), including both of the sequential test periods when the lighted quadrants were switched. Flies may appear to show a positive PI before light onset because of the repeated tests: after the first test they continue to avoid the previously illuminated quadrants until the next test. Bar graphs show the final PI (average over last 5 s of light presentation) for each genotype at all light intensities (B) or low light intensity (C). Effects of bitter neuron subsets at other light intensities are shown in Figure 4—figure supplement 2. (D) Protocol to test the effect of bitter neuron activation on learned odor aversion. (E–F) Effects of activating all bitter-sensing neurons (E) or subsets of bitter neurons (F) on learned aversion (n=20–24 trials, 10–12 sets of flies). Learned preference was quantified as the PI for the conditioned odor during the test periods, calculated as: (# flies in CS+ quadrants – # flies in CS- quadrants)/total # flies. Negative values indicate aversion. Data from odor tests 1 and 2 are combined. Line graphs show the conditioned odor PI during the odor test periods (shading indicates odors on). Bar graphs show the final PI (average over last 5 s of test) for each genotype. In all panels, experimental effects not labeled with an asterisk are not significant, and Gal4 lines used were Gr33a-Gal4 (all bitter neurons), Gr58c-Gal4 (leg), Gr59c-Gal4 (labellum, S-a + I-a), Gr22f-Gal4 (labellum, S-b), and Gr9a-Gal4 (pharynx). Figure 4A has been adapted from Figure 2A from Aso et al., 2014b and Figure 4D has been adapted from Figure 1D from Aso and Rubin, 2016.

In addition to driving innate aversion, bitter taste can serve as a negative reinforcement signal to elicit learned aversion to odors (Das et al., 2014). We used previously established protocols (Aso and Rubin, 2016) to test whether bitter neuron activation could drive learned odor aversion. We delivered one odor (the conditioned stimulus, CS+) while activating bitter neurons, then delivered a different odor (the CS-) without neuronal activation, and finally allowed the flies to choose between quadrants containing the CS+ versus CS- (Figure 4D). As expected, pairing odor with the activation of all bitter neurons elicited aversion to the CS+ (Figure 4E). Activating two of the four bitter neuron subsets also elicited learned aversion (Figure 4F), including neurons in the leg and labellum. Labellar S-b neurons failed to elicit a significant effect, similar to their lack of an effect on innate aversion. Interestingly, pharyngeal neurons did not elicit a learned response despite driving strong innate aversion.

Together, these experiments demonstrate that different subsets of bitter neurons drive broad and largely overlapping behavioral responses (Figure 5A). The fact that all neuronal subsets strongly evoked at least one aversive behavior suggests that each set of neurons was effectively activated by Chrimson. Labellar S-b neurons had the sparsest effect and only affected three of the seven behavioral measures analyzed (Figure 5A), but we noted that the corresponding Gal4 line may have weaker expression than the other lines (based on weaker labeling of axonal projections in the brain and the inability to drive consistent trans-Tango labeling in the experiments below; data not shown). The other three subsets, residing in three different organs, all affected at least six of the seven behaviors tested, including proboscis extension, feeding, locomotion, and innate aversion (Figure 5A). Two of those three subsets (leg and labellar S-a + I-a neurons) also caused aversive learning.

Summary of behavioral effects elicited by optogenetic activation of bitter neuron subsets.

(A) Effects are color-coded by strength, relative to the effect of activating all bitter neurons with Gr33a-Gal4. All observed effects went in the same direction. See Materials and methods for details of how effects were quantified. Note that the ‘moderate’ effect of all subsets on feeding is primarily due to the fact that flies consumed less of the control food compared to activating all bitter neurons (leading to a smaller difference between control and opto stim), but all subsets exerted an almost complete suppression of feeding. (B) Model for how different inputs contribute to different types of behavior. Line widths represent the strength of the effect (weak, moderate, or strong). For simplicity, we are only defining four behavioral categories: feeding suppression includes both feeding and proboscis extension response (PER) experiments, and locomotion includes the three measures shown in the table. The strongest effect for each category was used. Results with I-a + S-a neurons (not S-b) were used to interpret the role of the labellum, since positive results are more informative than negative results. Source data is provided for the table in Figure 5A.

Figure 5—source data 1

Summary of behavioral effects elicited by optogenetic activation of bitter neuron subsets.

https://cdn.elifesciences.org/articles/84856/elife-84856-fig5-data1-v1.xlsx

These results generally support an integrative model of taste processing (Figure 1B, left) in which bitter inputs from different organs converge onto common downstream pathways to drive a diverse set of aversive behaviors. However, the fact that not all neuronal subsets elicited all behaviors, or preferentially elicited some behaviors over others, suggests that some downstream pathways may receive selective or preferential input from specific bitter-sensing cells. We refer to this model as ‘selective integration’ (Figure 5B), and it lies in between the two models shown in Figure 1B.

Bitter input from different organs is relayed to overlapping downstream regions

Our behavioral results suggest that taste circuits in the brain integrate bitter input from multiple organs. Where does this integration occur? Cross-organ integration could occur early in the taste circuit, potentially at the first synapse, or much later in the circuit. To test whether integration may occur at the first synapse, we used the transsynaptic tracing method trans-Tango (Talay et al., 2017) to identify neurons that are postsynaptic to bitter sensory cells, termed second-order bitter neurons. We note that we cannot use the recently published synaptic connectome of the fly brain (Scheffer et al., 2020) to identify second-order neurons because the connectome does not include the SEZ, the region containing bitter sensory axons. Although some gustatory neurons in the legs arborize in the ventral nerve cord, the axons of tarsal bitter-sensing neurons pass through the nerve cord without making arborizations typical of synaptic connections (Kwon et al., 2014). We therefore focused on second-order bitter pathways in the brain.

We first labeled the entire population of second-order bitter neurons by tracing neurons that are postsynaptic to Gr33a-Gal4-expressing cells. We observed many local neurons projecting within the SEZ as well as three major projections to the superior protocerebrum: a lateral, medial, and mediolateral tract (Figure 6A and Figure 6—figure supplement 1). This pattern is consistent with second-order bitter neurons labeled in other recent studies (Chen et al., 2019; Snell et al., 2022). Thus, at the first synapse, the bitter taste circuit diverges into at least three downstream pathways projecting out of the SEZ.

Figure 6 with 1 supplement see all
Bitter inputs from different organs are relayed to overlapping downstream pathways.

(A) The entire population of second-order bitter neurons was labeled by trans-Tango tracing from Gr33a-Gal4-expressing cells. Arrows denote the three major tracts (lateral, mediolateral, and medial) projecting out of the subesophageal zone (SEZ) to the dorsal brain. (B) Neurons postsynaptic to specific bitter neuron subsets were labeled with trans-Tango. Gal4 lines used were Gr58c-Gal4 (leg), Gr59c-Gal4 (labellar S-a + I-a), and Gr9a-Gal4 (pharynx). Tracing from labellar S-b cells labeled by Gr22f-Gal4 yielded very weak trans-Tango staining that could not be interpreted, so it is not shown here. (C) Summary of inputs from different subsets of bitter neurons onto second-order projection pathways. Dotted lines indicate weak input based on weak or inconsistent trans-Tango staining. See Figure 6—figure supplement 1 for additional trans-Tango images that include presynaptic Gal4 expression.

We next used trans-Tango to label neurons postsynaptic to specific subsets of bitter sensory cells, using the same Gal4 lines used to test behavior (Figure 6B and Figure 6—figure supplement 1). Tracing neurons postsynaptic to bitter cells in the leg labeled all three projection tracts, although the medial tract was labeled relatively weakly. Tracing from S-a + I-a neurons in the labellum also labeled all three tracts; the medial tract was labeled more strongly than when tracing from the leg, whereas the lateral tract was labeled more weakly. Finally, tracing neurons postsynaptic to pharyngeal bitter neurons labeled the medial tract almost exclusively, although the mediolateral tract occasionally showed weak labeling. A previous study tracing postsynaptic partners of pharyngeal taste neurons reported similar results, although the mediolateral tract was more clearly labeled in that study (Chen et al., 2019).

Together, these results show that: (1) bitter neurons in each organ synapse onto multiple second-order pathways, and (2) each projection tract receives input from multiple organs, although it is not clear whether these inputs converge onto the same individual neurons. Interestingly, bitter inputs show biased connectivity to second-order taste regions, with neurons in different organs preferentially connecting to different projection tracts (Figure 6C). This biased connectivity could allow downstream taste pathways to be selectively or preferentially activated by certain organs, as suggested by our behavioral results (Figure 5B).

mlSEZt neurons integrate bitter input from the leg and labellum

In order to study how specific types of second-order bitter neurons encode and integrate bitter information, we require genetic driver lines to label them. By searching the Janelia FlyLight expression database (Jenett et al., 2012), we identified multiple Gal4 lines that appeared to label second-order neurons comprising the mediolateral tract. We refer to these cells as mlSEZt neurons based on the terminology in Talay et al., 2017. We focus the rest of this study on mlSEZt cells, which represent one of the only types of second-order bitter neurons to be functionally studied.

As visible in the first line we identified, R29F12-Gal4, mlSEZt neurons have cell bodies just dorsal to the antennal lobes, a V-shaped projection in the SEZ, and a circular projection in the superior lateral protocerebrum (SLP) (Figure 7A, top). Colabeling revealed that the axons of bitter-sensing neurons overlap with the medial branch of the mlSEZt projection in the SEZ (Figure 7A, bottom). To confirm that mlSEZt neurons receive bitter input, we monitored their responses to taste stimulation using calcium imaging. The projections of mlSEZt neurons in the SEZ and SLP could be clearly identified based on their location and morphology. mlSEZt projections in both the SEZ and SLP responded to bitter stimulation of either the labellum or leg (Figure 7B–D), demonstrating that these neurons receive bitter input from multiple organs. Water or sugar applied to the labellum elicited a much weaker response than bitter in the SEZ (~50% of the bitter response; Figure 7B) and no response in the SLP (Figure 7C). Water or sugar responses were not observed in either the SEZ or SLP with tarsal stimulation (Figure 7B–C). The addition of sugar did not modulate the response to labellar bitter stimulation (Figure 7E), suggesting that mlSEZt neurons do not receive inhibition from the sugar pathway.

Figure 7 with 2 supplements see all
Second-order mlSEZt bitter neurons receive input from multiple organs.

(A) Top: R29F12-Gal4 expression pattern (maximum intensity projection created from images generated by the Janelia FlyLight Project Team). Brain slices at the far anterior and posterior edges are not included to maximize visibility of mlSEZt cells. Pink arrowhead shows mlSEZt cell bodies; white and yellow arrows show mlSEZt projections in the subesophageal zone (SEZ) and superior lateral protocerebrum (SLP), respectively. Box depicts approximate area of the SEZ shown in the bottom image. Bottom: SEZ image with mlSEZt neurons (R29F12-Gal4 driving UAS-TdTomato, red) colabeled with bitter-sensing neurons (Gr66a-lexA driving lexAop-GCaMP6f, green). Medial and lateral branches of the mlSEZt projections are denoted by white and yellow arrows, respectively. Based on morphology, most of the visible bitter sensory projections likely arise from the labellum (white arrowhead); bitter axons from the leg have a stick-like projection just lateral to the labellar projections (yellow arrowhead). (B–F) GCaMP responses of mlSEZt neurons. R29F12-Gal4 was used to drive UAS-GCaMP6f. For all imaging traces, the gray bar denotes 5 s taste presentation. Unless otherwise specified, the bitter stimulus was 10 mM denatonium and the sugar stimulus was 100 mM sucrose. (B–C) GCaMP responses of mlSEZt projections in the SEZ (B) or SLP (C) with taste stimuli applied to the labellum (left) or foreleg (right). Response magnitudes across these four conditions are not directly comparable because they include different sets of flies. (B) n=15–22 trials, 4 flies (labellar stimulation); n=15–19 trials, 4 flies (leg stimulation). (C) n=25–34 trials, 7 flies (labellar stimulation); n=12–19 trials, 3–4 flies (leg stimulation). (D) Responses of mlSEZt projections in the SEZ to labellar or tarsal bitter stimulation imaged in the same flies (n=11–14 trials, 3 flies). (E) Responses of mlSEZt projections in the SLP when sugar, bitter, or a sugar-bitter mixture was applied to the labellum (n=16 trials, 4 flies). (F) Responses of mlSEZt projections in the SLP when different bitter compounds were applied to the labellum (n=22–23 trials, 6 flies).

mlSEZt neurons showed different response dynamics when bitter was applied to the labellum versus the leg, as previously reported for bitter sensory neurons in these organs (Devineni et al., 2021). Bitter stimulation of the labellum elicited transient responses at bitter onset (ON response) and offset (OFF response), whereas stimulation of the leg elicited a more sustained ON response with no OFF response (Figure 7B–D). When comparing responses in the same flies (Figure 7D), labellar stimulation elicited stronger ON responses than tarsal stimulation (~70% versus ~30% ∆F/F).

Tarsal and labellar bitter stimulation also produced different spatial patterns of activation in the SEZ (Figure 7—figure supplement 1A–B). Responses to tarsal stimulation were localized to a small region located at the dorsolateral edge of the larger region activated by labellar stimulation. This is consistent with our colabeling experiment (Figure 7A, bottom) and the known projection patterns of tarsal and labellar bitter-sensing neurons (Wang et al., 2004; Kwon et al., 2014). Thus, bitter inputs from different organs likely activate different dendrites on mlSEZt neurons.

In the SLP, the spatial pattern of mlSEZt activation was similar when bitter was applied to either the leg or labellum (Figure 7—figure supplement 1C). We observed individual puncta, likely representing axon terminals of single mlSEZt cells, that were activated by bitter stimulation of either organ. These results suggest that individual mlSEZt cells receive convergent input from the labellum and leg. mlSEZt cell bodies, which are located far from either the SEZ or SLP projections (Figure 7A), did not respond to bitter taste (data not shown), preventing us from unequivocally imaging responses in single mlSEZt cells.

Finally, we tested whether bitter responses in mlSEZt neurons depend on the identity of the bitter compound and the recent history of bitter stimulation, as shown for labellar bitter sensory neurons (Devineni et al., 2021). Indeed, mlSEZt neurons showed compound-dependent dynamics that resembled those of sensory neurons. Relative to the strength of the ON response, denatonium and quinine elicited strong OFF responses, caffeine elicited a slightly weaker OFF response, and L-canavanine elicited no OFF response at all (Figure 7F). mlSEZt responses also showed an experience-dependent effect resembling that of bitter-sensing neurons: with repeated bitter stimulation, the bitter ON response habituated much more strongly than the OFF response (Figure 7—figure supplement 1D).

Together, these results strongly suggest that the mlSEZt neurons labeled by R29F12-Gal4 are second-order bitter neurons: they have an identical morphology to second-order bitter neurons labeled by trans-Tango, their projections overlap with the axon terminals of bitter-sensing neurons, and they show clear bitter responses. We will therefore refer to mlSEZt neurons as second-order bitter neurons, although it is formally possible that they are downstream bitter neurons that are not directly connected to bitter-sensing cells. The taste selectivity and response dynamics of mlSEZt neurons closely resemble those of bitter sensory neurons. The major transformation we observe is that mlSEZt neurons receive convergent bitter input from at least two different organs, the leg and labellum. This convergence may represent a substrate by which bitter neurons in different organs elicit common behavioral effects (Figure 5).

mlSEZt neurons regulate a subset of taste-related behaviors

Given the existence of several second-order bitter pathways (Figure 6), we wondered how individual types of second-order neurons like mlSEZt contribute to behavior. To conduct behavioral experiments with mlSEZt, we first used a split-Gal4 intersectional approach (Dionne et al., 2018) to generate a more specific driver line. We took advantage of a second Gal4 line that labeled the same bitter-selective mlSEZt neurons as R29F12-Gal4 (Figure 7—figure supplement 2A–B). The split-Gal4 approach enables us to exclusively label neurons present in both expression patterns, resulting in a highly specific mlSEZt line (Figure 8A). We verified that mlSEZt neurons labeled by the split-Gal4 line respond to bitter (Figure 7—figure supplement 2C). This line was used for all behavioral experiments.

Figure 8 with 3 supplements see all
Second-order mlSEZt neurons regulate a subset of taste-related behaviors.

(A) Expression pattern of split-Gal4 line (R29F12-AD+R55E01-DBD) labeling mlSEZt neurons with high specificity. Pink arrowhead shows mlSEZt cell bodies; white and yellow arrows denote their projections in the subesophageal zone (SEZ) and superior lateral protocerebrum (SLP), respectively. (B–H) mlSEZt neurons were activated or silenced using R29F12-R55E01 split-Gal4 driving UAS-Chrimson or UAS-GTACR, respectively. (B) Effect of activating mlSEZt neurons on proboscis extension response (PER) to 50 mM sucrose (n=4 experiments, 40 flies). (C) Effect of silencing mlSEZt neurons on bitter suppression of PER, tested by adding 10 mM quinine to 50 mM sucrose (n=5 experiments, 46–50 flies). Left graph shows PER in each condition (3 of the 9 conditions produced zero PER, so bars are at zero with zero error). Right graph shows the degree of bitter suppression based on the left graph, quantified as: 1 – (PER with bitter/PER without bitter). (D–E) Effect of activating (D; n=42–47 flies) or silencing (E; n=66–69 flies) mlSEZt neurons on feeding, as measured over 1 hr in the optoPAD (same setup as shown in Figure 2B and E, respectively). Food sources contained 100 mM sucrose for activation experiments (D) or 50 mM sucrose + 1 mM quinine for silencing experiments (E). Left graphs show cumulative feeding duration on the control and opto stim food sources for experimental flies. Values for the last time point (1 hr) were compared using paired t-tests. Cumulative feeding for controls is shown in Figure 8—figure supplement 1. Bar graphs on the right represent total feeding duration or sip duration on the control versus opto stim food for each genotype. Paired t-tests were used to compare values for control versus opto stim food. Comparisons not labeled with an asterisk are not significant. (F–H) Effects of mlSEZt neuron activation on locomotion (F; n=11–12 trials), innate preference (G; n=22–24 trials, 11–12 sets of flies), and learned odor preference (H; n=14–16 trials, 7–8 sets of flies). Effects of high intensity light stimulation are shown in panels F and G; for other intensities, see Figure 8—figure supplement 2.

We first tested the role of mlSEZt bitter neurons in regulating PER. Activating mlSEZt neurons using Chrimson suppressed PER to sugar (Figure 8B), similar to the effect of activating bitter-sensing neurons (Figure 2A). Silencing mlSEZt neurons using GTACR reduced the PER suppression elicited by quinine (Figure 8C). Thus, mlSEZt neurons partially mediate the aversive effect of bitter taste on proboscis extension. To determine whether this effect translates into a regulation of feeding over longer timescales, we used the optoPAD assay. mlSEZt activation suppressed feeding on sugar (Figure 8D and Figure 8—figure supplement 1A–B), consistent with its effect on PER. Certain feeding parameters were suppressed (e.g., total feeding duration, sip duration, number of feeding bouts) while others were not (e.g., number of sips). In contrast to the effect of activation, mlSEZt silencing did not significantly affect feeding on a sugar/bitter mixture (Figure 8E and Figure 8—figure supplement 1C–D). These results suggest that mlSEZt neurons suppress specific aspects of feeding behavior but act in parallel with other second-order bitter pathways, creating redundancy in the circuit.

In contrast to the effect on proboscis extension and feeding, activating mlSEZt neurons generally did not elicit locomotor changes, innate positional aversion, or learned aversion (Figure 8F–H and Figure 8—figure supplement 2). The only statistically significant effect in any of these behavioral assays, at any light intensity, was an increase in turning at the onset of high intensity light (Figure 8F); medium intensity light also elicited a similar but non-significant effect. Changes in forward velocity at light onset or offset did not differ from controls (Figure 8F and Figure 8—figure supplement 2A–B). We repeated the locomotor assay at an even higher light intensity (44 µW/mm2) and again observed a significant increase in turning, but not forward velocity (Figure 8—figure supplement 2D).

We also tested whether mlSEZt activity is required for bitter-sensing neurons to elicit locomotor responses, innate aversion, or learned aversion. We constitutively silenced mlSEZt neurons by expressing the inwardly rectifying potassium channel Kir2.1 (Baines et al., 2001), and we activated bitter neurons across the body by using Gr66a-lexA to drive Aop-Chrimson. None of the locomotor or preference behaviors elicited by bitter sensory activation were affected by mlSEZt silencing (Figure 8—figure supplement 3), indicating that mlSEZt neurons are not required for these behaviors.

Together, these data show that second-order mlSEZt neurons regulate a subset of the aversive behaviors elicited by bitter taste. Behaviors that were not affected by mlSEZt activation may be mediated by other second-order bitter pathways, or they may require the co-activation of multiple second-order pathways.

Further characterization of mlSEZt cell identity

Having identified a novel type of second-order bitter neuron that regulates aversive behavior, we sought to further characterize its identity and function. The morphology of mlSEZt neurons indicates that they belong to the CREa1 lineage (Yu et al., 2013). Neurons within this lineage have sexually dimorphic projections (Yu et al., 2013), and mlSEZt neurons share some resemblance with sexually dimorphic fruitless (fru)-expressing mAL (aDT2) neurons (Kimura et al., 2005; Yu et al., 2010; Cachero et al., 2010), which function in the male courtship circuit (Clowney et al., 2015). However, mAL neurons in males do not have a circular projection in the SLP, as mlSEZt cells do. Although our initial characterization of mlSEZt neurons was performed in females (Figures 78), mlSEZt neurons in males also show a circular SLP projection (Figure 9—figure supplement 1), indicating that they are distinct from fru+mAL neurons and do not appear to be sexually dimorphic.

We next asked whether mlSEZt cells are excitatory or inhibitory. The CREa1 lineage contains GABAergic neurons (Ito et al., 2013), and some neurons resembling mlSEZt in the FlyCircuit database express Gad1-Gal4, a GABAergic marker (e.g., Gad1-F-600213 and Gad1-F-400450) (Chiang et al., 2011). To confirm that mlSEZt cells are GABAergic, we colabeled them with the GABAergic marker VGAT-lexA and observed co-expression in most mlSEZt cells (Figure 9A and Figure 9—figure supplement 2A). While some mlSEZt cells did not show clear co-expression, it is possible that expression levels were too low to be detected. To rule out the possibility that some mlSEZt cells are excitatory, we also performed colabeling experiments with the cholinergic marker ChAT-lexA, as cholinergic neurons represent the major population of excitatory neurons in the fly brain. We did not observe any mlSEZt neurons that expressed ChAT-lexA (Figure 9A and Figure 9—figure supplement 2B). Thus, mlSEZt bitter neurons are GABAergic cells that convey feedforward inhibition to the SLP.

Figure 9 with 2 supplements see all
Downstream pathways from mlSEZt cells.

(A) Colabeling of mlSEZt cells with markers for GABAergic (top) or cholinergic (bottom) neurons. Each row shows mlSEZt labeling (red), the neurotransmitter marker (green), and both channels overlaid. Arrows point to the location of mlSEZt cells in each image. Flies contained R29F12-Gal4 driving UAS-TdT (red) and either VGAT-lexA (top) or ChAT-lexA (bottom) driving Aop-GCaMP6f (green). Additional images are shown in Figure 9—figure supplement 2. (B) Left: trans-Tango labeling of cells postsynaptic to mlSEZt neurons (using the mlSEZt split-Gal4 line). Right: Magnified images showing superior lateral protocerebrum (SLP) staining (top) and cell bodies (bottom). Many cell bodies are colabeled in both red and green, representing lateral connections between mlSEZt cells. (C) Left: Same image as panel B with only the red channel shown for greater visibility. Arrow shows a ventral projection out of the SLP. Right: Magnified image of the SMP showing a projection to the contralateral hemisphere (arrowhead) and ventral projection toward the SEZ (arrow). For clarity, magnified images in panels B and C represent a subset of the z-slices shown in the whole-brain picture, with brightness and contrast adjusted. (D) Putative mlSEZt neurons in the connectome. (E) Interconnectivity between mlSEZt neurons. Connections with at least 5 synapses are shown. Line weights represent connection strength (number of synapses; maximum = 30). (F) Top: Distribution of connection strength from mlSEZt cells to each 3N. The y-axis represents the total number of synapses each 3N receives from all 21 mlSEZt cells. Bottom left: Fraction of top 3Ns projecting to each brain region. Parenthetical ‘C’ refers to the specified region in the hemisphere contralateral to the mlSEZt projection. See Table 1 for abbreviations. Bottom right: Frontal brain slice showing the approximate locations of major regions targeted by 3Ns or 4Ns. Regions in parentheses are not visible in the specific plane shown but are located at that x-y position in an anterior or posterior plane. See Ito et al., 2014 or NeuPrint Explorer for precise region locations. (G) Connectivity between mlSEZt cells and top 3Ns, showing connections with at least 10 synapses (thus not all of the top 3Ns are represented). Line weights represent connection strength (number of synapses; maximum = 18). 3Ns are color-coded based on cell type name prefixes in the connectome, which reflect the cell body location or types defined in previous studies. (H) Left: All top 3Ns (red) shown along with four mlSEZt cells (gray). Right: Top five 3N cell types (red) that receive the strongest mlSEZt input, each shown along with three input mlSEZt cells (gray). (I) Top: Distribution of connection strength from top 3Ns to each 4N receiving ≥20 synapses. The y-axis represents the total number of synapses from top 3Ns onto each 4N. Bottom: Fraction of the top 30 4Ns projecting to each brain region. (J) Left: All top 4Ns (blue) shown along with four mlSEZt cells (gray). Right: Five of the top 4N cell types (blue). Each 4N is shown along with 2–3 input 3Ns (red) and 2–3 mlSEZt inputs to those 3Ns (gray).

Neural circuitry downstream of mlSEZt neurons

Higher-order taste pathways in the superior protocerebrum have not been identified. What downstream circuits do mlSEZt cells connect to? We used trans-Tango to label neurons postsynaptic to mlSEZt, which represent third-order bitter neurons (3Ns). Interestingly, mlSEZt neurons themselves were prominently labeled (Figure 9B), suggesting that they form connections with each other. We also observed dense staining in the SLP and the neighboring superior intermediate protocerebrum (SIP), with many cell bodies located laterally (Figure 9B–C). There were very few projections out of the SLP/SIP, suggesting that 3Ns primarily comprise local neurons. The few projections out of the SLP/SIP mainly arborize in the superior medial protocerebrum (SMP) (Figure 9C), the region innervated by second-order bitter neurons comprising the medial tract (Figure 6), which suggests potential cross-talk between different second-order pathways. We also observed a medial projection that crossed the midline and innervated the contralateral SLP/SIP (Figure 9C). Finally, we observed a single projection that descended vertically from the SMP along the midline, potentially innervating the SEZ, and another projection descending directly from the SLP to the ventral brain (Figure 9C).

The dense staining in these trans-Tango experiments made it difficult to discern the morphology of single neurons. To identify individual 3Ns downstream of mlSEZt cells, we used the synaptic connectome of the fly brain (the hemibrain connectome; Scheffer et al., 2020). Because this connectome does not include the SEZ, we were restricted to analyzing the morphology and connectivity of mlSEZt cells outside the SEZ. We identified 21 neurons whose morphology strongly resembled that of mlSEZt cells (Figure 9D). These neurons were classified in the connectome as three different cell types: mAL3A (6 cells), mAL3B (5 cells), and mAL4 (10 cells). All three cell types show a ring-shaped projection in the SLP, with subtle differences in their arborizations (Figure 9D). We will refer to these 21 cells as mlSEZt cells, but it is not clear whether bitter-responsive mlSEZt cells include all 21 cells and three cell types, or only a subset. mlSEZt cells show dense SLP arborizations (Figure 8A) characteristic of mAL3B or mAL4 cells (Figure 9D), but they may also include mAL3A cells with sparser arborizations.

We next used the connectome to identify 3Ns receiving input from these 21 mlSEZt cells. Using a connection threshold of 3 synapses, we identified 201 unique 3Ns. Interestingly, four of the ten 3Ns receiving the strongest mlSEZt input are in fact mlSEZt cells, revealing strong interconnectivity that is consistent with trans-Tango labeling (Figure 9B). These mlSEZt interconnections show specific connectivity motifs. There are many strong connections from mAL3A to mAL3B, a few connections in the opposite direction (mAL3B to mAL3A), a few connections from mAL3A or mAL3B to mAL4, and almost no connections from mAL4 to the other types (Figure 9E). These results suggest a hierarchy where mAL3A strongly influences mAL3B, mAL3A and mAL3B weakly influence mAL4, and mAL4 does not influence the other types. Because these cells are GABAergic, these interactions may mediate inhibitory cross-talk that dampens or gates the responses of other mlSEZt cells.

After excluding interconnections between mlSEZt cells, most of the 3Ns receiving the strongest mlSEZt input are neurons projecting locally within the SLP and SIP (Figure 9F–H and Table 1), consistent with trans-Tango staining (Figure 9B–C). Other areas targeted by top 3Ns include nearby regions such as the SMP, lateral horn, superior clamp, and anterior ventrolateral protocerebrum. Very few 3Ns have long-range projections. Different mlSEZt cell types show different connectivity patterns to the top 3Ns (Figure 9G). Interestingly, mAL3A cells provide the weakest input to the top 3Ns (Figure 9G) despite providing the strongest input to other mlSEZt cells (Figure 9E), suggesting a distinction between lateral and feedforward mlSEZt circuitry.

Table 1
List of 3Ns receiving at least 20 synapses from mlSEZt cells.

Percent input from mlSEZt refers to the percent of total input synapses the cell receives that come from mlSEZt cells. Target regions for each cell refer to areas where the cell has output synapses. All target regions are ipsilateral to the mlSEZt projections unless specified with ‘(C)’, denoting contralateral innervation. Abbreviations follow NeuPrint conventions: AOTU (anterior optic tubercle), AVLP (anterior ventrolateral protocerebrum), GOR (gorget), IB (inferior bridge), ICL (inferior clamp), LH (lateral horn), MB (mushroom body; includes all lobes and calyxes), PLP (posterior lateral protocerebrum), SCL (superior clamp), SIP (superior intermediate protocerebrum), SLP (superior lateral protocerebrum), SMP (superior medial protocerebrum), VES (vest).

3N cell typeCell ID# mlSEZt input cellsmlSEZt input types# mlSEZt synapses% Input from mlSEZtTarget regions
SLP1914209735997mAL3A, mAL3B646.6AVLP, SCL, SIP, SLP
SLP1323598926696mAL3A, mAL4542.1AVLP, LH, SCL, SIP, SLP
SMP3895751974828mAL3A, mAL3B,mAL4483.4AVLP, PLP, SCL, SIP, SLP, SMP
SLP1914213135638mAL3A, mAL3B, mAL4475.0AVLP, LH, SCL, SIP, SLP
aSP-g3B4216509825mAL3A4612.9AVLP, PLP, SCL, SIP, SLP, SMP, SMP(C)
LHAV4l13602367244mAL4413.3AVLP, LH, SCL, SIP, SLP
SLP179_b4206238735mAL3B, mAL4383.7SCL, SIP, SLP
LHAV2f2_b5747101215mAL3A, mAL3B, mAL4387.9AVLP, LH, SCL, SLP
LHAV2f2_a6047355254mAL3B, mAL4369.4AVLP, LH, SCL, SLP
LHAV2f2_b5733463245mAL3A, mAL4355.5AVLP, LH, SCL, SLP
aSP-g3A3299190365mAL3A, mAL3B, mAL4356.3SIP, SLP, SMP(C), SMP
LHAV1e13902710334mAL3A, mAL3B, mAL4352.2AVLP, LH, SCL, SIP, SLP
SLP015_e3933404024mAL3B, mAL4336.0SIP, SLP
SLP0112975197367mAL3A, mAL4332.2LH, SCL, SIP, SLP
SLP179_b3913111865mAL3B, mAL4323.3SIP, SLP
SLP015_c3592401445mAL3B, mAL4283.1AVLP, SCL, SIP, SLP
aSP-g3B4854303365mAL3A279.2AVLP, SCL, SIP, SLP, SMP
SLP1875785219412mAL3B2711.8SIP, SLP
SLP1876078209372mAL3B267.3AVLP, SCL, SLP
SLP3762982543844mAL3B, mAL4250.6SIP, SLP, SMP
SLP21658130111197mAL3A, mAL3B, mAL4251.0AOTU, AVLP, GOR, GOR(C), IB, ICL, LH, PLP, SCL, SIP, SLP, SMP, SMP(C)
SLP2156085340973mAL3B, mAL4241.1AVLP, PLP, SCL, SLP, VES
SLP015_c3599269234mAL3B, mAL4235.4SIP, SLP
SLP1874837118113mAL3B238.9SLP
SLP25958130407074mAL3B, mAL4231.7AVLP, SCL, SIP, SLP, SMP
AVLP0244209651173mAL3B221.2AVLP, LH, PVLP, SCL, SIP, SLP
LHCENT13288612824mAL4220.2AVLP, LH, MB, SCL, SIP, SLP, SMP
LHAV6a1058130472553mAL4226.5LH, SIP, SLP
SLP015_e3292066282mAL4213.2SIP, SLP
SMP5504526894943mAL3B, mAL4210.4AVLP, LH, PLP, SCL, SIP, SLP, SMP
SLP0184516631724mAL3A, mAL3B, mAL4203.1AVLP, LH, SCL, SIP, SLP
SLP05758130199554mAL3B, mAL4200.5AVLP, LH, PLP, SCL, SIP, SLP

Most of the top 3Ns (Table 1) have not been previously studied, but some have been identified in other connectomic studies. Two top 3N types (SLP259 and SLP018) are third-order olfactory neurons in a pheromone-sensing circuit (Taisz et al., 2022) and one top 3N, LHCENT1, connects the mushroom body to olfactory circuits in the lateral horn (Bates et al., 2020). Several other 3Ns are lateral horn neurons that may participate in olfactory processing (Schlegel et al., 2021). Thus, mlSEZt neurons likely provide input to neurons integrating olfactory and gustatory information. The inputs to another top 3N type, aSP-g3, were recently reconstructed using a whole-brain connectome dataset (Taisz et al., 2022; Zheng et al., 2018). mlSEZt cells (mAL3) as well as several other second-order taste neurons were identified as inputs to aSP-g3, suggesting that some 3Ns integrate input from multiple types of second-order neurons.

Next, we used the connectome to identify postsynaptic partners of 3Ns, which represent fourth-order bitter neurons (4Ns). Starting from the top 3Ns (those receiving ≥20 mlSEZt synapses; 32 cells), we identified over 1700 4Ns total and 322 4Ns that receive at least 20 synapses from the top 3N population (Figure 9I). The 4Ns include several mlSEZt cells, revealing feedback connections that likely modulate mlSEZt output. We focused on the 30 4Ns receiving the strongest 3N input, excluding feedback connections to mlSEZt cells (Table 2). Most of these top 4Ns send axonal projections to the SLP, SIP, or nearby areas, but collectively they target more diverse brain regions than the 3Ns (Figure 9I–J). 4N projections to the SMP are especially prominent, with 90% of top 4Ns sending output to this region. 60% of the top 4Ns project to the lateral horn and 43% project to the mushroom body, regions that mediate innate and learned olfactory processing, respectively. The top 4Ns include neurons that regulate egg-laying (oviIN and oviDNa; two oviDNb cells are also 4Ns but not in the top 30; Wang et al., 2020), mushroom body output neurons (Aso et al., 2014a), and the PPL201 (PPL2a) dopaminergic neuron (Li et al., 2020; Boto et al., 2019). The 322 4Ns receiving strong 3N input do not include PPL1 dopaminergic neurons, which convey bitter information to the mushroom body for aversive learning (Kirkhart and Scott, 2015), and this may correlate with the inability of mlSEZt cells to elicit learned aversion (Figure 8H).

Table 2
List of top 30 4Ns.

Top 30 4Ns were selected and sorted based on the total number of synapses from top 3Ns. Percent input from 3Ns refers to the percent of total input synapses the cell receives that come from the top 3Ns. Target regions for each cell refer to areas where the cell has output synapses. All target regions are ipsilateral to the mlSEZt projections unless specified with ‘(C)’, denoting contralateral innervation. Abbreviations not defined in Table 2: CAN (cantle), CRE (crepine), LAL (lateral accessory lobe), POC (posterior optic commissure), PVLP (posterior ventrolateral protocerebrum), SPS (superior posterior slope), mALT (medial antennal lobe tract).

4N cell typeCell ID# 3N input cells3N input cell types# synapses from 3Ns% input from 3NsTarget regions
SMP54829758058916AVLP024, LHAV1e1, SLP015_c, SLP015_e, SLP057, SLP179_b, SLP191, SLP216, SMP389, SMP550, aSP-g3A, aSP-g3B3636.3AVLP, LH, PLP, SCL, SIP, SLP, SMP
SLP2793605918608LHAV1e1, LHCENT1, SLP015_e, SLP018, SLP179_b, SLP216, SMP550, aSP-g3A3548.4SCL, SIP, SLP, SMP, SMP(C)
LHCENT45175062654LHAV4l1, LHCENT1, SLP057, SMP3892974.6AVLP, LH,, MB, PLP, SCL, SIP, SLP, SMP, mALT
DM1_lPN5426348181LHCENT12472.4AL, LH, MB, SCL, SLP, mALT
LHMB158130209882LHCENT1, SLP0572383.5CRE, LH, MB, SCL, SIP, SLP, SMP
SLP38829825861111AVLP024, SLP015_e, SLP018, SLP179_b, SLP191, SLP216, SMP389, aSP-g3B2033.1SCL, SIP, SLP, SMP, SMP(C)
SMP5504526894947AVLP024, LHAV1e1, SLP018, SLP216, SMP389, aSP-g3B2014.2AVLP, LH, PLP, SCL, SIP, SLP, SMP
oviIN4231011892SMP389, SMP5501900.8CAN, CRE, CRE(C), GOR, IB, LAL, SIP, SIP(C), SMP, SMP(C), SPS, VES
LHCENT933026894010AVLP024, LHAV1e1, LHCENT1, SLP132, SLP191, SMP389, aSP-g3A, aSP-g3B1771.4AOTU, AVLP, LH, MB, SCL, SIP, SLP, SMP
SLP212581298052914AVLP024, SLP015_e, SLP018, SLP057, SLP132, SLP191, SLP216, SMP389, SMP550, aSP-g3A, aSP-g3B1456.7SIP, SLP, SMP, SMP(C)
SMP1082982585133SLP057, SMP389, SMP5501450.6CRE, CRE(C), LAL, LH, MB, MB(C), SCL, SIP(C), SIP, SLP, SMP, SMP(C)
oviDNa5506556682SLP216, SMP55013712.3CRE, SCL, SIP, SLP, SMP, SMP(C), VES
MBON1858130208281LHCENT11321.0LH, MB, SCL, SIP, SLP
LHPD4c14216418599LHAV1e1, LHAV4l1, LHCENT1, SLP011, SLP018, SLP057, SLP132, SLP1871302.5LH, SCL, SIP, SLP, SMP
SLP1133905895916AVLP024, LHAV2f2_b, LHAV4l1, LHCENT1, SLP13212710.5LH, SIP, SLP, SMP
LHPV10b16047097272LHCENT1, SLP0571252.8CRE, LH, MB, PLP, PVLP, SCL, SIP, SLP, SMP
SMP1566737767693SLP216, SMP389, SMP5501242.1CRE, GOR(C), IB, ICL(C), ICL, LAL, MB, SMP, SMP(C), SPS, SPS(C)
SMP38558130837802SMP389, SMP5501242.0CRE, CRE(C), LAL, LAL(C), SCL, SIP, SIP(C), SMP, SMP(C),
SLP44032887047213LHAV1e1, SLP015_c, SLP015_e, SLP018, SLP057, SLP179_b, SLP191, SLP376, aSP-g3B1233.0LH, SCL, SIP, SLP, SMP, SMP(C)
LHPV5e13286110045LHAV1e1, LHAV4l1, LHCENT1, SLP015_e, SLP1321231.1CRE, CRE(C), LH, MB(C), SCL, SIP, SIP(C), SLP, SMP, SMP(C)
SMP0296040704333SLP216, SMP389, SMP5501206.7AVLP, LH, MB, PLP, SCL, SIP, SLP, SMP
SMP31158130493782SMP389, SMP5501177.5AVLP, ICL, PLP, SCL, SIP, SLP, SMP
PPL2013285337616LHAV1e1, LHCENT1, SLP011, SLP015_e, SLP057, SLP179_b1171.4AVLP, CRE, LH, MB, PLP, POC, SCL, SIP, SLP, SMP, mALT
SMP10958130096202SMP389, SMP5501071.7AOTU, CRE, CRE(C), LAL, LAL(C), MB, SIP, SMP, VES
SMP0295413478114SLP215, SLP216, SMP389, SMP5501065.8AVLP, MB, PLP, SCL, SIP, SLP, SMP
SMP50336131280812AVLP024, LHCENT1, SLP011, SLP015_e, SLP018, SLP132, SLP187, SLP215, SMP389, SMP550, aSP-g3B1021.5AVLP, CRE, LH, MB, PLP, SCL, SIP, SIP(C), SLP, SMP, SMP(C)
SLP1134212713058LHAV2f2_a, LHAV2f2_b, LHAV4l1, LHCENT1, SLP132, SLP18710210.0LH, SIP, SLP, SMP, SMP(C)
MBON184571964441LHCENT1973.6LH, MB(C), SCL, SIP, SIP(C), SLP
SLP14958130093129LHAV1e1, LHAV4l1, LHCENT1, SLP015_c, SLP018, SLP057, SLP179_b, SLP376974.3SIP, SLP, SMP
SLP441581307854213LHAV1e1, LHCENT1, SLP011, SLP015_c, SLP015_e, SLP057, SLP179_b, SLP191, SLP259, SLP376, aSP-g3B963.3AVLP, SCL, SIP, SLP, SMP

The population of 4Ns also included descending neurons, neurons that project to the ventral nerve cord and regulate body movements such as locomotion. Three of these descending neurons (types oviDNa and oviDNb) drive egg-laying behavior (Wang et al., 2020), while a fourth, DNp32, is activated during walking (Aimon et al., 2022). Given that mlSEZt activation elicits turning (Figure 8F), we searched for downstream connections to DNa02, a descending neuron that controls turning (Rayshubskiy et al., 2020). Individual mlSEZt neurons connected to DNa02 via two or three intermediate neurons. We identified many possible connection paths, but a strong and commonly observed path was mlSEZt → SMP389 → AOTU012 → DNa02, with the latter two connections being exceptionally strong (40 and 140 synapses, respectively).

Together, these analyses suggest that bitter information conveyed by mlSEZt cells is processed in the SLP and SIP before being relayed to other areas, including the SMP, lateral horn, mushroom body, and descending neurons. Strong lateral connections between mlSEZt cells and feedback connections from 3Ns to mlSEZt neurons suggest the presence of extensive local processing. Several 3Ns and 4Ns likely integrate bitter information from mlSEZt with olfactory input, and at least one 3N integrates input from multiple second-order taste neurons. These studies provide the first glimpse into circuits for bitter processing in the higher brain.

Discussion

In this study, we investigated how different bitter taste inputs are processed in the brain to guide behavior. We took advantage of highly specific genetic tools and a wide range of behavioral assays, representing a systematic and rigorous approach to answering this question. Optogenetic experiments revealed that bitter neurons in different organs elicit broad and highly overlapping effects on behavior, suggesting that these inputs converge onto common downstream pathways. However, we also observed behavioral differences that suggest that these inputs are ‘selectively integrated’ – downstream pathways may selectively or preferentially receive input from certain bitter neuron subsets. Transsynaptic tracing of second-order pathways supported this interpretation, revealing that bitter neurons in different organs connect to overlapping second-order pathways, but with biased connectivity.

To begin to understand how second-order bitter neurons integrate sensory input and regulate behavior, we characterized a new type of second-order bitter neuron, mlSEZt. We found that mlSEZt neurons receive convergent input from multiple organs and regulate a subset of taste-related behaviors. mlSEZt cells transmit feedforward inhibition to the SLP, and this information is then relayed to several areas including the SMP, lateral horn, and mushroom body. mlSEZt is one of the only types of second-order bitter neurons that has been functionally studied in the adult fly (Kim et al., 2017; Bohra et al., 2018). Moreover, our studies of downstream circuits from mlSEZt are the first to uncover third- and fourth-order taste pathways in the higher brain.

Roles of bitter neuron subsets

Conventional models have proposed that taste neurons in different organs have distinct behavioral roles (Dethier, 1976; Scott, 2018). We were therefore surprised to find that activating bitter neurons within the leg, labellum, or pharynx had broad and largely overlapping effects on behavior. Activating bitter neurons in any of the three organs suppressed proboscis extension and feeding, enhanced locomotor speed and turning, and elicited innate aversion. Neurons in the leg and labellum also elicited learned aversion. These results imply that bitter inputs from different organs converge in the brain to drive a common set of aversive behaviors. However, the effects of different neuronal subsets were not identical. For example, activating pharyngeal bitter neurons elicited stronger innate aversion than leg or labellar (S-a + I-a) neurons, but the latter two subsets elicited stronger learned aversion than pharyngeal neurons (Figures 4 and 5). Differences like these cannot be explained by technical issues such as the number of neurons activated or strength of Gal4 expression. Together these results suggest a model in which downstream pathways selectively integrate different bitter inputs, enabling specific organs to preferentially regulate certain aspects of behavior (Figure 5B).

The strong similarities between the effects of bitter neurons in different organs contrast with studies of sugar taste. Sugar neurons in different organs have different effects on locomotion, feeding responses, and egg-laying (Thoma et al., 2016; Murata et al., 2017; Chen et al., 2021; Chen et al., 2022). Moreover, there are specific differences between the roles of sugar and bitter neurons within an organ. For example, pharyngeal bitter neurons strongly regulate proboscis extension (Figure 2A), whereas pharyngeal sugar neurons do not (Chen et al., 2021). These findings suggest that inputs for appetitive and aversive tastes may be integrated in fundamentally different ways. Perhaps organ-specific roles for appetitive tastes are needed to generate the appropriate sequence of motor actions during feeding, whereas aversive behaviors are more general. It may be adaptive for any organ to be capable of driving nearly any type of aversive response.

We were also curious whether different subsets of bitter neurons within the same organ would show similar or distinct behavioral effects. Due to limitations in driver lines available when we initiated this study, we only compared two such subsets: labellar neurons belonging to the S-a + I-a classes versus the S-b class. S-a + I-a neurons drove aversive effects in all behavioral assays tested, whereas S-b neurons had a more limited effect. However, there are fewer S-b neurons (3) than S-a + I-a neurons (13) and the Gal4 line used to label S-b neurons seemed to be expressed more weakly, limiting our interpretation of these results. Supporting the notion of class-specific behavioral effects, a previous study showed that two types of pharyngeal bitter neurons regulate different aspects of feeding behavior (Chen et al., 2019).

In general, the existence of multiple bitter neuron classes in each organ, with different tuning profiles (Weiss et al., 2011), would seem to imply that they have different functional roles. If different classes drive distinct aversive behaviors, this would enable the fly to respond differently depending on which bitter compound is detected. Alternatively, input from different classes may be combinatorially integrated to encode bitter identity. In this case, different compounds could elicit different behavioral responses without each class having a distinct behavioral role.

Second-order pathways for bitter taste processing

Through transsynaptic tracing, we identified three major second-order pathways that relay bitter input to higher brain areas, which have also been described in other recent studies (Chen et al., 2019; Snell et al., 2022). Each pathway receives input from multiple organs, although our experiments do not confirm that different inputs converge onto the same individual cells. Labellar bitter neurons connect to all three projection tracts, whereas tarsal bitter neurons primarily connect to the lateral and mediolateral tracts, and pharyngeal bitter neurons primarily innervate the medial tract. This biased connectivity pattern may provide a neuronal substrate for our behavioral results: different bitter neuron subsets may elicit overlapping but non-identical behavioral effects because they activate overlapping but non-identical second-order taste pathways. For example, bitter neurons in the leg and labellum elicited learning while pharyngeal neurons did not, and this difference may relate to the observation that only leg and labellar neurons connected strongly to the lateral and mediolateral tracts.

Bitter-sensing neurons in different organs show striking differences in their axonal projection patterns (Wang et al., 2004; Kwon et al., 2014). We were therefore surprised that trans-Tango labeling did not reveal any obvious pathways receiving input from just one organ. In the sugar-sensing circuit, several organ-specific pathways have been identified. IN1 interneurons in the SEZ receive sugar input specifically from the pharynx and regulate sugar ingestion (Yapici et al., 2016). A set of sugar-responsive serotonergic neurons in the SEZ receives input from the labellum, but not the leg, and regulates sugar intake and insulin release (Yao and Scott, 2022). Some tarsal sugar-sensing neurons arborize in the ventral nerve cord (Kwon et al., 2014) and activate leg-specific ascending neurons (Kim et al., 2017), one of which regulates egg-laying behavior (Chen et al., 2022).

The new type of second-order bitter neuron that we characterized, mlSEZt, receives bitter input from both the leg and labellum. This is also the case for the only other second-order bitter projection neuron that has been previously studied, TPN3 (Kim et al., 2017). Bitter inputs from multiple organs are thus integrated at the first synapse, at least within certain second-order pathways. Aside from cross-organ integration, mlSEZt neurons do not show significant transformations of bitter encoding: their taste selectivity, response dynamics, and experience-dependent modulation closely resemble responses in sensory neurons. It will be interesting to compare bitter encoding across different second-order neurons. Different second-order pathways may convey different features of the taste stimulus, perhaps analogous to the parallel ‘what’ and ‘where’ pathways in the visual and auditory systems (Goodale and Milner, 1992; Lee, 2013). For example, different second-order neurons may transmit the bitter ON versus OFF response, thus encoding the start and end of the stimulus. Certain neurons may encode bitter identity while others integrate multiple taste inputs to encode the overall taste valence.

How do individual types of second-order bitter neurons contribute to behavior? Activating mlSEZt neurons revealed that they regulate feeding responses and turning but not locomotor speed, innate preference, or learning. The behaviors not affected by mlSEZt neurons may be regulated by other pathways, and one possibility is that different pathways strictly control different behaviors. This has been suggested in the mammalian taste system, in which a cortico-amygdalar projection regulates licking behavior but is dispensable for reporting the identity of the stimulus in a discrimination task (Wang et al., 2018). Similarly, distinct pathways in the olfactory system mediate innate versus learned odor responses (Bear et al., 2016; Amin and Lin, 2019). Having separate pathways for different behavioral responses allows a circuit to readily modulate individual behaviors depending on context or state.

Alternatively, different second-order pathways may have overlapping behavioral roles. Perhaps different pathways encode different features of the stimulus, as discussed above, and this information is combined downstream in specific ways appropriate for regulating each behavior. In support of this combinatorial model, we found that different bitter-sensing neurons elicited overlapping behavioral effects even when they had largely non-overlapping projection targets. Tarsal bitter neurons mainly project to the lateral and mediolateral second-order pathways, whereas pharyngeal bitter neurons mainly project to the medial pathway, but both subsets strongly affected feeding responses, locomotor speed, and turning. In addition, activation of either mlSEZt (Figure 8B) or TPN3 (Kim et al., 2017) suppresses proboscis extension, representing parallel second-order pathways with common behavioral roles. Thus, different second-order pathways may converge downstream to regulate common behaviors. In support of this model, many neurons downstream of mlSEZt (the mediolateral tract) project to the SMP (the target of the medial tract), suggesting potential convergence. Moreover, one of the 3Ns downstream of mlSEZt, aSP-g3, receives input from multiple second-order taste neurons (Taisz et al., 2022).

Third- and fourth-order taste pathways

The population of second-order taste neurons has only recently been identified (Talay et al., 2017; Chen et al., 2019; Snell et al., 2022), and almost nothing is known about how taste information is processed further downstream. To gain insight into downstream bitter processing, we traced neural circuits downstream of mlSEZt cells. We first observed that mlSEZt cells connect to each other, and these are mainly unidirectional connections suggestive of hierarchical interactions. Because mlSEZt cells are GABAergic, specific mlSEZt cells likely inhibit other mlSEZt cells and gate their ability to transmit feedforward information. It will be interesting to determine how this motif shapes information flow, and whether mlSEZt cells representing the sources and targets of these connections have different response properties. Interestingly, the mlSEZt subtype providing the strongest connections to other mlSEZt cells provides the weakest connections to top 3Ns, suggesting a division of labor between lateral and feedforward output from mlSEZt cells.

Different mlSEZt cells also show different connectivity patterns to 3Ns, suggesting that parallel subcircuits may exist downstream of the mlSEZt population. 3Ns represent a broad set of neurons targeting the SLP, SIP, and nearby regions such as the lateral horn. Because mlSEZt neurons are inhibitory, 3Ns should be inhibited by bitter taste and may therefore represent neurons promoting appetitive behaviors. Several 3Ns are likely to process olfactory information, suggesting that they integrate smell and taste inputs. As mentioned above, at least one 3N receives input from multiple second-order taste neurons (Taisz et al., 2022). It will be interesting to determine what kinds of taste input it integrates and whether second-order to 3N convergence is a common feature of the circuit.

While 3Ns mainly target a few brain regions, 4Ns project more broadly. 4Ns include mlSEZt cells, indicating the presence of feedback circuits that may facilitate or dampen mlSEZt transmission. 4Ns include neurons likely to participate in olfactory processing, again suggesting cross-talk between olfactory and taste pathways. Many of the top 4Ns project to the lateral horn and mushroom body, regions that mediate innate and learned olfactory processing, respectively. 90% of top 4Ns project to the SMP, a region that contains neurosecretory cells that regulate feeding (Nässel and Zandawala, 2020). 4Ns also include several neurons that regulate egg-laying, suggesting that mlSEZt may convey bitter information to modulate egg-laying decisions. Flies have been shown to display either a preference (Joseph and Heberlein, 2012) or aversion (Dweck et al., 2021) for laying eggs on bitter substrates, which may depend on the substrate composition or context. The activity of the egg-laying neuron oviDNb, one of three oviDNs in the 4N population, reflects the sugar concentration of the substrate and displays a rise-to-threshold signal that drives egg-laying (Vijayan et al., 2021). Thus, oviDNs are poised to integrate multiple types of taste input to control egg-laying behavior.

Aside from oviDNs, the population of 4Ns that we analyzed (those receiving ≥20 synapses from the 3N population; 322 cells) included only one other descending neuron, DNp32. This suggests that additional layers of processing occur before mlSEZt input reaches neurons controlling locomotor behaviors. We identified the turning neuron DNa02 as a fifth-order neuron. While descending neurons control leg and body movements, motor neurons controlling proboscis movements and food consumption are located in the SEZ (Schwarz et al., 2017), a region not included in the hemibrain connectome. Future analyses of whole-brain connectome datasets (e.g., Zheng et al., 2018) should enable us to identify connections between mlSEZt cells and motor neurons controlling feeding.

Principles of sensory integration

Within a single taste modality, whether and how different channels of taste input are integrated has been largely unexplored, especially for aversive tastes. Our results show that bitter inputs from different organs are integrated early in the circuit and drive an overlapping, but not identical, set of behaviors. It will be interesting to determine whether similar principles govern the integration of taste input in other taste modalities and other organisms. Humans and other mammals have taste neurons in multiple locations, including three distinct areas of the tongue, the soft palate, pharynx, and internal organs such as the gut (Liman et al., 2014; Depoortere, 2014; Travers and Nicklas, 1990). Moreover, mammalian bitter-sensing cells in the tongue show heterogeneous responses (Caicedo and Roper, 2001) and express different combinations of bitter receptors (Behrens et al., 2007; Voigt et al., 2012), similar to flies, suggesting the presence of different functional classes.

It will also be interesting to compare principles of taste processing and integration with other sensory systems. Like taste, the olfactory system contains well-defined sensory input channels. Because odors activate overlapping ensembles of sensory cells, inputs across the population must be integrated to determine odor identity and behavior, and this begins to occur at the second synapse (Groschner and Miesenböck, 2019). However, certain channels of olfactory information may represent ‘labeled-line’ circuits that are processed separately from other inputs (Haverkamp et al., 2018). The mechanosensory system similarly contains well-defined input channels that detect specific stimulus features (e.g., indentation or vibration), suggesting that these channels could potentially mediate distinct aspects of touch perception (Abraira and Ginty, 2013). However, recent studies suggest that inputs from different mechanosensory types are integrated, beginning at the first synapse, to create mixed representations (Emanuel et al., 2021; Chirila et al., 2022). Uncovering when and how sensory circuits integrate different inputs, and how this relates to perception and behavior, lies at the core of understanding how we interpret signals from the world.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Genetic reagent (Drosophila melanogaster)Gr33a-Gal4Moon et al., 2009BDSC: 31425
Genetic reagent (Drosophila melanogaster)Gr58c-Gal4Weiss et al., 2011BDSC: 57646
Genetic reagent (Drosophila melanogaster)Gr59c-Gal4Weiss et al., 2011BDSC: 57650
Genetic reagent (Drosophila melanogaster)Gr22f-Gal4Weiss et al., 2011BDSC: 57610
Genetic reagent (Drosophila melanogaster)Gr9a-Gal4Weiss et al., 2011BDSC: 57596
Genetic reagent (Drosophila melanogaster)R29F12-Gal4Jenett et al., 2012BDSC: 49495
Genetic reagent (Drosophila melanogaster)R55E01-Gal4Jenett et al., 2012BDSC: 39117
Genetic reagent (Drosophila melanogaster)R29F12-ADDionne et al., 2018BDSC: 71164
Genetic reagent (Drosophila melanogaster)R55E01-DBDDionne et al., 2018BDSC: 69662
Genetic reagent (Drosophila melanogaster)Gr66a-lexAThistle et al., 2012BDSC: 93024
Genetic reagent (Drosophila melanogaster)VGAT-lexADeng et al., 2019BDSC: 84441
Genetic reagent (Drosophila melanogaster)ChAT-lexADeng et al., 2019BDSC: 84379
Genetic reagent (Drosophila melanogaster)UAS-Chrimson-TdTDuistermars et al., 2018N/A
Genetic reagent (Drosophila melanogaster)UAS-GTACR1-TdTB NoroN/A
Genetic reagent (Drosophila melanogaster)UAS-Kir2.1Baines et al., 2001BDSC: 6595
Genetic reagent (Drosophila melanogaster)trans-Tango reporter (UAS-Myr-GFP, QUAS-mtdTomato; trans-Tango)Talay et al., 2017BDSC: 77124
Genetic reagent (Drosophila melanogaster)UAS-GCaMP6fChen et al., 2013BDSC: 42747
Genetic reagent (Drosophila melanogaster)UAS-TdTVK5D HattoriN/A
Genetic reagent (Drosophila melanogaster)UAS-TdTp40G Rubin and B PfeifferBDSC: 32222
Genetic reagent (Drosophila melanogaster)lexAop-GCaMP6fD Kim; Hattori et al., 2017BDSC 44277
AntibodyAnti-GFP (chicken polyclonal)Aves LabsCat# GFP-1020; RRID: AB_100002401:1000
AntibodyAnti-DsRed (rabbit polyclonal)ClontechCat# 632496; RRID: AB_100134831:500
AntibodyAnti-bruchpilot (nc82; mouse monoclonal)Development Studies Hybridoma BankCat# nc82; RRID: AB_23148661:10
AntibodyAlexa Fluor 488 (goat anti-chicken polyclonal)Life TechnologiesCat# A11039; RRID: AB_25340961:500
AntibodyAlexa Fluor 568 (goat anti-rabbit polyclonal)Life TechnologiesCat# A11036; RRID: AB_105635661:500
AntibodyAlexa Fluor 633 (goat anti-mouse polyclonal)Life TechnologiesCat# A21052; RRID: AB_25357191:500
Software, algorithmPrism, version 9GraphPadN/A
Software, algorithmMATLABMathworksN/A
Software, algorithmFlyTrackerCaltech; Eyjolfsdottir et al., 2014N/Ahttp://www.vision.caltech.edu/Tools/FlyTracker/

Fly stocks and maintenance

Request a detailed protocol

Flies were reared at 25°C on standard cornmeal or cornmeal-molasses food, with the exception of flies for trans-Tango experiments which were reared at 20°. Bitter Gal4 lines and UAS-Chrimson were outcrossed into a 2U wild-type background for at least five generations. Unless otherwise specified, experiments were performed on mated females, with the exception of trans-Tango experiments that used males (the location of transgenes on the X chromosome leads to enhanced expression in males). Behavioral assays were performed on 3- to 7-day-old flies, calcium imaging was performed on 2- to 3-week-old flies (to ensure strong GCaMP expression), trans-Tango staining was performed on ~4-week-old flies (to ensure strong labeling), and other immunostaining experiments were performed on 1- to 2-week-old flies. Flies used for optogenetic experiments were maintained in constant darkness and fed on food containing 1 mM all trans-retinal for 3–5 days prior to testing. PER experiments used 1 day starved flies. optoPAD experiments used flies starved for 1 day (neuronal activation with sweet substrate) or 2 days (neuronal silencing with sweet+bitter substrate). Flies were food-deprived by placing them in an empty vial with a wet piece of Kimwipe containing all 1 mM trans-retinal.

PER assay

Request a detailed protocol

PER was tested using previously described methods (Devineni et al., 2019). Briefly, flies were anesthetized on ice, immobilized on their backs with myristic acid, and the two anterior pairs of legs were glued down so that the labellum was accessible for taste stimulation. Flies recovered from gluing for 30–60 min in a humidified chamber. Flies were water-satiated before testing. Tastants were briefly applied to the labellum using a small piece of Kimwipe. Tastant concentrations are specified in the figure legends. For optogenetic stimulation, a red (617 nm) or green (530 nm) LED was manually turned on just before the taste stimulus and remained on for the duration of the stimulus.

Approximately 10 flies were sequentially tested in each experiment, and the percent of flies showing PER to each stimulus was manually recorded. Only full proboscis extensions were counted as PER. At the end of each experiment, flies were tested with a positive control (500 mM sucrose) and were excluded from analysis if they did not respond. For statistical analyses, each set of ~10 flies was considered to be a single data point (‘n’).

optoPAD assay

Request a detailed protocol

The optoPAD was purchased from Pavel Itskov at Easy Behavior (https://flypad.rocks/), who also provided code to run the assay using Bonsai and to process the data using MATLAB. The design of the optoPAD is described in Moreira et al., 2019, and data processing methods are described in Itskov et al., 2014. Experiments were performed as described in Moreira et al., 2019. Optogenetic activation and silencing were performed using 2.2 or 3.5 V stimulation, respectively. Light onset was triggered immediately upon detection of an interaction with the specified food source, and the light remained on for 1.5 s. Tastants were mixed into 1% agarose.

Locomotor and preference assays

Request a detailed protocol

The behavioral arena for testing locomotor and preference behaviors was built based on designs by the Rubin lab at the Janelia Research Campus (Aso et al., 2014b; Aso and Rubin, 2016). The arena contains a circular chamber with a glass cover, and flies are filmed from above using a USB camera (Point Grey). The infrared light for illumination and the red (627 nm) LED array for optogenetic stimulation are located just beneath the chamber. Each quadrant of the arena has an inlet for air flow, allowing for odor delivery, and air is removed through a vacuum port at the center of the arena. We did not use air flow for locomotor or innate preference assays. 20–30 flies were tested per trial. Flies were loaded into the chamber using mouth aspiration and were given at least 3 min to habituate before the experiment started. Flies were filmed at 30 frames/s.

For learning experiments using odor, 400 mL/min air was split into two streams that each flowed to one pair of opposing quadrants and could be odorized independently. Each stream was odorized by flowing air through bottles containing odor dilutions. Each stream was then split again to provide input to two opposing quadrants, representing a flow rate of 100 mL/min air into each quadrant. The odors used were 3-octanol (1:1000) and 4-methylcyclohexanol (1:750). Odors were diluted in mineral oil and 2 mL of each odor solution was pipetted onto a piece of thick filter paper (Thermo Fisher Scientific, #88615) placed inside of a glass odor bottle (Cole-Parmer, #EW-99535-16).

The effects of taste neuron activation on locomotion and innate preference were assayed sequentially in the same flies. To quantify locomotor effects, light stimulation was presented for 5 s. Flies were filmed for >1 min before stimulation and 2 min after stimulation, although only the 5 s prior to stimulation was used to quantify baseline behavior. Each of the three light intensities was delivered to the same flies in increasing order of intensity, with several minutes between stimulations to ensure that the behavior had recovered. After an additional rest period, the same flies were then tested for innate preference by delivering low intensity light stimulation to two opposing quadrants for 30 s. The flies then had a 30 s rest period without light, followed by low intensity light stimulation in the other two quadrants for 30 s. Switching the light quadrants ensured that we could assess light preference independently of any spatial bias. This protocol was then repeated sequentially with medium and high intensity light.

To test learned odor preference, the CS+ odor was presented in all quadrants for 1 min along with light stimulation. Light stimulation started 5 s after odor valve opening and was delivered in 1 Hz pulses, following the protocol used in previous learning studies (Aso and Rubin, 2016). After a 1 min rest, the CS- odor was presented alone for 1 min. Following a 1 min rest, the CS+ and CS- odors were simultaneously delivered to different sets of opposing quadrants for 1 min, allowing the flies to choose between the odors. After another 1 min rest, the two odors were presented again for 1 min but the odor quadrants were switched to control for any spatial bias. Which odor was used as the CS+ or CS- was counterbalanced across trials. The light intensity used for learning experiments was 30 µW/mm2.

Fly videos were analyzed using FlyTracker (Eyjolfsdottir et al., 2014), which quantified the position, forward velocity, and angular velocity of each fly at each time point. FlyTracker output was further analyzed in MATLAB to quantify preference for light or odor and to generate the traces shown in the figures. Preference index (PI) values were quantified in 1 s bins and calculated as (# flies in light quadrants – # flies in non-light quadrants)/total # flies. PI for the conditioned odor was calculated similarly as (# flies in CS + quadrants – # flies in CS- quadrants)/total # flies. Forward and angular velocities were averaged over 0.33 s bins (10 frames). To quantify locomotor changes at light onset, we averaged forward or angular velocity over the first 1 s of light presentation and compared these values to baseline values averaged over the 4 s preceding light onset. To quantify the change in speed after light stimulation, we averaged forward velocity over a 5 s period following light offset and compared it to the same baseline (pre-light) value described above. For statistical analyses of these data, each trial was considered to be a single data point (‘n’). Further analyses are described below in the ‘Statistical analyses’ section.

Immunostaining and validation of expression patterns

Request a detailed protocol

Immunostaining experiments were performed as previously described (Devineni et al., 2021). Briefly, brains were dissected in phosphate buffered saline (PBS), fixed for 15–20 min in 4% paraformaldehyde, washed multiple times with PBS containing 0.3% Triton X-100 (PBST), blocked with 5% normal goat serum for 1 hr, incubated with primary antibodies at 4°C for 2–3 days, washed in PBST, incubated with secondary antibodies at 4°C for 1–2 days, washed in PBST and PBS, and mounted in Vectashield. Primary antibodies used were chicken anti-GFP (1:1000), rabbit anti-DsRed (1:500), and mouse nc82 (1:10). Secondary antibodies used were Alexa Fluor 488 goat anti-chicken (1:500), Alexa Fluor 568 goat anti-rabbit (1:500), and Alexa Fluor 633 goat anti-mouse (1:500). To confirm expression of bitter sensory Gal4 lines in taste organs, endogenous expression of UAS-Chrim-TdT was imaged without immunostaining. Taste organs were removed, immersed in PBS, and imaged within 1–2 hr.

Images for trans-Tango staining and characterization of mlSEZt expression (Figure 6, Figure 7A, Figure 8A, Figure 9B, Figure 6—figure supplement 1, and Figure 9—figure supplement 1) were acquired on a Zeiss LSM 710 system. Imaging for validating sensory Gal4 expression (Figure 1C) and neurotransmitter colabeling (Figure 9A and Figure 9—figure supplement 2) was performed on a Keyence BZ-X810 microscope. All images were processed using Fiji.

Calcium imaging

Request a detailed protocol

Calcium imaging was performed as previously described (Devineni et al., 2019; Devineni et al., 2021). Flies were taped on their backs to a piece of clear tape in an imaging chamber. For proboscis taste stimulation, fine strands of tape were used to restrain the legs, secure the head, and immobilize the proboscis in an extended position. For leg taste stimulation, the two forelegs were immobilized using tape and parafilm with the distal segment exposed. An imaging window was cut on the anterior surface of the head, the antennae were removed, and the esophagus was cut. The brain was immersed in modified artificial hemolymph in which 15 mM ribose is substituted for sucrose and trehalose (Marella et al., 2006; Wang et al., 2003).

Images were acquired using a two-photon laser scanning microscope (Ultima, Bruker) equipped with an ultra-fast Ti:S laser (Chameleon Vision, Coherent) that is modulated by pockel cells (Conoptics). The excitation wavelength was 925 nm. Emitted photons were collected with a GaAsP photodiode detector (Hamamatsu) through a 60X water-immersion objective (Olympus). Images were acquired using the microscope software (PrairieView, Bruker). A single plane was chosen for imaging in each experiment. For most experiments, images were acquired at 256 by 256 pixels with a scanning rate of 3–4 Hz. The resolution was decreased and scanning rate was increased to ~6–7 Hz for experiments using repeated taste stimulation.

Tastants were delivered to the labellum or foreleg as previously described (Devineni et al., 2019; Devineni et al., 2021). Briefly, we used a custom-built solenoid pinch valve system controlled by MATLAB software via a data acquisition device (Measurement Computing). Solenoid pinch valves (Cole Parmer) were opened briefly (~10 ms) to create a small liquid drop at the end of a 5 µL glass capillary, positioned such that the drop would make contact with the labellum or leg. Tastants were removed by a vacuum line controlled by a solenoid pinch valve. Proper taste delivery was monitored using a side-mounted camera (Veho VMS-004). Five s taste stimulation was used for all experiments. At least three trials of each stimulus were given to each fly. Other than experiments explicitly using repeated taste stimulation, at least 1 min rest was given between trials to avoid habituation. Tastants were used at the following concentrations: 10 mM quinine, 10 mM denatonium, 100 mM caffeine, 25 mM L-canavanine, and 100 mM sucrose.

Calcium imaging data were analyzed using MATLAB code from previous studies (Devineni et al., 2019; Devineni et al., 2021). Images were registered within and across trials to correct for movement in the x-y plane using a sub-pixel registration algorithm. Regions of interest (ROIs) were drawn manually. Average pixel intensity within the ROI was calculated for each frame. The average signal for 20 frames preceding stimulus delivery was used as the baseline signal, and the ∆F/F (change in fluorescence divided by baseline) for each frame was then calculated.

Connectome analysis

Request a detailed protocol

mlSEZt neurons in the connectome were identified using NeuPrint Explorer (https://neuprint.janelia.org), an online interface for searching the connectome data. Neuron morphology images in Figure 9 were generated by NeuPrint Explorer. Connectivity analyses were performed using Python scripts to query the database. 3Ns (201 cells, including mlSEZt cells) were defined as neurons receiving at least three synapses from at least one mlSEZt neuron. Top 3Ns (32 cells) were defined as non-mlSEZt neurons receiving at least 20 total synapses from the mlSEZt population. 4Ns (1743 cells) were identified as neurons receiving at least three synapses from at least one top 3N, but we focused on strong 4Ns that receive at least 20 synapses from the top 3N population (322 cells) and also identified the top 30 4Ns that receive the strongest input (≥96 synapses). Graphs showing the distribution of connection strength (Figure 9F and I, top) represent data for all 3Ns (201 cells) or the strong 4Ns (322 cells). Graphs showing the brain regions targeted (Figure 9F and I, bottom) represent data for the top 3Ns or 4Ns (32 or 30 cells, respectively). Brain regions can be viewed on NeuPrint Explorer and are based on the nomenclature in Ito et al., 2014. The five 3Ns shown in Figure 9H represent examples of the top five cell types receiving the strongest total mlSEZt input (based on number of synapses), and they each receive input from at least four mlSEZt cells (see Table 1). The five 4Ns shown in Figure 9J represent examples of the top cell types that receive the strongest total input from top 3Ns (based on number of synapses) and also receive input from at least three 3Ns (see Table 2).

Statistical analyses

Request a detailed protocol

Statistical analyses were performed using GraphPad Prism, Version 9. Statistical tests and results are reported in the figure legends. Two groups were compared using paired or unpaired t-tests. More than two groups were compared using one-way ANOVA followed by Dunnett’s test comparing experimental flies to each control. All graphs represent mean ± SEM. Sample sizes are listed in the figure legends. No explicit power analyses were used to determine sample sizes prior to experimentation. Minimum sample sizes were decided prior to experimentation based on previous experience or pilot experiments revealing how many samples are typically sufficient to detect reasonable effect sizes. Additional samples were sometimes added if the initial results were inconclusive or more variable than expected, but never with the intent to make a non-significant p-value significant or vice versa. In general, the experimenter was not explicitly blinded to genotype during experimentation.

To generate the table in Figure 5A, the effects of activating bitter neuron subsets were compared to the effect of activating all bitter neurons. For each genotype and behavioral parameter, one-way ANOVAs were used to determine if there was an effect at any light intensity. If so, the strength of the effect was quantified by comparing the maximal effect at any intensity to the maximal effect observed with activating all bitter neurons. To do this, the difference between experimental flies and each control genotype was quantified, and the smaller of the two differences was used as a measure of the experimental effect. A strong or moderate effect was defined as an effect at least 80% or 40%, respectively, of the maximal value. Statistically significant effects that were less than 40% of the maximal value were defined as weak effects.

Data availability

Source data files for experiments and code for connectome analyses are publicly available at https://github.com/avdevineni/taste-activation, (copy archived at swh:1:rev:920e0cd3d5d03b02e30faa3129e0d2d6b7dad8d6).

References

  1. Book
    1. Dethier V
    (1976)
    The Hungry Fly
    Cambridge: Harvard University Press.
  2. Book
    1. Eyjolfsdottir E
    2. Branson S
    3. Burgos-Artizzu XP
    4. Hoopfer ED
    5. Schor J
    6. Anderson DJ
    7. Perona P
    (2014) Detecting social actions of fruit flies
    In: Fleet D, Pajdla T, Schiele B, Tuytelaars T, editors. Computer Vision – ECCV 2014. Cham: Springer International Publishing. pp. 772–787.
    https://doi.org/10.1007/978-3-319-10605-2

Decision letter

  1. Ilona C Grunwald Kadow
    Reviewing Editor; University of Bonn, Germany
  2. Claude Desplan
    Senior Editor; New York University, United States
  3. Alex C Keene
    Reviewer; Texas A&M University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for submitting the paper "Selective integration of diverse taste inputs within a single taste modality" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by a Senior Editor. The following individual involved in the review of your submission has agreed to reveal their identity: Alex C Keene (Reviewer #2).

Comments to the Authors:

We are sorry to say that, after consultation with the reviewers, we have decided that the work in its current form will not be considered further for publication by eLife.

While the reviewers agree that the topic of the study is of general interest, they also felt that the work does not reach the level of significance and novelty expected for an eLife article. Specifically, the reviewers find that the study is divided into two parts. The first part is overall convincing, but does not offer sufficient advances beyond the current understanding and literature in the field. The second part has potential and contains some very interesting data. Unfortunately, the reviewers felt that the data of this part are too preliminary at this point. The reviewers discussed several possibilities to help you improve your study given its potential. First, they recommend shortening and re-organizing the manuscript and center it on part two of the study. Second, they recommend strengthening the second part with additional experiments and controls. We hope that the detailed reviews below will be helpful.

Reviewer #1:

Deere et al. combine optogenetics, behavioral analysis and calcium imaging methods to carefully study the effects of bitter neuron activation on different body parts on behavior. While activation of bitter neurons in different organs drives mostly the same behavior, they find a few subtle differences in behavioral output upon neuronal activation between pharynx and labellum, for instance. Using trans-tango labeling and a split-Gal4 line the authors aim to explain their findings. The data are of high quality and contain interesting information for people in the field.

Line 155: The authors state that sugar neuron activation drives a stronger behavioral effect in starved as compared to fed flies. Given that the authors are comparing the two systems, it would be nice to add a possible explanation. For instance, this would suggest that hunger state modulates behavioral responses to sugar later in processing, but this is not/less the case for bitter.

Line 196: it would be interesting to hear and include the authors' interpretation of this result and what it says about the underpinning circuits.

Line 217ff: these are nice results and confirm previous observations, but it is unclear how far they support or fit into the narrative. I encourage the authors' to consider removing this part to make the paper a bit leaner and straight forward, and to present it only in the context of the next paragraph.

Line 241ff: the data indicates that the subset that diverges most is the labellar subset S-b. The authors talk about differences in expression levels and are careful to not compare the subsets directly. On the other hand, they plot the data in the same graphs so that one automatically compares them, but ok. I am a little worried about cell number. You are activating 13/19 vs. 3/19. As you state in the discussion, these number differences could explain the behavioral effects rather than being of different subtypes. To make it easier for the reader, it might be better to move the S-b data into the supplements and remove them from the comparison in the main figure to essentially compare legs vs. labellum vs. pharynx. If you carry out this comparison, then it's evident that the only difference is seen for learned odor preference (pharynx activation does not induce it). This makes perfect sense looking at the trans-tango data (no connections to higher brain centers, which I find very interesting).

Line 299ff: the trans-tango data are very nice, but have the authors considered integrating available EM data, e.g. from Neuprint, into their study? Such data could be used to trace the circuits further downstream. It would also help to test their main conclusion with other means, i.e. projection neurons integrate bitter taste from multiple organs. Is this true for other bitter/taste projection neurons?

Discussion:

It is surprising that the activation of the bitter projection neuron does not induce learning. What are the connections of these neurons in higher brain regions?

Your data are of very high quality and very convincing. I am, however, not quite convinced by the interpretation.

Looking at the data, I see that pharynx induces different behaviors (in particular learning) than the other subsets. Your trans-tango data provides a possible explanation. Your other data suggest that, indeed, leg and labellum drive very similar behavior (with perhaps the exception of 'post-light locomotion). This also fits to the trans-tango data overall.

Therefore, it is difficult for me to follow your conclusion, e.g. abstract: 'These results suggest that different bitter inputs are selectively integrated early in the circuit, enabling the pooling of sensory information, and the circuit then diverges into multiple pathways that may have different behavioral roles.'

Given the high quality of your data, I suggest condensing the manuscript and perhaps focussing on the differences between pharynx and the other neurons a little more. Moreover, using more inactivation experiments could be helpful, too, to distinguish the differences between these subsets in the different behaviors you study. Finally, the paper is divided into two parts, and I am not convinced that the sugar neuron data is needed for the second part and vice versa. It might make sense to split the manuscript into two.

Reviewer #2:

The authors tackle a particularly interesting question of how multiple taste cues from the same modality are integrated to form a behavioral response. The manuscript is scientifically sound, with well controlled experiments. The authors rely on optogenetic manipulations to define behavioral responses to different subsets of neurons. The field has tended to rely on relatively simple behavioral responses, such as PER, and a strength of this manuscript is the detailed characterization of locomotor behavior, learning, and reflexive feeding. This systematic characterization of bitter responsive neurons will be broadly useful to the field. In addition, the identification of second order neurons that selectively regulate PER provides a platform for studying integration. A number of concerns are noted that are readily addressable. First, there is a reliance on optogenetics that may not reflect the temporal or spatial patterns of natural compounds. The impact of the paper may be greater if these experiments were complemented with naturally occurring compounds. Second, as written much of the manuscript is descriptive, focusing on aspects that have previously been explored (such as the opposing valances of bitter and sugar neurons). Streamlining this part of the manuscript will make the significance more accessible. Overall, this manuscript provides a valuable resource for understanding how bitter tastants are coded, and will be of broad interest to groups researching sensory processing and behavioral regulation.

1. The paper focuses on the complexity of taste coding, however, it does not firmly establish why the optogenetic approach is needed. There are many limitations to this approach, including the fact that the subsets of neurons, and their temporal activation may not reflect those induced by food. Additional justification of the use of this system is important.

2. The conclusions of Figure 2 are unclear. One possibility is that the activation of bitter neurons induces locomotion, which results in less time on those quadrants. The other is that they are actively avoiding the quadrant. Without addressing this, it is not clear what this adds to the conclusions beyond the data presented in Figure 1.

3. The conclusion of the first four figures are factors that sweet and bitter neurons have oppositional valance, and that sugar is more potentially modulated by a huger state than bitter taste. These have been well described throughout the literature in multiple systems. At the authors' discretion, I suggest streamline these as it will make the novel aspects of the manuscript more accessible.

4. Are the GAL4 lines used to characterize distinct subsets of neurons exclusively expressed in these body regions or preferentially expressed in them? I presume the former, but it would be valuable to reference this and/or show expression if it has not been thoroughly described across all regions.

5. It would be helpful to directly compare subsets of sweet and bitter second order neurons in figure 7 to gain insight into whether the circuit design is shared between the tastants.

Recommendations for the authors:

1. In numerous cases the justification for the experimental approach is unclear. For example line 246 states 'We focused on bitter taste for these studies.' It would be helpful to know why the bitter taste was chosen over sweet.

2. The introduction is very well written and frames the question of how tastants are detected. However, the focus primarily on the sweet and bitter modalities, and discusses how these are sensed by Grs. The complexity described includes the more recent descriptions of other modalities and new classes of gustatory receptors. It seems important to describe these as well in order to convey how the complexity of the system that is being dissected.

3. Line 43 says 'like other sensory systems…' but the focus of the introduction is almost exclusively on taste. Much of this can be streamlined, and it would be helpful to add comparisons for context. For example, questions of sensory integration and combinatorial receptor activation have been addressed in much greater detail for olfaction and this may be worth mentioning.

4. Line 109: It's not clear what the first sentence means. Most tastes are likely to be feeding related. Is the point that most studies have used taste response to food components?

Reviewer #3:

This manuscript by Deere et al. characterizes the behavior effects of optogenetic activation of sugar and bitter sensing GRNs. While activation of sugar-sensitive GRNs causes a brief decrease in locomotion, activation of bitter-sensitive GRNs increases locomotor activity in flies. Furthermore, the impact of sugar-GRN activation, but not bitter-GRN activation, is enhanced by food deprivation. The authors, also showed, pairing neutral odors with optogenetic activation of sugar-GRNs or bitter-GRNs can induce positive or negative valence to neutral odors in preference experiments in subsequent trials. After characterizing behavioral effects of optogenetic manipulation of sugar and bitter GRNs, the authors next focus on activation of bitter-sensing GRNs in different organs and discuss the differences and similarities between their behavioral effects. The authors observe some overlap between the behavior effects elicited by bitter-sensing GRN activation in different organs, legs, pharynx, labellum, etc. therefore they hypothesized bitter GRNs might share the same downstream neural circuits. Using transsynaptic labeling, they identified a group of neurons that receive input from different bitter sensing neurons, the mlSEZt neurons, and they showed that mlSEZt neurons showed ON and OFF responses to bitter tastant s but not sugar using calcium imaging. Interestingly, the response dynamics of mlSEZt neurons show different responses to bitter compounds, and the ON response of mlSEZt neurons habituated faster than the OFF response in response to repeated stimulation.

Overall, this is a promising study, however, I have some major technical concerns about the way experiments have been conducted. Without having some clarification of GRN activity during optogenetic activation and anatomical characterization of the GRN drivers used in the study, it is hard to interpret what the results mean.

1. In all of the optogenetic activation experiments, the authors are using constant light stimulation. Previous papers (Inagaki 2014, Nature Methods) have shown that constant light stimulation causes suppression of action potentials in GRNs. Hence, it is not clear how the stimulus protocols used in the manuscript are impacting GRN activity. Authors need to show that either GRNs are actually activated and not shut down by the prolonged optogenetic stimulation protocols they used, via electrophysiology or calcium imaging. Alternatively, they can change their activation protocols by shortening the stimulus period or using pulses of light instead of constant light activation. Without having some evidence, the GRNs are actually active during the periods of optogenetic activation, it is hard to interpret the results of the experiments in Figure 1-5.

2. For sugar-GRN activations, it would be good to use another sugar-GRN, such as Gr5a to confirm Gr64f results.

3. In figure 5, the authors are activating subsets of bitter-GRNs and measuring the effects of optogenetic stimulations in various behaviors. I would like to see the expression patterns of GRN drivers used in these experiments in flies expressing Chrimson. Chrimson transgene used in these experiments has a Td-tomato tag, so authors can stain for this transgene and show us exactly which neurons they are stimulating. Without detailed anatomical characterization of the driver lines used in the experiments, it is again hard to interpret the results.

4. See comment #2. The same concern here as Figure 5. Please show the expression pattern of the source GAL4s that are used for transsynaptic labeling. Without knowing the GRN expression, we can't interpret the results of downstream neurons.

5. In figure 8, are the ON and OFF responses of mlSEZt neurons displayed by arbors of the same neurons, or different neurons? I.e., is there any internal difference between different mlSEZt neurons, in terms of their response dynamics to bitter compounds?

Recommendations for the authors:

1. In Figure 8 – —figure supplement 1C and D, it is better to include a diagram encircling the region of ROI in the brain. The arrows in Figure 8D are unclear.

2. For the experiments that involved optogenetic activation of both sugar sensing and bitter sensing neurons, how could you know that the behavior effect is not influenced by the possible different GAL4 expression levels of sugar sensing and bitter sensing GAL4 lines?

3. In Figure 1 – —figure supplement 1, there should be '(D)' in the last figure.

4. In Figure 8, adding experiments with Gr66a-lexA > CsChrimson, mlSEZt-Split (the cleanest line) > TNT might be helpful in figuring out whether mlSEZt neurons are necessary for the locomotor responses, innate and learned preference. Also, as a negative control, Gr5a-lexA > CsChrimson, mlSEZt-Split > TNT can be tested if reagents are available.

5. In Figure 8G, angular velocity significantly increases with high led intensity. Would it be possible to use higher led intensity to make sure that all mlSEZt populations are activated?

[Editors’ note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Selective integration of diverse taste inputs within a single taste modality" for further consideration by eLife. Your revised article has been evaluated by Claude Desplan (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

Please address all reviewer comments with a detailed response and/or changes to the manuscript text and figures. If you have the data available for points 6 and 7 of Reviewer 2, please include them. Otherwise, please change the text to address the fact that this driver is not optimal.

Reviewer #1 (Recommendations for the authors):

The authors have done a commendable job in revising and thereby strongly improving the manuscript. As previously stated, the data is of high quality, and the results are convincing. The new results including the connectomics analysis are very interesting and integrate the findings in early sensory processing with higher-order processing. Moreover, the new optoPAD data including neuronal inhibition is helpful to make the point of functional redundancy and, in combination with the later results, highlights a difference in redundancy at different levels of behavioral performance and learning.

The authors have addressed my previous concerns related to the first submission of this work in full. I am satisfied with their responses. In particular, I believe the paper in its current, more streamlined form is easier and more interesting to read after the removal of some of the data including the data on sugar sensing.

Reviewer #2 (Recommendations for the authors):

My previous comments have been addressed. Here are some issues that still remain.

Issues to be addressed in the revised version.

1. The images are too small in Figure 1C. I suggest making three columns: One for the labellum only, one for the legs only, and one for the pharynx only. For the labellum column, authors can zoom in to the labellum, and for the pharynx, they can only zoom in to the pharynx. It will make the results in 1D clearer. As it is, neurons are not clearly visible.

2. Figure 2D, I don't see the point of plotting opto-ctrl and getting negative values; I think it is easier to plot ctrl-opto and plot positive values rather than negative values. Nothing will change, except the plots will look better, in my opinion. Right now, it is confusing to see negative values. Same argument for Figure 2G. Also, some of the negative controls seem to have a high variance. Did the authors test for ANOVA and compare all controls and test animals with each other statistically? If yes, please indicate statistical significance in the plots.

3. I think Figure 6 is missing the VNC expression. Many leg neurons project to the VNC, showing only the brain Trans-tango signal is misleading.

4. I don't think there is any reason for 9 figures. Figures 5-6 can be combined or shown as supplementary figures.

5. I think it is too strong to say mlSEZt neurons are second-order bitter neurons. There is no experiment showing direct synaptic connectivity to bitter neurons. They can say bitter downstream neurons or show synaptic connectivity via EM or GRASP.

6. Authors use a 'dirty' GAL4 for functional imaging experiments rather than the 'clean' split line. I think it might be better to use the cleaner split-GAL4 for functional imaging since many neurons are labeled in SEZ and hard to know which neurons' projections are recorded during imaging.

7. Effects on behavior upon silencing or activating the clean split are much weaker. I think functional imaging is essential to demonstrate these are the real downstream bitter neurons.

8. Figure 9 is interesting, but it is not super relevant without really demonstrating whether mlSEZt neurons respond to bitter compounds. Also hard to demonstrate, the putative mLSEZt neurons they found in EM data are really mLSEZt neurons.

Reviewer #3 (Recommendations for the authors):

This revised manuscript has been significantly reorganized to focus on bitter taste. The streamlining of results and inclusion of clarifying data significantly strengthen this manuscript, and further our understanding of bitter taste processing. With these changes, the authors have addressed all of my previous concerns.

https://doi.org/10.7554/eLife.84856.sa1

Author response

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

While the reviewers agree that the topic of the study is of general interest, they also felt that the work does not reach the level of significance and novelty expected for an eLife article. Specifically, the reviewers find that the study is divided into two parts. The first part is overall convincing, but does not offer sufficient advances beyond the current understanding and literature in the field. The second part has potential and contains some very interesting data. Unfortunately, the reviewers felt that the data of this part are too preliminary at this point. The reviewers discussed several possibilities to help you improve your study given its potential. First, they recommend shortening and re-organizing the manuscript and center it on part two of the study. Second, they recommend strengthening the second part with additional experiments and controls. We hope that the detailed reviews below will be helpful.

We understand the reviewers’ perspective, and we have completely reworked the manuscript to address their comments and build a stronger story. This includes removing many experiments that the reviewers did not think were sufficiently novel or relevant, and adding many new results to strengthen and expand the story. We have made the following major revisions:

1) The paper now focuses entirely on the bitter pathway; all experiments with sugar neurons have been removed. Moreover, the paper focuses from the very beginning on the core question of how different bitter inputs are processed. Figures 1-4 of the original paper no longer exist in their previous form.

2) We have conducted new experiments using a high-resolution feeding assay (the optoPAD). Previously we only used proboscis extension as a simple readout of feeding initiation. The new experiments are important for testing more naturalistic feeding behavior in freely moving flies, as well as analyzing detailed parameters of feeding.

a) We activated bitter neuron subsets, and the results were consistent with our previous data using proboscis extension: all subsets strongly suppressed feeding, with certain feeding parameters affected more than others (Figure 2B-D).

b) We silenced bitter neuron subsets in the presence of natural bitter taste, which some reviewers asked for. In contrast to the activation experiments, silencing generally did not produce statistically significant effects compared to genetic controls (Figure 2E-G). Together with the activation data, these results support a model in which different bitter neuron subsets act in a parallel and largely redundant manner to regulate aversive behaviors.

c) We activated or silenced second-order mlSEZt neurons (Figure 8D-E). mlSEZt activation suppressed feeding, and we observed effects on certain feeding parameters (e.g. sip duration) but not others (e.g. number of sips). mlSEZt silencing did not have a significant effect. These findings suggest that mlSEZt cells suppress specific aspects of feeding behavior but act in parallel with other second-order bitter pathways, creating redundancy in the circuit.

3) We have presented control experiments to address concerns about the Gal4 lines and neuronal activation protocols.

a) Reviewers 2 and 3 asked about the specificity of Gal4 lines targeting bitter neuron subsets. We now emphasize that all of these Gal4 lines have been extensively characterized in previous studies, and we show our own images confirming their organ-specific expression (Figure 1C).

b) We have added new experiments to address concerns regarding the stimulation protocol, described further in our responses to specific reviewer comments (see below).

4) Given the reviewers’ interest in the second part of the paper, which focuses on downstream bitter pathways, we have expanded this part of the study. We reorganized the figures to bring some of the supplemental imaging experiments into the main figures, as they provide insight into how mlSEZt encodes taste, and we have added many new results:

a) We performed additional behavioral experiments with mlSEZt suggested by Reviewer 3. First, we tested the effect of higher intensity activation on locomotion, which revealed similar effects as the lower intensities. Second, we silenced mlSEZt while activating bitter-sensing neurons. We found that mlSEZt is not required for bitter neurons to affect locomotion, preference, or learning, as expected from the inability of mlSEZt activation to elicit these behaviors.

b) We used co-labeling experiments to show that mlSEZt cells are GABAergic, revealing that they transmit feedforward inhibition to the SLP. This has implications for the circuit analysis below.

c) We traced the neural circuits downstream from mlSEZt neurons using two methods, transTango and connectome analysis. These studies revealed that bitter information from mlSEZt cells is first processed in the SLP and SIP before being relayed to other areas, including the SMP, lateral horn, mushroom body, and descending neurons. There are strong interconnections between mlSEZt cells and feedback connections from third-order neurons to mlSEZt cells, suggesting extensive local processing. Several downstream neurons are poised to integrate bitter information with olfactory input, and at least one neuron integrates input from different second-order taste neurons. There are also connections to oviDN neurons that use sweet taste input to regulate egg-laying decisions, suggesting that they integrate sugar and bitter inputs. These studies are the first to uncover third- and fourth-order taste pathways in the higher brain.

Overall, we believe that these revisions have resulted in a much stronger manuscript that should reach the level of significance expected for eLife. It is striking how little is known about central taste processing in flies, and our study provides several new insights that will have broad interest to the field. This includes: (1) discovering that bitter neurons in different organs drive largely overlapping behaviors, which challenges the conventional model, (2) revealing principles of taste circuit connectivity at the first synapse, (3) characterizing the response properties and behavioral role of a new type of second-order bitter neuron, which remains one of the only second-order taste neurons to be studied, and (4) revealing circuits for taste processing in the higher brain.

Reviewer #1:

Deere et al. combine optogenetics, behavioral analysis and calcium imaging methods to carefully study the effects of bitter neuron activation on different body parts on behavior. While activation of bitter neurons in different organs drives mostly the same behavior, they find a few subtle differences in behavioral output upon neuronal activation between pharynx and labellum, for instance. Using trans-tango labeling and a split-Gal4 line the authors aim to explain their findings. The data are of high quality and contain interesting information for people in the field.

Line 155: The authors state that sugar neuron activation drives a stronger behavioral effect in starved as compared to fed flies. Given that the authors are comparing the two systems, it would be nice to add a possible explanation. For instance, this would suggest that hunger state modulates behavioral responses to sugar later in processing, but this is not/less the case for bitter.

We have removed all results related to sugar neuron activation and fed vs. starved comparisons, so this comment no longer applies.

Line 196: it would be interesting to hear and include the authors' interpretation of this result and what it says about the underpinning circuits.

This comment refers to results with sugar neuron activation, which have been removed from the paper.

Line 217ff: these are nice results and confirm previous observations, but it is unclear how far they support or fit into the narrative. I encourage the authors' to consider removing this part to make the paper a bit leaner and straight forward, and to present it only in the context of the next paragraph.

We have removed these results from the paper.

Line 241ff: the data indicates that the subset that diverges most is the labellar subset S-b. The authors talk about differences in expression levels and are careful to not compare the subsets directly. On the other hand, they plot the data in the same graphs so that one automatically compares them, but ok. I am a little worried about cell number. You are activating 13/19 vs. 3/19. As you state in the discussion, these number differences could explain the behavioral effects rather than being of different subtypes. To make it easier for the reader, it might be better to move the S-b data into the supplements and remove them from the comparison in the main figure to essentially compare legs vs. labellum vs. pharynx. If you carry out this comparison, then it's evident that the only difference is seen for learned odor preference (pharynx activation does not induce it). This makes perfect sense looking at the trans-tango data (no connections to higher brain centers, which I find very interesting).

We agree that the data would be cleaner if we ignore S-b activation, but we think it would be misleading to relegate that data to the supplemental figures without a strong rationale. We do not have strong evidence that the weak S-b effects are solely due to the low cell number; for example, pharyngeal activation using Gr9a-Gal4 only labels one cell and yet had a very strong effect on many behaviors.

To draw conclusions about the functions of different organs (e.g. in the text and Figure 5B model), we are considering the S-a + I-a labellar data rather than the S-b data, just as the reviewer suggests. We think this is justified because positive results are more informative than negative results.

We agree that the difference in behavior and connectivity for the pharynx versus the other organs is interesting. We have added a sentence to make the point that the reviewer mentions (lines 636-638): “For example, bitter neurons in the leg and labellum elicited learning while pharyngeal neurons did not, and this difference may relate to the observation that only leg and labellar neurons connected strongly to the lateral and mediolateral tracts.”

Line 299ff: the trans-tango data are very nice, but have the authors considered integrating available EM data, e.g. from Neuprint, into their study? Such data could be used to trace the circuits further downstream. It would also help to test their main conclusion with other means, i.e. projection neurons integrate bitter taste from multiple organs. Is this true for other bitter/taste projection neurons?

The NeuPrint (hemibrain) connectome, which is the only publicly available annotated connectome, does not include the SEZ, where bitter-sensing axons arborize. Thus, we cannot use the connectome to analyze connectivity to second-order neurons. We added a sentence mentioning this (lines 285-287).

However, we have now used the connectome to trace downstream circuits from mlSEZt neurons, as described above and shown in Figure 9. These studies reveal that input from mlSEZt cells is first processed in the SLP and SIP before being relayed to other areas, including the SMP, lateral horn, mushroom body, and descending neurons. We identified hierarchical interconnections between mlSEZt cells, prominent feedback circuits, downstream neurons that are poised to integrate olfactory and taste input, and neurons controlling egg-laying that may integrate sweet and bitter input.

Discussion:

It is surprising that the activation of the bitter projection neuron does not induce learning. What are the connections of these neurons in higher brain regions?

As described above, we have now conducted extensive studies to identify downstream circuits from mlSEZt neurons. Postsynaptic partners of these neurons are primarily confined to the SLP and SIP. Moving one synapse downstream, we do observe connections to the mushroom body (the learning and memory center), but not to PPL1 dopamine neurons that convey bitter reinforcement signals during learning. The absence of these connections may relate to the inability of mlSEZt to elicit learned aversion, and we have pointed this out in the text (lines 533-536).

Your data are of very high quality and very convincing. I am, however, not quite convinced by the interpretation.

Looking at the data, I see that pharynx induces different behaviors (in particular learning) than the other subsets. Your trans-tango data provides a possible explanation. Your other data suggest that, indeed, leg and labellum drive very similar behavior (with perhaps the exception of 'post-light locomotion). This also fits to the trans-tango data overall.

Therefore, it is difficult for me to follow your conclusion, e.g. abstract: 'These results suggest that different bitter inputs are selectively integrated early in the circuit, enabling the pooling of sensory information, and the circuit then diverges into multiple pathways that may have different behavioral roles.'

We apologize if our conclusions have not been presented clearly. In our view, the similarities between the three organs (they all affect proboscis extension, feeding, locomotion, and positional aversion) are more striking than the differences. However, we agree that the differences are important too, and we repeatedly mention them. Perhaps the level of similarity versus difference is a matter of interpretation (a glass half-full versus half-empty scenario), and by carefully laying out the results (e.g. the table in Figure 5A) the reader can form their own interpretation.

We believe that our conclusions are fully compatible with the reviewer’s interpretation. By saying that “different bitter inputs are selectively integrated early in the circuit”, we mean that these inputs are integrated within some pathways (e.g. all three organs project to the medial tract), but not every pathway must receive input from all organs. We refer to this as “selective integration”, and we have created a new diagram (Figure 5B) to visually depict this model. We have also added more explanations in the text to describe it. For example (lines 270-276):

“These results generally support an integrative model of taste processing (Figure 1B, left) in which bitter inputs from different organs converge onto common downstream pathways to drive a diverse set of aversive behaviors. However, the fact that not all neuronal subsets elicited all behaviors, or preferentially elicited some behaviors over others, suggests that some downstream pathways may receive selective or preferential input from specific bitter-sensing cells. We refer to this model as “selective integration” (Figure 5B), and it lies in between the two models shown in Figure 1B.”

The example that the reviewer mentions regarding the pharynx is the most clear case where a behavioral difference may be attributable to a difference in connectivity, and we have added a sentence to explicitly describe this point (lines 636-638).

Given the high quality of your data, I suggest condensing the manuscript and perhaps focussing on the differences between pharynx and the other neurons a little more. Moreover, using more inactivation experiments could be helpful, too, to distinguish the differences between these subsets in the different behaviors you study. Finally, the paper is divided into two parts, and I am not convinced that the sugar neuron data is needed for the second part and vice versa. It might make sense to split the manuscript into two.

In accordance with the reviewer’s suggestions, we have removed all of the sugar neuron data. We have also removed much of the basic characterization of bitter neuron activation (e.g. comparing starved versus fed) in order to focus on the role of different bitter neuron subsets.

In addition, we have added inactivation experiments where we silenced each subset of bitter neurons in a feeding assay (Figure 2E-G). We did not expect that this would have a strong effect because each subset alone is capable of suppressing feeding, suggesting that they act redundantly. Our results support this model: silencing bitter neuron subsets had relatively small and generally nonsignificant effects. We note that pharyngeal silencing showed the strongest trend toward increased feeding (lines 183-184), suggesting that there may be a stronger requirement for the pharynx than other organs.

We have not performed inactivation experiments with the other behaviors because this would require setting up new behavioral paradigms; our current setups only allow light stimulation. We would need to test natural bitter compounds and verify what kind of locomotor changes, preference, and learning they normally drive. We feel that setting up new assays is beyond the scope of this study. More importantly, our Gal4 lines do not allow us to silence 100% of bitter neurons in any given organ, and we do not anticipate observing strong effects by only partially silencing each organ (which is supported by our results in the feeding assay). This is the primary reason why we took the activation approach.

Reviewer #2:

The authors tackle a particularly interesting question of how multiple taste cues from the same modality are integrated to form a behavioral response. The manuscript is scientifically sound, with well controlled experiments. The authors rely on optogenetic manipulations to define behavioral responses to different subsets of neurons. The field has tended to rely on relatively simple behavioral responses, such as PER, and a strength of this manuscript is the detailed characterization of locomotor behavior, learning, and reflexive feeding. This systematic characterization of bitter responsive neurons will be broadly useful to the field. In addition, the identification of second order neurons that selectively regulate PER provides a platform for studying integration. A number of concerns are noted that are readily addressable. First, there is a reliance on optogenetics that may not reflect the temporal or spatial patterns of natural compounds. The impact of the paper may be greater if these experiments were complemented with naturally occurring compounds. Second, as written much of the manuscript is descriptive, focusing on aspects that have previously been explored (such as the opposing valances of bitter and sugar neurons). Streamlining this part of the manuscript will make the significance more accessible. Overall, this manuscript provides a valuable resource for understanding how bitter tastants are coded, and will be of broad interest to groups researching sensory processing and behavioral regulation.

We thank the reviewer for these thoughtful comments. We have completely reworked the manuscript to streamline the sections that the reviewer felt are too descriptive and not sufficiently novel. First, we removed all data regarding the sugar neurons, as these results are not directly relevant to the rest of the paper. Second, we now present the activation of all bitter-sensing neurons as merely a control to analyze the effects of bitter neuron subsets, rather than presenting these results as novel findings. These results are no longer described in separate figures.

Regarding the optogenetic approach and testing natural compounds, please see our response to point #1 below.

1. The paper focuses on the complexity of taste coding, however, it does not firmly establish why the optogenetic approach is needed. There are many limitations to this approach, including the fact that the subsets of neurons, and their temporal activation may not reflect those induced by food. Additional justification of the use of this system is important.

We have now added some additional justification to the text (lines 121-128). We have several reasons for taking the activation approach, as opposed to performing neuronal silencing while flies taste natural bitter compounds:

1) Our goal was to test which behaviors can be driven by each subset of bitter neurons, rather than which subsets are required for behavior. The former question (addressed by neuronal activation) should reveal which neurons are connected to which behavioral circuits, even if different neurons have redundant functions. The latter question (addressed by neuronal silencing) will only reveal behavioral effects if neurons do not have redundant functions.

2) Related to the point above: Our Gal4 lines do not label 100% of the bitter neurons in any given organ. We would not expect to observe strong effects using neuronal silencing if there are still some functional bitter neurons in each organ, since redundant functions are very likely. Indeed, our new feeding experiments support this: we see very strong effects with neuronal activation and weak, nonsignificant effects with silencing.

3) The activation approach allows us to target neurons within any organ and precisely control the timing and duration of the stimulus. Conversely, presenting freely moving flies with natural bitter taste means that we cannot control which organs are stimulated, nor the stimulus timing/ duration. For example, if flies do not ingest the substrate then pharyngeal neurons will never be stimulated, and we cannot assess whether pharyngeal neurons are capable of driving behavior. How each organ samples bitter under natural conditions and which organs are required for each behavior are interesting questions, but ones that we plan to address in future studies.

We completely agree that optogenetic activation may not recapitulate natural activation, but this is a limitation of all optogenetic activation studies. For this reason, we first activated the entire population of bitter-sensing neurons and ensured that the behavioral effects are consistent with what we expect from previous studies of bitter taste. The concordance between natural and optogenetically evoked behaviors suggests that we are not activating bitter neurons in a totally unnatural way. We have also added new data using a different activation protocol: we compared the effect of continuous and 50 Hz pulsed light stimulation, and the behavioral effects were nearly identical (Figure 3 —figure supplement 1C-D and Figure 4 —figure supplement 1B). These results suggest that the behavioral effects we observe are not specific to a particular stimulation protocol.

Finally, we have complemented our activation approach with inactivation experiments for one type of behavior, feeding. The weak effects of neuronal silencing, as compared with the very strong effects of neuronal activation, support the idea that bitter neurons function redundantly.

2. The conclusions of Figure 2 are unclear. One possibility is that the activation of bitter neurons induces locomotion, which results in less time on those quadrants. The other is that they are actively avoiding the quadrant. Without addressing this, it is not clear what this adds to the conclusions beyond the data presented in Figure 1.

Figures 1 and 2 no longer exist in their previous form, so we are not sure if the reviewer would still have the same concern. In the revised manuscript, the effects of activating all bitter neurons are no longer presented as novel results; they are simply a control to test activation of bitter subsets.

We completely agree that positional aversion and locomotor stimulation could be linked, as we discussed in the original text. However, they are still very different behaviors displayed over different timescales, and it is possible that an active choice process contributes to positional aversion. It is not obvious that each bitter neuron subset should have identical effects in both assays, so we think it is reasonable to test their effects separately even if the results do end up being correlated. (Other behavioral measures we test are also intrinsically correlated, such as proboscis extension and feeding.) We have edited the text to include this rationale (lines 230-234).

3. The conclusion of the first four figures are factors that sweet and bitter neurons have oppositional valance, and that sugar is more potentially modulated by a huger state than bitter taste. These have been well described throughout the literature in multiple systems. At the authors' discretion, I suggest streamline these as it will make the novel aspects of the manuscript more accessible.

We have removed this entire section of the paper and incorporated the bitter neuron results into other figures. All sugar neuron results have been removed as well as comparisons between hunger states. The paper now begins with the activation of bitter neuron subsets.

4. Are the GAL4 lines used to characterize distinct subsets of neurons exclusively expressed in these body regions or preferentially expressed in them? I presume the former, but it would be valuable to reference this and/or show expression if it has not been thoroughly described across all regions.

The Gal4 lines have been thoroughly characterized in previous studies, and they are exclusively expressed in either the leg, labellum, or pharynx. We cited the references in the original text but we have now edited the text to be more explicit about the specificity of these lines (lines 135-139). We have also added our own images confirming the organ specificity of expression (Figure 1C).

5. It would be helpful to directly compare subsets of sweet and bitter second order neurons in figure 7 to gain insight into whether the circuit design is shared between the tastants.

We agree this would be interesting, but this has already been done by Snell et al. (Curr Biol, 2022).

Recommendations for the authors:

1. In numerous cases the justification for the experimental approach is unclear. For example line 246 states 'We focused on bitter taste for these studies.' It would be helpful to know why the bitter taste was chosen over sweet.

We apologize for a lack of clarity, and we have tried to ensure that the justification for each experiment is clearly described. In the revised manuscript, we have rewritten the Introduction to justify why we are studying bitter taste. We first describe the known functional diversity of sweet and bitter neurons (within and across organs). We then describe the overall question (how different inputs within a modality are processed) and previous work on the sugar pathway, which supports the segregated processing model. Finally, we discuss how very few studies have addressed this question in the bitter pathway, creating a gap in knowledge that we aim to fill.

2. The introduction is very well written and frames the question of how tastants are detected. However, the focus primarily on the sweet and bitter modalities, and discusses how these are sensed by Grs. The complexity described includes the more recent descriptions of other modalities and new classes of gustatory receptors. It seems important to describe these as well in order to convey how the complexity of the system that is being dissected.

We have added a sentence mentioning other taste modalities (lines 56-59), but we did not want to add too much detail since our focus now is exclusively on the bitter taste pathway, and the introduction is already quite long.

3. Line 43 says 'like other sensory systems…' but the focus of the introduction is almost exclusively on taste. Much of this can be streamlined, and it would be helpful to add comparisons for context. For example, questions of sensory integration and combinatorial receptor activation have been addressed in much greater detail for olfaction and this may be worth mentioning.

The phrase “Like other sensory systems” was meant to be a transition from the broad first paragraph (discussing sensory systems in general) to focusing on taste specifically. We considered adding more examples from other senses to the Introduction, but the Introduction is already fairly long, and these examples seemed to distract from conveying the most relevant information. However, we have added a comparison to olfactory coding and other sensory modalities in the Discussion (lines 756-769).

4. Line 109: It's not clear what the first sentence means. Most tastes are likely to be feeding related. Is the point that most studies have used taste response to food components?

This sentence has been removed.

Reviewer #3:

This manuscript by Deere et al. characterizes the behavior effects of optogenetic activation of sugar and bitter sensing GRNs. While activation of sugar-sensitive GRNs causes a brief decrease in locomotion, activation of bitter-sensitive GRNs increases locomotor activity in flies. Furthermore, the impact of sugar-GRN activation, but not bitter-GRN activation, is enhanced by food deprivation. The authors, also showed, pairing neutral odors with optogenetic activation of sugar-GRNs or bitter-GRNs can induce positive or negative valence to neutral odors in preference experiments in subsequent trials. After characterizing behavioral effects of optogenetic manipulation of sugar and bitter GRNs, the authors next focus on activation of bitter-sensing GRNs in different organs and discuss the differences and similarities between their behavioral effects. The authors observe some overlap between the behavior effects elicited by bitter-sensing GRN activation in different organs, legs, pharynx, labellum, etc. therefore they hypothesized bitter GRNs might share the same downstream neural circuits. Using transsynaptic labeling, they identified a group of neurons that receive input from different bitter sensing neurons, the mlSEZt neurons, and they showed that mlSEZt neurons showed ON and OFF responses to bitter tastant s but not sugar using calcium imaging. Interestingly, the response dynamics of mlSEZt neurons show different responses to bitter compounds, and the ON response of mlSEZt neurons habituated faster than the OFF response in response to repeated stimulation.

Overall, this is a promising study, however, I have some major technical concerns about the way experiments have been conducted. Without having some clarification of GRN activity during optogenetic activation and anatomical characterization of the GRN drivers used in the study, it is hard to interpret what the results mean.

We thank the reviewer for their feedback. We believe that our responses below, in combination with new experiments and data, should fully address the reviewer’s concerns.

1. In all of the optogenetic activation experiments, the authors are using constant light stimulation. Previous papers (Inagaki 2014, Nature Methods) have shown that constant light stimulation causes suppression of action potentials in GRNs. Hence, it is not clear how the stimulus protocols used in the manuscript are impacting GRN activity. Authors need to show that either GRNs are actually activated and not shut down by the prolonged optogenetic stimulation protocols they used, via electrophysiology or calcium imaging. Alternatively, they can change their activation protocols by shortening the stimulus period or using pulses of light instead of constant light activation. Without having some evidence, the GRNs are actually active during the periods of optogenetic activation, it is hard to interpret the results of the experiments in Figure 1-5.

We agree that this is an important issue to consider. We first want to clarify that prolonged light stimulation is only a concern for the locomotion (5 sec stimulation) and innate aversion (30 sec stimulation) experiments. The learning experiments used 1 Hz light pulses, and the PER and feeding experiments used short light stimulations (1-1.5 sec). In addition, for locomotion assays we are primarily assessing changes at light onset (first 1 sec).

At the reviewer’s suggestion, we have now conducted experiments using 50 Hz pulsed light stimulation to activate all bitter-sensing neurons in the locomotor and choice assays (Figure 3 —figure supplement 1C-D and Figure 4 —figure supplement 1B). The behavioral effects of continuous versus pulsed activation were nearly identical, implying that continuous light activation does effectively activate the neurons. The same was true for sugar neurons, although these results are not included now that we are only focusing on bitter taste: either continuous or pulsed stimulation elicited the same behaviors.

In addition, it would be strange that we observed behavioral effects throughout the light stimulation period if the neurons were not actually activated. For 30 sec choice experiments, in no case does the level of aversion decrease during the stimulus period (Figure 4B-C), as would be expected if the neurons are shut down, and in most cases the aversion continuously increases.

We note that for the choice and learning experiments, we are emulating the exact protocols published by Aso and colleagues, whose results are highly regarded in the field (Aso et al., eLife 2014; Aso and Rubin, eLife 2016). Continuous 30 sec stimulation during choice assays also appears to be the protocol used to validate the effect of Chrimson in the original Chrimson paper (Klapoetke et al., 2014, Supp. Figure 16), as they do not mention anything about light pulses. In these papers, continuous 30 sec stimulation produced behavioral effects for a variety of neurons, including olfactory sensory neurons, olfactory projection neurons, bitter-sensing neurons, and mushroom body output neurons.

Finally, we note that the previous study mentioned by the reviewer (Inagaki et al., 2014) used a different optogenetic effector, ReachR. ReachR and Chrimson are derived from distinct opsins and are known to have different properties.

2. For sugar-GRN activations, it would be good to use another sugar-GRN, such as Gr5a to confirm Gr64f results.

All sugar neuron data have been removed from the paper, so this comment no longer applies.

3. In figure 5, the authors are activating subsets of bitter-GRNs and measuring the effects of optogenetic stimulations in various behaviors. I would like to see the expression patterns of GRN drivers used in these experiments in flies expressing Chrimson. Chrimson transgene used in these experiments has a Td-tomato tag, so authors can stain for this transgene and show us exactly which neurons they are stimulating. Without detailed anatomical characterization of the driver lines used in the experiments, it is again hard to interpret the results.

We completely agree that detailed anatomical characterization of the Gal4 lines is essential, and this characterization was performed by the previous studies we cited (Weiss et al., 2011; Ling et al., 2014; Chen and Dahanukar, 2017). They did the painstaking work of characterizing the presence or absence of expression in every labellar or tarsal sensilla and every cell type within the pharynx. We have also added our own images verifying the expression patterns of these lines, confirming that each line is expressed within only one of the three organs (Figure 1C).

4. See comment #2. The same concern here as Figure 5. Please show the expression pattern of the source GAL4s that are used for transsynaptic labeling. Without knowing the GRN expression, we can't interpret the results of downstream neurons.

We have added these images (Figure 6 —figure supplement 1). We observed the expected Gal4 expression patterns, with neurons in different organs showing different projection patterns that have been previously characterized (Kwon et al., 2014).

For further transparency, we have also added additional examples of trans-Tango staining for each line; this shows the consistency (and slight variability) between labeling in different brains.

5. In figure 8, are the ON and OFF responses of mlSEZt neurons displayed by arbors of the same neurons, or different neurons? I.e., is there any internal difference between different mlSEZt neurons, in terms of their response dynamics to bitter compounds?

We wish that we could compare the responses of different mlSEZt neurons, but this was not feasible because (1) the projections of different neurons are intermingled in the SEZ and SLP, and (2) the cell bodies did not show bitter responses. We think it is likely that the same cells generate both ON and OFF responses because this is the situation for bitter sensory neurons as well as downstream PPL1 dopaminergic neurons (Devineni et al., 2021).

Recommendations for the authors:

1. In Figure 8 – —figure supplement 1C and D, it is better to include a diagram encircling the region of ROI in the brain. The arrows in Figure 8D are unclear.

We have added an image showing the ROIs (Figure 7 —figure supplement 1A).

2. For the experiments that involved optogenetic activation of both sugar sensing and bitter sensing neurons, how could you know that the behavior effect is not influenced by the possible different GAL4 expression levels of sugar sensing and bitter sensing GAL4 lines?

These experiments have been removed from the paper, so this comment no longer applies.

3. In Figure 1 – —figure supplement 1, there should be '(D)' in the last figure.

This figure has been reorganized, so this comment no longer applies.

4. In Figure 8, adding experiments with Gr66a-lexA > CsChrimson, mlSEZt-Split (the cleanest line) > TNT might be helpful in figuring out whether mlSEZt neurons are necessary for the locomotor responses, innate and learned preference. Also, as a negative control, Gr5a-lexA > CsChrimson, mlSEZt-Split > TNT can be tested if reagents are available.

As suggested, we have performed new experiments testing whether mlSEZt neurons are required for the effects of bitter neuron activation on locomotion, innate choice, and learning (Figure 8 —figure supplement 3). We did not expect that mlSEZt neurons would be required because their activation generally does not impact these behaviors. Indeed, none of the locomotor or preference behaviors were affected by mlSEZt silencing.

5. In Figure 8G, angular velocity significantly increases with high led intensity. Would it be possible to use higher led intensity to make sure that all mlSEZt populations are activated?

We have now performed these experiments (Figure 8 —figure supplement 2D). Using an even higher light intensity (44 µW/mm2) elicited an increase in angular velocity, similar to the previously tested “high intensity” (35 µW/mm2) stimulation. This result strengthens the evidence that mlSEZt neurons do exert a real effect on turning, and we have modified the text accordingly so that we are not downplaying this result. The higher intensity still did not affect forward velocity.

[Editors’ note: what follows is the authors’ response to the second round of review.]

Please address all reviewer comments with a detailed response and/or changes to the manuscript text and figures. If you have the data available for points 6 and 7 of Reviewer 2, please include them. Otherwise, please change the text to address the fact that this driver is not optimal.

We have addressed all of the reviewers’ comments; please see our detailed responses below. Our revisions include changes to the text and figures.

Points 6 and 7 from Reviewer 2 request that we perform functional imaging of mlSEZt cells using the cleaner split-Gal4 line, which we had already performed and is shown in Figure 7 —figure supplement 2C.

Reviewer #1 (Recommendations for the authors):

The authors have done a commendable job in revising and thereby strongly improving the manuscript. As previously stated, the data is of high quality, and the results are convincing. The new results including the connectomics analysis are very interesting and integrate the findings in early sensory processing with higher-order processing. Moreover, the new optoPAD data including neuronal inhibition is helpful to make the point of functional redundancy and, in combination with the later results, highlights a difference in redundancy at different levels of behavioral performance and learning.

The authors have addressed my previous concerns related to the first submission of this work in full. I am satisfied with their responses. In particular, I believe the paper in its current, more streamlined form is easier and more interesting to read after the removal of some of the data including the data on sugar sensing.

Thank you for these comments and for helping us improve the paper!

Reviewer #2 (Recommendations for the authors):

My previous comments have been addressed. Here are some issues that still remain.

Issues to be addressed in the revised version.

1. The images are too small in Figure 1C. I suggest making three columns: One for the labellum only, one for the legs only, and one for the pharynx only. For the labellum column, authors can zoom in to the labellum, and for the pharynx, they can only zoom in to the pharynx. It will make the results in 1D clearer. As it is, neurons are not clearly visible.

We have modified the figure as suggested. We now show the labellum and pharynx separately, and images of all three organs have been cropped and enlarged to make the neurons more visible.

As an aside, we want to clarify that in Figure 1D, the classes and numbers of neurons labeled in each organ are based on previous studies that thoroughly characterized these Gal4 lines (Weiss et al., 2011; Ling et al., 2014; Chen and Dahanukar, 2017). This is stated in the figure legend. Our images confirm that each line is expressed in the correct organ-specific pattern and the cell numbers are similar, but we have not rigorously quantified the number and subtypes of labeled neurons.

2. Figure 2D, I don't see the point of plotting opto-ctrl and getting negative values; I think it is easier to plot ctrl-opto and plot positive values rather than negative values. Nothing will change, except the plots will look better, in my opinion. Right now, it is confusing to see negative values. Same argument for Figure 2G.

To us, the most important thing is to show the direction of the effect and to keep the convention consistent across figures. The effects of activating or silencing all bitter neurons go in opposite directions, so one of these two experiments is going to have a negative value regardless of whether we choose to plot opto-ctrl or ctrl-opto. It is true that more of the Gal4 lines show an effect with activation rather than silencing, hence more negative bars than positive bars, but either way the reader is going to need to interpret negative bars somewhere in the figure.

In choosing between opto-ctrl or ctrl-opto, we find opto-ctrl to be more intuitive since we consider the control to be the reference point, which should be subtracted. We understand that others may have a different opinion, but we believe the reader will be able to interpret the data either way.

Also, some of the negative controls seem to have a high variance.

Yes, we have learned that individual flies are highly variable in this assay! Even without optogenetic stimulation, flies choosing between two identical food sources often have a strong preference for one food or the other. Any optogenetic manipulations must override this innate variability in order to produce an effect. For each experiment we tested a minimum of 40 flies per genotype, which is a similar or higher number than other papers that used this assay.

Did the authors test for ANOVA and compare all controls and test animals with each other statistically? If yes, please indicate statistical significance in the plots.

Yes, for every feeding parameter the experimental flies were compared to both controls by one-way ANOVA. All significant effects are denoted with asterisks. We decided not to add “n.s.” to all nonsignificant bars to avoid cluttering up the figure, as was stated in the legend: “Experimental effects not labeled with an asterisk are not significant”. This is the same convention we used in the other behavior figures.

We have now edited the legend to be more explicit: “For each feeding measure, experimental flies were compared to both controls, and effects not labeled with an asterisk are not significant.” If the reviewer feels that “n.s.” needs to be added to each nonsignificant bar then we can add this, but we thought the lack of significance is implied by the lack of an asterisk (and stated in the legend).

3. I think Figure 6 is missing the VNC expression. Many leg neurons project to the VNC, showing only the brain Trans-tango signal is misleading.

Although certain types of tarsal gustatory neurons are known to arborize within the VNC (e.g. sugar-sensing and pheromone-sensing cells), previous studies suggest that tarsal bitter-sensing neurons pass through the VNC without making synaptic arborizations there (Kwon et al., 2014). For this reason, we chose to focus on second-order pathways in the brain and did not stain the VNC for trans-Tango expression. Nevertheless, we agree that this is a limitation that we should acknowledge explicitly, and we have added two sentences to that effect (lines 291-294).

4. I don't think there is any reason for 9 figures. Figures 5-6 can be combined or shown as supplementary figures.

We see the reviewer’s point, and initially we also were reluctant to include 9 figures. However, we believe that the current organization works best for conveying the story by ensuring that each figure contains a conceptually linked set of data. We agree that Figures 5 and 6 are small enough to be combined, but they are conceptually quite different and it might be confusing to combine them. Keeping them separate should not present an extra burden to the reader since we are presenting the same number of panels either way.

We know that the majority of readers do not look at the supplemental figures, so we are reluctant to put anything critical in the supplement. Even if it is just a summary/model like Figure 5, it is important to understanding the story.

5. I think it is too strong to say mlSEZt neurons are second-order bitter neurons. There is no experiment showing direct synaptic connectivity to bitter neurons. They can say bitter downstream neurons or show synaptic connectivity via EM or GRASP.

We have modified the text to address this concern. In the Results section where we first characterize mlSEZt, we now state (lines 383-388):

“Together, these results strongly suggest that the mlSEZt neurons labeled by R29F12-Gal4 are second-order bitter neurons: they have an identical morphology to second-order bitter neurons labeled by trans-Tango, their projections overlap with the axon terminals of bitter-sensing neurons, and they show clear bitter responses. We will therefore refer to mlSEZt neurons as second-order bitter neurons, although it is formally possible that they are downstream bitter neurons that are not directly connected to bitter-sensing cells.”

In the Abstract and earlier parts of the text that readers will encounter before getting to this section, we have modified the wording to refer to mlSEZt as a “downstream” or “putative second-order” bitter neuron. We hope that these modifications, taken together, ensure that the language reflects the strength of the evidence.

6. Authors use a 'dirty' GAL4 for functional imaging experiments rather than the 'clean' split line. I think it might be better to use the cleaner split-GAL4 for functional imaging since many neurons are labeled in SEZ and hard to know which neurons' projections are recorded during imaging.

Although we initially performed functional imaging using the less specific R29F12 line, we also performed imaging using the cleaner split-Gal4 line. This experiment is shown in Figure 7 —figure supplement 2C and mentioned in the text (lines 408-410). The neurons labeled in the clean line respond to bitter taste and show the same dynamics as the less specific line.

In addition, we are confident that we are imaging the correct neurons even in the “dirty” R29F12 line. Even though the expression pattern may look dirty when compressed into a maximum intensity projection, when examining expression in all three dimensions it is clear that the mlSEZt projections in the relevant SEZ or SLP regions do not overlap with other neurons. Our functional imaging methods allow for high spatial resolution in all three dimensions (using two-photon imaging with a 60X objective). We are able to unequivocally identify mlSEZt cells based on their morphology and trace their projections into the SEZ or SLP. To convey this to the reader, we have added back a sentence that was removed in the previous version (lines 338-339): “The projections of mlSEZt neurons in the SEZ and SLP could be clearly identified based on their location and morphology.”

7. Effects on behavior upon silencing or activating the clean split are much weaker. I think functional imaging is essential to demonstrate these are the real downstream bitter neurons.

As mentioned above, we have imaged the mlSEZt neurons in the clean split-Gal4 line and shown that they respond to bitter taste (Figure 7 —figure supplement 2C). We agree that it is critical to know that these neurons are bitter-responsive.

8. Figure 9 is interesting, but it is not super relevant without really demonstrating whether mlSEZt neurons respond to bitter compounds. Also hard to demonstrate, the putative mLSEZt neurons they found in EM data are really mLSEZt neurons.

As mentioned above, we have shown that the mlSEZt neurons in the clean split-Gal4 line do respond to bitter taste (Figure 7 —figure supplement 2C).

Regarding the connectome data, we used both automated and manual searches to identify every neuron with even a vaguely similar morphology to mlSEZt cells, and after manually sorting through hundreds of cells, we are fairly confident that we have identified all neurons that have the correct morphology. Thus, the population of 21 neurons we identified should include the bitter-responsive mlSEZt cells. Moreover, the third-order neurons that we identified using the connectome (Figure 9H) closely resemble the postsynaptic neurons to mlSEZt labeled by trans-Tango (Figure 9B-C), suggesting that we have identified the correct mlSEZt neurons.

However, it is possible that not all of the 21 cells we identified represent bitter-responsive mlSEZt cells, as we mention in the text (lines 501-502). This is a caveat that needs to be acknowledged, but we don’t think it undermines our overall results given that the population of third-order neurons closely matches the trans-Tango staining, and individual downstream neurons would need to be functionally validated in future studies anyway. In addition, our finding that the 21 putative mlSEZt cells are highly interconnected suggests that most of the downstream neurons we identified are likely to be at least indirectly influenced by true mlSEZt cells.

Reviewer #3 (Recommendations for the authors):

This revised manuscript has been significantly reorganized to focus on bitter taste. The streamlining of results and inclusion of clarifying data significantly strengthen this manuscript, and further our understanding of bitter taste processing. With these changes, the authors have addressed all of my previous concerns.

We appreciate this feedback and thank the reviewer for helping us improve the paper.

https://doi.org/10.7554/eLife.84856.sa2

Article and author information

Author details

  1. Julia U Deere

    Zuckerman Mind Brain Behavior Institute, Columbia University, New York, United States
    Present address
    Rockefeller University, New York, United States
    Contribution
    Formal analysis, Investigation, behavioral experiments; brain dissections and immunostaining
    Competing interests
    No competing interests declared
  2. Arvin A Sarkissian

    Neuroscience Graduate Program, Emory University, Atlanta, United States
    Contribution
    Software, Formal analysis, Investigation, Visualization, Writing – original draft, Writing – review and editing, connectome analysis
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6663-2989
  3. Meifeng Yang

    Department of Biology, Emory University, Atlanta, United States
    Contribution
    Formal analysis, Investigation, behavioral experiments
    Competing interests
    No competing interests declared
  4. Hannah A Uttley

    Zuckerman Mind Brain Behavior Institute, Columbia University, New York, United States
    Contribution
    Formal analysis, Investigation, behavioral experiments; brain dissections and immunostaining
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0573-6546
  5. Nicole Martinez Santana

    Zuckerman Mind Brain Behavior Institute, Columbia University, New York, United States
    Contribution
    Investigation, behavioral experiments
    Competing interests
    No competing interests declared
  6. Lam Nguyen

    Department of Biology, Emory University, Atlanta, United States
    Contribution
    Formal analysis, Investigation, behavioral experiments
    Competing interests
    No competing interests declared
  7. Kaushiki Ravi

    Department of Biology, Emory University, Atlanta, United States
    Contribution
    Formal analysis, Methodology, behavioral experiments
    Competing interests
    No competing interests declared
  8. Anita V Devineni

    1. Zuckerman Mind Brain Behavior Institute, Columbia University, New York, United States
    2. Neuroscience Graduate Program, Emory University, Atlanta, United States
    3. Department of Biology, Emory University, Atlanta, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing, brain dissection and immunostaining; calcium imaging
    For correspondence
    anita.devineni@emory.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9540-8655

Funding

Whitehall Foundation (2022-08-017)

  • Anita V Devineni

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Richard Axel for his generous support and helpful discussions. We thank Barbara Noro for general advice, sharing unpublished fly lines and code, and providing manuscript feedback. We thank Katie Shakman for the initial setup of the behavioral arena and some analysis code, Tanya Tabachnik and Rick Hormigo for technical assistance with the behavioral arena, Pavel Masek for help with setting up the optoPAD, Jan Hawes for assistance with setting up a new lab, and Deepti Suchindran and Crystal Wang for assisting with pilot experiments in the new lab. We thank Yu-Chieh David Chen and Chris Rodgers for comments on the manuscript. We thank John Carlson, Gilad Barnea, the Janelia Research Center, and the Bloomington Drosophila Stock Center (BDSC) for providing fly strains. This work was supported by internal funding to the Axel lab, startup funding from Emory University (AVD), and a grant from the Whitehall Foundation (AVD).

Senior Editor

  1. Claude Desplan, New York University, United States

Reviewing Editor

  1. Ilona C Grunwald Kadow, University of Bonn, Germany

Reviewer

  1. Alex C Keene, Texas A&M University, United States

Publication history

  1. Preprint posted: February 10, 2022 (view preprint)
  2. Received: November 11, 2022
  3. Accepted: January 10, 2023
  4. Version of Record published: January 24, 2023 (version 1)

Copyright

© 2023, Deere et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 340
    Page views
  • 49
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Julia U Deere
  2. Arvin A Sarkissian
  3. Meifeng Yang
  4. Hannah A Uttley
  5. Nicole Martinez Santana
  6. Lam Nguyen
  7. Kaushiki Ravi
  8. Anita V Devineni
(2023)
Selective integration of diverse taste inputs within a single taste modality
eLife 12:e84856.
https://doi.org/10.7554/eLife.84856