Parallel Neural Pathways Mediate CO2 Avoidance Responses in Drosophila

See allHide authors and affiliations

Science  14 Jun 2013:
Vol. 340, Issue 6138, pp. 1338-1341
DOI: 10.1126/science.1236693

Too Much or Too Little

An important task of the nervous system is to distribute information appropriately throughout the brain. The olfactory and gustatory systems of Drosophila provide good models for understanding these processes and the underlying mechanisms (see the Perspective by Su and Carlson). Lin et al. (p. 1338) mapped out the circuit that detects carbon dioxide (CO2), an important environmental and communication signal for fruit flies. Two distinct classes of projection neurons mediate avoidance of high and low concentrations of CO2, while a third class, comprising inhibitory neurons, shuts down the low-concentration pathway at high concentrations. In contrast to other basic taste qualities, salt is innately attractive at low concentrations, but aversive at high concentrations. The mechanisms underlying salt detection are poorly understood in any species mainly because of a lack of specific molecular tools. Zhang et al. (p. 1334) discovered that Drosophila uses two types of gustatory receptor neurons to distinguish between high and low concentrations of salt. One type is activated maximally by low salt and induces attractive feeding behavior. The other class of receptors is activated primarily by high salt and leads to avoidance behavior.


Different stimulus intensities elicit distinct perceptions, implying that input signals are either conveyed through an overlapping but distinct subpopulation of sensory neurons or channeled into divergent brain circuits according to intensity. In Drosophila, carbon dioxide (CO2) is detected by a single type of olfactory sensory neuron, but information is conveyed to higher brain centers through second-order projection neurons (PNs). Two distinct pathways, PNv-1 and PNv-2, are necessary and sufficient for avoidance responses to low and high CO2 concentrations, respectively. Whereas low concentrations activate PNv-1, high concentrations activate both PNvs and GABAergic PNv-3, which may inhibit PNv-1 pathway-mediated avoidance behavior. Channeling a sensory input into distinct neural pathways allows the perception of an odor to be further modulated by both stimulus intensity and context.

Insects detect odor with olfactory sensory neurons (OSNs), which converge to the antennal lobe (AL) before conveyance to the mushroom body (MB) and lateral horn (LH) via stereotyped projection neurons (PNs) (14). In Drosophila, carbon dioxide (CO2) concentrations lower than 2% activate only one type of OSN that expresses Gr21a and Gr63a receptors and projects to a single V-glomerulus (59). We examined the morphology and functionality of PNs innervating the V-glomerulus (PNvs) with regard to CO2 responses.

We expressed a photoactivatable green fluorescent protein (PaGFP) in ~60% of neurons using Cha-Gal4>UAS-PaGFP flies (1012) and labeled candidate PNvs by means of targeted photoconversion (fig. S1A). Although nearby tracts could have been labeled, the circuits were complemented and validated by browsing single neuron representations in the FlyCircuit database (13). Up to 12 heteromorphic PNvs with different morphologies may link the V-glomerulus and higher brain centers (figs. S1B and S2) via the inner, medial, and outer antennocerebral tracts (iACT, mACT, and oACT, respectively).

To assess the functional roles of these PNvs, we identified 50 Gal4 lines that labeled the V-glomerulus, including seven putative PNvs. We used genetic mosaic analyses to resolve individual neurons, using either repressible cell marker (MARCM) (14), FLP-out (2), or Brainbow (fig. S3) techniques to identify four genetically addressable PNvs: PNv-1 (VT33008-Gal4, VT1606-Gal4, VT31497-Gal4, and VT48643-Gal4) (Fig. 1, A to D), which links the bilateral V-glomeruli via oACT to the lateral horn (LH) and calyx (Cal); PNv-2 (E0044-Gal4) (Fig. 1E), which connects a single V-glomerulus via iACT to the bilateral superior dorsofrontal protocerebrum (SDFP); PNv-3 (VT12760-Gal4) (Fig. 1F), which innervates all glomeruli of a single AL and projects via mACT to the LH, inner dorsolateral protocerebrum (IDLP), and SDFP; and PNv-4 (E0564-Gal4) (Fig. 1G), which links two ALs via oACT to the SDFP, superpeduncular protocerebrum (SPP), caudal ventrolateral protocerebrum (CVLP), IDLP, and LH. Their termini are primarily localized to the SDFP and LH (fig. S4) (13).

Fig. 1 PNvs structural and functional connectivity.

(Aa to Ga) Expression patterns of seven different PNv-Gal4s with V-glomerulus (dotted circle) innervations. Gal4 neurons were labeled with UAS-mCD8::GFP (green), and brain neuropils were immunostained with a DLG antibody (magenta). (Ab to Gb) Putative dendrites in the V-glomerulus labeled with UAS-Dscam[17.1]::GFP (green). Gal4 neurons were labeled with UAS-mKO (magenta). (Ac to Gc) GRASP visualization of structural contacts between L5131-LexA and PNv-Gal4s. Scale bar, 50 μm. (Ad to Gd) Morphology of a single PNv. (Ae to Ge) GCaMP changes (ΔF/F0) in the V-glomerulus in response to odor stimuli. Each value is mean ± SEM (n = 6 to 10 samples, *P < 0.05, **P < 0.01, ***P < 0.001). More detailed legends of this and the other figures are available in the supplementary materials.

Using Dscam[exon17.1]::GFP as a dendritic marker (15), we demonstrated that they all project into the V-glomerulus (Fig. 1, Ab–Gb). We used the GRASP (green fluorescent protein reconstitution across synaptic partners) technique to assess whether these dendrites receive input from CO2 OSNs (16, 17). Half of the split-GFP GRASP reporter was expressed in OSNv neurons by using L5131-LexA (fig. S5Aa), which specifically labels OSNs (fig. S5Ab) that are Gr21a-nlsDsRed–positive (fig. S5B), innervate the V-glomerulus (fig. S5Ca), and respond to 0.5% CO2 (fig. S5C, b and c). The other half was expressed in distinct PNvs by using an appropriate Gal4 driver. In all cases, GRASP signals were observed in the V-glomerulus (Fig. 1, Ac to Gc). In the control experiment using Mz19-Gal4 expressed in PNs innervating several other glomeruli (fig. S5D), GRASP signals were absent in the V-glomerulus (fig. S5E).

Monitoring functional responses in the V-glomerulus with a genetically encoded calcium indicator (GCaMP) revealed that all four PNv types responded to CO2 but not air, methylcyclohexanol (MCH), or octanol (OCT). PNv-1 and PNv-4 responded equally to 0.5 and 2% CO2, whereas PNv-2 and PNv-3 responded dose-dependently (Fig. 1, Ae to Ge). Quantitative fluorescence measurement showed that basal GCaMP expression driven by seven PNv-Gal4 lines varied more than twofold (fig. S6A). A functional curve to different CO2 concentrations showed that CO2-response kinetics were independent of Gal4 efficiency (fig. S6B). Although GCaMP imaging is not fast enough to monitor temporal PN responses (18), the degrees of GCaMP change were always PNv-1 > PNv-2 > PNv-4 > PNv-3 (fig. S6B). PNv-1 has significantly more dendrites than other PNvs (Fig. 1, Ad to Gd, and fig. S7). PNv-1 and PNv-2 only accounted for one neuron in each hemisphere (Fig. 1 and fig. S16). The three isomorphic PNv-3s exhibited similar CO2-response kinetics (fig. S8), which is consistent with direct intracellular recording showing that DM6 PNs with similar morphologies innervating the same glomerulus exhibit similar odor-response properties (19).

Are any of these CO2-responsive PNvs necessary for avoidance behavior? Avoidance to 0.5% CO2 was impaired when PNv-1 neurotransmission was acutely disrupted with UAS-shits1 at 30°C (Fig. 2A) but not 21°C (Fig. 2B). These flies exhibited normal avoidance to MCH and OCT at 30°C (fig. S9), and 0.5% CO2 avoidance was normal when PNv-2, PNv-3, or PNv-4 neurotransmission was disrupted at 30°C (Fig. 2A). A different pattern emerged for 2% CO2. Avoidance behavior was normal when neurotransmission from PNv-1, PNv-3, or PNv-4 was blocked (Fig. 2C). Instead, 2% CO2 avoidance was impaired when PNv-2 neurotransmission was disrupted with UAS-shits1 at 30°C (Fig. 2C) but not 21°C (Fig. 2D). We verified this by silencing neural activity with adult-stage–specific expression of inwardly rectifying potassium channel Kir2.1 (20). Silencing PNv-1 only impaired 0.5% CO2 avoidance, and silencing PNv-2 only impaired 2% CO2 avoidance. Silencing PNv-3 or PNv-4 did not affect CO2 avoidance (fig. S10).

Fig. 2 PNv-1 and PNv-2 outputs are necessary to elicit avoidance to 0.5 and 2% CO2, respectively.

(A and B) Avoidance response to 0.5% CO2 in PNv-Gal4>UAS-shits1 flies at 30°C (A) and 21°C (B). (C and D) Avoidance response to 2% CO2 in PNv-Gal4>UAS-shits1 flies at 30°C (C) and 21°C (D). Seven different PNv-Gal4 drivers were used: PNv-1a (VT33008-Gal4), PNv-1b (VT1606-Gal4), PNv-1c (VT31497-Gal4), PNv-1d (VT48643-Gal4), PNv-2 (E0044-Gal4), PNv-3 (VT12760-Gal4), and PNv-4 (E0564-Gal4). Each value is mean ± SEM (n = 6 to 8 experiments, **P < 0.01, ***P < 0.001).

For PNv-2, we identified E0044-Gal4 after screening thousands of Gal4 expression patterns, including those in FlyLight (21) and BrainBase ( Though E0044-Gal4 expression is specific in the brain (Fig. 1Ea), it is important to confirm that the observed changes are not due to other neuronal types. E0044-Gal4 was also expressed in paired dorsal anterior lateral (DAL) neurons, a few ascending subesophageal ganglion neurons, several descending pars intercerebralis neurons, some sensory input neurons terminating in the ALs and antennal mechanosensory motor center (AMMC), and brain-surface glial cells (fig. S11A). First, 2% CO2 avoidance was impaired by blocking neurotransmission with UAS-shits1 from E0044-Gal4 (fig. S12A) but normalized when Gal4 expression was inhibited by Cha-Gal80 (fig. S12B), indicating that Cha+ neurons are involved. Second, impaired 2% CO2 avoidance was unaffected by blocking PNv-2 neurotransmission in flies carrying ey-flp/E0044-Gal4;UAS>shits1>stop;cry-Gal80 transgenes that exclude DAL and sensory neurons (fig. S11, B and C) (22). Third, blocking neurotransmission in DAL neurons with specific and strong G0431-Gal4 driver did not impair 2% CO2 avoidance (fig. S11, D to H). Last, GCaMP responses to 2% CO2 were only evident in PNv-2s in the V-glomerulus (dendrites) and SDFP (axons) (fig. S11I).

Would activation of either PNv-1 or PNv-2 alone with a blue light-gated ion channel channelrhodopsin-2 (ChR2) (23, 24) be sufficient to elicit CO2 avoidance behavior? We used a modified optogenetic T-maze (8) (fig. S13A) equipped with blue and yellow light-emitting diodes (LEDs) (fig. S13B) and an air-cooling system (fig. S13C) (25). Intact antennae were necessary for avoidance behavior elicited by Gr21a-Gal4 neurons but unnecessary for attraction behavior mediated by Gr5a-Gal4 neurons (Fig. 3A). A similar avoidance response was observed upon optogenetic activation of PNv-1 (using VT31497-Gal4 and VT1606-Gal4) or PNv-2 (using E0044-Gal4), thus mimicking the response to CO2 exposure (Fig. 3, B and C). Avoidance behavior persisted without antennae. ChR2 activation of PNv-3 or PNv-4 did not elicit avoidance (fig. S14). Because neither blocking neurotransmission with shits1 nor silencing neural activity with Kir2.1 completely abolished CO2 avoidance, other PNv types (fig. S2) may drive partial CO2 avoidance behavior.

Fig. 3 Activation of PNv-1 or PNv-2 alone elicits avoidance behavior in an optogenetic T-maze.

(A) Phototaxis responses to ChR2 activation of specific Gal4 neurons. (B) Avoidance responses to ChR2 activation of PNv-2. E0044-Gal4>UAS-ChR2 flies avoided the blue light. (C) Avoidance responses to ChR2 activation of PNv-1. Blue and yellow bars indicate the blue and yellow LED arms, respectively. Each value is mean ± SEM (n = 6 to 8 experiments, *P < 0.05, ***P < 0.001).

At 0.5% CO2, PNv-1 was functionally more responsive than was PNv-2 (fig. S6B), explaining why only the former contributes to the low CO2 avoidance response. However, 2% CO2 activates both PNv classes. Why does 2% CO2 avoidance require PNv-2 but not PNv-1? The PNv-1 response remained higher at 2% CO2 (fig. S6). Blocking neurotransmission from γ-aminobutyric acid (GABA)–releasing (GABAergic) local neurons (LNs) in the ALs did not impair 2% CO2 avoidance (fig. S15). One possibility is that the PNv-1 pathway is gated downstream at higher CO2 levels. PNv-3s are GABAergic (Fig. 4, A and B), but PNv-1, PNv-2, and PNv-4 are not (fig. S16). PNv-3 and PNv-2 are highly activated by higher CO2 concentrations (Fig. 1, Ee and Fe), and LH imaging revealed that PNv-3 axon terminals were insensitive to 0.5% but highly sensitive to 2% CO2 (Fig. 4C). This suggests a model in which PNv-3 might inhibit and gate the PNv-1 pathway in the LH. In this scenario, only the PNv-1 pathway would be active at low CO2 concentrations, and higher levels would recruit PNv-3 and PNv-2, blocking the PNv-1 pathway while activating the PNv-2 pathway to elicit avoidance behavior (Fig. 4D).

Fig. 4 PNv-3 gates CO2 avoidance neural pathways.

(A and B) PNv-3 (arrows) labeled with UAS-mCD8::GFP (green) in VT12760-Gal4 and GH146-Gal4 were anti-GABA immunopositive (magenta). Scale bar, 25 μm. (C) PNv-3s exhibited broadly tuned responses to odorants. Each value is mean ± SEM (n = 6 samples, ***P < 0.001). MSC, methyl salicylate; IA, isopentyl acetate; EP, ethyl propionate; EA, ethyl acetate; GA, geranyl acetate; ACV, apple cider vinegar. (D) Shunting inhibition model showing brain circuit routing for 0.5% (green) and 2% CO2 (magenta). Excitatory and inhibitory stimuli are represented by arrows and “T” bars, respectively. Weak responses are represented as dashed lines. (E) Avoidance responses to 0.5 or 2% CO2 when ChR2 in PNv-3 (VT12760-Gal4) was activated by blue light. Blue and yellow bars indicate blue and yellow LED arms, respectively. Each value is mean ± SEM (n = 6 experiments, **P < 0.01). (F) Inhibition of PNv-1 pathway by PNv-3, not GABAergic LNs, during 2% CO2 avoidance. Neurotransmission outputs of PNv-2 in E0044-Gal4 and PNv-3 in VT12760-Gal4 (a) and GH146-Gal4 (b) were acutely blocked separately or together by UAS-shits1 at 30°C. Each value is mean ± SEM (n = 6 to 8 experiments, ***P < 0.001).

This model makes two key predictions. First, optogenetic PNv-3 activation should inhibit 0.5% but not 2.0% CO2 avoidance. Second, PNv-3 silencing should leave the PNv-1 pathway active at 2% CO2. Indeed, ChR2 activation of GABAergic PNv-3 impaired avoidance to 0.5% but not 2% CO2 (Fig. 4E), mimicking the effect of blocking PNv-1 alone (Fig. 2A). Conversely, acutely blocking PNv-3 neurotransmission did not impair avoidance responses, but PNv-2 activity was no longer necessary for 2% CO2 avoidance. In contrast, PNv-2 activity remained necessary for 2% CO2 avoidance when GABAergic LNs in the AL were acutely blocked (Fig. 4F). Direct PNv-3 inhibition to PNv-1 is unlikely because PNv-1 dendrites in the V-glomerulus (Fig. 1, Ae to De, and fig. S6) and axonal terminals in the LH (fig. S17) were not less responsive at high CO2 levels when PNv-3 was active. This model does not exclude the possibility of PNv-3 inhibition on possible redundant PNv pathways necessary for 0.5% CO2 avoidance (Fig. 4D).

What benefit might be gained by channeling CO2 signals through distinct PN pathways to elicit the same avoidance behavior? Although CO2 is part of a deterrent signal issued by other flies under stress conditions (5), high levels of CO2 can also indicate food sources. Some food odorants can directly inhibit CO2-sensitive neurons in the antenna (26). Behavior state is also critical, and in-flight Drosophila exhibits CO2 attraction (27). The PN circuitry described here would allow concentration-dependent CO2 avoidance. PNv-3s may also modulate other olfactory behaviors because they innervate many glomeruli (Fig. 1F) and respond to multiple odorants (Fig. 4C).

Supplementary Materials

Materials and Methods

Figs. S1 to S17


References and Notes

  1. Drosophila detects CO2 through olfactory and gustatory systems, inducing avoidance and approach behaviors, respectively. Concentrations higher than 2% CO2 dissolved in antennal lymph fluid can produce acid that is detected by acid-sensing IR64a neurons.
  2. The blue and yellow lights were balanced so that flies were naturally attracted to visible light distributed equally between two arms (fig. S13D). We validated effectiveness by expressing ChR2 in Gr5a-Gal4 neurons (required for sucrose attraction) or Gr21a-Gal4 neurons (required for CO2 avoidance); as expected, these flies were attracted or repelled, respectively, by the blue light.
  3. Acknowledgments: We thank J. Lippincott-Schwartz, D. F. Reiff, K. Scott, T. Lee, G. Struhl, A. Fiala, and the Bloomington Stock Center for fly stocks. We also thank G. S. B. Suh for helpful comments on the manuscript, Y.-T. Huang and S.-T. Wu for generating and testing Brainbow flies, and the National Center for High-performance Computing for FlyCircuit images. This work was supported by grants from the National Science Council and Ministry of Education in Taiwan.

Stay Connected to Science

Navigate This Article