Detection of Near-Atmospheric Concentrations of CO2 by an Olfactory Subsystem in the Mouse

See allHide authors and affiliations

Science  17 Aug 2007:
Vol. 317, Issue 5840, pp. 953-957
DOI: 10.1126/science.1144233


Carbon dioxide (CO2) is an important environmental cue for many organisms but is odorless to humans. It remains unclear whether the mammalian olfactory system can detect CO2 at concentrations around the average atmospheric level (0.038%). We demonstrated the expression of carbonic anhydrase type II (CAII), an enzyme that catabolizes CO2, in a subset of mouse olfactory neurons that express guanylyl cyclase D (GC-D+ neurons) and project axons to necklace glomeruli in the olfactory bulb. Exposure to CO2 activated these GC-D+ neurons, and exposure of a mouse to CO2 activated bulbar neurons associated with necklace glomeruli. Behavioral tests revealed CO2 detection thresholds of ∼0.066%, and this sensitive CO2 detection required CAII activity. We conclude that mice detect CO2 at near-atmospheric concentrations through the olfactory subsystem of GC-D+ neurons.

CO2 is an olfactory stimulus for many invertebrates (1, 2). CO2 levels fluctuate locally with biological activities, such as animal respiration, plant photosynthesis, and the decomposition of organic matter. CO2 signals regulate many insect innate behaviors, such as seeking food and hosts, avoiding stressful environments, and ovipositioning (36). CO2 has no discernable odor to humans, but at high concentrations (>30%), it produces a pungent trigeminal sensation in the nasopharynx (7). Carbonic anhydrase (CA), an enzyme that is implicated in CO2 sensing by peripheral systems such as carotid chemoreceptors (2, 8), is expressed in a subset of olfactory sensory neurons (OSNs) in several vertebrate species (8, 9). Studies indicate that rats can detect CO2 at levels above 0.5% (10, 11). It remains unknown whether mammals can detect CO2 at concentrations near the atmospheric level (0.038%), and if so, whether this detection is mediated by a specialized olfactory subsystem.

In the mouse olfactory epithelium (OE), conventional OSNs use adenosine 3′-5′ monophosphate (cAMP) in odorant-evoked signal transduction, but a minor population of neurons appears to use guanosine 3′-5′ monophosphate (cGMP) (1214). Unlike conventional OSNs, this minor population of cells expresses phosphodiesterase 2A (PDE2A), guanylyl cyclase D (GC-D), and cGMP-sensitive cyclic nucleotide-gated (CNG) channels (1214). They project axons to the necklace glomeruli, a set of glomeruli that form a “necklace” in the caudal end of the olfactory bulb (OB) (13, 15). The necklace glomeruli have been implicated in detecting suckling pheromones and pheromonal compounds (16, 17). Here we show that mice can sense CO2 at near-atmospheric levels through the subset of olfactory neurons that express PDE2A and GC-D.

We first examined the gene expression profile of neurons expressing PDE2A (PDE2A+ neurons) by single-cell serial analysis of gene expression of dissociated cells from the caudal OE of mice (18). We found that PDE2A+ neurons expressed high levels of CA type II (CAII), whereas conventional OSNs did not (fig. S1, A to C). Expression of CAII/Car2 mRNA by a subset of OE cells was verified by in situ hybridization (fig. S1D). Immunostaining showed abundant CAII expression by a small population of cells in the caudal OE. Most of these CAII+ cells clustered with bilateral symmetry in the cul-de-sac regions within the caudal recesses of the nasal cavity (Fig. 1A). CAII localized with PDE2A in the OE (Fig. 1, B to D) and in ∼20 glomeruli more than 30 μm in diameter in the caudal OB (Fig. 1, E to H; and fig. S2, n = 6 mice). Because PDE2A in turn is coexpressed with GC-D (13, 14), we analyzed a mouse strain with a targeted mutation in the GC-D locus that produces bicistronic messages causing cotranslation of GC-D with a fusion protein of tau and green fluorescent protein (GFP) (GCD-ITG mice). CAII immunoreactivity and GFP intrinsic fluorescence completely overlapped in the OE (Fig. 1, I to K) and OB (Fig. 1, L to N, n = 2 mice). Henceforth, we refer to the subset of CAII+–PDE2A+–GC-D+ neurons as GC-D+ neurons.

Fig. 1.

CAII immunoreactivity in GC-D+ neurons and necklace glomeruli. (A) Bilaterally symmetric distribution of CAII+ cells (arrows) in the OE. (B) High-power view of CAII immunoreactivity in a CAII+ cluster. (C) PDE2A immunoreactivity in the same section as (B). (D) Overlay of (B) and (C). (E) Acoronal section of the caudal OB showing CAII+ glomeruli (arrows) with largely bilateral symmetry. (F) High-power view of CAII immunoreactivity within the dashed box in (E). (G) PDE2A immunoreactivity. (H) Overlay of (F) and (G). Blue, DAPI labeling. (I to K) CAII expression overlaps with GFP labeling in the OE of GCD-ITG mice. (I) CAII immunoreactivity. (J) GFP immunoreactivity within the same region as (I). (K) Overlay of (I) and (J). (L to N) CAII expression overlaps with GFP labeling in the OB of GCD-ITG mice. (L) CAII+ glomeruli. (M) GFP+ glomeruli within the same region as (L). (N) Overlay of (L) and (M). Scale bars in (B), 20 μm; (F), 100 μm; (I), 10 μm; and (L), 50 μm.

CA catalyzes the reversible reaction of CO2 + H2O [rlhar2] HCO3 + H+ (19, 20). Because CA is implicated in CO2 sensing, we tested whether GC-D+ neurons respond to CO2 using calcium imaging in an intact OE preparation from GCD-ITG mice (18, 21). OSNs were incubated with rhod-2 acetoxymethyl ester (AM), a calcium-sensitive red fluorescent dye. GC-D+ neurons were identified by intrinsic green fluorescence of GFP in their dendritic knobs and somata (Fig. 2A), and the uptake of rhod-2 AM dye was confirmed by red fluorescence (Fig. 2B). Tissue was continuously superfused with oxygenated HEPES-based Ringer solution. CO2 was delivered by superfusing CO2-bubbled Ringer solution into the imaging chamber, resulting in peak CO2 concentrations of 0.4 to 6.4 mM as calculated by CO2 solubility (22). After CO2 application, we observed fluorescence enhancement in both dendritic knobs and underlying somata of GC-D+ neurons (Fig. 2C, n = 428 cells). CO2 responses of GC-D+ neurons were dose-dependent, with substantial activation starting at 1.0 mM CO2 (Fig. 2D). GC-D+ neurons were not activated by CO2-free Ringer solution with adjusted pH or bicarbonate levels (Fig. 2E and fig. S3A). When tested with 6.4 mM CO2, most GC-D cells were not activated [n = 256 out of 260 (256/260) tested], and a few of them were only weakly activated (fig. S4, n = 4/260).

Fig. 2.

CO2 activates GC-D+ neurons, and this activation requires CA activity and the opening of CNG channels. (A) GFP labeling of a cluster of GC-D+ neurons within an intact epithelial preparation from a GCD-ITG mouse. The arrow points to a dendritic knob. (B) Uptake of the calcium-sensitive dye rhod-2 AM into GC-D+ neurons within the same region as (A). (C) Map of normalized fluorescence changes (ΔF/F) within the same region as (A) and (B) showing activation of GC-D+ neurons after CO2 application. The color scale at the right indicates a ΔF/F range of 0 to 20%. For all GC-D+ neurons tested, ΔF/F = 11.1 ± 0.4% (mean ± SEM, n = 428 cells, 8 mice). (D) Dose-response curve of the calcium signal to CO2 (n = 123 cells). (E) Group data showing that GC-D+ neurons respond to 6.4 mM CO2 but not to pH-adjusted and bicarbonate controls (ΔF/F = 0.7 ± 0.1% for pH-adjusted control, n = 123 neurons, P < 0.001, t test, CO2 versus pH-adjusted control; ΔF/F = 0.4 ± 0.2% for bicarbonate control, n = 170 neurons, P < 0.001, t test, CO2 versus bicarbonate). Error bars in this and the following figures indicate SEM. (F) Group data showing blockade of CO2 responses by the CA inhibitor AZ (ΔF/F = 12.8 ± 0.5% in baseline conditions versus 0.5 ± 0.1% in AZ, n = 225 neurons, P < 0.001, t test). (G) Group data showing the absence of CO2 responses in Ca2+-free Ringer solution (ΔF/F = 13.0 ± 0.5% in normal Ringer solution versus 0.0 ± 0.1% for Ca2+-free Ringer, n = 118 neurons, P < 0.001, t test). (H and I) A single example (H) and group data (I) showing the reversible blockade of CO2 responses by l-cis-diltiazem (ΔF/F = 12.3 ± 1.1% in baseline, 1.8 ± 0.3% during diltiazem application, and 6.9 ± 0.5% during washout; n = 225; P < 0.001; t test). Scale bar in (A), 10 μm; in (H), 50 s and a ΔF/F of 10%.

To examine the role of CA in CO2 sensing by GC-D+ neurons, we applied acetazolamide (AZ, 1 mM), a CA enzymatic inhibitor. AZ completely eliminated the CO2 responses (Fig. 2F and fig. S3B). Calcium signals in response to CO2 were absent in Ca2+-free Ringer solution, indicating that calcium signals were due to external calcium influx (Fig. 2G and fig. S3C). Low concentrations of l-cis-diltiazem (10 μM), a CNG channel blocker (23), reversibly blocked the response of GC-D+ neurons to CO2 (Fig. 2, H and I). Thus, CO2 specifically activates GC-D+ neurons, and this activation requires CA enzymatic activity and the opening of CNG channels.

To test CO2 sensitivity in vivo, we recorded from bulbar neurons associated with the necklace glomeruli after exposing anesthetized mice to CO2 in the airflow delivered to their nostrils (18). Out of ∼5000 cells tested with 0.5% CO2 in the gas phase, excitatory responses were found in 260 cells, almost all of which were near the necklace glomeruli. By juxtacellular labeling, we identified five cells with dendritic processes that were delineated clearly and extended into the necklace glomeruli (Fig. 3A). These five cells were activated by external CO2 at levels above 0.1% (Fig. 3, B and D). Excitatory responses to CO2 pulses were characterized by initial bursting with short latency [306.3 ± 22.6 ms (mean ± SEM, n = 24 cells)], vigorous firing throughout the stimulus, and brief inhibition after stimulus termination (Fig. 3C). Control pulses (0% CO2 and 20% O2) with the same flow rate had no effect, confirming that the responses were evoked by CO2 but not by increased airflow. The CO2 responses of bulbar neurons were dose-dependent (Fig. 3, D and E). For the 18 most sensitive cells tested with a range of concentrations, significant activation was observed at 0.1% CO2. Response amplitudes rose steeply between 0.1 and 0.3% CO2 and saturated near 0.5% (Fig. 3E). Lower sensitivity to CO2 was observed in some neurons, with saturation above 2% (n = 30 cells). Activation of GC-D+ glomeruli by CO2 was further confirmed by immunostaining against c-Fos, a marker of neuronal activation in the olfactory bulb (17) (fig. S5).

Fig. 3.

CO2 response pattern and CO2 sensitivity of bulbar neurons in vivo evoked by exposing mice to CO2 in the airflow delivered to the nostrils. (A) Morphology of a cell with dendrites (red) extending into a CAII+ glomerulus (green). DAPI labeling (blue) delineates glomeruli. (B) Physiological trace of the cell in (A) showing vigorous response to 0.5% CO2. Horizontal bar at bottom indicates a 2-s CO2 pulse. The bottom trace indicates the respiratory rhythm monitored simultaneously. (C) Response pattern of another bulbar cell to a 0.5% CO2 pulse. (D) Dose-response curve of the cell in (A). (E) Dose-response curves of the 18 most sensitive neurons (connected lines of different colors). The thick red curve represents fitting with a sigmoid function. The inset shows a zoomed view between 0 and 0.5% CO2. Significant responses were observed at 0.1% CO2 (46.0 ± 0.5% of maximum responses, P < 0.001 compared to baseline, t test). Scale bar in (A), 20 μm.

Finally, we used behavioral assays to examine the sensitivity of CO2 detection by the mouse (18). Water-deprived adult mice were trained to lick a port for water delivery during 0.5% CO2 pulses and to not lick during control pulses with the same flow rate and oxygen concentrations (fig. S6) (18). Training over ∼10 days resulted in stable performance at levels of >90% correct responses (Fig. 4A, n = 10 mice). During tests, response accuracy remained high for test stimuli with ≥0.1% CO2 but fell sharply to near-chance level as CO2 levels decreased further (Fig. 4B). The response ratio at different CO2 concentrations fitted to a Weilbull psychometric function reporting a CO2 detection threshold of –1.18 ± 0.07 log units or 0.066% CO2 (Fig. 4B and fig. S7, n = 8 mice) (24), which is just above the average CO2 concentration in the atmosphere. The behavioral threshold also matched the response sensitivity of bulbar neurons (Fig. 3E). When the OE was ablated by nasal irrigation with zinc sulfate (50 μl at a concentration of 5%), mice were unable to detect 0.5% CO2 pulses (Fig. 4C). Consistent with OSN regeneration, they slowly regained their ability in 1 to 2 weeks (25).

Fig. 4.

Mice detect CO2 with behavioral thresholds near atmospheric levels. (A) Learning curve for CO2 detection by wild-type mice. The dashed line indicates 90% correct. (B) Mean response ratio to different CO2 concentrations from test sessions over 7 days. The curve indicates data fitting with a Weibull psychometric function. The dashed line indicates near-chance response level to 0% CO2. (C) After lesion of the OE with 5% zinc sulfate (day 0), the mouse's ability to detect 0.5% CO2 was eliminated over the following day (percent correct = 46.2 ± 4.1%, n = 5 mice, P < 0.001; within group t test) and then gradually recovered within 7 to 14 days. (D) CO2 leads to avoidance behavior. Bars indicate the avoidance index revealed by T-maze tests. Avoidance of CO2 was abolished by lesions of the OE after nasal irrigation with 5% zinc sulfate. **P < 0.01; ***P < 0.001, t test. (E) Learning curves of both wild-type (gray line and squares) and CAII mutant mice (black line and circles) for odorant and 0.5% CO2 detection. AA, amyl acetate; PMK, phenylmethylketone, also known as acetophenone. ***P < 0.001 for comparison between wild-type and mutant mice.

CO2 repels Drosophila (4) but attracts mosquitoes to hosts (3). In a T-maze assay, we found that mice avoided CO2 at concentrations as low as 0.2% (n = 31 mice, P <0.01, t test). The avoidance became substantially more marked at higher concentrations (Fig. 4D, 0.4 to 3.2% CO2). Lesions of the OE by nasal irrigation with zinc sulfate eliminated the avoidance of CO2 (Fig. 4D). The effects of lesions on mice both trained and untrained for CO2 detection, in combination with the fast reaction time for CO2 detection (fig. S8) and the low CO2 sensitivity of the trigeminal system (fig. S9), demonstrate that the OE has a direct role in CO2 detection.

Because CAII is expressed exclusively in GC-D+ neurons and its activity is essential for CO2 responses in these neurons (Figs. 1 and 2), we examined the behavioral phenotype of mice homozygous for a null mutation in the CAII/Car2 gene that was induced by chemical mutagenesis with N-ethyl-N-nitrosourea (fig. S10) (20, 26). These mutant mice were unable to detect CO2 at concentrations below 1% (fig. S11), but their learning curve and detection threshold for detecting amyl acetate (0.10% saturated vapor) were indistinguishable from those of wild-type mice (Fig. 4E, n = 6 for mutant and 7 for wild-type mice) (18). After learning the behavioral paradigm with amyl acetate, both wild-type and CAII mutant mice learned to switch to acetophenone as the new rewarding stimulus (Fig. 4E). Whereas wild-type mice switched to CO2 as the new rewarding stimulus in less than 50 trials, CAII mutant mice could not do so even after intensive training (Fig. 4E). Thus, CAII mutant mice have profound deficits in detecting CO2 but not other odorants. To confirm whether CA in the OE is critical for the sensitive detection of CO2, we applied the CA inhibitor AZ (50 μl at a concentration of 15%) to the nasal cavities of trained wild-type mice. The application of AZ but not of saline reversibly abolished the animals' ability to detect CO2 but not amyl acetate (fig. S12). Conventional OSNs require CNG channels consisting of CNGA2 subunits for odorant detection, and CNGA2 knockout mice are profoundly compromised in their sense of smell (17, 27). However, CNGA2 mutant mice had apparently normal CO2 learning curves and detection thresholds (fig. S13, n = 4 mice). Together, these results demonstrate that sensitive CO2 detection is exclusively mediated by GC-D+ neurons.

In this study, we identify the GC-D+ neurons as the olfactory subsystem that mediates CO2 detection in mice. It remains possible that GC-D+ neurons can detect other chemicals. CO2 responses in the GC-D+ neurons require CA enzymatic activity and the opening of CNG channels. We hypothesize that CAII converts CO2, a highly diffusible gas, into intracellular HCO 3 and H+, and that these ions in turn enhance cGMP levels and trigger the opening of CNG channels in GC-D+ neurons. The CO2-responsive neurons in insects provide a model system for studying innate behavior linked to a small number of neurons (4, 6). Similarly, GC-D+ neurons may be used as a model system for investigating innate avoidance behavior in mammals. Mammals have adapted to low-CO2 environments (∼0.03%) for millions of years (28). Some studies predict that the global CO2 level will exceed the detection threshold for mice within one century (29). Rising atmospheric CO2 levels may have ecological and ethological impacts on mammalian behavior.

Supporting Online Material

Materials and Methods

Figs. S1 to S13


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article