Understanding central processing requires precise monitoring of neural activity across populations of identified neurons in the intact brain. In the present study, we used recently optimized variants of the genetically encoded calcium sensor GCaMP (GCaMP3 and GCaMPG5G) to image activity among genetically and anatomically defined neuronal populations in the olfactory bulb (OB), including two types of GABAergic interneurons (periglomerular [PG] and short axon [SA] cells) and OB output neurons (mitral/tufted [MT] cells) projecting to the piriform cortex. We first established that changes in neuronal spiking can be related accurately to GCaMP fluorescence changes via a simple quantitative relationship over a large dynamic range. We next used in vivo two-photon imaging from individual neurons and epifluorescence signals reflecting population-level activity to investigate the spatiotemporal representation of odorants across these neuron types in anesthetized and awake mice. Under anesthesia, individual PG and SA cells showed temporally simple responses and little spontaneous activity, whereas MT cells were spontaneously active and showed diverse temporal responses. At the population level, response patterns of PG, SA, and MT cells were surprisingly similar to those imaged from sensory inputs, with shared odorant-specific topography across the dorsal OB and inhalation-coupled temporal dynamics. During wakefulness, PG and SA cell responses increased in magnitude but remained temporally simple, whereas those of MT cells changed to complex spatiotemporal patterns reflecting restricted excitation and widespread inhibition. These results suggest multiple circuit elements with distinct roles in transforming odor representations in the OB and provide a framework for further study of early olfactory processing using optical and genetic tools.
The olfactory bulb (OB) is an obligatory link between sensory input carried by olfactory receptor neurons (ORNs) and brain areas underlying odor perception and thus mediates the initial processing of olfactory information. The OB includes several classes of GABAergic local interneurons, including periglomerular (PG) cells, short axon (SA) cells, and granule cells, at least one class of glutamatergic local interneuron (external tufted cells), and several classes of principal output neurons (mitral/tufted [MT] cells) (Wachowiak and Shipley, 2006). Understanding how sensory inputs drive postsynaptic activity across these circuit elements and how the OB network transforms primary sensory representations is central to understanding mammalian olfactory processing.
Among ORNs, odor representations consist of odorant-specific and temporally dynamic patterns of input to OB glomeruli. Although these input patterns have been well characterized using imaging (Wachowiak and Cohen, 2001; Bozza et al., 2004; Verhagen et al., 2007; Ma et al., 2012), odor representations among defined populations of postsynaptic OB neurons have only begun to be described. Responses of individual MT cells have been characterized extensively using electrophysiological recordings, yet there are few descriptions of how other neuron types respond to odorants (Wellis and Scott, 1990; Tan et al., 2010; Kato et al., 2012). In addition, to compare odor representations at specific stages of processing directly within the OB, it is useful to monitor activity across many neurons of a given cell type under identical conditions, a goal that can be achieved efficiently using optical reporters of neural activity.
A few prior studies imaging postsynaptic odor representations in vivo have relied on voltage-sensitive dyes (Spors et al., 2006) or transgenic GCaMP expression (Chaigneau et al., 2007; Fletcher et al., 2009), approaches that lack clear cell-type specificity. A recent report using the genetically encoded Ca2+ sensor GCaMP3 expressed separately in MT and granule cells found a strong divergence in the response properties of these two populations and striking modulation of responsiveness by wakefulness and experience (Kato et al., 2012). How additional OB neuron populations represent odor information and how these representations compare with those of ORN inputs remain unclear.
In the present study, we used recently developed GCaMP variants with improved performance in vivo (Tian et al., 2009; Akerboom et al., 2012) to visualize how odor information is represented among three distinct subpopulations of OB neurons. We expressed the GCaMP variants GCaMP3 and GCaMP5G selectively in GABAergic PG interneurons, GABAergic and dopaminergic SA cells, and MT cells projecting to specific cortical targets. We also established a reliable quantitative relationship between GCaMP signals and spiking activity in OB neurons, compared spatiotemporal representations of odor information across ORNs, PG, SA, and MT cells, and found that the response properties of each of these neuronal populations are distinct and are differentially modified by anesthesia and wakefulness. Our results establish a framework for the continued study of olfactory information processing circuits in the intact animal.
Materials and Methods
Experiments were performed on transgenic mice expressing Cre recombinase (Cre) in defined neuronal populations. For selective expression of the fluorescent calcium indicators GCaMP3 and GCaMP5G, mice were either crossed to the Ai38 GCaMP3 reporter line or injected with a viral vector (see below). The mouse strains used were as follows: GAD2-IRES-Cre (Taniguchi et al., 2011), JAX Stock #010802 (The Jackson Laboratory); TH-Cre, JAX Stock #008601 (The Jackson Laboratory); PCdh21-Cre (Nagai et al., 2005), The GENSAT Project (Gong et al., 2007), Stock #030952-UCD (Mutant Mouse Regional Resource Centers); Ai38 (Zariwala et al., 2012), JAX Stock #014538 (The Jackson Laboratory); Ai9 (Madisen et al., 2010), JAX Stock #007905 (The Jackson Laboratory); and olfactory marker protein (OMP)-synaptopHluorin (spH; Bozza et al., 2004), JAX Stock #4946 (The Jackson Laboratory), backcrossed into an albino B6 background (JAX Stock #000058; The Jackson Laboratory). Mice used for OMP-spH imaging were heterozygous for the OMP-spH allele. All animals were <6 months of age by completion of data collection. Mice were housed up to three per cage and kept on a 12/12 h light/dark cycle. Food and water was available ad libitum except while the animals were restrained for imaging. All procedures were performed following National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the University of Utah institutional animal care and use committee.
Viral vector expression.
Viral vectors driving Cre-dependent expression of GCaMP3 (Tian et al., 2009) or GCaMP5G (Akerboom et al., 2012) were obtained from the University of Pennsylvania Vector Core (AAV2/1 serotype) and injected either directly into the dorsal OB or in designated targets within the primary olfactory cortex. All mice used for virus injection were homozygous for the allele driving Cre expression. For injection, mice were anesthetized with ketamine (70 mg/kg) and medetomidine (Domitor, 1 mg/kg; Pfizer), placed in a stereotaxic headholder, and a circular craniotomy (∼1 mm) was made over the injection site. Dorsal OB injections were performed using glass pipettes lowered to a depth of 200–300 μm; injections into anterior piriform cortex (aPC) used 33 gauge metal needles positioned at (in millimeters relative to bregma): 2.34 AP, +1.5 LM, −3.125 DV; medial amygdala. Virus was injected using a programmable syringe controller (QSI, Stoelting) at a rate of 0.1 μl/min. Large volumes of virus (0.5–1 μl) were injected to achieve widespread infection. Mice received atipamezole (Antisedan, 1 mg/kg s.c.; Pfizer) at the end of surgery to antagonize the medetomidine-induced effects and to accelerate recovery from anesthesia. Mice were given carprofen (Rimadyl, 5 mg/kg; Pfizer) as an analgesic immediately before surgery, and carprofen-supplemented food (2 mg/tablet) was provided for 4 d after surgery. Mice were single housed after surgery and typically imaged 14–28 d after virus injection.
In vitro imaging and electrophysiology.
OB slices were prepared from adult mice according to established techniques. Briefly, mice were anesthetized with isoflurane and decapitated. The brain was removed while submerged in ice-cold, oxygenated artificial CSF containing the following (in mm): 125 NaCl, 25 NaHCO3, 25 d-glucose, 2.5 KCl, 2 CaCl2, 1.25 NaH2PO4, and 1 MgCl2 (buffered to pH 7.4 with 95% O2/5% CO2). Sagittal slices of the OB were cut at a thickness of 300 μm with a vibrating microtome (VT1200; Leica). After a >1 h incubation at room temperature (∼22°C), slices were transferred to the stage of a custom-built microscope equipped with bright-field and two-photon fluorescence optics. Two- to 4 MΩ glass electrodes were drawn on a horizontal puller (P97; Sutter Instruments) and filled with an intracellular solution consisting of the following (in mm): 120 K-gluconate, 20 KCl, 10 HEPES, 7 diTrisPhCr, 4 Na2ATP, 2 MgCl2, 0.3 Tris-GTP, 0.2 EGTA, and buffered to pH 7.3 with KOH (Sigma-Aldrich). The internal solution contained 0.6% biocytin (Invitrogen) by weight in some experiments. Electrophysiological recordings were performed with a current–clamp amplifier (Multiclamp 700A; Molecular Devices), and data were acquired using RTXI software (www.rtxi.org). All recordings were conducted at 33°C. Fluorescence imaging was performed using targeted path scanning (Lillis et al., 2008) implemented on a custom-built two-photon microscope using a pulsed Ti:Sapphire laser (Chameleon Ultra II; Coherent) as a light source. Imaging was performed with a 20×, 0.95 numerical aperture (NA) objective (Olympus), and emitted fluorescence was filtered with a 520/65 nm emission filter (Semrock) and collected with low-noise photomultiplier tubes (H7360–01; Hamamatsu). Path scan rates depended on scan geometry in individual experiments, but ranged from 40 to 150 Hz.
Analysis of in vitro imaging data.
All analyses were performed using custom scripts written with MATLAB. Regions of interest along path scans were determined manually and pixels within each region were weighted equally. Fluorescence time series were low-pass filtered using a fourth-order Butterworth filter with a cutoff frequency of 10 Hz and converted to ΔF/F according to ΔF/F = (F − F0)/F0, where F0 was taken to be the mean fluorescence over each trial. Linear photobleaching trends were manually subtracted from raw data in a small number of experiments. Spike rate waveforms were computed by convolving binary spike trains with a double exponential function of the form exp(−t/toff) − exp(−t/ton) normalized to have unitary area. Correlation coefficients represent the fraction of variance explained by a linear model constrained to intersect the origin. Brute-force parameter searches were performed at resolutions of 1 and 10 ms for rise and decay times, respectively. Exponents were determined using the nlinfit function in MATLAB. Peak error was calculated as the fractional difference between the maximal spike rate and the maximal fluorescence averaged over a 200 ms time window around the peak of the fluorescence for each current step.
Odorants were presented as dilutions from saturated vapor in cleaned, humidified air using a custom olfactometer under computer control (Bozza et al., 2004; Verhagen et al., 2007). Odorants were typically presented for 4 s. All odorants were obtained at 95–99% purity from Sigma-Aldrich and stored under nitrogen. The concentration of the odorants ranged from 0.1% to 2% saturated vapor.
In vivo two photon imaging.
Animals were prepared for two-photon imaging in the same manner as described for wide-field imaging experiments, except that a craniotomy covering ∼1/2 of the dorsal surface of the OB was performed and covered with 2.5% low-melting-point agarose in Ringer's solution and a glass coverslip. Imaging was performed with a two-photon microscope (MOM; Sutter Instruments) coupled to a pulsed Ti:Sapphire laser (Mai Tai HP; Spectra-Physics) and controlled by Scanimage 3.7 (Pologruto et al., 2003). In all experiments, imaging was performed through a 20× 0.95 NA objective and emitted light was collected by multialkali photomultiplier tubes (R6357; Hamamatsu). Fluorescence videos were acquired at 7.5 Hz in most experiments. Fluorescence time series were extracted and analyzed with custom MATLAB scripts. In all cases, ΔF/F was calculated as in in vitro experiments. Pseudocolor glomerular-level activation maps reflect single trials in which videos were spatially filtered using a Gaussian window with a SD of 1.25 pixels and temporally filtered using a fourth-order Butterworth filter with a cutoff frequency of 0.25 Hz.
For acute in vivo experiments, mice were anesthetized with pentobarbital (50 mg/kg) or isoflurane (0.5–2% in O2). There was no difference observed between the response properties measured under these two anesthetics, so the data were pooled. Body temperature and heart rate were maintained at 37°C and ∼400 beats per minute. Unless otherwise stated, a double tracheotomy was performed and an artificial inhalation paradigm used (Wachowiak and Cohen, 2001; Spors et al., 2006); isoflurane was delivered directly to the tracheotomy tube, bypassing the nasal cavity. Animals were secured in a custom head holder or a stereotaxic device (Kopf Instruments) for further procedures and imaging, which followed previously established protocols (Wachowiak and Cohen, 2001; Bozza et al., 2004; Spors et al., 2006). Imaging in awake, head-fixed mice was performed using an identical optical setup that accommodated a custom restraint and behavioral training apparatus (described below). Wide-field epifluorescence signals were acquired either through thinned bone or after removal of the bone overlying the dorsal OB. The optical setup included an Olympus BX51 illumination turret, an Olympus 4× 0.28 NA objective, a filter set optimized for GFP (exciter: HQ480/40, dichroic: Q505LP, emitter: HQ535/50; Semrock), and a 470 nm LED source (Thorlabs) or 150W Xenon arc lamp (Opti-quip). Light was attenuated using neutral density filters. Optical signals were acquired at a 256 × 256 pixel resolution and a frame rate of 25 Hz (7 Hz for OMP-spH mice) and digitized at 14-bit resolution using a back-illuminated CCD camera (NeuroCCD, SM-256; RedShirtImaging), then synchronized with other experimental signals (respiration, odor control) and stored to disk using Neuroplex software (RedShirtImaging).
For experiments with different anesthesia levels (Fig. 9), animals were maintained on 0.5% isoflurane during the preparation. Recordings were started after an equilibration phase of 1 min (for 1% isoflurane) or were made immediately after adjusting the isoflurane level to 0% or 2%. Levels were returned to 0.5% immediately after data acquisition–typically after <1 min.
Awake, head-fixed preparation.
Epifluorescence imaging in awake, head-fixed mice was adapted from our previously published procedures in rats (Verhagen et al., 2007). Approximately 2 weeks after virus injection, a custom head bolt was affixed to the skull with its posterior edge at lambda using dental acrylic, a cannula guide (#C313G; Plastics One) was implanted through the nasal bone (position: 1 mm anterior of frontal/nasal fissure and 1 mm lateral) into the dorsal recess (0.5 mm depth) for subsequent monitoring of intranasal airflow, and a semichronic window for optical recordings was made by thinning the bone over the OB. All steps were performed in a single surgical procedure under ketamine/medetomidine anesthesia. Aseptic techniques were used throughout the procedure and local anesthetic (bupivacaine, 1%; Sigma-Aldrich) was applied to all incision areas. Mice received the same postsurgical treatment as for injections and were allowed to recover for 1 d before imaging.
Beginning 1–5 d after the surgical procedure, mice were acclimated to the head-fixed imaging apparatus, which consisted of a crossbar with custom female recess for securing the head bolt and a small PVC tube surrounding the mouse. An odor delivery port was positioned ∼5 mm in front of the animal's snout, and a lick spout made available for water delivery. The restraint setup was placed inside a box to limit external light and the imaging objective was extended through a port in the roof of the box. A fan at the rear of the animal removed odorants after presentation. Respiration was measured as intranasal airflow via a thermocouple (emtss-010 g-12; Omega Engineering) inserted through the cannula. Thermocouple signals were amplified (100–1000×, DP-301 differential amplifier; Warner Instruments), low-pass-filtered at 50 or 100 Hz, and digitized at 500 Hz. Odorant presentation, olfactometer control, water delivery, and data acquisition triggers were controlled with software custom written in LabVIEW (National Instruments).
For calcium imaging from ORNs, wild-type albino B6 mice were first trained in a go/no-go two-odor discrimination task for which they received water reward (Verhagen et al., 2007). After successful training, mice were anesthetized with ketamine/medetomidine and their ORNs were loaded with Calcium Green-1 dextran or Oregon Green BAPTA-1 dextran (both 10 kDa) using established protocols (Wachowiak and Cohen, 2001) in a noninvasive procedure. Collection of imaging data began 3–7 d after loading and during performance of the discrimination task. For imaging from postsynaptic neurons, mice were simply placed in the restraint apparatus with no operant conditioning and persistent limb movement or severely attenuated respiration was used as an indicator of excessive stress. In all cases, a single imaging session lasted for up to 60 min and data were collected over as many as five consecutive daily sessions. To compare optical signals during wakefulness and anesthesia (Fig. 9), head-fixed mice were briefly anesthetized with isoflurane and their responses to odorants were imaged in the 1–2 min after anesthetic was removed.
Epifluorescence data analysis.
Basic processing and analysis of optical signals followed protocols described previously for epifluorescence imaging from ORNs (Wachowiak and Cohen, 2001; Verhagen et al., 2007; Wesson et al., 2008; Carey et al., 2009). Initial data processing included, where applicable, subtraction of nonodor, “blank” trials to eliminate any photobleaching (this was minimal or nonexistent for GCaMP signals under most illumination conditions), extracting fluorescence time courses from visually selected regions of interest (ROIs), up-sampling of optical signals to 100 Hz to match the acquisition rate of respiratory signals, and offline correction of small movements in data from head-fixed mice. In anesthetized preparations, 3–8 repeated trials were averaged before analysis. For display in the figures, odorant response maps were scaled from 0 to 95% of the maximal ΔF for that map and the pixel resolution was doubled (to 512 × 512) using bilinear interpolation; epifluorescence traces were smoothed with a low-pass Gaussian kernel with a cutoff of 6–8 Hz.
Peak odorant-evoked response amplitudes were measured from ROIs consisting of 9–12 pixels centered on well defined signal foci (Verhagen et al., 2007) and computed as the average signal in 5 frames centered around the peak of the inhalation-evoked response, minus the 5-frame average before odorant onset (ΔF), then divided by the resting fluorescence measured from the first 5 frames of the trial (ΔF/F). For response maps (Fig. 7), ΔF values were calculated from temporal averages of the 80 frames (3.2 s) before odorant onset and 80 frames after the first inhalation of odorant for GCaMP signals, and of 20 frames before odorant onset and at the end of odorant presentation for spH signals. For autocorrelation analysis, these maps were demeaned and the resulting autocorrelations normalized to their peak amplitude. Autocorrelation widths were determined as the average diameter of a contour line drawn at 75% of the maximum value to generate glomerular widths that were consistent with previously reported values (∼90 μm for spH maps) (Bozza et al., 2004). To generate consensus odorant response topographies (Fig. 5), response maps were interpolated to 512 × 512 pixels, smoothed with a 6 × 6 pixel (52 μm) mean kernel, and aligned across preparations using the caudal sinus and midline of the dorsal bulbar region (Bozza et al., 2004), after which each map was thresholded at 40% maximal amplitude, further smoothed (Gaussian kernel with an SD of 12 pixels), normalized in amplitude, and responses to the same odorant averaged across preparations.
For analysis of anesthesia effects (Fig. 9), peak response amplitudes from 1–6 strongly activated foci were measured in each preparation. Measurement of signal onset latencies and rise times relative to inhalation were made as described previously for rat optical signals using a custom algorithm that fit the optical signals to a double sigmoid function (Wesson et al., 2008). The algorithm was modified slightly to fit the higher sniffing frequencies in awake mice. Onset time was defined as the time the fitted optical signal reached 10% of its maximal amplitude (t10); rise-time was defined as the time from 10–90% of maximal amplitude. Response onset latencies in awake mice were measured relative to the start of the first inhalation after odorant onset as reported by an upward deflection in the thermocouple signal. For descriptive statistics, latencies and rise times were measured from multiple foci and multiple odorants and averaged across inhalations of odorant in one session before computing a grand average across animals. For comparison of anteromedial versus caudal-lateral dynamics, latencies and rise times were measured for the first inhalation of odorant and measurements were made for 3–4 ROIs in each domain and then averaged to obtain a single latency and rise time value per domain per preparation. Relative values between the domains in each preparation were plotted and compared using a paired t test (Fig. 6). All analyses were performed using software custom written in MATLAB or LabVIEW. All statistical tests were performed with MATLAB or Origin and were two-sided, unpaired t tests unless stated otherwise. Summary data are reported as mean ± SD unless noted otherwise.
Histology and immunohistochemistry.
Mice were deeply anesthetized with an overdose of sodium pentobarbital and perfused with PBS followed by 4% paraformaldehyde in PBS. Heads were postfixed overnight at 4°C before brains were extracted and stored in PBS until they were processed further. For histological controls, brains were embedded in 5% agarose and 200 μm coronal sections in coronal orientation were obtained with a Vibratome. For immunohistochemistry, brains were cryoprotected in 30% sucrose in PBS. OBs were embedded in optimal cutting temperature compound (Tissue-Tek OCT Compound; Sakura-Fintek) and coronal slices (15–30 μm) were cut on a cryostat. Slices were first rinsed with PBS and nonspecific binding was blocked with 5% normal goat serum diluted in PBS with 0.1% Triton X-100 for 1 h at room temperature before incubation with primary antibody (rabbit anti-Tbx21,1:1000; kindly provided by Y. Yoshihara, RIKEN) or rabbit anti-TH (Bioscience Research Reagents #AB152, 1:500; Millipore) in PBS with 0.01% Triton-X overnight at 4°C. The primary antibody was detected with an Alexa Fluor 543-conjugated goat anti-rabbit secondary antibody (A-11010, 1:1000; Invitrogen) incubated for 1 h at room temperature. Intrinsic GCaMP expression was enhanced in cryosections of reporter animals by counterstaining with a FITC-conjugated GFP antibody (#ab6662, 1:150; Abcam) for 4 h at room temperature. Sections were analyzed with an Olympus IX70 FV1000 or FV10i confocal laser scanning inverted microscope at 10×, 20×, or 60× magnification. Colocalization of protein markers and the distribution of labeled cells in different OB layers was quantified by manually counting cells from confocal stacks taken from 3–5 representative sections of the dorsal OB in each animal.
Expression of GCaMP3 and GCaMP5G in defined OB neuron subpopulations
We used mouse lines expressing Cre recombinase in genetically defined neuronal populations to achieve selective expression of genetically encoded calcium indicators in three major classes of OB neurons. GAD2-Cre mice (Taniguchi et al., 2011) were used to drive expression in GABAergic interneurons expressing the 65 kDa variant of glutamic acid decarboxylase (GAD65). The GAD65+ chemotype comprises ∼60% of granule cells and 40% of all juxtaglomerular interneurons (Parrish-Aungst et al., 2007). Among the juxtaglomerular interneurons, nearly all GAD65+ neurons have processes that branch extensively within a single glomerulus and so fit the classical definition of PG cells (Kiyokage et al., 2010). TH-Cre mice were used to drive expression in a second, morphologically distinct class of juxtaglomerular interneurons that coexpress, along with TH, the 67 kDa GAD isoform (Kosaka and Kosaka, 2008; Kiyokage et al., 2010). TH+/GAD67+ interneurons innervate multiple glomeruli, are the primary source of interglomerular connections in the OB, and fit the classical description of SA cells (Kosaka and Kosaka, 2008; Kiyokage et al., 2010). Finally, PCdh21-Cre mice were used to drive expression in MT cells, which represent the primary output stream from the OB (Shipley and Adamek, 1984; Nagai et al., 2005; Mitsui et al., 2011) (Kato et al., 2012). The PCdh21 protein is a member of the nonclassical cadherins that is expressed strongly in the OB and weakly, if at all, elsewhere in the forebrain (Nagai et al., 2005).
Two variants of the circularly permuted calcium sensor GCaMP, GCaMP3 (Tian et al., 2009) and GCaMP5G (Akerboom et al., 2012), were expressed using either a Cre-dependent viral vector (Atasoy et al., 2008) or by crossing with the Cre-dependent GCaMP3 reporter line Ai38 (Zariwala et al., 2012). We first characterized the expression and basic functional properties of the GCaMP transgenes using both expression modalities, because the two methods can result in different levels of expression and, potentially, different distributions across cell types due to effects of viral serotype and developmental sequences of Cre expression. Figure 1 shows the expression of GCaMP3 in reporter crosses and after virus injection into the OB for GAD2-Cre (Fig. 1A), TH-Cre (Fig. 1B), and PCdh21-Cre mice (Fig. 1C). Viral expression patterns were variable in spatial extent, depth, and density and included a subset of the neurons expressing GCaMP3 in reporter crosses. In GAD2-Cre mice, expression in reporter crosses appeared stronger in granule cells than in PG cells, although PG cells were still apparent in fixed sections (Fig. 1A) and in vivo (see below). Expression in TH-Cre reporter crosses included, in addition to numerous juxtaglomerular neurons, smaller numbers of neurons with somata in the outer external plexiform layer (EPL; ∼10% of juxtaglomerular cell number, 211 vs 1741 cells, 3 mice) and deep EPL/superficial granule cell layers (∼15% of juxtaglomerular cell number, 290 vs 1741 cells) and in the accessory OB (data not shown), whereas virus injection led to expression almost exclusively in juxtaglomerular neurons (Fig. 1B).
In PCdh21-Cre mice, reporter-driven expression of GCaMP3 or tdTomato (using the Ai9 reporter line) led to expression in MT cells and clear expression in ORNs (Fig. 1C). In both the Ai38 and Ai9 lines, expression appeared strongest in neurons with somata in the superficial EPL (presumed tufted cells), leading to the brightest fluorescence in their lateral dendrites just beneath the glomerular layer, along with a sparse unidentified population of neurons with small somata in the deep EPL (Fig. 1C). Virus-driven expression of GCaMP3 was absent in ORNs and deep EPL neurons. We observed no expression in the granule cell layer.
The PCdh21 promoter also likely drives expression in glutamatergic neurons that remain intrinsic to the OB, such as external tufted and some superficial tufted cells (Shipley et al., 2004; Schoenfeld et al., 1985; Hayar et al., 2004; Nagai et al., 2005; Mitsui et al., 2011). Therefore, we used an anatomical approach to ensure GCaMP expression solely in PCdh21+ OB output neurons. To achieve this, virus was injected into aPC, leading to robust and widespread expression of GCaMP in mitral cells and a minority of tufted cells throughout the OB, with no expression in somata in aPC (Fig. 1D). This retrograde infection approach allowed us to exclude potential contributions of intrinsic PCdh21+ cell types and to image activity selectively from OB neurons projecting to cortex.
We further assayed cell-type specificity of expression by immunostaining for TH in GAD2-Cre and TH-Cre:GCaMP3 reporter mice and for the MT-cell-specific marker Tbx21 (Nagai et al., 2005; Mitsui et al., 2011) in GAD2-Cre and TH-Cre:GCaMP3 reporter mice and in PCdh21-Cre mice expressing GCaMPs after retrograde viral infection. In GAD2-Cre and TH-Cre reporter crosses, GCaMP3-expressing juxtaglomerular neurons were Tbx21− (GAD2-Cre: 4/508 cells, n = 3 mice; TH-Cre: 0/405 cells, n = 3 mice; Fig. 2A,B). In GAD2-Cre:GCaMP3 reporter mice stained for TH, a minority (14%) of juxtaglomerular neurons showed colabeling (91/627 cells, n = 3 mice; Fig. 2C), which is consistent with previous estimates of a small overlap between GAD65 and TH-expressing neurons (Kiyokage et al., 2010). In TH-Cre:GCaMP3 reporter crosses, 90% of GCaMP3-expressing neurons in the glomerular layer showed TH immunoreactivity (228/256 cells, n = 3 mice; Fig. 2D), whereas those in the EPL were TH− (7/181 were TH+ cells). Finally, in PCdh21-Cre mice expressing GCaMP after retrograde infection, ∼95% of GCaMP-expressing neurons were positive for Tbx21 (214/225 cells, n = 3 mice; Fig. 2E). Therefore, GAD2-Cre, TH-Cre, and PCdh21-Cre mice allow for selective expression of GCaMPs or other transgenes in distinct and largely nonoverlapping populations of OB neurons.
Characterization of GCaMPs as reporters of neural activity
Before imaging OB activity in vivo, we first characterized the relationship between neural activity and GCaMP fluorescence in vitro using simultaneous whole-cell or cell-attached recordings and two-photon Ca2+ imaging in acute slices of the OB (Fig. 3Ai). Alternatively, we used piriform cortex slices in which GCaMP5G was expressed using a Cre-independent vector as a comparison (for details, see Fig. 3 legend). As expected, peak GCaMP fluorescence increased when increasing numbers of spikes were evoked by square current steps (Fig. 3Aii,Aiii). For GCaMP5G- and GCaMP3-expressing neurons, the mean decay time constant after a spike burst was 1.19 ± 0.73 s (n = 8). The threshold number of action potentials required to evoke a suprathreshold GCaMP signal (>4 SDs above zero) was lower for GCaMP5G-expressing MT cells compared with GCaMP5G-expressing piriform cortex neurons and reporter-driven GCaMP3-expressing MT cells, with a minimum of ∼6 action potentials during a 500 ms current step (Fig. 3Aiv). In general, however, spike detection threshold depended upon the number of spikes, spike frequency, and whether spike bursts were superimposed on periods of low-frequency spontaneous spiking.
We next established a general relationship between GCaMP fluorescence and changes in spike rate using three parameters: two time constants, τon and τoff, accounting for the kinetics of increase and decrease in fluorescence after an action potential, and an exponent a to account for observed nonlinearities in the relationship between spike number and fluorescence change (Akerboom et al., 2012). Optimization of these three parameters (see Materials and Methods) allowed us to obtain excellent fits between filtered spike rates and GCaMP fluorescence changes (ΔF) both for step-evoked spike bursts (analogous to the inhalation-evoked bursts that often occur in OB neurons in vivo) and for periods of irregular spiking (Fig. 3B). Across recorded mitral cells (n = 8) in which GCaMP3 or GCaMP5G was virally expressed, the mean correlation coefficient (r2) between ΔF and spike rate after convolution of spike times with the derived time constants and linearization of the fluorescence using the derived exponent was 0.93 ± 0.02. The optimal rise and decay time constants for the GCaMP signal were 0.11 ± 0.05 and 3.09 ± 1.26 s, respectively (ranges, τon: 0.0012–0.18 s; τoff: 1.53–5.24 s). The mean value of exponent a in these cells was 1.62 ± 0.28 (range, 1.18–1.91), suggesting a supralinear relationship between spike rate and ΔF, which is consistent with a recent characterization of a related GCaMP5 variant (Akerboom et al., 2012) but different from recent characterizations of GCaMP3 (Tian et al., 2009; Borghuis et al., 2011).
The apparent supralinear nature of the spike rate–fluorescence relationship might arise from a thresholding effect in which ΔF decreases to zero at some nonzero spike rate and the resulting discontinuity is blurred by noise or spatial averaging to yield a concave curve. To test this possibility, we also derived the optimal time constants and exponent from the same time-varying spike trains but using only epochs in which the measured spike rate was >40% of the maximal value, a region that should be unaffected by threshold effects. These fits led to exponents not significantly different from unity (0.95 ± 0.18; p = 0.38, one-sample t test). Therefore, threshold effects during periods of infrequent spiking constitute one explanation for the apparent supralinear relationship between spike rate and GCaMP fluorescence observed here and described previously (Akerboom et al., 2012). Finally, in both GCaMP3- and GCaMP5G-expressing MT cells, evidence of saturation was observed only rarely except when high spike rates were elicited for prolonged periods of time (Fig. 3C). In fact, in many cells, spiking failed and depolarization block was reached before fluorescence responses saturated (data not shown).
Spike rate/ΔF fit parameters varied somewhat with expression modality and intracellular compartment. Neurons from GCaMP3-expressing reporter mice (n = 10) had modestly lower τoff constants, which is consistent with lower expression levels and with previous reports (Zariwala et al., 2012; Fig. 3D), and the spike rate/ΔF relationship appeared linear in these neurons, with an optimal exponent not significantly different from unity (1.07 ± 0.18, p = 0.26). We also found that GCaMP dynamics were faster when imaged from the dendritic versus the somatic compartment of the same neuron, with dendritic τon and τoff reduced by 85.1 ± 12.1% and 44.0 ± 13.3%, respectively (p < 0.05, paired t test, n = 4) from values derived from GCaMP signals imaged simultaneously from the soma (Fig. 3E).
Because spike rate/ΔF fit parameters are unknown for each cell during a typical imaging experiment, we also estimated the accuracy with which spike rate changes can be inferred from ΔF measurements using mean values for parameters derived from paired imaging and spike recordings. Applying the mean values reported above to all neurons, the mean correlation coefficient between linearized fluorescence and convolved spike rate was only modestly reduced, to 0.86 ± 0.07. The mean error in estimating the spike rate change after a current step was similar for optimally derived parameters versus mean parameters (0.14 ± 0.09 vs 0.13 ± 0.07, respectively). Overall, this analysis indicates that, with appropriate initial characterization, τon and τoff values, along with a nonlinearity exponent derived from a subset of paired recordings, can be used to estimate firing rate changes from a different subset of imaged neurons with reasonable accuracy over a wide dynamic range.
Two-photon imaging of spontaneous and evoked activity in identified neurons in vivo
We next evaluated the utility of GCaMPs as reporters of spontaneous and odorant-evoked activity in these three OB neuron populations in vivo. In all mouse lines, virus-driven expression of GCaMP3 led to brighter resting fluorescence than virus-driven GCaMP5G or GCaMP3 reporter crosses, which is consistent with the lower basal fluorescence of GCaMP5G (Akerboom et al., 2012) and the lower expression levels of the GCaMP3 reporter (Zariwala et al., 2012). Many neurons expressing GCaMP5G were only transiently visible during periods of elevated activity due the lower basal fluorescence level.
Spontaneous and odorant-evoked fluorescence signals were apparent in the apical tuft, lateral dendrites, and somata of PCdh21+ MT cells. In the glomerular layer, odorant-evoked GCaMP signals were confined to distinct, odorant-specific combinations of glomeruli and could show distinct temporal dynamics across glomeruli (Fig. 4Ai). In the mitral cell layer, spontaneous transients, presumably reflecting spike bursts, were observed frequently; such transients were often highly correlated between a subset of somata in the same field of view (Fig. 4Aii), which is consistent with the idea that “sister” MT cells innervating the same glomerulus are electrically coupled and synchronously active (Schoppa and Westbrook, 2001; Christie and Westbrook, 2006; Ma and Lowe, 2010). Fitting the decay phase of spontaneous transients in MT cells led to decay time constants similar to those measured from MT cells in vitro after current step-evoked spike bursts (0.95 ± 0.24 s, n = 6 cells, p = 0.24). Also in agreement with the in vitro experiments, fluorescence transients decayed with faster kinetics in mitral cell dendrites than in somata (Fig. 4Aiv). Odorant-evoked GCaMP signals in mitral cell somata, like those in dendritic tufts, were often dominated by large-amplitude transients linked to inhalation, but across the population (n = 83 cell-odorant pairs, 8 mice) were temporally diverse and included increases occurring after odorant offset (Fig. 4Aiii). Other temporal response patterns included slower ramps of activity and prolonged activity outlasting odorant presentation (Fig. 4Aii,Aiii). Odorant-evoked response patterns were specific for a soma-odorant pair and were consistent across trials. Therefore, GCaMPs expressed in PCdh21-Cre mice allow for the characterization of odorant specificity, response dynamics, and synchrony of activity across multiple individual OB output neurons.
In GAD2-Cre mice, presumptive PG cell somata in the glomerular layer were readily apparent after viral expression of GCaMP3 or GCaMP5G and were dim but discernible at high resolution in GCaMP3 reporter crosses (Fig. 4B). In TH-Cre:GCaMP3 reporter crosses, presumptive SA cell somata were easily identifiable in the glomerular layer (Fig. 4C). With GAD2-Cre and TH-Cre animals, individual glomeruli were not clearly distinguished from resting fluorescence at low magnification due to the presence of somata and interglomerular processes (Fig. 4B,C). However, odorant presentation nonetheless resulted in focal activation of discrete regions confirmed to be single glomeruli at higher magnification (Fig. 4B,C). GCaMP-expressing somata of GAD2-Cre and TH-Cre mice rarely exhibited spontaneous fluorescence transients, but responded robustly and specifically to odorant presentation (Fig. 4B,C). In contrast to MT cell responses, however, PG cell response dynamics (n = 138 cell-odor pairs, 3 mice) were simple, consisting of a transient increase limited to the duration of odorant presentation (Fig. 4B). GCaMP signals imaged from TH+ SA cell somata showed moderate temporal complexity, with all responsive cells (n = 97 cell-odor pairs, four mice) increasing fluorescence upon odorant presentation, but with cell cohorts in some fields of view (presumably associated with the same glomerulus) showing responses far outlasting the stimulus (Fig. 4Ciii). Therefore, these two distinct juxtaglomerular interneuron populations show distinct properties of responsiveness to odorant stimulation.
Topography of odorant representations among OB neuron subtypes
We next characterized population-level patterns of odorant-evoked activity among GAD2+ PG cells, TH+ SA cells, and PCdh21+ MT cells using low-magnification epifluorescence imaging across the dorsal OB. Peak odorant-evoked GCaMP signals ranged from 10% to 185% ΔF/F with epifluorescence; fluorescent changes were highest with GCaMP5G, which is consistent with the recent characterization of GCaMP5 variants (Akerboom et al., 2012). At the level of sensory input, odorants are represented as spatially organized patterns of input to OB glomeruli that have been well characterized using several presynaptic imaging methods (Wachowiak and Cohen, 2001; Bozza et al., 2004; Soucy et al., 2009; Ma et al., 2012). To visualize how these representations translate across different populations of OB neurons, we constructed spatial response maps of GCaMP signals evoked by a panel of odorants and compared these with patterns of ORN input using spH expressed in ORNs (Bozza et al., 2004; Fig. 5A). Response maps were imaged in GAD2-Cre and TH-Cre mouse lines using GCaMP3 reporter expression to avoid confounds of heterogeneous expression and potential tissue damage arising from local virus injection. Retrograde viral expression was used in PCdh21-Cre mice to selectively image activation of MT cells projecting to piriform cortex.
SpH-derived response maps reflected ORN input to glomeruli located in distinct domains of the dorsal OB that matched those previously characterized for organic acids, ketones, esters, and amines (Bozza et al., 2004; Takahashi et al., 2004; Bozza et al., 2009; Pacifico et al., 2012; Fig. 5A), and increasing odorant concentration recruited input to larger numbers of glomeruli and evoked larger-amplitude peak responses (Fig. 5A). Response maps imaged from GAD2-Cre mice and TH-Cre mice also consisted primarily of discrete glomerular foci and only a minor diffuse signal and showed a similar concentration-dependent recruitment of glomerular number and response amplitude (Fig. 5A). Maps of PCdh21+ MT cells were noticeably more diffuse, with widespread GCaMP signals becoming more prominent at suprathreshold concentrations (Fig. 5A). In all three cell types, odorants evoked responses with similar topography across the dorsal OB, with peak responses in domains that were highly overlapping for each odorant (Fig. 5B).
Spatially diffuse GCaMP signals may reflect the lateral spread of activity from the sites of glomerular ORN input. For example, the extensive interglomerular branching of TH+ SA cells and the extensive lateral spread of MT lateral dendrites in the EPL are both hypothesized to mediate lateral inhibition within the OB (Yokoi et al., 1995; Aungst et al., 2003). Conversely, GAD2+ PG cells are largely uniglomerular. To compare the degree to which these cell types might transmit (or receive) excitatory inputs laterally, we measured the spatial contrast of epifluorescence GCaMP signals mapped across the dorsal OB for the same set of odorants tested in TH-Cre, GAD2-Cre, and PCdh21-Cre mice using autocorrelation analysis (see Materials and Methods). For comparison and to control for diffuse signals arising from scattered and out of focus light, we also analyzed spH response maps in which activated ORN axons remain confined to a single glomerulus. At low odorant concentrations that typically evoke input to only a few glomeruli for a given odorant, response maps from GAD2+ and TH+ GCaMP signals showed little evidence of lateral signal spread, with autocorrelation widths that were not significantly different from those of spH response maps (spH: 88.5 ± 22.1 μm, n = 18 maps, 9 bulbs; GAD2: 77.6 ± 26.7 μm, p = 0.98, n = 21 maps, 5 bulbs; TH: 71.4 ± 43.4 μm, p = 0.82, n = 26 maps, 9 bulbs; p > 0.8 for both comparisons, one-way ANOVA and Tukey-Kramer honestly significant difference criterion). MT cell (PCdh21+)-derived response maps were significantly more diffuse (177.3 ± 50.2, n = 32 mice, 11 bulbs, p < 0.001). At suprathreshold concentrations of odorants that evoke stronger input to numerous glomeruli (e.g., ethyl butyrate; Fig. 5), autocorrelation widths were larger in all cases (GAD2: 112.7 ± 43.2 μm, n = 15 maps, 11 bulbs; TH: 125.5 ± 25.5 μm, n = 7 maps, 7 bulbs; PCdh21: 222.05 ± 51.2 μm, n = 13 maps, 13 bulbs), but, again, only MT cell (PCdh21+)-derived response maps displayed significantly larger autocorrelation widths compared with those of spH response maps (n = 18 maps, 18 bulbs; for comparison between GAD2, TH, and Pcdh21 versus spH: p = 0.62, p = 0.40, and p < 0.001, respectively, one-way ANOVA). The most likely source of the stronger diffuse GCaMP signals in PCdh21+ animals is action potential-mediated calcium influx into the lateral dendrites of tufted cells (Xiong and Chen, 2002; Debarbieux et al., 2003), which form an extensive network just below the glomerular layer (Fig. 1C, Fig. 3A). This conclusion is consistent with the observation that GCaMP signals mapped with two-photon imaging from the glomerular layer remained confined to discrete glomeruli (Fig. 4A).
Inhalation-linked dynamics of odor representations across cell type
Because odor information is also encoded in the temporal patterns of activity relative to respiration (Chaput, 1986; Spors et al., 2006; Cury and Uchida, 2010; Shusterman et al., 2011), we evaluated the degree to which GCaMPs can report respiration-linked dynamics among the different OB neuron populations and compared these dynamics against those reported by synthetic Ca2+ indicators loaded into ORNs (see Materials and Methods). Synthetic indicators loaded into ORN axon terminals can report differential onset dynamics for odorant-glomerulus pairs of ∼100 ms (Spors et al., 2006; Fig. 6A). Inhalation-locked modulation was prominent in GCaMP signals imaged in GAD2-Cre, TH-Cre, and PCdh21-Cre mice (Fig. 6B–D). Response latencies relative to inhalation onset were similar for GCaMP signals originating in all three cell types and for presynaptic Ca2+ signals measured from ORNs: ORNs, 180 ± 31 ms (n = 5 mice, 183 responses); GAD2+, 172 ± 42 ms (n = 9 mice, 958 responses); TH+, 195 ± 28 ms (n = 3 mice; 320 responses); PCdh21+, 202 ± 66 ms (n = 15 mice, 1096 responses).
Odorant- and glomerulus-specific differences in the dynamics of the inhalation-evoked GCaMP signal were apparent in all three cell types (Fig. 6B–D) and mimicked those measured from ORN terminals (Fig. 6A) in terms of both the magnitude and the topography of latency differences. For example, in GAD2-Cre, TH-Cre, and PCdh21-Cre mice, the latency and rise times of the GCaMP signal evoked by ethyl butyrate were significantly faster for caudal-lateral glomeruli than for anteromedial glomeruli and showed a greater modulation between inhalations (Fig. 6E), a result identical to that reported with presynaptic imaging (Spors et al., 2006). Therefore, despite their slower dynamics relative to synthetic Ca2+ indicators (Akerboom et al., 2012), GCaMP3 and GCaMP5G are able to report relative timing differences in odorant-evoked activity on a scale thought to be relevant for coding odor information (Cury and Uchida, 2010; Shusterman et al., 2011). In addition, the close correspondence in these dynamics between presynaptic and three different sources of postsynaptic activity suggest that the intrinsic and inhalation-driven dynamics of sensory input persist through multiple stages of synaptic processing in the OB.
Modulation of OB neuron responsiveness by wakefulness and anesthesia
Wakefulness can dramatically alter response properties and excitability of OB neurons compared with anesthetized states (Rinberg et al., 2006; Vincis et al., 2012), and a recent report using GCaMP imaging demonstrated that MT and granule cells are differentially modulated by wakefulness (Kato et al., 2012). To address how odor representations among MT cells and additional cell types are modulated by wakefulness at the population level, we imaged odorant-evoked activity among GAD2+ PG cells, TH+ SA cells, and PCdh21+MT cells using epifluorescence imaging in the awake, head-fixed mouse. For comparison, and to evaluate for the first time the temporal structure of sensory input to the OB of the awake mouse, we also performed imaging from ORNs loaded with a synthetic Ca2+ indicator.
Imaging presynaptic Ca2+ signals from ORNs in wild-type mice revealed robust (up to 15% ΔF/F) odorant-evoked calcium signals that were transient and tightly coupled to inhalation, with onset latencies of 153 ± 61 ms (n = 6 mice, 158 sniffs; Fig. 7A), similar to presynaptic ORN signals imaged in head-fixed rats (Verhagen et al., 2007). Presynaptic calcium signals evoked by inhalation in the absence of odorant could not be detected. In both GAD2-Cre and TH-Cre mice expressing virally driven GCaMP5G or reporter-driven GCaMP3, odorants evoked strong fluorescence increases (15–60% ΔF/F) that were also tightly coupled to inhalation. Onset latencies were similar to those imaged from ORNs (GAD2+, 161 ± 59 ms, n = 3 mice, 481 sniffs; TH+, 144 ± 15 ms, n = 4 mice, 615 sniffs). In both mouse lines, responses were reliable across repeated odorant presentations (Fig. 7B,C). Fluorescence decreases in response to odorant were never observed. In both lines, inhalation-evoked signals in anteromedial regions showed less modulation by the respiratory cycle, slower decay during the exhalation phase, and more summation across successive inhalations of odorant (Fig. 7B,C), recapitulating temporal response patterns seen under anesthesia (Fig. 6). In most regions, respiratory modulation was apparent at breathing rates of 3–4 Hz typical of quiescent mice. Both GAD2+ and TH+ GCaMP5G signals also showed respiratory modulation in the absence of odorant, consisting of small inhalation-driven responses apparent in a few regions of the OB (Fig. 7B). More striking, however, were larger fluorescence increases evoked by single large-amplitude inhalations, as well as longer-lasting increases linked to occasional bouts of high-frequency sniffing (Fig. 7B,C).
In contrast, in awake PCdh21-Cre mice expressing GCaMP3 (n = 2) or GCaMP5G (n = 2) in MT cells, the fluorescence response to odorant differed dramatically from that observed in anesthetized mice, consisting predominately of decreases in fluorescence of up to 12% ΔF/F, superimposed with transient, small-amplitude increases (max of 3% ΔF/F; Fig. 7D). Responses could consist of multiple phases of increase or decrease and often included distinct responses to odorant offset, all of which were reliably repeated across presentations (Fig. 7D, Fig. 8). Respiratory modulation in the absence of odorant was observed occasionally and, similar to GAD2+ and TH+ GCaMP5G signals, bursts of high-frequency sniffing elicited widespread fluorescence increases (data not shown). Although different odorants evoked distinct spatiotemporal patterns of fluorescence changes across the dorsal OB, in general, fluorescence increases were localized to restricted foci and decreases were spatially widespread (Fig. 8). In PCdh21-Cre mice expressing virally driven GCaMP3 in MT cells (n = 2 animals), odorant-evoked responses consisted entirely of widespread fluorescence decreases, with no detectable increases (n = 5 odorants tested, 10 sessions).
To compare differences between response patterns in awake versus anesthetized states directly, we briefly (1–3 min) anesthetized head-fixed mice with isoflurane and then imaged GCaMP signals immediately after removal of isoflurane and as the animal returned to the waking state over the next 1–2 min. In GAD2-Cre and TH-Cre mice, isoflurane-induced anesthesia led to a substantial decrease in the amplitude of odorant-evoked responses (Fig. 9A,B), with GAD2+ responses and TH+ responses being reduced to 45 ± 23% (n = 3 mice) and 27 ± 17% (n = 4 mice), respectively, of initial peak amplitudes. Isoflurane also led to a decrease in resting fluorescence across the dorsal OB to 78 ± 7% (GAD2+) and 80 ± 2% (TH+) of initial values. In PCdh21+ MT cells, the effect of short-term isoflurane anesthesia also reduced resting fluorescence to 77 ± 3% of awake levels (n = 4 mice), but its effect on odorant-evoked responses was more complex, consisting primarily of a reduction or elimination of the widespread fluorescence decreases apparent in the awake state, along with an emergence of slower fluorescence increases in focal locations (Fig. 9C). In all lines, both odorant-evoked and resting fluorescence returned to preanesthesia levels within 1–3 min after removal of isoflurane.
One difficulty in comparing sensory-evoked signals between awake and anesthetized animals is that the strength and frequency of inhalation also varies substantially between states. Therefore, in a separate set of animals, we compared responses at different depths of isoflurane anesthesia while using artificial inhalation to sample odorants consistently. Anesthesia depth affected response magnitude significantly in all three cell types (one-way ANOVA, GAD2-Cre, F(2,87) = 135.3, p ≪ 0.001; TH-Cre, F(2,57) = 86.9; p ≪ 0.001, PCdh21-Cre, F(2,45) = 39.8, p ≪ 0.001). Interestingly, the direction of the effect of anesthesia differed among cell types, with decreasing anesthesia depth leading to increased response amplitudes in GAD2+ and TH+ neurons (Fig. 9D,E) and decreased responses in PCdh21+ MT cells (Fig. 9F). Resting fluorescence was also modestly dependent on anesthesia depth, with fluorescence under 2% isoflurane at 85 ± 8% (GAD2+, n = 4 mice), 90 ± 4% (TH+, n = 5 mice), and 88 ± 6% (Pcdh21+, n = 4 mice) of the levels seen under 0.5% isoflurane. Therefore, anesthesia alters the responsiveness of OB neurons in a manner that is distinct for different cell types and is independent of its effects on respiration.
In the present study, we used optical reporters of neural activity and genetic tools for cell-type-specific expression to begin investigating early olfactory processing at the single-neuron and population level in anesthetized and awake mice. We compared the responses of genetically specified classes of OB neurons, including MT cells that project to the olfactory cortex and, for the first time to our knowledge, PG and SA cells that mediate intraglomerular and interglomerular processing, with those of sensory afferents. We also demonstrate our ability to achieve combinatorial selectivity of expression based on both genetic markers and axonal projection target using readily available recombinant AAV-type viral vectors.
We characterized the fidelity of GCaMP3 and GCaMP5G as reporters of spiking activity in OB neurons, adding to previous characterizations of these reporters in other brain areas (Tian et al., 2009; Borghuis et al., 2011; Yamada et al., 2011; Akerboom et al., 2012). Although not tested here, those studies reported no significant impact of GCaMP expression on the electrophysiological or evoked response properties of GCaMP-expressing neurons using either the long-term reporter expression or short-term viral expression modes. GCaMP fluorescence changes could be related quantitatively to changes in spike rate by accounting for the dynamics of the GCaMP reporter and for any nonlinearity in the relationship between spike frequency and fluorescence. In a given neuron, this relationship is robust, allowing for accurate inference of spiking activity over a relatively large dynamic range, a feature that is useful for detecting and quantifying changes in activity under different conditions, for example, due to different stimuli, activation of modulatory centers, or changes in behavioral state. A caveat is that the time constants and linearity of the fluorescence/spiking relationship can vary for the same neuron type depending on expression modality and can vary for the same reporter expressed in different neuron types. Therefore, it is important to characterize these parameters in a given population and with a given expression modality.
GCaMP signals could be measured reliably from dendritic processes and from cell bodies extending from the glomerular layer to as deep as the mitral cell layer in vivo. Simultaneous imaging from multiple somata revealed that, among MT cells, spontaneous fluorescence transients occurred synchronously among a subset of cells, which is consistent with sister mitral cells innervating the same glomerulus (Christie and Westbrook, 2006; Dhawale et al., 2010). In response to odorant stimulation, individual MT cells showed distinct odorant specificities and diverse temporal patterns of activation, which is consistent with a recent report using the same GCaMP imaging approach (Kato et al., 2012). In contrast, GAD2+ PG cells and TH+ SA cells, while odorant specific, showed little spontaneous activity and substantially less temporal diversity. The relatively simple response dynamics of PG and SA cells appeared to be distinct from those reported for granule cells, which show a large range of response latencies when imaged using GCaMP3 (Kato et al., 2012). These observations highlight the diversity of response properties among different OB subtypes and highlight the importance of recording from identified circuit elements to understand how sensory inputs are transformed within the OB in vivo.
Topography and dynamics of odor representations across cell type
At the population level, patterns of odorant-evoked calcium signals were similar for ORN inputs, GAD2+ PG cells, TH+ SA cells, and PCdh21+ MT cells. For all cell types, odorant response maps included foci located in stereotyped topographic domains (Uchida et al., 2000; Takahashi et al., 2004; Bozza et al., 2009; Pacifico et al., 2012), and increasing the odorant concentration increased response amplitudes and recruited more widespread signals. In addition, small temporal differences in the inhalation-linked dynamics of ORN inputs to topographically distinct glomeruli were also apparent in GCaMP signals imaged from each cell type.
One explanation for these similarities is that GCaMP signals imaged from the glomerular layer mainly report synaptic input from ORNs rather than output from postsynaptic neurons. This explanation seems unlikely for PG cells given their compact, uniglomerular morphology (Kiyokage et al., 2010) and given that GCaMP signals imaged directly from PG cell somata showed similar amplitude and dynamics as those imaged from the glomerular neuropil. Calcium signals in the apical tuft of MT cells are more likely to diverge from patterns of spike output. However, action potentials in MT cells back-propagate and drive Ca2+ influx into the apical tuft in vivo (Charpak et al., 2001; Debarbieux et al., 2003), and in vitro experiments suggest that synaptically driven Ca2+ influx generates significantly smaller Ca2+ signals in the apical tuft than do action potentials (Zhou et al., 2006), indicating that apical calcium signals at least partially reflect MT spiking. Finally, PG and MT cells transmit information within the glomerular neuropil via dendrodendritic synapses and, therefore, Ca2+ signals imaged from these neurons are likely reliable indicators of local output affecting glomerular processing (Murphy et al., 2005).
The substantial GCaMP signals arising from MT cell lateral dendrites also implies that such signals should have been apparent from SA cells, because their somata and extensive interglomerular projections remain confined to the glomerular layer (Kiyokage et al., 2010). It is therefore notable that GCaMP signals imaged from SA cells showed significantly less spread across the dorsal OB than did those imaged from MT cells, because SA cell processes and MT cell lateral dendrites have similarly sparse branching patterns and SA processes are more superficial. Potential explanations for this difference include differences in calcium channel expression or action potential propagation along SA cell processes versus MT cell lateral dendrites. Indeed, whether the extensive interglomerular branching of SA cells consists of axonal or dendritic processes remains unclear (Kiyokage et al., 2010). Direct visualization of lateral signal propagation along SA versus MT cell processes using two-photon imaging may be necessary to resolve this question.
The circuit elements of the OB network imaged here have been proposed to play multiple roles in transforming afferent odor representations, including sparsening representations through intraglomerular or interglomerular inhibition, concentration-invariant normalization of OB output patterns, and decorrelating sensory inputs and OB outputs (Friedrich and Laurent, 2001; Cleland and Sethupathy, 2006; Cleland et al., 2007; Yaksi et al., 2007; Cleland et al., 2012). Two-photon imaging of individual MT cells did reveal evidence for a decorrelation between ORN inputs and OB outputs, because some cells displayed temporal response patterns that were distinct from ORN input patterns and from neighboring cells. In addition, population-level MT responses in the awake mouse showed sparse excitation and temporal complexity not apparent in sensory inputs or in PG or SA cell responses. Further work will be necessary to understand how such transformations emerge from the OB network, for example, by simultaneously imaging activity in different neuron subtypes using spectrally distinct GCaMP variants (Zhao et al., 2011; Ohkura et al., 2012).
Effects of anesthesia and wakefulness on OB neuron responses
Our data expand on a recent report showing that anesthesia strongly and differentially modulates the responsiveness of MT and granule cells (Kato et al., 2012). In the present study, we confirmed that wakefulness leads to a reduction in excitatory MT cell responses and also found that PG and SA cell responses are enhanced during wakefulness, an effect similar to that seen for granule cells (Kato et al., 2012) and one consistent with the hypothesis that the strength of glomerular inhibition is an important regulator of MT cell response properties (Gire and Schoppa, 2009; Shao et al., 2012). We also found that this modulation is rapid, occurring within the 1–2 min necessary for the removal of brief isoflurane anesthesia, and that it is independent of changes in inhalation strength. Therefore, three major classes of GABAergic OB interneurons, PG cells, SA cells, and granule cells (Kato et al., 2012), show enhanced responsiveness in the waking versus the anesthetized state. The source of modulation of PG and SA cell excitability during wakefulness may be distinct from that affecting granule cell excitability, because a major source of granule cell modulation, centrifugal inputs from olfactory cortex (Balu et al., 2007; Gao and Strowbridge, 2009), does not strongly innervate the glomerular layer. Therefore, multiple circuit mechanisms likely underlie the reduced MT cell responsiveness during wakefulness compared with anesthesia. Similarly, distinct circuit mechanisms likely underlie the modulation of MT response properties during different behavioral states or as a function of experience (Kato et al., 2012).
We also found that odorant-evoked MT population responses in the awake mouse consisted of sparse and transient increases and, notably, widespread decreases in fluorescence. We interpret the fluorescence decreases that dominated awake MT response maps as reflecting inhibition of ongoing MT activity. The ability to report both excitation and inhibition was dependent on the choice of reporter: GCaMP3-expressing MT cells showed only decreases in the awake mouse (but showed increases under anesthesia), whereas GCaMP5G-expressing neurons exhibited responses in both directions. The predominance of inhibition in MT GCaMP signals during wakefulness suggests that inhibitory pathways in the OB are critical in shaping OB output patterns. The ability to optically map this inhibition among distinct cell types in vivo will be useful in further understanding the functional organization of processing within the OB.
This work was supported by the National Institutes of Health (Grants DC06441 and 2DC012718 to M.W.; Grant 1F32DC012718-01 to M.N.E.) and the Deutsche Forschungsgemeinschaft (to M.R.). We thank C. Zabawa for technical support and Drs. Looger, Akerboom, and Kim and the Genetically Encoded Calcium Indicator Project at Janelia Farm Research Campus in collaboration with Penn Vector Core for providing GCaMP-expressing viruses.
The authors declare no competing financial interests.
- Correspondence should be addressed to Matt Wachowiak, Brain Institute, Sorenson Molecular Biotechnology Building, 36 S. Wasatch Drive, University of Utah, Salt Lake City, UT 84112.