We measured the spatiotemporal aspects of the odor-induced population response in the turtle olfactory bulb using a voltage-sensitive dye, RH414, and a 464-element photodiode array. In contrast with previous studies of population activity using local field potential recordings, we distinguished four signals in the response. The one called DC covered almost the entire area of the olfactory bulb; in addition, three oscillations, named rostral, middle, and caudal according to their locations, occurred over broad regions of the bulb. In a typical odor-induced response, the DC signal appeared almost immediately after the start of the stimulus, followed by the middle oscillation, the rostral oscillation, and last, the caudal oscillation. The initial frequencies of the three oscillations were 14.1, 13.0, and 6.6 Hz, respectively. When the rostral and caudal oscillations occurred together, their frequencies differed by a factor of 1.99 ± 0.01.
The following evidence suggests that the four signals are functionally independent: (1) in different animals some signals could be easily detected whereas others were undetectable; (2) the four signals had different latencies and frequencies; (3) the signals occurred in different locations and propagated in different directions; (4) the signals responded differently to changes in odor concentration; (5) the signals had different shapes; and (6) the rostral and caudal signals added in a simple, linear manner in regions where the location of the two signals overlapped. However, the finding that the frequency of the rostral oscillation is precisely two times that of the caudal oscillation suggests a significant relationship between the two.
The location of the caudal oscillation in the bulb changed from cycle to cycle, implying that different groups of neurons are active in different cycles. This result is consistent with the earlier findings in the olfactory system of the locust (Wehr and Laurent, 1996).
Our results suggest an additional complexity of parallel processing of olfactory input by multiple functional population domains.
Odor stimuli have long been known to induce stereotyped local field-potential (LFP) responses in the olfactory bulb consisting of sinusoidal oscillations of 10–80 Hz riding on top of a slow “DC” signal. Since their first discovery in the hedgehog (Adrian, 1942), odor-induced oscillations have been observed across phylogenetically distant species, including locust, (Laurent and Naraghi, 1994), frog (Ottoson, 1959b), turtle (Beuerman, 1975, 1977), rabbit (Adrian, 1950), and monkey (Hughes and Mazurowski, 1962).
Despite its ubiquity in many brain regions, the roles and functions of oscillation are not well understood. In the olfactory system, odor-induced oscillations have been hypothesized (1) to be the consequence of mechanical stimulation of the olfactory receptors (Adrian 1942), (2) to encode and distinguish odors by different oscillation modes (Li and Hopfield, 1989), (3) to provide a mechanism for the brain to encode the intensity of the sensory stimuli (Hopfield, 1995), (4) to reflect “preferred states of activity” for reinforced odors (Freeman and Grajski, 1987), (5) to allow representation of stimuli by the firing sequences of neural assemblies (Wehr and Laurent, 1996), and (6) to be important in fine odor discriminations (Stopfer et al., 1997).
We examined the spatial and temporal properties of odor-induced oscillations in the hope that this will facilitate the understanding of their functional roles. Earlier studies were conducted using a 64-element local field potential electrode array (Freeman, 1978). Because the current sources driving these potentials can be below the surface, the spatial patterns will be smoothed by the volume-conductor properties of the cortical tissue. Freeman (1978) reported that spatial Fourier transforms had a sharp cutoff at approximately one cycle per millimeter, suggesting spatial resolution on the order of 1 mm [also see Bullock and McClune (1989)]. In contrast, the resolution of an optical recording using voltage-sensitive dyes appears to be substantially better. The optical resolution is limited to ∼200 μm by light scattering and by signals from layers that are out of focus (Orbach and Cohen, 1983). Preliminary data directly comparing voltage-sensitive dye signals and local field potential measurements from the surface of the turtle visual cortex also suggested that the linear spatial resolution of the optical recording was five times better (two-dimensional resolution 25 times better) (J. Prechtl, D. Kleinfeld, and L. B. Cohen, unpublished results) (Wu et al., 1998;Lam et al., 2000).
The olfactory bulb was stained by placing a solution of the voltage-sensitive dye on the bulb for 60 min. We think that this procedure stains all of the membranes in the bulb. At the magnification that we used, each detector in the photodiode array received light from a “column” of the bulb, where the x and ydimensions in the object plane are 170 × 170 μm2. Thus, the signals represent the population average of the change in membrane potential in the many neurons (thousands) and processes whose light falls on each detector.
The odor-induced oscillations revealed by our measurements were somewhat complex. We were able, however, to divide the response into one DC signal and three different oscillatory signals. We have begun a characterization of the three oscillatory signals using the following parameters: location, frequency, latency, signal shape, propagation direction, concentration dependence, and initiation site. Because odor stimuli are also known to elicit field potential oscillations in the olfactory epithelium (Ottoson 1959b; Takagi and Shibuya, 1960), the electro-olfactogram was measured simultaneously in a few experiments for comparison with the oscillations in the bulb.
All of the figures are presented with rostral to the right and lateral up.
MATERIALS AND METHODS
Animals. Two species of box turtle, Terepene carolina and T. ornata, were used. We detected an optical response to odors in 37 preparations; 17 were performed usingT. carolina and 20 used T. ornata. Because no species difference was detected, the results from both species are considered together. The number of trials in which we detected optical signals in response to odor presentation ranged from 4 to 52 in the 37 preparations. The animals were obtained from Charles D. Sullivan Co. (Nashville, TN) and were kept at 10°C. They were warmed to room temperature and watered twice a week. The animals weighed between 280 and 800 gm.
Turtle saline. The composition of the turtle saline was modified from Prechtl et al. (1997) and contained (in mm): NaCl 96.5, KCl 2.6, MgCl2 2.0, NaHCO3 31.5, CaCl2 4.0, dextrose 10. These chemicals were obtained from Sigma (St. Louis, MO). The saline was bubbled with 5% O2/95% CO2, resulting in a pH of 7.0–7.2.
Dissection. The turtles were first anesthetized by placing them in ice for 2 hr. Lidocaine (0.4–0.6 ml, 1% w/v solution in saline) was then applied under the skin around the craniotomy site as a local anesthetic. Tubocurarine (3 mg/kg) was injected into the intraperitoneal cavity to partially paralyze the animals throughout the surgery and the experiment. A craniotomy was performed over the olfactory bulb. The dura and arachnoid matter were then carefully removed to facilitate staining. A segment of polyethylene tubing (outer diameter, 2 mm; inner diameter, 1 mm) was inserted into the outlet of the nasal cavity in the roof of the mouth and fixed in place by Krazy Glue and epoxy. The preparation was then allowed to warm to room temperature; during this time the bulb was stained by placing a solution of the dye on the bulb for 60 min. To reduce movement artifacts during the optical recording, the animals were partially paralyzed using curare and restrained by clamping the tip of the nose. Most of the measurements were performed on this unanesthetized preparation. To test the effect of a general anesthetic on the signals, urethane (1.5 gm/kg) was injected into the intraperitoneal cavity in 11 animals after an initial set of optical recordings. After a wait of 30 min, we tested for odor-induced responses. In seven of these animals, the odor-induced responses did not appear to be significantly affected by the urethane. In the remaining four, optical responses were no longer detected. We do not know whether the loss of response in these four animals was caused by the anesthetic, rundown of the preparation, or other causes. The individual figure legends indicate whether urethane was used. The experimental protocol was approved by the Yale Animal Care and Use Committee and the Marine Biological Laboratory Institutional Animal Care and Use Committee.
Odor delivery. The design of the odor delivery system (olfactometer) was copied from Kauer and Moulton (1974) with minor modifications (Fig.1A). Cleaned and desiccated carrier gas, air with 1% CO2, and laboratory air saturated with odorant vapor were injected into and mixed in the inner tube of a double-barrel odor applicator. The flow rate of the air–CO2 was controlled by a flow meter and fixed at 300 ml/min. The flow rate of the odor vapor was adjusted by the speed of the syringe pump to give the desired final concentrations of the odorant. The outer tube of the applicator (Fig.1A) was normally under suction (1500 ml/min) to remove the odorant and keep it from reaching the nose. At a command pulse, this suction was turned off to release a square-pulse of odor.
In addition, suction (300 ml/min) on the segment of tubing inserted into the nasal outlet was controlled by a separate solenoid valve. This suction was normally off and was switched on 1–2 sec before the odor pulse. This ensured that the odor would be drawn into the nasal cavity and also allowed us to test for optical signals elicited by room air. The nasal outlet suction was maintained for 15–20 sec after the end of the trial to remove remaining odorant from the nasal cavity. Because a response to air was rare and not seen in the experiments presented in Figures 2-4, 6-8, and 10-12, these figures show only a brief interval before the odor response. Most of the components of the olfactometer were fabricated from Teflon or glass to reduce cross-contamination between odors.
Output of the odor from the applicator was monitored by measuring the CO2 in the carrier gas with a CO2 detector (Beckman Medical Gas Analyzer, LB-2; Beckman, Schiller Park, IL) (Fig. 1 A) (Kauer and Shepherd, 1975). Figure 1 B shows an example of such a measurement. In this measurement, the length of the tubing from the mouth of the odor delivery device to the probe of the detector was 10 cm, the inner diameter was 1 mm, and the flow rate was 300 ml/min. The top trace in Figure 1 B represents the command pulse sent to the solenoid pump controlling odor delivery; thebottom trace is the CO2 level detected by the gas analyzer. The odor output from the olfactometer was approximately square-shaped and had a latency of ∼100 msec from the onset of the command pulse to the solenoid controlling odor delivery. The horizontal bars indicating odor application in Figures2-4, 6-8, and 10-12 show the timing of this command pulse. The CO2 concentration in the outflow from the nose of the animals was monitored during the experiments to ascertain that odorant had passed through the nasal cavity.
In most experiments the odorant was delivered to a single naris, and the optical recordings were made from the ipsilateral bulb. Recordings from the contralateral bulb are shown in Figure 12.
Odorant. The odorant most often used was cineole. Isoamyl acetate, butyric acid, and pyridine were used in a few experiments. The final concentrations of odorant are indicated in the figures and ranged from 0.17 to 15% of saturation. All odorants were purchased from Sigma.
Dye staining. The exposed olfactory bulb was stained by covering it with dye solution for 60 min. Excess dye was then washed away with turtle saline. During the experiment, the brain was kept moist by washing with saline between trials.
In initial experiments, several voltage-sensitive dyes were screened on an in vitro preparation. The olfactory bulb was removed and stimulated by electric shocks to the nerve (no oscillations were detected in response to the shocks). The optical response to the shock consists of a fast peak and a slow signal (Orbach and Cohen, 1983). Both the signal size and penetration of the dye into the bulb were measured for each dye; the results are shown in Table1. RH414, 0.01–0.2 mg/ml in saline (Grinvald et al., 1994) (T-1111; Molecular Probes, Eugene, OR) (Fig. 1 D) penetrated throughout the thickness of the bulb and exhibited a relatively large signal. It was used in all of the experiments reported here. The dye staining appeared to be uniform in the different layers of the bulb, suggesting that the dye stains all cell types approximately equally.
Optical imaging. One difficulty of a voltage-sensitive dye measurement is the small signal size. The signals reported here had peak sizes between 10−4 and 2 ×· 10−3of the resting fluorescence. To measure signals this small we optimized the optics (Fig. 1C) for the measurements at low magnification. Because the fluorescence intensity in an epifluorescence measurement is proportional to the fourth power of the objective numerical aperture (NA) (Inoue, 1986) and conventional microscope optics have small numerical apertures at low magnifications, we assembled a 4× microscope based on a 25 mm focal length, 0.95 f, C-mount, camera lens (used with the C-mount end facing the preparation) (Salama, 1988; Ratzlaff and Grinvald, 1991; Kleinfeld and Delaney, 1996). With a magnification of 4×, the intensity reaching the photodetector was 100 times larger with this lens than with a conventional 4×, 0.16 NA microscope lens.
Fluorescence was measured and analyzed using NeuroPlex, a 464-element photodiode array system (RedShirtImaging, LLC, Fairfield, CT). The preparation was illuminated using a 100 W tungsten–halogen lamp. The excitation filter was 520 ± 45 nm. A 590 nm long-pass dichroic mirror (Omega Optical, Brattleboro, VT) was used to reflect the excitation light onto the preparation. The secondary filter was an RG610 long-pass filter (Schott Optical Glass, Duryea, PA). The signal from each of the 464 photodiodes was amplified by an individual amplifier. The cutoffs of the single-pole RC high-pass filter and the low-pass four-pole switched-capacitance Bessel filter in each amplifier were set to 0.07 and 125 Hz, respectively. We recorded the data at a frame rate of 250 Hz. Additional details of the apparatus are given inWu and Cohen (1993) and Wu et al. (1998). All of the measurements were performed at room temperature (20–25°C).
In a few experiments, high-resolution images of the fluorescence of the preparations were recorded with a CCD camera (Dage MTI RC300, Michigan City, IN). The images of a calibration pattern were recorded using both the photodiode array and the CCD camera to allow alignment of the images from the two systems.
Local field potential recording from the olfactory epithelium and the bulb. In a few preliminary experiments, local field-potential recordings from the bulb or the olfactory epithelium [electro-olfactogram (EOG)] were performed together with the optical recordings. For the bulb measurements, a small area of the pia matter was carefully removed with a pair of fine tweezers, and a glass microelectrode filled with 1 m NaCl (2–10 MΩ resistance) was inserted into the bulb with a micromanipulator. The optimal locations and depths for the electrode recording were found by trial and error. Figure 2 illustrates a simultaneous local field potential and voltage-sensitive dye recording from the bulb.
The electro-olfactogram was recorded by a electrode placed on the surface of the olfactory epithelium. The electrode consisted of a Teflon-coated silver wire (0.25 mm diameter) stripped at the end of the coating. The signal was compared with a ground electrode placed in the mouth of the turtle, amplified, and bandpass-filtered (10–1000 Hz). Figure 10 illustrates a simultaneous recording of the electro-olfactogram and voltage-sensitive dye signals from the bulb.
Data analysis. NeuroPlex software was used to digitally filter, analyze, and display the data. The high-pass filter was a numerical simulation of an RC circuit, and the low-pass filter was a Gaussian. The fractional change of the resting light intensity (ΔF/F) was calculated and plotted as traces in the figures. For the pseudocolor displays, colors were assigned so that red represents signals that were larger than 70 or 80% of the largest signal on any detector. In Figure 4 we used black contour lines to delineate the areas of the bulb where the signals are larger than 20% of the largest signal.
The latency and initial frequency of the oscillations were determined as follows. After filtering (3–30 Hz), the time between the onset of the odor command pulse and the oscillations or DC signal was measured (for latencies), and the time between the first two peaks of the oscillations was measured (for initial frequency) (Fig.2 A, arrows numbered1–4). The average latency and frequency of several trials in each preparation were calculated, and the means ± SEM of these averages across the experimental animals are presented in the text and in Table 2.
We fit the three oscillations numerically to provide a quantitative representation of the shape of the signals (see Fig. 7). The best approximation to the shape of the oscillation, e.g., power of the sine (p) and change in frequency (r), was found by visual inspection. The optimal constants of the envelopes for the oscillations were found using programs written in Mathematica.
Comparison of local field-potential and optical recordings from the olfactory bulb
In preliminary experiments we compared LFP measurements (the method used earlier) and voltage-sensitive dye recordings. Confirming previous observations (Beuerman, 1975, 1977), odor stimuli elicited local field-potential responses consisting of a slow DC signal and “sinusoidal” oscillations riding on top of the DC signal. The voltage-sensitive dye recordings had a similar character. The LFP recordings from two animals are shown (after high-pass filtering to enhance the oscillations) in Figure 2(top traces) together with simultaneously made optical recordings from individual detectors (bottom traces). In Figure 2 A the two recordings are relatively simple and appear to be highly correlated. However, the local field potential occasionally exhibited complicated patterns that suggested multiple frequency components (Fig. 2 B, top trace); there is a higher-frequency oscillation (∼14 Hz) followed by an oscillation with a longer latency and lower frequency (arrowheads, ∼5 Hz). Optical recordings from a rostral and a caudal region of the bulb are shown in the bottom part of Figure 2 B. The optical signals from these two regions of the bulb have very different signals. A simple, high-frequency oscillation was seen in the rostral region. The oscillations in the caudal region were more complex; some components of these complex oscillations matched the low-frequency signal seen in the LFP recording (arrowheads).
These results imply that the odor-induced response consists of multiple signals that are located in different regions of the olfactory bulb. The data presented below provide evidence for three different oscillatory signals in addition to the slow DC component. These three oscillations were named rostral, middle, and caudal according to the regions of the bulb where they occurred.
Multiple components of the odor-induced response
In Figure 3, the recordings from seven selected diodes in a single trial are shown. The location of these diodes is indicated on the image by the numbered squares on the left. In rostral locations (detectors1 and 2), we found a single oscillation with a relatively high frequency. On a diode from a middle location (detector4), there appeared to be a relatively brief, short-latency oscillation, and on a diode from the caudal bulb (detector 7), the oscillation consisted of a low-frequency, long-latency oscillation. In areas between two regions, the recorded oscillations were combinations of two signals: rostral/middle in detector 3 and middle/caudal in detectors5 and 6. After high-pass filtering, the DC signal appeared as a single peak; it was present in all seven locations. Figure 4 A shows the time course of an unfiltered recording (from the rostral region). The DC signal rose to a plateau and then continued for a period of seconds. The low-frequency noise in our recordings prevented us from determining the time course of the return of the DC signal to the baseline. After a delay, the rostral oscillations appeared on the DC response.
Figure 4 provides additional data from the same trial illustrated in Figure 3. The time courses from four detectors from this trial together with multiple-frame images indicating the position and propagation during one cycle of each oscillation are shown. The middle, caudal, and rostral oscillations (Fig. 4 B–D) are shown after the DC signal was reduced with a high-pass filter. Four generally observed characteristics of the three oscillations can be seen in this trial. The rostral oscillation had a medium latency and a high frequency (Fig. 4 D). The middle oscillation (Fig. 4 B) had a short latency and a frequency that was similar to the rostral oscillation. The caudal oscillation had a lower frequency and the longest latency (Fig. 4 C). In addition to differences in frequency and latency, the three oscillations also had different shapes: the rostral and caudal oscillations had relatively sharp peaks whereas the middle oscillation was more sinusoidal (see below). Finally, the frequency of each of the three oscillations appeared to decrease over cycles (see below).
The pseudocolor images in Figure 4 show the generally observed location and the change in location over time of the four components. In these multi-frame images, the red color and the area enclosed by the black line indicate the areas where the signals are larger than 80 and 20% of the maximum signal, respectively. The DC signal was detected over most of the ipsilateral olfactory bulb. The rostral signal (D) initiated rostrally and propagated in the caudal direction, the middle signal (B) did not appear to propagate, and the caudal signal (C) appeared to propagate in a lateral–caudal direction. We attempted to measure the average propagation velocity of the three oscillations. In every trial the rostral oscillation originated at the most anterior/rostral part of the preparation and propagated caudally at a mean speed of 0.12 ± 0.01 mm/msec (n = 8). Because the signal-to-noise ratios for the middle and caudal oscillations were generally lower, detecting time shifts was more difficult. Clear propagation of the caudal oscillation could be measured in two animals. The mean speed was 0.18 mm/msec. In these instances, the caudal oscillation appeared to propagate in every direction from a somewhat medial point of origin. We could not detect significant time delays in measurements of the middle oscillation. The propagation velocity of the rostral oscillation is similar to the propagation velocity of the action potential in the incoming olfactory receptor axons.
In a typical odor-induced response, the DC signal appeared almost immediately after the stimulus, followed by the middle oscillation. The rostral and caudal oscillations had longer latencies. The latency of the DC signal from the start of the odor command pulse was 250 ± 10 msec (n = 35). The rostral oscillation had an onset latency of 750 ± 40 msec (n = 29). The middle oscillation had a latency of 580 ± 20 msec (n = 22). The latency of the caudal oscillation was 1100 ± 50 msec (n = 16) (Table 2). Note that ∼100 msec of these latencies reflect the time between the onset of the odor command pulse and the arrival of the odorant at the epithelium. Thus the actual latencies are ∼100 msec shorter than the numbers presented. One-way ANOVA between the latencies of the four signals yielded a significant, overall difference: F (3,15) = 138,p < 0.0001. Fisher's protected least significant difference (PLSD) post hoc test was performed to compare the four group means in pairs, and they were all significantly different from one another (p < 0.0001).
The rostral oscillation had an initial instantaneous frequency of 14.1 ± 0.3 Hz (n = 29). The initial frequency of the middle oscillation was 13.0 ± 0.50 Hz (n = 22), and the initial frequency of the caudal oscillation was 6.6 ± 0.3 Hz (n = 16) (Table 2). One-way ANOVA comparing the frequency of the rostral, middle, and caudal oscillations also gave a significant difference: F (2,15) = 91, p < 0.0001. Post hoc pairwise comparison using Fisher's PLSD showed that the difference between the rostral and caudal oscillations and middle and caudal oscillations was significant: p < 0.0001. The difference in frequency between the rostral and middle oscillations was also significant by Fisher's PLSD (p < 0.05), but it was not significant by the more conservative Tukey-Kramer honestly significant difference (HSD): p > 0.05.
As suggested by the results in Figures 3 and 4, the frequency of the three oscillations appeared to decrease over cycles. Figure5 is a plot of the instantaneous frequency of the rostral (A), middle (B), and caudal (C) oscillations over cycle number from multiple trials in one preparation. Clearly, all three oscillations had a higher initial frequency that decreased over time.
The location of the initiation site of the caudal oscillation often varied from cycle to cycle. The pseudocolor images in the bottom of Figure 6 illustrate the positions of the initiation sites during asmall portion of three cycles of the rostral and caudal oscillation that occurred in one trial. The simultaneously measured oscillations from two individual detectors are shown at the top. Three cycles (indicated by the numbers) from the two oscillations are displayed in multiple-frame pseudocolor images at bottom. The ovals mark the position of the caudal initiation site, and the squares denote the location of the rostral initiation site in cycle 1. There were substantial changes in the position of the initiation site of the caudal oscillation in cycles 2 and 3, whereas the initiation site of the rostral oscillations was relatively stationary. In 7 of 10 animals analyzed, the initiation site of the caudal oscillations changed in different cycles. With only a few exceptions, the initiation site moved from rostral toward caudal. The trial-to-trial variability within one preparation was small. On the other hand, the positions of the initiation site of the rostral and middle oscillations were relatively stable over successive cycles in all 10 animals.
We examined the distribution of the durations of the three oscillations. Much of the distribution of the durations of the caudal oscillations could be fit by a power law (which could indicate a nonlinear effect), although a Gaussian distribution could not be excluded. The distribution of the durations of middle and rostral oscillations was better fit by a Gaussian distribution.
Fitting the signal shapes of the three oscillations
Numerical fitting was performed to provide a quantitative characterization of the shapes of the three oscillations. The general form of the fit of the optical signal F(t) that we used was:Equation 1where t is time and A,B(t), and C(t) together were used to describe the envelope of the oscillations as a slowly changing function of time. The shape and frequency of the oscillation are described in the following part of the equation:Equation 2The sine function describes the periodic, oscillatory nature of the traces. The exponential p of the sine function determines the sharpness of the rise and fall of the signal, andr determines the rate of decrease of the oscillation frequency. The exponential function for the rate of decrease of frequency was used for convenience; other functions (e.g., a power law) would fit the data equally well.
Figure 7 illustrates four examples of the fitting. In each of the four panels of Figure 7, the original traces (top), the expanded original signal and the overlaid fits (middle), and the difference between the fit and the data (together with the recording before and after the fitting interval,bottom), are shown. As indicated by the middleand bottom traces of Figure 7, we were able to achieve a reasonable approximation of the raw data. The bottom traces, nevertheless, have errors that are larger than and different from the background noise (except for the middle oscillation in Fig.7 A). Some of these differences were attributable to detectable amounts of another type of oscillation (rostral when fitting caudal, and vice versa) that was not taken into account during fitting.
Clear differences in shape among the three oscillations can be seen. Both the rostral and caudal oscillations had sharper peaks than the middle oscillation. The optimal values of p for the rostral oscillation ranged from 4 to 6 (six fits), whereas the caudal oscillation was even peakier (p ranged from 10 to 16, four fits). The middle oscillation was symmetrical and was best fitted by p equal to 2 (four fits). One explanation for the differences of the values of p is that the symmetrical middle oscillation represents a subthreshold sinusoidal change in membrane potential, whereas the peakier rostral and caudal oscillations represent a combination of a sinusoidal subthreshold oscillation and action potentials elicited during the depolarizing phase of the oscillation.
In the region of the bulb where the rostral and caudal oscillations overlapped, we detected more complicated signals (Fig. 7 D). Nonetheless, this signal is well fit by a linear combination of the rostral and caudal oscillations. The fit in Figure 7 D is the sum of 0.42 times the caudal oscillation in C and 0.71 times the rostral oscillation in B.
Effects of odor concentration on the four signals
Figure 8 shows examples of the four signals in response to two different concentrations of cineole, 1.7% (B) and 10% (A) of saturation. Traces from selected diodes are presented on the left, and the images of activity at selected time points are shown on the right to illustrate the spatial extent of the signals. Among the three oscillations, the middle oscillation seemed to be the least affected by the concentration change. The amplitude of the rostral and caudal oscillation markedly decreased at the lower concentration. In contrast, after normalization, the spatial spread and the locations of all four signals were apparently unaffected by the reduction of odorant concentration (Fig. 8, right panel).
Additional measurements over a larger range of concentrations were made in one animal. Figure 9 is a plot of the amplitude of the oscillation versus the percentage of saturation of cineole over a range from 0.17 to 15%. Again, the middle oscillation seemed to be nearly unaffected by the concentration changes. Again in this preparation, the area of the bulb occupied by the oscillations remained large, even at the lowest concentration at which a signal could be detected.
The results from other animals were generally in agreement with the results illustrated in Figures 8 and 9. In four of five other instances, the middle oscillation was not changed in size when the concentration of odorant changed, and seven of eight animals had smaller caudal and rostral oscillations when the concentration was lowered.
Relationships between signals
We could not detect all four signals in all of the animals. Furthermore, the detection of the four signals seemed to be uncorrelated. Table 3 shows the seven combinations that we found (out of 15 possible combinations). Any one of the signals could be found, whereas others were undetectable. In some of these instances the detected signals had large signal-to-noise ratios, whereas other signals were undetectable. Among the 35 experiments in which the DC signal was detected, 8 had no detectable rostral oscillation, 15 had no middle oscillation, and 21 had no caudal oscillation. We detected all four signals in only 11 of the animals. Results from all 37 animals were included in the summary statistics in the left portion of Table 2.
Relationships between the rostral and caudal oscillations
The means from all animals in Table 2 indicate that the rostral and caudal oscillations differ in initial frequency by a factor that is close to but not identical to two. This comparison is indirect for two reasons. First, the groups of animals used for the two measurements were not identical, and second, because the latencies of the two oscillations were different, the initial frequencies were not measured at the same time. We made a direct comparison by measuring the instantaneous frequency at a time when both oscillations were present. In this circumstance the frequency of the rostral oscillation was 1.99 ± 0.01 (n = 5) times the frequency of the caudal oscillation. Thus the frequencies differ by a factor that is very close to 2.
We determined the phase relationship between the two signals by measuring the time between each peak of the caudal oscillation and the nearest peak of the rostral oscillation. In five animals the mean phase relationship between the rostral and caudal oscillations differed markedly, with a range from 0 to 80° (the trial-to-trial variability in one preparation was much smaller). In addition, there was a modest cycle-to-cycle variability in the phase relationship. The mean variation in the timing of the peaks between the caudal and rostral oscillations in 10 trials (five animals) was 6.2 ± 0.8 msec. An example of this variability in phase is shown in Figure 6(bottom), where in the cycle numbered 1, the beginning of the rostral and the caudal oscillations are separated by four frames (16 msec), whereas in cycles 2 and 3they are separated by only one frame (4 msec). To determine whether this variability could result from the noisiness of the signals, we also measured the variation between the peaks of rostral oscillations at two different points on the olfactory bulb. Here the mean variation was significantly smaller: 3.5 ± 0.7 msec. Thus, there is excess variability in phase between the rostral and caudal signals over and above the noise in the measurement.
Comparison of the rostral oscillation and the EOG
In four experiments we compared the time course of the oscillations measured with an electrode positioned on the olfactory epithelium (electro-olfactogram) and the oscillations recorded from the bulb with a voltage-sensitive dye. As can be seen in Figure10, the epithelium oscillation and the rostral oscillation are similar in frequency. Nevertheless, Figure 10shows differences between the two oscillations. First, in this instance the rostral oscillation has a longer duration than the electro-olfactogram oscillation (in other instances the rostral oscillation was equal in duration or shorter than the electro-olfactogram oscillation). Second, although the frequencies of the two oscillations are similar, they are not identical. The two vertical lines drawn at time 1 indicate that the peak of the rostral oscillation (A) occurs at about the midpoint of the two peaks of the electro-olfactogram, whereas later in the epoch, at time 2, the peak of the rostral oscillation (B) occurs near the beginning of the electro-olfactogram cycle. We examined 16 trials from four preparations, and in each case the frequency of the electro-olfactogram slowed in comparison to the rostral oscillation. The mean relative phase shift per cycle in the four animals was 5.8 ± 0.6°, significantly larger than zero,t(3) = 10.2, p < 0.005.
Dynamic origins of the caudal oscillation
Figure 11 shows two different ways that the caudal oscillation emerged. In four animals, the caudal oscillation emerged after a period-doubling of high-frequency oscillations. The top trace in Figure 11 A shows an example of this behavior. Although the high-frequency portion of this oscillation has a frequency that is similar to the rostral oscillation (Fig. 11 A, bottom trace), the relative phase of the two oscillations is shifted by 180° (seearrows), a much larger shift than could be accounted for by the rostral oscillation propagation delay; thus they are not the same oscillation recorded at two locations.
Most other animals did not have such high-frequency oscillations before the onset of the caudal oscillation (e.g., Fig. 11 B). However, it is possible that in those animals, the period-doubling transition could be much faster and thus not detected. We do not know what experimental parameters cause the difference between the results in Figure 11, A and B.
In the visual system, Crevier and Meister (1998) reported period-doubling in the electroretinogram of the salamander and the human during flickering visual stimulation. In contrast to the oscillations and period-doubling in the olfactory bulb that occur with no known temporal pattern in the stimulus, the period-doubling in the retina occurs at exactly one-half of the period of the flickering visual stimulus.
Comparison of ipsilateral and contralateral responses
We compared ipsilateral and contralateral responses to odor application to the ipsilateral naris in seven preparations. Figure12 illustrates the results from one of these experiments. The time courses of the responses from rostral and caudal areas in the two hemispheres are presented in Figure 12. The DC signal was detectable in both hemispheres, although it was smaller on the contralateral side. In contrast, oscillation was not detected in the contralateral hemisphere. In five of the seven preparations, the contralateral bulb had a DC signal (always smaller than ipsilateral) in response to the odor. In the remaining two preparations, a contralateral DC response was not detected. Oscillations were not detected in the contralateral hemisphere in any preparation. These results confirm previous field-potential measurements showing that oscillation was not elicited by stimulation of the contralateral olfactory mucosa (Adrian, 1942; Von Baumgarten et al., 1962).
Response to air
In 2 of the 37 preparations, we detected oscillations in response to room air. The air-induced oscillation had an intermediate frequency (mean of 8.1 Hz). The locations of the signals measured in response to air differed in the two preparations. In one it was similar in location to the rostral oscillation; in the other it was more caudal. We do not know whether the turtles were responding to low concentrations of odors in the room air or to the mechanical stimulation from the air inflow. Our result, that responses to air were rarely seen, is in contrast to the data from mammalian olfactory bulbs in which responses to air are more usual (Adrian, 1950; Freeman, 1978).
In addition to the DC signal, we have identified and partially characterized three different oscillations that occur in the turtle olfactory bulb in response to odors. The simultaneous occurrence of the three oscillations implies that there are multiple functional domains in the olfactory bulb that are processing olfactory information in parallel. The identification of the oscillations is a first step in an effort to try to establish the functional role of these population events in olfactory processing. Characterizing each of the oscillations requires specification of a number of parameters. We are presently attempting to determine whether some of these parameters are dependent on odor quality (Y. W. Lam, L. B. Cohen, M. Wachowiak, and M. Zochowski, unpublished results).
There was variability in the response to odorants. This included cycle-to-cycle variability (Fig. 6), trial-to-trial variability (Figs.5, 9), and animal-to-animal variability (Fig. 11, Tables 2, 3). The animal-to-animal variability was quite large. Similarly large animal-to-animal variability has been observed in other species [e.g.,Aplysia californica (Wu et al., 1994) and humans (Ojemann et al., 1989)].
Independence of the four signals
Many, but not all, of our results suggest that the four signals are separate processes independently induced by odor stimuli. First, the detection of the four signals is uncorrelated to a large extent (Table 3). In many of the animals, only one or two of the four signals could be detected. This result could not be trivially explained as experimental artifacts of poor condition of the preparation or uneven staining because in many instances the signals that were present were large. Moreover, when the olfactory bulb was sectioned after the experiment, the staining was found to be deep and uniform. Second, the oscillations have different latencies (Table 2) and durations and thus, even in preparations where all three are present, they can exist independently. Third, the oscillation have different frequencies (Table2). Fourth, the four signals occur in different locations and propagate in different directions (Fig. 4). Fifth, they respond differently to changes in odor concentration (Figs. 8, 9). Sixth, the oscillations have different signal shapes; the middle oscillation is a simple sinusoid, whereas the rostral and caudal oscillations have relatively sharp peaks that required the sine to be raised to a power of 6 or 12 (Fig. 7, Table 2). Seventh, the numerical fit in Figure 7 Dsuggested that rostral and caudal oscillations added together in a simple, linear manner in the regions where the two signals overlapped. On the other hand, one result suggests a likely relationship between the rostral and caudal oscillations. When they occur together, the frequency of the two oscillations differs by a factor that is very close to 2 (1.99 ± 0.01).
Comparison with other results
We are not aware of previous reports of multiple oscillatory signals of the odor-induced response in the olfactory bulb. However, earlier local field-potential traces (like that of Fig.2 B) are consistent with this possibility [e.g.,Ottoson, 1959a (Fig. 2 C); Delaney and Hall, 1996 (Fig.4 A)]. It remains to be seen whether future improvements in the spatiotemporal resolution of the voltage-sensitive dye measurements will result in the identification of additional oscillations.
Freeman and Di Prisco (Freeman, 1978; Di Prisco and Freeman, 1985; Freeman and Di Prisco, 1986) used a two-dimensional array of 64 local field-potential electrodes to study the spatiotemporal response of the rabbit olfactory bulb to air and odor stimuli. Our results on the turtle differ from theirs in that we found three apparently independent oscillations that were spatially separated, whereas Freeman and Di Prisco (1986) reported that the oscillations had the same waveform and a single dominant frequency everywhere in the bulb. Although it is likely that this discrepancy arises from species differences between the rabbit and turtle olfactory systems, the fact that we detected multiple signals in the turtle may result from the substantially greater spatial resolution of the voltage-sensitive dye recording.
Friedrich and Korsching (1997, 1998) made optical recordings of the activity of the olfactory receptor axons in the olfactory bulb of the zebrafish. They used anterograde labeling with calcium and voltage-sensitive dyes that restricts the dye and signal to the axons and terminals of olfactory receptor neurons. They found odorant-induced signals that were highly localized to small regions of the olfactory bulb. Similarly, Rubin and Katz (1999) measured intrinsic signals from the bulb (that are thought to be related to changes in blood flow) and also found odorant-induced signals that were localized to small regions. In contrast, we found signals that were much more global in their spatial extent even at the lowest odorant concentrations that resulted in a detectable signal. With our present ability to detect oscillations, the oscillations have the all-or-none characteristic of occurring over a large area whenever they exist.
In the salamander, we and others carried out voltage-sensitive dye measurements similar to the ones described here (Kauer et al., 1987; Kauer, 1988; Cinelli et al., 1995). However, oscillations were not detected in the salamander. More recently, Dorries and Kauer (1996)found oscillations in local field-potential recordings from the salamander. Perhaps these oscillations in the salamander involve fewer neurons than the oscillations in the turtle and were thus not detected in the earlier voltage-sensitive dye recording.
Prechtl et al. (1997) made similar voltage-sensitive dye measurements of the population signals in turtle visual cortex in response to visual stimuli. They found signals that were much more complex than those described in this paper. Nonetheless, many of the signals in visual cortex appeared to be propagating events. Thus, propagation may be a general feature of stimulus-induced oscillations in sensory systems.
Origin of signals
Assuming that the voltage-sensitive dye stains all membranes equally, the size of population signals will be proportional to cell membrane area times the change in membrane potential. Thus, it would be useful to know the relative membrane areas contributed by the of the cell populations in the bulb. In the rabbit the ratio of the number of interneurons to mitral/tufted cells is high: 20:1 for periglomerular versus mitral cells and 100:1 for granule versus mitral cells (Shepherd, 1972). The anatomy of the turtle bulb is similar to that of mammals (Johnston, 1915; Beuerman, 1977; Greer et al., 1981). However, cell counts are not directly related to the membrane area, and furthermore, we do not know the relative membrane contribution of the presynaptic axons and terminals. Nonetheless, the large numbers of interneurons suggest that important contributions to our measurement will come from interneurons and their processes.
We found clear differences in location between the three oscillations. Thus far there are no suggestions of an anatomical substrate for these differences (Johnston, 1915; Beuerman, 1977; Greer et al., 1981).
Extracellular electrode recordings of oscillations in the olfactory epithelium (EOG) and nerve in response to odor stimuli have been reported in the turtle (Beuerman, 1975) and the frog and toad (Ottoson, 1959b; Takagi and Shibuya, 1960). We made simultaneous recordings of the EOG and the voltage-sensitive dye signals from the bulb. Our data (Fig. 10) indicate that the EOG and the rostral signal have a similar frequency. However, the two signals are different in that the frequency of the EOG signal always slowed relative to the rostral oscillation. In addition, the two signals often do not have the same latency and duration. We conclude that it is possible that the rostral signal is related to the EOG oscillation but that the relationship is not simple. Additional experiments are needed to determine whether the presynaptic axons and/or terminals of the olfactory neurons contribute to the rostral signal.
Temporal encoding of odors
Several different hypotheses explain how odors are encoded by the nervous system. Spatial hypotheses include (1) odor encoding by the locations of a few highly activated and specific glomeruli (Sharp et al., 1975; Mori et al., 1992), (2) the spatial pattern of a large number of glomeruli activated by the odor (Cinelli et al., 1995), or (3) both of the above (Cinelli et al., 1995; Friedrich and Korsching, 1998; Rubin and Katz, 1999). On the other hand, in additional to the spatial patterns, the temporal sequences in which the cells/glomeruli were activated could also carry information about the odor (Delaney et al., 1994; Gervais et al., 1996; Laurent et al., 1996; Stopfer et al., 1997). Insect neurons fired in an odor-specific succession of assemblies synchronized with local field potential oscillations (Laurent et al., 1996; Stopfer et al., 1997). Figure 6 shows that during the caudal oscillation, differentlocations of the olfactory bulb (and thus different assemblies of cells) are activated in different cycles. We detected these cycle-to-cycle differences in the caudal oscillation in 7 of 10 preparations. This fraction is probably an underestimate because we could only determine the initiation site in two dimensions. There may have been additional variation in thez-direction. Thus, this aspect of our results is consistent with the results from intracellular recordings in locust (Laurent et al., 1996) and with the idea that odors are represented by odor-specific successions of neuronal assemblies (Laurent et al., 1996;Stopfer et al., 1997). Neural network models that can recognize time sequences have been proposed (Kleinfeld, 1986; Sompolinsky and Kanter, 1986; Tank and Hopfield, 1987).
In view of the ubiquity of stimulus-induced oscillation across species and sensory modalities, it is reasonable to speculate that oscillations may have an important role(s) in perception. Our data show that the odor-induced oscillations in the olfactory bulb are substantially more complicated than had been anticipated.
A number of questions remain. What is the cellular origin of the DC signal and the three oscillations? Are these three oscillations functionally independent processes that can be separately evoked by manipulating the stimulus conditions, such as odor concentration, odor types, sensitization, habituation, or more complex forms of learning? Finally, what are the functional roles of the four signals in olfactory perception? Additional experiments will be required to answer these questions.
This work was supported in part by Grant NS08437 from the National Institute of Neurological Disorders and Stroke, and a Brown-Coxe fellowship from the Yale University School of Medicine. We thank Charles Greer, John Kauer, Bill Ross, Brian Salzberg, Sid Simon, and Dejan Zecevic for helpful suggestions on this manuscript. We are grateful to David Senseman for the loan of the 464-element photodiode array and to John Kauer for help with the construction of the olfactometer.
Correspondence should be addressed to Dr. Ying-Wan Lam, Department of Cellular and Molecular Physiology, Yale University School of Medicine, 333 Cedar Street, New Haven, CT 06510. E-mail:.