The hyperpolarization-activated cation current (Ih) is expressed in nearly every principal neuron1,2,3,4,5,6 and in some interneurons7. It's involvement in neuronal signaling has been extensively demonstrated and includes the generation of intrinsic resonance8,9, synaptic integration1,2,5, homeostatic regulation of neuronal excitability10 and synaptic plasticity11,12. More recently evidence has accumulated that Ih not only exerts a role at resting membrane potentials but can also affect the active properties of neurons. For example, in stellate cells of the entorhinal cortex, Ih controls the spike pattern by promoting recovery of the action potential afterhyperpolarization6. Both the input and output properties of fast-spiking basket cells in the dentate gyrus are shaped by the presence of HCN-channels7.

In the olfactory bulb, mitral cells integrate synaptic activity from the olfactory nerve13 and the local bulb circuitry. In mammals in vivo, mitral cell action potential output is typically binned into bursts that are controlled by the respiratory rhythm where timing of action potentials within cycles is proposed important for olfactory processing14,15,16,17,18. However, the mitral cell action potential patterns are shaped not only by complex integration of the synaptic activity in the network of the bulb, but also by the intrinsic membrane currents present in the mitral cells19,20,21,22. In most principal neurons of the central nervous system the Ih current and/or its corresponding sag has been identified and characterized in detail. However in mitral cells evidence for functional Ih currents is indirect. The lack of clear electrophysiological evidence of sag expression in mitral cells has led to the assumption that mitral cells are one of the few principal cell types that do not express the Ih current23 though two studies do provide evidence for a small, membrane potential sag21,24. Immunogold localization of the HCN1 subunit in rat shows a moderate signal in external tufted cells whereas mitral cells have undetectable HCN1 expression25. However the presence of other HCN subunits both at the mRNA and protein levels have later been detected in rat and mouse olfactory bulb26,27.

Here we biophysically characterize the Ih current and its distribution in mitral cells. We reveal a striking diversity in the amount of Ih sag observed across the mitral cell population. In the proportion of mitral cells which express sag we find a slow, ZD7288-sensitive current which activates at voltages beyond −75 mV. Furthermore we found that the degree of sag in a mitral cell substantially impacts its spiking probability and patterning suggestive of an important role of Ih in olfactory sensory coding.


Diversity in hyperpolarization-evoked sag potentials in mitral cells

This study was motivated by our initial in vivo observations that when we rapidly hyperpolarized the membrane potential of mitral cells to beyond −80 mV for more than 600 ms we would often observe a depolarization, resembling the Ih sag reported in many other cell types28,29. However across cells, the amount of sag recorded in vivo varied considerably (sag range: −3.4 – 22 mV, mean ± S.D. = 6.6 ± 7.5 mV, n = 13, Fig. 1). One potential explanation for this diversity is that by using blind recordings in vivo, we were sampling from more than one class of cell. For example, deep tufted cells that exhibit the sag phenotype reside near the mitral cell layer25,30. Also, a direct inhibitory effect of ketamine on HCN1 and HCN2 channel subunits expressed in HEK cells and on the Ih current in dendrites of pyramidal neurons31 has been noted. Such potential caveats could therefore account for the observed variability of our in vivo -recorded data set. To examine mitral cell Ih specifically and in detail, we therefore moved to the in vitro olfactory bulb preparation that allows us to unambiguously identify mitral cells based on the location of their cell body in the mitral cell monolayer. Recording in slices also permit us to extend our observations beyond the dorsal region of the bulb where our in vivo recordings were carried out.

Figure 1
figure 1

Variability of hyperpolarization-evoked sag potentials recorded in vivo.

A1 Example of hyperpolarization-induced sag potentials recorded from a mitral cell on the dorsal surface of the rat olfactory bulb (average of 20 sweeps for each hyperpolarizing current injection step). A2 Plot of the peak (*) vs. the steady state (**) shows the voltage-dependency and variability of sag amplitude (n = 13).

Using DIC microscopy our in vitro targeted recordings from cells in the mitral cell layer showed similar variability in the amount of sag observed (sag range: −14 – 29 mV, mean ± S.D. = 2.6 ± 8.9 mV, n = 83, Fig. 2A). There appeared no obvious correspondence between the amplitude of recorded sag and the location of the cell in the olfactory bulb slice. The location and identity of recorded cells was further confirmed with DAPI stains of the recorded slices. In all cases (n = 28 cells, n = 15 slices) the soma of biocytin-labeled and recorded neurons resided in the mitral cell layer (Figure 2B). Based on biocytin staining all cells were relatively intact (n = 40 cells) such that there appeared no obvious anatomical explanation (for example, gross dendritic amputation) for sag diversity. Thus the variability in sag expression in vitro appears entirely consistent with our in vivo data (Figure 2C, D) where the morphology of mitral cells is unperturbed.

Figure 2
figure 2

Variability of hyperpolarization-evoked sag potentials recorded in vitro.

A Voltage responses to hyperpolarizing current injections in a sag expressing (black) and a no-sag expressing cell (blue). Dashed red boxes show the region of the traces that has been magnified on the left to show the exponential time course of the charging of the membrane at the beginning of the current injection. The sag cell was injected with a maximum of −300 pA decreasing by −50 pA for every sweep. The no sag cell was injected with a maximum of −1777 pA decreasing in steps of −355 pA. B Example of the sag amplitude variability across mitral cells (GL = glomerular layer, EPL = external plexiform layer, MCL = mitral cell layer, GCL = granule cell layer. Scale bar is 100 µm). C Peak vs. steady state voltage showing the voltage-dependency of the sag and its variable amplitude between cells (n = 83). The sag (black) and no sag (blue) examples are the cells shown in A. D Distribution of sag expressing cells recorded both in vitro and in vivo.

Biophysical characterization and compartmental distribution of a hyperpolarization-activated current

To assess the biophysical nature of the current underlying the membrane potential sag we performed whole-cell voltage-clamp recordings in slices in the presence of a cocktail of both synaptic and ion channel blockers that allow us to isolate a potential Ih current (see methods). These recordings revealed activation of a current at negative potentials (Fig. 3A, B) that gradually increased over time to reach steady state (> 400 ms, Fig. 3C) becoming larger with increasing negative voltage steps (V½ = −88 ± 2 mV, n = 7, Fig. 3B). The reversal potential of the current was −32 ± 2 mV, suggestive of a cationic conductance with a permeability of K+ ions over Na+ of 31 (PNa/PK = 0.30 ± 0.04, n = 7, Fig. 3D) characteristic of HCN channels. The voltage-dependent component of this current was blocked by the Ih blocker, ZD7288. However we also found that a substantial component (46 ± 3%, n = 5, measured at potentials > −105 mV) of the ZD7288-sensitive current activated instantaneously and was linearly dependent on voltage (see traces Fig. 3A). The amplitude of this instantaneous current (IInst) strongly correlated to that of the voltage dependent (Ih) component (r > 0.9, for voltage steps > 85 mV, Fig. 3E). The biophysical properties of this current in mitral cells are typical of slow HCN currents found in other neurons7. This indicates that HCN channels in mitral cells are likely to give rise to both a voltage-dependent and independent current component32,33.

Figure 3
figure 3

Characterization of a hyperpolarization-evoked current.

A In the presence of synaptic and ion channel blockers (see methods) the cell was voltage-clamped at −50 mV and held between −50 and −140 mV in steps 10 mV for 2 seconds. Subsequently ZD7288 was applied, thus the traces examples are the ZD7288 sensitive current exclusively. B Activation curve of the tail current (measured at arrow in A) obtained from recordings as in A. Sigmoid fitting indicates half maximal activation at V½ = −88 ± 2, n = 7. The activation (C1) and deactivation (C2) times were determined by single exponential fitting (red). C3 Pooled data showing activation (grey) and deactivation (black) time constants (n = 7). D The reversal potential of Ih was determined to −32 ± 2 mV (n = 7) by linear extrapolation to the peak of the tail current from clamping at potentials between −85 and −35 mV after full activation of the current for 2 seconds. E Plot showing the correlation between the voltage-sensitive Ih and the instantaneous (Iinst) current (r > 0.9, n = 5).

Due to the influence of Ih on synaptic integration in particular, much attention has focused on the functional expression of the current in somatic and dendritic compartments1,5. To determine the spatial distribution of the Ih current within mitral cells we next performed cell-attached dendritic voltage-clamp experiments with pipettes of the same size using an internal solution that isolates the macroscopic Ih current (see methods). In the cell-attached configuration we recorded from membrane patches at the soma, along the apical and at the most distal point of the apical dendrite. On average the Ih specific current was largest (and also the most variable) at the very distal apical dendritic recording site. In contrast, the current was on average, almost absent midway along the apical dendrite and in the patches recorded at the soma (mean ± S.D.: 26.5 ± 23.3 pA (tuft), 1.4 ± 2.6 pA (middle), 3.5 ± 4.9 pA (soma), n = 20, Fig. 4A2). This indicates that the HCN channels are non-uniformly distributed along this somatic-dendritic axis with, on average, the highest yet variable density observed in the very distal dendritic region.

Figure 4
figure 4

Distribution of I h current along the soma -apical dendritic axis.

A1 Example of the macroscopic Ih current recorded in a cell-attached patch at the branch point of the apical dendrite and the tuft. The patch potential was clamped at −60 mV and hyperpolarized in steps of 5 mV. Peak tail currents were obtained by stepping back to the holding potential after 1500 ms. A2 Peak tail currents (recorded as in A1, arrow) recorded in cell-attached patches in the soma (n = 7, soma), along the apical dendrite (n = 7, middle) and the most distal region of the apical dendrite (n = 6, tuft).

Sodium action potentials fully propagate along the entire length of mitral cell apical dendrites19,34. However attenuation of subthreshold signals in mitral cell apical dendrites has not been directly assessed using electrophysiological recordings though voltage dye imaging experiments indicate very little attenuation along the apical dendrite24. Differential signal propagation along apical dendrites with different anatomical and electrotonic lengths may substantially contribute to the cell-to-cell variability in sag potential amplitude recorded at the soma. Using dual whole-cell recordings from the apical dendrite (> 150 µm from the soma) and the soma we therefore injected hyperpolarizing current steps (to 68.3 ± 5.5mV steady-state) in the dendrite and observed an average attenuation across cells of 8.3 ± 2.5 % (n = 4, apical to soma). This small degree of cell-to-cell variability (2.5%) in dendritic attenuation between mitral cells cannot account for the broad sag distribution (range > 30 mV) recorded at the soma.

The sag potential as a proxy for Ih

It is commonly accepted that the signature of the functional expression of HCN channels in neurons is that of a current step-evoked membrane potential sag observed at hyperpolarized membrane potentials35,36. To determined the correspondence between the Ih current and sag in mitral cells we first determined the sensitivity of the hyperpolarization-induced sag to the Ih blocker ZD7288 (Sagctrl: 7.37 ± 3 mV; SagZD: 0.03 ± 0.12 mV, n = 12, p < 0.05, 10 – 40 μM ZD7288, Fig. 5A and B). The ZD7288 block of the sag was accompanied by an input resistance increase (Rin-ctrl: 89 ± 20 MΩ; Rin-ZD: 108 ± 38 MΩ, n = 12, p < 0.05).

Figure 5
figure 5

The membrane potential sag and its relation to the I h current.

A Example membrane voltage traces highlighting the effect of ZD7288 on the sag potential. The current injection steps are −356 pA and −712 pA (mean ± S.D., sagctrl: 7.37 ± 3 mV vs. sagZD: 0.03 ± 0.12 mV, n = 12, 10 – 40 μM ZD7288, p < 0.05). Dashed red box show the regions of the control and ZD7288 traces that have been enlarged to observe the bridge test pulse. B Plot of the peak vs. the steady state voltage under control and following application of ZD7288. Block of the sag was observed for all recorded voltages (n = 4 cells). C Left: Example of the sag potential recorded in current-clamp and the current recorded in Ih isolation cocktail under voltage-clamp conditions in the same cell (at −100 mV) Right: Plot of the relation between sag amplitude and the voltage-clamp recorded Ih current in the same cell (r = −0.78, p < 0.05). The data point corresponding to the example traces is represented in the graph by the open circle.

Previously, voltage-clamp experiments performed at the mitral cell soma have been successfully executed to quantify the reversal potential of inhibitory synaptic currents arising in the apical tuft37. However, since voltage-clamp recordings from large neurons with complex dendritic arborisations involves space-clamp errors38,39 we minimized such errors by working in an Ih isolation cocktail (see methods) where the majority of other conductances, including synaptic currents, were blocked. Under these conditions we determined whether there was any relation between the size of the sag potential and the magnitude of the Ih current by switching from current- to voltage-clamp mode in the same cell after wash-in of the Ih isolating cocktail (n = 9, Fig. 5C). We found a significant correlation between the amplitude of the membrane voltage sag and the Ih current (slope = −20 pA/mV, r = −0.78, p < 0.05, Fig. 5C) indicating that the Ih sag potential recorded in current-clamp mode may be used as a proxy for the Ih current density in the cell. In the remainder of our study we therefore use the sag amplitude to evaluate the functional role of Ih in mitral cells. This approach has the advantage that it allows us to monitor Ih and assess its impact on the active properties of mitral cells under more physiologically-relevant conditions in current-clamp mode.

Ih sag co-varies with other intrinsic properties

The fraction of mitral cells that express measurable sag is approximately 60 percent (58 %, n = 83, Fig. 2). While sampling cells we observed a tendency for mitral cells with large amounts of sag to spike regularly whereas cells with little or no sag were reluctant to spike and prone to stutter during sustained depolarization (Fig. 6A1). To capture this phenomenon we recorded from cells at the two ends of the sag distribution; -a group of highly expressing sag cells (sag > 3.5 mV, n = 15) versus the strictly no sag expressing cells (sag = < −2.4 mV, n = 17). As spike clustering is sensitive to the membrane voltage from where the depolarization is executed22 the resting membrane potential of all cells in this series of experiments were standardized at −50 mV using direct current injection (sag cells: −50 ± 2,2 mV, n = 15; no sag cells: −50 ± 1,7 mV, n = 17, DC ± 200 pA, p = 0.4). Our choice of −50 mV as the holding voltage is based on an average break-in potential of −52.5 ± 0.41 mV (n = 65 cells) observed for our in vitro recordings.

Figure 6
figure 6

Sag expression and the effect of ZD7288 on spike regularity.

A1 Example current-voltage relation of a sag (left) and a no sag expressing mitral cell (right). A2 Raster plots of spiking in cells shown in A1 during injection of depolarizing current up to 450 pA in steps of 50 pA. B Plot of the coefficient of variation of the mean inter-spike interval (CVISI) against current injection amplitudes ranging from 135 – 450 pA for the sag and no sag groups (mean ± S.E.M. of n = 15 sag cells; n = 17 no sag cells, p-values < 0.05 are marked with *). C Left: Example traces of spiking in a sag expressing mitral cell during a 1500 ms sustained current injection of 350 pA before (black) and after (red) ZD7288 application (left). Right: Summary of the CVISI in control and in the presence of ZD7288 (n = 8 cells).

Stepwise square pulse current injections (0 – 450 pA, Δ50 pA, Fig. 6A2) show that the regular and irregular activity of sag versus non-sag cells persists over a large range of current step amplitudes (Fig. 6A2). To analyze this spiking behavior we determined the coefficient of variation of the mean inter-spike interval (CVISI) over the duration of the current injection (Fig. 6B). Mitral cells with no sag stutter significantly more than those cells with large amounts of Ih sag (mean ± S.D., CVISI from inj135–445pA: sag cells: 0.61 ± 0.08 vs. no sag cells: 1.48 ± 0.28, n = 13–15, p < 0.05, Fig. 6B). We also found that pharmacologically blocking sag had a tendency to produce irregular spiking, thereby transforming a regular spike train into a more sparse firing pattern (Fig. 6C, black trace vs. red trace). The difference in the CVISI before and after ZD-block was found to be significant (mean ± S.E.M.: CVISI, ctrl = 0.18 ± 0.07 vs. CVISI, ZD = 1.14 ± 0.43, p < 0.05, n = 8, Fig. 6C).

To explore potential mechanisms of Ih regulation of spiking we next investigated the current threshold for spiking in sag and no sag mitral cells. The minimum amount of sustained current necessary to initiate an AP (the rheobase, Fig. 6A2 and 7A) was 2.5 times lower in the sag expressing mitral cells than the non-sag cells (mean ± S.E.M., sag cells: 71 ± 14 pA vs. no sag cells: 184 ± 22 pA, n = 15–17, p < 0.05, Fig. 7A). To assess this rheobase effect we generated frequency-current (F/I) response curves for the two groups of cells selected from each end of the sag distribution. We found that cells with no sag had a more sigmoid F/I-curve compared to sag cells that had F/I curves that were shifted to the left and almost linear for small amplitude current injections (Fig. 7B and C). Since membrane noise can influence the shape of the input-output relation by increasing a neurons' input sensitivity to weak signals40 we also assessed membrane noise (at −50 mV). As in other cell types41 we found significantly more variance in membrane potential in sag, compared to no sag cells (mean ± S.D., sag cells: S.D. = 509 ± 49 µV vs. no sag cells: S.D. = 241 ± 44 µV, n = 14, p < 0.05, Fig. 7D and E). A power spectral density analysis shows that low frequency membrane fluctuations are significantly attenuated in no sag cell compared to sag cells (Fig. 7D). However, the large voltage noise recorded in the sag cells seems an unlikely explanation for the different shapes of the sag versus no sag F/I-curves. We reasoned that the size of these membrane fluctuations cannot explain the linearization at the base of the sag cell F/I-curve (add 500 µV noise to an average no sag cell of Rin at 50 MΩ, that is ± 5 pA, which is 14 times lower than the rheobase current of the sag cells).

Figure 7
figure 7

Sag expression co-varies with other intrinsic properties.

A Left: Examples traces for a no sag and sag cell in response to the minimum current injection necessary to generate the 1st spike. Cells were held at −50 mV. Right: Summary of the difference in the rheobase between sag and no sag cells (mean ± S.E.M., sag: n = 15, no sag: n = 17, p < 0.05). B Plot of the average firing rate vs. current step amplitude in a sag (black) and no sag cell (blue). Data points fitted with sigmoid functions. C Pooled F/I curves from the sag vs. no sag cells (n = 15 for each group). D Example traces (insert) and the corresponding power spectral density of a sag (black) and no sag cell (blue) held at −50 mV for 2500 ms. E The standard deviation of the voltage fluctuation around −50 mV in recordings from sag and no sag cells (mean ± S.D.; n = 14, p < 0.05). F The sag amplitude plotted against input resistance. The data points were fitted with a straight line (black dashed, 2.6 MΩ/mV, r = 0.634, p < 0.05). Blue and black filled circles indicate the cells used for the sag and no sag groups analyzed in detail in this study. Dashed horizontal rectangles indicate four input resistance value bins that contain a relatively high number of cells that highlight a broad range of sag amplitude values.

When comparing the input resistance of mitral cells we found it to be an intrinsic property that correlates with sag amplitude in a rather counter-intuitive manner (i.e. Rin increases with sag amplitude, 2.6 MΩ/mV, Fig. 7F). Also, although the input resistance and sag amplitude co-varied, for a given resistance value the range of recorded sag potential amplitude remained extremely broad (Fig. 7F). Firstly, this indicates that mitral cell input resistance cannot be attributed solely to Ih expression (or lack thereof) but is rather dominated by several intrinsic currents and their interplay. We suggest that the differences in the spike threshold in extreme no-sag and sag expressing cells largely reflects differential regulation of other intrinsic currents. Secondly, cell to cell differences in membrane resistivity does not explain the substantial variability in sag potential amplitude recorded across mitral cells.

Ih and AP signaling evoked under in vivo -like conditions

Thus far we have examined the influence of Ih on the firing patterns evoked by non-physiological depolarizing current steps. However, during olfactory processing in vivo where slow, “sniff-coupled” subthreshold membrane oscillations impact cell spiking14,15,16,17,18 mitral cells experience a slow voltage ramp. Thus, the contribution of activation and inactivation of voltage-gated channels will be substantially different than during large, instantaneous depolarisations21,22. To explore the possible influence of Ih under more physiological conditions we used a current injection protocol that reproduces the “sniff-coupled” membrane potential fluctuations observed during our in vivo recordings (Fig. 8A1). In this way we could create more physiological conditions and present the same input patterns to sag versus no sag cells. The average membrane potential (sag: −47.7 ± 0.6 mV, no sag: −47.6 ± 2.2, p = 0.95) was maintained by current injection so that for all cells, on average, approximately 3 spikes were evoked during a chosen reference cycle (see Fig. 8A1 and methods). Injection of 25 consecutively recorded depolarizing cycles repeated five times for each cell resulted in inherently different response patterns for sag and non-sag cells (Fig. 8A2). Compared to no sag cells, cells with sag showed consistently high spiking probability in many cycles (eg cycle 2, 3, 6, 7, 11, 13; overall mean spike probability per cycle ± S.D. sag cells: 0.8 ± 0.11 vs. no sag cells: 0.45 ± 0.22, p < 0.05; Fig. 8A2). The fraction of cycles where sag and no-sag cells fired 3 or more spikes was found to be very similar (Fig. 8B). However sag cells were more likely to fire single and double spike responses since cells without sag remained subthreshold during a vastly increased number of inhalation cycles (Fig. 8B). This suggests that during odor sampling in response to depolarizing synaptic input14,15 whether or not a given mitral cell expresses Ih sag will impact the likelihood of the onset of firing.

Figure 8
figure 8

Sag and spike output during physiologically-relevant fluctuations in membrane potential.

A1 Mitral cell membrane potential (black trace top, only 4 cycles are shown for clarity) obtained via in vivo whole-cell recording showing nasal inhalation-coupled subthreshold oscillations. This waveform was delivered via current injection into either sag (black) or no sag (blue) cells in vitro. Bottom: example traces highlighting mitral cell responses to oscillatory current injection. A2 Raster plots obtained from spiking responses to injection of the same in vivo stimulus (total of 25 cycles) repeated five times in each of the five sag (top panels, black) and no sag cells (bottom panels, blue). Cells are ranked from 1 to 5 based on their overall firing probability (highest to lowest). Beneath each group of raster plots is a histogram showing the mean number of spikes evoked within each cycle. B Pie chart showing the fraction of cycles that failed to evoked spikes for both groups of cells. Bar graphs showing the probabilities for none, one, two, three or four or more spikes discharged per cycle.


In this study we investigated the occurrence, cellular distribution and functional implications of Ih -mediated sag in olfactory bulb mitral cells. Both in vivo and in vitro we observed heterogeneity in the amount of Ih sag recorded in response to hyperpolarizing current steps that was found to be mediated by currents with voltage relations, kinetics and pharmacological sensitivity indicative of HCN channels36. Our biophysical characterization indicates that HCN channels (possibly the slow HCN channel types (HCN2-4)) are expressed in mitral cells26,27 and that they mediate the depolarizing current responsible for the sag potential. Our data support the idea that the expression of Ih and its impact (either direct or indirect) on the active properties of mitral cells reflects a functional continuum spanning cells with large amplitude sag prone to spike regularly, to stuttering cells containing little or no sag and a high current threshold.

Across the mitral cell population the amplitude of Ih sag is highly variable. Several studies in other brain regions have noted substantial variation in the amount of Ih expressed within the same cell type. For example, the Ih conductance in hippocampal CA3b pyramidal neurons shows large variance in its expression that appears unrelated to a concurrent large variation in the passive properties of those cells42. In the entorhinal cortex, stellate cells exhibit differences in the frequency of subthreshold intrinsic oscillations along the dorsal-ventral axis43, a property that is diminished in HCN1 knockout mice44. Similarly, the synaptic integrative properties of the stellate neurons are tuned to the organization of grid cell firing fields, a difference that is accounted for by a dorsal-ventral expression gradient of Ih and leak channels45. In this study we did not observe any obvious relation between the amount of Ih sag recorded in mitral cells and their anatomical location across the bulb.

Why neurons within the same morphological class of cell might exhibit substantially different biophysical properties, is not well understood46. It seems likely that even beyond development, plasticity mechanisms47,48,49 operating within hundreds of milliseconds to minutes and even hours time frame will continue to influence channel expression and overall excitability48,50. The heterogeneous distribution of Ih between mitral cells may therefore arise from modulation of channel kinetics, changes in the total number of ion channels in the membrane or be due to more complex mechanisms such as shifts in ion channel distribution51.

If the overall intrinsic profile of a mitral cell reflects ongoing active modulation of cell excitability what kind of activity might shape it? Examples exists where the Ih current within dendrites of principal neurons is regulated by glutamatergic synaptic input10 and post-synaptic firing12. In the olfactory bulb variable excitability may arise through membership to specific glomeruli that are known to receive functionally distinct sensory input. Interestingly, juxtaglomerular cells in the glomerular layer that receive the same sensory input as mitral cells show high cell-to-cell variability in HCN1 expression25. Consistent with the idea that Ih may play a role in the integration of olfactory signals we find that the largest cell-attached currents were recorded immediately adjacent to the apical tuft that provides the anatomical locus for direct sensory inputs onto mitral cells13. However, in this study we did not attempt to obtain any cell-attached recordings from mitral cell lateral dendrites or axon and thus cannot exclude the presence of HCN channels in those locations.

The fact that the input resistance of mitral cells with Ih is higher than those cells without sag is non-intuitive. This indicates that other conductances contribute substantially to the relation between membrane resistivity and the Ih sag. Bath application of ZD7288 while recording from sag cells shows that there is a correlation between the presence of active Ih channels and action potential firing in mitral cells. This may reflect a direct Ih-mediated mechanism. An inward conductance with a reversal potential of around −30 mV could potentially affect the rate at which the action potential threshold is reached. However, since the Ih current reverses near the action potential threshold, the impact on spiking will be minimal due to a potential lack of driving force during the activation of the sodium conductance. Rather such an effect would be most substantial just prior to the opening of the sodium channels. Similarly, during ongoing spiking, Ih may increase the likelihood for repeated firing by promoting recovery from the after hyperpolarization52. With a midpoint of activation at −88 mV, a typical half-activation value of Ih observed in many other neurons5,33,53, the open probability of the HCN channel is however likely to be quite low at membrane potentials around action potential threshold. It is also plausible that the instantaneous component of the HCN current activated at more depolarized potentials than the voltage-sensitive component may contribute. The presence of a ZD7288-sensitive and voltage-independent conductance associated with HCN expression has been reported in other cell types7,33 and with heterologous expression of HCN2 channels32,54.

We have found that while the input resistance and the amplitude of sag co-varies, for a given narrow input resistance range the sag distribution remains very broad. Together with the observation that the cell-to-cell variance in attenuation of subthreshold hyperpolarizing signals in the apical dendrite is extremely small, it appears that the variability in sag recorded in our experiments reflects a genuine broad distribution of Ih levels across the mitral cell population. We also suggest that the relation between sag potential and spiking could be indirect, such that high expression of Ih influences or is influenced by regulation of other membrane conductances53,55. Such counterbalancing of ion channel expression within individual mitral cells has previously been described21. Similarly, stellate cells in the enthorinal cortex express both potassium leak channels and Ih45. Additionally, evidence of converse regulation of Ih and TASK channels is found in recordings from thalamocortical relay neurons53. However, investigating the molecular identity of a conductance that may be co-expressed with the Ih current was not the objective of this study.

Individual mitral cells receive odor-specific and concentration-dependent patterns of synaptic input14,56. At this time we do not know the extent to which the heterogeneity in the Ih current is sensitive to the processing of sensory input. However, our observation on the variability of Ih and other functional properties provides some support to a recent study that highlights the fact that intrinsic diversity improves the information content of the spike trains across the mitral cell population57. In our study we show that the amount of sag expressed in mitral cells impacts their input-output response to depolarizing steps of current. Sag expressing cells have an almost linear response function for low current amplitudes and show increased sensitivity to depolarizing current (F-I curve is shifted left). Similarly, we find that for in vivo- like current injection patterns, sag cells exhibit an increased propensity to fire low numbers of action potentials while the non-sag cells tend to remain silent.

Given that it is possible to identify the neuronal networks to which individual mitral cells belong and that input to bulb can be readily manipulated it seems the olfactory bulb may prove an ideal model system to study the impact of sensory processing on neuronal diversity. Furthermore, the possibility of behavioral and in vivo imaging and electrophysiological experiments make it feasible to directly study neuronal diversity in the context of sensory representation and olfactory performance. Here we show that heterogeneity in Ih across mitral cells provides a biophysically broad dynamic range for integrating synaptic input from the periphery. In this way, mitral cells may use Ih to adjust their gain to maintain optimal information transmission to downstream structures.



All procedures involving laboratory animals were approved by the UK Home Office. In vivo whole-cell patch-clamp recordings from olfactory bulb mitral cells were carried out in Sprague-Dawley rats (P21-25) anaesthetized with a mixture of ketamine/xylazine (50/5 mg kg−1)(Sigma) delivered by intraperitoneal injection58. During the experiment the rat was breathing freely, its body temperature was kept at 35–37°C and the anesthesia was supplemented according to requirements. Respiration was recorded using a piezo-electric strap. After head-fixation with dental cement (Jet Denture Repair, a craniotomy was performed over either of the bulbs and the dura was removed. The criteria used to identify mitral cells were a minimum recording depth of 180–200 μm, respiration-coupled membrane oscillations and an input resistance between 40–200 MΩ (103 ± 28, mean ± S.D., n = 13)56,59. The membrane voltage was recorded with the Axoclamp 2B amplifier (Molecular Devices) and the data acquired with the Nclamp/Neuromatic software (J. Rothman, built to run under Igor Pro (Wavemetrics) using an ITC-18 interface (HEKA Electronics). The in vivo voltage signal was low-pass filtered at a frequency of 6 kHz and sampled at 20 kHz. Glass pipettes where pulled to a resistance of 5–8 MΩ and whole-cell recordings with > 40 MΩ access resistance were discarded. The internal solution for both whole-cell in vivo and in vitro recordings contained (mM): 130 methanesulphonic acid, 10 Hepes, 7 KCl, 0.05 EGTA, 2 Na2ATP, 2 MgATP, 0.5 Na2GTP and 0.4% biocytin, KOH titrated to pH 7.2. The membrane voltage was not corrected for the junction potential.

In vitro whole-cell recordings were carried out on acute slices of the rat olfactory bulb (P20-P35, Sprague-Dawley). After decapitation of the animal, the olfactory bulbs were surgically removed and horizontal slices were cut at a thickness of 300 µm in ice-cold artificial cerebrospinal fluid. The slices were incubated at 35°C for 45 minutes and subsequently equilibrated to room temperature. The slices were transferred to the recording chamber and the cells were visualized under a 40× water immersion objective on an upright Zeiss Axioscope II. Only mitral cells with their somata residing in the mitral cell layer were recorded in this study. Current and voltage recordings were performed with a Multiclamp 700A amplifier (Molecular Devices) using acquisition hardware and software described above. During voltage-clamp recording the signal was Bessel-filtered at 3 kHz and during current-clamp at 10 kHz. Both signals were digitized above 20 kHz. The external solution contained (in mM): 125 NaCl, 2.5 KCl, 2 CaCl2, 1 MgCl2, 25 NaHCO3, 1.25 NaH2PO4 and 25 D-glucose, perfused with 95% O2 and 5% CO2. To isolate Ih during whole-cell voltage-clamp recordings we used an external solution without NaH2PO4 and CaCl2 containing (in mM): 105 NaCl, 12.5 KCl, 26 NaHCO3, 1 MgCl2, 1 CoCl2, 10 Tetraethylammonium-Cl (TeaCl), 1 BaCl2, 5 4-aminopyridine (4-AP), 0.001 TTX, 0.05 picrotoxin, 0.01 NBQX, 0.01 D-APV.

For all current-clamp recordings the bridge was carefully balanced and the capacitance of the pipette was compensated during recording. The whole-cell currents recorded from voltage-clamping of the mitral cells in figure 3 were compensated for series resistance (16 ± 5 MΩ, n = 7) errors by 40–50% on-line. To predict the voltage error of a typical voltage-clamp experiment done in the Ih isolation solution we performed double somatic recordings, where one electrode was used as a simple voltage follower. The maximum voltage error measured was less than 10 mV at membrane potentials more negative than −130 mV. We suspect that we achieve very satisfactory voltage-clamp and therefore reasonably accurate estimates of the Ih current amplitude due to the block of non- Ih currents with the isolation cocktail and because the Ih current is itself a very slow conductance. No leak subtraction was performed when voltage-clamping. Rather the voltage-clamp traces of figure 3 were obtained by digital subtraction of the current recorded in the Ih isolation cocktail before and after application of ZD7288. In this way we minimize any potential non-specific effect of ZD7288. To minimize any wash-in or wash-out effects caused by intracellular dialysis we typically recorded sag data in the current-clamp experiments within 1–4 minutes following break-in.

To isolate the Ih current in membrane patches the pipette contained the following solution (in mM): 120 KCl, 2 CaCl2, 1 MgCl2, 10 Hepes, 20 TeaCl, 5 4-aminopyridine, 1 BaCl2, 0.02 CdCl2 and 0.001 TTX. For cell-attached recordings the external solution contained 200 nM TTX to prevent spontaneous firing of the mitral cells. Patch pipettes used in the cell-attached patch-clamp recordings all had a tip resistance of 10 MΩ to obtain membrane patches of similar sizes. Signals were filtered at 1–2 kHz and sampled at 25 kHz. No leak subtraction of the cell-attached currents was done but all patches with a leak > 20 pA were discarded. The recording location was separated into three groups: 1) the middle of the soma (soma), 2) at least 75 µm away from the soma and 50 µm from the glomerulus layer (middle of the dendrite) and 3) at the boarder of the glomerular layer or the tuft branch-point (tuft)(see Fig. 4A2).

ZD7288 (4-(N-ethyl-N-phenylamino)-1,2-dimethyl-6-(methylamino)- pyridinum chloride (Tocris) was used at a concentration of 10–40 μM. In the analysis of mitral cell spiking the experiments were carried out in the presence of synaptic blockers: 25 µM D-APV, 10 µM NBQX and 50 µM picrotoxin (Sigma). All slice recordings were done at physiological temperature by heating the external solution with an in-line solution heater (TC-324B, Warner Instruments). For the in vivo current injection experiments performed in figure 8 we used cycle 4 as a spiking reference point across cells since this cycle had one of the largest peaks and reliably produced more than 1 spike. Current injection could therefore be more easily adjusted to maintain comparable levels of excitability (approximately 3 spikes per cycle; sag: 3.4 ± 0.4 spikes, no sag: 2.6 ± 0.7 spikes, p > 0.05).


The structure of the mitral cells we have recorded from in vitro was obtained by filling the cell with biocytin through the whole-cell pipette and followed by DAB processing. Immediately after recording, the pipettes were carefully withdrawn and the slice was fixed for a minimum of 24 hours in a 4% paraformaldehyde phosphate buffered solution at 4°C. The membranes were permeabilized with pre-cooled 100% methanol at −20°C for 10 minutes. Endogenous peroxidase activity was blocked with 1% H2O2/10% methanol and the signal was amplified by incubation for 18–24 hours in the ABC kit solution (Vectastain, Vector Laboratories). The slices were washed thoroughly with phosphate buffer between each step. After DAB amplification (1 mg/ml, 15 mins at room temperature) the stain was developed by adding H2O2. The progress of the staining was followed under the microscope to ensure a dark brown coloration of the dendrites60. Lastly the slices were mounted using Mowiol 4–88 or glycerol/DAPI (4′,6-diamidino-2-phenylindole, Sigma) to stain the nuclei for visualization of bulb layers (see Fig. 2B).

Data analysis

The sag was measured as the difference between the peak and steady state voltage at the end of the current injection step. In cells expressing no sag the negative “peak” voltage was set at 100 ms after the start of the hyperpolarizing current injection. To record the Ih sag in vivo, the mitral cell was prevented from spiking with injection of continuous negative current. A negative current pulse (5–20 repetitions, 1500 ms) was injected and the sag amplitude was measured on the average trace. To determine sag variability between cells (e.g. Fig. 2) a voltage trace closest to a steady state voltage of −90 ± 3 mV and −90 ± 5 was selected for in vitro and in vivo, respectively.

The data points of figure 3B are the mean ± S.E.M. of tail currents (arrow in Fig. 3A) that were normalized to the top of the plateau of a sigmoid function. The midpoint of activation, V½, of Ih was subsequently determined by individually fitting each cell to the following Boltzmann function: Itail = A/(1 + exp[(V − V½)/slope], where Itail is the normalized peak tail current, A is the maximum tail current. The time constants of activation, τact and deactivation, τde-act, of Ih were determined by fitting to the single exponential equation: I(t) = Ioffset + Aexp(−t/τ), where Ioffset is the current at steady state.

Data were analyzed using Microsoft Excel and Igor Pro. Statistical significance was tested using a Student's t-test with a significance level at p < 0.05. In figure 3E, 5C1 and 7F the data points were fitted with a straight line and the Pearson's correlation coefficient (r) was used to test if the slopes were significantly different (p < 0.05). The power spectral density (Fig. 7D) of the voltage (2.5 sec) was generated by using the Power Spectral Density (square-window type) procedure in Igor Pro.