Local axon collaterals of projection neurons provide a source of feedback inhibition proportional to the output of the SNr. For this inhibition to regulate the output of the SNr it must be sufficient to suppress activity even in the presence of strong, phasic activation of the projection neurons. Phasic activation of the SNr population occurs, for example, at the onset of salient sensory cues (Pan et al., 2013; Schmidt et al., 2013; Figure 3—figure supplement 1). Thus, to determine whether local inhibition was sufficient to regulate the gain of the SNr network, we used ChR2 stimulation to drive repetitive somatic spiking in the projection neuron network. This recruits a population of SNr neurons with a time course and distribution of responses similar to that evoked by conditioned stimuli (Pan et al., 2013; Figure 3—figure supplement 1). We determined the consequences of local inhibition by comparing activity evoked when inhibition was intact with activity evoked following pharmacological blockade of inhibition.

Whole cell current-clamp recordings from individual SNr projection neurons were obtained from brain slices of Thy1-ChR2 mice in the presence of excitatory synaptic transmission blockers (D-AP5 and NBQX; Figure 3A). Wide-field illumination through a 10X objective was used to stimulate activity throughout the SNr network. Direct light-evoked spiking in the recorded neuron was substantially, or in some cases completely, suppressed under control conditions (Figure 3B,C). However, reliable light-evoked spiking was always present following application of the GABA_A receptor antagonist gabazine (Gbz) to block local inhibition (Figure 3D). The suppression of evoked spiking was consistent across stimulus durations within a cell (Figure 3E), whereas the magnitude of suppression was more idiosyncratic across cells for a given stimulus condition (Figure 3F).

Figure 3 with 2 supplements see all

Download asset Open asset

The ability of feedback inhibition to suppress spiking more effectively with increasing stimulus duration (Figure 3E) implies a divisive gain control. To quantify the gain effect across the population, we compared the normalized response to photostimulation of increasing duration both in the presence and absence of inhibitory synaptic transmission (Figure 3G). We found that the response of the population showed a significant increase as a function of stimulus duration and the magnitude of the increase was significantly reduced by the presence of feedback inhibition (two-factor ANOVA; p<0.05). Divisive gain control is characterized by a suppression of spiking at large stimuli but little to no effect on the response to weak or absent stimuli. Consistent with a divisive gain control, we found that baseline firing was unaffected by removal of inhibition (Figure 3G, open circles). To contrast a feedback gain control with the effect predicted for subtractive inhibition, we examined recordings from dopamine neurons of the SN that do not express ChR2, but are strongly inhibited by ChR2 expressing projection neurons (Pan et al., 2013). For dopamine neurons, we observed a constant suppression of spiking across the range of stimuli used (Pan et al., 2013). If the reduction in spiking observed during stimulation was the result of feedback inhibition one would predict that the inhibition should onset after the onset of the population response and truncate the response present in the absence of feedback. Consistent with this prediction we found that suppression of spiking was characterized by a significant decrease in the duration of the evoked spiking (Figure 3H). This effect on the duration could be observed in many individual neuron responses (Figure 3F).

Our in vivo results suggested that the extent of suppression of transient activation in SNr neurons is proportional to the estimated activation of the network (Figure 1). Anatomical studies indicate that the vast majority of projection neurons elaborate collaterals within the SNr, however, these collaterals can form relatively few (∼10) putative synaptic contacts (Mailly et al., 2003). Moreover, we found that unitary release events produced relatively weak mIPSCs (∼150 pS) (Figure 3—figure supplement 2). Taken together, these data imply that the inhibition observed results from the activation of approximately 50–100 presynaptic inputs. Given that the GABAergic neurons in the SNr are thought to be exclusively projection neurons (Deniau et al., 2007b), this is consistent with the finding that projection neuron collaterals form 79.4 ± 96.1 (SD) boutons per neuron within the SNr of the rat (Mailly et al., 2003).

Assuming a modest or low probability of paired connections, our connectivity estimates imply that the extent of activation across a large population of SNr neurons would determine the extent of feedback inhibition consistent with our observation in vivo. While we do not have a direct measure of the total extent of activation of the SNr by our photostimulation, we note that neurons recorded simultaneously experience the same activation state of the network. Thus, we reasoned that the extent of feedback inhibition in a pair of recorded neurons should be correlated if feedback inhibition is proportional to the total activation of the network. Consistent with this prediction, we found that there was a significant correlation (Pearson's correlation, p<0.01 permutation test) in paired recordings (Figure 3I). These results suggest that a given SNr projection neuron receives input from a spatially diffuse collection of other SNr projection neurons.

The data above are consistent with the claim that a large population of SNr projection neurons must be recruited to fire within a relatively small time window (5–20 ms) in order to achieve robust feedback inhibition and divisive gain effects (Figure 3G). These results were obtained in the Thy1-ChR2 mouse where all neurons in the SNr express ChR2 (Figure 2). This would suggest that if a local subset of the SNr was expressing ChR2, the divisive gain effect should be present, but reduced in magnitude analogous to the smaller effects observed when less of the network was recruited in the Thy1-ChR2 preparation (Figure 3G). Indeed, we found that when the same experiment was repeated in ChR2+ nigral neurons from virally infected Gad2-ChR2 mice a divisive gain effect was observed, but reduced in magnitude (p<0.05; two-factor ANOVA; 15% reduction in the saturated response).

To alter the gain of a response to activation of the network requires a change in the slope of the curve. As described above, we observed that there was a substantial effect of feedback inhibition in the strongly activated SNr circuit, but no effect in the absence of stimulation—resulting in a change in the slope of the response to stimulation. However, it is confusing how a strongly coupled inhibitory network of tonically active neurons could exhibit no effect of feedback even in the absence of stimulation. We first asked whether the rate of spontaneous IPSCs was consistent with our estimate, and a prior anatomical estimate (Mailly et al., 2003), of >50 inputs from other SNr projection neurons. The expected rate of spontaneous IPSCs would thus be approximately:

where, R_uIPSCs is the predicted rate of unitary IPSCs (uIPSCs), N_pre is the number of presynaptic inputs (release sites), R_pre is the mean firing rate of presynaptic inputs, and P_release is the effective release probability across all release sites. Thus, with a relatively low release probability (<0.5), we would predict 75–300 Hz of uIPSCs. This corresponds well to the rate of uIPSCs estimated directly from voltage clamp recordings (Figure 4A–C). Consistent with this estimate, we also show that repetitive stimulation of SNr collaterals fails to completely depress transmission (Figure 4—figure supplement 2) consistent with a vesicular release probability low enough to allow vesicle recycling to keep pace with release. Such a mechanism has been described in detail for Purkinje cell synapses (Telgkamp et al., 2004). These observations suggest that there is indeed a substantial background rate of IPSCs that, when pharmacologically blocked, has no significant effect on the tonic firing of SNr projection neurons (Atherton and Bevan, 2005).

Figure 4 with 2 supplements see all

Download asset Open asset

The question as we posed it—how can a strongly coupled inhibitory network of tonically active neurons exhibit no effect of feedback under basal conditions?—implies that tonic spiking is the problem. Alternatively, tonic spiking could be the solution. For a neuron to repetitively fire it must, upon the return from a spike, exhibit a net membrane current that is inward and thus drives the membrane towards spike threshold (Raman and Bean, 1999). This implies a positive slope conductance combined with a net inward current below threshold (Nolan et al., 2003)—in other words, the conductances that drive repetitive firing oppose hyperpolarizing currents in the perithreshold regime. Combined with a reduced driving force of inhibition near threshold, these biophysical features suggest that SNr neurons are much less sensitive to inhibition than to excitation in this regime. To test this hypothesis explicitly, we performed dynamic clamp experiments in which we systematically varied the balance between a high background rate of IPSCs and EPSCs (Figure 4D–F). Indeed, we found that the sensitivity of the spike rate to increasing inhibition was much reduced compared to the sensitivity to increasing excitation. Stimulation strongly biased towards an inhibitory conductance often exhibited no effect on the mean spike rate relative to tonic levels. Consistent with the mechanistic model described above, we found that the slope conductance in the perithreshold regime was indeed nonlinear with a sharp positive slope near the inhibitory reversal potential (Figure 4—figure supplement 1). Moreover, we found that measured biophysical properties (e.g., slope conductance, spike threshold) were unaffected by pharmacological blockade of inhibition (Figure 4—figure supplement 1).

Collateral inhibition resulted in a strong suppression of evoked spiking and was sufficient to truncate evoked responses, often after only a few milliseconds. This suggested that collateral inhibition provided substantial inhibition that onset rapidly following stimulation. However, it is possible that the transient effect could also reflect properties of the photostimulation. To distinguish these possibilities, we examined the kinetics of feedback inhibition and compared it to the main source of feed forward inhibition to the SNr from the striatum.

We made intracellular recordings from individual SNr projection neurons in Gad2-ChR2 mice to probe the properties of local feedback inhibition (Figure 5A,B) and from Drd1a-cre mice which were injected with a virus expressing a cre-dependent ChR2 transgene into the striatum to target the D1 receptor expressing medium spiny neurons which send axons directly into the SNr (Gerfen, 1988) (Drd1a-ChR2; Figure 5C). Postsynaptic neurons were recorded in the voltage clamp configuration with a holding potential of ∼+20 mV (reversal potential of the ChR2 current; Figure 2H–J) to isolate inhibitory postsynaptic currents (IPSCs). Slices were perfused with antagonists of excitatory synaptic transmission. Repeated pulses (10 Hz) of wide-field photostimulation elicited stimulus locked IPSCs in 15/23 SNr GABA neurons in the Gad2-ChR2 mouse line (Figure 5B) and 18/24 SNr GABA neurons in the Drd1a-ChR2 mouse (Figure 5C). Outward currents recorded following photostimulation of both inputs were completely abolished by application of Gbz (Figure 5B,C; p<0.001). Evoked IPSCs from feedback nigral collaterals recorded in Gad2-ChR2 mouse exhibited rapid kinetics characterized by short, monosynaptic latencies (Figure 5D; 1.93 ± 0.02 ms), rapid 10–90% rise times (Figure 5E; 0.53 ± 0.01 ms) and rapid decay time constants (τ) (Figure 5F; 5.64 ± 0.14 ms). In contrast to the intranigral inhibitory synapses, we found that striatonigral IPSCs recorded in Drd1a-ChR2 mouse had significantly longer latencies (Figure 5D; 2.51 ± 0.02 ms p<0.001), slower 10–90% rise times (Figure 5E; 0.75 ± 0.014 ms, p<0.001) and slower decay τ (Figure 5F; 9.00 ± 0.14 ms, p<0.001). Perhaps most surprisingly, we observed that with saturating stimulation, intranigral synapses contributed as large or greater inhibition than the major afferent source of inhibitory input, the direct pathway (Figure 5G).

Figure 5 with 1 supplement see all

Download asset Open asset

To probe the short-term plasticity properties of feed forward (Drd1a-ChR2) and feedback (Gad2-ChR2) inhibition to the SNr, we analyzed the amplitude of successive IPSCs. We found that collateral synapses within the SNr exhibited paired pulse ratios (PPR) less than 1 (Figure 5H; PPR = 0.91 ± 0.03). While, in direct contrast, the striatonigral synapse was modestly facilitating (Figure 5H; PPR = 1.17 ± 0.05). This latter observation was consistent with a previous study that used extracellular stimulation of the direct pathway (Connelly et al., 2010). The depressing nature of the PPR for local feedback inhibition is unlikely to reflect desensitization of ChR2 as repeated pulses were all suprathreshold under our stimulus conditions (Figure 2). These results are thus consistent with a rapid onset of feedback inhibition sufficient to truncate sustained activation of the SNr population.

We did not find any statistically significant differences between the kinetic properties seen between the IPSCs measured in slices from Gad2-ChR2 mice compared with those measured from Thy1-ChR2 mice (Figure 5—figure supplement 1). This is consistent with both approaches selectively or predominantly activating SNr projection neurons. By contrast, the amplitude of IPSCs evoked by maximal stimulation was significantly reduced in slices taken from Gad2-ChR2 mice compared to those taken from Thy1-ChR2 mice (p<0.05; unpaired two-tailed t test; Figure 5—figure supplement 1). These differences presumably reflect the non-homogeneous expression of ChR2 in the virally infected SNr of Gad2-ChR2 mice.

Our results demonstrate a previously unappreciated potency of feedback inhibition in the SNr (Figure 5G). Collateral synapses provide sufficient inhibition to regulate the gain of the output of the basal ganglia even during strong activation of the network (Figure 3G). However, anatomical reconstruction of individual axons suggests a sparse connectivity within the SNr (Mailly et al., 2003). The modest amplitude of individual mIPSCs (∼150 pS; Figure 3—figure supplement 2) and the low cell density of the SNr (Gerfen, 2004) (∼30,000 neurons in a ∼4 mm³ volume [Oorschot, 1996]) imply that feedback inhibition derives from a substantial volume. The maximal amplitude of evoked IPSCs was 10,000 pS. If we assume a <1% connection probability then we would predict that inhibition would be derived from neurons in a ∼600 μm radius from a given postsynaptic neuron. Consistent with such a model our wide-field stimulation experiments suggested that feedback inhibition magnitude scaled similarly for pairs of neurons (Figure 3I). While reconstructions of individual axons have been studied in detail, such results cannot be used to reliably infer the convergence of input onto an individual projection neuron. Moreover, light-level anatomy data cannot reliably predict the functional impact of feedback inhibition. Thus, we next adapted the ChR2-assisited circuit mapping (CRACM) (Petreanu et al., 2007; Wang et al., 2007) method developed for the neocortex to study the interconnectivity within nigral microcircuit.

Using a 10X objective, it was possible to contain the entire extent of the SN within a single field of view. We positioned either 81-point or 140-point grids of stimulation sites to cover the SN (Figure 6A). To achieve high spatial resolution of ChR2 activation, we used a focused 470 nm laser beam that could be rapidly re-positioned to each point on the grid in a pseudorandom sequence that avoided nearest neighbors. The duration of the light pulse was gated so as to deliver brief (<1 ms) pulses of light at each stimulation site. To obtain reliable and spatially homogeneous expression of ChR2 across nigral projection neurons, we used slices from the Thy1-ChR2 transgenic mouse line. Relative to wide-field stimulation or stimulation targeting axonal fibers, we reduced the maximal power of the laser using neutral density filters and performed calibration experiments to find intensities that would evoke precise, time-locked spikes in a small number of neurons with somatodendritic arbors surrounding the stimulation site (Figure 6—figure supplement 1). Our data were consistent with a requirement for propagating action potentials to elicit postsynaptic responses and we found no evidence in recorded neurons of direct axonal stimulation under these conditions (Figure 6—figure supplement 1). Furthermore, we focused on the rising phase and initial peak of IPSCs to bias our analysis towards transmission that resulted from highly reliable, low jitter spikes initiated at each stimulus site.

Figure 6 with 1 supplement see all

Download asset Open asset

To measure the spatial organization of local inhibitory connectivity within the SNr, postsynaptic GABA neurons were clamped to V_h = +20 mV in the presence of glutamate receptor antagonists. The peak amplitude of light-evoked IPSCs was determined for each stimulus position and response maps of IPSC amplitude as a function of stimulus position were generated (Figure 6A–C). While the viral-overexpression preparations do not provide homogeneous expression, we found that the length scale of the intranigral microcircuit estimated using the Gad2-ChR2 mouse was consistent with that from the Thy1-ChR2 mouse (Figure 6C). To characterize the spatial organization of input to individual neurons, we characterized the ‘receptive field’ of feedback inhibition as the center of mass (COM) and the 25% isoinhibition contour (ISO; Figure 6D). Finally, the majority of the somatodendritic arbor of each neuron was reconstructed from image stacks acquired on a two-photon microscope (Figure 6D) that allowed us to estimate the COM of the dendritic arbor.

It has been proposed previously that collateral inhibition may be largely confined to topographic boundaries defined by feed forward input from the striatonigral pathway (Mailly et al., 2003). However, if feedback and feed forward input were organized in register, then feedback could not produce a signal proportional to the ‘global’ activation of the network. This would imply a collection of parallel channels each of which could exhibit strong feedback. By contrast, our recording data in vivo (Figure 1) suggested that feedback was proportional to the average activation across a large spatial extent of the SN. Thus, we next asked whether the feedback intranigral inhibition was organized in register with feed forward inhibition from the striatum.

If feedback inhibition were organized in register with feed forward inhibition, then we would predict that (1) the somatodendritic position of the postsynaptic neuron should predict the location of the inhibitory receptive field and (2) the spatial extent of inhibitory receptive fields should be matched to the topographic boundaries defined by feed forward inhibition. In contrast to the first prediction, we found that there was no correlation between the location of individual neurons and the source of the strongest inhibition within the SNr. In other words, we found that neurons with non-overlapping dendritic arbors could have largely overlapping inhibitory receptive fields (Figure 6D). We found no correlation between relative somatic position and the correlation of the inhibitory receptive fields in neither coronal (n = 14) nor sagittal (n = 16) slices (Figure 6E–F).

To assess whether local inhibition in the SNr observed topographic boundaries defined by afferent inhibition, we generated double transgenic mice in which ChR2 was expressed under control of the Thy1 promoter and cre-recombinase was expressed under control of the D1 receptor (hereafter referred to as Drd1a-cre x Thy1-ChR2). We then made focal injections of a cre-dependent virus expressing a red fluorescent protein into the dorsal striatum 2–3 weeks prior to performing circuit mapping experiments in midbrain slices (Figure 7—figure supplement 1). Clear axonal labeling could be readily observed in the SNr (Figure 7A; Figure 7—figure supplement 1). The striatonigral projection exhibited the characteristic ‘dual nature’ that has been observed following focal tracer injections in rats (Gerfen, 2004). In each slice, we performed circuit mapping for three to six projection neurons at a range of distances from the axonal termination fields (Figure 7A). We found that the correlation in the maps obtained from individual neurons was a monotonically decreasing function of distance between neurons (Figure 7B). However, we found no clear organization between the striatonigral projection and the maps of feedback inhibition (Figure 7A,C). The organization of the local inhibitory circuit in relation to the boundaries of striatonigral axonal tracing suggests that there could be a partial separation between the regions that receive input from the medial and lateral striatum (Figure 7D, Figure 7—figure supplement 2). However, the consistent fall off of the correlation in inhibitory response maps across neurons (Figure 7B) suggests that projection neurons receive feedback inhibition from a diffuse microcircuit and independent of the discontinuous topography of feed forward input.

Figure 7 with 2 supplements see all

Download asset Open asset

The potency and diffuse organization of feedback inhibition suggested that collateral synapses from SNr projection neurons could be a powerful determinant of the activity of the SNr even in the presence of ongoing input from afferent sources. The in vitro preparation presumably reduces or eliminates structured activity in afferent sources of input that could either directly compete with or modulate the consequence of feedback inhibition. Moreover, the length constant of inhibition was similar in the sagittal and coronal planes (Figure 6E,F) and thus, the potency of inhibition measured in vitro is, if anything, an underestimate of the impact of intranigral inhibition. We next asked whether we could use a complementary approach to measure the extent of the intranigral microcircuit in the awake mouse. By contrast to the mapping experiments above where we measured at one location (neuron) and stimulated many other locations, we next used either a silicon probe electrode array (Figure 8A) or wire array (Figure 8—figure supplement 1) with integrated optical fibers to stimulate at one (somewhat diffuse due to light scattering) location while measuring the spiking activity of SNr neurons at many neighboring locations.

Figure 8 with 1 supplement see all

Download asset Open asset

Following implantation of a 64-site silicon probe array with integrated optical fiber into the SN of Gad2-ChR2 mice (Figure 8A), we observed single units with narrow waveforms and high baseline firing rates characteristic of SNr projection neurons (Figure 8B). The distribution of baseline firing rates obtained in vivo was closely matched to the same distribution obtained using on-cell recordings from identified SNr projection neurons in vitro (Figure 8C). We also confirmed that brief stimulation with a light pulse, like the stimulus used in the gain experiments in Figure 3, produced population responses similar in time course to those measured during response to natural stimuli (Figure 3—figure supplement 1). We found that in vivo, as observed in vitro, stimulus pulses of increasing duration produced mixed responses in the SNr population characterized by units with direct excitation and a subsequent delayed inhibition (Figure 8B, upper example) as well as units with a pure suppression of firing that onset with a short delay (∼5 ms latency; Figure 8B, lower example).

A relatively constant level of inhibition combined with increasing excitatory drive can produce subtractive effects on the output spiking of a neuron. By contrast, divisive gain effects require that inhibition be recruited in proportion to changes in excitatory drive. Our in vitro data were consistent with a divisive gain effect on nigral output due to intranigral inhibition. To distinguish these two possibilities in vivo, we examined neurons that exhibited apparent direct excitation (short-latency increase in stimulus evoked firing) with those that exhibited inhibition (stimulus evoked suppression of firing below baseline). We found that inhibition and excitation were recruited with a similar dependence on stimulus intensity consistent with a model in which intranigral inhibition produces a divisive gain effect on the output of SNr projection neurons (Figure 8D).

Finally, our circuit mapping data (Figures 6 and 7) suggested that functional inhibition extended for hundreds of microns within the SNr. To estimate the spatial extent of inhibition in vivo, we next examined the position of excitatory and inhibitory responses on the electrode array. The location of each inhibitory and excitatory response is plotted as a function of position for the 20 ms and 50 ms stimulus conditions as a function of electrode position, maximal intensity projection, or cumulative histogram as a function of distance from the focus of excitation (Figure 8E). Taken together, these data indicate that functional inhibition can extend for hundreds of microns beyond the focus of excitation in vivo as was observed in vitro.

In our experiments, we used two approaches to achieve cell-type specific expression of ChR2 in the GABAergic projection neurons of the SNr. Both viral-mediated infection of GABAergic neurons in the SNr and the Thy1-ChR2 transgenic mouse line exhibited cell-type specific expression of ChR2 in GABAergic neurons of the SN, but not in dopamine neurons. GABAergic projection neurons and dopamine neurons are thought to be the only two cell types present in the rodent SN. Importantly, both approaches had indistinguishable properties of inhibition onto both GABAergic projection neurons (Figure 5—figure supplement 1) and onto dopamine neurons (Pan et al., 2013). Moreover, viral mediated expression allowed for the best possible quantitative comparison between feed forward and feedback pathways by ensuring that expression of the ChR2 was controlled by a common promoter (Figure 5). Thus, for these results the only significant differences between the virally-mediated and transgenic expression of ChR2 in the SNr was the fraction of the population found to express ChR2 and the resultant magnitude of feedback inhibition. While viral-mediated expression could be used to estimate the divergence of inhibitory projections from a single stimulation site (Figure 8), it was not suitable for comprehensive mapping of convergence onto individual SNr neurons (Figures 6 and 7). The concern in a transgenic mouse line is that ChR2 expression could be present in afferent fibers. To control for this, we blocked excitatory transmission, confirmed that light-activated currents had kinetics expected of ChR2 positive neurons, and confirmed that neurons of the major source of inhibitory input, the striatonigral pathway, did not express ChR2. Consistent with these observations it has previously been suggested that inhibitory fibers in this Thy1-ChR2 mouse line do not express ChR2 (Gradinaru et al., 2009). However, in addition to expression in the SNr, we did also observe that neurons of the external globus pallidus (GPe) express ChR2. The GPe is the source of a projection to the subthalamic nucleus and a more modest projection to the SNr (Bolam et al., 2000; Gerfen, 2004; ). Nonetheless, we found that potent feedback inhibition could be observed in Gad2-ChR2 mice and that stimulation of GPe axons yielded relatively less inhibition in SNr neurons than either the striatonigral or intranigral pathways under our stimulus conditions. Finally, we used laser powers that were attenuated relative to the powers necessary to directly stimulate severed axons in these mice (data available from the authors on request), consistent with our observation that only perisomatic stimulation was sufficient to evoke reliable spiking in SNr neurons (Figure 6—figure supplement 1).

Unlike many other circuits in which gain control has been studied, projection neurons of the SNr are spontaneously active. Divisive gain control requires that there is little effect of inhibition in the absence of stimulation. In the case of gain control mediated by an interneuron this can be achieved in a number of ways, for example, through facilitating inhibitory synaptic transmission (Silver, 2010; Figure 9). However, in spontaneously active neurons, it is less clear how to prevent feedback inhibition from altering the baseline firing rate as we observe here. One possibility suggested by our data is that a broad distribution of firing rates (Figure 8) combined with relatively weak individual connections (Figure 4) could produce inhibition that is essentially tonic and too small to significantly affect intrinsic currents that drive repetitive spiking (Figure 4—figure supplement 1) and so is counteracted via subthreshold inward currents necessary for repetitive firing. We provided support for such a model by demonstrating that there is, indeed, a high frequency (200 Hz) of spontaneous IPSCs bombarding SNr projection neurons (Figure 4). Further consistent with this model, simulation of a steady background rate of IPSCs using dynamic clamp revealed a relative insensitivity of spike rate to a tonic net inhibitory background of inputs (Figure 4). This balance in the tonic firing rate can be disrupted by synchronously recruiting neighboring neurons to produce a rapid inhibition that, combined with the positive feedback produced by inhibition and subsequent disinhibition, overcomes via de-activation (Nolan et al., 2003) the inward currents that drive the membrane potential towards threshold. This mechanism combining intrinsic properties and synaptic properties is sufficient for a population of spontaneously active inhibitory neurons to implement divisive gain control in the apparent absence of interneurons (Deniau et al., 2007b). To our knowledge this represents a novel circuit mechanism for divisive gain control (Silver, 2010; Figure 9).

Figure 9

Download asset Open asset

One can think of two possible regimes in which the SNr may operate and each has distinct implications for the function and role of intranigral inhibition. On the one hand, it has been argued for some time that the feed forward pathways of the basal ganglia are topographically organized and largely independent (Mink, 1996; Haber, 2003; Hikosaka, 2007). Movement is thus thought to occur when a focal population of projection neurons becomes inhibited by feed forward input and thus disinhibits downstream pre-motor structures. From this perspective, the diffuse intranigral microcircuit could act to release neighboring projection neurons from intranigral inhibition and thereby suppress unintended movements. Thus, during focal activation of the SNr, collateral inhibition may be thought of as a mechanism for contrast enhancement akin to the role of lateral inhibition in sensory systems. This mechanism may be reflected in the time locked, bidirectional changes in the firing of SNr projection neurons that are commonly observed prior to and during movement in mice (Pan et al., 2013; Fan et al., 2012) and primates (Turner and Anderson, 1997; Nevet et al., 2007).

On the other hand, while the direct striatonigral pathway is topographically organized there is considerable divergence in the corticostriatal input in mice (Pan et al., 2010). In addition, there may be less precise topographic organization of the indirect pathway that enters the SNr via the subthalamic nucleus (Bolam et al., 2000). The subthalamic nucleus also receives direct cortical input and ascending input from the midbrain (Coizet et al., 2009) and hindbrain (Bevan and Bolam, 1995; Winn, 2006). The topographic organization of these ascending pathways is less well understood. Regardless of the topographic precision, the inputs that arrive at the SNr from the subthalamic nucleus convey a great diversity of information and likely exhibit a diversity of dynamics. While there are mechanisms that could maintain activity within a fixed dynamic range in upstream structures (Silver, 2010), individual neurons that constitute the output of cortical areas project to multiple subcortical structures (Kita and Kita, 2012). It is therefore unlikely that the dynamic range of even the cortical output is appropriate for the diverse computations performed in all target structures. To effectively control the basal ganglia output in the presence of such diverse input dynamics and anatomical divergence would seem to require coordinated processing across functional domains. Thus, divisive gain control supplied by a diffuse but potent inhibitory microcircuit could be well suited to ensure that activity remains within a fixed dynamic range.

We favor a model in which these contrasting descriptions of the role of the intranigral microcircuit are two aspects of its function that can be engaged in different input regimes. The diffuse organization that could produce lateral inhibition in some regimes under the animal's control (i.e., activation of specific well learnt actions), but, may be necessary to produce divisive gain control in other regimes (i.e., ‘global’ activation of the SNr by salient stimuli). While it is not currently possible to monitor nigral dynamics during selective manipulation of feedback, but not feed forward or efferent inhibition, such an approach will be required to definitively test the predictions of our work.

Simple geometrical considerations suggest that the intranigral microcircuit integrates functionally distinct information on a large scale. Cortical afferents to the basal ganglia are derived from an estimated 17 million neurons (Zheng and Wilson, 2002) spanning the majority of the roughly 100 mm³ volume of neocortex. These inputs are funneled through the basal ganglia and will ultimately terminate on approximately 30,000 projection neurons (Oorschot, 1996) within the roughly 4 mm² volume of the SN. Dendrites of nigral projection neurons and intranigral inhibition that extends over hundreds of microns (Grofova et al., 1982; Mailly et al., 2001) could therefore shape activity derived from cortical inputs separated by several millimeters.

Existing functional models focus on the feed forward structure of processing within the intrinsic basal ganglia circuitry (e.g., Albin et al., 1989; DeLong, 2000; Mink, 1996). However, our observation that intranigral inhibition is strong relative to the major source of feed forward input suggests that local processing of diverse streams of information in the SN could be critical for generating dynamics in the basal ganglia output. We hypothesize that the diverse impairments characteristic of pathological disruption of basal ganglia function could reflect, in part, a control system operating outside of a stable regime. Although there is evidence of perturbed dynamics in the SNr in disease models (Ibanez-Sandoval et al., 2007; Samadi et al., 2008; Wang et al., 2010), the specific contribution of the intranigral microcircuit to the diverse behavioral impairments observed in diseases afflicting the basal ganglia circuit remains unclear.

For in vitro experiments, adult transgenic mice (10–30 weeks old) expressing either; ChR2-YFP fusion gene under the control of the mouse thymus cell antigen 1 promoter (Line 18, Stock #007612; Jackson Labs, Bar Harbor, Maine, ‘Thy1’ mice), cre-recombinase under the control of the glutamic acid decarboxylase 2 gene (Stock #010802; Jackson Labs, ‘GADcre’ mice) or cre-recombinase under the control of the dopamine receptor D1A (GENSAT, Gong et al., 2010, Rockefeller University, ‘Drd1a-cre’ mice). For in vivo electrophysiology experiments, four adult (30 g, 3–6 months old) Thy1 mice were used. All animals were handled in accordance with guidelines approved by the Institutional Animal Care and Use Committee of Janelia Farm Research Campus. The experimenter was not blinded to genotype.

Mice were housed in a temperature- and humidity-controlled room maintained on a reversed 12 hr light/dark cycle. For in vivo physiology experiments, mice were housed individually. For in vitro experiments, mice were group housed (1–5 mice per cage).

We used 3 adeno-associated viruses (AAV, serotype 2/1) to achieve either conditional expression of ChR2 and tdTomato or pan-neuronal expression of eGFP and tdTomato. Viruses were produced at the Molecular Biology Shared resource of Janelia Farm Research Campus. Where indicated similar viruses can be obtained publicly from the Gene Therapy Program at the University of Pennsylvania (http://www.med.upenn.edu/gtp/vectorcore/Catalogue.shtml). Conditional ChR2 expression was achieved with AAV2/1 SYN-FLEX-ChR2-GFP analogous to AV-1-18917P. Conditional tdTomato expression was achieved with AAV2/1 CAG-FLEX-tdTomato-WPRE-bGH available as AV-1-ALL864. Pan-neuronal expression of tdTomato and eGFP were obtained via AAV2/1 SYN-[tdTomato/eGFP]-WPRE-SV40 available as AV-1-PV1696.

Viruses were injected into the striatum (STR) of Drd1a-cre mice, globus pallidus (GP) or substantia nigra (SN) of GADcre mice, in a fashion similar to that previously described (Atasoy et al., 2008). Briefly, under deep anesthesia, a small craniotomy was made over the STR (0.5 mm anterior-posterior, 1–2 mm medial-lateral, −2.5 mm dorso-ventral), GP (−0.45 mm anterior-posterior, −1.8 mm medial-lateral, −3.6 mm dorsal-ventral) or SN (−3 mm anterior-posterior, 1 mm medial-lateral, −4.2 mm dorso-ventral). A glass pipette was used to pressure inject small volumes of virus (20–100 nl per injection site). Animals were allowed to recover for at least 2 weeks following surgery.

Briefly, adult mice were deeply anaesthetized under isoflurane, decapitated, and the brains were dissected out into ice-cold modified artificial cerebral spinal fluid (aCSF) (52.5 mM NaCl, 100 mM sucrose, 26 mM NaHCO₃, 25 mM glucose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 1 mM CaCl₂, 5 mM MgCl_2, and 100 μM kynurenic acid) that had been saturated with 95% O₂/5% CO₂. 300 μM thick coronal and sagittal slices (as indicated in the text) were cut (Leica VT1200S; Leica Microsystems, Germany), transferred to a holding chamber and incubated at 35°C for 30 min in modified aCSF (119 mM NaCl, 25 mM NaHCO₃, 28 mM glucose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 1.4 mM CaCl₂, 1 mM MgCl₂, 3 mM sodium pyruvate, 400 μM ascorbate, and 100 μM kynurenic acid, saturated with 95% O₂/5% CO₂) and then stored at 21°C.

For recordings, slices were transferred to a recordings chamber perfused with modified aCSF (119 mM NaCl, 25 mM NaHCO₃, 11 mM glucose, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 1.4 mM CaCl₂, 1 mM MgCl₂, 3 mM sodium pyruvate, 400 μM ascorbate, saturated with 95% O₂/5% CO₂) and maintained at 32–34°C at a flow rate of 2–3 ml min⁻¹. For mIPSC experiments, extracellular Ca²⁺ was replaced with 2 mM Sr²⁺ to desynchronize release. Patch pipettes (resistance = 5–8 MΩ) were pulled on a laser micropipette puller (Model P-2000; Sutter Instrument) and filled with one of the following intracellular solutions: Current-clamp recordings of spike activity used a potassium gluconate-based intracellular solution (137.5 mM potassium gluconate, 2.5 mM KCl, 10 mM HEPES, 4 mM NaCl, 0.3 mM GTP, 4 mM ATP, 10 mM phosphocreatine, pH 7.5). Voltage-clamp recordings for IPSC measurements used a CeMeSO₄-based intracellular solution (114 mM CeMeSO₄, 4 mM NaCl, 10 mM HEPES, 5 mM QX314.Cl, 0.3 mM GTP, 4 mM ATP, 10 mM phosphocreatine, pH 7.5). Alexa Fluor 488 or Alexa Fluor 568 was commonly added to intracellular solution to aid cell visualization and post hoc reconstruction. In some experiments the following were added as indicated in the text: 10 μM CNQX or 5 μM NBQX, 50 μM D-AP5, 10 μM gabazine (Gbz), 0.5 μM tetrodotoxin (TTX). All drugs were obtained from Tocris Biosciences. Intracellular recordings were made using a MultiClamp700B amplifier (Molecular Devices) interfaced to a computer using an analog to digital converter (PCI-6259; National Instruments) controlled by custom written scripts (to be made available at http://dudmanlab.org/) in Igor Pro (Wavemetrics). Photostimulation was carried out using a dual scan head raster scanning confocal microscope and control software developed by Prairie Systems, and incorporated into a BX51 upright microscope (Olympus America).

Individual neurons were patched under DIC optics with a water-immersion 40X objective. Spiking was measured in the cell-attached configuration. The spiking frequency and action potential waveform were used to classify neurons as DA or GABA as described previously (Pan et al., 2013). Upon break in while diffusion of QX314 was allowed time to progress, negative voltage clamp steps were used to measure the hyperpolarization-activated inward (Ih) current. The presence of detectable inward currents was diagnostic for DA neurons. In current clamp recordings lacking QX314 in the internal solution the intracellular spike waveform and spontaneous firing frequency were further used to confirm cell identity. Analysis of postsynaptic currents (direct photocurrents, sIPSC, mIPSCs and evoke IPSCs) and spiking was performed using custom written analysis code in Igor Pro (Wavemetrics). Peak current amplitude was measured as the peak synaptic current relative to the baseline holding current preceding each stimulus. Tonic current amplitude was measured as the peak evoked synaptic current relative to the holding current preceding each train. The conductance (g) underlying IPSCs were calculated from g = IPSC_peak/(V_m − E_CI), where IPSC_peak is the peak amplitude of the IPSC and V_m is the holding voltage. The equilibrium potential of GABA_A current (E_CI) was estimated at −70 mV from the Nernst equation. Rise time constants of postsynaptic currents were measured by finding the 20–80% slope of the rising phase of the stimulus-evoked current. Decay time constant of postsynaptic currents were measured by fitting a single exponential to the decay phase of the stimulus-evoked currents.

Spikes were detected at the threshold of maximum acceleration. Phase plots were constructed by plotting the first derivative of the somatic membrane potential (dV/dt) vs the somatic membrane potential for the average spike waveform. The membrane potential at which phase plot slope reached 10 mV∙ms⁻¹ was denoted the voltage threshold and a linear fit was used to calculate the slope. The perithreshold slope was calculated as the slope of the ‘kink’ defined as the slope of dV/dt for 7 ms after the peak of the perithreshold dV/dt.

For dynamic clamp experiments, individual postsynaptic conductances were generated using IGOR Pro (Wavemetrics), from the sum of two exponentials with rise tau and decay tau derived from measured IPSC and EPSC rise and decay kinetics (rise tau = 0.5 ms, decay tau = 5 ms). The times of individual events were computed by sampling a Poisson distribution in which the rate of IPSC and EPSC events were independently changed from 1000 to 5000 Hz to generate different balances of excitation and inhibition. The convolved waveforms for excitation and inhibition were computed independently and passed to a custom made, digital dynamic clamp (update rate 30 kHz; to be described elsewhere) assuming reversal potentials of −70 mV and 0 mV for inhibition and excitation, respectively.

The optics were designed to minimize the spread of the laser in the x, y dimensions of the focal plane while accentuating the focus in z by underfilling the back aperture of the objective. Stimulation intensity was controlled by pulse duration (0.2–1 ms). Stimulation typically consisted of 9 × 9 and 10 × 14 maps of stimulation sites with independent stimuli being delivered in a pseudo-random (non-neighbor) sequence at an interstimulus interval of ≥150 ms and values reflect the average of 3–4 repetitions of the mapping experiment for each cell. Stimulation strength was modulated by gating the laser at maximal power (473 nm, AixiZ or 488 nm, BlueSky Research) with varying durations using timing signals from an external pulse controller (PrairieView software) and the internal power modulation circuitry of the laser or an external Pockels cell (Conoptics) with indistinguishable results. Wide-field activation of ChR2 was accomplished using blue LED (470 nm, ThorLabs) transmitted through the fluorescence light path of the BX51 microscope. LED intensity and timing were controlled through a variable current source (ThorLabs). Stimulus families (input/output curves) were delivered in a pseudorandom order and repeated 3–10 times per cell.

Analysis was performed using custom written routines for Igor Pro (Wavemetrics) and Matlab R2011a (Math Works). Analysis of the full field photostimulation was performed using standard analysis metrics as described in the text. To attempt to minimize the variability in estimates of short-term plasticity stimulation was performed at the half-maximum stimulus intensity determined by generation of an input–output function at the beginning of the experiment. The analysis of circuit mapping experiments was more complicated and is described briefly below and demonstrated more explicitly in Figure 6. Briefly, averages of 3–10 multisite photostimulation experiments were used in all analyses. The moment of photostimulation was determined by thresholding a photodiode signal positioned in a parallel light path to the stimulation light path. Galvonometer position signals were recovered from the PrairieView software and aligned using a transmitted light laser scanning DIC image of the brain slice. Offline analysis routines automatically detected the orientation of the stimulus grid and applied a rotation to put that grid into 0 rotation orientation. The dorsal and medial edges of the grid were manually annotated and used to flip or further rotate all grids into a common reference frame. The average dimensions of the SN are 1.67 mm wide by 1 mm tall. We found that the average of all grids had an identical (1.67–1) aspect ratio. There was modest variation (∼10%) in the dimensions of the SNr grids. All grids were linearly stretched or compacted to the same mean aspect ratio and cell positions moved accordingly. Qualitatively similar results were obtained in the presence and absence of warping. Full depth maps were generated by convolving the response amplitude at individual stimulus positions with an empirically-estimated Gaussian response function. An isocontour of the resulting image was generated at the half maximum level using the ‘contour’ function supplied by Matlab. The center of mass (COM) was also calculated as the vector average of the Euclidean distance to the stimulus position weighted by response magnitude.

Reconstructions of recorded neurons were derived from two-photon fluorescent image stacks using the semi-automated software generously provided by Ting Zhao (Janelia Farm Research Campus, HHMI) (Zhao et al., 2011). The COM of the dendrites was calculated as the vector average with the weights defined by the width of the dendritic branch segment at the end of the vector position. Data were then loaded into Matlab for display and scaling.

Recordings were performed using either a 32-microwire arrays (CD Neural Technologies) or a 64-channel silicon probe array (NeuroNexus Technologies). Electrode arrays were stereotaxically implanted under anesthesia (isoflurane; 1.5–2.5% in O₂) in mice that had been previously fitted with a plastic head restraint and held in place by a custom head fixation system (Osborne and Dudman, 2014). Design files and details about the manufacture and use of our head restraint system are available online (http://dudmanlab.org/html/rivets.html). Electrode arrays were targeted to the SN of the ventral midbrain (3.0–4.5 mm posterior to bregma, 0.5–2.0 mm lateral to midline and >3.5 mm below the surface of skull). Electrode arrays were maintained in position by a micromanipulator (Sutter Instruments or Scientifica) and connected to the recording systems via a flexible wire coupling and connector. For optogenetic experiments, a 200 μm core multimode fiber (ThorLabs) was affixed near the central recording wires of a 32 channel array or to one shank of the silicon probe array as indicated in Figure 8A. The entire array was slowly lowered in to the midbrain. Following >1 hr of recovery single unit recordings were obtained from alert, but quietly resting mice. Single cell isolation was performed offline using Offline Sorter (Plexon Technologies) and standard techniques. Analysis of stimulus-evoked responses were calculated and presented using Matlab 2011a.

The spike data in Figure 1 are a subset of recording sessions (all sessions with ≥8 simultaneously recorded, putative GABAergic cells) from mice performing an auditory trace conditioning task described in detail previously (Pan et al., 2013). Briefly, mice were trained to consume sweetened water rewards delivered from a port placed on one wall of a behavior box. A speaker placed behind one wall of the box delivered pure tones (10 kHz; 500 ms duration) as conditioned stimuli (CS). Water rewards were delivered 2.5 s following CS onset. This data set included 599 single units recorded across sessions in which 5 to 21 units were recorded simultaneously. For each session, we computed the firing rate of the population of units prior to the onset of the CS and the transient response in the 200 ms following CS onset and subtracted the mean response across all trials. For each trial and all units recorded in a given session, we then computed the population response excepting the ith unit (PRE_population) and the response of the ith unit to the CS (RESP_single). For individual sessions, we determined the correlation between PRE_population and RESP_single for all units in the session. Significance of the correlation was determined using a permutation test. For the entire population, we plotted PRE_population vs RESP_single for all units, all trials. The data were binned into 20 equally spaced bins and mean data for all bins with more than five samples was plotted and fit with a sigmoid function using Igor Pro.

All statistical tests were performed using the statistics package from Matlab 2011a (Math Works). Paired comparisons were performed using the student's t test (all results were also confirmed with a non-parametric ranksum test). Multiple comparisons were performed using ANOVA. Significance was defined as p<0.05 unless otherwise indicated. Averaged data are presented as mean ± standard error of the mean (SEM), unless otherwise specified.

The funder had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Susan Jones, Vivek Jayaraman, Alla Karpova, Albert Lee, Jeff Magee, Gabe Murphy, and Nelson Spruston provided critical feedback at various stages of preparation of the manuscript and progression of the project. We are indebted to the extensive feedback from our colleagues following presentation of this work at internal seminars on the Janelia Farm Research Campus.

JB is a graduate scholar in the Cambridge-Janelia Farm Graduate Program. JTD is a JFRC Group Leader of the Howard Hughes Medical Institute. This work was supported by funding from the Howard Hughes Medical Institute. The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to dudmanj@janelia.hhmi.org.

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#08-36, 11-69) of the Janelia Farm Research Campus. The animal care and use program at Janelia Farm Research Campus is accredited by The Association for Assessment and Accreditation of Laboratory Animal Care, International (AAALACi).

© 2014, Brown et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.