| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
1Department of Bioengineering and 2Department of Psychiatry, University of Pennsylvania, Philadelphia, Pennsylvania
Submitted 26 March 2007; accepted in final form 3 October 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
; Tepper et al. 2004).
The effects of DA on the MSN have been extensively studied and depend on the class of receptor expressed by the cell. Most MSNs coexpress two or more species of receptors, although D1 and D2 receptors (D1R and D2R, respectively) are the most prevalent. D2, D3, and D4 receptors are pharmacologically similar, as are D1 and D5 receptors (Vallone et al. 2000). Research to date suggests that striatal MSNs express primarily either D1Rs or D2Rs (Gerfen et al. 1990; Le Moine and Bloch 1996; Maurice et al. 1999; Surmeier et al. 1996; Yung et al. 1995). The specific effects of D3-, D4-, and D5-receptor activation on MSN channels have not been extensively investigated.
DA modulates several intrinsic and synaptic channels of MSNs, including sodium, potassium, and calcium species, as well as 
-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) and N-methyl-D-aspartate (NMDA) receptors (Nicola et al. 2000). Generally, the direction of modulation (increase or decrease of conductance) for each channel is dependent on the type of DA receptor stimulated. Presumably, the effect of DA modulation on MSN function is the result of a combination of several individual intrinsic and synaptic channel modulations. However, most studies that sought to examine the effects of net DA modulation have failed to generate consistently reproducible, widely accepted results, whereas several other studies have led to the development of hypotheses that are difficult to examine experimentally.
Using a 189-compartment computational model of the nucleus accumbens core MSN (Wolf et al. 2005b), we investigate three previously proposed effects of dopaminergic modulation on the integrative properties of striatal MSNs, as well as one novel hypothesis. The first effect we investigated is the hypothesis that D1R-mediated modulation increases the nonlinearity of MSN cell output in response to synaptic input (Gruber et al. 2003; Hernández-López et al. 1997; Nicola et al. 2000). This hypothesis is based on the observation that MSN cells in in vivo anesthetized preparations oscillate between a hyperpolarized membrane potential (down-state) and a depolarized plateau potential in which the cell may generate action potentials (up-state) (Goto and O'Donnell 2001; Stern et al. 1997; Tseng et al. 2001; Wickens and Wilson 1998; Wilson and Kawaguchi 1996). In this context, DA-enhanced nonlinearity would increase the tendency for the cell to dwell in one of these two states, potentially contributing to gating (O'Donnell and Grace 1995), pattern recognition (Houk 1995), or credit assignment (Kerr and Plenz 2002, 2004) at the network level. The second proposed effect is that DA activation of D1Rs increases striatal output, whereas D2R activation reduces striatal output (Albin et al. 1989; Bamford et al. 2004; Cepeda and Levine 1998; Delong 1990; Gonon 1997; Goto and Grace 2005b). This hypothesis underlies an influential model of basal ganglia function, in which DA regulates the balance of the direct, D1R-expressing, movement-facilitatory pathway and the indirect, D2R-expressing, movement-inhibitory pathway (Albin et al. 1989; Delong 1990). In this model, loss of DA innervation to the striatum, as in Parkinson's disease, biases control of the basal ganglia output toward the movement-inhibiting indirect pathway, resulting in deficits in movement initiation and execution. The third previously proposed effect hypothesizes that DA acts as an input filtering mechanism, suppressing weak inputs while permitting or even enhancing stronger inputs (Cepeda and Levine 1998; Hjelmstad 2004; Nicola et al. 2000, 2004). This could enhance the signal-to-noise ratio of MSN inputs. We also investigated a novel hypothesis, that dopamine might affect the integration time window of the MSN. In this scenario, dopamine could alter the integrative behavior of striatal cells, shifting their behavior in the direction of either integration or coincidence detection. Such an effect could presumably modulate the overall integration of inputs in the corticostriatal network, which may affect the behavioral output of the system.
We found that dopaminergic modulation had no effect on nonlinearity or bistability of the MSN, except at very high, apparently nonphysiological levels of NMDA conductance. DA modulation was able to regulate neuronal excitability and input filtering in the MSN and was also capable of modulating the temporal integrative properties of the MSN. In these cases, the synaptic effects of DA modulation counteracted and overcame the intrinsic effects of DA modulation.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
; Hines and Carnevale 1997). Simulations were performed on a dual 2.5-GHz Power Macintosh G5 (Apple Computers, Cupertino, CA) or in parallel on a 12-node cluster with dual 2.8-GHz processors per node (Penguin Computing, San Francisco, CA). Data analysis was performed using MATLAB (The MathWorks, Natick, MA). Morphology and physiology of the model
The MSN model has been previously described in detail (Wolf et al. 2005b), so we focus on the most salient aspects of the model in this section. Specifics of the model, including channel parameters and cell morphology, are described in more detail in the supplementary material.1 Cell dimensions (dendritic length and diameter, soma size) and passive properties were set to match published values (O'Donnell and Grace 1993; Wilson 1992). The model consists of 189 compartments and includes almost all intrinsic currents known to be expressed in the MSN, including: fast (NaF) and persistent sodium (NaP); fast-inactivating (KAf) and slow-inactivating (KAs) A-type, 4-aminopyridine (4-AP)–resistant, persistent delayed-rectifying (KRP), and inward-rectifying (KIR) potassium currents; large-conductance (BK) and small-conductance (SK) calcium-dependent potassium currents; N- (CaN), P/Q- (CaP/Q), R- (CaR), and L-type (Cav1.2) high-voltage–activated calcium channels; and T- (CaT) and L-type (Cav1.3) low-voltage–activated calcium channels. These channels were distributed throughout the cell in accordance with published data when possible. If not known, channels were assumed to be distributed uniformly throughout the cell unless this resulted in nonphysiological behavior (see Wolf et al. 2005b). All biophysical and kinetic properties for each channel in the model were taken directly from published data (Wolf et al. 2005b). Channel kinetics and voltage dependencies from channels isolated in striatal MSN cells were used when available. Spines were not explicitly modeled, but we accounted for their contribution to membrane area (Segev and Burke 1998). Each tertiary dendrite consisted of 11 compartments to ensure spatial accuracy, and inputs were placed in the middle of the appropriate compartment to acquire second-order correct solutions (Carnevale and Hines 2005).
The internal calcium concentration in a thin shell just inside the cell membrane was tracked for each compartment. BK and SK currents were regulated by calcium influx by N-, P/Q-, and R-type calcium channels, whereas the remaining calcium currents contributed to a separate pool based on published experimental results (Vilchis et al. 2000).
High-calcium retuning of the model
The model was tuned by hand to in vitro data, changing only the conductance and subcellular localization of each of the channels (except for NaF activation/inactivation), which involved an extensive exploration of the parameter space and the selection of the tunings that best fit in vitro data. Because the level of calcium expression in MSN cells can vary significantly (Bargas et al. 1994; Churchill and Macvicar 1998; Hoehn et al. 1993), and DA has been shown to modulate several calcium channels, we created a new version of the model with approximately tenfold higher calcium channel expression than that of the "low-calcium" version presented previously (Wolf et al. 2005b; see Table 1 for comparison). We ran all experiments with both tunings to more fully explore the range of effects of DA on MSNs. The results of experiments on the high-calcium tuning are presented in the main figures and the results using the previous, low-calcium tuning (Wolf et al. 2005b) are included in the supplemental material (Supplementary Figs. 2–5) for reference.
|
). First, calcium channel expression was increased to support calcium spiking. Second, SK current expression was changed from uniform expression throughout the cell to expression in the secondary and tertiary dendrites only, in agreement with studies suggesting that SK channels are expressed primarily in dendritic spines (Faber et al. 2005; Ngo-Anh et al. 2005; Obermair et al. 2003). Third, a rapidly activating, delayed rectifier Kv1.3 (KDR) current was added at uniform conductance throughout the cell. Fourth, the cell was retuned after these changes to match in vitro current-clamp data, including spike shape, frequency response, and subthreshold membrane response of a nucleus accumbens core cell (see RESULTS). This included changing the maximum conductance of many channels and implementing a –2-mV shift of the fast sodium channel (Table 1).
Calcium channel density in the cell was increased for all classes of calcium in comparison to the low-calcium version (Wolf et al. 2005b) (see Table 1). Our previous model was based on studies of acutely dissociated cells (Churchill and Macvicar 1998) and represented an estimate of whole cell calcium current levels. The fact that cellular expression of calcium channels may be largely dendritic (Carter and Sabatini 2004; Day et al. 2006; Kerr and Plenz 2002; Olson et al. 2005) suggests that our previous estimate may represent a cell with relatively low levels of calcium expression. Further, at least some striatal cells exhibit calcium spiking after 4-aminopyridine (4-AP) and tetrodotoxin (TTX) application (O'Donnell and Grace 1993), which was not supported in the previous model. Because cells exhibiting calcium spikes presumably represent cells with relatively high levels of calcium expression, we increased calcium expression approximately tenfold and retuned the rest of the model's parameters to support calcium spiking to verify the robustness of our results. Future experimental work will be necessary to provide a more accurate determination of the range of expression of calcium currents in the adult striatal neuron.
D1 modulation decreases the SK current (by CaN and CaQ reduction) and increases the Cav1.3 calcium current; both of these modulations can enhance the tendency of the model to spike in doublets. To our knowledge, D1 modulation has not been observed to induce doublets, but at the extreme modulation levels that we study here we occasionally observed doublets at high-input levels. To address this, we added a Kv1.3, fast-activating delayed-rectifier potassium current (KDR) throughout the cell (Erisir et al. 1999). It is similar to the KRP current (Nisenbaum et al. 1996) in that it is a tetraethylammonium (TEA)-sensitive, delayed-rectifier current. However, it activates about fivefold faster and at more hyperpolarized potentials, allowing it to suppress doublets without significantly affecting subthreshold activity. We used the Kv1.3 because it has been well characterized in a computational model (Erisir et al. 1999) and behaves similarly to the Kv1.1 and Kv1.6 channels (Coetzee et al. 1999), which have been detected in MSN cells using mRNA assays (Shen et al. 2004). Accordingly, we implemented the Kv1.3 channel as a substitute for the Kv1.1/1.6 channels and suggest that these currents may represent a portion of the TEA-sensitive, delayed-rectifier current in MSNs. The KDR channel was inserted at a uniform conductance of 5.0 x 10–4 S/cm2 throughout the cell.
Dopaminergic modulation
Dopamine has been reported to modulate a number of channels in MSNs. We performed a thorough review of the literature on dopaminergic modulation in both dorsal and ventral striatum, by both D1 and D2 receptors (Table 2). Although historically theories regarding the net effects of dopaminergic modulation on MSNs have been conflicting (Nicola et al. 2000), most studies on DA modulation of individual channels agree on both the direction and magnitude of modulation. The dopaminergic modulation conditions we used for our study are listed in Table 3; these modulations are drawn entirely from the studies listed in the literature review in Table 2. Values in Table 3 are listed as percentage changes of the baseline maximum conductance or as voltage shifts (in millivolts) for the appropriate channel parameter. An in-depth discussion of the rationale behind the modulations we used is provided in the APPENDIX.
|
|
We created four modulation conditions—D1 Intrinsic, D1 All, D2 Intrinsic, and D2 All—based entirely on published results (Table 3). The D1 and D2 Intrinsic conditions consisted solely of intrinsic channel (Na, K, Ca) modulations. The D1 and D2 All conditions consisted of the appropriate intrinsic modulation condition with synaptic (AMPA, NMDA) modulations included. These same modulation conditions were used for all experiments and applied uniformly to the entire cell. We studied D1 and D2 modulations in isolation, but did not investigate the condition in which these receptors may be coexpressed on the same cell.
For the nonlinearity study (Fig. 2) we investigated the possibility that very high levels of NMDA could induce bistability. We created a modulation condition (labeled "Nonphysiological") that included an NMDA conductance of 500% of the baseline, KAf 155% of the baseline, and KIR 350% of the baseline; the increases in KAf and KIR currents were needed to maintain the spike threshold of this condition near that of the unmodulated model.
|
Explicit glutamatergic and GABAergic synapses were modeled using a modified two-state synapse with time constants set to published values (Chapman et al. 2003; Galarreta and Hestrin 1997; Gotz et al. 1997). Each glutamatergic synapse consisted of an AMPA and NMDA pair receiving the same input train. Glutamatergic synapses were placed throughout the dendrites, in accordance with published results (Gracy et al. 1999; Wilson 1992). GABAergic synapses were distributed throughout the cell but clustered near the soma in agreement with physiological data (Fujiyama et al. 2000; Pickel and Heras 1996). AMPA and NMDA channels contributed to the calcium pool not associated with the SK/BK currents: 10% of NMDA current and 0.5% of AMPA current were designated as calcium currents as described in previous studies (Burnashev et al. 1995). AMPA (Myme et al. 2003), NMDA (Dalby and Mody 2003), and 
-aminobutyric acid (GABA) (Nusser et al. 1998) conductance levels were set to published values.
Synaptic inputs were modeled using a modified version of the NetStim object provided in the NEURON package. Each synapse (AMPA/NMDA or GABA) received an independent spike train generated using MATLAB. Each spike train was generated using the following algorithm: first, a constant interspike interval (ISI) train was generated at the desired frequency. Each spike was then pulled anew from a Gaussian distribution centered at the original spike time. The resulting train was then randomly shifted; this process was repeated for each of the 168 total synapses. Input was generated by using a large shift (one ISI) and a large SD (1/4 of the ISI). In our experience, MSNs rarely spike at more than 10 Hz, which corresponds to a maximum physiological input frequency of 1,350 Hz (see RESULTS). Accordingly we did not investigate synaptic input at higher frequencies. The ratio of glutamatergic inputs to GABA inputs was held constant at roughly 1:1 for all simulations (Blackwell et al. 2003).
Simulations
Calcium spiking was examined by simulating the application of 50 µM 4-AP and 2 µM TTX during a current injection of 0.6 nA (O'Donnell and Grace 1993). To do this, we multiplied the KAf conductance by 0.9 (Song et al. 1998), KAs by 0.4 (Russell et al. 1994), KRP by 0.8 (Nisenbaum et al. 1996), and KDR by 0.6 (Coetzee et al. 1999). The NaF conductance was adjusted to match in vitro behavior, which required multiplying the baseline conductance by 0.25 (Fig. 1B).
|
|
; Myme et al. 2003; Nusser et al. 1998).
|
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
The model is a stylized representation of the nucleus accumbens core medium spiny neuron, with 189 compartments, branched dendrites, and explicit synapses (Fig. 1A, inset). The model was tuned to match the in vitro response to current injection of a nucleus accumbens core MSN isolated from an adult rat (Fig. 1A, left). Because the level of calcium expression in MSN cells can vary significantly (Bargas et al. 1994; Churchill and Macvicar 1998; Hoehn et al. 1993), and DA has been shown to modulate several calcium channels, we created a new tuning of the model with approximately tenfold higher calcium channel expression than that of the "low-calcium" version developed previously (Wolf et al. 2005b; see Table 1 for comparison). With this increased level of calcium, the model was able to approximate reports of calcium spiking in MSN cells after TTX and 4-AP application (Fig. 1B). We ran all experiments with both tunings to more fully explore the range of effects of DA on MSNs. We found that although the effects of DA on MSNs were generally more appreciable in the high-calcium tuning, the overall results of the experiments were the same. Results of experiments on the high-calcium tuning are presented in the main figures and results using the previous, low-calcium tuning are included in the supplemental material (Supplementary Figs. 2–5) for reference.
The high-calcium tuning of the model matched the frequency–current (F–I; gray line in Fig. 1C, inset) behavior of a representative in vitro MSN (Fig. 1C, inset, thick black line). This version of the model (Fig. 1C, gray trace) is also within the SD of the averaged current–voltage (I–V) response of seven MSNs (Fig. 1C, solid black trace). Interestingly, the high-calcium tuning (Fig. 1C, inset, gray trace) matches the F–I response of the representative cell better than the low-calcium version (Fig. 1C, inset, dashed black line), but has slightly more outward rectification (Fig. 1C, gray line) at depolarized potentials than the low-calcium version (Fig. 1C, dashed black line). This is primarily the result of redistribution of the SK current from uniform expression throughout the cell to only the secondary and tertiary dendrites.
Modulation conditions
Following a review of published D1R- and D2R-mediated modulations of MSN ionic channels (Table 2), we created four modulation conditions for use in this report (Table 3). These conditions represent the expected effects of solely D1 or solely D2 modulation of MSN channels (see METHODS). As much as possible, we used maximum reported levels of modulation because this should most clearly demonstrate the effects of D1R- and D2R-mediated modulation on MSN behavior. The D1 Intrinsic and D2 Intrinsic conditions include modulations of intrinsic channels only, whereas the D1 All and D2 All conditions account for synaptic modulations as well as intrinsic modulations (Table 3).
The D1 Intrinsic condition resulted in three primary changes: a delayed first spike, a relatively short interspike interval between the first and second spikes, and a reduced number of action potentials for a given current injection (Fig. 1D, middle). These observations agree with in vitro data (Hernández-López et al. 1997). The delay to first spike was primarily the result of increasing the KIR current. The relatively short interspike interval between the first and second spikes after the start of the current injection resulted primarily from the more hyperpolarized activation of the Cav1.3 current. The D2 Intrinsic condition resulted in deeper interspike troughs (Fig. 1D, right) and increased spiking, in agreement with previous observations in vitro (Akaike et al. 1987).
D1R-mediated modulation and nonlinearity in the MSN
It was previously proposed that D1R activation enhances nonlinearity in the MSN's response to synaptic input, perhaps even inducing bistable membrane behavior (see DISCUSSION for our definitions of bistability, nonlinearity, and hysteresis) (Gruber et al. 2003; Hernández-López et al. 1997; Nicola et al. 2000). We examined this hypothesis in several ways. First, we examined the model's response to synaptic input applied at equally spaced increments from low to high frequency (Fig. 2A, inset). In this manner, we were able to examine whether the cell exhibited nonlinearity in response to the linear steps in input frequency (Fig. 2A; traces represent the average membrane potential for 18 trials). For clarity, we also plotted the average somatic potential of the model over the last 50 ms of each trace in Fig. 2A against the corresponding synaptic input frequency (Fig. 2B). Neither the unmodulated, D1 Intrinsic, or D1 All conditions exhibited any nonlinearity (Fig. 2, A and B). To see whether two distinct populations of membrane potential could be discerned across all 18 trials of 21 input frequencies, we created histograms of the membrane potential for each condition (Fig. 2C). Neither the unmodulated, D1 Intrinsic, nor the D1 All conditions appeared to express two appreciably distinct states for the membrane potential.
Following a recent study describing NMDA-dependent bistability in an MSN-like two-compartment computational model (Kepecs and Raghavachari 2007), we examined whether very high levels of NMDA might be able to induce nonlinearity or bistability (Fig. 2, A–C, bottom). We found that an NMDA:AMPA ratio >2.5:1 was required to induce two substantially distinct membrane potential states (Fig. 2C, bottom); this value appears to be outside the range of physiological ratios (see DISCUSSION; Myme et al. 2003). NMDA:AMPA ratios between 1:1 and 2.5:1 were not capable of inducing this behavior (data not shown).
We examined the response of the model for increased hysteresis after D1R modulation. To do this we ramped the frequency of synaptic input up and back down symmetrically (Fig. 2D, left) and averaged the membrane response across 40 trials of different inputs. Plotting the averaged cell voltage versus time on the down-slope of the ramp against the averaged cell voltage versus time on the up-slope of the ramp allows the paths to be compared (Fig. 2D). A bistable cell would have a highly asymmetric response to a symmetric input ramp, staying in the up-state even after removal of the input. The model MSN exhibits a small amount of asymmetry in the unmodulated condition, indicating that it is somewhat hysteretic in this condition (Fig. 2D, middle). D1 All modulation did not noticeably enhance this asymmetry (Fig. 2D, right). We previously showed that hysteresis in response to current injection is minimal compared with hysteresis in response to synaptic input (Wolf et al. 2005b); this is also the case for the D1-modulated cell in response to current injection (data not shown).
Taken together, these data suggest that D1 modulation does not enhance nonlinearity, except at potentially nonphysiological levels of NMDA. We found no evidence supporting MSN bistability under any modulation condition.
Excitatory/inhibitory properties of DA modulation
DA has long been hypothesized to be either excitatory or inhibitory based on the type of receptor activated. We focus on the hypothesis that D1R activation increases MSN activity, whereas D2R activation decreases MSN activity because these ideas underlie the most influential model of basal ganglia function to date (Albin et al. 1989; Delong 1990). Contrary to this idea, we found that the D2 Intrinsic condition increases spiking in response to current injection at all levels (Fig. 3A, solid green line) relative to the unmodulated condition (Fig. 3A, back line). Interestingly, the D1 Intrinsic condition is neither wholly excitatory nor inhibitory, but rather increases the slope of the F–I curve (Fig. 3A, red line). A change in gain response, and possible induction of nonlinearity, has been proposed as an important effect of D1 modulation on MSN function (Gruber et al. 2003; Hernández-López et al. 1997; Nicola et al. 2000). This has specifically been proposed to be the result of increases in KIR and Cav1.3 (Gruber et al. 2003). We found that KIR and Cav1.3 modulations contributed to the gain change, as did the CaP/Q and CaN modulations. The effects of D2 Intrinsic modulation are critically dependent on NaF modulation. As discussed in the APPENDIX, D2R activation may either increase or decrease sodium current (by conductance changes and inactivation curve shifts), depending on the recording method (Surmeier et al. 1992). We used a net increase in sodium (10% increase in conductance with a +3-mV shift of the inactivation curve) to agree with previous descriptions finding D2R activation to be mildly excitatory in response to current injection (Akaike et al. 1987; Higashi et al. 1989; Yim and Mogenson 1988). However, decreasing the net sodium current can substantially decrease spiking (not shown). Accordingly, the net effect of D2 Intrinsic modulation can be either excitatory or inhibitory, depending on the magnitude and direction of sodium modulation.
We compared the effects of DA modulation in our model to previously published reports. Hernández-López et al. (1997) reported that holding the cell at –82 mV (resting membrane potential) and activating D1 receptors (0.3-nA current injection, 300-ms duration) decreases spiking 62.5%. In our model, holding the cell at –87 mV, injecting 0.3-nA current for 300 ms, and simulating D1 modulation decreases spiking 20%. These authors also showed that holding the cell at –57 mV and activating D1 receptors increases spiking 34%, whereas in our model, holding the cell at –57 mV and simulating D1 modulation increase spiking 33%. Importantly, increasing the NaF modulation to a 17% reduction in conductance can match the 60% reduction in spiking reported by Hernández-López et al. (1997). Akaike et al. (1987) reported excitation after DA application that is sensitive to D2 receptor blockade. With a 0.3-nA current injection, they found that spiking is increased from 0 spikes per 300 ms to 5 spikes per 300 ms. In our model, D2 modulation increases spiking from 0 to 2 spikes (0.2375-nA injection). Accordingly, D1 excitation is the same in our model as in these experiments, whereas D1 inhibition and D2 excitation are somewhat more mild.
To explore the effects of D1R- and D2R-mediated modulation on the MSN response to synaptic input, we calculated the model spiking frequency versus synaptic input frequency (Fig. 3B; also see Supplementary Fig. 1 for examples of model response). We found that the model spikes in response to synaptic input frequencies in the range of 850–1,400 Hz, which corresponds very well with other reports of frequency values for "up-state generation" (806 ± 188 Hz: Blackwell et al. 2003; 
600 Hz: Wilson 1992). As seen in the current injection experiments, the D1 Intrinsic condition induced a change in gain of the spiking response (Fig. 3B, solid red line) compared with the unmodulated response (Fig. 3B, black line) to synaptic input. At low synaptic input frequencies (<1,200 Hz), D1 Intrinsic modulation was inhibitory, whereas at frequencies >1,200 Hz D1 Intrinsic modulation became slightly excitatory. The D1 All condition significantly increased the spike frequency at all synaptic input frequencies (Fig. 3B, dashed red line) and slightly increased the change in gain beyond that in the D1 Intrinsic condition. Analogous to the current injection experiments, the D2 Intrinsic condition increased excitability of the cell in response to synaptic input (Fig. 3B, solid green line). The D2 All condition reduced the excitability of the model in response to synaptic input (Fig. 3B, dashed green line). Thus our results indicate that the net effect of D1R modulation is to excite MSN cells, whereas the net effect of D2R activation is to inhibit MSN cells in response to synaptic input.
Effects of dopamine MSN filtering of synaptic inputs
It has been suggested that DA might change the way that MSNs filter synaptic inputs, enhancing large synaptic inputs while filtering out smaller synaptic inputs (Bamford et al. 2004; Nicola et al. 2004). To investigate this possibility, we stimulated a tertiary dendrite with different-sized glutamatergic inputs while the cell was receiving synaptic input (1,050 Hz). We stimulated at a position located halfway out along the dendrite (Fig. 4A, inset). Evoked synaptic potential input sizes were expressed relative to a baseline conductance level from previous reports (Dalby and Mody 2003; Myme et al. 2003). Varying input size from 0.2- to 2.2-fold baseline revealed an approximately linear relationship between the size of the input and the measured somatic depolarization (Fig. 4A, black line). Neither D1 Intrinsic (Fig. 4A, solid red line) nor D2 Intrinsic (Fig. 4A, solid green line) modulation changed the linearity of this relationship significantly. Both conditions increased the magnitude of the depolarization measured at the soma, with larger inputs more enhanced relative to smaller inputs. Including synaptic effects diminished, but did not reverse, these relationships (D1 All, dashed red line; D2 All, dashed green line). The boost of inputs by D2 modulation was the result of an increase in sodium current; the boost by D1 modulation was mostly the result of decreased SK current (not shown).
We also examined the change in depolarization at the soma as a synaptic input was moved to progressively more distal locations on a tertiary dendrite (Fig. 4B, inset). D2 Intrinsic modulation again increased the magnitude of the depolarization measured at the soma for all distances (Fig. 4B, solid green curve) compared with the unmodulated condition (Fig. 4B, black curve), whereas D1 Intrinsic modulation appeared to have no effect (Fig. 4B, solid red curve). Including synaptic effects diminished this increase in magnitude for D2 modulation (D2 All, Fig. 4B, dashed green curve), but had no effect on D1 modulation (D1 All, Fig. 4B, dashed red curve). Based on these results, we conclude that DA modulation can preferentially enhance large inputs.
Dopamine and temporal integration properties of the MSN
Previous studies have shown that MSN somatic voltage appears to closely reflect integrated synaptic input over an approximately 50-ms timescale (Wolf et al. 2005b). We therefore investigated the possibility that DA modulation might change the integration time window of the MSN. We simulated a prolonged, nonspiking up-state and compared the resulting MSN somatic voltage (Fig. 5A, top, black trace) to the synaptic input frequency calculated using a sliding window of various widths (Fig. 5A, top, red trace). On calculating the covariance of the two (see METHODS), we found that the unmodulated MSN correlates best with input frequency binned over 50 ms (Fig. 5B, black trace). D1 Intrinsic modulation (Fig. 5A, middle) decreased the bin size with maximum correlation to 40 ms (Fig. 5B, solid red trace), whereas D2 Intrinsic modulation (Fig. 5A, bottom) increased the bin size with maximum correlation to 60 ms (Fig. 5B, solid green trace). Combining synaptic modulations with intrinsic modulations, as in the D1 and D2 All conditions, reversed this trend. D1 All modulation (Fig. 5B, dashed red trace) increased the bin size with maximum correlation to 60 ms, whereas D2 All modulation (Fig. 5B, dashed green trace) returned the bin size with maximum correlation to 50 ms.
We also investigated whether DA modulation affected the response of the model cell to inputs of different intensities and different degrees of coherence. First, we stimulated different numbers of glutamatergic synapses, randomly distributed throughout the cell, every 200 ms (Fig. 5C, inset), and calculated the probability of the stimulation eliciting a spike (see METHODS). Our results indicate that for the same stimulation (i.e., same set of synapses), D1 Intrinsic modulation (Fig. 5C, solid red trace) decreases the probability of the model spiking and D2 Intrinsic modulation (Fig. 5C, solid green trace) increases the probability of the model spiking in response to the stimulation, relative to the unmodulated condition (Fig. 5C, black trace). Combining synaptic modulations with intrinsic modulations reverses this trend. D1 All modulation (Fig. 5C, dashed red trace) increases the probability of the model spiking and D2 All modulation (Fig. 5C, dashed green trace) decreases the probability of the model spiking in response to the stimulation. Next, we examined how coherent these inputs needed to be to elicit a spike by varying the duration of the time window in which a given number of synapses was activated (Fig. 5D). As expected, increasing the duration of the time window generally decreased the probability of the model spiking in response to the stimulation. However, in general, D2 Intrinsic modulation increased the ability of the model to spike in response to wider stimulation windows (Fig. 5D, solid green line), whereas D1 Intrinsic modulation decreased this ability (Fig. 5D, solid red line), relative to the unmodulated condition (Fig. 5D, black line). Including synaptic modulations reversed this trend, with D1 All (Fig. 5D, dashed red line) modulation generally improving the ability of the cell to integrate temporally dispersed input and D2 All (Fig. 5D, dashed green line) modulation impairing this ability. These experiments suggest that DA modulation can alter the temporal integration properties of the MSN.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONAPPENDIXGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
Historically, studies of the net effects of dopamine on MSN function have yielded a variety of results that are either conflicting or difficult to combine into a unifying hypothesis for dopaminergic modulation (Nicola et al. 2000), even though studies on individual channel modulations are mostly consistent (Table 2). We used three approaches with our model in an attempt to integrate previous experimental results on the net effects of DA modulation. First, we used parameters derived from maximal physiological levels of dopaminergic modulation (Table 3) based strictly on previously published results (Table 2). Second, to be as objective as possible, we used modulation conditions that included all reported channel modulations for D1 or D2 receptors, rather than subjectively selecting certain channel modulations for inclusion in each subtype. Third, we applied modulation conditions to both a low-calcium (Wolf et al. 2005b) and a high-calcium tuned version of the MSN model. These tunings represent significantly different cells, yet both give the same results, suggesting that our findings may be generally true for MSN cells. Still, it will be necessary for future studies to thoroughly and systematically examine the full range of potential modulation conditions, applied to a number of distinctly tuned MSN cells, to ensure that this is the case.
We examined DA modulation of MSNs by breaking DA action down into either D1- or D2-receptor–mediated effects. Some MSNs may coexpress D1 and D2 receptors (Aizman et al. 2000; Surmeier et al. 1992, 1996). However, because D1 and D2 receptors mediate opposite effects on nearly every MSN channel (Table 3), we studied D1R- and D2R-mediated effects in isolation to most clearly illustrate their actions. It is possible that we were not able to capture the full effects of DA on the MSN by looking at solely D1R- or D2R-mediated effects—at least one study in ventral striatum has shown that D1 and D2 receptors on MSNs must be coactivated to modulate KAs (Hopf et al. 2003). However, the relative paucity of information on coactivation-dependent modulations prevents us from further addressing this possibility.
To examine the potential effects of various modulations, we created conditions incorporating results from studies in both dorsal and ventral striatum and applied these modulations to a model of the ventral striatal medium spiny neuron. Some reports have suggested that DA modulates ventral and dorsal cells differently (Nicola et al. 2000). However, a review of the literature on dopaminergic modulation of MSNs revealed that almost all studies in both dorsal and ventral striatum find consistent direction and magnitude of modulation for individual channels (see APPENDIX and Table 2 for more detail). Accordingly, we suggest that the data presented here are true for MSN cells from both dorsal and ventral striatum. Further study will be necessary to determine whether net DA modulation of MSNs is significantly different in the core, shell, and dorsal striatum.
DA has been shown to modulate GABAergic receptors in the striatum (Centonze et al. 2002; Flores-Hernández et al. 2000, 2002; Guzman et al. 2003; Hernández-Echeagaray et al. 2006; Hjelmstad 2004; Pennartz et al. 1992; Taverna et al. 2005). We do not include GABAergic modulation in this study for two reasons. First, as we reported previously (Wolf et al. 2005b), changes in the GABA conductance do not have an appreciable effect on the function of our model because of the way in which we have simulated GABAergic input as a desynchronized signal. Potentially, GABA inputs need to be synchronized to have a strong effect on MSN activity (Tepper et al. 2004). Second, DA appears to affect interneuron inputs and MSN collaterals differently; we do not distinguish between these two GABA sources in the present model. However, future studies will explore the effects of DA modulation of GABA on striatal network function.
Effects of D1R-mediated modulation on MSN nonlinearity and bistability
Bistability is the ability of a cell to remain in either of two membrane potentials indefinitely in the absence of external input. Because these states act as attractors, a bistable cell will exhibit a strongly nonlinear steady-state membrane potential profile in response to linear increases in synaptic input frequency (Fig. 2, A and B) because it will reside almost solely in these two states and switch abruptly between them. A bistable cell will also exhibit pronounced hysteresis, which is the characteristic of a cell to follow different voltage paths between two states depending on the direction of state transition (Fig. 2D)—this is caused by the tendency of the cell to maintain its current state as long as possible (Booth et al. 1997). In contrast, bimodality refers to a cell exhibiting two ranges (or modes) of membrane potentials (Fig. 2C), regardless of the mechanism (intrinsic properties or synaptic input). A bimodal cell may or may not also exhibit nonlinearity and hysteresis. The concept that MSNs might be bistable has been highly influential on functional theories of the basal ganglia. Several models of striatal/basal ganglia function have been built on the idea that striatal MSNs are inherently bistable, or become bistable after D1 modulation. Two of these models posit the basal ganglia as an action-selection mechanism, in which striatal MSN bistability enables the cortico-basal ganglionic loop to function as a pattern detector (Beiser and Houk 1998) or enhances the duration and intensity of striatal activity (Gruber et al. 2003). Another important model suggests that the ventral striatum gates cortical output based on limbic input, with MSNs transmitting information in the up-state but not in the down-state (Grace 2000; O'Donnell and Grace 1995). The concept that MSNs are intrinsically bistable arose from intracellular recordings performed in vivo in anesthetized animals, in which MSNs oscillate between spiking up-states and quiescent down-states with sharp transitions (Goto and O'Donnell 2001; Stern et al. 1997; Tseng et al. 2001; Wickens and Wilson 1998; Wilson and Kawaguchi 1996).
We did not observe bistability in our model, under any modulation condition (Fig. 2). The only condition in which we observed nonlinearity or bimodality was with a very high level of NMDA conductance [fivefold the baseline NMDA conductance, or an NMDA:AMPA (N:A) ratio of 2.5:1]. This observation agrees with a recent study in a two-compartment, MSN-like model that observes bimodality with a 4:1 N:A ratio (Kepecs and Raghavachari 2007). The exact N:A ratio in corticostriatal synapses can vary, but appear to remain below 1:1 in normal animals (Beurrier and Malenka 2002; Li et al. 2004; Popescu et al. 2007; Thomas et al. 2001), increasing to as high as 1.5:1 in cocaine-treated animals (Thomas et al. 2001). Studies of synapses onto cortical pyramidal neurons have found N:A ratios ranging from 0.2:1 to as high as 7.7:1 (Myme et al. 2003). However, the authors of this review note that studies of the N:A ratio using extracellularly evoked synaptic potentials/currents reliably report higher N:A ratios (probably because of suboptimal space clamping), as do studies using younger animals (p1–p15). Upon excluding these studies, the N:A ratios range from 0.2:1 to 1.2:1 (Myme et al. 2003). Taken together, these reports suggest that N:A ratios 
1.5:1 (which was not enough to induce bimodal behavior in our model) may not naturally occur in MSNs. We therefore suggest that NMDA-induced bistability is not likely to occur in MSNs under normal conditions. Because our model and others have suggested that the N:A ratio is extremely important in defining the behavioral response of MSNs to synaptic input and, because changes in this ratio have been hypothesized to occur in various disease states, it is crucial to further examine these ratios in adult animals in various areas of the striatum.
Although MSNs demonstrate bimodal membrane potentials under certain conditions, it is possible that this ability may not be functionally significant for striatal processing in the awake state. Recent in vivo intracellular recordings of MSNs found that although MSNs exhibited bimodal behavior during slow-wave sleep and anesthesia, the membrane potential distribution in the awake rat was clearly unimodal and centered at –61 mV (Mahon et al. 2006). In vivo studies in our lab have also indicated that in the awake state, hippocampal and cortical inputs to the accumbens sum sublinearly (Wolf et al. 2005a), not supralinearly, as would be expected if the bimodal membrane potential was responsible for a gating effect in the awake animal.
Effects of dopamine modulation on MSN excitability, filtering, and temporal integration
One influential hypothesis of basal ganglia function proposes that D1R activation excites MSN cells in the D1R-expressing, movement-facilitating, direct pathway, whereas D2R activation inhibits MSN cells in the D2R-expressing, movement-suppressing, indirect pathway (Albin et al. 1989; Delong 1990). We found that D1 modulation of intrinsic properties changed the slope of the frequency–current relationship, so that D1 intrinsic modulation could be inhibitory or excitatory, depending on the current injection amplitude (Fig. 3); simultaneous D1 modulation of both intrinsic and synaptic modulations was solely excitatory. Conversely, we found that D2 modulation of intrinsic properties was excitatory, although this was dependent on the direction and magnitude of sodium modulation, which is not precisely known; including synaptic modulations caused D2 modulation to be inhibitory. Accordingly, our results demonstrate that D1R-mediated modulation increases the activity of MSNs, whereas D2R-mediated modulation decreases the activity of MSNs (primarily as the result of synaptic modulations). This agrees with the classical model of the basal ganglia and supports the suggestion that loss of DA input to the striatum, as in Parkinson's disease, could alter the activity levels of D1- and D2-expressing MSNs and their downstream projections.
It has been suggested that dopamine acting at D1 receptors may differentially affect MSNs based on the current membrane potential, with hyperpolarized MSNs being inhibited and depolarized neurons being excited by D1R activation (Hernández-López et al. 1997; Nicola et al. 2000; Pacheco-Cano et al. 1996). We found that D1R activation changed the gain of the response of the MSN to current injection, so that at smaller current injections the MSN could be inhibited relative to the unmodulated state, whereas at larger current injections the cell could be excited (Fig. 3A). This gain change also occurred in response to synaptic input (Fig. 3B) and for both tunings of the cell (Supplementary Fig. 3). Accordingly, our findings support the possibility that D1R modulation of intrinsic properties might differentially affect the excitability of MSNs, exciting some while inhibiting others, based on the level of input to each MSN. It is important to note that simultaneous D1R modulation of both intrinsic and synaptic properties would be expected to solely cause MSN excitation, in agreement with the classical model of the basal ganglia.
It was previously proposed that dopamine may filter inputs to the MSN by enhancing the contrast between large and small synaptic inputs (Nicola et al. 2004). We found that DA modulation of intrinsic MSN properties enhanced the propagation of synaptic inputs to the soma, with larger synaptic inputs enhanced more than smaller ones (Fig. 4). However, including synaptic modulations diminished this effect. Still, the D2R-mediated up-regulation of sodium in the dendrites during D2 Intrinsic modulation should also increase backpropagation of action potentials into the dendrites, which would presumably affect the induction of synaptic plasticity in MSNs. However, these effects appear to be less significant than the effect of DA on MSN excitability.
We hypothesized that DA might affect the temporal integration properties of MSNs. Our results indicate that D1 modulation of intrinsic properties decreases the integration time window of the MSN, whereas D2 modulation of intrinsic properties increases this window. However, including synaptic effects again reverses this relationship, with the D1 All condition increasing the integration window and the D2 All condition returning the integration window to the unmodulated value. Regardless, the MSN appears to integrate synaptic inputs over a time window of 40 to 60 ms, suggesting that it functions more as an input integrator than as a classical coincidence detector. Given not only the very large number of glutamatergic inputs (5,000–15,000) each MSN receives, but also the ability of each MSN to respond to a minimum of 100 distinct cellular ensembles (Wolf et al. 2005b), this suggests that the MSN might function as a pattern detector, integrating and classifying patterns of cortical/subcortical inputs as part of the corticostriatal action selection mechanism. In this sense, dopaminergic modulation of the temporal integration window of MSNs might subtly regulate striatal integration of input from cortical ensembles.
Lack of synergy between intrinsic and synaptic effects
In all of the preceding cases, the synaptic effects of DA modulation counteracted the effects of DA modulation of intrinsic properties. Whether intrinsic and synaptic modulations always occur simultaneously is not known. However, it seems highly likely that both would occur at the same time because dopamine is a highly divergent signal, with one DA neuron targeting approximately 400 MSNs, 30% of DA neurons responding similarly to any given novel or salient event, and with the apparent ability of DA to diffuse out of the synaptic cleft (Arbuthnott and Wickens 2006; Hyland et al. 2002; Schultz 1998). One possibility is that the intrinsic and synaptic modulations counteract each other to maintain precisely balanced regulatory control over the MSN's integrative properties.
Another possibility is that both tonic and phasic dopamine release differentially affect intrinsic and synaptic properties of the MSN. Dopamine neurons exhibit two basic activity modes—regular spiking and phasic bursting, which might initiate different downstream signaling mechanisms (Grace 1991). In this light, tonic spiking by DA neurons has been shown to maintain extrasynaptic DA levels at nanomolar concentrations, whereas bursting by DA neurons can boost DA concentration to micromolar levels, but possibly only within the synaptic cleft (Arbuthnott and Wickens 2006; Floresco et al. 2003; Phillips and Wightman 2004). Tonic, low-concentration DA levels might primarily influence intrinsic modulations at extrasynaptic sites, whereas phasic, high-concentration DA levels might control synaptic modulations. In this sense, the intrinsic effects of DA modulation could dominate during regular spiking, whereas the synaptic effects might override the intrinsic effects after short-term bursts by DA cells. Nonetheless, the intrinsic effects of DA modulation appear to be much less significant than the synaptic effects of DA modulation because the synaptic effects tended to dominate the net effect of DA modulation when the two were combined.
Possible D1-mediated regulation of calcium in dendritic spines
With the exception of its regulation of excitability, DA's effects on the integrative properties of the MSN appear to be surprisingly weak. Specifically, although we report that dopaminergic modulation can lead to changes in input filtering and temporal integration properties of the MSN, these effects do not appear to be very significant, at least at the single-cell level. Further, given that distinct MSN cells will inevitably express different levels and combinat
40 ms following the center of the stimulation window (inset). In general, D1 Intrinsic (solid red) modulation decreases the ability of the model to integrate synaptic inputs over larger time windows, whereas D2 Intrinsic (solid green) modulation increases its ability to do so relative to the unmodulated condition (black). Including synaptic modulations reverses this trend, with D1 All (dashed red) modulation generally improving the ability of the cell to integrate temporally dispersed input and D2 All (dashed green) modulation impairing this ability. These experiments suggest that DA modulation can alter the temporal integration window of the MSN.