Regional Variation in Striatal Dopamine Spillover and Release Plasticity

Abstract

Recent optical observations of dopamine at axon terminals and kinetic modeling of evoked dopamine responses measured by fast scan cyclic voltammetry (FSCV) support local restriction of dopamine diffusion at synaptic release sites. Yet, how this diffusion barrier affects synaptic and volume transmission is unknown. Here, a deficiency in a previous kinetic model’s fitting of stimulus trains is remedied by replacing an earlier assumption that dopamine transporters (DATs) are present only on the outer side of the diffusion barrier with the assumption that they are present on both sides. This is consistent with the known distribution of DATs, which does not show obvious DAT-free zones proximal to dopamine release sites. A simultaneous multi-fitting strategy is then shown to enable unique model fits to sets of evoked dopamine FSCV responses acquired in vivo or in brain slices. This data analysis technique permits, for the first time, the calculation of the fraction of dopamine which spills over from what appears to be the perisynaptic space, as well as other parameters such as dopamine release, release plasticity, and uptake. This analysis shows that dopamine’s diffusion away from its release sites is remarkably hindered (τ = 5 s), but dopamine responses are rapid because of DAT activity. Furthermore, the new analysis reveals that uptake inhibitors can inhibit dopamine release during a stimulus train, apparently by depleting the releasable pool. It is suggested that ongoing uptake is critical for maintaining ongoing synaptic dopamine release, and that the previously reported and also herein claimed increase of the initial dopamine release of some uptake inhibitors might be an important mechanism in addiction. Finally, brain mapping data reveal that the diffusion barrier is conserved, but there are variations in perisynaptic uptake, volume transmission and release plasticity within the rat striatum. Therefore, an analysis paradigm is developed to quantify previously unmeasured features of brain dopaminergic transmission and to reveal regional functional differences among dopamine synapses.

Graphical Abstract

INTRODUCTION

Dopamine is an important neuromodulator¹ that controls reward², learning³, memory⁴ and movement⁵. Pathologies of dopamine transmission include drug addiction⁶, schizophrenia⁷ and Parkinson’s disease⁸, which are all significant ongoing medical treatment challenges. Both improved basic understanding of dopamine transmission and more refined strategies for treating medical conditions can be achieved by improving our understanding of the mechanistic nature of dopamine transmission at synapses within the brain. Dopamine is known to engage in both synaptic⁹ and volume^10–11 transmission; synaptic transmission⁹ refers to chemical signals actuated upon receptor targets by only one axon terminal in close proximity, while in volume transmission^10–11, the chemical signals from axon terminals diffuse so that receptors at a distance from the release site can be activated. However, how and to what extent these two transmission modes are implemented throughout the brain is incompletely understood.

Prior work produced a simple Restricted Diffusion (RD) kinetic model^12–14 (Equations 1 and 2) that revealed that after release, dopamine is spatially restricted within an inner extracellular compartment for a period of at least hundreds of milliseconds prior to mass transport into the outer compartment where a carbon fiber electrode can detect the transmitter. This was surprising because the exit time of dopamine from the synapse was expected to occur in microseconds by unhindered diffusion from release sites.^15,16 The conclusions of the RD model have in our view recently been supported by fluorescent dopamine sensors^17–19, which have collected data that is consistent with sampling dopamine signals from an inner compartment. Specifically, fluorescent protein-based sensors expressed on DA neurons report one to two orders of magnitude higher DA concentrations than that reported away from release sites by FSCV¹⁷, which is consistent with optical DA measurements being made at least partially in the previously hypothesized inner compartment for DA transmission. Furthermore, nanotube-based fluorescence sensors¹⁹ suggest that released DA is localized to cellular structures instead of diffusing throughout the optically sampled tissue volume, which is also consistent with an inner compartment defined by a physical diffusion barrier around perisynaptic spaces. Therefore, these observations suggest a biophysical location for the kinetic RD model-derived inner compartment - perisynaptic volumes around sites of dopamine release.

The Restricted Diffusion Model (4 Parameter, 4P)

$$\frac{dD{A}_{ic}}{dt}=R_p\cdot f\cdot e^{-k_{R}t}-D{A}_{ic}\cdot k_T$$ Eqn 1

$$\frac{d{[DA]}_{oc}}{dt}=\frac{D{A}_{ic}\cdot k_T}{V_{oc}}-{[DA]}_{oc}\cdot k_U$$ Eqn 2

In the RD model, dopamine is released by stimulus pulses into an “Inner Compartment (IC)”, after which it subsequently is transferred by unidirectional mass transport into the “Outer Compartment (OC)”. Dopamine uptake occurs within the OC. We define the terms below:

R_p is the amount of dopamine released from an arbitrary volume for the initial stimulus pulse

f represents the stimulus frequency (Hz)

k_R is a plasticity factor that affects release (R_p) over time during a stimulus,

k_T is a first order rate constant that controls mass transport of DA from the IC to the OC

k_U is the first order rate of DA reuptake from the OC.

However, the RD model displays a problematic feature concerning extracted biophysical parameter values: the obtained parameter values all change when stimulus trains of different lengths are used at a given dopamine FSCV recording site.^12,13 This suggests that the previous kinetic model (RD Model, 4P) is either missing important terms, lacks appropriate terms to describe stimulation-dependency of these terms, or both. One term that the previous model lacked was a term for dopamine transporter mediated reuptake within the Inner Compartment, which by definition contains the site of DA release. Striatal dopamine transporters (DATs)^20–22 are transmembrane proteins that transport dopamine into the cytosol by extracting energy from the Na⁺ membrane gradient. This active transport is the primary mechanism of elimination of dopamine from the extracellular space. Prior work suggests that DATs are present proximal to DA release sites, and there is no clear evidence that zones free of DAT are systemically present throughout the striatum.^20–22 This suggests that including an uptake term within the Inner Compartment might appropriately remedy the fitting deficiencies in the prior model. We found that we were able to resolve the problem of parameter values inappropriately changing with stimulus length by incorporating an uptake term within the model’s Inner Compartment. With the incorporation of a term to describe perisynaptic DATs into the RD model, brain mapping and drug data are analyzed to reveal new insights into the mechanistic functioning of dopamine terminals with an emphasis on the roles of the perisynaptic diffusion barrier, release plasticity and spillover, which controls the ratio of volume transmission^10,11 to synaptic transmission^9,23.

RESULTS & DISCUSSION

Dopamine uptake before and after passage through the perisynaptic barrier

The original RD model^12,13 is shown in Eqn 1 and 2 and Figure 1A. This version of the RD model contains either 3 or 4 parameters depending on the length of the stimulus pulse train used to evoke a dopamine response (the k_R release plasticity factor, which alters the amount of dopamine released in response to a stimulus pulse over time, applies only to fitting dopamine responses arising from stimulus trains of more than one pulse). Essentially, the model posits that dopamine is released into an inner compartment (IC) and then crosses a diffusion-slowing barrier to enter the outer compartment (OC), where it encounters the FSCV carbon fiber electrode and is taken up by DAT. This original RD model makes excellent (R² >0.99) fits to individual observed dopamine FSCV responses under drug naive conditions and with exposure to nearly all drugs in vivo^{12,13, 24} and in brain slices^{14, 25} (Fig 1A, bottom).

Fig 1. Dopamine Reuptake In Inner And Outer Model Compartments.

A,B,C (top). Schematic cartoons showing the RD model before and after the inclusion of inner compartment uptake. Each model is matched with the set of traces immediately below it. A, bottom. A representative timecourse of the electrical stimulus (10 Hz, 10 pulse) evoked DA response before and after the introduction of 10 μM GBR12909 to a brain slice. The response amplitude and duration grow with exposure to GBR12909. Black lines are 4 parameter RD model fits to data with all parameters freely floating (R_p, k_U, k_T, k_R). Data traces were sampled at t = −5 min predrug, 5, 10, 30, 40, 50, and 70 min postdrug. The colors of the data traces with respect to time are in the order of the rainbow, red, orange, yellow, green, blue, indigo, and violet. The k_T values, in order from predrug to postdrug, were 1.34, 1.40, 0.95, 0.99, 0.40, 0.35, 0.21. R², in order = 0.99, 0.98, 0.99, 0.99, 0.97, 0.96, 0.96. B, bottom. As with (see A), but with the k_T parameter constrained to a single value for the fits. R², in order = 0.98, 0.97, 0.99, 0.99, 0.87, 0.80, 0.75. C, bottom. As with (see A), but the inclusion of an uptake term in the inner compartment (k_Ui) as a 5th adjustable model parameter permits all traces to be objectively fit with the same k_T value (k_T =0.21). R², in order = 0.99, 0.98, 0.99, 0.99, 0.97, 0.96, 0.96.

However, there are cases in which the original RD model fails to fit evoked dopamine data. One example is with hundreds of stimulus pulses delivered at 110 Hz frequency.²⁶ This is well outside the physiological range of dopamine neurons, but illustrates that the original RD model cannot fully account for all possible dopamine responses to any arbitrary electrical stimulus with any set of parameter values. Furthermore, we observed that adding the dopamine uptake inhibitor GBR12909 to a brain slice and fitting the evoked dopamine responses over time (Fig 1A bottom, Eqn 1 + Eqn 2) produced changes in the k_T parameter (Fig 1A, see legend). This is a problem, because there is no obvious reason for a drug to be able to affect the intrinsic rate of diffusion from one compartment to another, which is described by k_T. In fact, we asserted that at a given recording site, this passive transport parameter value should not change in response to any uptake inhibitor. A version of our model fitting algorithm that forced the k_T value to be the same for all time points during the experiment, while permitting all other parameter values to float, showed that the experimental data could not be fit with the original 4 parameter (4P) RD model (Fig 1B, Eqn 1 + Eqn 2). However, the experimental data could be fit by demanding that k_T is not affected by the drug and adding an uptake term k_Ui to the model’s inner compartment to reflect the presence of DAT at release sites^20–22 (Fig 1C bottom, Eqn 3 + Eqn 2). This suggests that perisynaptic DAT contributes to the kinetics of dopamine FSCV responses.

$$\frac{dD{A}_{ic}}{dt}=R_p\cdot f\cdot e^{-k_{R}t}-D{A}_{ic}\cdot k_T-D{A}_{ic}\cdot k_{Ui}$$ Eqn 3

Equation 3 adds a new inner compartment uptake term, k_Ui, which also removes DA from the IC alongside k_T. However, instead of removing the DA to the OC where it is detected by the FSCV electrode, it simply removes it from the extracellular space entirely, where it has no further impact with respect to the model.

Despite the improvement with inclusion of the inner uptake term k_Ui in Fig. 1C, the 5 parameter model still fails when the length of the stimulus train is changed in a single preparation (Fig. 2A–C). The 5 parameter RD model already contains one plasticity parameter that accounts for time dependent change in dopamine release per stimulus pulse (k_R) over the duration of the stimulus train, but this is not adequate for fitting stimulus trains of different lengths with the RD model with the same parameter values. However, dopamine reuptake is known to be subject to various processes which would directly or indirectly create progressive depression of uptake during a stimulus; these processes include progressive fractional occupation of the DAT with rising dopamine concentration, as might be predicted by the Michaelis-Menten model²⁷, and also Ca²⁺-dependent mechanisms^28,29 which could lower DAT activity. Both processes would be presumed to be indirectly engaged by electrical stimulation to some degree. Therefore, a term to modulate inner compartment uptake, k_kui, was included in the model to describe the stimulation dependence of uptake:

$$\frac{dD{A}_{ic}}{dt}=R_p\cdot f\cdot e^{-k_{R}t}-D{A}_{ic}\cdot k_T-D{A}_{ic}\cdot k_{Ui}\cdot e^{-k_{kUi}}$$ Eqn 4

Fig 2. Dopamine Inner Compartment Reuptake Rate Is Stimulus Dependent.

A. A block diagram of Equations 3& 2. Equation 3 modifies the original RD model by placing a term for uptake into the model’s Inner Compartment (k_Ui). B. Averaged post-nomifensine (10 μM) responses (orange) (n=24), recorded at dorsal striatal sites (n=8) in brain slices from 3 animals, and 5 parameter model (black) (see A) fit. The three stimulus conditions shown here (1 pulse, 5 pulses @ 10 Hz, and 10 pulses @ 10 Hz) are fit with the same set of model parameters for all three stimulus conditions. Although shown together, each stimulus was conducted 5 minutes after the previous one to permit the slice to recover and conditions to be equivalent. The “break” symbols below the traces illustrate this in this and the later panels. R² = 0.954. C. A representative set of 3 responses (12 pulses @ 60 Hz, 60 pulses @ 60 Hz, and 180 pulses @ 60 Hz) from a nucleus accumbens slow domain measured in vivo, before (orange) and after nomifensine (blue) (20 mg/kg, i.p.). For each drug state, the 3 responses were simultaneously fit with the same set of 5 parameter model (black) (see A) parameter values. As before, each response was separated in time by 5 minutes, although they are shown together. R² naive= 0.980, R² nomi= 0.996. D. A block diagram of Equation 4 & 2. Equation 4 adds a term to modulate the rate of inner compartment uptake (k_kUi). This is necessary to achieve the superior fits shown in parts (E) and (F). E. Averaged post-nomifensine (10 μM) responses (orange) (n=24), recorded at dorsal striatal sites (n=8) in brain slices from 3 animals, and 6 parameter model (see F) fit (black lines). R² = 0.998. The data are identical to that in (B). F. The same data set as (C), fit with a 6 parameter model (see D). R² naive= 0.992, R² nomi= 0.997. G. The k_T values of dopamine mass transport determined by simultaneous objective fitting of multiple dopamine responses at the same recording site under different stimulus conditions with the 6 parameter model of 2F. All parameter values were permitted to float for this fitting. Fast NAc nomi: N=7 in vivo experiments from 7 animals 25 min after i.p. injection of nomifensine (20 mg/kg), DA responses resulting from stimuli of 12, 60, and 180 pulses at @ 60Hz simultaneously fit, avg R²= 0.998. Slow NAc nomi: N=7 in vivo experiments from 7 animals 25 min after i.p. injection of nomifensine (20 mg/kg),DA responses resulting from stimuli of 12, 60, and 180 pulses at @ 60Hz simultaneously fit, avg R²= 0.997. DS Fast nomi: N=7 in vivo experiments from 7 animals 25 min after i.p. injection of nomifensine (20 mg/kg), DA responses resulting from stimuli of 12 and 180 pulses at @ 60Hz simultaneously fit, avg R²= 0.999. DS slice nomi: N=8 brain slice experiments, ~30 minutes after beginning perfusion with 10 μM nomifensine. DA responses fit from 1, 5, and 10 pulse stimuli @ 10 Hz. Avg R²= 0.998. NAc slice nomi: N=6 brain slice experiments, ~30 minutes after beginning perfusion with 10 μM nomifensine. DA responses fit from 1, 5, and 10 pulse stimuli @ 10 Hz. Avg R²= 0.993. The k_T values are not significantly different by ANOVA.

This version of the RD model, with 6 freely adjustable parameters, was incapable of converging on unique fits for single DA response curves, but requiring that the exact same set of parameter values simultaneously fit responses arising from multiple lengths of stimulus pulse trains produced unique fits to data sets (Fig 2E, F), and for several other data sets for which the fit is not shown but for which we report the fitting-derived k_T parameter and average R² value of the fit (Fig 2G). Strikingly, the average k_T values determined over many sets of different conditions by this 6 parameter fully objective multifitting method (Fig 2G) were about the same (~0.2) as that reported in Figure 1, which implies an average uptake-free mean lifetime of 5 seconds for dopamine within the perisynaptic space.^23,30 Therefore, this analysis suggests that dopamine diffusion out of the perisynaptic inner compartment is markedly hindered and the faster dopamine kinetics observed in vivo and in brain slices reflect attenuation by DAT activity.

Data fitting also produced values for k_kUi that were generally small (no more than ~0.02 in vivo and no more than ~0.15 in brain slices). However, as Figure 2D–F shows, this small decay in the inner compartment uptake during the stimulus created by the k_kUi was very effective at improving simultaneous RD model fits to different lengths of stimuli collected at the same recording sites with the exact same set of 6 parameter values, in brain slices and in vivo respectively.

Measuring regional differences in spillover and release plasticity

Objective fitting of data to a model with 6 adjustable parameters is time consuming and requires, at a minimum, sets of stimulus trains to be performed because it is not possible to achieve unique fits with this 6 parameter model to individual, single dopamine responses. This greatly increases data collection requirements for parameter extraction, requires the assumption that the recording site is invariant over time, and could make some experimental designs infeasible. Therefore, it is advantageous to reduce model flexibility by determining parameter values in advance of model fitting. Figure 2G illustrates an opportunity to do that, as it was found that the average value of the k_T parameter was about 0.2 s⁻¹ (corresponding to a time constant or mean lifetime of 5 s) in a wide variety of cases, including with and without drug and in brain slices and in vivo. We therefore fixed the value of k_T to be 0.2 for the rest of our analysis. In addition, we focused on short stimulus lengths for which the small value of the k_kui parameter would not matter much and the k_kui parameter was fixed to appropriate values in vivo or in vitro as determined by multiple length stimulus train fitting (Fig 2). Finally, we focused on stimulus trains of 1 second length. This enabled the RD model to be cast with 4 adjustable parameters and 2 fixed parameters, where it could achieve unique fits to short stimuli (i.e., as in Fig 3).

Fig 3. Dopamine Uptake Inhibitors Increase Spillover.

Effects of uptake inhibitors on outer compartment uptake, inner compartment uptake, and spillover. All fitting to extract these parameters was done with k_T set to 0.2 and the k_kUi set to 0.01 for in vivo bupropion experiments, 0.075 for drug naive GBR12909 slice experiments, and 0.15 for GBR12909 slice experiments. A. The effect of GBR12909 (10 μM) over time on outer compartment uptake. n=8 slices for most time points. B. The effect of GBR12909 (10 μM) over time on inner compartment uptake. n=8 slices for most time points. C. The effect of GBR12909 (10 μM) over time on spillover fraction. n=8 slices for most time points. D. The effect of bupropion (80 mg/kg i.p.) over time on outer compartment uptake. n=19 animals. E. The effect of bupropion (80 mg/kg i.p.) over time on inner compartment uptake. n=19 animals. F. The effect of bupropion (80 mg/kg i.p.) over time on spillover fraction. n=19 animals. G. The effect of nomifensine (20 mg/kg i.p.) over time on outer compartment uptake. n=14 animals. H. The effect of nomifensine (20 mg/kg i.p.) over time on inner compartment uptake. n=14 animals I. The effect of nomifensine (20 mg/kg i.p.) over time on spillover fraction. n=14 animals.

Mathematical models are most robust when they satisfy independent tests of their validity. Because by definition, dopamine either leaves the perisynaptic space by k_T or is removed from it by k_Ui, spillover fraction from the dopamine release sites can be calculated by Eqn 5. We predicted that blocking the k_Ui with an uptake inhibitor should therefore increase the fraction of dopamine that spills over. Therefore, we tested the hypothesis that blocking dopamine uptake with three different DAT blockers in vivo and in vitro would increase spillover. Testing the hypothesis both in brain slices and in vivo ruled out the possibility of confounding effects from isoflurane or dependence on intact brain circuits.

$$\frac{k_T}{\left(k_T+k_{Ui}\right)}=\mathit{\text{Spillover Fraction}}$$ Eqn 5

Figure 3A, 3D, and 3G show that GBR12909 in brain slices, bupropion in vivo, and nomifensine in vivo all lower uptake in the outer compartment. In addition, Figure 3B, 3E, and 3H show that under the same conditions, these drugs also lower uptake in the inner compartment. In turn, this lowering of inner compartment uptake corresponds to an increase in spillover as a result of exposure to these drugs (Fig. 3C, F, I). This feature of the model is consistent with the conclusion that the updated model can be used to measure DA spillover in vivo and in vitro.

Having developed an approach to measure spillover, dopamine responses throughout the striatum of anesthetized rats were analyzed to deduce spillover underlying volume transmission in different regions. This analysis also allows for comparison in release plasticity, which is reflected in the k_R parameter. (Fig 4). Figure 4A uses Eqn 5 and a mapping data set to calculate the average spillover fraction at each recording site for which modelable data was obtained by the mapping experiment. A higher DA spillover fraction increases the proportion of DA volume transmission to DA wired transmission. There are a few apparent “hot spots”, which may reflect local anatomical specializations for increased spillover fraction, and thereby increased volume transmission. Compared to the dopamine terminal architecture of the striatum, this map is relatively crude, consisting of 400 micron increments and single electrode track penetrations. Nevertheless, we present it as the first map of its kind, and offer that the kinetic analysis techniques presented in this work should be transferable to higher spatial resolution methods such as fluorescence sensor experiments^17–19 in order to build a more detailed map of spillover in the future.

Fig 4. Dopamine Release Plasticity And Spillover Vary Throughout the Striatum.

A. A map of short term release plasticity (k_R) in the striatum (in vivo). Each spatial coordinate consists of a sampling of measurements from 8 rats. (n=100 of 296 total measurements from 8 rats). The 100 measurements with the best signal to noise ratio were used to construct this map. Gray block regions have no data. All measurements were all recorded 10 minutes apart. ML = mediolateral axis, DV = dorsoventral axis, AP = anterioposterior axis. 0,0 - ML,AP is located at bregma, and DV depth refers to depth below the brain surface. k_R is an exponential decay constant applied to the release parameter R_p, so positive values (blue) represent depression of release and negative values (red) represent facilitation of release. ML = mediolateral axis, DV = dorsoventral axis, AP = anterioposterior axis. 0,0 - ML,AP is located at bregma, and DV depth refers to depth below the brain surface. B. The average calculated release plasticity (k_R) measured in each anatomical quadrant. An ANOVA identifies that at least one region is different than the others (***P < 0.001) while the post-test identifies that region as the DLS. C. As in 3A, a map of spillover fraction throughout the striatum, constructed from the same data. D. As in 3B, sorting calculated spillover fraction by quadrant. An ANOVA does not identify any significant differences among these groups.

Figure 4B illustrates that, even though there may be localized hot spots, DA spillover does not vary on average among anatomical quadrants of the brain - Dorsolateral, Dorsomedial, Ventrolateral, and Ventromedial. This is noteworthy in the context of prior work^31,32 to ascertain the functional roles of these anatomical regions, as it suggests that the balance between dopamine volume transmission and synaptic transmission is not likely to establish these functional roles.

Figure 4C shows the short term release plasticity throughout the striatum, which is calculated from the same data set as the spillover fraction of Fig 4A. As with spillover fraction, there appear to be “hot spots” of short term release depression. Unlike with spillover fraction, these “hot spots” of short term release depression are clustered in the dorsolateral striatum, leading to a significant regional difference (Fig 4D). The differences in dopamine short term release plasticity (k_R parameter) reflect different ratios of single pulse to pulse train DA amplitudes. This is in agreement with earlier model-independent work in brain slices³³. This agreement further builds confidence in the correctness of the analysis, and shows that variations discovered in slices also apply to the intact brain. This shows that differing release plasticity is correlated with different anatomically segregated functions of dopamine. Therefore, dopamine release plasticity is a possible candidate for mechanistic encoding of these differing functional roles.

Striatal dopamine FSCV responses have been discovered to arise from nonautoinhibited (fast) and autoinhibited (slow) domains, which exist on spatial scales exceeding the 200 micron length of a carbon electrode.³⁴ In light of this realization, we examined how this intrinsic autoinhibition is related to differences in spillover. Figure 5A shows a map of 296 evoked dopamine responses recorded throughout the striatum of 8 rats. It is immediately obvious from Figure 5A that fast or non-autoinhibited recording sites are much more prevalent in the lateral striatum and in particular the dorsolateral striatum (also see Fig 5D). We found in Fig 5B that autoinhibited sites predictably exhibit short term facilitation³⁵, while non autoinhibited sites show short-term depression. We also found that autoinhibited sites have more than double the spillover fraction of nonautoinhibited sites (Fig 5C). These differences in spillover and release plasticity between autoinhibited and nonautoinhibited sites were also consistent within anatomical striatal quadrants (Fig 5E and 5F).

Fig 5. Spatial Distribution of Fast And Slow Dopamine Domains Explains Spatial Differences in Release Plasticity and Spillover.

A. A map of prevalence of fast and slow DA recording sites in the striatum (in vivo). Each spatial coordinate consists of measurements from 8 rats. (n=296 measurements from 8 rats). Measurements were all recorded 10 minutes apart. Red represents observed fast DA response (cite) while dark blue represents observed slow DA responses. Light blue represents DA responses which were too small or slow to exceed the limit of detection. The area occupied by each color at each spatial coordinate is proportionate (by circle radius) to the frequency of DA response types observed over the data set at that coordinate. Fast DA sites are defined as those which show a dopamine response above the limit of detection within 200 ms, while Slow DA sites do not. ML = mediolateral axis, DV = dorsoventral axis, AP = anterioposterior axis. 0,0 - ML,AP is located at bregma, and DV depth refers to depth below the brain surface. B. A significant difference (P <0.0001, Welch’s t test) in short term release plasticity (k_R) between Fast (n=63) and Slow (n=37) modeled DA responses. C. A significant difference (P <0.001, Welch’s t test) in spillover fraction between Fast (n=63) and Slow (n=37) modeled DA responses. D. Significant differences in Fast DA domain abundance among anatomical quadrants (P <0.001 ANOVA). E. Significant differences between in short term release plasticity between Fast and Slow DA domains by anatomical quadrant. (DLS, Fast n = 35, Slow n = 9, P<0.001 Welch’s t test. DMS, Fast n = 12, Slow n = 14, P<0.01 Welch’s t test. VLS, Fast n = 14, Slow n = 10, P<0.001 Welch’s t test. VMS, Fast n = 2, Slow n = 6, not significant, Welch’s t test. F. Significant differences between in spillover between Fast and Slow DA domains by anatomical quadrant. (DLS, Fast n = 35, Slow n = 9, P<0.05 Welch’s t test. DMS, Fast n = 12, Slow n = 14, P<0.01 Welch’s t test. VLS, Fast n = 14, Slow n = 10, not significant Welch’s t test. VMS, Fast n = 2, Slow n = 6, not significant, Welch’s t test.

It is not clear from this analysis why the slow sites, or the ones with intrinsic autoinhibition, have a greater spillover fraction. One possible explanation is that that the autoinhibited sites are autoinhibited due to a higher basal concentration of dopamine present there, and that in addition to causing autoinhibition, this basal concentration partially occupies the DAT, causing inner compartment uptake in the slow sites to be reduced. It also stands to reason that more spillover and therefore reduced uptake within the perisynaptic space proximal to autoreceptors ought to increase autoinhibition as well. What is certain is that these slow sites will exhibit a greater ratio of volume transmission to wired transmission than the fast sites, due to their higher spillover fraction.

Because we were not able to use every single FSCV data recording to generate the spillover map in Figure 4 due to signal to noise issues (see Methods, and Fig S1), we wanted to further analyze our data to test if spillover fraction and release plasticity were correlated. The strategy used for this purpose was to calculate the average release plasticity and spillover parameters for fast and slow sites within each anatomical quadrant (Fig 5E, 5F), and then to multiply these values by the frequency of fast and slow sites observed at each recording site. This is beneficial because it is possible to tell if a recording site has fast or slow kinetics even if the signal produced is of too low signal to noise ratio to be amenable to modeling. This is also based on the assumption that the release plasticity and spillover fraction for fast and slow sites is not different between the sites that have robust enough signal to noise ratios to model and those that do not. When maps of release plasticity (Fig 6A) and spillover fraction (Fig 6B) are constructed in this way, the release plasticity map appears similar to the map constructed in Figure 4C, but the spillover map shows that the high proportion of low-spillover fast sites in the DLS predicts lower overall spillover there. When the values on these maps that are generated in this way are tested for correlation, a correlation coefficient of 0.789 is found (Fig 6C). This reflects the differences found in both release plasticity and spillover fraction (Fig 5B and 5C) and the observed frequency of fast and slow sites at each recording site (Fig 5A), thus suggesting that release plasticity and spillover fraction are related. We argue that this relationship is of biological origin rather than as a consequence of over-parameterization due to the way in which the model was constructed as shown in Fig 1 and Fig 2, and also because this fitting strategy produces unique fits. Further evidence that this parameter correlation is of biological origin and not a mathematical artifact is shown below (Fig 7) - the correlation is not, therefore, a mathematical artifact. It is logical that increased spillover resulting from reduced k_Ui would be associated with higher local DA concentrations in both the inner and outer compartment. This would lead to more intrinsic autoinhibition, reflected in the lower k_R parameter (Fig 6C).

Fig 6. Spatial Distribution of Fast And Slow Dopamine Domains Explains Spatial Differences in Release Plasticity and Spillover.

A. A map of short term release plasticity generated by, at each coordinate, adding the product of the fast site frequency (f/8) at each coordinate and the average fast site release plasticity for that quadrant (Fig 4E), and the product of the slow site frequency (s/8) at each coordinate and the average slow site release plasticity for that quadrant (Fig 4E). For the purposes of generating this map, sites which were not fast were assumed to be slow. B. A map of spillover generated by, at each coordinate, adding the product of the fast site frequency (f/8) at each coordinate and the average fast site spillover for that quadrant (Fig 4F), and the product of the slow site frequency (s/8) at each coordinate and the average slow site spillover for that quadrant (Fig 4F). For the purposes of generating this map, sites which were not fast were assumed to be slow. C. The correlation between the two maps generated in A & B.

Fig 7. Dopamine Uptake Inhibitors Cause Release Depression by Blocking Vesicle Refilling.

Throughout the figure, Red represents Fast DA domains while Blue represents Slow DA domains. Where normalization is used within a figure panel (G-I), all curves are normalized to the initial, drug free stimulus pulse. A. The influence of GBR12909 (10 μM) on brain slices (n=8, but fewer at some time points shown). The k_R parameter increases. B. The influence of bupropion (80 mg/kg i.p.) on the release plasticity factor k_R at fast (n=9) and slow (n=10) recording sites in the dorsal striatum, as determined by multiple fitting with the 6 parameter model with all parameters allowed to float except for k_T, which was fixed to 0.2. The k_R parameter increases. C. The influence of nomifensine (20 mg/kg i.p.) on the release plasticity factor k_R at fast (n=7) and slow (n=7) recording sites in the nucleus accumbens, as determined by multiple fitting with the 6 parameter model as in Figure 2D. The k_R parameter increases. D. A plot of spillover fraction versus release plasticity for the data shown in (A). E. A plot of spillover fraction versus release plasticity for the data shown in (B). F. A plot of spillover fraction versus release plasticity for the data shown in (C). G. A plot of the normalized DA release over the course of the 1 second stimulus train for t = 0 (black dots) and t = 50 min (green dots). H. A plot of the normalized DA release over the course of the 1 second stimulus train for t = 0 (red and blue light dots) and t = 37 min (red and blue heavy dots). I. A plot of the normalized DA release over the course of the 3 second stimulus train for t = 0 (red and blue light dots) and t = 25 min (red and blue heavy dots).

Dopamine release depends on continuous uptake

Short-term release plasticity (k_R) is under the control not only of autoinhibition, but also the depletion of the releasable pool. Vesicular release of the antipsychotic drug cyamemazine is able to deplete the releasable dopamine pool completely²⁵, which would correspond to a very large and positive k_R value. In addition, upon protracted stimulus additional dopamine stimulus pulses will simply cease to be effective^36,37. This apparent depletion of the releasable pool cannot be a regulatory effect due to autoinhibition because raclopride does not change the kinetics of protracted dopamine stimulus responses, but only scales up the amplitude³⁸. The depletion of the releasable pool demonstrated under the above extraordinary circumstances led us to consider the contributions of the perisynaptic barrier and dopamine uptake to the normal maintenance of releasable pool.

Specifically, the observation that the k_T parameter was so small that dopamine would take ~5 seconds to exit the inner compartment (Fig 2G) raised the possibility that the dopamine terminals were taking advantage of the perisynaptic barrier to rapidly refill the synaptic vesicles after kiss and run fusion³⁹ without losing much to other cells. This would enable repeated and relatively long-lasting synaptic transmission to be actuated while reducing loss of the DA molecule to the environment by diffusion and thus volume transmission. We therefore decided to test for this effect by blocking the refilling of synaptic vesicles from released dopamine via block of DAT, and seeing how much release depression this could cause.

Figure 7A–C show this short term release depression effect for GBR12909, bupropion, and nomifensine. Interestingly, detecting this effect is not dependent on the 6 parameter model we developed here to measure spillover, as can be seen by Fig S2C. We found that this type of short term release depression was positively correlated with spillover (Fig 7D–F). This is not surprising, because dopamine that is spilling over is not being taken up and loaded into vesicles for release on subsequent stimulus pulses. This illustrates the different associations of release plasticity with exhaustion of release capacity (Fig 7G–I) and intrinsic autoinhibition (Fig 6). This also supports, as previously mentioned, that the correlated parameters of spillover fraction and k_R are of biological origin rather than from overfitting, since the correlations between release plasticity and spillover occur in opposite directions under the different sets of experimental conditions in Fig 6 and Fig 7.

Importantly, these data (see the Supplemental for bupropion data) are consistent with our hypothesis that blocking the DAT would prevent the vesicles from refilling and that this would be relevant to release on the subsecond timescale of a stimulus pulse train. Therefore, we conclude that dopamine reuptake inhibitors not only boost initial dopamine levels, but remove the capacity of the DA terminals to respond normally to further electrical inputs by preventing refilling of the DA vesicles.

We observed both a previously described effect of uptake inhibitors boosting initial release^14,40 (Fig 7I, nomifensine, in vivo, fast NAc DA domains), and unchanged initial release (Fig 7G, GBR12909, slice, Fig 7H, bupropion, in vivo, fast and slow DS DA domains, Fig 7I, nomifensine, in vivo, slow NAc DA domains). We report the release per pulse in Fig 7G, H, and I normalized to the largest naive DA release per pulse in each of the panels. As has been previously reported⁴¹, we see anatomical differences in uptake inhibitor effects (Fig 7I). Uptake inhibitors that are notorious drugs of abuse, such as cocaine, have been previously suggested to increase DA release^14,40. The uptake inhibitors that are here claimed to not increase DA release, such as bupropion and vanoxerine (GBR12909), have therapeutic uses and are not DEA-scheduled. Nomifensine is also not DEA-scheduled, but only appears to increase initial DA release in the fast dopamine domains; initial DA release in the slow domains appears to be unaffected. The domain specificity of effects on DA release by DEA-scheduled uptake inhibitors has not been investigated. These interesting preliminary findings argue for a future, more inclusive in vivo study to investigate the effects on initial dopamine release of different dopamine reuptake inhibitors with well characterized abuse potentials in fast and slow dopamine domains. The determination of the mechanism of action of selective boosting of initial dopamine release in some anatomically discrete striatal domains is also important, as this may well prove to be relevant to drug addiction as well as questions of basic function.

CONCLUSION

Dopamine receptor activation depends on biophysical parameters such as dopamine release, passive transport, active transport, and release plasticity. However, these parameters cannot be directly measured in living brains. Fully understanding the mechanistic basis of dopamine signaling therefore requires a quantitative theory or model which can accurately reproduce any possible dopamine response in terms of biophysical parameters with a unique set of values. The updated model describes dopamine as being subject to DAT-mediated uptake on both sides of a perisynaptic diffusion barrier between the sites of dopamine release and the bulk extracellular space. This enables, for the first time, the calculation of the timecourse and regulation of spillover from the perisynaptic space. The model remarkably claims that an average transit time of ~5s is required for dopamine molecules to cross this diffusion barrier, in the absence of DAT activity. Analysis of mapping data sets with this model reveals hotspots in spillover and release plasticity, and significantly greater short term depression of dopamine release in the dorsolateral striatum than in other striatal subregions. Analysis of drug effect data sets supports the conclusion that dopamine reuptake is required to maintain sustained release and therefore blocking uptake with a drug interferes with sustained dopamine release. The analysis also raises the possibility that a domain specific effect on initial release by certain dopamine reuptake inhibitors may contribute to abuse liability.

METHODS

Brain Slice Preparation

Animal protocols and care procedures were in accordance with the National Institutes of Health Guide for the care and use of laboratory animals and approved by the University of Pittsburgh Institutional Animal Care and Use Committee. Whole coronal striatal slices (300 μm thick, taken 0.3–1.2mm caudal of the anterior most portion of the striatum) were prepared in icy aCSF¹⁴ from male Sprague-Dawley rats (14–40 days postnatal). Slices were superfused with 29–31 °C aCSF (95% O₂/5% CO₂) containing 0.0005% v/v DMSO or 0.0005% v/v DMSO and 10 μM GBR12909. FSCV was performed at carbon fiber microelectrodes^42–44 (7 μm diameter, 200 μm length) with HDCV software⁴⁵ on a WaveNeuro potentiostat (Pine Instruments). Carbon fibers were fully inserted into the slice at a 20° angle with the tip positioned 100 μm below the slice surface. DA release was evoked through optically isolated electrical stimulation (800 μA, biphasic, 2 ms per phase) performed with a stimulus isolator (A365, WPI) under the control of the potentiostat, and pulse generators (A310, WPI; AM-2100, AM Systems). Current was passed in 1 second, 10 Hz trains ~2 mm between electrode poles (MS303–1-A-SPLELECTSS, Plastics One) (13) in contact with the brain surface. When multiple sites were measured, the simulating electrode was moved so that the area bracketed by the stim electrode poles was did not overlap and the carbon fiber electrode was implanted equidistant between the stimulating electrode poles. DA responses were observed throughout the striatum. Background subtracted DA peak current time series were converted⁴⁶ to concentrations.

Small Animal Surgeries

All procedures involving animals were approved by the University of Pittsburgh Animal Care and Use Committee. Rats (male, Sprague-Dawley, 250–450g, Charles River Inc., Wilmington, MA) were anesthetized with isoflurane (2.5% by volume O₂), placed in a stereotaxic frame (David Kopf, Tujunga, CA), and connected to an isothermal blanket (Harvard Apparatus, Holliston, MA). Carbon fiber electrodes and stimulating electrodes (MS303/a, Plastics One, Roanoke, VA) were implanted in the dorsal striatum and ipsilateral medial forebrain bundle. The stimulus waveform was a biphasic constant current square wave (2 ms pulses, 60 Hz, 250 μA, 200 ms, 1s, or 3 s in duration) delivered with a stimulus isolation unit (Neurolog 800, Digitimer, Letchworth Garden City, UK). For Figs 3 and 8, evoked responses were recorded before and after i.p. administration of 80 mg/kg bupropion hydrochloride (Sigma-Aldrich, St Louis MO). For Figs 2, 3, and 8, evoked responses evoked responses were recorded before and after i.p. administration of 20 mg/kg nomifensine maleate (Sigma-Aldrich, St Louis MO).

FSCV

Carbon fiber electrodes (7 μm in diameter and 200 μm in length) were prepared with T650 fibers (Cytec LLC, Piedmont, SC, USA). The electrodes detected DA via a 400 V/s waveform beginning at 0.1V, rising to 1.3V, falling to −0.5V, and rising again to the 0.1V resting potential. This waveform was applied at 10 Hz. We have previously used this waveform to electrochemically pretreat electrodes, and it was used in this study to obtain good sensitivity from the 1.3V oxidation limit. FSCV was performed with a fast-scan potentiostat (EI-400, out of production) and CVTarHeels software (courtesy Prof. Michael Heien, University of Arizona). In vivo FSCV calibration was performed in a homemade flow cell using DA (Sigma, St Louis, MO, USA) dissolved in N₂ purged artificial cerebrospinal fluid (142 mM NaCl, 1.2 mM CaCl₂, 2.7 mM KCl, 1.0 mM MgCl₂, 2.0 mM NaH₂PO₄, pH 7.4). Slice FSCV calibration was performed with the previously used method of Sombers et al.

Voltammetry Data Analysis

Objective Classification of Kinetic DA Types

Fast DA sites are those which produce an identifiable dopamine response to 12 pulses of 250 uA electrical stimulus of the medial forebrain bundle. If they do not do this, but eventually respond to electrical stimulus if more pulses than 12 are administered, they are deemed to have slow DA overflow kinetics.

Hang-up Correction

The hang-up correction was explained in detail by Walters et al, 2015^{13 11}. Briefly, the algorithm assumes that DA undergoes first order adsorption and desorption at the surface of the FSCV electrode according to the following rate expression: $\frac{dH}{dt}=k_{on}C-k_{off}{\Gamma }_{DA}$ Equation 1, which is used to construct a hang-up signal component, H(t), by curve fitting to the hang-up segment of the measured response. The correction is performed by subtracting the calculated signal component from the measured response. In performing the hang-up correction, it is important to avoid distorting DA’s apparent kinetics. This could occur, for example, by curve-fitting H(t) to the measured response before the time where the measured response is caused solely by hang-up. To avoid this outcome, we fit H(t) to later and later segments of the reponse until H(t) stops changing.

Mapping Data Selection

It is not possible in a mapping experiment to search for robust dopamine responses to study exclusively, as has been the case in many past studies. The inclusion of “small responses”, which occur throughout the brain along with the large ones, creates variations in signal-to-noise that confounds modeling (see Figure S1A and S1B) by compromising the accuracy of extracted parameters. In addition, signal averaging of dissimilar responses is not a panacea for the problem of low signal to noise ratio because this can at least theoretically cause systematic distortions in the extracted values if dopamine responses arising from dissimilar parameter values are added together. (Fig S1C). Therefore, our approach to construct maps of spillover and other parameters such as release plasticity was to only analyze a group of responses above a signal to noise ratio with a certain cutoff. While this approach was useful in finding a number of significant differences across the data set, it should not be interpreted as a comprehensive striatal dopamine map, as many recorded responses were too small and noisy for analysis. Given that there are at present no proven technologies which can measure any arbitrarily small dopamine response in vivo, this remains a general problem for the field, in that we cannot hope to analyze features of brain behavior below our measurement resolution.

Mathematical Modeling

The DA kinetic model has been explained and used in prior recent reports from our laboratory. It is intended to provide a generic description of restricted diffusion in the brain extracellular space. To do so, it treats the extracellular space as if it were divided into an inner and outer compartment. The model postulates that DA is released into the inner compartment and undergoes restricted diffusion to the outer compartment where it is detected by the FSCV electrode. Uptake then removes DA from the outer compartment. The model is composed of two coupled differential equations. One equation describes Inner Compartment dopamine dynamics and one describes Outer Compartment dopamine dynamics. There are three versions of the Inner Compartment equation used in the text but only one version of the Outer Compartment equation

Inner Compartment Dopamine Dynamics

$$\frac{dD{A}_{ic}}{dt}=R_p\cdot f\cdot e^{-k_{R}t}-D{A}_{ic}\cdot k_T\left(4\text{ parameter RD model version}\right)$$

$$\frac{dD{A}_{ic}}{dt}=R_p\cdot f\cdot e^{-k_{R}t}-D{A}_{ic}\cdot k_T-D{A}_{ic}\cdot k_{Ui}\left(5\text{ parameter RD model version}\right)$$

$$\frac{dD{A}_{ic}}{dt}=R_p\cdot f\cdot e^{-k_{R}t}-D{A}_{ic}\cdot k_T-D{A}_{ic}\cdot k_{Ui}\cdot e^{-k_{kUi}}\left(6\text{ parameter RD model version}\right)$$

Outer Compartment Dopamine Dynamics

$$\frac{d{[DA]}_{oc}}{dt}=\frac{D{A}_{ic}\cdot k_T}{V_{oc}}-{[DA]}_{oc}\cdot k_U\left(4,5,\text{ and }6\text{ parameter RD model version}\right)$$

There are up to 6 adjustable parameters depending on the model used; R_Prepresents the moles of DA released per stimulus pulse, k_R is a first order rate constant that modifies DA release, k_T is a first-order rate constant for transport between the compartments, k_Ui is first order uptake from the inner compartment, k_kUi is a first order decay term operational during the stimulus only for inner compartment uptake, and k_U is a first-order rate constant for DA uptake. There are two fixed parameters; V_oc is the volume of the outer compartment (16 μm³, see Walters et al 2014)^{12 10} and f is the stimulus frequency.

The model is fit to the data via a direct search algorithm which has been previously described for a three and four dimensional space and which is scalable to any dimensionality of parameter space. The specific conditions of fitting and cases where fixed parameters were used are noted in the text and figure legends.

Statistics

Statistical analyses were performed in GraphPad Prism and are described in the figure legends.