Animal behavior is influenced by the release of neuromodulators such as dopamine (DA), which signal behavioral variables that are relevant to the functioning of circuits brainwide. Projections from dopaminergic nuclei to the striatum and cortex, for example, play important roles in reinforcement learning, decision-making, and motor control. Loss of DA or dysfunction of its target circuits has been linked to disorders such as Parkinson’s disease, schizophrenia, and addiction (
1–
3).
Much work has been devoted to determining how neural representations of behavioral states are encoded in the firing patterns of neuromodulatory neurons (
4–
9), but very little is known about how the precise release of neuromodulators alters the function of their target circuits (
10,
11). To address this problem, an essential step is to monitor the spatiotemporal dynamics of neuromodulatory signals in target circuits while also measuring and manipulating the elements of the circuit during behavior.
Analytical techniques such as microdialysis and electrochemical microsensors have provided useful insights about neuromodulator presence (
12,
13) but suffer from poor spatial and/or temporal resolution and cannot be targeted to cells of interest. Optical approaches such as injected cell-based systems (CNiFERs) (
14) and reporter gene–based iTango (
15) can reveal DA release with high molecular specificity. However, these systems are limited by poor temporal resolution (seconds to hours), preventing direct detection of DA release events that occur on a subsecond time scale (
16,
17).
High-quality single fluorescence protein (FP)–based sensors that report calcium or glutamate transients with subsecond temporal resolution have recently been developed and are widely used (
18,
19). Here, we report the development of a set of single FP–based DA sensors, named dLight1, that enables imaging of DA transients with high spatiotemporal resolution in behaving animals.
Sensor engineering
Sensitive optical readout of changes in DA concentration was achieved by directly coupling the DA binding–induced conformational changes in human DA receptors to changes in the fluorescence intensity of circularly permuted green fluorescent protein (cpGFP). We did this by replacing the third intracellular loop (IL3) of the human dopamine D1 receptor (DRD1), D2 receptor (DRD2), and D4 receptor (DRD4) with a cpGFP module from the genetically encoded calcium indicator GCaMP6 (
Fig. 1A).
To determine the insertion site of cpGFP in IL3 that produces maximal coupling of ligand-induced conformational changes to cpGFP fluorescence, we aligned the sequences of DRD1 and DRD4 with that of the β
2 adrenergic receptor (B2AR) (
Fig. 1B), for which both active and inactive structure are available (
20). The initial variant, obtained by inserting a cpGFP module with original linker sequences (LSSLE-cpGFP-LPDQL) between Lys
232 and Lys
269 of DRD1, was well expressed at the plasma membrane of human embryonic kidney (HEK293) cells and showed a fluorescence decrease (Δ
F/Fmax = –19.4 ± 0.02%) in response to puffed DA (fig. S1A). To obtain a positive-response sensor, we screened a library of 585 variants in HEK cells (
Fig. 1C and fig. S1B). The variant with the largest positive fluorescence response (max Δ
F/Fmax = 230 ± 9%) and excellent membrane localization was named dLight1.1 (
Fig. 1D). In situ DA titration on HEK cells revealed submicromolar apparent affinity of dLight1.1 (affinity constant
Kd = 330 ± 30 nM;
Fig. 1E).
We next sought to further tune the dynamic range and affinity of the sensor. Mutation of Phe
129, a highly conserved residue among many G protein–coupled receptors (GPCRs) (
21), into Ala (dLight1.2) slightly increased dynamic range (max Δ
F/Fmax = 340 ± 20%,
Kd = 770 ± 10 nM;
Fig. 1, D and E). Optimizing the cpGFP insertion site in dLight1.1 and dLight1.2 (fig. S1, C to G) greatly increased the dynamic range but also reduced the affinity to micromolar range (dLight1.3a: Δ
F/Fmax = 660 ± 30%,
Kd = 2300 ± 20 nM, fig. S2, A and B; dLight1.3b: Δ
F/Fmax = 930 ± 30%,
Kd = 1680 ± 10 nM;
Fig. 1, D and E). Insertion of the cpGFP module into DRD4 and DRD2 produced dLight1.4 and dLight1.5, respectively, which exhibited nanomolar affinity with a relatively small dynamic range [dLight1.4: Δ
F/Fmax = 170 ± 10%,
Kd = 4.1 ± 0.2 nM,
Fig. 1, B, D, and E; dLight1.5: DA, Δ
F/Fmax = 180 ± 10%,
Kd = 110 ± 10 nM; quinpirole (synthetic agonist of D2 dopamine receptors), Δ
F/Fmax = 124 ± 19%, fig. S2, A to C]. In addition, we engineered a control sensor by incorporating a D103A mutation in dLight1.1 to abolish DA binding (control sensor: Δ
F/F = 0.4 ± 4%,
Fig. 1E) (
22). Because dLight1.1 and dLight1.2 produced large responses at low DA concentration (e.g., 100 nM) without approaching response saturation (
Fig. 1E, inset) and had submicromolar affinity, we further characterized these two sensors.
Sensor characterization
These two sensors showed peak emissions at 516 nm and 920 nm for one- and two-photon illumination in HEK cells, respectively (fig. S3). In situ titration on dissociated hippocampal neurons and on HEK293 cells showed similar apparent affinities to DA (
Fig. 1E and fig. S4, A to C). Single 5-ms pulses of uncaged DA were robustly detected on the dendrites of cultured neurons, and the fluorescence response tracked uncaging pulse duration (fig. S4, D to F). In cultured hippocampal slices, dLight1 could reliably detect submicromolar DA concentration changes at dendrites and single dendritic spines (fig. S4, G to I).
We then investigated the endogenous and pharmacological molecular specificity of the sensor. dLight1 was less sensitive to norepinephrine and epinephrine than to DA by factors of ~70 and ~40, respectively; negligible responses were observed to all other neuromodulators tested (fig. S5). The amplitude of the response to each pharmacological compound reflected the efficacy of drugs on the wild-type receptors, with the largest response to the full agonist dihydrexidine (Δ
F/F = 300 ± 10%), followed by partial agonists (
Fig. 1F). The response to DA was abolished in the presence of the DRD1 antagonists SKF-83566 and SCH-23390 but was unaffected by the DRD2 antagonists haloperidol and sulpiride (
Fig. 1F).
To investigate the possible interference of sensor expression with G protein signaling, we first measured the effect of sensor expression on the ligand-induced cyclic adenosine monophosphate (cAMP) response (fig. S6) (
23). Transiently transfected dLight1.1 and dLight1.2 triggered no significant cAMP response in HEK cells, similar to the negative control (EGFP), whereas wild-type DRD1 receptor significantly did (fig. S6A). The conversion of DRD1 to a fluorescent sensor thus apparently blocked the scaffold’s ability to bind G protein and trigger the signaling cascade. When introduced into a cell line that endogenously expressed DRD1 (U2OS), dLight1 did not significantly alter the dose-response curve for DA (
P = 0.96, fig. S6B). dLight1 also showed a significant reduction in agonist-induced internalization, a readout of DRD1 engagement of β-arrestin (
24), when compared to wild-type DRD1 (fig. S6C). Total internal reflectance fluorescence (TIRF) imaging verified that dLight1 remained diffusely distributed in the plasma membrane, without any detectable internalization, during a complete cycle of ligand-dependent fluorescence change (fig. S6, D to F). Taken together, these results indicate that the dLight sensors are suitable for use on the cell membrane without affecting endogenous signaling through G proteins or engagement of β-arrestins.
Two-photon imaging of DA release in dorsal striatum ex vivo and in vivo
We next used dLight1 to measure the time course and concentration of endogenous DA release triggered by electrical stimulation and drug modification in acute striatal slices with two-photon imaging (
Fig. 2A). Two to four weeks after injection of an adeno-associated virus encoding dLight1 (AAV9.
hSynapsin1.dLight1.2) into the dorsal striatum, we observed both broadly distributed and localized fluorescence transients across the field of view (
Fig. 2, B and C, and fig. S8, A to C) in response to a single electrical stimulus. Fast line scan at these hotspots (
Fig. 2C) revealed a rapid onset of fluorescence increase (rise τ
1/2 = 9.5 ± 1.1 ms) followed by a plateaued peak (averaged Δ
F/F = 220 ± 50%) for about 150 ms, which decayed to baseline in about 400 ms (decay τ
1/2 = 90 ± 11 ms,
Fig. 2D). We observed robust and reproducible fluorescent transients to low-frequency stimuli over a prolonged imaging period, whereas subsequent higher-frequency stimuli elicited significantly smaller responses (
Fig. 2, E and F), indicating strong depression from an initially high probability of release. Blockade of DA reuptake with cocaine significantly prolonged the decay of fluorescence from peak to baseline (
Fig. 2, G and H), but with equivocal effect on response amplitude (
Fig. 2, G and H). Application of the competitive antagonist SKF83566 eliminated the responses (fig. S8F), confirming that fluorescent signals are indeed attributable to DA binding.
We next used dLight1 to estimate released DA concentration induced by a brief electrical stimulus. By comparison with a concentration-response curve (fig. S8, D, E, and G), the fluorescence response suggested a DA release of 10 to 30 μM (
Fig. 2I), which is one to two orders of magnitude higher than previously reported in ventral striatum using fast-scan cyclic voltammetry (FSCV) (
25) and is similar to that reported by measuring DRD2 activation (
26). Addition of saturating amphetamine (10 μM in the presence of 400 μM sulpiride) increased tonic DA to 3.3 μM (fig. S8, F and G).
We then examined the action of known modulators of DA release using dLight1 (
Fig. 2, J to L). Activation of D2 autoreceptors with quinpirole decreased the electrically evoked fluorescence transients; this effect was significantly reversed by the application of sulpiride (
Fig. 2J). Perfusion with a κ-opioid receptor agonist (U69,593) caused a small decrease in the amplitude, which was completely blocked by naloxone (
Fig. 2J). We then imaged the effects of nicotinic receptor activation in mediating the probability of DA release. Blockade of nicotinic receptors with hexamethonium profoundly reduced the fluorescence transient, which depended on the number of stimuli (
Fig. 2, K and L). In the absence of hexamethonium, the amplitude of the fluorescence remained consistent regardless of the stimulation protocol (
Fig. 2, K and L) (
27).
Next, we asked whether dLight1 could reliably report DA signals associated with mouse locomotion in dorsal striatum, which was labeled with AAV1.
hSynapsin1.dLight1.1/1.2 and AAV1.
hSynapsin1.flex.tdTomato. We measured DA transients with two-photon imaging during rest and self-initiated locomotion (fig. S9). Consistent with in vivo two-photon calcium imaging of substantia nigra pars compacta (SNc) axon terminals in dorsal striatum (
10), dLight1 reliably showed widespread and synchronous subsecond transients associated with spontaneous locomotion, which was clearly distinguishable from motion artifacts (fig. S9, A to E). The DA transients were rapidly and bidirectionally modulated with respect to locomotion. Accelerations were associated with an increase and decelerations with a decrease in fluorescence (peak mean cross-correlation 240 ms fig. S9, F to L).
In summary, dLight1 faithfully and directly reports the time course and concentration of local DA release and drug-dependent modulatory effects on DA release in an acute striatum slice. In addition, dLight1 enables direct visualization of locomotion-triggered DA release in behaving mice.
Deep-brain recording of DA dynamics simultaneously with optogenetics or calcium imaging
The nucleus accumbens (NAc) receives projections from dopaminergic neurons in the ventral tegmental area (VTA). To directly probe DA release in freely moving mice, we delivered AAV9.
CAG.dLight1.1 or AAV9.
CAG.control_sensor in NAc, followed by fiber photometry imaging (
Fig. 3 and fig. S10, A and B). dLight1 revealed visible spontaneous DA transients, which were absent in the imaging sessions using the control sensor (fig. S10C).
To optically activate VTA dopaminergic neurons, we infected VTA of
TH::IRES-Cre mice with AAV5.
hSynapsin1.flex.ChrimsonR.tdTomato (
28) (
Fig. 3, A to C, fig. S11, A and B, and fig. S12, A and D). The high temporal resolution of dLight1 enabled detection of individual peaks of DA transients in response to 5-, 10-, and 20-Hz photostimulation (
Fig. 3D and fig. S13, A to C). The amplitude of fluorescence increase was correlated with the frequency of photostimulation (
Fig. 3, D and F). In contrast, no fluorescence changes were observed with the control sensor using 20-Hz stimuli (
Fig. 3, D and E). Relative to saline-injected controls, systemic administration of SCH-23390 significantly reduced optogenetically induced dLight1 responses, whereas the reuptake inhibitor GBR-12909 enhanced them (
Fig. 3, G and H).
Next, we examined whether dLight1 can report inhibition of DA transients. To induce transient inhibition of VTA dopaminergic neurons, we optogenetically stimulated VTA γ-aminobutyric acid–releasing (GABAergic) neurons in
VGAT::IRES-Cre mice (
29) (
Fig. 3I). Histology confirmed ChrimsonR expression in VTA GABAergic neurons (fig. S12, B, C, and E). We observed rapid and reversible reductions in dLight1 fluorescence in response to VTA GABAergic neuron photoactivation at 40 Hz (
Fig. 3, J and K, and fig. S13D), indicating that dLight1 can report bidirectional changes in local DA release.
Motivationally salient stimuli modify DA neuron firing and downstream NAc activity (
9,
29,
30). To link the DA release to local neuronal activity, we performed dual-color measurements with dLight1 and the red-shifted calcium indicator jRGECO1a (
31) in lateral core/shell regions (
Fig. 3L and figs. S10B, S11A, and S12F). When mice voluntarily consumed a reward (50 μl of 5% sucrose), we observed a concordant increase of DA concentration and local population activity (
Fig. 3, M and N, and fig. S13E), similar to a class of NAc single units showing excitation upon reward (
32). In contrast, footshocks suppressed DA release while enhancing local neuronal activity, indicating dissociation between DA dynamics and local circuit activity (
Fig. 3, O and P, and fig. S13F).
Chronic imaging of DA dynamics throughout cue-reward learning
We next examined the utility of dLight1 in reporting modulation of DA signaling in response to conditioned stimuli (CS) and unconditioned stimuli (US) throughout Pavlovian conditioning (
Fig. 4A) (
8,
33,
34). Mice successfully learned to associate the predictive cues to the reward, as shown by increasing numbers of licks during CS over the course of training and by decreasing numbers of licks during extinction learning (
Fig. 4B).
Repeated fiber photometry recordings in NAc revealed two types of DA transients modulated during associative learning: increased DA response to the predictive cues and decreased response to reward consumption across sessions. In the first session, a small and time-locked phasic DA signal was present at the CS onset, whereas after US the DA signal was larger and also more temporally spread (
Fig. 4, C and D), consistent with US consumption onsets being highly variable at early stages (fig. S14, A and B). Aligning to consumption onset revealed large DA signal to the US at the first session (
Fig. 4C and fig. S14A). Upon repeated cue-reward pairings, the amplitude of CS response significantly increased (
Fig. 4, C, D, F, and H, and fig. S14C). On the other hand, US response, when aligned to the consumption onset, showed a monotonic decrease across learning sessions (
Fig. 4, G and H, and fig. S14D) (
9,
33). During extinction, we observed an attenuated phasic CS response (
Fig. 4, E, F, and H). The amplitude of the phasic CS response was correlated with CS-triggered licking behavior during both learning and extinction sessions (fig. S14E).
We further investigated whether dLight1 can report signals correlated with “reward prediction error” (
4). After the animals had fully learned CS-US association, mice underwent “unexpected reward availability” sessions (in which the US was occasionally made available without the CS) between normal paired trials (
Fig. 4I). Unexpected availability of reward elicited significantly higher fluorescence than did expected consumption (
Fig. 4, J and K). In the “unexpected reward omission” session, where the US was occasionally omitted after the predictive CS, fluorescence decreased below the pre-CS baseline after the time at which the US would have normally become available after CS presentation (
Fig. 4, L and M).
Cellular-level imaging of functionally heterogeneous DA transients in mouse cortex
Finally, we tested whether two-photon imaging with dLight1 could reveal the spatiotemporal release of DA associated with reward in the cortex. The cortex receives projection axons from both SNc and VTA. Inputs from these nuclei carry distinct dopaminergic signals influencing motor control and reward learning, respectively (
10,
35). To demonstrate the utility of dLight1 in detecting behavior-related DA signals, we broadly labeled frontal/motor cortex with AAV9.
hSynapsin1.dLight1.2, followed by two-photon imaging of dLight1-expressing layer 2/3 neurons in head-fixed mice. The animals had fully learned a visuomotor association task that required them to run in response to a visual “Go” cue in order to receive a water reward (
Fig. 5, A and B). We observed task-related DA transients, distinguishable from motion artifacts (fig. S15), across cell-sized regions of interest (ROIs) across the field of view (
Fig. 5C and fig. S16).
Aligning the DA transients to trial/standstill phase onset, we found two types of task-relevant DA responses during the reward expectation and reward delivery intervals. An average of 63% of responsive ROIs showed significantly increased DA transients that correlated with reward, which were abolished by unexpected reward omission (20% of randomly selected trials) (
Fig. 5D, right). A subset of ROIs (~37%) showed significantly increased DA transients that lasted during the short phase of “Go” stimulus presentation for both rewarded and nonrewarded trials (
Fig. 5D, left). These transient increases during the stimulus presentation phase were not caused by the stimulus appearance itself, because no significant increase in DA levels was observed during miss trials during which the animal saw the stimulus but did not respond (
Fig. 5D, yellow traces).
To investigate whether these early responses shown in 37% of ROIs reflect increased DA levels during reward expectation or correlate with locomotion, we aligned the trials at running onset (
Fig. 5E, group averages; fig. S16G, single ROIs) and compared the DA transients of runs triggered by the “Go” stimulus (when the animals expected a reward) with spontaneous runs that erroneously occurred during the standstill phase (with no reward expectation). A small subset of responsive ROIs (5%) showed significant increases in DA transients during reward expectation but not spontaneous running (
Fig. 5E, center), whereas the other 32% of ROIs correlated with locomotion (
Fig. 5E, left). The 63% of ROIs responsive to reward only (
Fig. 5D, right) also showed increased DA transients during the early stimulus presentation phase consisting of both locomotion- and reward expectation–related responses (
Fig. 5E, right). All three types of responses were consistently seen across animals. Comparing the heterogeneity of response transients between layer 1 and layers 2/3 of cortical area M1 (fig. S16, E and F), we found that layer 2/3 showed more ROIs active during reward. A similar number of ROIs responded to locomotion and reward expectation in both layers (fig. S16H). Mesocortical dopaminergic projections are thus spatially intermingled, and activation of these inputs leads to spatiotemporally heterogeneous DA signals in the cortex whose dynamics depends on motor behavior, reward expectation, and consumption.
Conclusion
We developed and applied a new class of genetically encoded indicators that overcome major barriers of current methods to permit high-resolution imaging of DA dynamics in acute brain slices and in behaving mice. The submicromolar affinity and fast kinetics of dLight1 offer fast temporal resolution (10 ms on, 100 ms off) to detect the physiologically or behaviorally relevant DA transients with higher molecular specificity relative to existing electrochemical or cell-based probes (
14). For example, in NAc of freely behaving mice, longitudinal measurements revealed different changes in time-resolved DA signals encoding either predictive cue or reward consumption across learning.
The disparate contributions of synaptic, extrasynaptic, and spillover DA events to circuit function are not addressable without fast, robust, and genetically encoded sensors. In a dorsal striatal slice, dLight1 reliably detected the concentration and time course of DA transients and their modifications by pharmacological compounds. The rapid rise of fluorescence (10 ms) and the peak concentration (10 to 30 μM) of DA after electrical stimulation indicates that the initial measures of DA are closely associated with the site of release (
26). The decline of fluorescence, particularly in the presence of cocaine, results primarily from reuptake and diffusion of DA away from release sites.
dLight1 also permits measurement of functionally heterogeneous DA transients at the cellular level with high spatial resolution. In the cortex, two-photon imaging with dLight1 revealed a DA transient map with spatially distributed, functionally heterogeneous DA signals during a visuomotor learning task. Simultaneous calcium imaging can further determine how spatiotemporal differences in DA levels relate to ongoing neural activity and influence associative learning or goal-directed behavior.
dLight1.1 and dLight1.2 are optimized sensor variants that can be immediately applied to ex vivo or in vivo studies, as they offer a good balance between dynamic range and affinity. Other dLight variants may be suitable for measuring synaptic release (dLight1.3) or tonic DA transients (dLight1.4). Given the broadly tunable affinity and dynamic range of dLight1, protein engineering and high-throughput screening efforts can further optimize the signal-to-noise ratio and molecular specificity (
36) as well as the performance of other neuromodulator indicators.
In combination with calcium imaging and optogenetics, our sensors are well poised to permit direct functional analysis of how the spatiotemporal coding of neuromodulatory signaling mediates the plasticity and function of target circuits.
Acknowledgments
We thank B. P. McGrew for assistance during in vitro sensor library screening; H. Cheng for producing AAV virus; L. Lavis (Janelia Research Campus) for providing NV-caged DA; E. Carey for performing cortical viral vector injections; and B. Mensh for critical advice in writing and revising the manuscript. This project was made possible with generous help from L. Looger (Janelia Research Campus).
Funding: Supported by NIH BRAIN Initiative U01NS090604 and U01NS013522, DP2MH107056 (L.T.); DP2NS083038, R01NS085938, P30CA014195 (A.N.); BRAIN Initiative U01NS013522 (J.T.W. and M.v.Z.); BRAIN Initiative U01NS094247 and R01NS104944 (H.Z.); NIH 1R01MH110556 (D.D.); and NIH DP2NS087949, NIH/NIA R01AG047664 (V.G.). K.M. is a DFG research fellow and recipient of a Catharina Foundation postdoctoral scholar award. V.G. is a Heritage Principal Investigator supported by the Heritage Medical Research Institute.
Author contributions: L.T. and T.P. conceived the project; T.P. and L.T. designed, screened, and optimized sensors and characterized them in mammalian cells and cultured neurons; R.L. simulated the structure of the sensor; A.M. and M.v.Z. characterized signaling properties of the sensor; W.-H.X. and H.Z. characterized the sensor in organotypic brain slices; J.T.W. performed characterization in acute brain slices; M.W.H. and D.D. characterized the sensor in vivo in the dorsal striatum; J.R.C. and V.G. performed fiber photometry recordings coupled with optogenetic manipulations, calcium imaging, and behavioral experiments in NAc, analyzed the data, and prepared the related figures and text; M.J.J. performed hybridization chain reaction experiments and prepared relevant figures and text with input from J.R.C. and V.G.; K.M., R.W.F., and A.N. performed the two-photon imaging experiments in the cortex of behaving mice, analyzed the data, and prepared the related figures and text; all authors analyzed the data; L.T. led the project; and L.T. and T.P. wrote the paper with contributions from all authors.
Competing interests: L.T., R.L., and T.P. have submitted a provisional patent application on sensor engineering.
Data and materials availability: All DNA and viruses have been deposited in NCBI (accession number MH244549-MH244561), ADDGENE, and the University of Pennsylvania Vector Core. All DNA plasmids and viruses are available from UC Davis or designated repository under a material transfer agreement. Computer codes are deposited in github (
https://github.com/GradinaruLab/dLight1/). All other data needed to evaluate the conclusion in the paper are present in the paper or the supplementary materials.