D₁ intrinsic activity and functional selectivity affect working memory in prefrontal cortex

Abstract

Dopamine D₁ agonists enhance cognition, but the role of different signaling pathways (e.g., cAMP or β-arrestin) is unclear. The current study compared 2-methyldihydrexidine and CY208,243, drugs with different degrees of both D₁ intrinsic activity and functional selectivity. 2-Methyldihydrexidine is a full agonist at adenylate cyclase and a super-agonist at β-arrestin recruitment, whereas CY208,243 has relatively high intrinsic activity at adenylate cyclase, but much lower at β-arrestin recruitment. Both drugs decreased, albeit in dissimilar ways, the firing rate of neurons in prefrontal cortex sensitive to outcome-related aspects of a working memory task. 2-Methyldihydrexidine was superior to CY208,243 in prospectively enhancing similarity and retrospectively distinguishing differences between correct and error outcomes based on firing rates, enhancing the micro-network measured by oscillations of spikes and local field potentials, and improving behavioral performance. This study is the first to examine how ligand signaling bias affects both behavioral and neurophysiological endpoints in the intact animal. The data show that maximal enhancement of cognition via D₁ activation occurred with a pattern of signaling that involved full unbiased intrinsic activity, or agonists with high β-arrestin activity.

Introduction

The prefrontal cortex (PFC) guides higher-order cognitive function through representational knowledge. Its neuronal activities have properties of cognitive representation that embody strategies for performing memory-guided behaviors^1–3 Not surprisingly, working memory (WM) deficits are observed commonly in many neurologic and psychiatric disorders, as well as in normal aging. It is clear that dopamine D₁ receptors (D₁R) have a critical role in the regulation of WM in PFC. D₁Rs stabilize PFC neuronal representations of WM, and either hyper- or hypo-activity of D₁Rs can impair WM.⁴ Several D₁R intracellular signaling pathways may be involved in this regulation, including cAMP synthesis⁵ and signaling initiated by β-arrestin recruitment.^6–8 Increased cAMP concentration has been the canonical mechanism of D₁R activation. Optimal amounts of cAMP synthesis improve neuronal representations of WM in PFC; firing is decreased in response to noise input, therefore the signal-to-noise ratio is enhanced.⁵ Recently, several studies of β-arrestin recruitment suggest that this mechanism may be possibly critical for the treatment of memory-related disorders.^6–8 Despite the evidence of cognitive enhancement resulting from D₁R stimulation, there is no centrally available D₁ agonist yet approved for this indication.

One strategy for discovering novel drugs with better therapeutic indices is known as functional selectivity or biased signaling.⁹ This involves the search for compounds that are receptor selective, but that only activate a subset of the signaling pathways of the targeted receptor. As an example, in schizophrenia, antagonism of the dopamine D₂ receptor is a hallmark for approved antipsychotic drugs, whereas D₂ agonists will induce or exacerbate psychotic symptoms. Conversely, aripiprazole, a compound with functionally selective D₂ properties (including full agonism in some systems), is an effective antipsychotic.^10,11 More recently, there have been several reports of functionally selective/biased opioid receptor ligands in which undesirable effects are decreased.^12,13 The relative importance of cAMP and β-arrestin pathways¹⁴ for a given behavior is fundamental to understanding how GPCR signaling bias can be leveraged to improve therapeutic indices. Thus, we sought to determine how D₁ activation of these pathways might regulate WM representation in the PFC. We performed in vitro screening and identified two D₁ selective compounds with similar intrinsic activity at D₁ G protein (cAMP-based) signaling, but with markedly different β-arrestin activation, that permitted us to do the first investigation on the relative roles of these two signaling pathways. Our results suggest that activation of both cAMP and β-arrestin may be required for efficient cognitive enhancement at the single unit, the micro-network, and behavioral levels.

Materials and Methods

Materials

2-methyldihydrexidine (2MDHX) was synthesized by modifications of published procedures¹⁵ (Supplementary Figure 1). CY208,243 (CY208) was purchased from Tocris (Minneapolis, MN) and SCH23390 was purchased from Sigma/RBI (St. Louis MO). SCH39166 (ecicopam) was a gift from Dr. Richard Chipkin of Psyadonrx, Inc. (Germantown MD). The compounds in the second independent biochemical experiment were synthesized by Pfizer Central Research (Cambridge MA) or purchased from Sigma or Tocris.

In vitro methods

Adenylate cyclase (cAMP) assay

Two independent experiments using different cells lines and different sources of compounds (vide supra) were conducted (Supplementary Table 1). The first experiment (data in Figure 1, Supplementary Figure 2; Table 1), used Chinese hamster ovary (CHO) cells stably transfected with human D₁R. And the radioimmunoassay for cAMP was carried out with minor modifications of previously published procedures¹⁶. In the second experiment (data in Supplementary Table 2), a stable clone (HEK293T cells) expressing high levels of hD1R was generated and frozen until screening. A homogeneous time-resolved fluorescence (HTRF) competitive immunoassay (Cisbio International 62AM4PEJ) was used to detect cAMP.

Figure 1. D₁ intrinsic activity and functional selectivity of 2MDHX and CY208.

(a & b) Representative curves of induction of β-arrestin activation at D₁R (a) and D₁R-mediated adenylate cyclase activation (b) by 2MDHX and CY208. All data are expressed as a percentage of dopamine activity. (c & d) Population analysis showed 2MDHX and CY208 have functional selectivity at β-arrestin vs. cAMP based on E_max (c) and EC₅₀ (d). Grey lines represent mean±SEM. Star indicates the comparison between 2MDHX vs CY208. Pound indicates the comparison between β-arrestin vs cAMP. (e - g) Averaged (mean±SEM) bias scaled by (f) Δlog(RA) and (g) ΔΔlog(RA). Each individual sample was shown in (e), along with the dopamine as reference. (h & i) D₁ antagonist SCH23390 (SCH) completely blocked the increases of β-arrestin (h) and cAMP (i) induced by 2MDHX and CY208. *, # indicates P<0.05; **, ## indicates P<0.005; ***, ### indicates P<0.001.

Table 1.

Intrinsic activity and functional selectivity of tested compounds.

Ligand	CAMP				β-arrestin				ΔΔlog(RA)#
	Emax, %	EC50, nM	n	Δlog(RA)*	Emax, %	EC50, nM	n	Δlog(RA)*
DA	101 ± 3	159 ± 62	20	0.0 ± 0.1	107 ± 4	373 ± 93	9	0.0 ± 0.1	0.0 ± 0.1
2MDHX	100 ± 8	151 ± 53	14	−0.1 ± 0.2	145 ± 17	302 ± 106	6	0.2 ± 0.1	0.3 ± 0.2
CY208	85 ± 7	69 ± 28	8	0.0 ± 0.1	50 ± 5	676 ± 116	7	−0.7 ± 0.1	−0.7 ± 0.1
SKF83959	38 ± 8	115 ± 18	2	−0.7 ± 0.1	40 ± 0	161 ± 6	3	−0.2 ± 0.1	0.5 ± 0.1
SKF38393	46 ± 5	289 ± 158	7	−0.6 ± 0.2	59 ± 5	451 ± 123	6	−0.4 ± 0.1	0.2 ± 0.2
Dihydrexidine	98 ± 2	98 ± 28	5	−0.1 ± 0.2	131 ± 6	1023 ± 288	3	−0.5 ± 0.1	−0.3 ± 0.1
LEK-8829	61 ± 15	127 ± 47	4	−0.5 ± 0.2	14 ± 4	506 ± 112	4	−1.2 ± 0.1	−0.7 ± 0.1
Dinapsoline	127 ± 13	54 ± 14	2	0.1 ± 0.1	107 ± 3	427 ± 109	4	−0.1± 0.1	−0.3 ± 0.1

^*Reference compound is dopamine (DA).

^#Bias vector is β-arrestin over cAMP.

Data obtained from experiment one. Each value represents mean ± se. Maximal efficacy is expressed relative to dopamine.

Assessment of β-arrestin activation

Two independent experiments using different sources of compounds (vide supra) were conducted. In both experiments, PathHunter® βarrestin assay (DiscoveRx, Fremont, CA) was used to measure β-arrestin activation in live CHO-hD₁ cells as previously reported¹⁶.

In vivo methods

Total seventeen male adult Sprague-Dawley rats (Charles River Laboratories, Wilmington, MA) were used. All animal care and surgical procedures were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and Penn State Hershey Animal Resources Program and were reviewed and approved by IACUC of Penn State College of Medicine.

Delayed alternation response test in T-maze

A standard T-maze combined with a Limelight video recording system (Actimetrics, Coulbourn Instruments, Whitehall, PA) was used. Pre-defined zone and grids (Supplementary Figure 3a) were used to assess the behavior of a subject as decision making (staying in zone A) and choice (passing grids i or ii), and mark the reference time (RT; i.e., the time for choice behavior). The detailed test procedure has been described recently.¹ Briefly, the rats were trained to alternate visits two arms of the maze after a delayed time, and only a correct choice led to a reward. When the rats were performing at the appropriate baseline (60–80% correct), they were first subcutaneously administrated vehicle (0.1% ascorbate). If performance was still at baseline, then various doses (0.1–1000 nmol/kg) of 2MDHX or CY208 were tested. All of the drug administration was twenty minutes before the task. In some experiments, an additional pretreatment of SCH39116 (0.01 mg/kg) was given one hour before the task. No subject was retested unless a minimum of five days had passed since the prior drug treatment. The duration of rats staying at the intersection was recorded as the time to make a decision. The correct rate of choice and the speed of rats running in the maze (from gate opening to intersection reaching) were calculated after the testing session.

Electrophysiological recordings

Six rats were unilaterally implanted with microwire electrodes into the PFC as described previously¹. Briefly, a microwire electrode array was implanted at 3.0 mm rostral to bregma, 0.4 mm lateral to bregma, and vertically lowered to a depth of 3.5 mm to brain surface, according to the rat brain atlas¹⁷ (Supplementary Figure 3b). The microwire electrode array was made from 25 µm stainless steel wire coated by polyimide (H-ML) having an impedance of 600–900 kΩ, arranged in a 2 × 4 configuration with 250 µm between electrodes (MicroProbes, Gaithersburg, MD), and the recording procedure follows a published protocol¹. Briefly, the OmniPlex Neural Data Acquisition System (Plexon, Dallas, TX) was used to record neural signal, and the RT signal from Limelight synchronized neural and behavioral recording. Wideband signal (digitized at 40 kHz), spike signal (high-pass filtered by a cutoff of 250 Hz), and local field potential (LFP, low-pass filtered by a cutoff of 250 Hz and downsampled to 1 kHz) were all recorded for later off-line analysis. Action potentials were detected and sorted both on-line and off-line via Offline Sorter (Plexon, Dallas, TX) to get better unit isolation results. Units with the signal to noise ratio greater than 2 were used for analysis. Waveform correlation coefficients were calculated via linear regression¹⁸ to identify single units across recording sessions from microwire electrode that remained stationary (Supplementary Figure 3c). The drug effects were tested at a dose of 10 nmol/kg which was shown to be the most effective dosage to improve cognition in a pilot dose-response study.

Data analysis

All data are reported as mean ± SEM unless otherwise specified. Depending on the analysis, we used MATLAB (MathWorks, Natick MA), SAS (Cary NC), SPSS (IBM, Armonk NY), and/or Prism (GraphPad, San Diego, CA) as noted for individual experiments.

Functional assays

The dose-response curve was calculated by a sigmoidal dose-response equation with a standard slope to obtain apparent potency (EC₅₀) and maximal efficacy of intrinsic activity (E_max). ANOVA was used to examine the difference of EC₅₀ and the difference of E_max between compounds (2MDHX vs. CY208) and between signaling pathways (β-arrestin vs. cAMP). We used currently accepted methods to calculate a quantitative measure of the signaling bias of the test compounds.^19–21 First, relative activity scale was calculated as log(RA)=log(E_max/EC₅₀). Then the bias was quantified as ∆∆log(RA) = ∆ log(RA) _beta- ∆log(RA) _cAMP, where ∆log(RA)=log(RA)_drug-log(RA)_dopamine. Dopamine was the reference compound for both cAMP and β-arrestin assays, and the vector of the bias was β-arrestin versus cAMP. This method is reported to decrease observation of assay bias, as well as system bias innate to the assays used.^19–21

Single unit activity

For the analyses of neural sensitivities to the behavioral events, spike counts were binned (0.1 s) around RT (±2 s) in each trial for each unit. Each trial in the task was divided into three epochs: baseline, prior and post choice. Prior and post choice epochs are one second before or after RT which reflects the mnemonic component of the task.¹ Baseline epoch is the combination of one second before prior epoch and one second after post epoch. ANOVA was used to examine the outcome-related activity with regard to (i) different epochs of the task (baseline, prior vs. post) and (ii) different outcome (correct vs. error). Outcome sensitivity was scaled by calculating the sensitivity index (d’), the ability to distinguish error from correct based on firing rates during the prior or post choice epochs, where $d^{\prime }=\frac{mea{n}_{error}-mea{n}_{correct}}{\sqrt{\left(s{d}_{error}^2+s{d}_{correct}^2\right)/2}}$. To examine drug effects, ANOVA was used to compare differences in firing rate and differences in d’ among drug conditions and to analyze interactions between condition _{(control, 2MDHX vs. CY208)} and outcome _{(correct vs. error)}. To construct peri-event spike histograms for groups of units, z-scores for each unit were calculated as $\frac{F{R}_i-F{R}_{mean}}{sd}$, where FR_i is the firing rate in the i^th bin of the peri-event period, FR_mean is the mean firing rate over baseline epoch, and sd is the standard deviation of firing rate for baseline epoch, with values averaged for all trials. These per-unit normalized z-scores were then averaged for groups. Paired ANOVA was used for population analysis. Tukey’s multiple comparison was used for post hoc test.

Spectral analysis

Spectral analysis was done using an open source toolbox, EEGLAB.²² The dataset included single unit spike and LFPs. Spike counts were first binned (1 ms) for each unit and then normalized to z-scores. LFPs were first normalized to z-scores and then averaged among all recorded channels for each animal. Peri-event matrices were then created around RT (±1 s) for each unit and each animal. The spike-spike and LFP-spike coherence was calculated by the function:

$$\left[coh,\sim ,\sim ,freqsout,\sim ,\sim ,\sim ,\sim ,\sim \right]=newcrossf\left(mat1,mat2,2001,\left[-10001000\right],1000,0{,}^{\prime }baselin{e}^{\prime },NaN\right).$$

The LFP oscillation power was calculated by the function:

$$\left[spectra,freqs,\sim ,\sim ,\sim \right]=spectopo\left(mat,0,1000\right).$$

For each frequency, paired ANOVA was used to compare differences in coherence and power among drug conditions. Tukey’s multiple comparison was used for post hoc testing.

Behavioral data

In the behavioral test, rats served as their own controls, thus repeated ANOVA was used to test whether the correct rate of choice, the time for decision making, and the speed of running were changed by administrations of drugs. The effects of dose were tested by pairwise multiple post comparison.

Results

Functional characterization of D₁-selective ligands:

To select ligands for this study, we screened a variety of known D₁-selective ligands in vitro, including some ligands recently reported as having high functional bias²³ (Table 1; Supplementary Table 2). We did not reproduce the high bias reported for several phenylbenzazepines, and therefore did not utilize them. We selected 2-methyldihydrexidine (2MDHX) based on its full intrinsic activity at adenylate cyclase [Figure 1a-d; Table 1; maximal efficacy (E_max) =100±8%; apparent potency (EC₅₀) =151±53 nM], and even higher intrinsic activity than dopamine at β-arrestin recruitment (E_max=145±17%; EC₅₀=302±106 nM). The second ligand selected was CY208,243 (CY208) based on its high intrinsic activity at adenylate cyclase (E_max=85±7%; EC₅₀=69±28 nM), but lower intrinsic activity at β-arrestin recruitment (E_max=50±5%; EC₅₀=676±116 nM). The D₁ antagonist SCH23390 completely blocked the increases of cAMP and β-arrestin for both ligands (Figure 1h-i; cAMP, P_{2MDHX*2MDHX+SCH23390}=0.01, P_{CY208*CY208+SCH23390=0.03}; β-arrestin, P_{2MDHX*2MDHX+SCH23390}=0.0007, P_{CY208*CY208+SCH23390}<0.0001), consistent with a D1-mediated effect.

We found that 2MDHX was more efficacious (higher E_max; Figure 1c-d; P_2MDHX*CY208<0.001) and more potent (lower EC₅₀; P_2MDHX*CY208=0.04) than CY208 at β-arrestin recruitment, whereas at cAMP synthesis the efficacy (P_2MDHX*CY208=0.22) and potency (P_2MDHX*CY208=0.40) were not significantly different. These were also reflected by their normalized relative activity values [∆log(RA); Figure 1f; β-arrestin, 2MDHX=0.2±0.1 vs. CY208=−0.7±0.1, P_{2MDHX*dopamine}=0.51, P_{CY208*dopamine}=0.01; cAMP, 2MDHX=−0.1±0.2 vs. CY208=0.0±0.1, P_{2MDHX*dopamine}=0.59, P_{CY208*dopamine}=0.98]. Of particular importance is the fact that, when comparing between signaling pathways, 2MDHX is slightly biased towards β-arrestin (Figure 1c-d; E_max, P_{cAMP*β-arrestin}=0.002; EC₅₀, P_{cAMP*β-arrestin}=0.16), whereas CY208 is slightly biased against β-arrestin (E_max, P_{cAMP*β-arrestin}=0.02; EC₅₀, P_{cAMP*β-arrestin}<0.001). These were also reflected by their bias vectors [ ∆∆log(RA) ; Figure 1g; 2MDHX=0.3±0.2 vs. CY208=−0.7±0.1, P_{2MDHX*dopamine}=0.03, P_{CY208*dopamine}<0.001]. Although it would have been ideal to have a completely biased compound, these are rare (cf., Discussion); the differences we found, however, were adequate to justify the use of 2MDHX and CY208 as paired probes to evaluate the possible role of signaling bias in the regulation of WM representation in the PFC.

Effects of 2MDHX and CY208 on single unit activity in PFC by systemic administration to rats performing the delayed alternation response (DAR) task:

PFC neurons encode outcome-related information by firing differently between correct and error trials.^1–3 As expected, when rats performed the DAR task, neurons increased their firing by ca. 1 sec near the time of choice. The change in firing was classified as increase before outcome (prospective neurons) or after outcome (retrospective neurons). We found that a large portion of recorded neurons fired higher in error trials than in correct trials, and we recorded only three neurons that had higher firing during correct outcome, consistent with previous reports that the neuronal activity in PFC tends to predict premature errors significantly better than correct trials.^1–3 Thus, the analysis focused on the former group.

Prospective neurons fired more before making error choice (correct=35±2 vs. error=42±2 spikes/s; P<0.0001), and both 2MDHX and CY208 decreased the firing of these neurons (Figure 2a-b, Supplementary Figure 4a). The major decrease was before error outcome (2MDHX=35±2 vs. CY208=35±2 spikes/s; P_{control*2MDHX}<0.0001, P_{control*CY208}<0.0001). The decrease also occurred before the correct outcome, but was a small effect (2MDHX=34±2 vs. CY208=32±2 spikes/s; P_{control*2MDHX}<0.05, P_{control*CY208}<0.0001). Interestingly, the sensitivity index (d’) of error vs. correct based on firing was significantly decreased by 2MDHX more than CY208 (Figure 2c; control=0.4±0.1, 2MDHX=0.2±0.03, CY208=0.4±0.1; P_{control*2MDHX}<0.0001, P_{control*CY208}=0.59, P_2MDHX*CY208<0.0001). This was also reflected by the interaction analysis between drug and outcome based on firing (Figure 2b; P_{(2MDHX,CY208)*(correct,error)}=0.0002). Conversely, retrospective neurons fired more after making error choice (correct=35±2 vs. error=41±2 spikes/s; P<0.0001). Both 2MDHX and CY208 decreased their firing (Figure 2d-e, Supplementary Figure 4b) with the largest decrease occurring after error outcome (2MDHX=35±2 vs. CY208=32±2 spikes/s; P_{control*2MDHX}<0.0001, P_{control*CY208}<0.0001). The decrease also occurred after the correct outcome but was a small effect (2MDHX=33±2 cs. CY208=31±2 spikes/s; P_{control*2MDHX}<0.005, P_{control*CY208}<0.0001). As opposed to prospective neurons, d’ of retrospective neurons was decreased less by 2MDHX than by CY208 (Figure 2f; control=0.5±0.1, 2MDHX=0.3±0.1, CY208=0.1±0.03; P_{control*2MDHX}=0.006, P_{control*CY208}<0.0001, P_2MDHX*CY208<0.0001). And there was a significant interaction between drug and outcome based on firing (Figure 2e; P_{(2MDHX,CY208)*(correct,error)}<0.0001).

Figure 2. Effects of 2MDHX and CY208 on single unit activity in PFC. (a & d).

Both 2MDHX and CY208 decreased firing rate (FR) of prospective neurons (a) and retrospective neurons (d). (Upper) Normalized FR for every single unit for control vs. 2MDHX and CY208. (Lower) Averaged FR of all units. (b & e) There was a significant interaction between drug and outcome based on FR for prospective neurons (b) and retrospective neurons (e). Star indicates post hoc comparison between control vs 2MDHX (red) or between control vs CY208 (blue). Pound indicates the interaction between drugs (2MDHX vs CY208) and outcomes (correct vs error). (c & f) d’of error vs. correct based on FR was significantly decreased by 2MDHX but not CY208 for prospective neurons (c), whereas it was less decreased by 2MDHX than CY208 for retrospective neurons (f). Grey lines represent each unit. *, # indicates P<0.05; **, ## indicates P<0.005; ***, ### indicates P<0.0001.

Spectral analysis of single unit spikes and local field potentials (LFPs) of PFC neuronal networks:

Single unit firing rates were synchronized with each other (Figure 3a; n=141 pairs), and 2MDHX significantly increased the coherences at all frequencies (Pcontrol_*2MDHX<0.05 for each frequency except 16 Hz where P_{control*2MDHX}=0.07). CY208 had no significant effects on delta (1–4 Hz), theta (5–8 Hz), and alpha bands (9–12 Hz, P_{control*CY208}>0.05 for each frequency), but decreased coherences of beta (27 Hz) and gamma bands (35–37 Hz, P_{control*CY208}<0.05 for each frequency). There was a significant difference between 2MDHX and CY208 regarding the change of spike-spike synchronization (P_2MDHX*CY208<0.0001). In addition, the oscillation power of LFPs was also increased by 2MDHX (Figure 3b; n=6 animals), especially at theta (7–9 Hz, Supplementary Figure 5) and beta bands (15–16 Hz, P_{control*2MDHX}<0.05 for each frequency). CY208 did not cause a significant increase (P_{control*CY208}=0.10). There was synchronization between the oscillation of single unit spikes and the oscillation of LFPs (Figure 3c; n=44 pairs). 2MDHX maintained this synchronization (P_{control*2MDHX}>0.05 for each frequency), whereas CY208 decreased it, especially at theta (6 Hz), beta (16–18 Hz), and gamma bands (35, 49–51 Hz, P_{control*CY208}<0.05 for each frequency). There was a significant difference between 2MDHX and CY208 regarding the change of LFP-spike synchronization (P_2MDHX*CY208<0.0001).

Figure 3. Effects of 2MDHX and CY208 on the neuronal network in PFC.

(a) 2MDHX significantly increased spike-spike synchronization between units measured by coherence. CY208 did not increase but even decreased it at beta and gamma bands (27, 35–37 Hz). Each line is averaged coherence of all paired units at each frequency. Star indicates P<0.05 for post hoc comparison between control vs 2MDHX (red) or between control vs CY208 (blue). (b) 2MDHX but not CY208 significantly increased the power density of LFP oscillation at theta and beta bands (7–9, 15–16 Hz, zoom in at top right corner). Same format as in A. (c) 2MDHX maintained whereas CY208 significantly decreased synchronization between LFP oscillation and unit spike oscillation at theta, beta, and gamma bands (6, 16–18, 35, 49–51 Hz). Same format as in (a).

Changes in behavioral performance caused by 2MDHX and CY208:

Both compounds produced an ‘inverted-U’ dose response in which the optimal dose (10 nmol/kg) enhanced the correct rate, whereas higher doses had no significant effect or tended to impair (Figure 4a; Supplementary Table 3; 2MDHX, P_0*10nmol/kg=0.004; CY208, P_0*10nmol/kg=0.003; all other doses P>0.05). There was no difference between 2MDHX and CY208 in enhancing the accuracy of performance (P>0.05 for all doses). Conversely, 2MDHX significantly decreased the duration of decision making at the optimal dose (10 nmol/kg; Figure 4b; Supplementary Table 3; P_0*10nmol/kg=0.04; all other doses P>0.05), whereas CY208 did not (P>0.05 for all doses). 2MDHX tended to quicken decision making better than CY208 (P_2MDHX*CY208=0.08), and the change of duration was not related to motor function as reflected by no changes in running speed^24–28 (Figure 4c; Supplementary Table 3; P>0.05 for both drugs at all doses). The cognitive improvement induced by either 2MDHX (Figure 4d; P_{control*2MDHX}=0.01) or CY208 (P_{control*CY208}=0.005) was abolished by pretreatment with the D₁ antagonist SCH39166 (P_{control*2MDHX+SCH39166}=0.53; P_{control*CY208+SCH39166}=0.60), consistent with these effects being mediated by the D₁R.⁵

Figure 4. Effects of 2MDHX and CY208 on cognitive performance.

(a) Both 2MDHX and CY208 produced an inverted-U dose effect on DAR task whereby low doses improved but high doses had no effect or tended to impair the accuracy of performance as measured by correct rate. (b) 2MDHX but not CY208 improved decision making at a low dose as measured by duration, whereas at high dose they both tended to impair it. (c) 2MDHX and CY208 did not change motor function as measured by running speed. Star indicates the comparison between control vs 2MDHX (red) or the comparison between control vs CY208 (blue). (d) Pretreatment of D₁ antagonist SCH39166 (SCH) blocked cognitive enhancement induced by 2MDHX or CY208. * indicates P<0.05. ** indicates P<0.01.

Discussion

This study was, to our knowledge, the first to examine how the signaling bias of a drug affected both behavioral and neurophysiological endpoints in the intact animal. D₁ agonists have been reported to change neuronal WM representation in PFC by decreasing firing,⁵ and this occurred in our study for both 2MDHX and CY208. Interestingly, along with the overall decrease, dissimilar effects were observed for 2MDHX versus CY208. 2MDHX was superior to CY208 because it prospectively enhanced similarity and retrospectively maintained the difference in firing between correct and error trials. As the discernment of firing between correct and error trials, measured by d’ in this study, encodes the outcome-related information,^1–3 we interpret our findings as follows. First, neurons in PFC fire at an appropriate level to achieve a correct outcome with greater firing predicting error outcomes.^29–31 2MDHX reduced firing to the ‘correct outcome level’, increasing the possibility of a correct outcome. CY208 caused similar changes, but of lower magnitude than 2MDHX. Second, once error outcomes appear, neurons in PFC increase firing, providing an adjusting feedback.^2,29–32 2MDHX maintained the higher firing after error outcome better than CY208, increasing the possibility of correct outcome for the next trial. Thus, although both compounds positively affected these processes, 2MDHX produced greater improvement. The differential enhancement at the single neuron level did not lead to dramatically improved behavioral performance. There was similar performance accuracy, but only 2MDHX shortened decision-making. The two most likely explanations are either that there were inadequate differences in signaling bias between these two compounds to have dramatic behavioral effects, and/or that multiple D₁-mediated signaling pathways influence behavioral outcome and one can compensate for another.

Along with single neuron activity, the neuronal micro-network measured by the oscillations of LFPs and spikes was also affected by 2MDHX and CY208. The strongest influences were for theta, beta, and gamma bands that are associated with WM, and that are impaired in many neuropsychiatric disorders.^2,33,34 Interestingly, 2MDHX clearly enhanced or maintained coherence and power of oscillation, whereas CY208 had no significant effects or, at some frequencies, actually impaired these measures. Thus, both compounds affected neuronal activity, but 2MDHX produced greater cortical physiological improvements including both single neuron and micro-network. The important question, then, is whether these changes are due to signaling differences at the D₁ receptor, or other mechanisms.

Do the current data reflect actions at D₁ receptors?

Pharmacological studies often have great potential to translate to the clinic, as well as provide a relatively rapid route to understanding underlying mechanisms. As an example, dihydrexidine, the parent compound of 2MDHX, has been shown to have positive effects on cognition in murine^35,36 and in primates, both non-human³⁷ and human.³⁸ Yet one of the major mechanistic limitations can be uncertainties from the possible contribution from off-target effects. With regards the current study, there are two inherent issues. First, although we have used the term “D₁” throughout this manuscript, we are keenly aware that there are no known orthosteric ligands that have marked selectivity for the D₁ vs. the highly homologous D₅ dopamine receptor in rats. In the current studies, this is not a confound - in murine species, neither D₅ binding sites, nor D₅ mRNA, are found in the cortex, but rather are largely confined to the hippocampus.^39,40 In primates, however, there are D₅ receptors co-localized with D₁ in the prefrontal cortex, although the density of D₁ is far higher,^41,42 so this will be an interesting issue to resolve as studies are translated to higher species.

We have also considered the effect of interactions at other receptors, especially the D₂ dopamine receptor that represents a potential off-target for these two compounds. Two lines of evidence suggest that the results were not affected by D₂ actions. In many pharmacological studies, the drug doses or concentrations that are used are likely to engage secondary targets (e.g., see ⁴³). In the current study, the maximal enhancement of performance for both compounds was seen at 10 nmol/kg which, if the dose was absorbed into the circulation instantaneously and not metabolized, would equate to a maximal whole body concentration of ca. 10 nM. The actual concentration would, of course, be far less, meaning that receptor occupancy would be very low based on the 9 nM K_I of 2MDHX.¹⁵ The next highest affinity for 2MDHX is for the dopamine D₂ at which 2MDHX is more than 30-fold D₁:D₂ selective.¹⁵ This would lead to a prediction of essentially negligible receptor occupancy at the D₂ or other receptors at the low doses used in these experiments. There are additional data and discussion in the Supplementary material, but we recognize that future studies using orthogonal methods will be important to test these hypotheses further.

What have we learned about the role of D₁ signaling in cognitive function?

The current study has emphasized two signaling pathways clearly linked to D₁R: cAMP synthesis and β-arrestin recruitment. As reported previously,^15,44 both 2MDHX and CY208 have high intrinsic activity at the canonical cAMP pathway in assay systems that can detect partial agonists like SKF38393. Conversely, there was a marked difference found at β-arrestin recruitment. Some D₁ agonists have previously been shown to have bias towards cAMP vs. receptor desensitization⁴⁵ or internalization.^46,47 In the current data for β-arrestin recruitment, CY208 was a low intrinsic activity partial agonist, whereas 2MDHX was a “super agonist”. Although the degree of bias was incomplete, these findings permitted us to compare these two compounds as probes to study the possible role of D₁ signaling bias in cognition (Supplementary text). Our data clearly show that 2MDHX and CY208, two drugs with different D₁ signaling bias, differentially influence cognitive regulation. The pressing question is whether it is the bias toward β-arrestin recruitment, or the subtle differences in cAMP activation, that are responsible for differences in WM representation in the PFC.

A prior study⁵ found that stimulation of D₁Rs regulates cognition via cAMP synthesis such that a cAMP inhibitor blocks the effects of D₁ agonists on neuronal activity in PFC and cognitive performance. In the current study, both 2MDHX and CY208 had high intrinsic activity at cAMP synthesis, suggesting that the overall decrease of firing requires cAMP synthesis. Along with the overall decrease, 2MDHX and CY208 caused different patterns of decrease, with the discernment of correct and error trials based on firing differently affected by 2MDHX and CY208. Neuronal oscillation was also influenced differently by 2MDHX and CY208. As intrinsic activity at β-arrestin recruitment was the largest difference between 2MDHX and CY208, it is reasonable to hypothesize that β-arrestin related signaling may also be important for neuronal WM representation in PFC.

It is clear that β-arrestin can be an important modulator of dopamine function. Rats or mice with mutations that eliminate β-arrestin recruitment have less locomotor activity,⁴⁸ more dyskinesia-like behavior,⁶ enhanced adiposity,⁴⁹ and impaired memory reconsolidation.⁸ Excessive stimulation of D₁R (e.g., during stress) impairs cognition,⁴ thus greater β-arrestin recruitment may increase the internalization and desensitization of D₁Rs,⁵⁰ and therefore reduce D₁R detrimental actions. On the other hand, β-arrestin also can evoke signaling (e.g., ERK/MAP kinase signaling and Akt/GSK3 signaling¹⁴), implying that β-arrestin-related signaling may play a role in the differentiation of 2MDHX and CY208 influencing on cognition.

It is, however, also possible that subtle differences in a G protein-mediated pathway are the defining mechanism that explains the differences between these two compounds. A previous non-human primate study compared a full D₁ agonist (dihydrexidine) with another partial agonist, and concluded that only the full agonist could improve cognition of young monkey, whereas both worked on old animals,³⁷ highlighting the importance of canonical cAMP signaling of D₁. The initial D₁ characterization of CY208 was done in bovine retina preparations in which, unlike the current study, CY208 was reported as a partial agonist with intrinsic activity not significantly different than the known partial agonist SKF38393 (68 vs. 58%). To our knowledge, there have been no studies that have specifically looked at how reduced receptor reserve^51,52 might affect the intrinsic activity of CY208 in rat brain, hence, the role of canonical signaling via the D₁ receptor cannot be ruled out as a crucial factor.

Although this study, like those with many other receptor systems, focused on cAMP and β-arrestin signaling, the results only associate the behavioral and physiological differences with the in vitro pattern of signaling. Although our data provide a very strong case for these effects being due to actions at the D₁ receptor, they do not provide direct evidence that it is actually cAMP or β-arrestin that mediate the differences (much as is the case for studies with other receptor systems). Bias is likely to be seen in many signaling pathways affected by this receptor, and to different degrees. Nonetheless, when in vitro functional selectivity leads to unexpected functional changes in situ, it makes elucidation of the exact signaling mechanism even more important both heuristically and in the discovery of more efficacious new drugs.

Clinic implications and future research

Since the first report using the full agonist dihydrexidine,⁵³ D₁ agonists have been widely reported to improve cognitive performance, but with biphasic dose-response curves (i.e., performance was improved by low doses, but impaired by high doses⁴). This was found for both 2MDHX and CY208, with ca. 10 nmol/kg (ca. 3 µg/kg) being the most effective dosage for cognitive enhancement in rats. Of note, although performance accuracy was improved by both 2MDHX and CY208, only 2MDHX improved the quickness of decision making. D₁ agonists have been hypothesized to be useful in several neuropsychiatric and neurological disorders,^4,54 but despite their promise, none have yet been approved clinically although at least one is now in later-stage clinical development.^55,56 In particular, the hypothesis that D₁ activation could improve cognitive function has received some support from some clinical studies,³⁸ but not others,⁵⁷ although these studies are difficult to interpret because the D₁ agonist that was used (dihydrexidine) has an exceptionally short half-life in humans, limiting the types of studies that can be done. New compounds with better pharmacokinetic properties (e.g., 2MDHX) will be needed to explore this approach, but the current data also suggest that functionally selective compounds may affect therapeutic specificity. Of note, it was known that a full D₁ agonist (i.e., dihydrexidine) would improve working memory in both young and aged primates, whereas a partial agonist (i.e., SKF38393) only was effective in aged primates.³⁷ The current experiments were performed on young adult rats and show that even in young animals, drugs with similar canonical intrinsic activities have different physiological and behavioral effects. It would be interesting to determine if dopamine-depleted or aged rats had an amplified difference to these two compounds based on reports of full versus partial agonists.³⁷

The current findings provide the first neurophysiological evidence that the functional selectivity of D₁ signaling has an effect on cognitive function. There are numerous experimental avenues that are left to explore, but this study is the first to demonstrate that the pattern of signaling engagement of the D₁ receptor will be critical to maximizing its cognitive-enhancing effects. The results underscore the importance of both the discovery of novel D₁ ligands and further exploration of the involved mechanisms.