Functional dynamics of dopamine synthesis during monetary reward and punishment processing

Synthesis model

To assess task-relevant changes in dopamine signaling during cognitive performance we developed a novel approach, based on the dynamic regulation of neurotransmitter synthesis. As most neurotransmitters cannot pass the blood brain barrier, they are synthetized in the brain through precursor molecules. For dopamine, the main pathway is the conversion of l-tyrosine to l-3,4-dihydroxyphenylalanine (DOPA) via the enzyme tyrosine hydroxylase, and then to dopamine by aromatic amino acid decarboxylase (AADC). Importantly, these enzymes are subject to fast-acting regulatory mechanisms. Tyrosine hydroxylase and AADC activities increase with neuronal firing in order to refill the synaptic vesicles with de novo synthetized neurotransmitter after stimulus-induced dopamine release and are further regulated by activation or blockade of dopamine receptors.^30–34 Moreover, the radioligand 6-[¹⁸F]FDOPA can be incorporated into this synthesis chain, as it is a substrate for AADC, rapidly forming 6-[¹⁸F]F-dopamine. The radioligand is thus specific to the dopaminergic pathway^35,36 and represents an established approximation for dopamine synthesis rates.^34,37 Taken together, the evidence suggests that stimulus-induced activation of dopamine synthesis is also reflected in a proportionally increased radioligand binding (Figure 1(a) and (b)).

OPEN IN VIEWER

Figure 1.

Synthesis model: (a) the neurotransmitter dopamine (DA) is synthetized from its precursor dihydroxyphenylalanine (DOPA) by the enzyme aromatic amino acid decarboxylase (AADC). Use of the radioligand 6-[¹⁸F]FDOPA as substrate for AADC is a well-established approach to estimate dopamine synthesis rates at baseline; (b) neuronal stimulation leads to dopamine release, but also increases AADC activity to refill synaptic vesicles with de novo synthetized neurotransmitter, which in turn is reflected in higher radioligand uptake as indicated by arrow thickness; (c) the proof of concept experiment showed a marked increase in striatal dopamine synthesis rates Ki during performance of the monetary incentive delay task. The ventral striatum (VStr) region of interest is outlined in black and indicated by arrows, exhibiting increases in K_i = 0.017 and 0.022/min from baseline for two subjects.

This hypothesis can be directly tested by the application of 6-[¹⁸F]FDOPA within the framework of functional PET imaging.^28,38 Similar to fMRI, fPET employs cognitive paradigms in repeated periods of task performance with an alternating control condition, thereby enabling the assessment of task-induced changes of multiple conditions within a single measurement. The radioligand 6-[¹⁸F]FDOPA is particularly suited for this application, as a bolus + infusion protocol ²⁹ further emphasizes its apparently irreversible binding characteristics^37,39 (Suppl. Fig. S1b and e), which in turn allows to identify task-specific changes in dopamine synthesis with high temporal resolution.

Participants

In total, 41 healthy participants were recruited for this study. Three subjects were excluded as the fPET measurement failed for technical reasons or urinary urgency. Two subjects participated in the proof of concept experiment (age 19.8 and 20.8 years, both female). The main study included 36 participants, who underwent behavioral testing with the MID task (24.5 ± 4.3 years, 18 female). Of those, 16 participants also completed fPET and fMRI examinations (24.8 ± 4.8 years, 7 female). Men and women did not differ regarding their age in the full sample (n = 36, p = 0.69) or the imaging subsample (n = 16, p = 0.51). Please see supplement for further details.

After detailed explanation of the study protocol, all participants gave written informed consent. Participants were insured and reimbursed for their participation. The study was approved by the Ethics Committee (ethics number: 2259/2017) of the Medical University of Vienna and procedures were carried out in accordance with the Declaration of Helsinki.

Cognitive task

Reward and punishment processing was assessed using the well-established ¹ and previously employed ⁴⁰ MID task. Here, participants aim to maximize gain and avoid loss by fast reaction upon presentation of a target stimulus.

As the crucial aspect of the paradigm is the time limit of the reaction, we employed an adaptive algorithm to control the probability for gain and loss. First, the initial reaction time was individually determined directly before each testing procedure (imaging/behavior). Second, the time limit was decreased (increased) during the paradigm if the reaction was fast enough (too slow), to maintain a probability of approximately 0.5. Third, for the main study the time limit was increased (decreased) in the beginning and middle of each task block, which enabled separation of gain and loss by increasing (decreasing) the probability for each condition. The last step allowed assessment of both conditions in a single scan. Please see supplement for a detailed task description.

Magnetic resonance imaging

MRI data was obtained on a 3T Magnetom Prisma scanner (Siemens Healthineers) using a 64 channel head coil. A structural MRI was acquired with a T1-weighted MPRAGE sequence (TE/TR = 2.29/2300 ms, voxel size = 0.94 mm isotropic, 5.3 min), which was used to exclude gross neurological abnormalities and for spatial normalization of fPET data. fMRI data was acquired using an EPI sequence (TE/TR = 30/2050 ms, voxel size = 2.1 × 2.1 × 2.8 mm + 0.7 mm slice gap).

Positron emission tomography

The radioligand was freshly prepared every day by Iason GmbH or BSM Diagnostica GmbH. One hour before start of the fPET measurement, each participant received 150 mg carbidopa p.o. to block peripheral metabolism of the radioligand by aromatic amino acid decarboxylase. ³⁷ fPET imaging was carried out using an Advance PET scanner (GE Healthcare). The radioligand 6-[¹⁸F]FDOPA was administered in a bolus + constant infusion protocol (ratio 20:80) similar to previously described procedures^28,29 (see supplement). During the scan the MID task was carried out at 10 (except for the PoC experiments), 20, 30 and 40 min after start of the radioligand application, each lasting for 5 min. Otherwise, a crosshair was presented and subjects were instructed to keep their eyes open and avoid focusing on anything specific (in particular not the task).

Blood sampling

Arterial blood samples were drawn from the radial artery (see supplement). Manual samples of plasma to whole blood ratio were fitted with a linear function. Correction for radioactive metabolites was based on previous literature, assuming that the only relevant metabolite is 3-O-methyl-6-[¹⁸F]FDOPA (3-OMFD) after carbidopa pretreatment.^37,39,41 Thus, the 3-OMFD fraction was taken from previous bolus data and modified to match our bolus + infusion protocol (see supplementary methods and Suppl. Fig. S1d). By calculation, the protocol with a bolus:infusion ratio of 20:80 indicated a 51.4% reduction of the 3-OMFD fraction (area under the curve) as compared to a pure bolus. The final arterial input function was then obtained by multiplication of the whole blood curve with the plasma to whole blood ratio and the parent fraction. The 3-OMFD input function was calculated likewise by using the 3-OMFD fraction.

Quantification of dopamine synthesis rates

Image preprocessing was done as described previously ²⁹ using SPM12 and default parameters unless specified otherwise. fPET images were corrected for head motion (quality = 1, registered to mean) and the resulting mean image was coregistered to the T1-weighted structural MRI. The structural scan was spatially normalized to MNI space and the resulting transformation matrices (coregistration and normalization) were applied to the dynamic fPET data. Images were smoothed with an 8 mm Gaussian kernel, masked to include only gray matter voxels and a low-pass filter was applied with the cutoff frequency set to 2.5 min.

6-[¹⁸F]FDOPA time activity curves (TAC) were corrected for the 3-OMFD component using the occipital cortex as reference region. We employed a mathematical correction procedure as this avoids overcorrection compared to simple subtraction of the raw reference TAC.^42,43 Briefly, the 6-[¹⁸F]FDOPA reference TAC was extracted with the Harvard-Oxford atlas and adjusted for potential task effects and movement using the same general linear model as described below. The reference TAC was fitted with a one-tissue compartment model in PMOD 3.5. The brain TAC representing the 3-OMFD component was then calculated as convolution of the 3-OMFD arterial input function with the impulse response function given by the fitted values of K₁ and k₂. The estimated 3-OMFD TAC was then subtracted from every brain voxel. This procedure assumes that the reference region is devoid of specific binding, that the distribution volumes of 3-OMFD and 6-[¹⁸F]FDOPA are equal in the reference region and that the distribution volume of 3-OMFD is uniform across the brain. A fixed whole blood component of 5% was used for all calculations.

The general linear model was used to separate task effects from baseline synthesis (Suppl. Fig. S1c). This included one regressor for each task block (except for the PoC experiments where a single task regressor was used) with a slope of 1 kBq/frame, one representing baseline dopamine synthesis and one for head motion (first principal component of the six motion regressors). As discussed in our previous work, ²⁸ such a task regressor assumes that task changes are constant throughout a block. The baseline was defined as average time course of all gray matter voxels, excluding those activated in the corresponding fMRI acquisition (contrast success > failure, p < 0.001 uncorrected) and those identified in a recent meta-analysis of the MID task (contrasts reward/loss anticipation and reward outcome, Suppl. Fig. S2). ¹ The Gjedde–Patlak plot was then applied to compute the net influx constant K_i as index of dopamine synthesis for baseline and task effects separately (Suppl. Fig. S1e). The slope was fitted from t* = 25 min, which is half of the scan time. The four task blocks were finally weighted according to task performance (actual gain/possible gain, similar for loss) and averaged to obtain task specific K_i for gain and loss. To assess the specificity of the findings, task-specific changes in dopamine synthesis rates were also calculated as percent signal change from baseline with

PSC Ki = Ki task / Ki baseline * 1 00

(1)

Neuronal activation

Task-induced neuronal activation was computed as described previously using SPM12. ⁴⁰ fMRI BOLD images were corrected for slice timing differences (reference: middle slice) and head motion (quality = 1, registered to mean), spatially normalized to MNI space and smoothed with an 8 mm Gaussian kernel. Neuronal activation was estimated across the two runs with the general linear model including one regressor for each cue (gain, loss, neutral), one for the target stimulus and one for each of the potential outcomes (gain, omitted gain, loss, avoided loss, neutral) as well as several nuisance regressors (motion, white matter, cerebrospinal fluid). To obtain an index of reward outcome ¹ which is as similar to fPET as possible, parameter estimates were combined as (gain + avoided loss) – (omitted gain + loss). Percent signal changes were computed as

PSC fMRI = β task / β baseline * 1 00 * peak

(2)

Statistical analysis

All statistical tests were two-sided and corrected for multiple comparisons with the Bonferroni-Holm procedure (e.g., when testing multiple conditions and/or groups) and the reported p-values have been adjusted accordingly.

For behavioral data, the accumulated amount of money that was gained and lost during the corresponding task blocks of the MID were assessed with one sample t-tests against zero, whereas sex differences were computed by independent samples t-tests. Due to the adaptive nature of the MID task the reaction times were normalized to the mean by subtracting the average reaction time within each block. Differences in reaction times were evaluated by repeated measures ANOVA with the factors sex and amount. Post-hoc t-tests were used to assess sex differences for each amount of money. Furthermore, we modeled the relationship between reaction time and amount with a stepwise linear regression up to second-order polynomial functions. Stepwise regression choses the model that best explains the data based on statistical significance. This was done across the entire group (n = 36) to test for a general relationship. Subsequently, parameters of the resulting models were also estimated individually for the fPET subjects (n = 16) to assess the correlation with task-specific changes in dopamine synthesis using Spearman’s correlation (since n < 10 in each group for fPET).

For imaging parameters, the primary region of interest was the VStr due to its pivotal importance in reward processing.^2,3 Therefore, values of K_i and PSC_fMRI were extracted for this region using the Harvard Oxford atlas as provided in FSL (termed “nucleus accumbens” in the atlas). For comparison, a functional definition of the VStr was also employed (neuronal activation of reward outcome ¹ within the striatum), which comprised 2.55 cm³ (in contrast to the nucleus accumbens of the Harvard-Oxford atlas with only 1.38 cm³). Task-specific changes in dopamine synthesis rates were evaluated by one sample t-tests against zero for gain and loss separately. Similarly, for K_i and PSC_fMRI the difference of gain vs. loss was calculated and assessed by one sample t-tests against zero (i.e., being identical to a paired samples t-tests). Finally, sex differences in K_i and PSC_fMRI were addressed using an independent samples t-test.

In an exploratory analysis K_i values were extracted from the caudate and putamen as defined by the Harvard-Oxford atlas and investigated in the same manner as the VStr.

Proof of concept

To assess the feasibility of the proposed synthesis model an initial PoC experiment was conducted. Two subjects underwent fPET imaging with the radioligand 6-[¹⁸F]FDOPA while performing the MID task. In both subjects the task induced substantial increases in VStr dopamine synthesis of K_i = 0.017 and 0.022/min from baseline (Figure 1(c)), supporting the feasibility of the approach to assess task-specific changes in dopamine neurotransmission.

Behavioral data

Since the PoC experiment combined monetary gain and loss within a task block, the main study specifically aimed to disentangle these two effects on a behavioral (n = 36) and neurobiological level (n = 16, see below). The task was extended to four blocks and each of them manipulated to enable the separate assessment of monetary gain and loss.

Behavioral data showed that average monetary gain and loss were significantly different from zero (all t = 10.0 to 12.6, p = 1.7 × 10⁻⁸ to 2 × 10⁻⁹, Figure 2(a)), indicating successful task manipulation. Women gained significantly more than men (5.6 ± 2.4 € vs. 4.2 ± 1.7 €, t = 2.1, p = 0.047), but both groups showed similar loss (–5.3 ± 1.9 € vs. –5.5 ± 1.9 €, p = 0.8).

OPEN IN VIEWER

Figure 2.

Behavioral data (blue = men, red = women): (a) the monetary incentive delay task was manipulated by modifying the reaction time limit for a successful trial completion, which enabled separate assessment of monetary gain and loss. During the gain task block women earned significantly more money than men (5.6 ± 2.4 € vs. 4.2 ± 1.7 €, t = 2.1, *p = 0.047), but both groups showed similar loss (–5.3 ± 1.9 € vs. –5.5 ± 1.9 €, p = 0.8). Lines represent average accumulated monetary amount at each trial; (b) the association between the individually normalized reaction times and the gained/lost amount of money was modelled by a linear relationship in men (reaction time = –0.18 – 2.36 * amount, p_linear = 0.0012, p_quadratic = 0.11), where a steeper negative slope indicated faster reaction time and thus higher sensitivity for reward. In contrast, the association was characterized by an inverted U-shaped function in women (reaction time = 4.31 – 0.95 * amount – 1.16 * amount ² , p_linear = 0.2, p_quadratic = 0.001). Since these two functions exhibit the most pronounced difference for high amounts of loss, a strong negative quadratic term was interpreted as high sensitivity for punishment. Circles denote average values for each amount and lines are model fits across the entire data set (n = 18 women and 18 men, see Suppl. Fig. S3 for details), reaction times are mean centered due to the adaptive nature of the MID task.

The difference in monetary gain was also reflected in the normalized reaction times, with a main effect of sex (F_(1,34) = 6.9, p = 0.013) and amount (F_(5,170) = 4.4, p < 0.001) as well as a trend for an interaction effect sex * amount (F_(5,170) = 2.0, p = 0.08). Post-hoc t-test indicated that this seemed to be driven by the –3 € condition with women showing a faster reaction than men (t = 2.0, p = 0.049).

We further aimed to model the behavioral response in more detail, as the relationship between reaction time and amount for each group. In men, this was best described by a negative linear function (reaction time = –0.18 – 2.36 * amount, p_linear = 0.0012, p_quadratic = 0.11), with a faster reaction for higher monetary gains (Figure 2(b), Suppl. Fig S3). In contrast, the association in women was characterized by an inverted u-shaped function (reaction time = 4.31 – 0.95 * amount – 1.16 * amount ² , p_linear = 0.2, p_quadratic < 0.001), with faster reaction times for high amounts of loss as compared to men. These distinct relationships for men and women were also obtained for the imaging subsample (men p_linear = 0.009, p_quadratic = 0.08; women: p_linear = 0.7, p_quadratic = 0.03). Thus, we interpreted the linear (quadratic) term for men (women) as index for reward (punishment) sensitivity, that is, the more negative the parameter, the faster the reaction time for high gain (loss).

Functional dynamics in dopamine synthesis

To assess reward-specific changes in dopamine synthesis, 16 of the above subjects also underwent fPET with the radioligand 6-[¹⁸F]FDOPA (seven female). The MID task yielded increased dopamine synthesis rates in the VStr during gain (men: K_i = 0.014 ± 0.004/min, women: K_i = 0.012 ± 0.004/min) and loss (men: K_i = 0.009 ± 0.005/min, women: K_i = 0.019 ± 0.003/min, all t = 6.0 to 16.1, all p < 0.001, Figure 3(b) OPEN IN VIEWER

OPEN IN VIEWER

Figure 4.

Comparison between fPET and fMRI: (a) in men the task-specific changes in VStr dopamine synthesis K_i were higher for gain than for loss (⁽*⁾p = 0.06). In contrast, women showed the opposite pattern with higher changes in dopamine synthesis during loss vs. gain (***p < 0.001), leading to a significant difference between the two groups (t = 4.1, ^###p < 0.001); (b) although neuronal activation obtained with BOLD fMRI indeed showed robust VStr signal changes for the contrast gain versus loss for men and women (t = 5.5 to 6.9, ***p = 0.0005 to 0.0006), there was no significant difference between the two groups (p = 0.4). Boxplots indicate median values (center line), upper and lower quartiles (box limits) and 1.5 * interquartile range (whiskers).

Exploratory assessment of differences between the first and second task blocks, showed a trendwise increase in dopamine synthesis for gain in men (p = 0.05). However, no such difference was observed for women or the loss condition for both sexes (all p > 0.25), indicating no systematic influence of the task timing on dopamine synthesis.

Proceeding from the distinct models to characterize the behavioral response of monetary gain and loss (Figure 2(b)), we assessed the relationship between individual model parameters (linear and quadratic terms) and task-specific dopamine synthesis rates. This resulted in an association in men between the linear term and VStr dopamine synthesis during gain (n = 9, ρ = –0.67, p = 0.059, Figure 3(c)). On the other hand, the quadratic term in women was positively associated with VStr dopamine synthesis of gain vs. loss (n = 7, ρ = 0.79, p = 0.048, Figure 3(d)).

Sex differences

Finally, VStr dopamine synthesis rates between gain vs. loss were significantly higher for men than women (t = 4.1, p < 0.001, Figure 4(a)). This sex difference was similarly present when using a functional delineation of the VStr ¹ (t = 3.9, p = 0.002), for percent signal change from baseline (t = 3.8, p = 0.002) and without weighting by task performance (t = 3.1, p = 0.009).

Exploratory analysis showed a sex difference for the putamen (t = 3.3, p = 0.005). This was driven by higher dopamine synthesis for loss than gain in women (t = –3.3, p = 0.03), as men did not show any difference between the two conditions (p = 0.2, Suppl. Fig. S4). The caudate indicated no sex difference and also no difference between gain and loss (all p > 0.1). Furthermore, no significant associations between synthesis rates and behavioral values for putamen and caudate were found (all p > 0.2).

For direct comparison we also assessed neuronal activation, where the same subjects as in the fPET experiment also underwent fMRI. In line with previous reports ¹ we observed robust neuronal activation in the VStr for gain vs. loss in men and women (t = 5.5 to 6.9, all p < 0.001, Figure 4(b)). As expected, there was however no significant sex difference in activation for the atlas-based (p = 0.4) or the functional delineation of the VStr (p = 0.3).

Similar to another study, ¹⁹ our method was able to replicate previously observed sex differences in baseline VStr dopamine synthesis (men: K_i = 0.009 ± 0.001/min, women: K_i = 0.012 ± 0.002/min, t = 3.0, p = 0.009). It is however unlikely that these baseline differences affect the task-specific estimates (see limitations).