Levodopa suppresses grid-like activity and impairs spatial learning in novel environments in healthy young adults

Abstract

Accumulated evidence from animal studies suggests a role for the neuromodulator dopamine in memory processes, particularly under conditions of novelty or reward. Our understanding of how dopaminergic modulation impacts spatial representations and spatial memory in humans remains limited. Recent evidence suggests age-specific regulation effects of dopamine pharmacology on activity in the medial temporal lobe, a key region for spatial memory. To which degree this modulation affects spatially patterned medial temporal representations remains unclear. We reanalyzed recent data from a pharmacological dopamine challenge during functional brain imaging combined with a virtual object-location memory paradigm to assess the effect of Levodopa, a dopamine precursor, on grid-like activity in the entorhinal cortex. We found that Levodopa impaired grid cell-like representations in a sample of young adults (n = 55, age = 26–35 years) in a novel environment, accompanied by reduced spatial memory performance. We observed no such impairment when Levodopa was delivered to participants who had prior experience with the task. These results are consistent with a role of dopamine in modulating the encoding of novel spatial experiences. Our results suggest that dopamine signaling may play a larger role in shaping ongoing spatial representations than previously thought.

Introduction

Dopaminergic signaling, which is pervasive in the brain, is widely implicated in a multitude of cognitive processes, ranging from decision-making, learning and memory, including working memory and spatial memory, to reward prediction (Vijayraghavan et al. 2007; Bromberg-Martin et al. 2010; Lisman et al. 2011; Kempadoo et al. 2016; Schultz et al. 2017). The medial temporal lobe, important for the encoding and retrieval of spatial and episodic memory (Burgess et al. 2002), receives dopaminergic input via multiple sources, both from the ventral tegmental area (VTA) and via co-release with noradrenaline from the locus coeruleus (Kempadoo et al. 2016; Takeuchi et al. 2016; Duszkiewicz et al. 2019). Evidence from animal studies shows that dopamine signaling in the hippocampus and entorhinal cortex (EC) enhances the formation of episodic memory and the learning of novel spatial information. This occurs via binding to dopamine receptors of the D1/D5 type (Lisman et al. 2011; Kempadoo et al. 2016). Relatedly, multiple lines of evidence have linked the spatially correlated neuronal activity in the hippocampal formation to dopaminergic signaling. For instance, dopaminergic transmission was found to affect both the stability (Kentros et al. 2004) and the flexibility of place cell firing (Tran et al. 2008) as well as the reorganization of place cell activity during the learning of new reward locations or spatial rules (Retailleau and Morris 2018; Kaufman et al. 2020).

Complementing hippocampal place cells, the EC harbors neurons representing multiple aspects of space, including grid-like firing patterns (Hafting et al. 2005), which are assumed to support path integration (McNaughton et al. 2006). Other entorhinal neurons represent head direction (Sargolini et al. 2006), speed (Kropff et al. 2015), and the distance to borders (Solstad et al. 2008). Whether dopamine modulates these spatial firing pattern remains unclear. In this regard, recent studies have described a reorganization of entorhinal grid cell activity in tasks requiring memory for reward locations (Boccara et al. 2019; Butler et al. 2019), consistent with a potential influence of dopaminergic reward-related signaling on grid cell firing. One line of explanation for how dopamine might drive these changes in spatial firing is inspired by the temporal firing pattern of dopaminergic neurons in the VTA, which resembles prediction error signals (Schultz et al. 1993; Gardner et al. 2018), well suited to support learning of reward locations (Foster et al. 2000; Gönner et al. 2017; Sosa and Giocomo 2021). A yet different mechanism by which dopamine may impact spatial representations is its modulatory effect on the excitability of both entorhinal principal cells and interneurons (Cilz et al. 2013; Jin et al. 2019). This likely alters the delicate balance between excitation and inhibition that appears necessary for grid cell activity (Christensen et al. 2021; Sosa and Giocomo 2021).

While the role of dopamine in spatial learning is well established in the animal literature, there is a scarcity of studies addressing the role of dopamine in human spatial learning, particularly in healthy young adults. Indirect evidence for a dopaminergic influence comes from studies which showed that dopamine impacts working memory function in humans (Cools and D’Esposito 2011; Garrett et al. 2015), which plays a key role in spatial learning (Weisberg and Newcombe 2016; He et al. 2021). In this regard, the inverted U-function hypothesis of dopamine suggests that increasing dopamine availability beyond optimal levels (expected to be present in healthy young adults) may impair performance (Cools and D’Esposito 2011). However, only few studies have directly addressed effects of dopamine on spatial navigation. In Parkinson’s disease patients, dopaminergic medication showed differential effects on striatal versus hippocampal spatial learning strategies (Thurm et al. 2016). A recent pharmaco-imaging study used a virtual spatial navigation task to assess the effects of L-Dopa on spatial learning and memory in a sample of healthy young and older adults. This study found that effects of L-Dopa on spatial learning and medial temporal lobe activity depended on age and the order of L-Dopa/Placebo administration (Baeuchl et al. 2023).

An underexplored question is whether L-Dopa’s effect on spatial memory might reflect its impact on neural representations of space (cf. Koch et al. 2022) that may implicate also grid cells, in addition to place cells. A potential avenue to address this issue is measuring entorhinal grid-like activity using functional magnetic resonance imaging (fMRI), recognized as a non-invasive way of characterizing navigation-related neural activity patterns in humans. Thus far, fMRI studies of grid-like activity in humans have linked grid-like activity to performance in spatial memory and navigation tasks (Doeller et al. 2010; Kunz et al. 2015;Stangl et al. 2018 ; Bierbrauer et al. 2020). Grid-like activity was found to be reduced in young adults carrying a genetic risk factor for Alzheimer’s disease (Kunz et al. 2015) and in older adults (Stangl et al. 2018), two populations characterized by deficits in path integration abilities (Stangl et al. 2018; Bierbrauer et al. 2020). Further, human grid-like codes reflect the boundaries of virtual environments in a manner similar to rodent grid cell firing (He and Brown 2019). Taken together, these findings indicate that grid-like activity measured using fMRI is a viable marker of representations of space in the human EC.

Extending this line of prior research, in the present study we examined the effects of L-Dopa on grid-like activity and object-location memory in a young adult sample. A question of particular interest is the direction of a potential effect on task performance and grid-like representations when dopaminergic tone is pharmacologically increased. Given that healthy young adults are expected to have optimal levels of dopaminergic signaling, in light of the inverted U-function hypothesis (Vijayraghavan et al. 2007; Cools and D’Esposito 2011), we expected adverse effects of L-Dopa on spatial memory performance in young adults. Further, via dopaminergic modulation of excitability of both principal cells and interneurons in medial EC (Cilz et al. 2013; Jin et al. 2019), L-Dopa may also modify the excitation-inhibition balance that is thought to support grid cell activity, thereby likely impairing spatial learning.

Answering how grid-like activity and object-place memory are affected by a pharmacological intervention targeting dopaminergic transmission will further our understanding of the role of dopamine in spatial memory processes.

Materials and methods

Subjects

This experiment was part of a larger study investigating the effects of dopamine modulation on spatial learning and decision-making.

For the overall purposes of the study, a total of 5,927 young and older adults drawn from the Dresden city registry’s population database were contacted. Out of these, 672 were interested in the study and could complete telephone screening, with 187 subjects fulfilling screening criteria. Eligibility criteria were (1) fluent command of German, (2) no history of mental disorders in the past 12 months, (3) no lifetime history of neurological disorders or seizures, (4) no antidepressants or neuroleptics taken in the past 12 months, no anxiolytics or hypnotics taken in the past 14 days, nor any other drug affecting brain dopamine levels taken within three days prior to each study session, (5) normal or corrected-to-normal vision, (6) no pregnancy or breastfeeding, and (7) no MRI or L-Dopa contraindications.

In the present analysis, we focus on data from younger adults for the following reasons: In older adults, the reduced amplitude and higher noise levels of BOLD signals (D’Esposito et al. 1999; Huettel et al. 2001; D’Esposito et al. 2003; Gauthier et al. 2013) adds to signal loss in medial temporal lobe areas (Olman et al. 2009). In combination with the present analysis, which rests on accurate estimation of grid orientation, the use of a dedicated MRI sequence optimized for medial temporal lobe areas is warranted (Stangl et al. 2018). In younger adults, however, this type of analysis has been previously used on images acquired with a standard MRI sequence (Doeller et al. 2010; Horner et al. 2016). A total of 64 young participants completed all experimental sessions. Of these, six participants had to be excluded from analysis because of incomplete task execution, technical issues during fMRI scanning, and/or prolonged nausea (a common side effect of L-Dopa) during and after fMRI scanning. Further, we excluded three of the remaining participants for whom excessive signal dropout occurred in the EC. For these three participants, the fraction of EC voxels with a mean signal above the masking threshold (0.8) was less than 25% in at least one scanning session. Our final sample consisted of n = 55 healthy young adults (34 males, 21 females, age range: 26–35 years, M = 31.3, SD = 3.0 years).

Eligible participants took part in three experimental sessions three to six weeks apart. On the first day of testing, participants performed behavioral tests assessing general cognitive abilities. On the following two experimental sessions, participants were administered either L-Dopa (Madopar; 150 mg plus 75 mg booster) or a Placebo (P-Tabletten) and performed three computer experiments, two of which were performed during fMRI scanning and one afterwards. Participants were paid for their participation. However, the exact amount the participants received depended on the points gained during additional tasks and ranged from 118 to 196€.

Written informed consent was obtained prior to participation in the experiment. The experiment was approved by the ethics committee of Technische Universität Dresden (EK440202012) and conducted in accordance with the Helsinki declaration. The present study is a secondary analysis of a subsample of participants, i.e. focusing only on young adults for which results using other analysis methods that focused on different research questions about effects of brain aging on spatial memory were previously reported (Koch et al. 2022; Baeuchl et al. 2023).

Task

We used a computerized object-location memory task (Fig. 1) that has been described in previous studies (Schuck et al. 2015; Thurm et al. 2016; Hilliard et al. 2019; Glöckner et al. 2021). Participants navigated in an arena consisting of a grassy plane surrounded by mountains serving as distal cues which were projected to infinity. A custom-made MR-compatible joystick was used for navigation. A circular wall served as an arena boundary (diameter: 80 virtual meters [vm]). Furthermore, a traffic cone served as a salient intra-maze cue.

Fig. 1.

(A) Participant’s view of the environment. (B) Schematic diagram of the object-location memory task. During fMRI scanning, participants performed an object-location task in a circular virtual environment in which the locations of five objects had to be learned. After an initial encoding phase in which objects were visible and had to be picked up by the participant, a feedback phase followed in which participants were cued with the image of one object per trial and had to visit that object’s remembered location. In the transfer phase, the effect of two manipulations on spatial memory was tested: In the location cue shift condition, the salient intra-maze cue was displaced by a fixed direction and distance. In the boundary enlargement condition, the arena boundary was shifted outwards, enlarging its radius by 20%. Images (A) and (B) modified from Hilliard et al. (2019), CC-BY license. (C) The drop error for participants receiving Placebo treatment in session 1 decreases across feedback trials. Bars indicate bootstrapped 95% confidence intervals around the mean. (D) Significant grid-like activity (i.e. 6-fold symmetric activation) in the right EC for participants who received Placebo in session 1. Control periodicities are not significantly different from zero. (E) Higher grid magnitude is associated with reduced drop error for participants who received Placebo in session 1. Pearson correlation: R = −0.35, P = 0.038, CI(r) = [−0.61, −0.02].

The task was implemented in UnrealEngine2 (Epic Games). In each scanning session, participants had to remember the locations of five distinct objects. Different sets of objects were paired with different unique locations in the two sessions, totaling ten distinct objects. The task consisted of three phases: First, in the encoding phase, each object was only presented once, one by one, and participants were asked to indicate that they encoded each object’s location by virtually walking over it. Before the next object was shown, participants were relocated to a position near the center of the environment. Next, in the feedback phase, objects were no longer visible at their location in the arena. Participants were cued with the image of an object and had to navigate to its remembered location. Participants pressed a button when they reached the remembered location (referred to as “drop”). Feedback was provided in the following manner: If the drop location fell within a radius of 5 vm around the correct location, the trial was scored as a perfect hit. Otherwise, the object was shown at its correct location and the participant had to collect the object again. Each object was cued six times, totaling 30 feedback runs. Performance in the task was measured via the Euclidean distance between each drop location and the object’s correct location, which we refer to as “distance to location” and which reflects the error in object-location memory. All behavioral responses including the participant’s position in the environment were logged every 100 ms.

Finally, in the transfer phase, two manipulations were applied to the environment to test for an influence of intra-maze cues versus boundary information on object-location memory (cf. Schuck et al. 2015; Thurm et al. 2016; Glöckner et al. 2021): In the location cue shift condition, the intra-maze cue was shifted to a different position inside the virtual arena. This shift remained constant across all trials of the location cue shift condition. The other manipulation, termed boundary transfer condition, was included for the purposes of other studies (cf. Schuck et al. 2015; Thurm et al. 2016; Glöckner et al. 2021) but was not relevant to the present analysis because of potential confounding effects, which we detail below. In the boundary condition, the arena boundary was shifted outwards, increasing its radius by 20%. Altogether, participants performed 20 trials in the transfer phase (each object was cued twice per transfer condition). In both conditions of the transfer phase, each trial started with one of the objects being cued. Irrespective of the manipulations, participants had to navigate to the location where they remembered the object and then indicate that location by pressing a key, but without receiving any feedback. Given that the manipulation in the boundary transfer condition creates a potential confound for the analysis of grid-like activity, as the expansion of boundaries has been shown to impact grid-cell firing in rodents (Barry et al. 2012), we focused our analysis solely on the feedback phase together with those trials of the transfer phase that did not involve an increase in diameter of space (i.e. the location-cue shift condition). In addition, we excluded the five encoding trials of each session from analysis for the following reason: In the encoding phase, navigation was not continuous across encoding trials, but participants were relocated to a position near the center of the environment at the beginning of each trial. Excluding this very brief phase for initial encoding (duration of 1–2 minutes on average) is also consistent with previous studies on grid-like activity (e.g. Kunz et al. 2015). To summarize, the analysis was focused on trials with continuous between-trial navigation within a spatial environment of constant size (diameter).

Drug administration

The pharmacological intervention followed a randomized double-blind crossover design, with an additional control group receiving Placebo treatment in both sessions. Therefore, participants were randomly assigned either to the group receiving L-Dopa in the first session and Placebo in the second session (“L-Dopa-starters”), or to the group receiving Placebo in the first session and L-Dopa in the second session (“Placebo-starters”), or to the “Placebo-Placebo” group receiving Placebo in both sessions. The corresponding random probabilities for group assignment were 2/5 each for the intervention groups, and 1/5 for the Placebo-Placebo group. For details about the resulting group sizes, see Table 1. Following a previous study (Kroemer et al. 2019), and to maximize dopamine availability throughout the task, the medication was administered in two doses: A first dose of 150 mg/37.5 mg L-Dopa/benserazide (Madopar; Levodopa and Berazidhydrochlorid; Roche) was administered 15 minutes before participants underwent fMRI scanning while performing another task which is not part of this manuscript. 110 minutes after the first dose, and 25 minutes before the onset of the object-location memory task, participants received a booster dose of 75 mg/18.75 mg L-Dopa/benserazide.

Table 1.

Sample characteristics by drug administration group.

Group	N	Age (years)	Gender
L-Dopa-starters	20	31.0 ± 3.0	8 females
Placebo-starters	22	31.3 ± 2.9	8 females
Placebo-Placebo	13	31.9 ± 3.3	5 females

fMRI acquisition

A 3T Siemens Magnetom Trio scanner was used to acquire imaging data. At the start of the first MRI session, T1-weighted structural scans were collected (voxel size: 0.8 × 0.8 × 0.8 mm³, repetition time (TR) = 2,400 ms, echo time (TE) = 2.19 ms, inversion time (TI) = 1,000 ms, acquisition matrix = 320 × 320 × 240, field of view (FOV) = 272 mm, flip angle = 8^°, bandwidth = 210 Hz/Px). Further, at the beginning of the second MRI session, a T2-weighted structural image was acquired (0.8 × 0.8 × 0.8 mm³, TR = 3,200 ms, TE = 565 ms, acquisition matrix = 320 × 320 × 240, FOV = 272 mm, bandwidth = 744 Hz/Px). During the task, T2*-weighted BOLD images were acquired using an echo-planar imaging (EPI) pulse sequence with the following parameters: TR =2.36 seconds, TE = 25 ms, slice thickness = 2.5 mm, in-plane resolution: 3 mm × 3 mm, image matrix: 64 × 64, number of slices = 48, FOV = 192 mm, flip angle = 80°, GRAPPA parallel imaging, acceleration factor = 2, ascending acquisition. Further, a field map has been acquired using the following parameters: TR = 532 ms, TE 1: 5.32 ms, TE 2: 7.78 ms, FOV: 192 mm, acquisition matrix: 64 × 64, number of slices = 48, voxel dimensions: 3.5 mm.

fMRI preprocessing

All MRI preprocessing was performed using SPM12 (Statistical Parametric Mapping; www.  fil.ion.ucl.ac.uk/spm). The realignment algorithm offered by SPM12 was used to estimate displacement of functional scan images stemming from head movements. The resulting six motion parameters describing translational and rotational head movements were entered as regressors of no interest in all subsequent GLM analyses. Following spatial realignment, images were slice time corrected and functional images were co-registered to the T1-weighted structural image. Images were smoothed using a kernel with 4 mm full width at half-maximum. Further analyses were performed in each participant’s native space.

ROI masks

Given our research question and hypotheses, we took the region of interest (ROI) approach and focused our analyses on subregions of the hippocampal formation. To label voxels in EC, we used the automatic segmentation of hippocampal subfields (ASHS) software package in conjunction with the PMC-T1 atlas (Xie et al. 2019). This multi-atlas approach labels subfields of the medial temporal lobe based on T1-weighted images (Supplemental Fig. S2). We carefully double-checked correct registration to the functional images by visual inspection. In particular, we ensured that the EC mask images generated by ASHS contained the same voxel-to-world transformation matrix as defined in the T1-weighted scan image before down sampling these masks to the resolution of the functional images. In line with previous studies which reported the strongest effect of grid-like activity in the right EC (Doeller et al. 2010; Kunz et al. 2015; He and Brown 2019), we chose the right EC as the main region of interest, and the left EC and the right posterior hippocampus (pHC) as control regions. Masks for the pHC were similarly obtained from the ASHS segmentation.

fMRI analysis of grid-like activity

We used the GridCAT Toolbox (Stangl et al. 2017) based on SPM12 and Matlab to analyze grid-like activity, which implements the analysis developed in Doeller et al. (2010). The analysis logic is as follows:

First, movement epochs were extracted from the log files. Movement epochs lasting < 1 second were excluded from the analysis. In the desktop VR environment, participants navigated with a joystick. This allowed participants to walk either straight lines or smooth curves, at an angular speed of approx. 15°/second. Any curved paths were linearized as follows: When the accumulated turning angle exceeded 15°, the path was split into separate movement events such that the turning angle within each event did not exceed 15°. Each of these partial events was treated as linear movement into the average movement direction, where the average is taken across all directions logged within the partial event. This approach was chosen to balance between minimizing directional error introduced by linearization and avoiding event durations below 1 second.

The total experiment was subdivided into time bins of 330 TRs = 779 seconds = 13 minutes. Data from the first half of each time bin was used to estimate the putative grid orientation, and the second half of each time bin was used to compute the contrast of aligned versus misaligned translation events, relative to the estimated grid orientation. Time bins were created in an overlapping fashion, with a 50% overlap between successive time bins (Kunz et al. 2015). To ensure a fully orthogonal design despite partially overlapping time bins, odd and even time bins were analyzed separately. The total result was computed as a weighted average of BOLD contrasts for odd and even time bins, weighted by their respective counts.

To avoid potential confounding effects, our analysis focused on both the feedback phase and the location cue shift condition of the transfer phase but omitted both the encoding phase and the boundary transfer phase (for details, see subsection “Task” above). Regressors of no interest were added for short movements (< 1 second), intertrial intervals, and cue periods.

The analysis assumes that the magnitude of grid-like activity, Y, is a sinusoidal function of 6-fold periodic movement direction θ: Y (θ) ∼ cos(6[θ(t) − φ]). In a first step, the analysis seeks to estimate the putative grid orientation, φ. In the first GLM (GLM1), two parametric regressors for cos(6θ) and sin(6θ) are applied, where θ = θ(t) denotes the participant’s movement direction at time t. This leads to two beta estimates: β_cos, β_sin. The grid orientation per voxel is then estimated as (Doeller et al. 2010).

Next, in a second GLM (GLM2), data from the second half of each time bin is used to compute a contrast of movement aligned to the putative mean grid orientation within a given ROI (i.e. events with a direction differing less than 15° from the mean grid orientation) versus movement misaligned with the mean grid orientation. Resulting contrast values were then averaged across time bins as described above.

To compute contrasts for control models (5- and 7-fold symmetry), the same approach is followed in analogy.

Behavioral analysis

For each trial, we calculated the distance between the correct object location and the location at which the object was dropped by the subject (distance to location). Larger distances indicate larger errors in object-location memory.

Statistical analyses

Statistical analyses were performed using R 4.2.1 (www.r-project.org). We used two-sided one-sample t-tests to assess whether contrasts relating to 5-, 6-, and 7-fold symmetries were different from zero.

To assess the effect of the pharmacological intervention, we used linear mixed-effects models, implemented in the lme4 package, version 1.1-30, and the lmerTest package, v3.1-3. Note that participants from the Placebo-Placebo group could not be added to the linear mixed model, as this would have required a fully crossed design with an additional group receiving L-Dopa in both sessions. Given the limited sample size, we refrain from analyzing session effects in this group. However, based on previous findings, we do not expect session differences in terms of grid-like activity (Constantinescu et al. 2016). Additionally, in the case that fitting the model resulted in a singularity, i.e. the random-effect variance estimate was nearly zero, we first ensured that the random effects were reduced in complexity as far as possible. We then used a fully Bayesian method to estimate linear mixed-effects models with regularization via informative priors, implemented in the brms package, v2.17.0 (Bürkner 2017). Specifically, for the random effects estimates, we used the default prior given by a Student-t-distribution with df = 3, μ = 0, and σ = 2.5. Uniform priors were used for the fixed effects estimates. For sampling, we used 12 chains and 3,000 iterations. We used the bayestestR package, v0.13.0 (Makowski et al. 2019) to calculate p-values for the existence of an effect based on the resulting posterior distributions.

Further, we used the emmeans package (v1.8.0) to compute post-hoc contrasts based on estimated marginal means.

Results

Grid-like activity at baseline

We used established methods to measure grid-cell-like activity in human participants during fMRI, which is reminiscent of the hexagonal symmetry of entorhinal grid cells measured in rodent electrophysiology (Doeller et al. 2010; Kunz et al. 2015; Stangl et al. 2017). We first analyzed behavioral performance and grid-like activity for all participants receiving Placebo in the first fMRI session (i.e. pooling participants from the Placebo-starter group with the Placebo-Placebo control group) to establish the grid-like activity in our sample at baseline (i.e. without prior experience in the spatial learning task and not under L-Dopa influence). Spatial learning in the object-location memory task improved substantially within the first session, as can be seen in a rapid reduction of the distance to location across trials (Fig. 1C). Of note, we observed significant grid-like activity in the right EC of these participants (Fig. 1D), i.e. the BOLD contrast for trajectories aligned to the putative grid orientation (assuming 6-fold symmetry) versus trajectories misaligned to the same orientation was significantly greater than zero (t₃₄ = 2.10, p = 0.043, Cohen’s d = 0.36, 95% confidence interval for d: [−0.34, 1.05].) When assessing control periodicities (5- and 7-fold), the aligned-misaligned contrast was not significantly different from zero (5-fold: t₃₄ = −0.33, p = 0.74, 7-fold: t₃₄ = 1.00, p = 0.32. Fig. 1D). Further, when repeating the analysis in control ROIs, we observed no significant grid-like activity in the left EC or in the right posterior hippocampus (Supplemental Fig. S1). Moreover, we found an association between the magnitude of grid-like activity and performance: Higher grid-like activity was associated with improved task performance (i.e. lower distance to location; Fig. 1E). However, as correlations with small samples sizes have large confidence intervals, this result should be interpreted with caution (Schönbrodt and Perugini 2013).

Effects of L-Dopa

We next evaluated the effect of the pharmacological intervention on grid-like activity and behavioral performance in the two groups of participants who received L-Dopa (Fig. 2A). We used a linear mixed-effects model (Model 1.1: Grid magnitude ∼ intervention × intervention order, with intervention as the within-subject fixed effect, intervention order (L-Dopa starters vs. Placebo-starters) as the between-subjects fixed effect, and random intercepts for subjects). As fitting the model resulted in nearly zero estimated variance for the random intercepts, we additionally fitted a fully Bayesian linear mixed-effects model involving the same structure of fixed and random effects, which uses a regularizing prior for the random effects to avoid zero variance (see Methods). As both models were highly similar in their results, we here report the results of the standard linear mixed-effect model for simplicity. Results for the fully Bayesian model are provided in Supplemental Table S2. In the right EC in which 6-fold grid-like activity was observed at baseline, we observed a significant main effect of intervention (F_1,80 = 4.92, p = 0.029, other p-values > 0.31).

Fig. 2.

(A) For L-Dopa-starters, the magnitude of grid-like activity in the right EC is significantly higher under Placebo than under L-Dopa. For Placebo-starters, grid-like activity does not differ significantly between L-Dopa and Placebo. Error bars show bootstrapped 95% confidence intervals. (B) Distance to location across interventions and sessions. For L-Dopa-starters, the distance to location is significantly lower under Placebo than under L-Dopa. For Placebo-starters, the distance to location is not significantly different between L-Dopa and Placebo. (C) Magnitude of contrasts for control models (5- and 7-fold symmetries) for L-Dopa-starters in both intervention sessions. (D) Same for L-Dopa-starters.

Next, to assess the intervention effect within each group, we calculated post-hoc contrasts based on estimated marginal means. These contrasts revealed that for L-Dopa starters, grid-like activity was significantly lower under L-Dopa than under Placebo (contrast estimate = −0.39, t₄₀ = −2.25, p = 0.03). For Placebo-starters, however, we observed no significant effect of the L-Dopa intervention (p = 0.40). Contrasts for control symmetries were not significantly different from zero in either treatment, both in the L-Dopa starter group and in Placebo-starters (Fig. 2C, D). Taken together, this suggests that L-Dopa attenuates grid-like activity, most prominently in novel environments. Similarly, we used a linear mixed-effects model to assess the relation between behavioral performance and the pharmacological intervention (Fig. 2B; Model 2.1: Drop error ∼ intervention × intervention order, with intervention and intervention order as fixed effects, and random intercepts for subjects). We observed a significant interaction of intervention and intervention order (F_1,40 = 16.53, p = 2.2 × 10⁻⁴). Post-hoc contrasts revealed a significantly higher drop error for LDopa starters under L-Dopa than under Placebo (contrast estimate = 6.87, t₄₀ = 3.78, p = 0.003, Tukey-adjusted for pairwise comparisons, all other p-values > 0.05).

Further, as a control analysis, we tested whether participants’ navigation behavior was altered as an effect of L-Dopa, in a way that might constitute a confound for the analysis of grid-like activity. In participants’ trajectories, we observed no effect of the intervention on the number of curves they made or on the radii of their curves (Supplemental Figs. S3 and S4). Moreover, we observed no overall systematic differences in the directional distributions between 6-fold versus 5- or 7-fold symmetries (Supplemental Fig. S5). Similarly, we found no overall effect of L-Dopa on directional distributions that would confound our analyses (Supplemental Fig. S6). This shows that L-Dopa impaired performance in the object-location memory task specifically in participants performing the task in a novel environment.

Discussion

We tested our hypothesis that L-Dopa modulates both spatial representations and spatial memory in healthy young adults. Our main finding is that L-Dopa impairs both grid-like activity in the EC and object-location memory performance in young adults. This impairment is most pronounced when L-Dopa is administered in a novel environment. These results support the hypothesis that dopaminergic signaling modulates the formation and within-session maintenance of both grid-cell-like representations and object-place associations.

We first discuss our findings relative to theories of the role of dopamine in memory. The role of dopamine in hippocampus-dependent memory has been addressed both from a perspective of ”online” hippocampal place representations, supporting ongoing behavior, and of ”offline” post-encoding consolidation processes, influencing memory persistence.

First, a prominent line of research has linked dopaminergic signaling to attention to visuospatial cues as a means to stabilize both hippocampal place representations and spatial memory (Kempadoo et al. 2016, Kentros et al. 2004; for a review, see Muzzio et al. 2009). By this account, our observation of impaired performance in object-location memory could be interpreted as reflecting deficits in attending to distal cues and boundaries that provide location information. However, such an attention deficit most likely does not account for the reduction in grid-like activity we observed under L-Dopa, as grid cell firing in rodents is more strongly influenced by self-motion than visual cues (Chen et al. 2019). We discuss alternative explanations below.

Second, the role of dopamine in hippocampus-dependent memory has been widely characterized in relation to “early” and “late” synaptic long-term potentiation (LTP): While D1-like dopamine receptors have been shown to strongly modulate late LTP, related to memory retention across a 1-day delay (O’Carroll et al. 2006; Lisman et al. 2011; Chowdhury et al. 2012), the dopaminergic modulation of early LTP and memory retention has been regarded as moderate (Lisman et al. 2011). While our study did not assess any potential effects of L-Dopa on memory retention across 24 hours, there is also evidence for acute effects of dopaminergic modulation on memory encoding, compatible with our behavioral findings. Pezze and Bast (2012) reported that short-term spatial memory at a 30 minute delay was impaired by a D1 receptor agonist, using a delayed-matching-to-place watermaze task. In a similar task, Nai et al. (2010) observed impairments in learning when a peptide was administered which interfered with interactions between D1-like receptors and NMDA receptors. Taken together, these results agree with our findings that dopaminergic modulation impacts memory processes on task-relevant timescales.

Despite the established role of dopamine in memory formation, we are unaware of any evidence whether dopamine is similarly involved both in the initial formation of a representation of a new environment (e.g. Monaco et al. 2014) and in the formation of object-place associations (for review, see Connor and Knierim 2017). The fact that L-Dopa impaired memory performance specifically in a novel environment suggests that “configural” learning, i.e. the formation of an allocentric representation of the environment based on environmental geometry and distal cues (Rudy and Sutherland 1995), is more strongly impaired than the formation of object-place associations in an already familiar environment. However, we cannot exclude the possibility that L-Dopa impairs performance similarly when the task is already familiar, an effect which might be counteracted by learning-related improvements in task performance. Future studies should aim to address this issue, e.g. by including another control group receiving L-Dopa in both sessions. Clarifying this issue would also be important for computational theories of spatial memory, in which dopamine often acts as a prediction error signal for the learning of reward locations (Schultz et al. 1993; Foster et al. 2000; Erdem and Hasselmo 2012; Gönner et al. 2017), but in which experience-dependent effects of dopamine do not yet play a role.

A further prominent theory implicates dopaminergic signaling in the generation of novelty responses, which support memory persistence (Lisman and Grace 2005). In this regard, Bunzeck et al. (2013) have reported that L-Dopa alters novelty responses during image viewing in the hippocampus and parahippocampal cortex, which are both strongly implicated in spatial memory formation. In our object-location memory task, reduced neural responses to sensory stimuli of a novel spatial environment may potentially further contribute to impaired memory encoding.

At the neurophysiological level, there is ample evidence that dopamine exerts effects on entorhinal cortical neural activity via multiple mechanisms, including a modulation of intrinsic neuronal excitability (Jin et al. 2019), synaptic transmission (Cilz et al. 2013), persistent firing patterns (Batallán-Burrowes and Chapman 2018), and network activity (Mayne et al. 2013); for a review, see Sosa and Giocomo (2021). All of these factors are likely capable of altering the fine balance between excitation and inhibition in the entorhinal grid cell network, suggesting that increasing dopamine beyond optimal levels could potentially lead to a loss in grid cell spatial specificity (cf. Christensen et al. 2021).

A complementary theory to interpret the deficits we observed under L-Dopa is the inverted U-function hypothesis of dopamine (Vijayraghavan et al. 2007; Cools and D’Esposito 2011): Given that healthy young adults are expected to have optimal levels of dopaminergic signaling, increasing dopamine availability by L-Dopa will induce a suboptimal level of this neuromodulator. Potential consequences of excessive dopamine are primarily thought to affect the balance between stability and flexibility of working memory computations: Increased stability would result in reduced distractibility, at the expense of a reduced capability to flexibly update working memory content relative to current task demands, and vice versa (Cools and D’Esposito 2011). For the domain of spatial navigation, Lester et al. (2017) have summarized that impairments in working memory, stemming from altered processing in prefrontal areas, can disrupt representations of self-location and navigational goals. While this opens up the possibility that the impairments we observed might be partly owing to altered spatial working memory computations, our study did not directly assess whether L-Dopa modulated working memory function in general.

We now turn to review our findings relative to previous fMRI studies of grid-like activity. While there is agreement in existing studies about a predictive relationship between path integration performance and grid-like activity (Stangl et al. 2018; Bierbrauer et al. 2020), results about an association to object-location memory are somewhat more variable. Kunz et al. (2015) reported that higher grid-like activity was predictive of improved spatial memory performance in a population of young adults, including APOE-ε4 carriers at genetic risk of Alzheimer’s disease. In contrast, Stangl et al. (2018) found no association between grid-like activity and object-location memory performance. It remains possible, however, that these differences may partly be owing to task differences (i.e. the number of different object-location pairs to be remembered, and the amount of training prior to fMRI scanning). As both our task design and our results about an association between grid-like activity and memory performance in the absence of L-Dopa are broadly similar to the study by Kunz et al. (2015), this suggests that grid-like activity may predict memory performance specifically during the learning phase of a task, and/or when a larger number of object-location pairs must be remembered, a condition which requires greater spatial pattern separation.

For a sample of young adults at genetic risk of Alzheimer’s disease, Bierbrauer et al. (2020) have reported that the magnitude of grid-like activity was associated with path integration performance, but only in the absence of spatial cues such as landmarks or boundaries which would permit compensatory strategies. As the task used in our study was not dependent on pure path integration, the findings of Bierbrauer et al. (2020) suggest that an impairment of path integration reflected in reduced grid-like activity is not the primary driver of the impairments in object-location memory we observed as an effect of L-Dopa. Taken together, it seems plausible that L-Dopa alters either the perceptual or the mnemonic processing of spatial cues such as boundaries, distal cues, and landmarks.

Regarding determinants of grid cell firing, our findings add to a growing body of literature which emphasizes the variability and malleability of grid cell firing and grid-like activity. While entorhinal grid cells were initially interpreted as providing a universal spatial metric supporting navigation-related computations (Hafting et al. 2005), recent research has identified several variable factors which influence grid cell firing patterns. This includes changes in environmental layout (Derdikman et al., 2009, Krupic et al. 2015), task demands (Boccara et al. 2019; Butler et al. 2019), developmental phase (Langston et al. 2010; Wills et al. 2010; Stangl et al. 2018), and physiological integrity of the EC (Gil et al. 2018; Christensen et al. 2021). Our findings add to these influencing factors by highlighting the sensitivity of grid-like activity to pharmacological interventions of the dopaminergic system. In summary, our results support the notion that grid cells and grid-like activity patterns are a sensitive marker for changes in environmental, behavioral, developmental, and physiological conditions.

We discuss several limitations of our study. Our methodology does not permit to characterize the neurophysiological mechanisms underlying the suppression of grid-like activity by L-Dopa. Future work is necessary to discriminate between potential effects on path integration, place coding, object-place associations, or representations of navigational goals. Further, we did not test memory retention at a 24 hour delay. Finally, our chosen age range of 25–35 years overlaps with a period of extensive rewiring of dopaminergic inputs to the prefrontal cortex, which occurs up to an age of about 30 years (Petanjek et al. 2011). This might induce a source of variability in our data.

Conclusion

Our results highlight the important role of dopaminergic modulation of spatial representations and memory processes. Candidate mechanisms for the effects we observed range from changes in cellular excitability and modulation of synaptic plasticity to system-level attentional effects. Future work should attempt to disentangle these potential mechanisms.

CRediT statement

Lorenz Gönner (Conceptualization, Investigation, Methodology, Software, Writing—original draft), Christian Baeuchl (Data curation, Investigation, Methodology, Writing—review and editing), Franka Glöckner (Data curation, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Writing—review and editing), Philipp Riedel (Writing—review and editing), Michael Smolka (Funding acquisition, Resources, Supervision, Writing—review and editing), Shu-Chen Li (Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing—review and editing).

Funding

This work has been funded by the Deutsche Forschungsgemeinschaft (DFG project number 178833530 [SFB 940], grant to S.C.L., F.G., and M.S.).

Conflict of interest statement: None declared.

Data availability

Data underlying this paper will be made available at https://osf.io/suj5x/.