Dopamine Related Genes Differentially Affect Declarative Long-Term Memory in Healthy Humans

In humans, monetary reward can promote behavioral performance including response times, accuracy, and subsequent recognition memory. Recent studies have shown that the dopaminergic system plays an essential role here, but the link to interindividual differences remains unclear. To further investigate this issue, we focused on previously described polymorphisms of genes affecting dopaminergic neurotransmission: DAT1 40 base pair (bp), DAT1 30 bp, DRD4 48 bp, and cannabinoid receptor type 1 (CNR1). Specifically, 669 healthy humans participated in a delayed recognition memory paradigm on two consecutive days. On the first day, male vs. female faces served as cues predicting an immediate monetary reward upon correct button presses. Subsequently, participants performed a remember/know recognition memory task on the same day and 1 day later. As predicted, reward increased accuracy and accelerated response times, which were modulated by DAT 30 bp. However, reward did not promote subsequent recognition memory performance and there was no interaction with any genotype tested here. Importantly, there were differential effects of genotype on declarative long-term memory independent of reward: (a) DAT1 40 bp was linked to the quality of memory with a more pronounced difference between recollection and familiarity in the heterozygous and homozygous 10-R as compared to homozygous 9-R; (b) DAT1 30 bp was linked to memory decay, which was most pronounced in homozygous 4-R; (c) DRD4 48 bp was linked to overall recognition memory with higher performance in the short allele group; and (d) CNR1 was linked to overall memory with reduced performance in the homozygous short group. These findings give new insights into how polymorphisms, which are related to dopaminergic neuromodulation, differentially affect long-term recognition memory performance.


INTRODUCTION
The dopaminergic system serves several, yet interconnected, functions. On the one hand, a functional loop between the medial temporal lobe and midbrain dopamine (DA) neurons drives the encoding of novel information into long-term memory (Lisman and Grace, 2005;Bunzeck and Düzel, 2006;Düzel et al., 2009;Bunzeck et al., 2014). On the other hand, DA neurons are known to be critical for reward processing (Schultz et al., 1997;Fiorillo et al., 2003;Tobler et al., 2005). Therefore, it is not surprising that reward motivation not only accelerates response times (Knutson et al., 2001;Bayer et al., 2013;Steiger and Bunzeck, 2017) and enhances physical effort (Pessiglione et al., 2007), but also improves subsequent long-term memory (Wittmann et al., 2005;Adcock et al., 2006). Evidence for a link between reward processing and invigoration of behavior through dopaminergic neuromodulation comes from computational models and empirical studies in animals and humans (Niv et al., 2007;Guitart-Masip et al., 2011;Dayan, 2012;Beierholm et al., 2013;Steiger and Bunzeck, 2017). Importantly, specific genes that affect dopaminergic neuromodulation have also been identified; yet the link between genetic polymorphisms and rewarddependent long-term memory remains less clear.
In a typical reward-dependent long-term memory task, a cue predicts monetary rewards for a given behavior, for instance, correct category judgments. The to-be-learned information, such as an image, often follows the cue, or it is a cue itself (for instance the categories living vs. non-living indicate reward vs. no reward). Subsequently, the effect of reward can be tested with free recall (in the case of words), or with scores of recollection and familiarity based recognition memory. Dual-process models assume that recognition can be associated with specific details or associations of the encoding episode (i.e., recollection), or in the absence of such recollective experience (i.e., familiarity). Support for dual-process models (Yonelinas et al., 1996(Yonelinas et al., , 2010 comes from functional imaging studies suggesting that different regions of the medial temporal lobe are involved in recognition memory depending on task demands and type of information (Diana et al., 2007). In particular, while the hippocampus and posterior parahippocampal gyrus are closely associated with recollection, the anterior parahippocampal gyrus is more associated with familiarity (Diana et al., 2007).
While several studies reported a promoting effect of reward on behavior, some could not replicate it (Callan and Schweighofer, 2008;Ariely et al., 2009;Sharifian et al., 2017;Steiger and Bunzeck, 2017), and others have even reported detrimental effects of reward on behavior (Mobbs et al., 2009;Chib et al., 2012;Kuhbandner et al., 2016). At the psychological level, this has been related to over motivation or anxiety in the context of reward (Callan and Schweighofer, 2008;Mobbs et al., 2009). However, interindividual differences in cognitive performance could also be due to genetic predisposition affecting dopaminergic neuromodulation. Indeed, genetic polymorphisms are known to affect the availability of neurotransmitters in the central nervous system, its retention time in the synaptic cleft through transporter availability, and receptor density (Breedlove et al., 2010). Therefore, it is reasonable to further investigate the relationship between genetic polymorphisms and performance in a reward-based declarative memory task.
Four a priori selected polymorphisms were investigated here: DAT1 40 base pair (bp), DAT1 30 bp, DRD4 48 bp, and cannabinoid receptor type 1 (CNR1). The human DA transporter (DAT1) exhibits several functional mutations, including a 40 bp variable number of tandem repeat (VNTR) polymorphism with the most common 9-repeat (9-R) and 10-repeat (10-R) alleles (Mitchell et al., 2000;Simsek et al., 2006). The 9-R allele of the DAT1 polymorphism has been associated with lower availability of DAT1 (Heinz et al., 2000;Cheon et al., 2005) possibly leading to increased striatal DA in the synaptic cleft (Schuck et al., 2013). Those findings are supported by in vitro (VanNess et al., 2005) and in vivo research . Only a few studies found contrary results, showing that 10-R carriers had lower availability of DAT1 (Jacobsen et al., 2000;van Dyck et al., 2005). At the behavioral level, the 10-R allele has been associated with risk-taking, suggesting a possible link to reward processing (Mata et al., 2012), and worse memory performance in some studies (Simon et al., 2011;Li et al., 2013). In a reward-based memory task, 10-R homozygotes remembered rewarded pictures better compared to neutral ones, which could be linked to differences in hemodynamic activity within the striatum and hippocampus (Wittmann et al., 2013); since this study only included 24 participants, a replication of a link between DAT1 and reward-based learning appears reasonable. Finally, Raczka et al. (2011) could show that 9R carriers are quicker in learning fear extinction, possibly related to a higher dopaminergic prediction error signal when an expected aversive unconditioned stimulus does not occur in extinction (see also Kalisch et al., 2019). Taken together, research on the DAT1 40 bp polymorphism suggests a critical role not only for learning and memory but also for reward processing.
The present study investigated the effects of reward in the context of a two-day recognition memory paradigm. We expected reward to accelerate RT and increase accuracy during encoding, and to promote subsequent long-term recognition memory. Moreover, we investigated whether genotypes, that affect the dopaminergic system (DAT1 40 bp, the DAT1 30 bp, the DRD4 48 bp, and the CNR1), interact with those measures.

Participants
In total, 690 healthy humans participated in this study, which was part of a large behavioral test-battery, over 2 days. However, due to incomplete data in the behavioral tasks (see below) or lack of genotyping, 21 participants had to be excluded. Thus, the final sample included 669 subjects [mean age (±SD) = 24.37 (±3.39) years, 478 female]. The testing and data acquisition included several behavioral paradigms, questionnaires, and neuropsychological tests. Some of the results have already been reported elsewhere (Haaker et al., , 2017Lonsdorf et al., 2015;Schümann et al., 2018). On both days, participants arrived in the laboratory at 9 AM; on day 1, they provided a urine sample for drug testing as well as a saliva sample for genotyping and DNA extraction.

Experimental Paradigm
The experiment took place on two consecutive days. On the first day, subjects participated in an encoding task and a subsequent recognition memory test, which was repeated on day two (with different stimuli, see below). During encoding, they were presented with 80 male and female faces (in random order) for 1 s, followed by a fixation cross for 2 s (Figure 1). Subjects were instructed to indicate via button presses as quickly and correctly as possible whether the face was male or female. One of both categories was associated with a reward, which was directly presented upon correct responses (1e). Incorrect responses and correct responses to the not-rewarded category led to no reward feedback (0e). The rewarded category was counterbalanced between participants. In a short training session, reward probability was 100% but it was 80% in the actual experiment. Participants were not explicitly instructed about which sex (male vs. female) was linked to a reward but quickly learned it throughout the training session. The encoding phase (including training) lasted approximately 20 min.
Shortly after encoding, participants took part in a first retrieval test. Here, 40 studied (i.e., old) faces from the encoding task were intermixed with 40 unstudied (i.e., new) faces. Each image was presented for 1.5 s and participants had up to 4 s to decide whether the face was ''old'' or ''new.'' Following a new response, they indicated whether they were ''sure'' or ''guessed'' in their decision. If they had recognized the face from before (i.e., old response), they had 4 s to decide if they ''remembered,'' ''knew,'' or just ''guessed.'' They chose ''remember'' when they recognized a picture and could recollect specific thoughts or associations linked to the study episode (recollection). They chose ''know'' when they recognized the picture but were not able to recall specific details or associations related to the study episode (familiarity). ''Guessed'' had to be pressed when they did not know whether a picture was old or new. Participants were carefully instructed, orally, and in writing, about the meaning of each response option.
On the second day, participants took part in a second recognition memory task, which was identical to the recognition memory test on day 1, except that it included the second, unseen, half of the studied items and 40 new unstudied distractors.
The frontal-view photographs of unfamiliar adult faces were used before  and taken from ''The Psychological Image Collection at Stirling (PICS, http://pics.psych.stir.ac.uk/'').
Genotyping DNA samples were extracted by the Department of Human Genetics of the University Medical Center Hamburg-Eppendorf. Subjects were genotyped for the DAT1 40 bp, DAT1 30 bp, DRD4 48 bp, and CNR1 polymorphism. Genotyping was performed by Bioglobe (Hamburg, Germany). To detect the SNPs, the iPLEX method and the MALDI-TOF mass spectrometry were used on the MassARRAY system. A standard protocol, which was recommended by the system supplier, was used for most FIGURE 1 | Experimental design. (A) During encoding, participants responded to male vs. female faces via button presses. One category was, upon correct categorization, rewarded; in this case female faces. (B) Retrieval, on day 1 and day 2, followed a remember/know recollection paradigm, see text. The frontal-view photographs of unfamiliar adult faces were used before  and taken from "The Psychological Image Collection at Stirling (PICS, http://pics.psych.stir.ac.uk/)". iPLEX reactions. It produces allele-specific analytes in a primer extension reaction applying a primer directly adjacent to the SNP site. The length of the extended primers is identical for any allele of interest so that detection and allelic discrimination rely on the present mass differences of each nucleotide base. For data acquisition, a MassARRAY Analyzer Compact was used before an automated data analysis with TYPER RT software version 3.4.
A polymerase chain reaction (PCR) amplification was used to analyze the different lengths of the VNTR. A fluorescent dye was attached to one of the primers and purification was performed. The obtained PCR product was separated by capillary electrophoresis on an ABI 3500 XL sequencing instrument. With the fluorescence signal, the length of the product was obtained and used to extrapolate the alleles from the raw data.

Statistical Analysis
For the statistical analysis, IBM SPSS Statistics for Mac (Version 22.0) was used. Results were considered to be significant at p < 0.05 (in combination with Bonferroni correction for multiple comparisons when applicable, see ''Results'' section). As a measure of effect size, partial η 2 is reported. Outliers were not excluded due to the large sample size. If the normal distribution, tested with the Shapiro-Wilk-test, was violated, nonparametric tests were used. The homogeneity of variances was examined with the Levene-test. When the sphericity assumption, measured with the Mauchly sphericity test, was violated, Greenhouse-Geisser correction was used.
For the encoding phase, mean RT was calculated for each participant for the rewarded and not-rewarded trials. The number of correct responses in the rewarded and non-rewarded trials was used as a measure of accuracy.
For the retrieval phase, corrected hit rates (CHR) were calculated for rewarded and not-rewarded stimuli separately, taking into account also the time of retrieval (day 1 vs. day 2) and type of memory (remember vs. know). This resulted in eight dependent variables per participant. CHR was calculated separately for remember responses (CHR-rem) and know responses (CHR-know) based on hit rates (percentage of correct old classifications) minus false alarm rate (i.e., false classification of old faces as new). Items classified as ''guessed'' were not included in any analysis. Genotypes were classified as described earlier and grouped based on previous studies.
The first set of analyses focused on RT, the number of correct responses, and memory performance independent of genotypes. Since RT and the number of correct answers for rewarded and not-rewarded trails were not normally distributed, the Wilcoxon-Test was used as a nonparametric test. For further analyses, a condition (rewarded/not-rewarded) × genotype ANOVA was performed to investigate whether genotype groups differed in RT and accuracy depending on the reward. In a final set of ANOVAs, memory performance was used as dependent variable. The factors included: genotype (betweensubject variable), condition (rewarded/not-rewarded), time (day 1/day 2), and quality of memory (CHR-rem/CHR-know). To further investigate significant effects, Bonferroni corrected post hoc t-tests were performed.
Genotype groups were also compared regarding sociodemographic variables. For metric variables, an ANOVA was used and for alternative and nominal data the χ 2 -test.

Other Polymorphisms
Three other polymorphisms were analyzed post hoc in an exploratory fashion. However, they were not planned to be part of this manuscript and will, therefore, not be reported in detail here: dopamine receptor D4 polymorphism 120 bp, noradrenergic receptor polymorphism α2B, and serotonin transporter-linked polymorphic region: 5-HTTLPR.

Genotyping and Group Descriptions
Genotyping was successful in most participants, group sizes range from 646 to 662. Distributions for the different polymorphism are shown in Table 1. Groups did not differ in age (p's > 0.058), gender (p's > 0.061), BMI (p's > 0.582), smoking (p's > 0.210), alcohol consumption regarding glasses/week (p's > 0.083; DRD4 48 bp: p = 0.038) and years of consumption (p's > 0.085) and cannabis consumption (p's > 0.137), see Table 1. Note that the p-values reported here and in Table 1 are uncorrected p-values.

Reaction Time and Accuracy
Reaction time and accuracy analysis included n = 669 subjects. Both variables were not normally distributed (p < 0.001); therefore, the non-parametric Wilcoxon-test was used. As expected, participants responded significantly faster in rewarded (778.42 ms, SD = 249.89) as compared to not-rewarded trials [804.61 ms, SD = 259.45, p < 0.001, r (Wilcoxon test effect size) = 0.312], see Figure 2A; and they were more accurate in the rewarded 37.19 (2.69) as compared to the not-rewarded 36.30 (3.19) trials (p < 0.001, r = 0.227, Figure 2B).
Bonferroni corrected post hoc analyses (pairwise comparisons) revealed a significant difference between recollection and familiarity scores (CHR) on both days (p's < 0.006), and that memory performance (both categories) differed significantly between days (p's < 0.003), see Figure 3. Specifically, memory was lower on day 2 compared with day 1, and CHR-rem were higher as compared to CHR-know. The interaction between time × quality of memory was driven by a larger difference between CHR-rem vs. CHR-know on day 1 as compared to day 2.

DA Transporter 40 bp Polymorphism
In the following analysis, n = 656 participants were included. Subjects were subdivided into three groups: homozygous 9-R allele, heterozygous and homozygous 10-R allele. For information about socio-demographic variables, see Table 1.

DA Transporter 30 bp Polymorphism
In the following analysis, n = 662 participants were included. Subjects were subdivided into three groups: 4-R homozygotes, heterozygotes, and 5-R homozygotes. For information about socio-demographic variables, see Table 1.

DA Receptor D4 48 bp Polymorphism
In the following analysis, n = 646 participants were included. We subdivided the sample into two groups based on the 7-R allele: participants carrying at least one 7-R allele were categorized as ''long'' while all others were categorized as ''short'' group (Asghari et al., 1995;Eisenberg et al., 2007;Dreber et al., 2009). For information about socio-demographic variables, see Table 1.
There was a significant time × quality of memory interaction (F (2,651) = 14.591, p < 0.001, η 2 = 0.022) and a significant reward × quality of memory × genotype interaction (F (1,651) = 3.868, p = 0.050, η 2 = 0.008). This latter interaction was further explored with post hoc tests, FIGURE 5 | (A) Reaction time of the DAT1 30 bp genotype. A significant interaction between group and reward was driven by faster responses for rewarded vs. not-rewarded items in the heterozygous and homozygous 5-R group (p < 0.001) but not in the homozygous 4-R group (p = 0.882). (B) Differences in memory performance of the DAT1 30 bp genotype. Recognition memory performance decreased over time. However, this difference was most pronounced in the homozygous 4-R group. * p < 0.05; * * * p < 0.001.

FIGURE 6 | (A)
Memory performance for DRD4 48 bp polymorphism. Based on the main effect of group (see text), a post hoc t-test revealed a trend for higher CHR for the short allele carriers (7-repeat allele absent) as compared to the long allele carriers (t 644 = −1.928, p = 0.054). (B) Reward × quality of memory × genotype interaction. Bonferroni corrected post hoc analyses revealed significantly lower CHR-rem in the long as compared to the short allele group for not-rewarded items only (p = 0.026). There were no differences between groups in all other categories (p's > 0.209). Short refers to 7-allele absent, and long to 7-allele present in the DRD4 48 bp, * p < 0.05. which indicated a difference between genotype groups regarding non rewarded recollection-based memory performance, see Figure 6B. No other interactions were statistically significant (p > 0.3).
As in all previous analyses, there was a significant interaction between time × quality of memory (F (1,658) = 6.684, p = 0.010, η 2 = 0.010) but no significant main effect of reward. All other interactions were not statistically significant (p > 0.16).

DISCUSSION
In a reward-based long-term recognition memory task and a sample of 669 healthy human adults, we can show that reward accelerates response times and accuracy. While only one genotype polymorphism, namely DAT 30 bp, could be linked to reward dependent response times during encoding, reward had no impact on declarative long-term recognition memory and there was no interaction with any of the genotypes tested here. However, there were differential effects of genotype polymorphisms on declarative long-term memory: (a) DAT1 40 bp was linked to the quality of memory with a more pronounced difference between recollection and familiarity in the heterozygous and homozygous 10R as compared to homozygous 9-R; (b) DAT1 30 bp was linked to memory decay, which was most pronounced in homozygous 4-R; (c) DRD4 48 bp was linked to overall recognition memory with higher performance in the short allele group; and (d) CNR1 was linked to overall memory with reduced performance in the homozygous short group. These findings give new insights into how polymorphisms, that are related to dopaminergic neuromodulation, affect different aspects of long-term recognition memory performance.
As expected, cues that predict monetary rewards invigorate and drive behavioral performance (Figure 2). Specifically, this was expressed in faster response times and higher accuracy during encoding, which is in accordance with previous results (Knutson et al., 2001;Pessiglione et al., 2007;Bayer et al., 2013;Steiger and Bunzeck, 2017). At the physiological level, these effects may be linked to the dopaminergic system as suggested by computational models and empirical studies in animals and human subjects (Niv et al., 2007;Guitart-Masip et al., 2011;Dayan, 2012;Beierholm et al., 2013). For instance, individual reward-related response times in older participants could be predicted by the structural integrity of the dopaminergic substantia nigra/ventral tegmental area (SN/VTA) as measure by magnetization transfer imaging (Steiger and Bunzeck, 2017). In line with a DA-reward hypothesis (Wise, 1982), here we found a link between response times and DAT 30 bp. The heterozygous and homozygous FIGURE 7 | Memory performance in the CNR1 polymorphism. Based on a main effect of group (see text), Bonferroni corrected post hoc t-tests revealed significant differences in CHR between homozygous short and heterozygous group (p = 0.018) but not between homozygous short and long (p = 0.144) and homozygous long and heterozygous group (p = 0.615) for overall memory performance. CHR, corrected hit rates; CNR1, cannabinoid receptor type 1, * p < 0.05. 5-R groups showed significantly faster response times for rewarded trials as compared to not-rewarded trials; however, no such effect was observed in the homozygous 4-R group ( Figure 5A). This may suggest impaired neural responses to reward predicting cues in the homozygous 4-R group, which could be further investigated with functional brain imaging such as fMRI or PET.
In contrast to our predictions, there was no effect of reward on recognition memory, and no interaction with genotype. The former hypothesis was based on the assumption that reward not only drives response times and physical effort but also cognitive performance including long-term memory. For instance, retrieval performance was increased for rewarded scenes images (Wittmann et al., 2005;Adcock et al., 2006), words (Gruber and Otten, 2010), or photographs (Shigemune et al., 2010), and everyday objects with different motivational value (Schomaker and Wittmann, 2017). However, not all studies showed a positive effect of reward on cognition, including learning and memory (Callan and Schweighofer, 2008;Ariely et al., 2009;Sharifian et al., 2017;Steiger and Bunzeck, 2017), and some have even reported detrimental effects (Mobbs et al., 2009;Chib et al., 2012;Kuhbandner et al., 2016). While psychological explanations, for instance, related to over motivation or anxiety, might help to explain these opposing results (Callan and Schweighofer, 2008;Mobbs et al., 2009), they might also relate to differences in task design. Specifically, tasks with cues that initially indicate a reward for correct subsequent retrieval (Adcock et al., 2006;Gruber and Otten, 2010;Wittmann et al., 2013) appear to be more robust as compared to designs used here, in which a cue predicts an immediate reward for a correct response (see also Steiger and Bunzeck, 2017).
Along the same lines, although the effect of reward on memory has been shown for several stimulus materials (see above) and retention intervals-including 24 h (Krebs et al., 2009;Bunzeck et al., 2012), it remains unclear whether the faces used here might be special. In particular, we presented male and female faces without scalp hair and background information, which typically helps to form associations and therefore drives recollection. Moreover, the reward task was administered in the context of several other tasks (some of which also included reward) over 2 days (see ''Materials and Methods'' section), which might lead to interferences and therefore reduced the effects of reward on long-term recognition memory. Indeed, on average the CHRs were rather low (Figure 3). In any case, a parsimonious explanation for our absent interaction between reward-based long-term memory and genotype might relate to the overall absence of a reward effect on long-term memory. Therefore, further studies should include a task with a more robust reward effect.
The main finding of our study is that dopaminergic genotypes had differential effects on several aspects of declarative long-term recognition memory. First, there was a significant main effect of memory with higher recollection as compared to familiarity scores, and, importantly, this effect interacted with the DAT 40 bp genotype. Specifically, recollection was enhanced in heterozygous participants and carriers of the homozygous 10-R; however, there was no significant difference between recollection and familiarity scores in homozygous 9-R carriers (Figure 4). This observation is consistent with dual-process models of recognition memory suggesting a distinction between recollection and familiarity (Yonelinas et al., 2010). While the hippocampus and posterior parahippocampal gyrus are closely associated with recollection, the anterior parahippocampal gyrus is more associated with familiarity (Ranganath et al., 2004;Bowles et al., 2007;Diana et al., 2007;Sauvage et al., 2008;Vann et al., 2009;Martin et al., 2013). Therefore, the hippocampus appears to be more critical for recollection as compared to familiarity. Concerning genetic variations, the DAT 9-R allele has been associated with stronger SN/VTA activity in an episodic memory task, which was, however, not present at the behavioral level (Schott et al., 2006). In contrast, another study could link the DAT 10-R allele to hippocampal activity (Bertolino et al., 2008), which is more compatible with our findings of enhanced recollection in the homozygous 10-R but not the homozygous 9-R group. Therefore, our findings support dual-process models and they suggest that variations in the DAT1 40 bp polymorphisms contribute to interindividual differences in recollection-and familiaritybased recognition memory possibly via hippocampal activity and dopaminergic neurotransmission.
Second, recognition memory significantly decreased from day 1 to day 2, which is in line with the notion of a memory decay over time (e.g., Schandry, 2011), and, importantly, this effect interacted with the DAT1 30 bp polymorphisms ( Figure 5B). The decay of recognition memory over 2 days was most pronounced in the homozygous 4-R group, or, conversely, less pronounced in the homozygous 5-R group. Past research has associated the DAT1 30 bp polymorphism 5-R allele with impulsivity (Paloyelis et al., 2010), ADHD Simpson et al., 2010), and addiction O'Gara et al., 2007;Smirnova et al., 2011), possibly through the modulation of dopaminergic processes. The present findings suggest that the 4-R allele affects recognition memory by enhanced forgetting rates. At the physiological level, this might be related to higher expression of the DAT in 4-R carriers, leading to decreased DA in the synaptic cleft and, therefore, less neurotransmission. Conversely, homozygous 5-R carries may express more DAT, leading to temporally more stable memory representations.
Third, overall recognition memory performance was higher in short allele carriers of the DRD4 48 bp polymorphism as compared to long allele carriers (at least one 7-R allele, Figure 6). This is partly consistent with previous research demonstrating decreased working memory performance in children carrying the 7-R allele (Froehlich et al., 2007;Altink et al., 2012). In contrast, others found better performance in cognitive tasks, including short-term working memory, in 7-R allele carriers (Boonstra et al., 2008), which might be mediated by ADHD status and age (Altink et al., 2012). Alternatively, the DRD4 gene might be relevant for attention selection of highly relevant information. This has been suggested since long allele carriers demonstrated increased selective attention to ''high-priority items'' in a category learning and operation span task (Gorlick et al., 2015). A similar attention effect with advantages for long allele carriers has been shown in the context of emotional faces (Wells et al., 2013). Both studies (Wells et al., 2013;Gorlick et al., 2015) would predict a clear behavioral advantage for rewarded (i.e., high priority) items in the long allele group. However, this was not the case here. Together, in our study, the short allele carriers of the DRD4 48 bp polymorphism showed overall enhanced recognition memory performance, which fits the notion that DA plays a role in encoding novel information into long-term memory (Lisman and Grace, 2005;Lisman et al., 2011). However, this size of the effect-like most others we observed (see Table 1) was rather weak and needs to be replicated in future studies.
Fourth, variations in the CNR1 polymorphism were linked to overall memory with reduced performance in the homozygous short group (Figure 7A). More specifically, carriers of the short homozygous allele (≤12-R/≤12-R) showed decreased memory performance compared to the heterozygous group (≤12-R/>12). This effect contrasts a previous study showing increased performance in working memory in the short homozygous group (Ruiz-Contreras et al., 2013), but it supports the more general observation that the CNR1 polymorphism modulates memory performance (Ruiz-Contreras et al., 2011). At the physiological level, activation of CNR1 receptors in the ventral hippocampus of rats enhanced neuronal firing of DA neurons in the VTA and, at the same time, decreased activity of non-DA neurons; further, it increased reward salience and impaired social behaviors (Loureiro et al., 2015). Since the CNR1 polymorphism has also been associated with reward-related traits and disorders (Ponce et al., 2003;Martínez-Gras et al., 2006;Chavarría-Siles et al., 2008), we expected an interaction with reward processing. However, those studies have been inconsistent in their association with a specific allele. For instance, the allelic distribution of the CNR1 seems to be heterogeneous among populations (Ruiz-Contreras et al., 2011), and alleles that were associated with reward-related behavior were not even present in this sample so that their influence could not be detected. Taken together, in line with the notion of a link between CNR1 and the dopaminergic system, our findings suggest that CNR1 modulates long-term memory processes possibly through modulation of DA firing.
We would like to point out that several studies we mentioned above to interpret our findings do not necessarily relate to declarative long-term memory but instead to working memory (or other cognitive domains such as attention). According to traditional models, declarative long-term and working memory rely on partly overlapping but-importantly-distinct brain regions (Squire et al., 2004;Lara and Wallis, 2015;Squire and Dede, 2015;Miller et al., 2018). Specifically, while studies on working memory emphasize the role of the prefrontal cortex, models of declarative long-term memory focus on the medial temporal lobe. Under the assumption that genes and associated polymorphisms may also act in a regionally specific manner (e.g., Schott et al., 2006;Yacubian et al., 2007), the parallels we draw need to be treated with caution. To investigate this issue further, future studies are required.
The effects of DA are not only limited to novelty encoding, but it also modulates long-term memory via consolidation and retrieval activity. For instance, in Wittmann et al. (2005), the behavioral effects of reward on recognition memory were most pronounced after a retention interval of 3 weeks, which is consistent with a role of DA in the late phase of long-term potentiation (LTP, see e.g., Lisman and Grace, 2005). Along the same lines, research in animals could show that reward enhances hippocampal reactivation (Singer and Frank, 2009) and hippocampal DA regulates the persistence of long-term memory (Rossato et al., 2009). Finally, pharmacological modulation of the dopaminergic system in humans drives memory retrieval, which provides evidence for a role of DA in episodic memory retrieval (Clos et al., 2019a,b). Concerning our findings, this suggests that polymorphisms affecting dopaminergic neuromodulation may relate to encoding, consolidation, and/or retrieval. A clear distinction between all three components was not possible here and should, therefore, be addressed in future studies for instance with fMRI.
We acknowledge that several genotypes, other than the ones investigated here, also relate to dopaminergic neuromodulation and as such could impact learning and memory processes. These include, for instance, COMT Val158Met (Bilder et al., 2004); DRD2 (Richter et al., 2017), DRD3 , CHRNA4 (Markett et al., 2009), and DARPP-32 (Schuck et al., 2013;Persson et al., 2017). However, further investigating all potentially relevant genes is beyond the scope of the current study and would need to involve other methodological and statistical approaches.
Another important question that arises based on our results is the link and possible interaction between the investigated genes. Indeed, in a previous study, reward-related hemodynamic activity in the ventral striatum could be related to the combination of DAT and COMT polymorphisms (Yacubian et al., 2007). More recent work has confirmed such epistatic gene-gene interactions, for instance, by showing that COMT and DRD3 together modulate behavior in children with ADHD (Fageera et al., 2020) and that DRD4 in combination with COMT modulate the clinical responses to clozapine in schizophrenia patients (Rajagopal et al., 2018). In our study, the most apparent link to the investigated polymorphisms is their impact on dopaminergic neuromodulation. However, the exact processes, including the relationship to different types of memory, underlying brain regions, and possible gene-gene interactions, remain unclear and need to be addressed in future studies.
Finally, although the observed genotypic distribution of our SNPs did not significantly deviate from expectation according to the Hardy-Weinberg equilibrium, it should be noted that some effects appear to be driven by the less frequent homozygotes in three of the four variants investigated (DAT1 40 bp, DAT1 30 bp, CNR1), see Figures 4, 5, 7. Therefore, future work should replicate our findings.
Together, in a cohort of 669 healthy human adults, we can show that reward accelerates response times and accuracy, but it did not affect subsequent recognition memory. Four a priori selected genotypes, previously associated with the dopaminergic system, could be related to different aspects of recognition memory. Quality of memory was linked to DAT1 40 bp, memory decay was linked to DAT1, and overall recognition memory was linked to DRD4 48 bp and CNR1. As such, our findings give new insights into how interindividual differences in learning and memory processes relate to genes that modulate the activity of the dopaminergic system.

DATA AVAILABILITY STATEMENT
This manuscript contains previously unpublished data. The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the ethics committee of the Hamburg Medical Association. The participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
CL and NB analyzed the data and wrote the manuscript. DS acquired the data. RK, TS, and NB designed the experiment. DS, RK, and TS proofread the manuscript. All authors contributed to the article and approved the submitted version.

FUNDING
This work was supported by a State of Hamburg Excellence initiative (Landesexzellenzcluster 12/09 ''neurodapt'') and the Land Schleswig-Holstein within the funding program Open Access Publikations fonds.