In the 1920s, two psychologists taught a young child known as ‘Little Albert’ to fear a white rat. They did so by striking a metal bar with a hammer whenever the rat was present. After experiencing the rat and noise together on multiple occasions, Little Albert eventually began to cry whenever the rat appeared. However, he also showed a similar response to a number of other white furry objects, including a rabbit and even a fur coat. By applying knowledge about a familiar object to other similar stimuli, humans and other animals avoid having to learn about each and every stimulus from scratch. However, this stimulus generalization is only effective if it occurs to the correct degree: under- or over-generalization (as shown by Little Albert) can lead to behaviors that are less than optimal.

Nerve cells use molecules called neurotransmitters to communicate with each other. For example, one nerve cell might release a neurotransmitter called dopamine, which can be detected on the surface of another nerve cell by a protein called the dopamine D2 receptor. Research suggests that dopamine is involved in stimulus generalization in rat and pigeons, but the effects of dopamine have not been studied in humans. Kahnt and Tobler have now explored how the degree of stimulus generalization is determined in human adults.

For the experiments, volunteers viewed images that contained sets of parallel lines, while lying inside a “functional MRI” brain scanner. The volunteers learned to associate lines with a particular orientation (e.g. 39° from vertical) with receiving a reward, and lines with another specific orientation with the absence of a reward. The next day, half the volunteers were given a drug that blocks the dopamine D2 receptor, while the other half received a placebo. All volunteers were then asked to classify lines of different orientations as either “rewarded” or “non-rewarded”.

Both groups classified lines that were aligned similarly to the previously rewarded orientation as also rewarded. However, those given the D2 receptor blocker classified a narrower range of line orientations as rewarded than those who received the placebo. In other words, blocking the D2 receptors reduced stimulus generalization. It also reduced activity in a region of the brain called the hippocampus, and the extent to which this activity was connected to activity in another area called the midbrain.

Taken together, these results suggest that activation of dopamine D2 receptors in the hippocampus may determine the extent to which we generalize between stimuli. Given that over- and under-generalization is a feature of psychiatric disorders such as depression and anxiety, manipulating D2 receptor activity could have therapeutic benefits in these patients.

Generalization enables neural systems to apply stimulus-outcome associations that have been acquired for one particular stimulus to other, related stimuli. A key aspect of generalization is the degree to which learned associations are applied to novel stimuli: the width of generalization. Generalization relieves individuals from having to learn outcome predictions for every single stimulus from scratch before using them to guide behavior. However, generalizing too widely can be maladaptive because it leads to the indiscriminate approach (or avoidance) of stimuli that are unlikely to be associated with reward (or punishment). Indeed, aberrant generalization is implicated in several neuropsychiatric diseases including schizophrenia, anxiety disorders, depression, and drug abuse (Dunsmoor and Paz, 2015; Gotlib and Joormann, 2010; Lissek et al., 2014b; Lucantonio et al., 2015; Moustafa et al., 2010; Shohamy et al., 2010). Insights into the neurobiological mechanisms controlling the width of generalization are therefore important for understanding adaptive behavior and its disruption in these conditions.

Two basic forms of generalization can be distinguished based on what constitutes the relation among stimuli; associative and stimulus generalization. In the case of associative generalization, such as transitive inference and acquired equivalence, the associative relationship among stimuli determines similarity. This relationship can be established for example by sensory preconditioning (e.g., train stimuli A-B and B-C, test whether A comes to predict C) or by a common associate (e.g. train A-C and B-C, test whether A and B are associated). In contrast, in stimulus generalization the relationship among stimuli is based on the similarity along one or more perceptual dimensions (frequency of sounds, color, line orientation, etc.). Most of what we know about stimulus generalization comes from behavioral experiments utilizing intradimensional stimulus discrimination (Dunsmoor and LaBar, 2013; Ghirlanda and Enquist, 2003; Guttman and Kalish, 1956; Hanson, 1959). In these paradigms, one stimulus (e.g. one particular line orientation) is paired with reward (rewarded conditioned stimulus, CS+), while a second stimulus (e.g. a slightly different line orientation), which differs from the first in only one dimension, is paired with no reward (unrewarded conditioned stimulus, CS−). Generalization is then tested using a range of stimuli that vary along the defining stimulus dimension (e.g. line orientation). Although the test stimuli have never been paired with reward, animals and humans show robust generalization in that they respond to test stimuli that are similar to the CS+. Several psychological models have been developed based on these experiments (Ghirlanda and Enquist, 1998; Pearce, 1987; 1994; Shepard, 1987; McLaren et al., 2012), but the neurobiological mechanisms regulating the width of stimulus generalization have remained unknown.

Early research suggested a role for dopamine in mediating generalization by demonstrating that blockade of dopamine receptors during generalization tests alters response gradients in rats and pigeons (Lyons et al., 1973a; 1973b; Terrace, 1963). While the effects of dopamine have not been investigated in humans, evidence from neuroimaging suggests that dopaminoceptive and dopaminergic regions such as the striatum and the midbrain are involved in generalization (Kahnt et al., 2012; Wimmer et al., 2012; Wimmer and Shohamy, 2012). However, because standard neuroimaging relies on indirect measurements of neural responses, these studies were unable to inform questions about neurotransmitter-specific activity.

A candidate brain region for a prominent role in stimulus generalization is the hippocampus, based on its involvement in learning, memory, and associative generalization (Dickerson and Delgado, 2015; Eichenbaum, 2000; Frank et al., 2006; Kumaran, 2012; Norman and O'Reilly, 2003; Squire and Wixted, 2011). Specifically, the hippocampus is thought to contribute to associative generalization by representing higher-order relationships among different stimuli (Eichenbaum, 1999; Howard et al., 2005; Kumaran and McClelland, 2012). However, human imaging and animal lesion studies suggest that the hippocampus may also be involved in similarity-based stimulus generalization not requiring inference (Casasola et al., 2007; Kahnt et al., 2012; Lissek et al., 2014a; Solomon and Moore, 1975).

In the current study, we directly examined the role of dopamine and hippocampal processing in regulating the width of stimulus generalization. For this purpose, we combined a visual intradimensional discrimination task (Kahnt et al., 2012) with pharmacologic dopamine D2-receptor (D2R) blockade (Kahnt et al., 2015) and functional magnetic resonance imaging (fMRI). Our central hypothesis was that dopamine is involved in regulating the width of stimulus generalization via the modulation of similarity-based processing in the hippocampus. In order to test this hypothesis, we fit a computational model of stimulus generalization to the behavioral data, and examined the model parameters regulating the width of generalization. We then used model-based fMRI and functional connectivity analyses to identify the brain processes that potentially mediate the effects of dopamine on generalization. We predicted that D2R blockade during the test session would reduce the computational parameters governing the width of generalization, and thus lead to narrower generalization gradients. Moreover, we hypothesized that this reduction would be mirrored in hippocampal activity as well as in reduced functional coupling between the dopaminergic midbrain and the hippocampus.

The experiment was carried out on two consecutive days. On the first day, subjects underwent intradimensional discrimination training to learn stimulus-specific reward associations. On the second day, a generalization test was performed in extinction. All subjects received placebo on the first day, whereas on day two, one group received placebo (PP group) and the other group received 400 mg of the D2R blocker amisulpride (PA group) (Figure 1A). By verifying that the two groups learned equally well on day one, we ensured that any observed group differences in behavioral or neural responses during day two resulted from effects of dopamine on generalization at retrieval, independent of encoding-related effects. In the following sections we first describe the behavioral data during the training, and then examine how D2R blockade affected the neurocomputational mechanisms governing generalization using fMRI and a model of stimulus generalization.

Figure 1

Download asset Open asset

Robust acquisition of stimulus-outcome associations during discrimination training

The visual intradimensional discrimination task used the orientation of a Gabor patch as reward-relevant dimension (Figure 1B). Specifically, one orientation (39°) served as the CS+ and was paired with the delivery of 20 cents in 50% of trials, whereas a second orientation (51°) served as the CS− and was paired with no reward in all trials (Figure 1D). The association between orientation and reward was counterbalanced across subjects. In order to track the acquisition of stimulus-outcome associations, subjects were asked to make a discriminatory response after the stimulus was displayed, but before the outcome was shown. Subjects had to indicate whether the current stimulus was the rewarded stimulus, the non-rewarded stimulus, or whether they were unsure, by pressing buttons associated with +, -, or x, respectively. Subjects learned the discrimination within the first 50 trials and performed at high levels afterwards (Figure 2A). Performance increased as a function of time (two-way, group-by-time ANOVA, main effect of time, F(39,1716) = 18.98, P < 0.001), but did not differ between groups (main effect of group, F(1,44) = 0.41, P = 0.53; group-by-time interaction, F(39,1716) = 1.27, P = 0.13; two-way, group-by-CS type ANOVA, main effect of group, F(1,44) = 0.35, P = 0.56; group-by-CS type interaction, F(1,44) = 0.39, P = 0.53, Figure 2B). Moreover, learning-related activity in the ventral striatum (Delgado, 2007) did not differ between groups (see Figure 3), suggesting that both groups were also comparable in terms of neural responses during discrimination training. This demonstrates that, as expected, both groups acquired stimulus-outcome associations and performed at comparable levels during training. Accordingly, effects of D2R blockade during the generalization test session on the next day can be compared independent of potential group differences in discrimination training.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 3

Download asset Open asset

D2R blockade during test narrows generalization gradients

On the second day, subjects in the PP group received placebo, whereas subjects in the PA group received 400 mg of the D2R antagonist amisulpride (Rosenzweig et al., 2002). One hour later, subjects performed a generalization test session in extinction (without feedback). Specifically, on each trial, one of 15 orientations (17°–73°, Figure 1D) was presented, and subjects performed the same discrimination as during training (Figure 1C; the original CS+ and CS− orientations were not shown during the test). Responses described a bell-shaped gradient around the CS+ and revealed a peak shift (Derenne, 2010; Purtle, 1973; Wisniewski et al., 2009), such that subjects responded most frequently to an orientation that was never paired with reward (Figure 2C). Specifically, average responding was stronger to orientations on the side of the CS+ that was farther away from the CS− (paired t-test on responding to stimuli left vs. right of the CS+, t = 2.99, P = 0.006). Such peak shifts are typical for intradimensional discrimination with one CS+ and one CS−; they are thought to result from the summation of excitatory and inhibitory gradients around the CS+ and CS−, respectively, and have been observed across many species and stimuli (Ghirlanda and Enquist, 2003; Pearce et al., 2008; Spence, 1937).

Direct group comparisons of the individual data points along the generalization gradient did not reveal any significant differences (two-sample t-tests, all Ps > 0.29). However, visual inspection of the gradients suggested that the amisulpride group had a narrower gradient than the placebo group, with enhanced responding at the peak of the curve, reduced responding at both flanks, and enhanced responding at the tail of the curve. These shape features are parsimoniously described by the 4th moment of probability distributions, namely, their kurtosis. Accordingly, a test for differences in the kurtosis of group-specific distributions (Pearson type VII distribution, see Materials and methods) revealed a significantly greater kurtosis in the amisulpride group compared to placebo (PA: 6.73, PP: 3.29; permutation test, P = 0.043). This finding demonstrates that amisulpride narrowed the width and increased the peak of the behavioral generalization gradient, and suggests that D2R activity alters the neurocomputational processes that mechanistically control generalization behavior.

To further address this possibility, and to identify the specific computational parameters that are affected by D2R blockade, we utilized a mathematical model for similarity-based stimulus generalization (see Materials and methods and Figure 4). The model assumes that the reward prediction of a given stimulus reflects the integrated excitatory and inhibitory associations of that stimulus, plus the excitatory and inhibitory associations of stimuli that are similar to it (Pearce, 1987). Critically, associations of stimuli that are similar to the currently presented stimulus have a stronger contribution than the associations of dissimilar stimuli. Because the shape of the function determining the similarity between the currently presented stimulus and other stimuli (i.e. the generalization coefficient) is of critical importance (Ghirlanda and Enquist, 2003), we directly compared the most commonly used models, i.e. one with Gaussian (Kahnt et al., 2012) and the other with exponential similarity functions (Shepard, 1987). While the exact shape of the similarity functions differs between models, for both models, the extent to which inhibitory and excitatory associations generalize to the current stimulus is controlled by the width of the similarity functions (s_i and s_e) (Figure 4B). The larger s_i and s_e, the stronger the impact of the inhibitory and excitatory associations of dissimilar stimuli on the currently predicted reward, respectively, and thus the more generalization takes place.

Figure 4

Download asset Open asset

In order to assess the explanatory power of the Gaussian and exponential similarity functions, we directly compared their fit to the behavioral data. The free parameters of both models (width of inhibitory generalization coefficient, s_i; width of excitatory generalization coefficient, s_e; slope, β; offset, a; learning rate, α) were estimated for the entire group of subjects by maximizing the likelihood of subjects’ responses during the generalization test given the model (see Materials and methods). Visual inspections suggested that the model with the Gaussian similarity function fitted the behavioral data better than the model with the exponential similarity function (Figure 2—figure supplement 1). This was confirmed by a formal model comparison using the Akaike information criterion (AIC) and Bayesian information criterion (BIC) (Gaussian: AIC = 9586.4, BIC = 9629.2; Exponential: AIC = 9691.6, BIC = 9734.4). We also compared the fit of the two models by comparing the regression coefficients from a logistic regression of the trial-by-trial responses on the modeled P(+) responses. Although both models predicted behavioral responses reliably (Gaussian: t = 11.06, P < 0.001; exponential: t = 11.38, P < 0.001), we found significantly higher regression coefficients for the Gaussian model (paired t-test t = 5.34, P < 0.001). Taken together, this demonstrates that in our experiment, a Gaussian similarity function fits behavior better than an exponential similarity function.

To determine the effects of dopamine on generalization, in a next step, the free parameters of the Gaussian model were estimated separately for each group (see Materials and methods). As can be seen in Figure 2D, the model (with group-specific parameters) accurately reproduced responses in both groups, including the differences in the shape of the generalization gradients. Logistic regression coefficients were significantly different from zero (PA: t = 7.76, P < 0.001; PP: t = 7.95, P < 0.001), and did not differ between groups (t = -0.21, P = 0.84), suggesting that responses in both groups were well described by the model. Notably, comparing the model parameters between groups (Table 1), revealed significant group differences in the width of the excitatory generalization coefficient, with a smaller coefficient in the amisulpride group compared to the placebo group (permutation test, P = 0.035). The width of the inhibitory coefficient was also smaller in the amisulpride group, but this difference was not significant (P = 0.069). Importantly, the learning rate during test did not differ between groups (P = 0.21), demonstrating that additional learning during the test session, which might have been altered by amisulpride, cannot account for the differences in generalization gradients. In order to obtain individual estimates of model parameters, we re-estimated the model using a leave-one-out procedure (see Materials and methods). Comparing the resulting individual estimates between groups confirmed a significant difference in the width of the excitatory (two-sample t-test, t = -2.03, P = 0.024), and to a lesser degree, the inhibitory generalization coefficient (t = -1.59, P = 0.059, Figure 5). These results suggest that D2R blockade modulates the computational processes that control the width of stimulus generalization, resulting in narrower generalization.

Figure 5

Download asset Open asset

Table 1

Group	Inhibitory coefficient si	Excitatory coefficient se	Slope β	Offset a	Learning rate αtest
PA group N=25	20.121	17.599	3.093	0.370	0.002
PP group N=21	30.587	24.584	3.124	0.361	0.006
Difference PA-PP	-10.465	-6.985	-0.031	0.009	-0.004
*P-value (PA-PP)	0.069	0.035	0.495	0.462	0.214

1. Note: *P-value is based on 10.000 permutations.

In principle, the observed effects of amisulpride on the width of generalization could have resulted from a drug-induced improvement in perceptual orientation-discrimination. To control for such perceptual effects, subjects performed a challenging orientation discrimination task, once before the drug took effect and once after (see Materials and methods and Figure 6). Performance (percentage correct) on this task did not differ between groups (two-way, time-by-group ANOVA, main effect of group, F(1,43) = 0.99, P = 0.33) and there was no group-by-time interaction (F(1,43) = 0.42, P = 0.52). Moreover, while perceptual discrimination performance was reliable across time (correlation between pre- and post-drug performance, r = 0.52, P < 0.001), it was not related to the width of the estimated generalization coefficients (all Ps > 0.38). Taken together, this control analysis demonstrates that the effects of amisulpride on generalization cannot be explained by perceptual improvements in orientation discrimination per se.

Figure 6

Download asset Open asset

D2R blockade reduces similarity-based activity in the hippocampus

Having established an effect of D2R blockade on the computational processes that govern stimulus generalization, we next examined the neural circuits that mediate these changes. We first identified brain regions involved in generalization of reward predictions during retrieval. As a proxy of generalized value, we focused on prediction error responses derived from our model, which reflect the extent to which reward predictions have generalized from the original CS+ and CS− to the current stimulus (please note that because no outcomes were shown, prediction errors are perfectly but negatively correlated with expected value). Accordingly, to identify brain regions involved in similarity-based computations during generalization, we searched for regions in which fMRI activity correlated with generalized prediction errors at the time of the expected outcome. Based on previous empirical and modeling work linking hippocampal activity to the representation of relationships between stimuli and their predicted value (Kumaran et al., 2009; Kumaran and McClelland, 2012; Lee et al., 2012; Lissek et al., 2014a; Wimmer and Shohamy, 2012), we expected fMRI signals in the hippocampus to positively correlate with generalized prediction errors. In line with this hypothesis, across the entire group (one sample t-test) we found significant correlations in the bilateral hippocampus (extending into the parahippocampal gyrus, left, x = -30, y = -22, z = -16, t = 6.53, P = 0.001, FWE whole brain corrected; right, x = 33, y = -19, z = -16, t = 7.11, P < 0.001, FWE whole brain corrected, Figure 7A). Similar effects were found in the left amygdala (x = -24, y = -4, z = -19, t = 6.04, P = 0.006, FWE whole brain corrected) and the bilateral middle temporal gyrus (left, x = -48, y = -73, z = 14, t = 6.41, P = 0.002, FWE whole brain corrected; right, x = 28, y = -64, z = 14, t = 5.57, P = 0.022, FWE whole brain corrected). In addition, supporting recent work highlighting a role for medial (mPFC) and ventromedial PFC (vmPFC) in generalization (Dunsmoor and Paz, 2015; Onat and Buchel, 2015), at an uncorrected threshold of P < 0.001, we found a cluster in the mPFC (x = -3, y = 56, z = 8, t = 4.03) extending into the vmPFC.

Figure 7

Download asset Open asset

We next tested whether reduced behavioral generalization observed in the amisulpride group was paralleled by a decrease in generalization-related activity in the hippocampus. In line with this idea, we found significantly reduced activity in the hippocampus in the amisulpride relative to the placebo group (two-sample t-test, t = -2.12, P = 0.02, Figure 7B). To examine whether these effects of dopamine on generalization-related activity are specific to the hippocampus, as a control, we tested for similar group differences in the amygdala, middle temporal cortex and mPFC. No significant group differences were observed in the amygdala (P = 0.43), the middle temporal gyrus (P = 0.31), or the medial PFC (P = 0.98). However, post-hoc analyses directly comparing the effect of the drug in the hippocampus to the drug effect in the other regions (i.e. group-by-region interactions), demonstrated that while the effect of D2R blockade in the hippocampus was significantly stronger than in the mPFC (P = 0.02) similar interactions involving the amygdala (P = 0.097) and the middle temporal gyrus (P = 0.127) did not reach significance. These data suggest specificity of the effects of D2R blockade on similarity-based processing in the hippocampus relative to the mPFC, but not necessarily relative to the amygdala and middle temporal lobe.

Midbrain-hippocampal connectivity correlates with the width of generalization

In a next step, we examined the neural pathways on which DR2 blockade may mediate its effects on generalization. Given the anatomical origin of the dopaminergic projections to the hippocampus (Swanson, 1982), the relevance of hippocampal D2R for memory functions (Takahashi et al., 2008), and the modulation of hippocampal processing reported above, we hypothesized that amisulpride would reduce the functional connectivity between the midbrain and the hippocampus. In line with this prediction, a functional connectivity analysis with the midbrain as a seed region (Figure 8A, see Materials and methods) revealed decreased midbrain connectivity in the right hippocampus (x = 33, y = -19, z = -19, t = 3.68, P = 0.02, FWE small volume corrected, Figure 8B,C) and the left striatum (x = -9, y = 8, z = -19, t = 4.26, P = 0.001, FWE small volume corrected), in the amisulpride compared to the placebo group. This finding suggests that D2R blockade may modulate the functional connectivity between the midbrain and dopaminergic target regions such as the hippocampus and the striatum. To examine the specificity of these findings, we tested for similar drug-related effects on functional connectivity in the regions involved in similarity-based processing defined above. While connectivity estimates differed significantly between groups in middle temporal gyrus (P = 0.01), no drug effects were observed in the amygdala (P = 0.296) and the mPFC (P = 0.17). Accordingly, for all regions except the middle temporal gyrus (P = 0.13), the corresponding drug-by-region interactions were significant (all Ps < 0.05), suggesting that amisulpride-related decreases in midbrain connectivity are relatively specific to the hippocampus.

Figure 8

Download asset Open asset

Finally, we examined whether the relationship between D2R blockade and the width of generalization is associated with midbrain-hippocampal connectivity. Specifically, we tested the correlation between midbrain-hippocampal connectivity and the estimated width of generalization coefficients in the amisulpride group. This correlation was significant for the inhibitory generalization coefficient (r = 0.47, P = 0.01, Figure 8D), but not for the excitatory coefficient (r = 0.18, P = 0.19). We replicated these findings in our previous data set (Kahnt et al., 2012), showing that midbrain-hippocampal connectivity was significantly correlated with the inhibitory generalization coefficient (r = 0.44, P = 0.036), but not with the excitatory coefficient (r = 0.15, P = 0.49). Interestingly, this relationship was not observed for midbrain-striatal connections (P > 0.16), and was significantly stronger for midbrain-hippocampal compared to midbrain-striatal connectivity (Z = 2.31, P = 0.011), suggesting that the relation between dopamine and the width of generalization may arise primarily from midbrain-hippocampal connections. In summary, these findings suggest a link between dopamine-mediated midbrain-hippocampal coupling and the width of stimulus generalization in humans.

The degree to which individuals generalize outcome predictions across similar stimuli is important for adaptive behavior. Here, using D2R pharmacology, fMRI, and computational modeling, we demonstrate that D2R blockade results in narrower behavioral generalization gradients and changes in the computational parameters controlling the width of generalization. Moreover, D2R blockade altered similarity-based processing in the hippocampus and decreased the functional coupling between the midbrain and the hippocampus. This coupling was in turn related to the computational parameter controlling the width of generalization.

Previous empirical and modeling work suggests that the hippocampus contributes to generalization by detecting the relationship between items in memory (Eichenbaum, 1999; Howard et al., 2005; Kumaran and McClelland, 2012). However, this function was thought to only apply to associative forms of generalization involving higher-order relationships, whereas basic stimulus generalization involving perceptual similarity is suggested to be hippocampus-independent (Kumaran, 2012). Opposing this view, here we provide evidence that the hippocampus is involved in stimulus generalization. Accordingly, our findings suggest that hippocampal similarity computations are not restricted to detecting higher-order relationships among stimuli as previously thought, but can also exploit the perceptual similarity between stimuli to establish meaningful relationships. While the hippocampus may not be necessary for stimulus generalization per se, our data suggest that it facilitates generalization by allowing a flexible modulation of its width. Specifically, such flexibility could not be achieved if discrimination and generalization were entirely based on static, hippocampus-independent, stimulus-outcome associations.

Our results contribute to an ongoing debate regarding the time point at which hippocampal processes support generalization. Two alternative accounts suggest that the relationship among stimuli is established by the hippocampus either during encoding via overlapping neural codes (Eichenbaum, 1999; Howard et al., 2005), or at the time of retrieval by means of recurrent similarity computations that are based on separated neural codes (Kumaran and McClelland, 2012). The former proposal has received empirical support from studies showing that hippocampal activity during encoding is related to transfer performance at test (Shohamy and Wagner, 2008; Wimmer and Shohamy, 2012). In contrast, in line with the second account, our model assumes that reward predictions are generalized and integrated during test (Pearce, 1987). It should be noted though that this model is mathematically equivalent to a model in which generalization occurs at encoding (Kahnt et al., 2012), and thus, the model alone does not provide evidence for either account. However, only if generalization occurs at retrieval can the width of generalization be modulated after encoding has occurred, and thus, our results are only compatible with a retrieval-based account of generalization. Specifically, because dopamine receptor blockade modulated generalization gradients during test, without affecting additional learning, our pharmacological manipulation provides evidence that stimulus generalization occurs – at least in part – at retrieval.

Our data suggest that the functional connectivity between the dopaminergic midbrain and the hippocampus is related to the width of generalization. Specifically, we found that participants with D2R blockade showed decreased midbrain-hippocampal connectivity, which in turn correlated with the width of generalization. While the presence of dopamine receptors in the hippocampus is undisputed, it is worth keeping in mind that although direct dopaminergic innervation from the midbrain to the hippocampus is present (Gasbarri et al., 1994) and functionally relevant for memory stabilization (McNamara et al., 2014), this pathway is not very strong (Mingote et al., 2015; Swanson, 1982). Moreover, it is conceivable that some of the hippocampal dopamine is co-released from noradrenergic neurons, in which it may not have been completely metabolized.

As such, our data is in line with the idea that reduced activity of hippocampal D2R decreases the extent to which reward associations generalize across stimuli. By extension, enhanced dopaminergic activity in the hippocampus may increase the width of generalization. More specifically, dopamine release in the hippocampus could increase the likelihood that ensemble patterns representing similar stimuli are activated, which would effectively lower the threshold for similarity detection, and, by broadening the range of stimuli for which reward predictions are taken into account, facilitate generalization. Conversely, reduced levels of dopamine would decrease the likelihood of ensemble pattern activation, increase the threshold for similarity detection, and, by enhancing fine-tuned discrimination, reduce generalization. This provides a simple neurobiological mechanism by which dopamine may flexibly adjust generalization during retrieval. Such flexibility is highly adaptive as it allows for different levels of generalization based on the state of the organism or the environment. We speculate that tonic dopamine levels, similar to their enabling effects on movements (Schultz, 2007), play an enabling role in generalization. For instance, in situations where dopamine transmission is high, such as in novel environments (Ihalainen et al., 1999; Lisman and Grace, 2005), elevated levels of dopamine may not only support memory formation (Shohamy and Adcock, 2010), but also broaden generalization and thereby facilitate exploratory behavior.

Of note, although D2R blockade reduced generalization coefficients for both excitatory and inhibitory associations (albeit less reliably), midbrain-hippocampal connectivity was only correlated with the width of the inhibitory coefficient. This raises the possibility that only the effects of D2R blockade on the inhibitory coefficient are mediated via a modulation of midbrain-hippocampal connectivity, whereas the effects on the excitatory coefficient are mediated via a different, yet to be explored, mechanism. The dissociation between inhibitory and excitatory generalization coefficients is in line with the interpretation of previous results suggesting that administration of chlorpromazine, a nonspecific dopamine receptor antagonist, specifically reduces the strength of inhibitory associations in a dose-dependent manner (Lyons et al., 1973b; Terrace, 1963). Moreover, it adds to previous findings indicating that generalization involving appetitive and aversive outcomes may involve different mechanisms (Schechtman et al., 2010).

By revealing the effects of dopamine on generalization, our current results substantially extend those of our previous study (Kahnt et al., 2012). Specifically, the current experiment suggests an association between dopamine and the width of generalization in the hippocampus, and dissociates effects of generalization during retrieval vs. encoding, which could not be achieved with the previous design. However, whereas here we find that prediction errors correlate primarily with activity in the hippocampus, the previous study identified prediction error related activity primarily in the ventral striatum. Also, in the previous study functional connectivity between the striatum and the hippocampus was related to the modeled generalization coefficients, whereas our current results suggest a facilitating role of midbrain-hippocampal connections. Several notable differences in the design of both experiments might explain these discrepancies. Most importantly, the previous design consisted of multiple alternating training and testing blocks, all conducted within one session on a single day, whereas in the current study, training and test did not alternate but were conducted in distinct sessions that were separated by a 24 hr delay. Moreover, in the 2012 study, CS-outcome associations were deterministic (100% contingency), whereas the current experiment involved a 50% reinforcement schedule in order to slow down extinction during test. These differences could have shifted the primary focus of generalization-related processing from the striatum to the hippocampus.

In conclusion, here we propose a neurobiological mechanism for the control of stimulus generalization, in which midbrain dopamine changes similarity computations in the hippocampus, resulting in altered generalization gradients. As such, our results demonstrate that the width of stimulus generalization is not hard-wired but flexible, and can change under pharmacological interventions. Accordingly, our results have important clinical implications for a number of neuropsychiatric disorders in which generalization is disrupted. Specifically, aberrant generalization is implicated in depression, anxiety, and schizophrenia (Buss and Daniell, 1967; Gotlib and Joormann, 2010; Lissek et al., 2014b; Shohamy et al., 2010), and our findings indicate that blocking D2R activity may provide a potential treatment of overgeneralization in these disorders.

1. 1. Buss AH
   2. Daniell EF

   (1967) Stimulus generalization and schizophrenia

   Journal of Abnormal Psychology 72:50–53.

   https://doi.org/10.1037/h0020082

   • Google Scholar
2. 1. Büchel C
   2. Holmes AP
   3. Rees G
   4. Friston KJ

   (1998) Characterizing stimulus–response functions using nonlinear regressors in parametric fMRI experiments

   NeuroImage 8:140–148.

   https://doi.org/10.1006/nimg.1998.0351

   • Google Scholar
3. 1. Casasola C
   2. Mejía-Gervacio S
   3. Cruz-Pérez M
   4. Sánchez-Castillo H
   5. Velázquez-Martínez DN

   (2007) Participation of the dorsal hippocampus in stimulus discrimination with scopolamine

   Brain Research 1178:125–131.

   https://doi.org/10.1016/j.brainres.2007.08.003

   • Google Scholar
4. 1. Delgado MR

   (2007) Reward-related responses in the human striatum

   Annals of the New York Academy of Sciences 1104:70–88.

   https://doi.org/10.1196/annals.1390.002

   • Google Scholar
5. 1. Derenne A

   (2010) Shifts in postdiscrimination gradients within a stimulus dimension based on bilateral facial symmetry

   Journal of the Experimental Analysis of Behavior 93:485–494.

   https://doi.org/10.1901/jeab.2010.93-485

   • Google Scholar
6. 1. Dickerson KC
   2. Delgado MR

   (2015) Contributions of the hippocampus to feedback learning

   Cognitive, Affective, & Behavioral Neuroscience 15:861–877.

   https://doi.org/10.3758/s13415-015-0364-5

   • Google Scholar
7. 1. Dunsmoor JE
   2. LaBar KS

   (2013) Effects of discrimination training on fear generalization gradients and perceptual classification in humans

   Behavioral Neuroscience 127:350–356.

   https://doi.org/10.1037/a0031933

   • Google Scholar
8. 1. Dunsmoor JE
   2. Paz R

   (2015) Fear generalization and anxiety: behavioral and neural mechanisms

   Biological Psychiatry 78:336–343.

   https://doi.org/10.1016/j.biopsych.2015.04.010

   • Google Scholar
9. 1. Eichenbaum H

   (1999) The hippocampus and mechanisms of declarative memory

   Behavioural Brain Research 103:123–133.

   https://doi.org/10.1016/S0166-4328(99)00044-3

   • Google Scholar
10. 1. Eichenbaum H

    (2000) A cortical-hippocampal system for declarative memory

    Nature Reviews Neuroscience 1:41–50.

    https://doi.org/10.1038/35036213

    • Google Scholar
11. 1. Frank MJ
    2. O'Reilly RC
    3. Curran T

    (2006) When memory fails, intuition reigns: midazolam enhances implicit inference in humans

    Psychological Science 17:700–707.

    https://doi.org/10.1111/j.1467-9280.2006.01769.x

    • Google Scholar
12. 1. Gasbarri A
    2. Verney C
    3. Innocenzi R
    4. Campana E
    5. Pacitti C

    (1994) Mesolimbic dopaminergic neurons innervating the hippocampal formation in the rat: a combined retrograde tracing and immunohistochemical study

    Brain Research 668:71–79.

    https://doi.org/10.1016/0006-8993(94)90512-6

    • Google Scholar
13. 1. Ghirlanda S
    2. Enquist M

    (1998) Artificial neural networks as models of stimulus control

    Animal Behaviour 56:1383–1389.

    https://doi.org/10.1006/anbe.1998.0903

    • Google Scholar
14. 1. Ghirlanda S
    2. Enquist M

    (2003) A century of generalization

    Animal Behaviour 66:15–36.

    https://doi.org/10.1006/anbe.2003.2174

    • Google Scholar
15. 1. Gotlib IH
    2. Joormann J

    (2010) Cognition and depression: current status and future directions

    Annual Review of Clinical Psychology 6:285–312.

    https://doi.org/10.1146/annurev.clinpsy.121208.131305

    • Google Scholar
16. 1. Guttman N
    2. Kalish HI

    (1956) Discriminability and stimulus generalization

    Journal of Experimental Psychology 51:79–88.

    https://doi.org/10.1037/h0046219

    • Google Scholar
17. 1. Hanson HM

    (1959) Effects of discrimination training on stimulus generalization

    Journal of Experimental Psychology 58:321–334.

    https://doi.org/10.1037/h0042606

    • Google Scholar
18. 1. Howard MW
    2. Fotedar MS
    3. Datey AV
    4. Hasselmo ME

    (2005) The temporal context model in spatial navigation and relational learning: toward a common explanation of medial temporal lobe function across domains

    Psychological Review 112:75–116.

    https://doi.org/10.1037/0033-295X.112.1.75

    • Google Scholar
19. 1. Ihalainen JA
    2. Riekkinen Jr P
    3. Feenstra MGP

    (1999) Comparison of dopamine and noradrenaline release in mouse prefrontal cortex, striatum and hippocampus using microdialysis

    Neuroscience Letters 277:71–74.

    https://doi.org/10.1016/S0304-3940(99)00840-X

    • Google Scholar
20. 1. Jocham G
    2. Klein TA
    3. Ullsperger M

    (2011) Dopamine-mediated reinforcement learning signals in the striatum and ventromedial prefrontal cortex underlie value-based choices

    Journal of Neuroscience 31:1606–1613.

    https://doi.org/10.1523/JNEUROSCI.3904-10.2011

    • Google Scholar
21. 1. Kahnt T
    2. Grueschow M
    3. Speck O
    4. Haynes J-D

    (2011) Perceptual learning and decision-making in human medial frontal cortex

    Neuron 70:549–559.

    https://doi.org/10.1016/j.neuron.2011.02.054

    • Google Scholar
22. 1. Kahnt T
    2. Park SQ
    3. Burke CJ
    4. Tobler PN

    (2012) How glitter relates to gold: similarity-dependent reward prediction errors in the human striatum

    Journal of Neuroscience 32:16521–16529.

    https://doi.org/10.1523/JNEUROSCI.2383-12.2012

    • Google Scholar
23. 1. Kahnt T
    2. Weber SC
    3. Haker H
    4. Robbins TW
    5. Tobler PN

    (2015) Dopamine D2-receptor blockade enhances decoding of prefrontal signals in humans

    Journal of Neuroscience 35:4104–4111.

    https://doi.org/10.1523/JNEUROSCI.4182-14.2015

    • Google Scholar
24. 1. Kumaran D
    2. McClelland JL

    (2012) Generalization through the recurrent interaction of episodic memories: a model of the hippocampal system

    Psychological Review 119:573–616.

    https://doi.org/10.1037/a0028681

    • Google Scholar
25. 1. Kumaran D
    2. Summerfield JJ
    3. Hassabis D
    4. Maguire EA

    (2009) Tracking the emergence of conceptual knowledge during human decision making

    Neuron 63:889–901.

    https://doi.org/10.1016/j.neuron.2009.07.030

    • Google Scholar
26. 1. Kumaran D

    (2012) What representations and computations underpin the contribution of the hippocampus to generalization and inference?

    Frontiers in Human Neuroscience 6:.

    https://doi.org/10.3389/fnhum.2012.00157

    • Google Scholar
27. 1. Lee H
    2. Ghim J-W
    3. Kim H
    4. Lee D
    5. Jung M

    (2012) Hippocampal neural correlates for values of experienced events

    Journal of Neuroscience 32:15053–15065.

    https://doi.org/10.1523/JNEUROSCI.2806-12.2012

    • Google Scholar
28. 1. Lisman JE
    2. Grace AA

    (2005) The hippocampal-VTA loop: controlling the entry of information into long-term memory

    Neuron 46:703–713.

    https://doi.org/10.1016/j.neuron.2005.05.002

    • Google Scholar
29. 1. Lissek S
    2. Bradford DE
    3. Alvarez RP
    4. Burton P
    5. Espensen-Sturges T
    6. Reynolds RC
    7. Grillon C

    (2014b) Neural substrates of classically conditioned fear-generalization in humans: a parametric fMRI study

    Social Cognitive and Affective Neuroscience 9:1134–1142.

    https://doi.org/10.1093/scan/nst096

    • Google Scholar
30. 1. Lissek S
    2. Kaczkurkin AN
    3. Rabin S
    4. Geraci M
    5. Pine DS
    6. Grillon C

    (2014a) Generalized anxiety disorder is associated with overgeneralization of classically conditioned fear

    Biological Psychiatry 75:909–915.

    https://doi.org/10.1016/j.biopsych.2013.07.025

    • Google Scholar
31. 1. Lucantonio F
    2. Kambhampati S
    3. Haney RZ
    4. Atalayer D
    5. Rowland NE
    6. Shaham Y
    7. Schoenbaum G

    (2015) Effects of prior cocaine versus morphine or heroin self-administration on extinction learning driven by overexpectation versus omission of reward

    Biological Psychiatry 77:912–920.

    https://doi.org/10.1016/j.biopsych.2014.11.017

    • Google Scholar
32. 1. Lyons J
    2. Klipec WD
    3. Eirick R

    (1973a) The effect of chlorpromazine on the peak shift in the albino rat

    Physiological Psychology 1:165–168.

    https://doi.org/10.3758/BF03326895

    • Google Scholar
33. 1. Lyons J
    2. Klipec WD
    3. Steinsultz G

    (1973b) The effect of chlorpromazine on discrimination performance and the peak shift

    Physiological Psychology 1:121–124.

    https://doi.org/10.3758/BF03326882

    • Google Scholar
34. 1. McLaren DG
    2. Ries ML
    3. Xu G
    4. Johnson SC

    (2012) A generalized form of context-dependent psychophysiological interactions (gPPI): a comparison to standard approaches

    NeuroImage 61:1277–1286.

    https://doi.org/10.1016/j.neuroimage.2012.03.068

    • Google Scholar
35. 1. McNamara CG
    2. Tejero-Cantero Álvaro
    3. Trouche S
    4. Campo-Urriza N
    5. Dupret D

    (2014) Dopaminergic neurons promote hippocampal reactivation and spatial memory persistence

    Nature Neuroscience 17:1658–1660.

    https://doi.org/10.1038/nn.3843

    • Google Scholar
36. 1. Mingote S
    2. Chuhma N
    3. Kusnoor SV
    4. Field B
    5. Deutch AY
    6. Rayport S

    (2015) Functional connectome analysis of dopamine neuron glutamatergic connections in forebrain regions

    Journal of Neuroscience 35:16259–16271.

    https://doi.org/10.1523/JNEUROSCI.1674-15.2015

    • Google Scholar
37. 1. Moustafa AA
    2. Keri S
    3. Herzallah MM
    4. Myers CE
    5. Gluck MA

    (2010) A neural model of hippocampal–striatal interactions in associative learning and transfer generalization in various neurological and psychiatric patients

    Brain and Cognition 74:132–144.

    https://doi.org/10.1016/j.bandc.2010.07.013

    • Google Scholar
38. 1. Murty VP
    2. Shermohammed M
    3. Smith DV
    4. Carter RM
    5. Huettel SA
    6. Adcock RA

    (2014) Resting state networks distinguish human ventral tegmental area from substantia nigra

    NeuroImage 100:580–589.

    https://doi.org/10.1016/j.neuroimage.2014.06.047

    • Google Scholar
39. 1. Norman KA
    2. O'Reilly RC

    (2003) Modeling hippocampal and neocortical contributions to recognition memory: a complementary-learning-systems approach

    Psychological Review 110:611–646.

    https://doi.org/10.1037/0033-295X.110.4.611

    • Google Scholar
40. 1. Onat S
    2. Büchel C

    (2015) The neuronal basis of fear generalization in humans

    Nature Neuroscience 18:1811–1818.

    https://doi.org/10.1038/nn.4166

    • Google Scholar
41. 1. Pearce JM
    2. Esber GR
    3. George DN
    4. Haselgrove M

    (2008) The nature of discrimination learning in pigeons

    Learning & Behavior 36:188–199.

    https://doi.org/10.3758/LB.36.3.188

    • Google Scholar
42. 1. Pearce JM

    (1987) A model for stimulus generalization in pavlovian conditioning

    Psychological Review 94:61–73.

    https://doi.org/10.1037/0033-295X.94.1.61

    • Google Scholar
43. 1. Pearce JM

    (1994) Similarity and discrimination: a selective review and a connectionist model

    Psychological Review 101:587–607.

    https://doi.org/10.1037/0033-295X.101.4.587

    • Google Scholar
44. 1. Purtle RB

    (1973) Peak shift: a review

    Psychological Bulletin 80:408–421.

    https://doi.org/10.1037/h0035233

    • Google Scholar
45. 1. Rosenzweig P
    2. Canal M
    3. Patat A
    4. Bergougnan L
    5. Zieleniuk I
    6. Bianchetti G

    (2002) A review of the pharmacokinetics, tolerability and pharmacodynamics of amisulpride in healthy volunteers

    Human Psychopharmacology: Clinical and Experimental 17:1–13.

    https://doi.org/10.1002/hup.320

    • Google Scholar
46. 1. Schechtman E
    2. Laufer O
    3. Paz R

    (2010) Negative valence widens generalization of learning

    Journal of Neuroscience 30:10460–10464.

    https://doi.org/10.1523/JNEUROSCI.2377-10.2010

    • Google Scholar
47. 1. Schultz W

    (2007) Multiple dopamine functions at different time courses

    Annual Review of Neuroscience 30:259–288.

    https://doi.org/10.1146/annurev.neuro.28.061604.135722

    • Google Scholar
48. 1. Shepard R

    (1987) Toward a universal law of generalization for psychological science

    Science 237:1317–1323.

    https://doi.org/10.1126/science.3629243

    • Google Scholar
49. 1. Shohamy D
    2. Adcock RA

    (2010)

    Dopamine and adaptive memory

    Trends in Cognitive Sciences 14:464–472.

    • Google Scholar
50. 1. Shohamy D
    2. Mihalakos P
    3. Chin R
    4. Thomas B
    5. Wagner AD
    6. Tamminga C

    (2010) Learning and generalization in schizophrenia: effects of disease and antipsychotic drug treatment

    Biological Psychiatry 67:926–932.

    https://doi.org/10.1016/j.biopsych.2009.10.025

    • Google Scholar
51. 1. Shohamy D
    2. Wagner AD

    (2008) Integrating memories in the human brain: hippocampal-midbrain encoding of overlapping events

    Neuron 60:378–389.

    https://doi.org/10.1016/j.neuron.2008.09.023

    • Google Scholar
52. 1. Solomon PR
    2. Moore JW

    (1975) Latent inhibition and stimulus generalization of the classically conditioned nictitating membrane response in rabbits (oryctolagus cuniculus) following hippocampal ablation

    Journal of Comparative and Physiological Psychology 89:1192–1203.

    https://doi.org/10.1037/h0077183

    • Google Scholar
53. 1. Spence KW

    (1937) The differential response in animals to stimuli varying within a single dimension

    Psychological Review 44:430–444.

    https://doi.org/10.1037/h0062885

    • Google Scholar
54. 1. Squire LR
    2. Wixted JT

    (2011) The cognitive neuroscience of human memory since H.M

    Annual Review of Neuroscience 34:259–288.

    https://doi.org/10.1146/annurev-neuro-061010-113720

    • Google Scholar
55. 1. Swanson LW

    (1982) The projections of the ventral tegmental area and adjacent regions: a combined fluorescent retrograde tracer and immunofluorescence study in the rat

    Brain Research Bulletin 9:321–353.

    https://doi.org/10.1016/0361-9230(82)90145-9

    • Google Scholar
56. 1. Takahashi H
    2. Kato M
    3. Takano H
    4. Arakawa R
    5. Okumura M
    6. Otsuka T
    7. Kodaka F
    8. Hayashi M
    9. Okubo Y
    10. Ito H
    11. Suhara T

    (2008) Differential contributions of prefrontal and hippocampal dopamine D1 and D2 receptors in human cognitive functions

    Journal of Neuroscience 28:12032–12038.

    https://doi.org/10.1523/JNEUROSCI.3446-08.2008

    • Google Scholar
57. 1. Terrace HS

    (1963) Errorless discrimination learning in the pigeon: effects of chlorpromazine and imipramine

    Science 140:318–319.

    https://doi.org/10.1126/science.140.3564.318

    • Google Scholar
58. 1. Wimmer GE
    2. Daw ND
    3. Shohamy D

    (2012) Generalization of value in reinforcement learning by humans

    European Journal of Neuroscience 35:1092–1104.

    https://doi.org/10.1111/j.1460-9568.2012.08017.x

    • Google Scholar
59. 1. Wimmer GE
    2. Shohamy D

    (2012) Preference by association: how memory mechanisms in the hippocampus bias decisions

    Science 338:270–273.

    https://doi.org/10.1126/science.1223252

    • Google Scholar
60. 1. Wisniewski MG
    2. Church BA
    3. Mercado E

    (2009) Learning-related shifts in generalization gradients for complex sounds

    Learning & Behavior 37:325–335.

    https://doi.org/10.3758/LB.37.4.325

    • Google Scholar
61. 1. Wisniewski MG
    2. Radell ML
    3. Guillette LM
    4. Sturdy CB
    5. Mercado E

    (2012) Predicting shifts in generalization gradients with perceptrons

    Learning & Behavior 40:128–144.

    https://doi.org/10.3758/s13420-011-0050-6

    • Google Scholar

Author details

1. Thorsten Kahnt

   Department of Neurology, Northwestern University Feinberg School of Medicine, Chicago, United States

   Contribution

   TK, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   thorsten.kahnt@northwestern.edu

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-3575-2670

2. Philippe N Tobler

   Department of Economics, Laboratory for Social and Neural Systems Research, University of Zurich, Zürich, Switzerland

   Contribution

   PNT, Conception and design, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

• 3,230

  Page views

• 555

  Downloads

• 35

  Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.