Visual pathways for associative learning
in blindsight monkeys
Takakuwa, Norihiro
）octor of Philosophy
）epartment of Physiological Sciences
School of １ife Science
SO０EN）AI (The Graduate University for
Advanced Studies)
Summary
Visual pathways for associative learning
in blindsight monkeys
Takakuwa, Norihiro
SOKENDAI (The Graduate University for Advanced Studies)
School of Life Science
Department of Physiological Sciences
Introduction
To adapt behaviors in our changing environment, we have to continuously learn associations between various sensory stimuli, actions and subsequent outcomes. This form of learning is called associative learning, and the associative learning makes us predict future outcomes in relation to an ongoing event, in particular the outcome is reward or punishment. It is thought that activity of dopamine (DA) neurons in the substantia nigra pars compacta (SNc) plays a key role in the
associative learning by encoding a reward prediction error (responses to the reward itself (Schultz, et al. 1997; Fiorillo 2013)) or an expectation of future reward value (responses to the sensory predictors (Matsumoto and Hikosaka, 2009; Matsumoto and Hikosaka, 2007; Eshel et al. 2015)). The reward prediction error is that current reward is better or worse than they predicted.
Classical or Pavlovian conditioning is that a neutral sensory event predicts a reward. In pioneering studies by Schultz and colleagues, in Pavlovian conditioning, DA neurons initially respond to the unpredicted reward, but after learning the association between the conditioned stimuli (CS) and reward (UCS), the neurons respond only to the CS but not to UCS. If the CS predicts the reward but no reward is delivered, DA neurons pause at the time of reward delivery (Schultz et al. 1997). In their later study, the magnitude of short latency visual responses to a CS was correlated with an expectation of the reward value, an expectation of larger value reward leads to induction of larger responses (Tobler et al. 2005).
However, it is still unclear which input pathways mediate visual information to encode the reward expectation or reward prediction error signals in DA neurons. Regarding short latency visual CS responses, Redgrave and colleagues reported the existence of direct projection from the superior colliculus (SC) to the SNc in rodents, cats and monkeys (Dommett et al. 2005; Comoli et al. 2003). Visual pathways from the retina to SNc would be roughly divided into two pathways; one is the cortical pathway via the lateral geniculate nucleus (LGN) and the primary visual cortex (V1), and
To investigate whether the subcortical visual pathway could mediate the afferent visual cue signal and is involved in associative learning or not, I used monkeys with unilateral V1 lesions. After damage to the V1, visual awareness is impaired in the lesion-affected visual field. However, it is known that visual stimulus presented in the affected visual field can trigger the behaviors which identify the location of the stimulus. The phenomenon is called “blindsight” and it has been shown that the monkeys with V1 lesion mimic human blindsight (Cowey and Stoerig 1995; Yoshida and Isa 2015).
Materials and Methods
Three adult Japanese monkeys (Macaca fuscata; all female, body weight 5-7 kg, Monkey K, Monkey U and Monkey T) with unilateral V1 lesions were used in this study. All experimental procedures were performed in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals and approved by the Committee for Animal Experiment at the National Institute of Natural Sciences.
After a 0.7 to 1.2 s fixation period, conditioned stimuli (CS) were presented during the fixation. Two kinds of CSs were used; one is a large reward conditioning stimulus (LR-CS) whereas the other was a small reward conditioning stimulus (SR-CS). These CSs were presented within a period of 1.0 s (monkey T) or 1.7 s (monkeys K and U). These CSs were discriminated by their presented positions, one CS (upper visual field) predicted an immediate and large reward (0.17 ml juice) during the CS-presentation period (0.7 s (monkey T) or 1.3 s (monkey K and U) after the CS onset), while the other (CS in the lower visual field) predicted a small reward (0.06 ml juice), 1.5 s after the CS offset.
To investigate whether the visual information mediated by the SC was used for expressing
conditioned responses, muscimol was injected into the ipsi-lesional SC of monkey K and monkey T. Before the injection, neural activity of the SC was electrophysiologically recorded, and the location of the LR-CS on the SC map was identified. Muscimol was injected into the local area of the SC with the receptive field around the position of LR-CS. The effect of muscimol was confirmed by failure in a visually guided saccadic eye movement task to the location of LR-CS.
To investigate whether DA neurons responds to the visual CS presented to lesion-affected visual field, I recorded the activity of DA neurons during the Pavlovian conditioning task. Monkey K and monkey T were used for the recording. I recorded the activity of neurons which responded to an unpredicted reward in and around the SNc, the location of which was confirmed in advance by MR images. DA neurons have characteristic features in terms of their spike activity; a low background
background firing rate was between 1 and 10 Hz, and its spike width (between the first negative peak and next positive peak) was broader than 0.45 ms.
To test whether the visual information driving DA neurons is mediated by the SC, muscimol was injected into the ipsi-lesional SC during recording the activity of DA neurons in the Pavlovian conditioning task. The procedure was the same as that in the experiments to examine the effects of muscimol injection on the performance of Pavlovian conditioning. After the control data on visually guided saccadic task and on Pavlovian conditioning task was collected, muscimol injection was started. The activity was continuously recorded from the same DA neuron while the Pavlovian conditioning task was performed before and after inactivation of the SC with muscimol.
Results
In this study, I first examined whether the “blindsight monkeys” could learn the association
between visual cues and subsequent reward in the Pavlovian conditioning paradigm. After several days of training, monkeys started anticipatory licking triggered by the visual CSs. These behavioral responses were different between LR trials and SR trials, which suggested that the monkeys
predicted the reward value indicated by each CS. These results indicated that surprisingly, the blindsight monkeys could learn the association between the CS presented to their affected visual field and reward. Next, I tested whether the visual signals mediated by the SC is critical for the associative learning. For this purpose, I performed pharmacological reversible blockade of the SC by microinjection of muscimol, a GABAA receptor agonist into the SC. The anticipatory responses were disappeared during SC inactivation. All results of these experiments suggested that the visual information through the SC is critical for associative learning in blindsight monkeys.
As the next step, to examine whether short latency responses of DA neurons to the visual CS still remains after V1 lesion, and whether the signal carries the information about the reward expectation, I conducted single unit recordings from DA neurons in the blindsight monkeys performing the Pavlovian conditioning task. The electroohysiological recording of DA neurons revealed that short- latency phasic responses were evoked by CSs prsented to the lesion affected visual field. These responses reflected differential reward values predicted by CSs in LR and SR trials. Furthermore, I tested the reversible blockade of the SC with muscimol to demonstrate whether the reward
predicting CS responses in the DA neurons were mediated by the SC. DA neural responses elicited by CS presentation to the lesion affected field were suppressed when the SC was inactivated by a local injection of muscimol.
These results demonstrate that conditioned visual cues can support Pavlovian conditioning in the absence of V1 cortex. Moreover, that phasic DA responses representing different CS values can be mediated via subcortical pathway throgh the SC. Results of all these experiments will update the
current understanding of the neural mechanism for emergence of reward predicting signals in DA neurons.
Discussion
Many studies indicated that midbrain DA neurons causally contribute to learning. The involvement of dopamine or DA neurons in learning has been well studied, however, it is still unclear how the value component of DA neurons is calculated. As I suggested in the previous section, it is important to investigate brain regions receiving the visual information from the SC.
Input pathways to DA neurons have been well studied, and Uchida and colleagues recently encompassed the brain regions which project to DA neurons. DA neurons mainly receive reward information from the striatum, amygdala, subthalamic nucleus, pedunculopontine nucleus, rostromedial reticular nucleus, and GABAergic neurons of pars reticulata of the substantia nigra (Watanabe-Uchida et al. 2012). On the other hands, in this study, the results suggested that the visual information via SC induce the response of DA neurons reflecting the predicted reward value at short latency.
There are some possible visual input pathways from the SC to DA neurons. The SC directory projects to subthalamic nucleus (Comoli et al. 2003) which was reported to mediate reward information (Espinosa-Parrilla et al. 2015). Because the subthalamic nucleus projects to DA neurons in the SNc, this pathway (SC-STN-SNc) could be responsible for the short-latency phasic activation of DA neurons reflecting reward value. On the other hand, the direct pathway from the SC to the DA neurons was also reported (Dommett et al. 2005). It is known that visual responses in the SC are also modulated by reward predicting visual stimulus (Ikeda et al. 2003), suggesting that this direct pathway could mediate short-latency phasic responses with reward value information in addition to the responses to salient visual events.
