國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Chapter Four: Discussion
In this present study, by the establishment of a probabilistic risky choice model in the rat, the neural substrates of risk-based decision making were systemically investigated by the use of excitotoxic lesion technique.
The results of Experiment 1a show that the manipulations of different EV’s set in CLR and PHR arm significantly affected the probabilistic risky choice made by the rat. When two chosen options had the same EV = 1 set for both CLR and PHR arm, the rat exhibited different patterns of choice behavior following different reward probabilities and magnitudes under each reward ratio. In addition, given in options with different EV’s (EV 0.5 vs. EV 1 or EV 2 vs. EV 1), the subject apparently chose the option which had higher EV. This finding implies that the rat may have a
“sense” to process basic EV in this probabilistic risky choice behavior.
The effects of amphetamine tested in Experiment 1b, showed a drug induced relatively risk-seeking choice behavior when the condition of EV set in both CLR and PHR arms at 1. These results provide an evidence to support the role of dopamine system is involved in the present behavioral task of probabilistic risky choice.
In Experiment 2, the results show that the lesion of NAC produced a relatively risk-averse choice behavioral effect. By contrast, the lesion of DLS as an anatomical control, had no such an effect on probabilistic risky choice behavior. These findings indicate the heterogeneity of behavioral function existed between the NAC and DLS on probabilistic risky choice. Regarding to the subarea of prefrontal cortex as manipulated, the OFC lesion produced a tendency of relatively risk-averse choice behavior by the rat showing a marginal significant decrease on the percentage of choosing PHR in the last 3 days of free choice phase. By contrast, lesion of mPFC did not alter the probabilistic risky choice behavior in this study.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Manipulations of different EV conditions on probabilistic risky choice behavior In the condition of EV=1 set in both CLR and PHR arm, the rat showed different preferences of choices over three different reward ratios. The rat preferred to choose PHR arm in the reward ratio of 1:2 but preferred to choose CLR arm in the reward ratio 1:4 and 1:8. In terms of the risk, the reward ratio 1:2 leads a relatively lower risky condition whereas the reward ratio 1:8 generates a higher risky condition (Tobler, O’Doherty, Dolan, & Schultz, 2007; Tobler et al., 2009). The results indicate that the subject may respond to the different degrees of risk set on the basis of the
probabilities of the reward presence. The rat performed in a relatively risk-seeking fashion in the lower risky condition, but became responding toward a more relatively risk-averse manner in the higher risky condition on the present task. Thus, this study demonstrates that the probabilistic risky choice behavior made by the rat can be risk dependent.
Further, to our knowledge, this is the first study emphasizing EV given in animal model of probabilistic risky choice behavior. The results suggested that the rat would choose the option which had relatively higher EV if the EV’s were set in different among the chosen options. For instance, in the condition of EV=0.5 set in PHR arm and EV=1 set in CLR arm, the rat significantly preferred to choose CLR arm. Conversely, in the condition of EV=2 set in PHR arm and EV=1 set in CLR arm, the rat significantly preferred to choose PHR arm. These results indicate that the rat may have a cognition-like function to process the EV.
However, whether the rat actually had an internal representation of EV is still a controversial issue. In this present study, a between-subject design was used on the EV factor. The comparisons of internal representation of different EV within a rat are then limited, because of the rat only experienced one of three EV’s conditions.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Thus, further investigations by a within-subject design may be helpful to verify the notion of rats’ internal EV representation.
As mentioned earlier, Tversky and Kahneman (1979) suggested that people show risk-averse attitude while in a monetary gain condition. In the present study, the rat chose the option of 2-pellet with 50% rather than 1-pellet with 100%, this result implied a relatively risk-seeking choice behavior in reward ratio 1:2 (see Figure 1).
Although there are several discrepancies in methodology between animal and human studies, the aforementioned result of the present study is somewhat intriguing. In this specific condition of reward ratio set on 1:2, but not on1:4 or 1:8, the subject may still prefer to take a relative lower risk for obtaining the reward that can reduce its hunger drive. The motivational state for the present animal subject is different from that of the human subject in the test of a monetary gain condition with probabilistic risk. It should also be noted that the risk-averse response was appeared in the reward condition of 1:4 and 1:8 in the present study. In human study, in contrast to the monetary gain condition, risk-averse response appears in the monetary loss condition.
Together, it may be an interesting issue to be further investigated with a more sophisticated experimental design.
Linear functions for EV representation
According to Cardinal and Howes (2005), the rat’s internal representation of probabilities may be possibly evaluated by “indifference probability.” By using a linear regression function, Cardinal and Howes (2005) suggested that the core of NAC-lesioned rats had higher indifference probabilities, the value of 1-pellet is equal to 4-pellet with 70%. Namely, the rat would not choose large/uncertain lever when the probability of obtained 4-pellet was below 70%. Thus, the estimated value was used to infer the rat with NAC lesion showing a relatively risk-averse choice behavior as compared to sham lesion group.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
In this present study, there will be three distinct indifference probabilities if we apply this linear function into three reward ratios. For example, in the reward ratio 1:2, there will be an indifference probability for the rat by regress the 2-pellet with probabilities of 25%, 50%, and 100%. Also, another two indifference probability values would be calculated for reward ratio of 1:4 and 1:8. In addition to reward ratio, the present study manipulated probability to adjust the EV. Thus, it can be a more complex linear function for representing the EV. In which, both reward
magnitude and probability would be necessarily included within the regression model.
Nevertheless, in considering the internal representation of EV, it may be possibly assessed in current data. While EV=1 was equally set in both CLR and PHR arm, the rat chose PHR arm significantly more than CLR arm in reward ratio 1:2. This result implies that the rat’s internal representation of EV toward 2-pellet with 50%
probability is higher than 1-pellet. In contrast, the internal EV toward 4-pellet with 25% probability and 8-pellet with 12.5% probability is lower than 1. Based on this implication, further study using linear regression model will be helpful for
investigating the rat’s internal representation of EV.
The roles of reward probability and reward magnitude within EV
EV is defined as a summation of the probability to obtain each of reward multiple with the reward magnitude as received. It is interesting to figure out which of the components may play the major role on the probabilistic risky choice behavior. In most of previous studies that investigated probabilistic risky choice task, the reward magnitude was kept in constant but changing the probabilities with sessions for the test (Mobini et al., 2002; Cardinal & Howes, 2005; St. Onge & Floresco, 2008, 2009).
These results indicated that the rat decreased choices of large but risky lever with the reward probabilities decreased (as the risk increased). In contrast, the results from the tests set by keeping the reward probabilities in constant but changing the reward
‧
magnitudes revealed that the rat chose the to respond for the option containing the larger reward (Cardinal & Howes, 2005; Zeeb, Robbins & Winstanley, 2009). A separate statistical analysis was conducted to clarify this issue for the current data.
As shown in the following table, three thicken frames represent the experimental conditions are characterized by three different reward magnitudes presented in the same reward probability (25%) set in the PHR arm.
Reward
Figure 26 shows the results of this assigned condition. As indicated by a
significant main effect on the reward magnitude from a two-way ANOVA (F(2, 15) = 5.691, p < 0.05), the percentage of choosing PHR was higher given in 8-pellet reward condition than that given in 2-pellet reward.
--- Figure 26
---
Together, with all of these results from the present studies and the others, it is still difficult to declare which of the two components within EV plays the key role in this kind of risk-based decision making. Because of both factors of the probability to
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
obtain reward and the reward magnitude as received are crucial for making a risky choice. Thus, the rat chooses among different EV options are likely based on the integration of both probabilities and magnitudes of the presented reward (Zeeb et al., 2009). This argument is in agreement with the notion that EV is a key factor as indicated by human fMRI studies showing the subject’s making choices under uncertainty (Schultz et al. 2008; Tobler et al. 2009).
The risk of “get nothing” versus the risk of “punishment”
The behavioral mechanisms to elucidate the risky choice may be more complex than what being thought with a certain experimental design used in the animal study (e.g. Cardinal and Howes, 2005). In present study, the risk could be regarded as “get nothing” rather than “loss something.” Namely, only the “gain” domain was
considered in the present task. However, this type of operational definition may not be the only way to delineate the risk. There were several other studies examining the risk by simultaneously manipulating both the “gain” and the “loss” in the design of animal behavioral tasks. As mentioned earlier, in a rodent model of IGT (van den Bos et al., 2006), the loss were represented for the animal by a bitter-tasted quinine-treated pellet. Another behavioral task developed by Simon, Gibert, Mayse, Bizon, and Setlow (2009) was set up to assess a larger reward but with the risk of punishment with footshock on the probabilistic discounting task. The results indicated that the rat, in a condition with a higher intensity of shock, chose to press more on the small reward and safe lever than the large reward but risky lever. Also, with the
probabilities of the punishment increased (ascending from 0%, 25%, 50%, 75%, to 100%), the rat decreased their preference toward large reward lever. From these studies, it is suggested that the concept of “gain and loss” can be assessed in animal models. And, the risk to obtain a punishment can be addressed as a “loss” and is indeed with the impact to influence the subject’s choice. Despite this highlight, there
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
is a further concern in terms of EV to this issue. Namely, it would then be difficult to measure the values of punishments contrast to the values of reward. For example, as suggested by van den Bos et al. (2006), the starving rat still ate the quinine-treated pellet. Also, in Simon et al. (2009), the rat showed a wide individual variation toward footshocks.
Regardless whether only the gain domain or both the gain and loss being manipulated, a similarity of behavioral choice pattern is found in between the aforementioned studies and the present study using a probabilistic risky choice without presenting any punishment for the “loss.” The rat shifted their preferences from choosing large reward toward small reward as the probabilities of obtaining large reward decreased or the probabilities of punishment increased in the large reward site. Thus, “get nothing” of the present task is presumably regard as a kind punishment for the subject, which is true if it is starved under a food-deprived condition.
Effects of amphetamine on probabilistic risky choice behavior
The effects of amphetamine on probabilistic risky choice behavior were
systemically assessed in Experiment 1b. In the conditions of EV=0.5 or EV=2 set in PHR arm, the amphetamine did not affect any behavioral responding on probabilistic risky choice. One way to explain the negative results of amphetamine treatment may be attributed to the “floor effect” and “ceiling effect” derived from the EV set in PHR arm. Namely, the risk perception in either condition is too rigid to influence
behavior response on this task. In the condition of EV=1 equally set in both CLR and PHR arm, the results indicate that the rat injected with the high dose of
amphetamine treatment (1.0 mg/kg) showed a relatively risk-seeking fashion of choice behavior. That is, psychostimulant drug alter the choice behavior showing relatively more risk-seeking compared to saline control group.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
From those reported in a previous study by St. Onge and Floresco (2008),
amphetamine induced increases in probabilistic risky choice were significantly higher than saline control group on the reward probabilities of 25% and 12.5%. Thus, the rat with amphetamine treatment showed relatively risk-seeking only when the risk was relatively higher. In this present study, we found a similar tendency on reward ratio 1:8 which is a relatively higher risky condition (the top panel of Figure 8).
However, based on our results, this notion is limited due to the dose-by-reward-ratio interaction effect was not significantly confirmed. In this regard, if a further reward ratio 1:16 or a higher dose of amphetamine could be extendedly tested, it would then be able to verify the effects of amphetamine affected risky choice behavior only on higher risky condition.
In addition, a chronic effect of repeated amphetamine treatment on probabilistic risky choice behavior had increased risky choice after drug exposure, whereas these treatments did not affect effort-based decision making (Floresco & Whelan, 2009).
The effects of amphetamine induced relatively risk-seeking on this probabilistic risky choice may attribute to the drug effects on the mesolimbic dopamine systems because amphetamine is a general dopamine agonist in terms of pharmacology (St.
Onge & Floresco, 2008). Another supportive evidence to this notion is that cocaine, as one of the psychostimulants but known as a serotonin transporter blocker, had no effects on this kind of behavior (Simon et al., 2009). Moreover, the different dopamine subtype receptors were involved in the probabilistic risky choice behavior (St. Onge and Floresco, 2008) and in a five-choice serial reaction time task (Zeeb et al., 2009; Winstanley, Zeeb, Bedard, Fu, Lai, Steele, Wong, 2010) in distinctive manners. In addition to dopamine subtype receptors, the role of dopamine transporter (DAT) is also critical on the risky choice behavior. In a series of investigations with DAT associated treatments (Adriani, Boyer, Gioiosa, Macri,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Dreyer, & Laviola, 2009; Adriani, Boyer, Leo, Canese, Podo, Perrone-Capano, Dreyer,
& Laviola, 2010), the effects of DAT over-expression increased the choices of
“large/luck-linked” reward rather than “small/sure” reward. Taken together, the midbrain dopamine system is highly involved in the probabilistic risky choice behavior. That the dopamine pre- and post-synaptic mechanisms potentially dissociable on risk-based decision making deserves for further investigations.
The lesion effects of striatal subareas on probabilistic risky choice behavior
In this study, one of the major findings is the lesion of the NAC, but not DLS, affect probabilistic risky choice behavior. As compared to the sham lesion control, the rat with excitotoxic lesion in the NAC became a relatively risk-averse to respond on this probabilistic risky choice task, namely choosing less PHR even in the lower risk condition. But such a behavioral alteration was not found in the subjects with DLS lesion. These results further suggested that the subareas of striatum contain heterogeneity to mediate behavioral functions.
From previous studies, the effects of DLS lesion are shown to impair the habit formation (Yin, Knowlton, & Balleine, 2004) and reduce the resistance to extinction (Castane, Theobald, & Robbins, 2010). Also, the DLS has been argued to be crucial for stimulus-response (S-R) learning (Horvitz, 2009; Anselme, 2010). To the best of our knowledge, the present study is the first to investigate the role of DLS on
probabilistic risky choice behavior. The current results show that the lesion of DLS had no effect on this choice task. This negative result cannot be attributed to those potential side effects derived by the excitotoxic lesion applied in the DLS. Since there were no differences between the DLS lesion group and the sham lesion control group on locomotor activity or discrimination test. However, with a further analysis, the average response time of completing a trial on the risky choice was longer in the DLS lesion group (3.41 ± 0.11 s) than sham lesion group (2.60 ± 0.04 s). The
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
increase of response time on DLS lesion group may be due to the impairment of motivation (Anselme, 2010), suggesting that the rat could be less motivated to obtain the reward. Another explanation for the lesion of DLS could be the impaired S-R association which leads to behavioral outcome with the increase of the response time (Horvitz, 2009). Despite these arguable confounding effects, it should be noted that the lesion of DLS did not significantly affect the probabilistic risky choice behavior in this present study.
As for the NAC-lesioned induced relatively risk-averse choice behavior, it may not due to the impairment of motor function because a hyperactivity reaction to an open field was revealed by a post-lesion locomotor activity test for the subjects. The average response time of completing a trial was no difference between NAC lesion group (3.00 ± 0.08 s) and sham lesion group (2.90 ± 0.06 s). In addition, there were no omission trials appeared in the probabilistic risky choice tasks. Further, the behavior alteration in the NAC-lesion subjects cannot be attributed to impairment of basic discriminate function. A discrimination test was conducted and the data revealed no differences between NAC lesion group and sham lesion group. The rat with NAC lesion still had the ability to distinguish reward magnitude from 1 pellet and 2 pellets. Whether the lesion effect of NAC or DLS may impair the learning ability to affect the probabilistic risky choice can be an issue to concern. The results of a three-way ANOVA revealed a significant main effect of day and a significant reward-ratio-by-day interaction on both of NAC and DLS analyses (Figure 14 and Figure 18). It is then indicate that the rat with lesion of NAC or DLS did learn by showing dramatic changes on choice in each of reward ratios across ten daily sessions.
Comparing to previous study, Cardinal and Howes (2005) reported that the rat with NAC core lesion showed relatively risk-averse on probabilistic discounting task
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
in the stable choice sessions of post-surgery. Consistent with Cardinal and Howes (2005), as revealed by the data presented in the last 3 days of free choice test, the rat with NAC lesion showed relatively risk-averse choice behavior.
Taken together, these data indicate that the NAC, but not DLS, is highly involved in behavioral performance on the probabilistic-based risky choice.
The lesion effects of prefrontal cortex subareas on probabilistic risky choice behavior The data regarding to the effects of OFC lesion on probabilistic risky choice behavior yielded only a significant main effect of lesion from ANOVA, indicating that the lesion groups decreased the percentage of choosing PHR compared to that of sham lesion group. As an anatomical control, mPFC lesion produced no difference
The lesion effects of prefrontal cortex subareas on probabilistic risky choice behavior The data regarding to the effects of OFC lesion on probabilistic risky choice behavior yielded only a significant main effect of lesion from ANOVA, indicating that the lesion groups decreased the percentage of choosing PHR compared to that of sham lesion group. As an anatomical control, mPFC lesion produced no difference