We present a neural network model of cognition in medicated and unmedicated patients with Parkinson’s disease (PD) in various learning and memory tasks. The model extends our prior models of the basal ganglia and PD with further modeling of the role of prefrontal cortex (PFC) dopamine in stimulus–response learning, reversal, and working memory. In our model, PD is associated with decreased dopamine levels in the basal ganglia and PFC, whereas dopamine medications increase dopamine levels in both brain structures. Simulation results suggest that dopamine medications impair stimulus–response learning in agreement with experimental data (Breitenstein et al., 2006 and Gotham et al., 1988). We show how decreased dopamine levels in the PFC in unmedicated PD patients are associated with impaired working memory performance, as seen experimentally (Costa et al., 2003, Lange et al., 1992, Moustafa, Sherman et al., 2008 and Owen et al., 1995). Further, our model simulations illustrate how increases in tonic dopamine levels in the PFC due to dopamine medications will enhance working memory, in accord with previous modeling and experimental results (Cohen et al., 2002, Durstewitz et al., 2000 and Wang et al., 2004). The model is also consistent with data reported in Cools, Barker, Sahakian, and Robbins (2001), who showed that dopamine medications impair reversal learning. In addition, our model shows that extended training of the reversal phase leads to enhanced reversal performance in medicated PD patients, which is a new, and as yet untested, prediction of the model. Overall, our model provides a unified account for performance in various behavioral tasks using common computational principles.
Parkinson’s disease (PD) is a neurodegenerative disorder associated with reduced dopamine levels in the basal ganglia, particularly the dorsal striatum (Kish et al., 1988 and Rinne et al., 2000). In addition to motor dysfunction, PD patients show impairment performing various cognitive tasks such as planning (Dagher et al., 1999 and Owen et al., 1998) and cognitive set shifting (Hayes, Davidson, Keele, & Rafal, 1998). PD patients also show impairment performing various working memory tasks, including delayed-response tasks (Partiot et al., 1996), the Wisconsin Card Sorting Task (Amos, 2000, Cooper et al., 1991, Lees and Smith, 1983, Owen et al., 1993 and Pickett et al., 1998), object and spatial span tasks (Gabrieli, Singh, Stebbins, & Goetz, 1996), as well as other working memory tasks (Lewis et al., 2003).
In reversal learning, subjects initially learn to associate different stimuli with different responses (stimulus–response learning), and subsequently learn to associate the same stimuli with the opposite responses (i.e., reversal). Experimental studies show that dopamine agonists, such as pergolide and bromocriptine, impair reversal learning in monkeys, PD patients, and healthy subjects (Cools et al., 2001, Jentsch et al., 2002 and Swainson et al., 2000). Cools et al. (2001) found that medicated PD patients on dopamine agonists are more impaired at reversal learning than unmedicated patients (also see Swainson et al., 2000). Jentsch et al. (2002) found that the administration of cocaine (dopamine reuptake inhibitor) to monkeys lead to impairment in reversal learning. Similar results were found with administering quinpirole (dopamine agonist) to rats (Boulougouris, Castane, & Robbins, 2009). It is hypothesized that dopamine medications might overdose the PFC and thus impair performance in reversal tasks (Cools et al., 2001). In line with this hypothesis, we show how simulating this dopamine ‘overdosing’ of the PFC due to the administration of dopaminergic medications impairs reversal performance in our model (see Experimental Procedures section for more details).
Dopamine medications (both precursors and agonists) are used to treat motor symptoms of PD (tremor, rigidity, and bradykinesia), but can either enhance or impair cognitive function (Cools et al., 2001, Feigin et al., 2003, Frank et al., 2004 and Swainson et al., 2000). For example, various studies show that dopamine medications impair stimulus–response learning in both PD patients (Gotham, Brown, & Marsden, 1988; Jahanshahi, Wilkinson, Gahir, Dharmaindra, & Lagnado, 2010) and healthy subjects (Breitenstein et al., 2006 and Pizzagalli et al., 2007). In stimulus–response learning tasks, subjects learn to associate the presentation of different stimuli with different responses based on corrective feedback. Unlike stimulus–response learning, many studies found that dopamine medications enhance working memory performance in PD patients as well as in Parkinsonian animal models (Costa et al., 2003, Lange et al., 1992 and Lewis et al., 2005; Owen, Sahakian, Hodges, Summers, & Polkey, 1995). It was also found that dopamine agonists enhance working memory performance in healthy subjects (Mehta, Swainson, Ogilvie, Sahakian, & Robbins, 2001).
The model we present here builds on our earlier models (Moustafa and Gluck, 2011 and Moustafa and Maida, 2007), and collectively addresses how PD and dopamine medications affect performance in stimulus–response learning, reversal, and working memory tasks. This and similar adaptive network, or “connectionist” theories of human learning are reminiscent of statistical learning theories, the most influential of which is Stimulus Sampling Theory, developed by the late W. K. Estes and colleagues ( Estes, 1961). Building on Estes’ work, we are able to extend “connectionist” theories to account for broader conception of associations among representation of events, thereby addressing the shortcomings of earlier approaches in this domain.
