Decision-Making, Antisaccade Models of
Basic Antisaccade Behavior
The antisaccade task was first developed to dissociate the stimulus location from the goal of the saccade. During a single trial of the antisaccade task two processes take place: (1) suppression of an erroneous prosaccade toward the peripheral stimulus and (2) generation of a volitional saccade to a position in the opposite direction (antisaccade) (Munoz and Everling 2004). In a single trial, a participant may express any of the following three oculomotor behaviors: (1) the subject makes an antisaccade (an eye movement in the opposite direction of the peripheral stimulus), or (2) the subject makes an erroneous prosaccade (an eye movement in the direction of the peripheral stimulus), or (3) the subject makes an erroneous prosaccade followed by a corrected antisaccade. An error is an eye movement toward the peripheral stimulus instead of the opposite direction. Antisaccade performance involves different metrics such as the mean and standard deviation of the saccade reaction time (SRT) of each eye movement as well as the error rate (Ettinger et al. 2003). Healthy participants typically fail to suppress erroneous prosaccades toward the target on about 20–25% of trials, before correctly saccading toward a location in the opposite direction (Fischer and Weber 1992; Everling and Fischer 1998; Smyrnis et al. 2002; Ettinger et al. 2003; Tatler and Hutton 2007). Unimodal skewed to the right distributions of antisaccades, erroneous prosaccades, and corrected antisaccades are observed. The mean and standard deviation of the antisaccade reaction time from a large cohort of 2006 healthy subjects is reported to be 270 ms and 56 ms, respectively (Evdokimidis et al. 2002). In the same group, the mean and standard deviation of the erroneous prosaccade reaction time is 208 ms and 46 ms, respectively, whereas the mean and standard deviation of the corrected antisaccade reaction time is 146 ms and 85 ms, respectively (Evdokimidis et al. 2002).
Antisaccade Oculomotor Circuit
The antisaccade oculomotor circuit consists of several cortical and subcortical areas including the frontal eye fields (FEF), supplementary eye fields (SEF), dorsolateral prefrontal cortex (DLPFC), anterior cingulate cortex (ACC), lateral intraparietal area (LIP) (parietal eye field (PEF) in the human), basal ganglia, thalamus, superior colliculus (SC), brainstem reticular formation, and cerebellum (Munoz and Everling 2004). Visual information is processed via the retino-geniculo-cortical pathway to primary visual cortex and from there to LIP/PEF and via the retinotectal pathway to the superficial layers of the SC (SCs). LIP is an area in the parietal cortex coding for space. LIP/PEF then projects to both the intermediate layers of the SC (SCi) and frontal cortical oculomotor areas including FEF, SEF, ACC, and DLPFC. FEF is critical for voluntary saccade execution. SEF is implicated in internally guided decision-making and sequencing of saccades. AAC plays a role in conflict resolution and error monitoring. DLPFC is critical in executive function, spatial working memory, and suppressing automated or reflexive responses. These frontal oculomotor areas project to SCi, an area important for decision-making, which then projects to the reticular formation of the brainstem to provide the necessary input to guide saccades.
Another pathway to SCi from the frontal oculomotor areas is through the basal ganglia structures via the direct, indirect, and hyperdirect pathways (Munoz and Everling 2004). Via the direct pathway, frontal areas project to the caudate nucleus (CD), which in turn inhibits the substantia nigra pars reticulata (SNr). SNr disinhibits the SCi and motor nuclei of thalamus, which project back to frontal cortex. Via the indirect pathway, CD projects to the external segment of the globus pallidus (GPe), which in turn projects to the subthalamic nucleus (STN). STN sends excitatory projections to SNr and GPe, which projects via GABAergic connections back to SNr. Via the hyperdirect pathway, cortical regions excite STN, which in turn excite SNr. These complex sets of excitatory and inhibitory projections within the basal ganglia provide a rich set of control mechanisms to help guide voluntary behavior (Coe and Munoz 2017).
Primate Neurophysiology Overview
A number of brain areas in monkeys are involved in the antisaccade task, including the DLPFC (Funahashi et al. 1993), the ACC (Phillips et al. 2010), the LIP (Gottlieb and Goldberg 1999), the SEF (Schlag-Rey et al. 1997; Amador et al. 2004), the FEF (Everling and Munoz 2000), and the SC (Everling et al. 1999). Understanding how neurons in these brain areas participate in the suppression of automatic responses and the generation of antisaccades is crucial for explaining the antisaccade behavior.
Single neuron recordings in FEF and SC have revealed the existence of two distinct and reciprocally activated populations of neurons: the fixation cells and the saccade cells. Fixation cells are tonically active during visual fixation, and their activity ceases when a saccade is executed. On the other hand, saccade cells are silent during fixation but discharge when the animal is making a saccade. In SC, two distinct types of saccade neurons have been recorded: buildup and burst cells (Munoz and Wurtz 1995a, b). A network of inhibitory interneurons is thought to control the reciprocal activation of fixation and saccade neurons (Munoz and Istvan 1998). During fixation in the gap condition (gap prosaccade versus gap antisaccade), fixation activity is greater in the antisaccade trials than in the prosaccade trials in both FEF and SC. This pattern of enhanced fixation activity explains the so-called anti-effect (Munoz and Everling 2004): longer reaction times on antisaccade trials than on prosaccade trials. A few milliseconds into the gap period, there is a drop in activity of fixation neurons and a slow buildup of low-frequency activity of a subset of saccade neurons in both SC (Everling et al. 1999) and FEF (Everling and Munoz 2000). The appearance of the visual stimulus in the right visual field leads to phasic activation of visually responsive saccade neurons in FEF and SC on the contralateral (left) side of the brain, and to phasic inhibition of saccade neurons on the ipsilateral (right) side of the brain. On the prosaccade trials, saccade neurons on the left side discharge a saccadic burst for the rightward prosaccade immediately after the visual phasic response. On antisaccade trials, the saccade neurons in the left FEF and SC are inhibited compared to saccade neurons in the right FEF and SC, which are active to drive the leftward antisaccade.
Recordings from neurons in PFC in rhesus monkeys trained to perform a delayed antisaccade task (Funahashi et al. 1993) revealed that most PFC neurons code the location of the visual stimulus in working memory, and this memory can be engaged to suppress and prescribe a response. Response-coding neurons in smaller percentages were also found, some of which increased their firing rate for the direction of the saccade (e.g., rightward saccade) irrespective of the stimulus location, and others coding for both stimulus location and saccade direction. Similarly, recordings in LIP (Gottlieb and Goldberg 1999) also revealed the majority of LIP neurons reliably coding for the encoded cue location, with only a very small minority encoding for the direction of the upcoming saccade.
Electrical microstimulation in dorsal AAC in monkeys performing alternating blocks of prosaccade and antisaccade trials (Philips et al. 2010) suggested a direct role of AAC in antisaccade performance. On antisaccade trials, microstimulation decreased SRTs for both ipsilateral- and contralateral-directed antisaccades. On the other hand on prosaccade trials, SRTs were increased for saccades contralateral to and decreased for saccades ipsilateral to the stimulated hemisphere.
Finally, recordings from SEF in monkeys (Amador et al. 2004; Schlag-Rey et al. 1997) showed that the vast majority of SEF movement neurons fired significantly more before antisaccades than before prosaccades. The level of their firing was predictive of the correct performance on antisaccades on individual trials.
Antisaccade Performance Across Lifespan
Antisaccade performance varies systematically with age (Munoz et al. 1998; Klein and Foerster 2001; Peltsch et al. 2011). Young children (5–8 years old) have slow SRTs, large intra-subject SRT variance, and the largest error rate in the anti-saccade task. Young adults (20–30 years of age) typically have the fastest SRTs, the lowest intra-subject variance in SRT, and the fewest direction errors. Elderly subjects (60–85 years of age) have slower SRTs than other subject groups. These results demonstrate very strong positive correlation of age and subject antisaccade performance, which may reflect different stages of normal development and degeneration in the nervous system. The dramatic improvement in antisaccade performance observed from the ages of 5 to 15 years is attributed to delayed maturation of the frontal lobes.
Antisaccade Performance in Disorders
Antisaccade performance has been investigated in many neurological and psychiatric disorders including attention deficit hyperactivity disorder (Munoz et al. 2003; Hakvoort Schwerdtfeger et al. 2013), fetal alcohol spectrum disorders (Green et al. 2007; Paolozza et al. 2013), Huntington disease (Peltsch et al. 2008), Parkinson’s disease (Chan et al. 2005; Cameron et al. 2012; Cameron et al. 2010; Amador et al. 2006; Antoniades et al. 2015), Alzheimer’s disease (Peltsch et al. 2014; Kaufman et al. 2012), mild cognitive impairment (Peltsch et al. 2014; Heuer et al. 2013), amyotrophic lateral sclerosis (Witiuk et al. 2014), bipolar disease (Soncin et al. 2016), schizophrenia (Levy et al. 2004; Zanelli et al. 2005; Theleritis et al. 2014), obsessive-compulsive disorder (Lennertz et al. 2012; Jahanshahi and Rothwell 2017), Tourette syndrome (Jahanshahi and Rothwell 2017), multiple sclerosis (Clough et al. 2015), depression (Carvalho et al. 2014; Malsert et al. 2012), epilepsy (Lunn et al. 2016), ventrolateral prefrontal damage (Hodgson et al. 2007), and frontotemporal dementia (Burrell et al. 2012; Boxer et al. 2012).
In particular, patients with frontal lobe lesions (Guitton et al. 1985; Pierrot-Deseilligny et al. 2002) and patients suffering from schizophrenia (Fukushima et al. 1988) make more antisaccade errors and their antisaccade latencies are more variable within and across subjects (Fukushima et al. 1988; Hutton et al. 1998; Karoumi et al. 1998; Brownstein et al. 2003). An increase in correct antisaccade mean latency in schizophrenia patients was recently reported (Damilou et al. 2016). Impaired antisaccade task performance has also been reported in patients with recent onset schizophrenia and first-episode schizophrenia (deWilde et al. 2008; Ettinger et al. 2004; Grootens et al. 2008; Hutton et al. 2002, 1998; Kirenskaya et al. 2013), chronic schizophrenia (Boudet et al. 2005; Curtis et al. 2001a; Fukushima et al. 1988; Behrwind et al. 2011), and remitted schizophrenia (Curtis et al. 2001b). Aberrant antisaccade performance has also been reported by first degree unaffected biological relatives of schizophrenia patients (Kang et al. 2011; Radant et al. 2010; Zanelli et al. 2009). The antisaccade performance deficit in schizophrenia patients is reported to be due to: (1) a deficit in top-down inhibition control of the erroneous response (Everling and Fischer 1998; Broerse et al. 2001; Brownstein et al. 2003; Curtis et al. 2001), (2) a deficit in response generation of the antisaccade (Everling and Fischer 1998; Broerse et al. 2001; Brownstein et al. 2003; Curtis et al. 2001), or (3) an emergent property of competing noisy decision accumulating processes (the erroneous prosaccade and the antisaccade) (Cutsuridis et al. 2014; Cutsuridis 2010).
Various psychopharmacological manipulations including administration of lorazepam (Green and King 1998; Green et al. 2000), risperidone (Burke and Reveley 2002; Hutton 2002), nicotine (Petrovsky et al. 2012; Rycroft et al. 2007; Depatie et al. 2002; Larrison-Faucher et al. 2004), amphetamine (Dursun et al. 1999), and modafinil (Rycroft et al. 2007) led to changes in the antisaccade performance of cohorts of patients. Risperidone has been observed to improve error rates in some schizophrenia patients (Burke and Reveley 2002; Hutton 2002). Nicotine administration in schizophrenia patients improves their antisaccade performance (Petrovsky et al. 2013; Depatie et al. 2002; Larrison-Faucher et al. 2004).
Antisaccade performance, on the other hand, of obsessive compulsive (OCD) patients has been variable and contradictory. Initial studies reported increased error rates in OCD patients compared to healthy controls, but no difference in their latencies of antisaccades (Tien et al. 1992). Other studies reported higher antisaccade latencies in OCD patients compared to healthy controls, while their error rate did not differ significantly (Maruff et al. 1999; van der Wee et al. 2006). Another study observed no differences in error rates and latencies of antisaccades between OCD patients and healthy subjects (Kloft et al. 2011). An increase in error rates and in latency of corrected antisaccades was recently reported (Damilou et al. 2016). It is speculated that the OCD antisaccade performance is due to a deficit in erroneous response inhibition control in the oculomotor circuit (Chamberlain et al. 2005; Everling and Fischer 1998; Broerse et al. 2001; Brownstein et al. 2003; Curtis et al. 2001).
Types of Theoretical Models of Antisaccade Performance
Accumulator models: In these models, the process of decision-making often involves a linear gradual accumulation of information concerning the various potential responses starting at some baseline level S0, which represents the prior expectation, at a constant rate r until it reaches a threshold ST, which represents the confidence level required before the commitment to a particular course of action. Once the decision signal crosses ST, then a response toward the target is initiated. Response time (RT) is measured as the time from the onset of the decision process till when the decision signal crosses ST. Often the rate of accumulation is assumed to vary randomly from trial to trial, with a mean μ and variance σ2 (Reddi and Carpenter 2000). Changes in the baseline level of activity, the rate of accumulation, or the threshold often result in changes in response latency. Prior expectation and level of activation of intention influence the baseline levels of activation.
Neural accumulator models: In these models, the accumulation process is represented by the firing rate of usually a population of neurons. Changes in the rate (slow or fast) of neural firing are usually nonlinear, often competing, and reflect the changes in the rate of accumulation in the linear accumulator.
Below I will review models of various degrees of antisaccade performance from both categories.
To address some of these shortcomings, Noorani and Carpenter (2014) extended their previous model (Noorani and Carpenter 2013) by including a RESTART mechanism (see Fig. 2b). In this case, when the PRO unit reached the threshold first, it restarted the ANTI unit allowing it to reach the threshold and generate the antisaccade response. Their new model successfully reproduced the “erroneous prosaccade followed by the corrected antisaccade” behavior, but failed now to reproduce the just erroneous prosaccades. This shortcoming was inherent in their model. The authors postulated that if the STOP signal did not prevent the erroneous prosaccade response, then the PRO unit will always restart the ANTI unit (Noorani and Carpenter 2014). This meant the erroneous prosaccades followed by corrected antisaccades will always be produced. If the STOP unit did prevent the PRO unit, then the ANTI unit would not restart (the corrected antisaccade will not be produced), and an antisaccade response would be generated (Noorani and Carpenter 2014). In either scenario, just an erroneous prosaccade response cannot be generated. The authors claimed in their studies participants never made any just erroneous prosaccades (private communication of the author with Roger Carpenter). However, psychophysical studies of a large group of 2006 participants performing the antisaccade task (Evdokimidis et al. 2002) reported that subjects do make the just erroneous prosaccades, but their response frequency is low. Another limitation of their new model was their consideration that the simulated latency of the corrected antisaccade is the result of the linear sum of latencies of the erroneous prosaccade and the antisaccade minus the latency of the STOP activity. This shortcoming was inherent in the model, because its units are considered linear encoders of the input information.
The three-unit antisaccade model was recently applied to a large sample of Huntington’s disease (HD) patients against healthy controls in an effort to quantitatively predict HD before symptom onset (Wiecki et al. 2016). Experimental RT distributions and error rates of pre-manifest individuals carrying the HD mutation (pre-HD), early symptomatic, and healthy controls performing the antisaccade conflict task were fit using the three-unit antisaccade model. Further machine learning analysis based on fitted model parameters revealed a key executive control parameter was predictive of HD prior to symptom onset, whereas response inhibition processes are impaired only after the motor symptoms are observed.
Neural Accumulator Models
Spiking Neuron Models
To uncover the ionic and synaptic mechanisms that produced the range of values of accumulation rates needed to produce the latency distributions of the erroneous prosaccades and antisaccades in Cutsuridis et al.’s (2007a) model, the same group (Cutsuridis et al. 2007b) introduced a multi-modular neural network model consisting of two cortical modules (FEF and LIP) that drove the SC module to decide the winning motor command to move the eyes (Fig. 4). Each cortical module was a network of Hodgkin-Huxley type excitatory and inhibitory neurons connected together. The SC module was the same as in Cutsuridis et al.’s (2007a) study. No connectivity was assumed between the cortical modules, although it has been experimentally observed (Schall 1997). Symmetric and asymmetric connection types were tried. Background noise and synaptic noise were also included in the cortical model neurons and in their connections to simulate homogeneous and heterogeneous neuronal firings. The population activity from each cortical network was extracted and a line was fitted to its ramping activity to estimate its slope. Variations in all model ionic and synaptic conductances were attempted to uncover which current(s) and what range of their conductance values reproduced the full range of slope values of the planned and reactive inputs to the SC model needed to reproduce the latency distributions and error rates of the virtual groups of participants in the Cutsuridis et al.’s (2007a) study. The model predicted that only conductance variations of the persistent Na+, NMDA and AMPA currents could produce the necessary slope variability in the cortical decision signals to reproduce the latency distributions and response probabilities of the virtual subjects.
Recently, a two-module spiking with competition network model of antisaccade performance was advanced by Lo and Wang (2016). The model consisting of sensorimotor remapping and action selection modules, the latter endowed by a “Stop” process through tonic inhibition, both under the modulation of rule-dependent control revealed the circuit mechanisms for the experimentally observed distributions of erroneous responses in the antisaccade task. In the model, fast errors resulted from failing to inhibit the quick automatic responses and therefore exhibited very short response times. Slow errors, on the other hand, were due to an incorrect decision in the remapping process and exhibited long response times comparable to those of correct antisaccade responses.
- Aichert DS, Derntl B, Wöstmann NM, Groß JK, Dehning S, Cerovecki A, Möller HJ, Habel U, Riedel M, Ettinger U (2013) Intact emotion-cognition interaction in schizophrenia patients and first-degree relatives: evidence from an emotional antisaccade task. Brain Cogn 82(3):329–336CrossRefPubMedCentralGoogle Scholar
- Antoniades C, Ettinger U, Gaymard B, Gilchrist I, Kristjánsson A, Kennard C, John Leigh R, Noorani I, Pouget P, Smyrnis N, Tarnowski A, Zee DS, Carpenter RH (2013) An internationally standardised antisaccade protocol. Vis Res 84:1–5. https://doi.org/10.1016/j.visres.2013.02.007CrossRefPubMedGoogle Scholar
- Cutsuridis V (2017b) A neural accumulator model of antisaccade performance of healthy controls and obsessive-compulsive disorder patients. In: vanVugt MK, Banks AP, Kennedy WG (eds) Proceedings of the 15th international conference on cognitive modeling. University of Warwick, Coventry, pp 85–90Google Scholar
- deWilde OM, Dingemans P, Boeree T, Linszen D (2008) Antisaccade deficit is present in young first-episode patients with schizophrenia but not in their healthy young siblings. Psychol Med 38:871–875Google Scholar
- Lennertz L, Rampacher F, Vogeley A, Schulze-Rauschenbach S, Pukrop R, Ruhrmann S et al (2012) Antisaccade performance in patients with obsessive-compulsive disorder and unaffected relatives: further evidence for impaired response inhibition as a candidate endophenotype. Eur Arch Psychiatry Clin Neurosci 262:625–634. https://doi.org/10.1007/s00406-012-0311-1CrossRefPubMedPubMedCentralGoogle Scholar
- Petrovsky N, Ettinger U, Quednow BB, Walter H, Schnell K, Kessler H, Mössner R, Maier W, Wagner M (2012) Nicotine differentially modulates antisaccade performance in healthy male non-smoking volunteers stratified for low and high accuracy. Psychopharmacology 221(1):27–38CrossRefPubMedCentralGoogle Scholar
- Petrovsky N, Ettinger U, Quednow BB, Landsberg MW, Drees J, Lennertz L, Frommann I, Heilmann K, Sträter B, Kessler H, Dahmen N, Mössner R, Maier W, Wagner M (2013) Nicotine enhances antisaccade performance in schizophrenia patients and healthy controls. Int J Neuropsychopharmacol 16(7):1473–1481CrossRefPubMedCentralGoogle Scholar
- Radant AD, Dobie DJ, Calkins ME, Olincy A, Braff DL, Cadenhead KS, Freedman R, Green MF, Greenwood TA, Gur RE, Gur RC, Light GA, Meichle SP, Millard SP, Mintz J, Nuechterlein KH, Schork NJ, Seidman LJ, Siever LJ, Silverman JM, Stone WS, Swerdlow NR, Tsuang MT, Turetsky BI, Tsuang DW (2010) Antisaccade performance in schizophrenia patients, their first-degree biological relatives, and community comparison subjects: data from the COGS study. Psychophysiology 47(5):846–856PubMedPubMedCentralGoogle Scholar
- Wiecki TV, Antoniades CA, Stevenson A, Kennard C, Borowsky B, Owen G, Leavitt B, Roos R, Durr A, Tabrizi SJ, Frank MJ (2016) A computational cognitive biomarker for early-stage Huntington’s disease. PLoS One 11(2):e0148409. https://doi.org/10.1371/journal.pone.0148409CrossRefPubMedPubMedCentralGoogle Scholar
- Frontal eye fields. http://www.scholarpedia.org/article/Frontal_eye_field
- Human saccadic eye movements. http://www.scholarpedia.org/article/Human_saccadic_eye_movements
- Antisaccade task. https://en.wikipedia.org/wiki/Antisaccade_task
- Two alternative forced choice. https://en.wikipedia.org/wiki/Two-alternative_forced_choice