Decoding effects of psychoactive drugs in a high-dimensional space of eye movements in monkeys.

Daily injections of phencyclidine (PCP) at 3.0 mg/kg or ketamine at 0.8 mg/kg in two macaque monkeys produced robust reductions in accuracy and increases in reaction time during high cognitive-demanding tasks. Saccades, smooth pursuit, and fixation stability were significantly impaired; involuntary microsaccades during fixation increased in amplitude and were biased toward the lower visual field. Pupillary response during cognitive tasks was reduced, and sensitivity to auditory cues increased as shown by auditory evoked potentials. A machine-learning classifier identified altered states induced by different drugs with 93% accuracy, and 90% accuracy when data from one monkey trained and tested the other, demonstrating data transferability across individuals.

Study at a glance

Abstract

Oculomotor behavior has been shown to be correlated with mental disorders in clinics, making it promising for disease diagnosis. Here we developed a thorough oculomotor test toolkit, involving saccade, smooth pursuit, and fixation, allowing the examination of multiple oculomotor parameters in monkey models induced by psychoactive drugs. Eye movements were recorded after daily injections of phencyclidine (PCP) (3.0 mg/kg), ketamine (0.8 mg/kg) or controlled saline in two macaque monkeys. Both drugs led to robust reduction in accuracy and increment in reaction time during high cognitive-demanding tasks. Saccades, smooth pursuit, and fixation stability were also significantly impaired. During fixation, the involuntary microsaccades exhibited increased amplitudes and were biased toward the lower visual field. Pupillary response was reduced during cognitive tasks. Both drugs also increased sensitivity to auditory cues as reflected in auditory evoked potentials (AEPs). Thus, our animal model induced by psychoactive drugs produced largely similar abnormalities to that in patients with schizophrenia. Importantly, a classifier based on dimension reduction and machine learning could reliably identify altered states induced by different drugs (PCP, ketamine and saline, accuracy = 93%). The high performance of the classifier was reserved even when data from one monkey were used for training and testing the other subject (averaged classification accuracy = 90%). Thus, despite heterogeneity in baseline oculomotor behavior between the two monkeys, our model allows data transferability across individuals, which could be beneficial for future evaluation of pharmaceutical or physical therapy validity.