The data here provide novel behavioral evidence that the lose-shift response is a product of sensorimotor systems, and that the regulation of such responding is compromised differently in men and women with high recreational cannabis use. A high proportion of lose-shift responding is sub-optimal in the present task because it is predictable, and can therefore be exploited by the computer opponent. Indeed, the propensity for lose-shift responding is negatively correlated with task performance here. Nonetheless, subjects engage this response above chance levels for several hundred trials before learning to suppress it. Similarly, humans exhibit considerable difficulty in generating random response sequences, preferring to alternate choices at a rate well above chance (Falk and Konold, 1997; Sun and Wang, 2012). Consequently, the lose-shift is likely a default strategy, consistent with previous work in humans (Ivan et al., 2018), and analogous to what has been observed in animals performing a similar task (Gruber and Thapa, 2016). As lose-shift responses eventually converge to chance levels in trials with no global changes, the probability of win-stay responses increase above chance levels. We found that this negative correlation between lose-shift and win-stay was significant and strikingly similar to rodents (Gruber and Thapa, 2016), suggesting that lose-shift and win-stay are expressed by neural systems in competition with one another.

We show here that participants overwhelmingly perform the lose-shift according to target position, rather than target identity. In other words, participants avoided the prior position of the previously chosen target when it was unrewarded. This novel observation reveals a strong spatial component to the lose-shift. These data are consistent with the notion that lose-shift is a product of sensorimotor systems. Loss-driven response shifting is reduced following lesions to sensorimotor striatum in animals (Skelin et al., 2014; Gruber et al., 2017; Thapa and Gruber, 2018) or damage to putamen/insula in humans (Danckert et al., 2011); these homologous structures are strongly involved with sensorimotor control. Moreover, decision times are lower for lose-shifting than for lose-stay responses, even when global changes in position (that require equally distant arm movements) are present. There are multiple reasons this may occur. First, the visual systems of the brain process information about spatial position independently from other object characteristics (Mishkin et al., 1983; Haxby et al., 1991). The ventral “where” pathway processes information more quickly than the “what” pathway. (Goodale and Milner, 1992; Deubel et al., 1998). Secondly, the ventral pathway may be used to compute actions prior to stimulus presentation. In the perceptual learning literature, activity in both the motor and visual cortices builds up prior to stimulus onset, and reflects stimulus expectation and the associated motor responses (de Lange et al., 2013). Moreover, pre-response fluctuations in beta-power motor activity are also predictive of choice alternation (i.e., lose-shift), regardless of associated motor action (Pape and Siegel, 2016). There is evidence that loops involving premotor cortex and the lateral striatum map vision and other sensory modalities into an egocentric space. The ventral premotor cortex contains neurons that both drive motor actions, but also encode locations of visual, tactile, and auditory stimuli (Fadiga et al., 2000). Consequently, they form a motor vocabulary for mapping several modalities into actions in a common egocentric space. Even when stimuli are removed, these neurons still respond to the position of remembered objects in relation to the body (Graziano and Gross, 1998). The putamen (LS in rodents) also contains these bimodal visuomotor cells (Graziano and Gross, 1996), and therefore has the capacity to mediate lose-shift from a remembered location. In the context of our study, spatial rearrangement of choice targets subverts this motor preparation, requiring choices to be recalculated following stimulus onset. This is evidenced by the increase in response times following local swaps and displacement. Interestingly, local swaps had a larger and more consistent effect on response times than did displacement, suggesting that a greater level of motor recalculation is required. Specifically, we speculate that it requires more time for the executive control to overcome the intrinsic inhibition of a previously unrewarded action (than a novel one) in the motor system in order to intentionally select it.

The influence of local and global changes in choice position also highlights the importance of egocentric and allocentric processing of space. While local changes in choice orientation modulate the lose-shift, these effects persist even when all choices are moved to a new global position relative to the observer. Conversely, the win-stay is much less affected by spatial position. Local changes do have an effect on behavior, but these are eliminated by concurrent global changes. These results highlight the importance of allocentric processing on the lose-shift. Choice positions are calculated relative to one another, allowing their associated values to be maintained across large global movements in choice position. Conversely, while processing of the win-stay is less reliant on spatial information, egocentric reference frames are more important than allocentric, where choice value is calculated relative to the subject. Consequently, local and global changes have a large effect on win-stay behavior.

In addition to driving different decision strategies, the putamen/LS and ventral striatum (VS, including the nucleus accumbens) also respond differently to psychoactive drugs. Relative to the VS, the LS exhibits a much higher density of dopamine transporters (Coulter et al., 1997), endocannabinoid receptors (Herkenham et al., 1991), opioid receptors (Benfenati et al., 1991), and alcohol-sensitive NMDA receptors (Liste et al., 1995). Consequently, THC administration temporarily increases dopamine release throughout the striatum, but particularly in the LS (Sakurai-Yamashita et al., 1989; Jentsch et al., 1998), resulting in diminished loss aversion. The effects of acute ethanol exposure are similar, though greater in the VS (Clarke et al., 2015; Vena et al., 2016). Conversely, long-term sensitization to alcohol and cannabis reduces availability of striatal dopamine receptors (Martinez et al., 2005; Budygin et al., 2007; Albrecht et al., 2013) and cannabinoid receptors (Villares, 2007), especially in the LS. Chronic exposure also inhibits the prefrontal cortex and anterior cingulate (Goldstein et al., 2007, 2009) which may result in an attenuated feedback-related negativity, an error signal generated in the cingulate necessary for behavioral adaptation following losses (Cohen and Ranganath, 2007). Instead, choice control is shifted to the LS (Everitt and Robbins, 2005, 2013; Everitt et al., 2008; Lucantonio et al., 2014). We expect this effect to impair the ability of participants to use executive control to suppress lose-shift responding by the sensorimotor systems, while having little effect on win-stay behavior. We do not have sufficient primary evidence to hypothesize how the change in receptor densities by repeated alcohol/THC exposure affects lose-shift processing within the LS and/or other components of the sensorimotor system.

In the present study, we find that self-reported level of recreational use of cannabis affects task performance, but that this differs on the basis of biological sex. Elevated cannabis use in men decreased spatial modulation of the lose-shift, possibly through dopaminergic desensitization of the LS. As seen in Figure 4, baseline lose-shifting is also reduced, falling below 50% in trial blocks 4 and 5. With this reduction, lose-shift responding after swaps increases to 63%. Therefore, either the calculation of the lose-shift is affected or its suppression by executive systems is enhanced in male cannabis users, while spatial processing remains unaffected. Conversely, female cannabis users exhibit decreased task performance, possibly due to weakened suppression of sensorimotor responding by the prefrontal cortex. Furthermore, they show a moderate and significant increase in baseline lose-shift responding [F_(1,51) = 4.109, p < 0.048], revealed by a mixed effects model between female controls and cannabis users in the no change condition (Figure 3C). It thus appears that females with high cannabis use exert less executive control over sensorimotor systems in our task.

While it is tempting to describe this sex difference as a consequence of different drug effects on the brain in males and females, several alternatives are also possible. For example, a confounding factor may be present that promotes high levels of recreational drug use and also impairs sensorimotor regulation. Unfortunately, the WHO ASSIST is not sufficient to infer whether these are the case in the present study. However, it is known that females are more susceptible to drug tolerance (including cannabis) and sensitization than are males (Robinson, 1988; Wakley et al., 2014). Drug use is also comorbid with mood and anxiety disorders, particularly depression (Zilberman et al., 2003), which causes heightened loss aversion (Beevers et al., 2013). These differences are possibly due to the effects of estrogen, which enhances striatal dopamine release in response to psychoactive drugs (Becker, 1999) and alters the effects of drugs on the prefrontal cortex. Females rats with high estrogen levels exhibit dysfunction of the prefrontal cortex relative to males and low-estrogen females when exposed to dopamine-enhancing drugs (Shansky et al., 2004). Estrogen also heightens the effects of cocaine and amphetamine, causing an abnormal BOLD response in rats (Febo et al., 2005; Sárvári et al., 2014). Alcohol and cannabis consumption also increase oestradiol levels, and can inhibit testosterone production in males (Kolodny et al., 1974; Maskarinec et al., 1978; Harclerode, 1984; Purohit, 2000; Yonker et al., 2005). In males, increased estrogen and reduced testosterone levels cause declines in spatial cognitive ability (Janowsky et al., 1994). Therefore, the heightened susceptibility of the PFC to the combined effects of estrogen and drug abuse provides an explanation for why only women with high ASSIST scores show a dominance of sensorimotor control, without compromising the spatial dependence of lose-shift. Specifically, this population had accelerated decision speeds, lower proportion of wins, and a tendency for lower entropy of response sequences. On the other hand, the lose-shift remained sensitive to swapping cue locations, which is similar to controls, but opposite of what is observed in males with high ASSIST scores.

Our analysis of behavior through a reinforcement learning framework also revealed a cannabis × sex interaction. Whereas the other analysis presented here focuses on the effects of the previous trial, the Q-learning model allowed us to examine effects that span many trials. It was the men who used cannabis who stood out in this analysis; they had increased learning when previous cues were displaced locally, and decreased learning when previous cues were switched globally. We expect such learning is part of a reinforcement learning scheme in “goal-directed” brain circuits linked more closely to executive function than sensorimotor control (Balleine and O'Doherty, 2010; Gruber and McDonald, 2012), suggesting that not only is there an enhanced suppression of sensorimotor control by executive function in male cannabis users, but that adaptation by the executive system is also different than the other groups. It is worth noting, however, that our sample (as is common in the field) was predominantly young university students, who are presumably well-educated and high functioning. We urge caution in extrapolating our findings to the general public.

The interpretation of data in this study faces several challenges besides the aforementioned limitations of the ASSIST. First, alcohol and cannabis use are highly concordant (Spearman's correlation of ρ = 0.533, p < 0.001 in our sample), and likely additive in their effects. Second, the sexually dimorphic effects observed here may be due to confounding interactions between drug use, IQ, and/or psychiatric disorders that have different prevalence among the sexes. However, the sexually dimorphic distribution of endocannabinoid receptors in the striatum and prefrontal cortex (De Fonseca et al., 1994) likely also play an important role. For instance, errors when reconstructing spatio-temporal sequences were reduced in men and increased in women following THC treatment (Makela et al., 2006). We previously reported that lose-shift is decreased by acute administration of THC in female rats (Wong et al., 2017a). It is possible that the down regulation of receptors in heavy chronic users may cause the inverse, which would be consistent with the data here.

In sum, the data presented here indicate that lose-shift responding is a useful gauge of the cognitive control over sensorimotor responding in humans, and that this is impacted differently in men and women that heavily use cannabis. These linkages are important factors to account for the impact of lose-shift responding in real-world economic decision making, such as gambling (Worthy et al., 2013; Abouzari et al., 2015), as well as clinical/laboratory testing of cognitive flexibility with tasks such as the Wisconsin Card Sorting Task that involve loss-based shifting of response policies.