Humans and rodents can associate actions with their outcomes and modify expectations when associations change. These cognitive adaptions accommodate change and presumably optimize decision-making. For example, we might modify our driving route when construction blocks our path or abstain from alcohol when we need to drive, expecting that both actions will deliver us safely home. Across species, medial prefrontal cortical regions are involved in linking actions with valued outcomes, but contributions of the orbitofrontal cortex remain contentious. One issue is that the orbitofrontal cortex occupies a large territory, yet is sometimes treated as a homologous structure. A related concern is the assumption that ventromedial subregions, like lateral regions, specialize in stimulus−outcome associations (linking cues, rather than actions, with likely outcomes) agnostic to action−outcome associations. Nevertheless, poor or aberrant decision-making is commonplace in neuropsychiatric illnesses, necessitating a full dissection of how action-outcome associations form, update, and solidify.

Parkes et al. [1] made important in-roads in resolving controversies. Rats were trained to associate two actions with a single food reward; then the actions were paired with different, unique rewards in the days preceding a devaluation test. Inactivation of the ventrolateral orbitofrontal cortex (VLO) either during the final training days or during the subsequent probe test blocked the ability of rats to choose actions based on the value of respective rewards. Meanwhile, VLO inactivation had no effect when action−outcome contingencies had not changed during training. In another investigation, inactivation of the VLO immediately following the violation of a familiar action−outcome association in mice occluded optimal responding in a later test, even when the VLO was back “on-line” [2]. Thus, the VLO appears necessary for stabilizing newly formed or updated action−outcome associations, which then guide future behavior.

Notably, linking actions with their outcomes involves dendritic spine plasticity on excitatory neurons in the VLO [2]. Specifically, updating action−outcome expectations reduces thin-type dendritic spines, considered immature, on layer V neurons. Meanwhile, the proportion of mushroom-shaped spines, considered mature, increases, potentially solidifying newly modified action−outcome associations to optimize future decision-making. Remarkably, inactivating VLO neurons upon the violation of familiar action−outcome associations not only blocks response updating, but also inhibits dendritic spine plasticity in an activity-dependent manner [2]. These patterns strongly suggest that dendritic spine plasticity on excitatory VLO neurons is necessary for forming action−outcome associations, consistent with evidence that orbitofrontal neurons are capable of forming and maintaining long-term reward-related memory to support behavioral adaptations [3].

Orbitofrontal neurons display a rich diversity of functionally distinct populations based on input/output patterns, many of which make unique contributions to flexible decision-making [3, 4]. A key question is thus: What inputs to the VLO help to form/update action−outcome associations? Are these inputs distinct from those supporting other associations, e.g., stimulus−outcome associations? Basolateral amygdala (BLA) projections are one candidate. BLA lesions alter the reward-related coding properties of orbitofrontal neurons [5], and BLA → orbitofrontal cortical connections appear necessary for certain forms of reinforcement learning [4], including in primates ([6] and references therein). Whether and how these “bottom-up” connections are involved in forming action−outcome associations should be resolved.

Funding and disclosure

DCL and SLG are supported by the National Institute of Mental Health and National Institute on Drug Abuse at the National Institutes of Health (grant numbers 044297, 117103, 117873). The authors declare no competing interests.