The orbitofrontal cortex (OFC) is often proposed to signal expected value, at least for rewards (Padoa-Schioppa and Conen, 2017, Wallis, 2012). Cues that predict reward (i.e. conditioned excitors) are generally reported to elicit more neural activity in OFC than cues that predict no reward, and cues (or cue combinations) that predict multiple rewards are reported to elicit even more activity than cues that predict one (Critchley and Rolls, 1996b; Takahashi, Chang, Lucantonio, Haney, Berg, Yau, Bonci, and Schoenbaum, 2013; Thorpe, Rolls, and Maddison, 1983). Similarly, neural activity in OFC is normally reported to scale with reward probability or relative reward values (Critchley and Rolls, 1996a, Kennerley et al., 2011, Tremblay and Schultz, 1999). This viewpoint is most fully realized in economic choice tasks, where both single-unit activity and the BOLD response in OFC are said to track pure value, independent of other information about the cues and outcomes (Levy and Glimcher, 2011, Padoa-Schioppa and Assad, 2006).

This raises the question of how neural activity in the OFC is modulated by cues that signal the reduction or omission of reward, i.e., conditioned inhibitors (for a recent review on conditioned inhibition, see Sosa and Ramirez, 2019). Several learning models envisage conditioned inhibitors as acquiring value of the opposite valence to that of conditioned excitors (Konorski, 1967, Pearce and Hall, 1980, Rescorla and Wagner, 1972, Sutton and Barto, 1990). Consistent with this, inhibitors of reward will evoke withdrawal responses in situations where conditioned excitors evoke approach (Wasserman, Franklin, and Hearst, 1974). In addition, inhibitors will drive a below-baseline suppression in the activity of midbrain dopaminergic neurons (Tobler, Dickinson, and Schultz, 2003)—the opposite response to that evoked by conditioned excitors. If the OFC is functioning primarily to tally up the likely reward expected, then the neural activity in response to a conditioned inhibitor should reflect this negative value. Specifically, once trained, an inhibitor should reduce activity in OFC neurons when compounded with an excitor, and it should evoke low or even suppressed activity when presented alone by comparison with a neutral cue, as reported for dopamine neurons. This should be apparent in the single unit activity and population responses, which should simply scale with expected value (value hypothesis).

However, OFC has also been implicated in constructing and updating a sensory-rich model of the associative structure of the environment that specifies the probability of future outcomes to guide behavior. This view aligns with proposals that conditioned inhibitors cancel or reduce outcome-specific expectancies generated by conditioned excitors in order to estimate outcome probability. However, in the absence of such backdrop expectancies inhibitors should be associatively inert (Konorski, 1948, Lysle and Fowler, 1985, Miller and Matzel, 1988, Rescorla, 1979). That is, conditioned inhibition depends on excitation not only for its acquisition but also for its expression. This dependence on excitation makes intuitive sense: if you do not expect to be gifted a cake at work since it is not your birthday, you will likely not notice all the predictors that you are not getting cake today. Consistent with this, evidence shows that conditioned inhibitors signal the absence of sensory-specific outcomes (Laurent, Wong, and Balleine, 2017) and will not extinguish their inhibitory properties if presented by themselves in the absence of such an excitatory background (Zimmer-Hart and Rescorla, 1974). If OFC represents conditioned inhibition in this fashion, then we would expect neural activity to a conditioned inhibitor to be likewise conditional. Single unit and population responses should reflect a decrement in outcome expectancy when the inhibitor is presented against the background of an excitor, but fail to distinguish the inhibitor from a neutral cue when presented alone. Similarly, ensemble responses should classify the inhibitor in a way that reflects this expectancy dependence (expectancy-dependent hypothesis).

Here, we assessed these predictions by recording from small groups of neurons in the lateral OFC of rats during training on a series of summation tests. Our design allowed for the simultaneous examination of the behavioural and neural effects of a conditioned inhibitor on the background of excitation as well as on its own. Rats showed negative summation on test, in keeping with the notion that they learned that the inhibitor signaled the absence of reward. Against this backdrop, we found unit and population responses that scaled with expected reward value both during training and summation tests when the inhibitors were paired with a conditioned excitor. Critically, however, the responses of these neurons did not differentiate between the conditioned inhibitor when presented by itself and a neutral cue. Further, when the ensemble patterns were analyzed, activity to the inhibitor did not classify according to putative value. Instead, it classified with a same-modality neutral cue when presented alone and as a unique item when presented in compound with a reward-predicting cue. This pattern of results suggests that OFC represents conditioned inhibition in a manner consistent with the expectancy-dependent hypothesis, and contrasts with what has been observed in the midbrain dopaminergic system (Tobler et al., 2003). Our findings support the notion that OFC is critical for building a model of associative relations and dependencies to guide behavior.