LOCUS COERULEUS SIGNALS SIGNED VISUOMOTOR PREDICTION ERRORS THAT DEPEND ON TASK CONTEXT

The locus coeruleus (LC) is thought to dynamically regulate learning rates across the brain by signalling unexpected uncertainty – violations of predictions beyond the expected range. A possible way to compute unexpected uncertainty globally is via the summation of prediction errors across modalities, alongside a subtraction of the expected rate of error. Consistent with this idea, the LC displays prediction error-like responses to a variety of stimuli, including rewards and visuomotor mismatches. If these responses result from computation of unexpected uncertainty, we expect that: 1) LC responses will generalise across modalities and increase when coincident errors occur, and 2) persistent increases in the ongoing rate of errors (i.e., high uncertainty) should become expected and have a subtractive influence on LC responses. We tested these ideas using fiber photometry recordings of LC calcium activity while mice were engaged in a range of different virtual reality (VR) paradigms. We found several phenomena in LC activity that are inconsistent with these predictions. First, LC responses to positive and negative visuomotor mismatch stimuli are heterogeneous across imaging sites. Second, increasing the uncertainty in visuomotor coupling results in a negligible short-term impact on LC mismatch responses. Finally, LC visuomotor mismatch responses are context-dependent, reducing in amplitude in the context of a spatial rewarded task. In sum, our results are inconsistent with the idea that the LC signals a global measure of unexpected uncertainty, and are more consistent with heterogeneous LC populations that signal signed prediction errors.