Generalization reveals asymmetric and history-dependent control networks for multi-finger dexterous movements

Finger dexterity, fundamental in our daily lives, is manifested by the generation of multi-finger and multi-directional patterns of muscles activity during various motor tasks, and further, by the generalization of learning in one context to other contexts. Tying shoelaces, for example, requires precise coordination of multiple fingers, some active primarily in the flexion direction, others in the extension direction, and some immobile. Nevertheless, whether the control processes of these actions are independent or interact and potentially generalize across each other, remains unclear. In a set of experiments, we sought to characterize the behavioral principles underlying the control process, learning and generalization of dexterous extension and flexion movements. We developed an isometric dexterity task that precisely measures dexterity in terms of finger individuation, force accuracy and temporal synchronization during finger flexion and extension. First, we investigated learning and generalization abilities across flexion and extension directions, both within and across hands. To do so, two groups of participants were trained for 3 days in either the flexion or extension direction. We found improvement in all dexterity measures in both groups following training, though finger extension generally exhibited inferior dexterity. Surprisingly, while the newly acquired finger extension abilities generalized to the untrained flexion direction, the newly acquired finger flexion abilities did not generalize to the untrained extension direction. Generalization biases of the finger flexion direction were also evident in the untrained hand. Next, we examined whether the asymmetric generalization pattern of multi-finger dexterous movements was history dependent. We thus recruited skilled musicians who showed increased baseline levels of dexterity in both directions and found that the degree to which learning generalizes between two contexts was affected by prior experience. Overall, our data indicate that control of multi-digit dexterous patterns is direction-specific in humans, supporting the hypothesis that control circuits for learning of finger flexion and extension are overlapped in that they partially, but asymmetrically, transfer between directions. This ability, however, is modular as it depends on hand use and the history of prior training.