Do prism and other adaptation paradigms really measure the same processes?

Abstract

Sensorimotor plasticity allows the nervous system to set up appropriate motor and sensory compensations when individuals face changing demands in a given motor task. A much-debated question in neuroscience research is the identification of processes that encompass this capacity of plasticity. Prism adaptation is the oldest experimental paradigm that has been used to achieve this goal (Helmholtz, 1867). Since 1990's, other paradigms have emerged such as visuomotor rotations or dynamical perturbations (inertial Coriolis forces, velocity-dependent force-field).

We compared these paradigms with respect to three specific methodological features: application of the perturbation, after-effects, and generalization. This work aimed to shed light on the following central issue: Do all these paradigms involve similar processes? We used generalization properties—a relevant feature associated with the credit assignment problem—to emphasize the involvement of different processes in “adaptation” paradigms. We therefore classified these processes based on the context specificity of elicited transformations.

This review reveals that the processes involved are closely linked to paradigm-related experimental conditions. Context-independent processes appear to be favored when errors are attributed to our own sensorimotor performance (prism, Coriolis) whereas context-dependent processes appear to be mostly mediated by attribution of errors to a specific external interface (visuomotor rotation, force-field). This work encourages researchers to consider the methodological aspects specific to each paradigm for future investigations of sensorimotor plasticity.

Introduction

Humans are remarkably able to produce a variety of accurate motor behaviors despite constantly changing demands. This capacity includes the ability to modify a known behavior to face new conditions (for example, reaching underwater, through magnifying glasses or under a reduced gravitational field). It also includes the ability to acquire a new motor behavior (for example riding a bicycle). A much-debated question in neurosciences research is to identify the processes that encompass this capacity of plasticity.

Recently, several reviews have focused on the classification of these adaptive processes using various terms to describe them, such as sensorimotor learning, motor adaptation (Bastian, 2008, Kitago and Krakauer, 2013, Krakauer and Mazzoni, 2011), motor learning (Bastian, 2008, Kitago and Krakauer, 2013), skill learning (Kitago and Krakauer, 2013, Makino et al., 2016), skill acquisition (Kitago & Krakauer, 2013), sensory perceptual learning (Bedford, 1993, Makino et al., 2016), sensorimotor associative learning (Makino et al., 2016), etc.

Despite the use of varying definitions and terms, these reviews share a couple of key features and refer to two main notions: “adaptation” and “learning”. Unfortunately, these terms are often used interchangeably and without precision. However, several attempts of definitions have been proposed. For Bastian (2008), “adaptation” broadly refers to the modification of a pre-existing pattern in response to altered conditions. “Learning” is the acquisition of a new motor program or skill. In addition, Bastian suggests that “adaptation” must imply “after-effects”. This means that participants cannot retrieve their prior behavior once adapted, unless they de-adapt. On the other hand, “learning” is associated with the possibility to store the new movement pattern, so it becomes immediately available in the appropriate context. These characteristics may imply differences concerning the context specificity of elicited transformations. In fact, as an adjustment of the sensorimotor system to new conditions, the consequences of adaptation should apply to different movements performed by the subject. Thus, it would not be limited to the target and conditions that were practiced during exposure to the perturbation, but should extend across space (Bedford, 1993, Redding et al., 2005, Torres-Oviedo and Bastian, 2012). Conversely, learning would be highly contextual. As such, learning would not extend to other types of action toward the environment or to other experimental conditions, as long as these conditions are completely different from those experienced during the learning session (e.g., learning to smash in tennis should not interfere with the ability of riding a bicycle) (Wolpert & Flanagan, 2010). Obviously, adaptation may also be context-dependent and learning a skill such as riding a bicycle is known to generalize to other types of bicycles (Braun, Aertsen, Wolpert, & Mehring, 2009). Thus, the classification between adaptation and learning might be confusing.

Generalization properties appear to be particularly interesting and meaningful in order to question the processes involved during adaptation paradigms. In fact, patterns of generalization provide clues about the nature of representational transformations set up by the nervous system to face a given perturbation (Poggio and Bizzi, 2004, Taylor and Ivry, 2012). Additionally, the possibility to transfer compensations acquired to other conditions might be useful in the field of neurorehabilitation. Therapists aim to design strategies that allow patients to reinvest compensations acquired during rehabilitation in other daily life situations. As such, disentangling the processes through their context-dependency should be of great practical interest. Taylor and Ivry (2012) defined the context by the layout of the target locations. More globally, the context could refer to the whole characteristics of the task performed by the subject during behavioral transformations. It therefore comprises environmental characteristics (e.g., layout of the targets locations, apparatus and settings, instructions) but also individual characteristics (e.g., postural configuration, effectors). As such, a context-dependent process implies transformations that are specific to the context in which they emerged (i.e., the perturbation/exposure context). Conversely, a context-independent process entails compensations that are transferable to other conditions, beyond the exposure context.

The oldest paradigm used to explore sensorimotor plasticity is the use of prism goggles to laterally shift the visual field (Helmholtz von, 1962, Helmoltz von, 1867). When a subject wearing prisms points quickly to a near object, he/she initially points to the prism-displaced image of the object, experiencing a pointing error. After tens of pointing attempts, the pointing error is gradually reduced close to zero. This experiment simply depicts the short-term plasticity of the central nervous system (CNS), which allows to adapt to changes in the relationships between visual inputs and corresponding motor outputs (for a review, see Redding et al., 2005).

Numerous other “adaptation” paradigms have been used to investigate the compensations set up in response to altered conditions. Among them, dynamical perturbations (either contactless Coriolis forces or force-field) and visuomotor rotation paradigms are the most commonly used. In contactless Coriolis force-field experiments, participants are required to perform reaching movements while sitting on a rotating chair (Coello et al., 1996, Lackner and Dizio, 1994, Lackner and DiZio, 2002). In force-field experiments, a manipulandum produces dynamic forces on the hand while subjects are performing reaching movement toward visual targets with this robotic arm (Shadmehr & Mussa-Ivaldi, 1994). In the case of visuomotor rotations, subjects’ index finger position is coupled with the position of a cursor on a screen while they are asked to perform reaching movement. The relationship between the hand position and the cursor position is then perturbed by imposing a clockwise or anti-clockwise rotation of the moving hand visual reafference (Krakauer et al., 2000, Krakauer, 2009, Prablanc et al., 1975).

In most contact force-field and visuomotor rotation paradigms, a critical feature is the subject's knowledge of a physical interacting interface (the manipulandum and its tactile feedback) between motor planning and endpoint visual representation. Moreover, in visuomotor rotation, the actual feedback is usually not a direct view of the hand position, but a virtual cursor image of the hand displayed on a screen. Conversely, contactless Coriolis force-field or prism paradigms are related to a contactless discrepancy between motor planning and visual reafference. Then, feedback is directly perceived.

Another methodological question deserves particular consideration concerning contact force-field and visuomotor rotation paradigms: how would the after-effects change if the contact with the manipulandum was suppressed while keeping the same structure of the adapted/learned movement? A few studies have considered this crucial question (Cothros et al., 2006, Kluzik et al., 2008) and concluded that most of the aftereffect, when present, resulted from the adaptation/learning of the interface properties rather than from the adaptation/learning of the subject's own sensorimotor system. By contrast, in prism adaptation, the after-effect's strength is the same when the prisms are removed or are replaced by null deviation sham prisms. In addition, the different paradigms are associated with various generalization properties. This could underlie the contribution of different processes.

Comparison of methodological aspects between “adaptation” paradigms may highlight the fact that adaptive processes actually involved are closely linked to the paradigm used. Through a comprehensive comparison between prisms and other adaptation paradigms, the scope of this review is to identify the contributions of context-dependent versus context-independent processes during these experiments. Particularly, we will focus on three methodological points associated with specific behavioral findings: application of the perturbation, conditions in which after-effects are tested, and assessment of generalization properties. We will discuss how these features could emphasize the possibility that different processes are actually involved. To achieve this, we will review the existing modelings of sensorimotor adaptation and the question of credit assignment. We will focus the review to four most-commonly used error-based paradigms that investigate reaching movements: prisms, Coriolis force-field, force-field and visuomotor rotations.