Pointing tasks involve the displacement of the hand, or working point of the upper limb toward an object of interest. In healthy subjects, this movement tends to be predominantly mediated via feed-forward control, as evidenced by the smooth, bell-shaped velocity profile described by the hand (Abend et al., 1982). Prehension tasks couple reach and grasp components, identified by the pioneering works of Jeannerod (1984). The pre-shaping of the finger aperture during reaching and the smooth peak velocity of the reaching hand show that control is anticipated as a function of the object position in space and of its intrinsic characteristics [review in Jeannerod (2009) see Smeets et al. (2019) for an alternative interpretation of the coupling]. In these types of prehensile tasks, the upper-limb may be seen to possess no less than 7 degrees of freedom (DoF) afforded by the rotational axes of the shoulder, elbow and wrist. The pointing task is defined by 6 DoF, corresponding to the 3D position and orientation of the object in space. As a result, there is kinematic redundancy at the level of the upper limb as the 7 DoF all contribute to the displacement of the hand for grasping in the 3D space (Desmurget and Prablanc, 1997). As initially proposed by Bernstein (Bernstein, 1967), control of the upper limb may be based on synergies which share the same spatio-temporal properties, and can be additively combined in a task specific way. There is still no agreement as to whether synergies are coordinated at the joint kinematic (Scholz et al., 2000; Yang et al., 2002) or muscular level (d’Avella and Bizzi, 2005; Ting and McKay, 2007; Bizzi et al., 2008). The Uncontrolled Manifold theory subscribes to the former, suggesting that synergies are flexible and allow automatic compensation between elements in order to stabilize the important task related variables such as the displacement of the endpoint (Yang et al., 2007; Latash, 2008; Martin et al., 2009). As proposed by Feldman, the neural basis of motor control, bridging the gap between physiology and biomechanics, could be the modulation of the stretch reflex threshold by the CST acting on motor neuron membrane potential (Feldman, 2015), consistent with the formation of synergies (Latash et al., 2010). This theory is disputed, and the reduction of redundancy by synergies has also been interpreted in the framework of optimal control (Guigon et al., 2007). However, there is accepted evidence that the primary motor cortex is responsible for the coordination of muscle and joints to generate the spatio-temporal form of goal directed movements (Kalaska, 2009; Capaday et al., 2013).

After stroke, patients generally exhibit decreased force and range of motion across joints with alterations in movement coordination, referred to as pathological synergies (Brunnstrom, 1970). Kinematic analysis underscores these direct consequences of the CST lesion and may equally serve to distinguish certain compensatory mechanisms (Figure 2).

The most prominent aspect of the CST lesion on kinematics is the duration of upper limb movements, with observable segmentation of the velocity profile (Trombly, 1993; Roby-Brami et al., 1997; van Dokkum et al., 2014; van Kordelaar et al., 2014; Buma et al., 2016). Smoothness metrics, such as the jerk, or spectral arc length, may be used to quantify this particular kinematic feature. This may provide an indication of the level of motor control, or ability to perform efficient movements (Rohrer et al., 2002; Balasubramanian et al., 2015; Buma et al., 2016). Through the course of stroke recovery, reduced smoothness might suggest problems in the central blending of sub-movements (Rohrer et al., 2002). Another hypothesis is that smoothness deficits reflect the suboptimal performance of secondary sensory-motor brain areas (Buma et al., 2016) recruited at the structural level to compensate for the faster and more synchronized CST.

Secondly, the normal flexible shoulder-elbow coordination is disrupted with anomalies in synergistic control at both joint (Levin, 1996; Reisman and Scholz, 2003; Roby-Brami et al., 2003b; Micera et al., 2005) and muscular levels (Beer et al., 2000; Cheung et al., 2012). The normal flexible inter-joint coordination is replaced by stereotypic synergies (Twitchell, 1951; Dewald and Beer, 2001). As a result, hand orientation at the time of grasping is also altered (Roby-Brami et al., 2003b; Sangole and Levin, 2009). It is likely that the disruption of the normally flexible kinematic and muscular synergies, is a direct consequence of the lesion of the primary motor area (Capaday et al., 2013). Abnormal stereotyped synergies are probably due to the reorganization of the neural pathways at the structural level, as described in the Section “Biological Change to Upper Limb Motor Pathways Following Stroke.” Changes in shoulder flexion and elbow extension post stroke limit the upper-limb workspace, and consequently the patient’s reaching abilities (Levin et al., 2002; Reisman and Scholz, 2003; Roby-Brami et al., 2003a). However, hemiparetic patients with mild impairment may retain the ability to use the abundance of DoF to stabilize the trajectory of the hand via automatic compensation of errors between DoF (Reisman and Scholz, 2003). This ability is less flexible than in healthy subjects, particularly when the trunk is involved to reach more distant targets (Reisman and Scholz, 2006).

In effect, coordination between the upper limb and trunk is modified, and stroke patients make excessive use of trunk flexion in forward reaching tasks (Roby-Brami et al., 1997; Cirstea and Levin, 2000; Levin et al., 2002). This is interpreted as a compensatory movement pattern in response to the shortening of the functional length of the limb, arising from impaired elbow extension (Cirstea and Levin, 2000; Roby-Brami et al., 2003a). Indeed, this voluntary control of the trunk remains relatively preserved in most stroke patients due to its bilateral control (Robertson and Roby-Brami, 2011). According to the context, patients may keenly adjust trunk rotation to target direction (Robertson and Roby-Brami, 2011) and to the possible voluntary triggering of pathological synergies (van Kordelaar et al., 2012; Levin et al., 2016). The involvement of the trunk in reaching is interpreted as the spontaneous adaptive use of the redundant and abundant DoF of the body allowing task accomplishment (displace the hand) despite the upper-limb impairment. The use of such compensations may, however, mask a patient’s actual movement abilities. Providing specific instructions to a patient, or use of a mechanical constraint to limit compensatory movement patterns can result in qualitatively better performance. For instance, when the trunk is blocked, a reaching movement within the arm workspace can be successfully executed (at greater effort) with an improved shoulder-elbow coordination (Michaelsen et al., 2001; Levin et al., 2004; Bakhti et al., 2018). Preference for compensatory movement patterns though, do risk becoming highly automated in what is called “learned non-use” (Taub et al., 2006), (Figure 2, also see section “Manipulation and Object Handling”). Once established, these behavioral changes may be difficult to break down, and limit long term clinical outcomes.

Analytical studies of hand dexterity focused on precision grip or reach to grasp actions have generally overlooked the great variety of hand configurations adapted to the multiple tasks of everyday life (Bullock et al., 2013). The great flexibility of the human hand to grasp objects (Kapandji, 1980) has been mainly studied using qualitative methods. Napier (Napier, 1956) proposed a dichotomous classification of grasping: “precision grip” and “power grasp.” Later, Iberall et al. (1986) added an intermediate “key grip” and proposed a systematic description of possible finger opposition configurations. This taxonomy of grasping was further developed in the context of anthropomorphic robotics (Feix et al., 2015).

The impairments due to stroke limit the possibilities of action on the environment and, as a consequence, the hemiparetic patient may develop alternative prehension strategies. In order to experimentally investigate this compensation phenomenon, patients must be free to act spontaneously, independent of unnecessary physical constraints or external instruction. Kinematic studies of reach to grasp indicate considerable differences in motor planning. In a study by Roby-Brami et al. (1997), two notable reaching patterns were distinguished. The first involved relatively direct movement of the hand along the sagittal axis toward the object, using the table as a support for the weight of the hand during reaching, this was referred to as a “sliding” strategy. The second pattern observed in this study was characterized by a greater amplitude of vertical movement as patients lifted the hand and descended upon the target object, an action described as “grasping from above” [similar patterns are also observed in tetraplegic patients; (Laffont et al., 2000)]. Importantly, each of these prehensile actions were found to be associated with the severity of hemiparesis, with the “sliding strategy” observed in patients with more proximal weakness and “grasping from above” in those with more distal weakness. These observations illustrate how distinct changes in motor planning emerge in response to the specificity of the motor impairment. Additionally, the ‘sliding strategy’ provides a simple example of how patients may spontaneously exploit features in the environment in order to carry out functional tasks in naturalistic activity.

Significant variation in grasp configuration is also observed in patients post-stroke. When displacing handheld objects, healthy subjects generally use a precision grip or multipulpar grasp including the thumb and a number of fingers according to the size of the object (Cesari and Newell, 1999). Stroke patients, however, appear to use these particular grasp configurations much less frequently when employing objects regularly used in daily life activities (Roby-Brami et al., 1997; Bensmail et al., 2010; Garcia Alvarez et al., 2017). For example, while the majority of healthy subjects used multipulpar grasps to take a spoon, water bottle or ball, the different stroke patients used various combinations of palmar and digito-palmar grasp configurations (Garcia Alvarez et al., 2017). Others still were seen to use a particular “raking” strategy, either with the four fingers and the palm parallel to the table, or with the ulnar aspect of the hand, the palm perpendicular to the table.

Again, the movement variability observed in these works are likely associated with the individual’s impairment and the means by which that person elects to overcome the associated movement limitation. More specifically, we propose that the preservation of precision grasps in some patients might be attributed to less severe lesions of the pyramidal tract or recovery due to cortical plasticity. In contrast, alternative “raking” grasp strategies could be archaic motor acts similar to those of monkeys who lack thumb opposition and have less developed cortico-spinal tracts (Maier et al., 2005). This interpretation is consistent with Jackson’s dissolution concept (York and Steinberg, 1995) and the hierarchical evolutionary organization of dexterity proposed by Bernstein (Bernstein, 1996). Second, we suggest that patients preferentially use standard grasp types if they can. Increased severity of the motor deficit will consequently limit the range of grasp-types possible for that patient, with the ultimate selection of a given hand configuration reflecting the specificity of their impairment (e.g., fine thumb control, spasticity, limitation of the thumb-forefinger opening …).

Finally, we propose that there are causal interactions between impairment and compensation at proximal (shoulder and elbow) and distal (hand and fingers) levels that determine the pose of the hand at the time of grasping. In hemiparetic patients, the hand is oriented more frontally, inclined downward with a variable axial rotation than that which is typical in healthy subjects (Roby-Brami et al., 2003b). Indeed, if a patient uses trunk compensation (flexion and internal axial rotation) or pathological patterns of shoulder and elbow coordination, the hand will be mechanically inclined downward, further constraining potential grasp configurations. Conversely, alternative grasping strategies may impose specific motor planning with atypical hand position and orientation relative to the object and, by consequence, an atypical reaching trajectory inducing deviant kinematic features. These close and complex functional interactions between the proximal and distal parts of the upper-limb may explain that, while cortico-spinal control is mainly distal (Maier et al., 2005), grasping is not more functionally impaired than reaching (Lang et al., 2005). The evolution of these prehensile compensatory strategies may account for the poor correlation between independent finger control and clinical tests of hand function in stroke patients (Raghavan et al., 2006).

In contrast to prehensile tasks aiming at stabilizing the object relative to the hand, manipulation requires the ability to move and rotate the object relative to the body, and supports the use of tools for actions on the environment. Manipulation imposes greater challenges to the sensorimotor system than grasping, including the anticipation of inertial forces and torques in response to variations in the position and orientation of the handheld object (Schneider et al., 2020). In-hand manipulation implies a particularly sophisticated form of manual dexterity where independent finger movements enable the displacement of the object with respect to the hand (Elliott and Connolly, 1984). A classification system has been adopted in order to describe a variety of actions during naturalistic activity in professional or household contexts (Bullock et al., 2013) as well as the use of prosthetic devices by amputees in their homes (Spiers et al., 2017). In daily life activities, performance of object handling tasks is conditioned by the characteristics of the individual and the disposition of the environment. Given the unstructured nature of daily life activity, effective motor solutions may be generated using any number of action sequences and postural configurations. Moreover, certain regularities in grasp transitions (e.g., top grasp to power grip) have been observed in daily life activities (Bullock et al., 2013), suggesting that use of a given hand configuration influences the subsequent prehensile activity. Even for highly repetitive assembly tasks, considerable variability in hand gestures can be observed across different actors (Brunet and Riff, 2009).

Generally speaking, the quantitative analysis of manipulation tasks in stroke patients has received relatively limited attention. Kinematic studies of drinking movements are one exception to this, and recommendations exist on the standardization of this task in order to provide reproducible clinical data (Alt Murphy et al., 2012; Kwakkel et al., 2019). Movement variables examined in these contexts remain, nonetheless, similar to those examined in reaching tasks, including movement duration, hand velocity and smoothness (Alt Murphy et al., 2012). Predictably, stroke patients with poor upper limb function tend to present with segmented velocity profiles and greater total duration for performance of the drinking task (Alt Murphy et al., 2012).

More recently, studies incorporating instrumented objects for measuring force exchanges during tasks involving object rotation have begun to provide additional perspective (Hermsdorfer et al., 2003). Our team designed an instrumented object to be easily manipulated by a person with an upper-limb movement disorder (Jarrassé et al., 2013). This device facilitates the analysis of the sequence of phases involved in object manipulation tasks (Martin-Brevet et al., 2017). Moreover, this method may also serve to highlight micro-errors occurring in action performance at the transitions between sub-goals (Seligman et al., 2014). For example, when using their hemiparetic arm, stroke patients experience greater difficulty with maintaining the vertical orientation of the handheld object, most notably in the transitions to/from a table (i.e. object lifting and object placement) (Parry et al., 2019). In addition, measurable “touch” and “push” errors observed in the form of force variations on lateral load cells prior to establishing grasp as well as increased downward force upon the object following placement. Ongoing research aims at characterizing how stroke patients perform object handling and regulate grip forces during prototypical rotational tasks (e.g., lifting a cup to the mouth, pointing a remote control).

Of course, in everyday life, most activities require some form of bimanual coordination. Moving with cooperative spatiotemporal precision, both hands have differentiated and specialized roles (i.e. the left hand holds and orients an object while the right hand performs an action on it) (Kantak et al., 2017). However, while there is a reasonable body of literature on bimanual organization through infant development (Fagard and Lockman, 2005) and primate evolution (Obhi, 2004), there is a paucity of literature on bimanual cooperative actions during daily life tasks in human adults. Existing studies on the subject have tended to focus most notably on the role of executive functions in movement planning (Gulde et al., 2019).

Despite their clinical interest, bimanual gestures remain largely unexplored in hemiparetic patients (Haaland et al., 2012). It is well known that stroke impairs both sides of the body since the less affected, ipsilesional side, also presents with weakness (Colebatch and Gandevia, 1989) and reduced hand dexterity (Cunha et al., 2017), generally proportional to the severity of the hemiparetic impairment (Maenza et al., 2020). Differential effects upon the ipsilesional hand are observed according to the cerebral hemisphere involved. Most notably, right sided lesions incur problems with visuospatial aspects of coordinated prehensile gestures while left sided lesions have greater effects upon planning and sequencing of the action sequence (Hermsdorfer et al., 1999).

The functional role of the less affected limb of stroke patients is controversial. On one hand, the patients may tend to use their less-affected hand for daily life activity as a compensatory strategy. As stated previously, this preferential use of the less-affected limb may inhibit implication of the hemiparetic counterpart in functional activity and thereby hinder recovery (Taub et al., 2006; Hidaka et al., 2012). It is often assumed that the improvement of the contralesional paretic arm by active rehabilitation will help patients in their (bimanual) daily activities (Johnson et al., 2011). However, some results suggest a limited transfer from unimanual training to bimanual activity (Johnson et al., 2011). In functional assessment of the upper limb during instrumental activities of daily living (IADLs), the use of both arms together favors performance when compared to modal use of the hemiparetic or less affected upper limb (Haaland et al., 2012). As proposed by Haaland et al. (2012) “rehabilitation therapy should focus on the ipsilesional as well as the contralesional arm.”

Taken together, these principles underscore how dexterity after stroke should be broadly understood as a skillful way to perform purposeful actions, either unimanually by using alternative reaching and grasping strategies or bimanually. Beyond the ability to carry out prototypical gestures with the hand and arm, dexterous function is something which engages both upper-limbs and likely the whole body.

Ideally, therapies should induce recovery at the level of the brain via cortical networks, once the extent of the lesion is stabilized following the acute period. Many promising neurobiologically inspired interventions have been proposed (non-invasive brain stimulation and neuro-technologies) but are yet to demonstrate their effectiveness in routine clinical practice (Sandrini and Cohen, 2013; Regenhardt et al., 2020). Regardless, activity-dependent neural plasticity remains the cornerstone of contemporary advances in rehabilitation practice (Nudo et al., 1996). As demonstrated by Nudo et al. (1996), following a lesion to the M1 cortical representation of the hand in monkeys, recovery of dexterous upper limb function occurred only among those monkeys who completed functional exercise by grasping food in feeding activities. The dimensions of the cortical representation for monkeys with no specific training program (spontaneous recovery) were found to diminish. This contraction of cortical maps supports the behavioral concept of learned non-use. Structural plasticity (synaptogenesis, axonal growth and branching) in regions proximal to the lesion (premotor area) was later demonstrated in animal studies (Dancause, 2006). Nudo’s observations are accepted as proof that plasticity of the cortical map is the neurobiological basis of true recovery. However, a video analysis of the same monkeys in a complementary article showed that some of them, in fact, used alternative grasping strategies (Friel and Nudo, 1998). In stroke patients, recovery through the subacute period is associated with changes in brain excitability, functional plasticity of cortical maps, and changes in connectivity in both hemispheres [review in Loubinoux et al. (2017)]. But the clinical consequences of these processes still remain unclear since functional neural plasticity does not necessarily lead to behavioral recovery (Buma et al., 2016; Peters et al., 2017). Structural plasticity probably contributes to improvements during long term rehabilitation but the relationships between the physical intervention and structural plasticity, and between structural plasticity and clinical outcome are still unclear [review in Sampaio-Baptista et al. (2018)].

In healthy subjects, the improvement of performance with repetition during sensory-motor learning is composed of two processes, occurring at different time scales: adaptation and skill learning (Kitago and Krakauer, 2013). Adaptation corresponds to the tuning of sensory-motor parameters to the actual situation after a relatively small number of repetitions; it relies on functional excitability changes and plasticity in the brain and cerebellum sensory-motor networks. Skill learning requires long-term practice, typically in a professional, sporting or artistic context, and can lead to structural changes in the central nervous system, as demonstrated in healthy subjects (Sampaio-Baptista et al., 2018). Contemporary rehabilitation methods inspired by motor learning paradigms are mostly based on the active repetition of meaningful movements, in contrast to classical neuro-developmental methods. The practice should be task specific, goal oriented, and motivating, with intense well-structured practice and provision of adequate sensory-motor feedback (Krakauer, 2006; Maier et al., 2019). Sensory-motor learning is often assisted by technology (e.g., virtual reality, robotics, and adapted video games) to provide more precise and standardized exercises and increased motivation thanks to engaging game design and user experience. However, it is still unclear if patients can truly recover after the acute, 3 month period of spontaneous recovery. Recent Cochrane meta-analyses showed that robotics and electromechanical devices could improve ADL, arm function and strength (Mehrholz et al., 2018) while the benefit of virtual reality was less convincing but significant when used in addition to standard care (Laver et al., 2017). Both studies underline the difficulties involved in evaluating the efficiency of these methods, as considerable variation is observed across interventions from one trial to another, and between the characteristics of the control intervention (in particular, “usual care,” which is still a “black box,” or matched intensity exercises).

The possibility of recovery at the impairment level is particularly debated. Negative findings could be due to a dose-effect (Winstein et al., 2019). The cumulated duration of training during usual trials is relatively low [18–36 h according to Ward et al. (2019)] and increasing the dosage up to 60 h improved the MAL but not function (WMFT) (Winstein et al., 2019). A recent study used a particularly intense schedule (300 h in 60 sessions with 5 h/day training) to compare three rehabilitation methods (robotics, functional electrical stimulation, and motor learning). The authors observed some significant and clinically relevant improvement at both impairment (Fugl-Meyer) and activity level, irrespective of the method used. Accordingly, a recent, non-controlled, retrospective study suggested that particularly intensive and prolonged therapies in stroke patients could lead to some improvement in the Fugl-Meyer score (Ward et al., 2019). Other studies suggest that training based on individual finger movements could induce some improvements that generalized to patient performance on Fugl-Meyer, hand function (ARAT) and activity (MAL) testing (Mawase et al., 2020). Similar approaches using highly specific hand training tasks (aiming, tapping, turning…) have also been found to improve manual dexterity when evaluated using the Fugl-Meyer test and hand activity test (TEMPA, Test d’Evaluation des Membres Supérieurs de Personnes âgées) (Platz et al., 2009; Platz and Lotze, 2018).

The most studied method to limit compensation is Constraint Induced Movement Therapy (CIMT) and its modified derivatives. The principle, proposed by Taub, is to impede the use of the less affected limb while soliciting the most affected limb with an intensive training program (Taub et al., 1999). This method is indicated only to patients who have already attained a certain functional threshold (clinically determined by taking into consideration factors such as partial recovery of wrist and finger extension) but has nonetheless provided a source of much hope (Sirtori et al., 2009). However, a recent meta-analysis of CIMT in chronic stroke patients reported “limited improvements in motor impairment and motor function, but that these benefits did not convincingly reduce disability” (Corbetta et al., 2015). In particular, a kinematic study showed no improvement in coordination during a 2D pointing task (Kitago et al., 2013). These inconsistencies could be due to individual differences, with some patients improving enough to use their limb, while others regressing following conclusion of the training period (Han et al., 2008). Improvement in daily life activity and self-reported arm use after constraint induced therapy could be attributed to the learning of new behavioral compensations, possibly involving both limbs and not to the recovery of the impairment [review in Kwakkel et al. (2015)].

When the use of the less affected limb is blocked during CIMT, people can still use compensatory motor patterns based on body redundancy, in particular the participation of the trunk. This is the equally true of conventional rehabilitation exercises, when the success of the task is only based on the displacement of the end-point. In contrast to classical rehabilitation methods such as Bobath (Levin and Panturin, 2011), these methods based on goal success seldom consider the quality of movement performance. As described in Section “Kinematics of Reach to Grasp Movements,” stroke patients may preferentially involve the trunk instead of exploiting shoulder and elbow rotation to displace the hand toward the target. The involvement of the trunk may be detrimental by inducing non-use phenomena of the upper-limb. Indeed, the limitation of trunk compensation by a physical restraint can improve shoulder-elbow coordination (Michaelsen et al., 2001). Limiting trunk compensation can thus “unmask latent potential recovery of upper extremity movement” (Wee et al., 2014). Training better movement coordination increases the efficiency of training as shown by the initial study (Michaelsen et al., 2006) and a meta-analysis which showed a moderate effect on the impairment and on the kinematics (Wee et al., 2014). These studies show that the quality of movement coordination during training is important and should be closely controlled either by trunk restraint, by Knowledge of Performance (KP) feedback given by the therapist (Cirstea and Levin, 2007) or by specific technology assisted KP, for example auditory feedback (Chen et al., 2016).

Some rehabilitation methods based on skill training aim at improving daily life activity, rather than reducing impairment through repeated movements (Winstein et al., 2016b). Rehabilitative task-oriented training can induce dosage-dependent improvements in reported motor activity (MAL) (Winstein et al., 2019). However, the effect of skill training is controversial since the interventions are difficult to systematize, and meta-analyses have shown only modest effects (French et al., 2010; Timmermans et al., 2010).

One difficulty is that very few studies have addressed the training of bimanual tasks, which are essential for daily life. A pilot study suggested that intensive training of the ipsilesional, less-affected, limb could improve its dexterity (Jebsen test) and could generalize to functional independence (Sainburg et al., 2016). Several studies developed symmetrical bilateral arm training with the perspective of assisting the paretic limb by the less affected limb thanks to interlimb coupling (Whitall et al., 2011), however, a meta-analysis did not show any clear neural or behavioral effects (Choo et al., 2015). Surprisingly, there are very few studies of bimanual rehabilitation methods with more functional, asymmetric manipulative tasks in hemiparetic patients despite their relevance for daily life activity. A better understanding of the use of both hands during manipulative actions is needed, particularly the effect of laterality and handedness.

Individual behavior during ADL according to the mantra “use it and improve it or lose it” is probably a key to better understanding of why some patients above a certain functional threshold continue to improve during follow-up while other regress (Hidaka et al., 2012).