SIR- Prehensile force control while grasping and lifting an object has been studied fairly extensively in children with hemiplegic cerebral palsy (CP). Children with hemiplegia have a decreased ability to couple the grip (squeeze) and load (vertical tangential) force during lifts (force synergy)1,2 and scale the amplitude of the force development in their involved hand.3 The decreased abilities reflect unilateral impairments in anticipatory control,3,4 which are reduced when lifting common objects (e.g. a cup) and following extensive practice with novel ones.4,5 Although lack of force coupling is also observed during pulling actions,6 the extent to which anticipatory control is impaired during well-practiced dynamic tasks is largely unknown. Walking with a hand-held object is a well-practiced dynamic action that requires precise coupling of fingertip forces in an anticipatory manner. During this task, the inertial forces (IF) acting on the hand-held object oscillate in a sinusoidal manner during the gait cycle, with the maxima occurring shortly after contact of the foot with the ground (initial contact), and the minima in mid-stance.7 In typically-developed individuals the grip force (GF) is actively timed and modulated simultaneously to the resulting IF oscillations.7–9,10 This force coupling reduces the need to maintain high grasping forces throughout the task (potentially leading to fatigue) by ensuring that the GF/IF ratio is sufficient at critical points (IF maxima) to prevent slips. The coupling is preserved across a variety of velocities and stepping/object accuracy constraints.7,8 It is not known whether such anticipatory fingertip force coupling is present in the involved or non-involved hand of children with hemiplegic CP. In the present study we tested the hypothesis that similar to unilateral anticipatory control impairments during lifts,3,4 the anticipatory force coupling in the involved hand while walking with a hand-held object is impaired. Eleven children (four males, seven females; age range 5–10y, mean age 7.6y) with a diagnosis of hemiplegic CP (six right, five left hemiparesis) participated in the study. The participants consisted of a sample of convenience taken from our database of individuals with hemiplegic CP who had participated or were interested in participating in hand intervention studies. The sample consisted of participants with the age representing the general age in which both anticipatory fingertip force control during lifting11 and force coupling during cyclical object movements12 are fairly well-developed in typically developing children. All were classified according to the Manual Ability Classification System13 as level I (n=4) or II (n=7). All participants walked without assistive devices and were classified as Gross Motor Function Classification System14 level I. Dexterity/speed was assessed using the Jebsen–Taylor test of hand function15 measured in seconds. On average, the participants required 177.7 seconds (SD 105.7) to complete the Jebsen–Taylor test using the involved hand and 38.4 seconds (SD 8.7) using the non-involved hand. With vision occluded, either one or two points of the DiskCriminator16 were placed perpendicular to the thumb and index fingerpads. The minimal distance participants could distinguish two discrete points in seven out of 10 trials (averaged across the index finger and thumb) was recorded.17 The mean two-point discrimination overall was within the normal range17 (mean 4mm [SD 2] in the involved hand; mean 3mm [SD 1] in the non-involved hand). All participants were required to be able to walk with the grip instrument without dropping it. Informed consent was obtained from all participants and their caregivers. The study was approved by the Teachers College, Columbia University Institutional Review Board. The experimental set-up and procedures have been reported in detail elsewhere.7 Briefly, the grip instrument (165g) consisted of two 3D force transducers (Nano F/T, ATI Industrial Automation, NC, USA). The transducers (diameter 1.7cm, 4.5cm apart, 0.05N resolution) were attached above the surface of a rectangular base (7.8 × 4.3 × 4.7cm) (Fig. 1a). The force transducers measured the GF normal to the grip surfaces, and the vertical load and horizontal forces tangential to the grip surfaces. The participants walked on a 9m level walkway. Average velocity was determined using kinematic analysis of markers placed on the heels of their shoes (VICON, Oxford Metrics Ltd., Oxford, UK) for eight participants and a stopwatch for three participants. Matlab 7.0 was used to produce an auditory start signal also used to synchronize the Vicon and force data. (a) Grip instrument (oblique view) with two force-transducers, arrows indicating the direction of grip forces (GF1 and GF2), load forces (LF1 and LF2), and horizontal forces (HF1 and HF2). Note: not drawn to scale. The net resultant of the load and horizontal forces was calculated to derive the inertial force (IF). (b) Stick figure depicting experimental set-up with participant walking on level 9m walkway holding grip instrument. (c) Grip force, inertial force, and grip and inertial force rate streams from a representative participant’s involved and non-involved hands seen while walking with the grip instrument. Vertical lines represent the time of initial contact of the foot. After washing their hands, participants stood at the beginning of the walkway and held the grip instrument between the thumb and index finger. They were instructed to hold the object upright and forward-oriented with their elbow flexed at approximately 90 degrees (Fig. 1b), and to start walking at a comfortable, self-selected speed after the start signal. After several practice trials, participants performed one walking trial with the grip instrument in the involved hand and another with the instrument in the non-involved hand (counterbalanced across participants). For comparison, participants were also asked to grasp and lift the object with each hand for 5 seconds without walking (‘standing trial’). Force data were sampled at 400 Hz (low-pass filtered at 6Hz) using SC/ZOOM (Department of Physiology, Umeå University, Sweden). Kinematic data were sampled at 120 Hz. The GF and IF forces were determined to be nearly identical in each digit indicating the object did not experience out of plane movements.10 Thus the averaged GF from each transducer and the IF (sum of the resultants of the horizontal and vertical load force at each force transducer) were measured. Force derivatives (dGF/dt and dIF/dt) were obtained using a SD 5-point numerical differentiation (12.5s moving average). To evaluate grasp control during steady-state walking a time window between the third initial contact after gait initiation and the third to last initial contact before gait termination was used. This allowed seven or more steps to be analyzed. The following measures were taken across this time period: (1) average gait velocity and; (2) peak-to-peak IF changes to ensure any differences in coupling are not simply due to differences in walking or the resulting IF changes for each hand; (3) time of maximum IF relative to initial contact (when the heel-marker transiently ceased moving in the vertical direction or the direction of progression) to determine the relationship between gait and IF; (4) GF/IF ratios throughout the steps analyzed and; (5) at the peak IF as a measure of grip force efficiency, and; (6) peak cross-correlation coefficients (r), and; (7) their corresponding time lags (ms) for the GF and IF rates across the analyzed steps to quantify the relationship between these measures. For the standing trial, the mean GF/IF ratio was measured during the middle 3 seconds. Cross-correlation coefficients and Pearson correlation coefficients for each participant were converted to Fisher’s z-scores18 to calculate means and for statistical analyses. The two conditions (involved and non-involved hand) were compared in all dependent measures using a paired t-test. Changes in the average GF/IF ratios between standing and walking were measured using a 2 (hand) × 2 (hold/walk) repeated measures analysis of variance (anova). Changes in the GF/IF ratios at the IF peak were compared with the mean ratios throughout the analyzed steps using a 2 (hand) × 2 (average/peak) repeated measures anova. Non-parametric statistics (not reported) yielded qualitatively identical results. Statistical significance was considered at the p<0.05 level. Fingertip forces and their derivatives during locomotion from a representative participant (involved and non-involved hand) are illustrated in Figure 1c. As seen in the figure and consistent with earlier studies,7–9 the IF fluctuate in a sinusoidal manner with the peaks occurring shortly after initial contact (vertical lines). Importantly, the IF and GF oscillate largely in parallel in the non-involved hand, whereas in the involved hand these forces do not. Table I shows the analyzed variables across all participants. The average gait velocity, and the resulting peak-to-peak IF difference while walking with the object in each hand, were nearly identical. The average duration between initial (foot) contact and the subsequent peak in IF was 101 millisecond and 135 millisecond for the involved and non-involved hand respectively. The average peak cross correlation (r) in the involved hand was significantly lower than in the non-involved hand (p<0.001), indicating a weaker force coupling. Furthermore, they had significantly longer cross-correlation time lags in the involved hand (positive values mean GF followed IF; p<0.001). When the temporal offset established with cross-correlation is not considered (i.e. at zero phase lag), the relationship was even weaker (r=0.31 in the involved hand; r=0.59 in the non-involved hand; p<0.001). The average GF/IF ratio in the involved hand was not significantly different than the ratio in the non-involved hand during walking and standing. However, the pattern in each hand differed, whereby the GF/IF ratio increased 54% from standing to walking in the involved hand but decreased 19% in the non-involved hand (hand × task interaction, F=7.86, p<0.05; Table I). As compared with the average throughout the gait cycles, at the point of maximum IF, the GF/IF ratio decreased significantly more in the involved hand (21%) than in the non-involved hand (11%; hand × task interaction, F=5.99, p<0.05; Table I), resulting in similar ratios at that point. Age was significantly correlated with the cross-correlation coefficient in the involved (r=0.73; p<0.01) but not the non-involved hand. Two-point discrimination of the involved hand was moderately correlated with the cross-correlation coefficient (r=0.67; p<0.05), but not the non-involved hand or with the GF/IF ratio in either hand (p>0.05). The Jebsen–Taylor test of hand function was not significantly correlated with the cross-correlation coefficients in either hand (p>0.05). Overall we found impairments in the grip-inertial force coupling in the involved hand compared with the non-involved hand. Specifically we found moderate correlations between the GF and IF that were closely linked in time for the non-involved hand, but not for the involved hand. Although the time lag in the latter case (57ms) is below the 60–90 millisecond time to be considered anticipatory,10 the weak relationship (r=0.39) means that this lag may be considered arbitrary. Thus, similar to more discrete tasks,1,4,5 there are impairments in anticipatory control that are specific to the involved hand while engaged in a routine, familiar task even for this group of participants with mild/moderate hand impairments. In order to maintain appropriate GF/IF ratios as the IF oscillates during gait, the GF must either overall be very high or must fluctuate in parallel with the IF7,19 since tracking the IF using feedback would involve delays that could result in drops. In a grip-lift task in healthy adults,20 the dynamics of a task along with the threshold of load change influences the GF safety margins. Thus participants could conceivably slow down while walking with the object in the involved hand to reduce the IF oscillations and the amount of coupling required.7 However, participants walked with similar gait velocities and had similar resulting peak-to-peak IF oscillations with the object in the involved and non-involved hand respectively. Participants showed similar force ratios in both hands during the standing trials, but increased the force ratios in the involved hand when walking compared with standing. This strategy of elevating the GF/IF ratios by employing overall high GF could partially alleviate the requirement of force coupling, but could lead to fatigue, especially given their paresis. Children with hemiplegia have been shown to have ‘global planning’ impairments that are independent of the effector used for tasks such as grip orientation.21–23 So how could such planning deficits be lateralized to the involved hand for the current study (e.g. effector-dependent)? The type of planning deficit may be task-dependent. Although force coupling is anticipatory,7–9,19 sensory information is used to establish and update representations of object properties and body dynamics for modulating the fingertip forces. The children with hemiplegia had varying levels of sensory impairments in the involved hand, but not the non-involved hand, as indicated by two-point discrimination (albeit an imperfect measure). These were moderately correlated with coupling strength. Although studies using digital anesthesia in healthy adults demonstrate undisrupted fingertip force coupling during movements of a hand-held object,24 individuals with complete peripheral deafferentation have varying degrees of impaired force coupling.25 Similar lateralized impairments are seen in the anticipatory scaling of fingertip forces based on sensory information signaling object properties during prior discrete lifts with the involved, but not the non-involved hand.26 Interestingly, sensory information obtained during lifts with the non-involved hand may be transferred for subsequent anticipatory fingertip force control with the involved one,26 suggesting the impairments are not simply due to motor execution problems. Conversely, despite a lack of anticipatory control in the involved hand, the sensory information obtained is sufficient to transfer for subsequent anticipatory control of the non-involved one.27 Thus, the impairments are not sensory-based per se, but instead reflect a lateralized impairment in the ability to integrate prior sensory (potentially reduced) information and current motor signals for anticipatory force scaling in the involved hand. Such lateralized planning impairments involving the use of sensory information to update representations of object properties and limb dynamics suggests impaired sensorimotor integration underlies the effector-specific findings of the present paper. These differential planning behaviors may have distinct neural bases. For example, planning tool use involves the left inferior frontal, inferior parietal, and posterior temporal cortices independent of the hand involved (i.e. global planning).28 In contrast, anticipatory coupling of grip and inertial forces, which appears to be effector-dependent, has been shown to involve the parietal cortex and posterior lateral cerebellum.29 Both systems appear to be affected in hemiplegia, with the latter being exclusive to the involved hand. The strength of the force coupling (r=0.64) in the non-involved hand was slightly lower than typically reported in adults (r=0.7–.9).7–9 This could reflect subtle impairments in the non-involved hand.4 However this is unlikely since the non-involved hand’s Jebsen–Taylor test of hand function30 time was within the normal range for the participant’s age. Furthermore, the strength of the coupling is dependent on the magnitude of IF changes,7,8 which are lower than in adults given reduced walking velocities and body mass in children (i.e. lower vertical ground reaction forces associated with initial contact). The object’s mass was also approximately half that as used previously in adults to increase the likelihood they could maintain their grip. Thus the coupling is within the expected range. Blank et al.12 studied the typical development of grip force modulation during repetitive vertical arm movements with a hand-held object. They found that temporal coupling is well-established by 4 years of age but that the gain modulation continues to develop. We did not find a correlation between the coupling and age in the non-involved hand, but did for the involved hand, suggesting coupling strength for the involved hand may improve during continued practice associated with development. A limitation of the study is that the sample was fairly small and homogeneous (Manual Ability Classification System levels I and II, Gross Motor Function Classification System level I), and thus we do not know if the coupling would be affected in both hands in individuals with greater gait impairments. To our knowledge this is the first study to examine the coordination of grasp and whole body movements in children with hemiplegic CP. The results suggest that in addition to impairments in grasp31 and gait,32 there are impairments in the coordination of the forces associated with the two motor behaviors. Thus while rehabilitation efforts may be directed individually to grasp and gait impairments, they should not solely focus on treating body parts, and instead include manual dexterity in a context suitable for independence in activities of daily living, including activities requiring coordination across the extremities such as walking with an object. This work was supported by a grant from the Thrasher Research Fund. The funding sources did not have any role in study design, data collection, analysis or data, or manuscript preparation. The corresponding author (AMG) had full access to all data and the final responsibility for the decision for submission. We thank the children and their families for participating, and Celine Crajé for helpful comments and Claudio Ferre for assistance with statistical analyses.