Thirty-five patients, 18–45 years of age, who met DSM-IV TR criteria (42) for schizophrenia were recruited at the Resource Center for Cognitive Remediation and Psychosocial Rehabilitation (C3RP), Sainte-Anne Hospital, Paris. Patients were clinically stabilized and under psychotropic medication for at least 1 month. Exclusion criteria: substance abuse/dependence, neurological disorders, participation in other cognitive remediation programs, resistance to neuroleptic treatment, electroconvulsive therapy in the previous 6 months, piano playing for a number of years.

Patients passed clinical and neuropsychological assessments, and comprehensive testing of manual dexterity. Twenty healthy age-matched subjects served as control group for the dexterity assessment. Table 1 lists clinical and demographic information. The study, approved by the local ethics committee (study no. 2011-A00454-37, Comité de Protection des Personnes, Ile de France 3), complied with the Declaration of Helsinki. Subjects provided written informed consent.

Cognitive remediation therapy (43–45), to alleviate dysexecutive impairments was completed by 15 patients. It lasted 14 weeks (40 sessions of 60 min each: 2 sessions/week in the C3RP, 12 sessions at home) and comprised exercises on executive functions including attention, cognitive flexibility, planning, and memory.

Clinical evaluation (Table 1) comprised PANSS (46), NSS (22), and the Simpson-Angus Scale for rating abnormal movements (47).

Executive Function (Table 2) was assessed by selective attention [D2 test (48)], working memory [digit and spatial span of Wechsler Adult Intelligence Scale-III, WAIS-III (49)], word processing speed (digit symbol-copy of WAIS-III), cognitive flexibility [Wisconsin Card-Sorting Test (WCST) (50, 51)], action planning [zoo map test (52)], problem solving [D-KEFS Tower Test (53)], and inhibition [Stroop (54, 55)]. The above neuropsychological tests have been shown to be reliable and valid in schizophrenia (i.e., with test–retest correlation R > 0.7 or Cronbach’s alpha > 0.7): selective attention, D2 test (56); WAIS-III (57); WCST (58); action planning, zoo map test (59); problem solving, D-KEFS Tower Test (60); Stroop (58).

Finger movements were measured with the FFM,¹ as previously described (40). Individual forces (of the index, middle, ring, and little finger) were sampled to a CED1401 (10 kHz sampling rate/digit) under Spike2v6,² which provided real-time visual display of digit forces, target instructions or target forces. The Moberg pick-up test (61) gave a complementary functional measure of dexterity.

Task performance was analyzed using MatlabV7.5 (MathWorks Inc., Natick, MA, USA). Force signals were smoothed (5 ms sliding window) and down-sampled to 100 Hz. The following task-specific performance measures were extracted trial-by-trial.

Respective measures were averaged/summed across trials and conditions.

Student’s t-test and Mann–Whitney U-test were used to test for group differences in parametric and non-parametric single-level variables. FFM measures were analyzed using repeated measures ANOVAs (Table 3). To extract individual profiles of dexterity we first selected the most discriminant “key-FFM” score (Table 3) in each task from group comparisons and then calculated individual z-scores for each key-FFM score, based on the control group’s performance: a z-score > 2 (>mean + 2SD, one-tailed) was considered out of normal range (i.e., a deficient score). A composite score (sum of the four key-FFM z-scores) was used for computing a receiver-operating curve to discriminate patients from control subjects based on the degree (or severity) to which dexterity was affected.

Spearman’s rank order or Pearson’s correlations were used to investigate relations between key-FFM performance measures and clinical or neuropsychological scores, and antipsychotic medication. Remediation effects on clinical/neuropsychological scores were tested with the Wilcoxon test. Statistical analysis was performed under Statistica©³ and level of significance set to P < 0.05. In correlation tests, significance level was corrected for multiple comparisons according to Benjamini and Hochberg (62).

Patient and control groups were similar in gender and age (Table 1). However, patients were significantly slower than control subjects in performing the pick-up test (Table 1, T = 2.74, P = 0.008), suggesting a qualitatively decreased level of dexterity. Full clinical details and neuropsychological scores are given in Table 2.

Patients with schizophrenia showed altered finger force control during force tracking (Figures 1B,C). Patients had significantly higher tracking error compared to controls (SZ_patients: 0.20 ± 0.06 N; controls: 0.14 ± 0.04 N; GROUP effect: F = 14.9, P = 0.0003; Figure 2A) in all conditions (FORCES, PHASES). Force variability (CV) was also significantly increased in patients (SZ_patients: 2.86 ± 1.08; controls: 2.02 ± 0.56; GROUP effect: F = 4.6, P = 0.04). No significant differences were found in timing (force onset and release onset) or in release duration.

Patients (N = 33; two patients were excluded due to data acquisition problems) showed a significantly decreased number of correct taps over the entire task compared to controls (SZ_patients: 4.04 ± 0.63; controls: 4.68 ± 0.25; GROUP effect: F = 18.9, P = 0.00007). This difference was even greater during recall (SZ_patients: 3.73 ± 1.00; controls: 4.74 ± 0.30; GROUP*PHASE effect: F = 6.47, P = 0.002; post hoc test P3: P < 0.000001; Figure 2B). During task progression, patients and controls increased the number of correct taps (TRIALS effect: F = 12.7, P < 0.000001; Figure 3A). When splitting the task into three consecutive phases [first (P1) and second (P2) half of the learning phase, and the recall phase (P3)], patients started the task with a significantly reduced number of correct taps compared to controls (SZ_patients: 3.89 ± 0.68; controls: 4.47 ± 0.48; GROUP/PHASE effect: F = 6.47, P = 0.002; post hoc test P1: P = 0.002; Figure 3B). Both groups increased their number of correct taps during the second learning phase (SZ_patients: 4.48 ± 0.56; post hoc test P1/P2: P = 0.000001; controls: 4.84 ± 0.20; post hoc test P1/P2: P = 0.013), and patients reached a similar success rate as controls (somewhat lower, but statistically non-significant, post hoc test P2: P = 0.06). During recall, controls maintained the same number of correct taps (controls: 4.74 ± 0.30; post hoc test P2/P3: P = 0.49) while the performance of the patients decreased to levels seen in the initial (P1) learning phase (SZ_patients: 3.73 ± 1.00; post hoc test P2/P3: P = 0.000001). Both groups made the same kind of errors during the task and patients had a significantly longer trial duration during recall compared to controls (SZ_patients: 2,674 ± 693 ms; controls: 2,027 ± 480 ms; GROUP effect: F = 12.9, P = 0.00001).

In the single-finger-tapping task, the average tap frequency was similar in patients and in control subjects, close to the 1, 2, and 3 Hz target frequency, during the cued as well as during the non-cued phase. However, patients showed a significantly increased tap interval variability compared to controls (SZ_patients: 94.5 ± 50 ms; controls: 60.1 ± 17.2 ms; GROUP effect: F = 8.8, P = 0.004; Figure 2C), and this difference was similar in all conditions (FINGER, FREQUENCY, PHASE). There was no group difference in other task measures (erroneous taps).

In the multi-finger-tapping task, patients showed a significantly reduced degree of finger individuation compared to controls (SZ_patients: 79 ± 20%; controls: 94 ± 6%; GROUP effect: F = 9.0, P = 0.004; Figure 2D). This was the case in all fingers and for single- or multi-finger tap configurations. Patients also had significantly longer dual-tap intervals (SZ_patients: 380 ± 18 ms; controls: 304 ± 24 ms; GROUP effect: F = 6.6, P = 0.01).

Individual dexterity profiles are shown as radar plots of the four key-FFM measures (Figure 4). Four types of profiles were distinguished according to the number of affected z-scores in each patient. “Profile_0” (N = 7 patients) with all z-scores <2 (i.e., non-affected), “profile_1” (N = 13) with one affected z-score >2, “profile_2” (N = 4) with two affected scores, and “profile_3” (N = 9) with three or four affected z-scores. Sequence recall was the most frequently affected component (N = 16), followed by degree of individuation, ramp error, and tap interval variability. Profile_0 showed homogeneous radar plots (Figure 4A) with all z-scores <2 (below pathology threshold). In profile_1 the majority of the patients had affected z-scores close to threshold (3 > z-score > 2; Figure 4B). In profile_2 and profile_3 (Figures 4C,D), most of the affected z-scores were >3 or >4, resulting in heterogeneous, large amplitude patterns. Qualitatively, patients with a large number (>2) of affected scores also showed highly abnormal values (z-score > 3; Figure 4). Furthermore, dexterity profiles showed strong heterogeneity with only 15% of patients showing abnormal values in a common set of components.

The heterogeneity among dexterity profiles suggested that each key-FFM component represents a specific element of control, independent from the other components. A quantitative estimate of their independence was obtained by pair-wise Pearson correlations: this showed that only one among the six pair-wise correlations was statistically significant, i.e., tap interval variability vs. degree of individuation (r = 0.54, P < 0.05, Dof = 32). This lack of systematic covariance between key-FFM scores indicates a large (but not perfect) independence among component measures.

The composite dexterity measure (sum of the four key-FFM scores) was analyzed using receiver-operating characteristics (ROC) to test for sensitivity and specificity (Figure 5). A log value of 1.2, corresponding to a score = 3.3, yielded best discrimination between patients and controls, with a positive predictive power = 0.93 and a negative predictive power = 0.9. In comparison, ROC analysis of each single key-FFM score provided lower discriminative power. Similarly, using the NSS total score as a means of discrimination (mean = 5 ± 2, threshold = 9, obtained from (22)) resulted in weaker positive predictive power (0.65).

For patients with schizophrenia, correlations were computed between key-FFM scores and clinical (PANSS; NSS) and neuropsychological (D2, WAIS-III) outcomes. Briefly, FFM sequence recall scores correlated with PANSS total scores (Figure 6A; r = 0.53) and with PANSS disorganization scores (r = 0.55, P < 0.05, Dof = 30). FFM tracking error correlated with NSS motor coordination (Figure 6B; r = 0.61), with NSS sensory integration (r = 0.52) and with NSS motor integration (r = 0.53, P < 0.05, Dof = 28). FFM tap interval variability did not show any correlations. FFM degree of individuation correlated with NSS sensory integration sub-score (r = 0.50, P < 0.05, Dof = 28), with different sub-scores of the D2 test of attention (Figure 6C; D2 GZ (r = −0.61), D2 KL (r = −0.63) and D2 GZ-F (r = −0.62), P < 0.05, Dof = 25), and with the WAIS-III working memory score (Figure 6D; r = −0.52, P < 0.05, Dof = 26). No relation was found between FFM measures and the scores of Stroop, D-KEFS Tower or BADS.

The chlorpromazine equivalent (CPZe) correlated positively with the composite dexterity measure (r = 0.50; P = 0.005; Dof = 28), but not with single key-FFM scores, except for ramp error (r = 0.47, P < 0.05, Dof = 28).

Additional multiple regression analyses showed that correlations between each of the three key-FFM scores (degree of individuation, tap interval variability, and sequence recall) with clinical and neuropsychological outcomes (Table 3) remained significant even with CPZe as a covariate, indicating that antipsychotic medication could not completely explain these relations. Furthermore, two different key-FFM scores (tap interval variability and individuation) were more affected in patients with prescribed anxiolytic medication (in 26% of patients), but not with mood stabilizers, antidepressants, or anticholinergic medication.

In the post-remediation condition, ANOVA showed a significantly decreased ramp error during force tracking (T1/T2 effect: F = 8.86, P = 0.009; Figure 7A), and also a significantly increased sequence recall score during sequential finger tapping (T1/T2 effect: F = 10.26, P = 0.005; Figure 7B). Remediation did not lead to significant changes in tap interval variability or degree of finger individuation.

Changes in clinical and neuropsychological scores following remediation are listed in Table 2. In particular, PANSS total score (Wilcoxon Z = 2.95, P = 0.003) and sub-scores (Negative: Z = 2.73, P = 0.006; General: Z = 2.73, P = 0.006; Disorganization: Z = 2.86, P = 0.004) decreased significantly after remediation, except for the Positive subscale. The total NSS score also decreased significantly after remediation (Z = 2.04, P = 0.04) but not the sub-scores. Among the neuropsychological scores, those related to attention (D2, Stroop) improved significantly.

In terms of force tracking, which requires visuomotor matching and fine on-line adjustments of low finger forces, clear deficits were found in patients with schizophrenia: they produced about 40% more error than controls, similar to deficits found in controlling power grip forces (30) and consistent with less accurate control of steady grip (32) and finger flexion force (28). However, grasp function (evaluated by the ratio of grip/load force) was not or only marginally affected in schizophrenia (37, 63), and when affected, it was considered a side effect of antipsychotic medication (10). There is thus clear evidence for a deficit in voluntary and visually controlled force modulation, but less so for force control during grasp. Visuomotor force tracking requires real-time sensorimotor mapping and integration, and the patients’ precision in this task correlated positively with NSS sub-scores in motor coordination, motor integration and sensory integration. This validates that these NSS sub-scores reflect lack of precision (clumsiness) in tasks requiring sensorimotor integration. Interestingly, visuomotor precision (ramp error) was the only key-FFM score to correlate with antipsychotic medication (CPZe).

The sequential finger-tapping task required successive activation of fingers to produce movement sequences in the context of procedural learning: patients with schizophrenia performed significantly less well than controls, in particular during recall. Nonetheless, during learning patients reached a level similar to control subjects, suggesting that altered memory processing and not learning might be the cause of this deficit. These data agree with previous results on motor sequence learning involving finger-thumb opposition in schizophrenia (64). However, motor sequencing is more generally affected, as shown by studies on grip control (38), planned grasping (63), and other tasks (e.g., NSS (65)). Sequence recall correlated with the PANSS (global and disorganization sub-score). This (rather weak) association (in line with (66)) might not be surprising, since cognitive aspects contribute far more to PANSS scores than sensorimotor components or deficient dexterity (67). The lack of correlations between the FFM recall score and working memory scores (WAIS-III) might be related to differences between cognitive and procedural learning (68, 69).

The finger-tapping task required primarily temporal motor coordination. Deficits in finger-tapping tasks, such as reduced maximal tapping speed [(18, 68, 70, 71), but not in Ref. (72)], higher than required tapping frequency (73), and higher tapping variability (5, 73) have been previously shown. We found increased tapping variability, but not increased tapping frequency, whether cued or not (most likely due to our more constrained task conditions). Finger-tapping performance has been considered as a potential endophenotype, with a typical performance gradient that increased from patients, to unaffected relatives and finally to control subjects (18, 70, 71). However, timing, as assessed with the FFM task, did not correlate to any of the clinical/neuropsychological tests or to antipsychotic medication.

The degree of finger individuation (assessed during multi-finger tapping) was significantly lower in patients with schizophrenia. Although finger individuation is a key feature of manual dexterity (74), to our knowledge, this has not been previously quantified in schizophrenia. However, the Purdue Pegboard test, which requires some (non-quantifiable) degree of finger independence, has repeatedly been used and might be compared to our degree of finger individuation: schizophrenia patients showed lower Pegboard scores than controls (31, 39, 72) and deficient scores were related to social functioning (38). In our case, degree of finger individuation correlated with sensory integration (NSS sub-score), with attention (D2 score), and with working memory (WAIS score). Tentatively, this may be interpreted as common cognitive operations involved in task-related finger individuation, requiring (i) visuomotor mapping (NSS) to determine the digits to be moved (and those not to), (ii) working memory (WAIS-III) to maintain this mapping, and (iii) focused attention (D2) since targets changed unpredictably. The lower scores in these basic cognitive operations may (in part) explain the resulting deficit in finger individuation, and are in line with neurocognitive deficits observed in schizophrenia (1, 75). This does not exclude that the motor command itself or its transmission via the corticospinal tract may be perturbed (76, 77).

Our sample of stabilized patients may have masked even stronger deficits in dexterity, likely to be present in more severely affected (refractory or acute phase) patients with schizophrenia (78). A larger sample may also permit identification of subgroups of dexterity profiles and investigation of their relation to neuropsychological/clinical subtypes [e.g., Ref. (79)]. The absence of drug-naïve patients prevented us from providing direct evidence against an antipsychotic medication effect on the dexterity scores. There is, however, indirect evidence (see below) and evidence from other studies against this assumption (28–31). Similar limitations (plus a significant dropout rate) affected the remediation protocol. Some of these issues and others, such as the specificity of deficient dexterity components, their potential use as trait or state markers, or their occurrence in “high risk” populations, might be addressed in future studies.

A number of findings speak against antipsychotic medication as a main cause of deficient dexterity in schizophrenia. First, ramp error decreased after cognitive remediation, while antipsychotic medication (CPZe) remained constant. Second, CPZe only correlated with ramp error but not with motor sequence recall, timing, or independence of finger movements. Third, the observed correlations between these three dexterity scores and clinical/neuropsychological scores (Table 3) were robust when adding CPZe as a covariate. This is consistent with previous studies on upper limb deficits vs. antipsychotic medication in schizophrenia (28–31).

Our results provide quantitative evidence at the group level that dexterity is affected in schizophrenia, that four different components of dexterity can be distinguished, that each is significantly affected, and that these components are largely independent of each other. Whether the heterogeneous profiles at the individual level will cluster into subtypes and correlate with clinical phenotypes in schizophrenia [e.g., Ref. (79)] needs further investigation. Discrimination between patients and controls using ROC showed that affected manual dexterity provided a measure with a sensitivity of 79% and a specificity of 90%. Thus dexterity, quantified with FFM, showed a better discrimination than total NSS score and showed similar predictive accuracy compared to gaze deficits (80, 81). Therefore, manual dexterity may well be considered a potential clinical marker in schizophrenia.

After cognitive remediation patients showed an improvement in main clinical and neuropsychological scores. Moreover, patients also improved significantly in two of the four main FFM scores, suggesting that these measures are sensitive enough to detect small variations in manual dexterity, only in part captured by NSS. Crucially, the remediation protocol did not include any FFM training, but used more general aspects of sensorimotor training. This suggests that tracking error and sequence recall, which varied as a function of remediation, may be more state-related [similar to NSS (82)], whereas degree of individuation and tap variability, remediation invariant, may be considered potential trait markers.

In conclusion, our results are consistent with deficits in manual dexterity being a genuine phenotype of the underlying pathophysiology in schizophrenia. Our data suggest that quantitative assessment of these dysfunctions may serve as useful clinical markers in schizophrenia.