Tactile stimulation of the right-hand index (D2) and ring (D4) finger was performed using a piezo-electric stimulation device (Piezostimulator, QuaeroSys, St. Johann, Germany) consisting of a control unit and two connected, custom-built stimulation modules. Each module was equipped with a 17-pin radial display (Figure 1A) consisting of one central pin surrounded by two 8-pin circles of radius 2.5 and 5.0 mm, respectively. Each pin could be controlled individually.

By simultaneously elevating up to 5 pins arranged on a straight line through the central pin, 4 different orientations could be presented (Figure 1B): parallel to the finger axis (0°), orthogonal to the finger axis (90°) and the diagonal orientations in between (45° and 135°). The stimulation displays were positioned below the fingertips such that the parallel pin orientation was oriented along the proximal-to-distal fingertip axis and the central pin was located slightly distal to the fingertip vortex. Subjects were instructed to keep their right hand relaxed and pronated throughout the experiment. Each orientation presentation lasted for 1 s and consisted of 10 pin-raising cycles (stimulus duration = 50 ms, inter-stimulus interval = 50 ms), resulting in a stimulation frequency of 10 Hz. Pins were set to maximum drive-out (1.5 mm, if no weight was applied onto them). Orientation stimuli were generated by raising only 4 pins randomly chosen from the 5 pins forming the given orientation (Figure 1C). This procedure was applied to prevent subjects from solely concentrating on individual pins for solving the task.

Stimulation and recording of responses were each controlled using Apple Macintosh computers running in-house real-time stimulation and data-acquisition software.

Per session, 40 blocks of 10 trials each were acquired. The first trial of each block started 2 s after cue presentation. Trials within a block were separated by an inter-trial time of 1 s. The paradigm is illustrated in Figure 2.

In each trial (Figure 2C), independent random sequences of distractor orientations (oblique) were presented simultaneously at the right-hand index (D2) and ring (D4) finger (Figure 2A). The number of distractor presentations (3–15, mean of 6) was gamma distributed ~Γ(7.5, 0.8). Distractor stimuli were separated by 100 ms. At some point, a stimulus parallel or orthogonal to the finger axis (the “target”) was presented at one of the locations. The subjects were instructed to give a speeded response by pressing the foot pedal. Upon response (if within the 1 s stimulus interval) or 100 ms after target presentation, a mask stimulus was presented at both locations for 1 s, generated by repeated presentation of every second pin of the 17 pins.

In order to guide attention, each block started with a tactile cue (3 s duration), which was always presented at the same location. Half of the subjects received the cue at D2, the other half at D4 (Figure 2B). The cue was location-informative, as the targets were displayed at the cued finger in 60% of the trials (Figure 2D). Targets at this finger were therefore called location-matched, whereas targets at the other finger (only 40%) were called location-unmatched. The cue was of either parallel or orthogonal orientation (generated by elevation of all 5 pins) and orientation informative for location-matched targets, as these had the same orientation as the cue in 90% of the trials; for location-unmatched targets, the cue was non-informative, with parallel and orthogonal targets being equally likely (Figure 2D). A target was referred to as orientation-matched/-unmatched, if its orientation agreed/disagreed with the cue orientation. Subjects were instructed to make use of the information provided by the cue (i.e., location and orientation).

After each block, subjects could choose whether to go on or take a break in order to be able to maintain their level of concentration and tactile sensitivity. They were instructed to take at least one break within 100 trials.

Only RTs between 250 and 1350 ms after target onset were used for analysis, as shorter RTs likely were responses to the previous distractor and longer ones responses to the mask. RTs of each subject were sorted into the different combinations of target location and target orientation. For each subject and for each target location separately, RTs from each session were normalized to the subject’s overall mean and standard deviation (SD) and pooled across the two recording sessions in the following way. First, the population mean and SD were calculated for each target location, both separately for each session and jointly for both sessions (resulting in grand mean and grand SD). Then, separately for each session, the RTs were transformed into z-scores (by subtraction of the session mean followed by division by the session SD). Finally, these z-scores were transformed into normalized RTs by multiplication with the grand SD followed by addition of the grand mean. These normalized RTs could then be pooled across the two sessions of a subject and were used for further analysis.

Statistical analysis of RTs between attentional conditions was performed in SPSS (version 16.0). A four-way mixed analysis of variance (ANOVA) was conducted with the across-subjects factor cue-location group (D2-cued/D4-cued subjects) and the three within-subject, target-property factors location validity (location matched/unmatched), target orientation (parallel/orthogonal), and orientation validity (orientation matched/unmatched). Significant two-way interactions were broken down by simple-effects analysis, i.e., by pooling RTs across all but the two interacting factors and then calculating post-hoc paired t-tests between two levels of one factor, separately for the two levels of the other factor. Cohen’s d (Cohen, 1988; Erdfelder et al., 1996) was used as a measure of effect size.

At the cued location, subjects reacted on average 117 ms faster than at the uncued location. This effect is visible in Figure 3, where RTs under identical sensory conditions are lower in the left (“location-matched”) compared to the right half (“location-unmatched”) of the plot. The effect is also visible in Figure 4. Statistically, it was confirmed by a significant main effect of location validity in the ANOVA (mean difference (M) = −117 ms, repeated-measures standard deviation (SD) = 89 ms, F_(1,18) = 35.6, p < 0.001). Hence, spatial allocation of attention resulted in decreased RTs at the cued location.

RTs to targets with the cued orientation were on average 60 ms faster than orientation-unmatched targets. This is apparent in Figure 3, as under otherwise identical conditions (neighboring same-marker data points without line separation) RTs to orientation-matched targets (left two values in each group of four) are lower than those to orientation-unmatched targets (right two values in each group of four). This observation was statistically confirmed by a significant main effect of orientation validity (M = −60 ms, SD = 46 ms, F_(1,18) = 32.8, p < 0.001).

The feature-based decrease in RTs was strong at the cued location, where responses to the cued orientation were 99 ms faster, but also detectable at the uncued location, where the RT difference amounted to 20 ms. Figure 4 shows this increase in RT between orientation-matched and orientation-unmatched targets both for location-matched (large increase) and for location-unmatched targets (small increase). Statistically, this was reflected in a significant interaction between the factors location validity and orientation validity (F_(1,18) = 48.1, p < 0.001). Follow-up simple-effects analysis confirmed that RTs to the cued orientation were significantly faster both at the cued and at the uncued location. While the effect was large for location-matched targets (M = −99 ms, SD = 66 ms, t₍₁₉₎ = −6.79, p < 0.001, effect size d = 1.5), it was of medium effect size for location-unmatched targets (M = −20 ms, SD = 23 ms, t₍₁₉₎ = −2.59, p = 0.018, d = 0.6). Hence, feature-based allocation of attention not only has an influence at the location where the feature-based cue is informative, but also at the location without any previous feature-based information, documenting a global effect of feature-based attention.

Responses to targets parallel to the finger axis were on average 76 ms faster than responses to orthogonal targets, as can be seen in Figure 5. The effect was statistically confirmed by a significant main effect of target orientation (M = −76 ms, SD = 47 ms, F_(1,18) = 51.9, p < 0.001). The RT difference between targets with parallel and orthogonal orientation points to an anisotropy in orientation processing or perception.

Subjects for whom the index finger (D2) formed the cued location tended to respond faster (729 ± 30 ms) than D4-cued subjects (753 ± 23 ms). That trend is reflected in Figure 3, where the black marker tends to be lower than its adjacent white marker. However, the trend did not reach statistical significance, as seen by the main effect of cue-location group (M = −73 ms, F_(1,18) = 3.5, p = 0.079).

All interactions between within-subject factors were far from significant (F_(1,18) < 1, p = 0.7), except for the already discussed interaction between location validity and orientation validity. Also, all interactions of the across-subject factor cue-location group with one, two, or three of the within-subject factors proved insignificant (F_(1,18) < 1.5, p = 0.2).

Our data show that responses to targets at the cued location are much faster compared to those at the uncued location, in line with several previous studies (Posner, 1978; Spence et al., 2000; Chica et al., 2007) reporting a behavioral RT effect of spatial cueing in touch.

In many other psychophysical studies, Posner-like RT paradigms have been used and whenever RTs decreased for validly- compared to invalidly-cued features or locations, this is typically attributed to an attentional acceleration of the processing of information from the attended location. However, improved perceptual performance is not necessarily evidence for more efficient cortical processing (a consequence of attentional selection; Duncan, 1980; Sperling, 1984). Decreased RTs as a function of the information by the cue can also result from lowering the amount of sensory information required for triggering a response, i.e., by decreasing the level of required certainty. As a target at the cued location was more likely (60%) than at the uncued location, the effect of faster RTs at the cued location might, at least partly, result from a higher expectancy or a higher likelihood of location-matched compared to location-unmatched targets.

Using a psychophysical paradigm designed to investigate global effects of feature-based attention our data show perceptual benefits when the subjects’ attention was directed to behaviorally-relevant tactile features. Responses are faster for orientation-matched targets at the location for which the feature-based cue was orientation-informative (cued location). However, similar to the spatial attention effects described above, it is unclear whether these effects are due to faster processing resulting from allocation of feature-based attention or due to lowering the amount of information required for triggering a response resulting from the higher probability for a matched target (90% vs. 10% at the cued location).

Crucially, however, decreased RTs were not only observed at the cued location but also at the uncued location, documenting a global effect of tactile feature-based attention. At the uncued location, different degrees of certainty between valid and invalid targets cannot account for the effect, as both conditions occurred equally often.

The global effect of feature-based attention we observed is similar to the one reported by psychophysical studies in vision (Rossi and Paradiso, 1995; Alais and Blake, 1999; Sàenz et al., 2003; Arman et al., 2006) and represents the first report of behavioral effects of feature-based attention in touch. Thereby, it extends prior observations (Sathian and Burton, 1991; Whang et al., 1991) showing evidence for feature detectors for grating and intensity changes in the tactile modality. Our findings are also well complemented by the only other study on tactile feature-based attention (Forster and Eimer, 2004), which reported cortical evidence for global effects. In their study, event-related potentials (ERPs) were recorded upon delivery of tactile stimuli presented to the right or left hand. Stimuli were of low or high frequency (first experiment) or of low or high intensity (second experiment). Subjects had to attend simultaneously to one of the stimulus locations and to one of the non-spatial features. ERP analysis revealed effects of feature-based attention (enhanced negativities to the attended frequency or intensity), independent of the current focus of spatial attention, suggesting a global effect of feature-based attention. Perceptual effects were not assessed in the cited study. Further imaging studies will be necessary to identify the cortical regions in which tactile feature-based attention operates.

Across subjects we observed a trend for lower RTs to targets presented to D2 compared to targets at D4. This finding is in agreement with a study by Vega-Bermudez and Johnson (2001) who reported that tactile acuity in a letter-recognition and in a grating-orientation discrimination task progressively declined from D2 to D4, suggesting anisotropic sensitivities between the fingers, possibly because of the more important role of the index finger (compared to the ring finger) in everyday hand-use. Duncan and Boynton (2007) further reported that this effect was reflected in the primary somatosensory cortex (SI), as the increase in tactile threshold from D2 to D4 was correlated with a decrease in digit area from D2 to D4 in SI.

There is an ongoing debate about anisotropic processing and perception of tactile orientations either aligned (“parallel” throughout this text) or orthogonal (“orthogonal” throughout here) to the finger axis. Lechelt (1988) reported better detection of deviations from orthogonal compared to parallel orientations. Similarly, Bensmaia et al. (2008b) observed a lower angular deviation-detection threshold for the orthogonal compared to the parallel orientation and a trend for better performance for orthogonal compared to bars parallel to the finger axis, both in an orientation-discrimination and in a convergence-detection task. Essock et al. (1997) reported that sensitivity to detection of gratings (vs. blanks) was best for gratings parallel to the finger axis and worst for orthogonal ones. The results could not be replicated by a similar study by Craig (1999), whereas Gibson and Craig (2005) found similar results for the finger location stimulated in our design (defined as fingerpad in their study). In a gap-detection and in a grating orientation (GR/OR) task, however, these authors could not find anisotropy between the parallel and the orthogonal orientation. In rhesus monkeys (DiCarlo and Johnson, 2000), “parallel” was reported less often as a neuron’s preferred orientation in the SI than other orientations (not statistically tested), whereas Bensmaia et al. (2008a) did not find any orientation to be overrepresented in SI. Alternatively, it has been suggested in a study on humans and monkeys that orientation parallel to the finger ridges is best detected (Wheat and Goodwin, 2000). However, in contrast to monkeys, the rigde pattern of fingerpads is not consistent across humans.

Our results show anisotropy between target orientations, as parallel targets led to significantly faster responses than orthogonal targets. Without being questioned, seven out of the 20 subjects stated that the detection of parallel targets was easier for them, while none claimed the contrary. Interpreting the results in light of previous studies, one could argue that faster responses to parallel targets resulted from worse detection of deviations from a parallel compared to those from an orthogonal standard orientation (Lechelt, 1988; Bensmaia et al., 2008b). Oblique-stimulus presentation before target appearance might further alter the subjects’ perception of parallel and orthogonal orientation. As even large deviations from the parallel orientation might still be categorized as “parallel”, parallel targets might have appeared clearer and hence were more quickly detectable. However, as we did not measure the exact amount of skin displacement for the parallel compared to the orthogonal orientation, the observed orientation anisotropy might also partly have resulted from differences in physical stimulus strength, potentially leading to a higher perceptual strength for the parallel orientation.