1. Neuroscience
Download icon

High-resolution imaging of skin deformation shows that afferents from human fingertips signal slip onset

  1. Benoit P Delhaye  Is a corresponding author
  2. Ewa Jarocka
  3. Allan Barrea
  4. Jean-Louis Thonnard
  5. Benoni Edin
  6. Philippe Lefèvre
  1. Institute of Information and Communication Technologies, Electronics and Applied Mathematics, Université catholique de Louvain, Belgium
  2. Institute of Neuroscience, Université catholique de Louvain, Belgium
  3. Department of Integrative Medical Biology, Umeå University, Sweden
Research Article
  • Cited 0
  • Views 778
  • Annotations
Cite this article as: eLife 2021;10:e64679 doi: 10.7554/eLife.64679

Abstract

Human tactile afferents provide essential feedback for grasp stability during dexterous object manipulation. Interacting forces between an object and the fingers induce slip events that are thought to provide information about grasp stability. To gain insight into this phenomenon, we made a transparent surface slip against a fixed fingerpad while monitoring skin deformation at the contact. Using microneurography, we simultaneously recorded the activity of single tactile afferents innervating the fingertips. This unique combination allowed us to describe how afferents respond to slip events and to relate their responses to surface deformations taking place inside their receptive fields. We found that all afferents were sensitive to slip events, but fast-adapting type I (FA-I) afferents in particular faithfully encoded compressive strain rates resulting from those slips. Given the high density of FA-I afferents in fingerpads, they are well suited to detect incipient slips and to provide essential information for the control of grip force during manipulation.

eLife digest

Each fingertip hosts thousands of nerve fibers that allow us to handle objects with great dexterity. These fibers relay the amount of friction between the skin and the item, and the brain uses this sensory feedback to adjust the grip as necessary. Yet, exactly how tactile nerve fibers encode information about friction remains largely unknown.

Previous research has suggested that friction might not be recorded per se in nerve signals to the brain. Instead, fibers in the finger pad might be responding to localized ‘partial slips’ that indicate an impending loss of grip. Indeed, when lifting an object, fingertips are loaded with a tangential force that puts strain on the skin, resulting in subtle local deformations. Nerve fibers might be able to detect these skin changes, prompting the brain to adjust an insecure grip before entirely losing grasp of an object.

However, technical challenges have made studying the way tactile nerve fibers respond to slippage and skin strain difficult. For the first time, Delhaye et al. have now investigated how these fibers respond to and encode information about the strain placed on fingertips as they are loaded tangentially. A custom-made imaging apparatus was paired with standard electrodes to record the activity of four different kinds of tactile nerve fibers in participants who had a fingertip placed against a plate of glass. The imaging focused on revealing changes in skin surface as tangential force was applied; the electrodes measured impulses from individual nerve fibers from the fingertip. While all the fibers responded during partial slips, fast-adapting type 1 nerves generated strong responses that signal a local loss of grip. Recordings showed that these fibers consistently encoded changes in the skin strain patterns, and were more sensitive to skin compressions related to slippage than to stretch.

These results show how tactile nerve fibers encode the subtle skin compressions created when fingers handle objects. The methods developed by Delhaye et al. could further be used to explore the response properties of tactile nerve fibers, sensory feedback and grip.

Introduction

The most fundamental requirement for dexterous manipulation is the ability to handle objects without slippage and dropping of the object. To ensure that an object is held safely in the hand, one must apply a sufficient amount of force to the object’s surface to counteract the forces tending to make it slip, for example, the object’s weight and inertia (Johansson and Flanagan, 2009). Exerting an excessive grip is inefficient and can result in crushing the object. Inversely, a minimal amount of force is required to avoid slip and is dictated by friction: a stronger grip is required for slippery surfaces and a looser grip is sufficient for sticky surfaces. A good strategy is then to adjust the grip to friction with an amount of force slightly above the minimum. Previous work has suggested that humans can quickly and accurately adapt to changes in friction (Westling and Johansson, 1984; Cadoret and Smith, 1996). Importantly, tactile feedback is necessary for grip adjustments to take place as disruption of this feedback abolishes the normal, fine-tuned grasp control and results in frequent object dropping despite excessive compensatory grip forces (Westling and Johansson, 1987; Augurelle et al., 2003; Witney et al., 2004). However, how information about friction is encoded by tactile afferents is largely unknown.

Friction information might be partly available at the initial contact. Anecdotal evidence suggests that when contacting surfaces of different frictions the strength of the initial burst of activity of the afferents varies as the surface is changed (Johansson and Westling, 1987). In these experiments, however, different frictional conditions were associated with different surface textures and thus do not necessarily imply that the afferent responses specifically represented friction between the fingerpads and contact surface. Friction might not be encoded per se, but instead, tactile afferents might respond to short and localized slip events that imply impending slip. There is indeed evidence that small, short-lasting slips occur during manipulation (Johansson and Westling, 1984). Those slips trigger strong afferent responses and elicit grip force adjustments (Johansson and Westling, 1987). In addition to the context of object manipulation, it has also been suggested that tactile slip detection plays an important role in a range of tactile tasks involving movements of surfaces relative to the skin (Gueorguiev et al., 2016; Schwarz, 2016).

Tactile slip is not instantaneous but develops progressively (Levesque and Hayward, 2003; Tada et al., 2006; André et al., 2011; Terekhov and Hayward, 2011; Delhaye et al., 2014; Barrea et al., 2018). As the surface-tangential force, that is, traction, of the fingerpad increases, ‘local’ slips begin at the periphery of the contact and progress toward the center until the last central ‘stuck’ point finally slips, that is, the instant of a ‘full slip’ or a global slip. We refer to the period between the beginning of the tangential loading and the instant of a full slip as the partial slip phase. Such a phenomenon gives rise to substantial local strain patterns in the slipping regions, near the boundary between the stuck and the slipping points (Delhaye et al., 2016). We hypothesized that information about these local deformations is carried by tactile afferents that inform the central nervous system about the contact state.

Single-unit recordings of primary tactile afferents, both in humans and monkeys, have shown that type I afferents respond strongly to local skin deformation (Johansson et al., 1982; Sripati et al., 2006; Saal et al., 2017). Those responses contain information about local geometric features such as edge orientation (Pruszynski and Johansson, 2014; Suresh et al., 2016; Delhaye et al., 2019). The most common stimuli used to evoke deformation of the skin are indentations and scanning with embossed geometric patterns or textures. Applying such stimuli makes it possible to relate the strength or the timing of the response to the topography or the statistics of the stimulus itself, but does not provide a mechanistic understanding of the nature of the response with respect to the local skin deformation at the mechanoreceptors themselves. Moreover, it is mostly unknown how tactile afferents respond to surface strains, that is, strains acting tangentially to the surface (as opposed to features indented perpendicularly to the skin surface). Afferent recordings in the hairy skin of the human hand have shown that afferents of all types strongly respond to local skin stretch and that the fast-adapting type I (FA-I) afferents also strongly respond when the stretch is released (Edin, 1992; Edin, 2004). Slowly-adapting type II (SA-II) afferents are also known to be sensitive to skin stretch, but relating their response to the exact local stretch pattern is complex given the large size of their receptive fields. How glabrous skin afferents, that is, those engaged in the contact with objects during manipulation, respond to local skin strain has, to our knowledge, never been studied. This is mainly due to the difficulties of applying well-controlled mechanical stimuli and measuring the strain at the same time.

To address this, we took advantage of a recently developed imaging system that can measure fingertip skin strain through a transparent material during tangential loading of the fingertip until slip occurs (Delhaye et al., 2014; Delhaye et al., 2016). While recording local strains with this system, we simultaneously recorded the activity of human tactile afferents innervating the fingertip (Figure 1A). This way we were able to relate local strains to responses of afferents with receptive fields inside the fingerpad contact area. We found that all tactile afferents in the fingertip responded to slip events, but that FA-I afferents in particular faithfully signaled local skin compressions related to the progression of slips. We suggest that FA-I afferents are primarily responsible for detecting changes in surface strains and that their discharges are a primary source of information for the central nervous system to, for instance, quickly adjust fingertip forces to different levels of friction.

Experimental setup.

(A) A robotic platform (left) was used to move a transparent plate of glass in contact with the fingerpad while the responses of single tactile afferents were recorded from the median nerve using microneurography (right). (B) The plate exerted a servo-controlled normal force of 4 N and was moved tangentially in one of four directions (U: ulnar; D: distal; R: radial; P: proximal). At the same time, a camera was used to image the contact area. All fingerprint images and strain heatmaps are shown using the same view, with the ulnar on the right. (C) From the fingerpad images, features (red dots) were sampled and tracked from frame to frame (the orange arrows show the features motion to the next frame). Features were then triangulated and the triangle strains were computed, leading to two axial strain components (exx and eyy) and a shear strain component (exy). Lower-right pictograms show how an initial triangle (in yellow) is deformed when experiencing positive (tensile, in blue) or negative (compressive, in red) strain.

Results

Slips were applied to the fingerpad using a robotic platform holding a transparent plate of glass that was either plain, yielding high friction, or covered with a hydrophobic coating yielding a lower friction. The plate first made contact with the fingerpad ('contact' in Figure 2A, C) and then moved tangentially at constant velocity until full slip in one of four different directions: ulnar, distal, radial, and proximal, and then moved back in the opposite direction until full slip occurred again (forward and backward, respectively, Figure 1B, Figure 2A, C). The normal component of the force was servo-controlled to be kept at 4 N, and the tangential component was developed as a consequence of the tangential movement of the plate. At the same time, we imaged the fingerpad contact and tracked numerous features on fingertip ridges as the slip progressed (Figure 1C). Finally, the plate was moved down ('release' in Figure 2A, C). We were able to precisely monitor skin strains from frame to frame (i.e., change in strain or strain rates) in the contact area during the transition from a fully stuck contact to a fully slipping contact (see also Delhaye et al., 2016). The skin strains were measured in the contact plane and expressed in terms of three independent components: two axial components aligned to the plate movements (exx and eyy) and one shear component (exy, Figure 1C). Importantly noted, the presence of local strains also indicates that the skin is locally slipping. That is, the limit between the stuck and slipping region is just preceding the front of the strain waves.

Figure 2 with 2 supplements see all
Experimental procedures and typical traces.

(A) Evolution of the global variables, the plate position (vertical in gray and horizontal in black) and the contact force (normal in gray and tangential in black), together with the afferent instantaneous firing rate (and the spikes) as a function of time for a typical fast-adapting type I (FA-I) unit in the high friction condition. The plate was pressed against the fingerpad (‘contact’), moved tangentially forward until the occurrence of a full slip, then moved backward, and finally retracted (‘release’). The partial slip phase is highlighted by the gray vertical boxes. Five repetitions are overlaid. The tangential movement was split into three phases: onset (lasting 100 ms), partial slip, and plateau. (B) Heatmaps of the evolution of the local surface strain rates in the contact area as a function of time during the tangential loading, for one of the five repetitions. The three strain components, axial along x and y, and shearing, are shown (colored triangles depict the deformation axes). Compression (negative strain) is in red. The location of the unit receptive field center is shown with a blue circle on the raw fingerpad image on the bottom left. (C, D) Same as in (A, B) for a typical slowly-adapting type I (SA-I) unit. For both units, insets show recorded, superimposed action potentials and their average shape represented by a dark line.

We focused the analyses on tangential loading movements, from the moment when the plate started to move tangentially and until it completed a forward or a backward movement. Each tangential loading movement was split into three sequential phases defined as follows: (1) the movement onset phase arbitrarily defined as the initial 100 ms; (2) the partial slip phase that lasted until the tangential force reached a plateau and the finger fully slipped; and (3) the plateau phase during which the finger was fully slipping against the glass surface and that lasted until the end of the tangential movement (Figure 2A, C). Strain changes were observed in the contact area during the partial slip phase in the form of two waves of opposite signs (Figure 2B, D). Remember that the strain wavefront is where the slip starts to occur. Those waves started moving at the onset of movement from opposite sides and from the periphery of the contact toward the center and disappeared at the point of full slip (Delhaye et al., 2016). Once the full slip was reached, the changes in strain faded away. The components of the strain changes corresponding to the direction of the plate movement were the largest in amplitude (Figure 2B, D, circled by a black box). Different movement directions elicited different patterns (i.e., compression, stretch, or shear) at all points in the contact area. For instance, the receptive field of a given afferent could be stretched along the proximal-distal axis for movements in the distal direction but compressed along the same axis for movements in the opposite (i.e., proximal) direction (Figure 2Beyy).

Tactile afferents strongly respond during partial slip

We used microneurography (Vallbo and Hagbarth, 1968) to record the activity of single units whose receptive fields were located at the fingertip (Figure 1A). We focused on FA-I and SA-I afferents, which respond to local deformation events and have small, well-defined receptive fields (Johansson and Vallbo, 1983). We also recorded from a few type II afferents (FA-II and SA-II). Sufficient recordings for data analysis (three out of five repetitions of each plate move direction) were obtained from 22 afferents (13 FA-I, 6 SA-I, 2 SA-II, and 1 FA-II). The locations of the receptive fields of all afferents are depicted in Figure 2—figure supplement 1. As expected, the afferents responded vigorously to contact (Figure 2, ‘contact’), but also responded in a variety of ways to the tangential loading (Figure 2, ‘forward’ and ‘backward’). First, we looked at the overall discharge pattern of the afferents. FA-I units showed a phasic response, with a burst of activity at the instant of contact, and another one during the tangential loading (e.g., Figure 2A). However, a majority of FA-I afferents responded mostly only during the partial slip phase, showing no or almost no response during the start (‘onset’ phase) and the end (‘plateau’ phase) of the tangential movement (for U115-04 in Figure 2A, there was one spike at the onset phase during one repeat). SA-I units instead presented a rather tonic response beginning at the initial contact that changed during the tangential loading phase by increasing or decreasing their firing rates (e.g., Figure 2C). The discharge patterns of all recorded afferents for all directions and frictions are reported in Figure 2—figure supplement 1. Video 1 and Video 2 show image recordings of the fingerpad during one trial, together with spike sound associated with the afferent responses, for the two example units shown in Figure 2.

Video 1
Image recordings of the fingerpad during one trial (distal, high friction), together with spike sound associated with the afferent responses (unit 115-04).
Video 2
Image recordings of the fingerpad during one trial (radial, high friction), together with spike sound associated with the afferent responses (unit 110-06).

Note that the tangential movement of the plate led to slight fluctuations in the normal force that could not be suppressed by the force controller (see Materials and methods). Those fluctuations did not evoke strong afferent responses. Indeed, the discharge rates were neither correlated to the normal force nor to its derivative (Figure 2—figure supplement 2). In fact, we considered a causal relationship untenable observing in Figure 2A that the afferent discharge seemed to follow the normal force fluctuation by ~100 ms in the distal direction but preceded it by ~200 ms in the proximal direction.

First, we describe how the tactile afferent responded with respect to ‘global’ stimulus parameters such as the movement phase or direction. Afferents were more active during the partial slip phase than during the two other tangential loading phases (i.e., onset and plateau), suggesting that this period is key to the afferent responses (Figure 3A). Indeed, for all four afferent types, the fraction of trials during which the afferent responded with at least one spike was higher for the partial slip phase than for the two others (one-way ANOVA with repeated measures, F(2,24) = 47.05, p<0.001 for FA-I units; F(2,10) = 13.97, p=0.001 for SA-I units; all p<0.05 for paired comparisons; for FA-II and SA-II afferents, we did not have enough data for such analysis). FA-I units were silent during the plateau, except when a stick-and-slip phenomenon was observed, in which case the firing phase was locked to the stick-and-slip events (e.g., see U109-04 during plateau phase in Figure 2—figure supplement 1).

Properties of afferents’ responses during tangential loading.

(A) The afferents were mostly active during the partial slip phase. The fraction of trials during which the afferents were active for each phase of the tangential loading (onset, partial, and plateau) for the forward and backward movements. Bars show the average across units, and the lines show individual afferents (n = 13, 6, 2, and 1 for fast-adapting type I [FA-I], slowly-adapting type I [SA-I], slowly-adapting type II [SA-II], and fast-adapting type II [FA-II], respectively). (B) Afferents discharge more in a given direction. Mean firing rates elicited by partial slip as a function of angle with respect to the preferred direction (‘North’) for each unit. Each line shows a different afferent, and the mean firing rate was averaged across repetitions, movements (‘fwd’ or ‘bwd’), and frictions. (C) Distribution of the afferent preferred global direction for each afferent type. (D) Mean firing rate for backward versus forward movements. One data point is shown for each afferent (n = 22) and each condition (8 = 4 direction × 2 friction) and averaged across repetition. The dashed gray line is the slope = 1. (E) Mean firing rate for low versus high friction. One data point is shown for each afferent (n = 22) and each condition (8 = 4 directions × 2 forward-backward) and averaged across repetition. The black line is the least square regression, and the dashed gray line is the slope = 1.

For each afferent, we defined its preferred global direction (i.e., ulnar, distal, radial, or proximal), namely, the plate movement direction for which the highest mean firing rate was observed during partial slip. Figure 3B shows the mean firing rate in the preferred global direction (labeled ‘North’) and in the remaining other directions with respect to the preferred one. We found that different movement directions elicited different responses, confirming previous observations (Birznieks et al., 2001). Some FA-I units tended to have a preferred-opposite pattern (5 out of 13), that is, the preferred (‘N’ for North in Figure 3B) and opposite directions (‘S’ for South in 3B) tended to elicit a stronger response than the two other directions. This was not observed for the other afferent types, which tended to discharge less in the direction opposite to the preferred direction. The distribution of the preferred global direction covered the four directions tested; however, we observed overall a slight preference for the proximal-distal axis (Figure 3C). Firing rates during partial slips were consistent across forward and backward (Figure 3D; correlation r = 0.97, n = 176, p<0.001), suggesting similar directional preference when the skin was already under tension (backward movement). While the firing rates were strongly correlated across friction levels (Figure 3E; correlation r = 0.95, n = 176, p<0.001), the firing rates across the whole population tended to be slightly lower for low friction (paired t-test, t(171) = 7.57, p<0.001) and the ratio of the mean firing rate during low and high friction condition was 0.82 ± 0.55 (mean ± std).

FA-I afferents respond to local skin strain rates

Next, we sought to relate the afferent discharge to the strain pattern, that is, the local events taking place inside the afferent receptive field. We observed that the discharge of FA-I units was strongly coupled to the compressive strain changes taking place in their receptive fields during the partial slip phase (Figure 4). This was particularly clear for U104-02, whose receptive field (depicted with a gray circle in the fingerpad heatmaps of Figure 4) was located on the proximal side of the contact area. Indeed, during a proximal movement (Figure 4A, left, and B, right), the plate movement elicited a compressive wave of strain changes (in red) along the y-axis (proximal-distal axis) moving from the proximal side of the contact area toward the center and crossing the afferent’s receptive field. This crossing elicited a strong discharge burst. Moreover, the low friction surface exhibited full slip at a lower level of tangential force, and therefore also earlier than the high friction surface, with respect to the movement onset. As a consequence, the compressive strain wave moved across the afferent receptive field earlier in this case and, strikingly, the discharge burst of the afferent also took place earlier, coinciding with the strain changes. This is even easier to observe for the backward movement (Figure 4B, right). In this case, due to the previous loading, the movement of the compressive wave came even earlier when the low friction condition was used and the discharge burst of the afferent coincided. Finally, we observed that the response evoked by a stretch wave, generated by a distal movement (Figure 4A, right, and B, left), was much weaker than its compressive counterpart. Still, the timing of the response was perfectly synchronized with the occurrence of the stretch in the receptive field. Note that a short burst was elicited at the onset of the movement in the distal direction in the backward case (Figure 4A, right). Such transient burst cannot be explained by our strain measurements and occurred in a small fraction of the trials and only in a few afferents (Figure 3A). Also note that part of the unit receptive field lost contact during the partial slip phase in the high friction case.

Fast-adapting type I (FA-I) responses during partial slip are related to local strain rates.

(A) Evolution of tangential force, afferent instantaneous firing rate (together with the spikes), and strain rates as a function of time during tangential loading in the proximal direction (forward) followed by a distal movement (backward). Data are shown for two different frictions, high in black and low in gray, and are aligned on the onset of movement. Five repetitions are shown, except for the strains for which one trial is shown. The compressive strain (negative) is in red. The color of the contact area contour indicates the friction condition, and the gray circle shows the location of the afferent receptive field. The horizontal lines between the heatmaps depict the partial slip phase as shown in Figure 2. (B) Same as in (A) but for distal movement (forward) followed by a proximal movement (backward). Inset shows recorded, superimposed action potentials and their average shape represented by a dark line.

To test the hypothesis that the responses of the tactile afferents are caused by specific ‘local’ strain patterns taking place inside the contact area, we took two different approaches. In the first model-free approach, we looked at the strain pattern observed at the time of each spike across all stimulus directions and frictions and computed a ‘spike-triggered average’ (STA, see Materials and methods) for all recorded FA-I and SA-I units. If the afferents with a receptive field in the contact area responded to local strains, we expected to observe a clear strain pattern associated with these units’ discharges. In contrast, for afferents with a receptive field outside the contact area, we expected no clear strain pattern at all. First, we used the strain rate norm (||e||) as a variable to estimate the STA. We found that, indeed, the average strains causing spikes in all recorded FA-I afferents had a clear, more or less annular (ring-like) pattern (Figure 5A). Such an annular pattern is expected from the stimulus, which is a strain wave in the form of an annulus and does not reflect the shape of the afferent receptive field but rather the correlations present in the strain patterns (Materials and methods). Importantly, however, the pattern overlapped the afferent’s receptive field and often peaked inside it. Furthermore, as expected, we did not observe such a clear pattern for the afferents having their receptive field outside the contact area (Figure 5, middle). Strikingly, clear patterns did not emerge for the SA-I afferents, suggesting that those afferents are less sensitive to the local surface strain changes (Figure 5A, bottom). Note that we repeated the same analysis using the total (cumulative) strain instead of the strain changes, and that again, we did not observe any clear pattern. The peak values of the STA are shown in Figure 5B in orange and show strong values for FA-I units inside the contact and much weaker values for other afferents, confirming the previous observations. The same STA analysis was repeated using the two principal strain components, one compressive and one tensile, separately to build two STA maps (see Materials and method). The results obtained are consistent with the STA obtained with strain rate norm, that is, that a clear pattern emerges only for FA-I afferents and that the STA peaks in the afferent receptive field (Figure 5—figure supplement 1). Moreover, we found that the compressive STA peaks were generally larger and more often found in the afferent receptive field than their tensile counterpart, suggesting that FA-I afferents are more sensitive to compression. This finding will be further supported in the next section.

Figure 5 with 2 supplements see all
Afferent responses to surface skin strains.

(A) Heatmaps of the spike-triggered average (STA, in orange) for the fast-adapting type I (FA-I) units inside the contact area (top row), the other FA-I units (middle row), and the slowly-adapting type I (SA-I) units (bottom row). The STA matrices were obtained from the norm of the strain (e) and normalized to the maximal value (reported in B). The gray contour shows the initial contact area, and the pink contour depicts the parts of the contact area that remained in contact for at least 50% of the partial slip phase. The black circle is the unit receptive field. (B) STA peak value (red) and linear regression model performance (green). For units in or on the border of the contact area, the maximum was taken inside the receptive field. For units outside the contact area, the maximum was taken anywhere inside the contact area. The STA of some units were undefined over the whole contact area (see Materials and methods); those units are not shown. The unit shown in (C) is highlighted with a black circle. (C) Time evolution of the strain rates inside the receptive field, the actual firing rate (represented as spike events convolved with a Gaussian kernel, see Materials and methods), and the predicted firing rate of an example unit during partial slip for each direction and each friction (black is high friction, gray is low friction).

In the second, model-based approach, we aimed to predict the afferent discharge rate from the skin strain measured in the contact area. Inspection of the data led us to assume that, first, strains of opposite signs might not contribute in the same manner to the discharge as skin stretch seemed to evoke weaker responses than skin compression (Figure 4). Second, multiple components might be needed to explain responses in all directions. Therefore, we first set out to test if the afferents’ firing patterns could be reliably predicted for the skin strains using a model with six distinct predictors, that is, the three strain components each half-wave rectified, using both the positive (stretch) and negative part (compression). The simplest model possible, a multiple linear regression including an intercept, was used. Our method was cross-validated, such that the regression models were fit on one friction condition and tested on the other (Materials and methods). In the subsequent validations, we used data from different directions and forward vs. backward movements and observed quantitatively similar results (see Figure 5—figure supplement 2). The results obtained using the model-based approach revealed similar trends as the first model-free method. The linear model could predict the discharge pattern of the FA-I afferents, but not the SA-I afferents. Heatmaps built from the cross-validated R2 (Figure 5—figure supplement 2) were qualitatively similar to those built from the STA. An example unit is shown in Figure 5C for an afferent with a substantial R2 (0.66). This unit maximally responded in the proximal direction, when a compressive wave was observed in its receptive field. The maximal values of the R2 found within the receptive field (or anywhere for the afferents outside the contact area) are shown in Figure 5B in green. As with the model-free approach, only the firing rates of FA-I inside the contact area could be predicted from the strain. The high R2 value for the SA-I afferent outside the contact area is because this particular afferent was either active at a constant firing rate or silent, generating two separate clouds of data points and thus driving up the R2 (Figure 2—figure supplement 1).

In summary, we observed that FA-I afferents respond to local strain patterns generated during partial slips.

Aspects of the skin strain rates encoded by the FA-I afferents

Having demonstrated that the recorded FA-I afferents respond to local strain patterns, we then sought to uncover what aspects of the strains were responsible for these responses. To that end, we aimed to predict FA-I afferent discharge rates with a subset of strain predictors and to compare their performance to the full model with six predictors. The analysis was performed only on the FA-I afferents for which we had optical measurements of the strains, that is, those having their receptive fields inside the contact area (n = 6, all shown in the top line of Figure 5A). Since all models are cross-validated, they can be compared irrespective of the number of predictors. First, we selected three single predictors that were invariant to the choice of a particular reference frame. The strain norm, informative about the intensity but neither the orientation nor the sign of the deformation, and the two principal strain components separately, the compressive (e1) and the tensile (e2), obtained from single-value decomposition of the strain tensor (Materials and methods), informative about the intensity and the sign of the deformation, irrespective of the orientation of the deformation. All those three single predictor models performed worse than the full model, as could be expected (Figure 6A, left). However, we found that the compressive principal component always outperformed the tensile one, suggesting that the afferents are more sensitive to compression than to stretch.

Figure 6 with 2 supplements see all
Aspect of the skin strain encoded by the fast-adapting type I (FA-I) afferents inside the contact area.

(A) Firing rate prediction performance (cross-validated R2) for different models with respect to the full model comprising six predictors (0 on the y-axis corresponds to the full model performance, lower is worse). The three first models (e, e1, and e2) are single predictors and rotation invariant. The two last models (compression and stretch) have one component, and the performance obtained with the best orientation is shown. The unit shown in (B) is highlighted with a black line. (B) Polar plot showing the performance of the prediction (cross-validated R2) of linear models based on single strain components (red for compression and blue for stretch) as a function of the rotation angle of the reference frame (from 0 to 90°) with respect to the radial-ulnar axis (0°). Data from an example afferent U104-02. The black lines show the frame rotation that yields the best performance (maximizing the sum of the R2 of the compressive and tensile models).

Next, we used each of the predictors of the full model separately. That is, the half-wave rectified positive and negative value of the three strain tensor components (exx, eyy, and exy). Given that those components are dependent on the choice of a particular reference frame orientation, we repeated the fitting procedure for multiple rotations of the reference frame equally spaced from 0 to 90° (Materials and methods). The results are shown in Figure 6B for an exemplar afferent, with the shear component ignored. In this figure, the performance of the prediction (R2) based on a single strain component (compressive in red and tensile in blue) is shown as a function of the reference frame rotation. This afferent seemed to have a preferred strain orientation close to 90° with respect to the radial-ulnar axis (i.e., along the proximal-distal axis), where the compressive component peaks. That is, the afferent seemed to encode preferentially 'local' compressive strain rate along a particular orientation. Indeed, as already described in Figure 4, this unit was responding strongly in the proximal condition, where a compressive strain wave along the proximal-distal orientation passed through its receptive field. Perpendicular to that orientation, the tensile component peaked as well but with a lower R2. This is expected since compression in one orientation generates stretch in the other at the same time because the volume is mostly conserved. The same plot as in Figure 6B is provided for all FA-I afferents and the three cross-validation methods in Figure 6—figure supplement 1. It is important to avoid the confusion between the units’ preferred direction described in Figure 3, which relates to the robot movement direction and the maximal firing rate of the afferent, and the strain orientation preference described here, which is related to the orientation of the local deformation in the afferent receptive field.

From these analyses, we draw two important conclusions. First, compressive strain change is a more effective stimulus than tensile strain change (Figure 6A, right), confirming the observation in the principal component analyses, namely, that the FA-I responses are mostly driven by compression. Second, FA-I afferents did not respond to compression in any orientation but rather to compression along a certain preferred strain orientation. The argument for this is twofold: (1) the performance of a single compressive component in a particular orientation was always higher than the performance of the compressive principal component e1 (Figure 6A) and (2) models with one single component along its preferred strain orientation performed as well as the full model comprising all the components, suggesting that this component is mainly responsible for driving the afferent response. Note that the FA-I afferents’ preferred strain orientation seemed to coincide with the local fingerprint orientation, but more data is needed to confirm this trend (Figure 2—figure supplement 2, correlation r = 0.74, p=0.10, n = 6).

In sum, our analyses strongly suggest that the FA-I are sensitive to local skin strains, more so to changes in compressive strain than tensile strain and that they respond maximally along a preferred strain orientation.

SA-I and SA-II responses are related to external forces

Finally, we asked how much the response of all types of afferents was related to the ‘global’ external 3D force vector. We computed the correlation between the afferent firing rates and each force components during the tangential loading (all three phases). We also computed the same correlations with the force rates. We found low correlation values for both force and force rates for the fast-adapting afferents (FA-I and FA-II). However, the correlations with tangential (horizontal) forces were high for slowly-adapting afferents (SA-I and SA-II), especially for those not inside the contact area (Figure 2—figure supplement 2). Their firing rates were however not related to the small vertical (normal) force fluctuations.

Discussion

This is, to our knowledge, the first study that investigates the relationship between the responses of tactile afferents to slippage and the detailed surface strain patterns in glabrous skin. Specifically, the results from our analyses imply that the local compression induced by partial slip generates strong responses from FA-I but not SA-I units. Given that the FA-I from the fingertip are mostly silent when the fingerpad is stuck to the object and even at the onset and offset of the plate movement, their vigorous responses to the changes in compressive strain provide a particularly salient signal that enables contact stability. It is reasonable to suspect that FA-I afferents play a crucial role in maintaining grasp stability since they constitute more than 50% of all afferents in the glabrous skin of a human fingertips (Johansson and Vallbo, 1979). Our results show that while partial slips give rise to strong responses, the friction per se is poorly encoded by FA-I. Finally, given that cortical somatosensory neurons are strong edge detectors (DiCarlo et al., 1998; Delhaye et al., 2018), the particularly salient signal from the FA-I afferents will likely be amplified downstream.

A previous study showed that information about grip safety, that is, the increment of tangential force needed to reach full slip, is present in the response of local tactile afferents at the level of a population of afferents, irrespective of the level of friction (Khamis et al., 2014). Given that all afferents respond to progressing slips, and that the timing of each FA-I afferent’s burst is dependent on how far from the center of contact the afferent is, it is not surprising that a linear combination of multiple FA-I units can provide a good estimate of the grip safety. While this previous study demonstrated that the information is present in the response of the population, our approach provides a mechanistic explanation and underlines the saliency of the grip safety signal.

An essential component of this work was the use of two different frictions. By using the same protocol with two different frictions, we were able to disambiguate the relative contribution of the external force and the local strain measurements to the afferent response. Indeed, until an instant very close to full slip in the low friction condition, the low and high friction materials lead to very similar normal and tangential force profiles, whereas the surface strain patterns inside the contact area have different timing and their progressions differ very early on. Thanks to this contrast, uniquely provided by the difference in friction, the earlier afferent response in the case of the low friction cannot be attributed to the change in tangential force itself. The fact that the afferent responses, particularly the FA-I units, are related to local slip and are therefore poor predictors of friction per se suggests that humans could be exquisitely sensitive to detect slip while being poor at estimating friction. That is, estimating friction would require the occurrence of partial slips, and the absence of slips would make such estimates impossible.

The FA-I afferent responses were best predicted by compressive strain, rather than tensile strain. Moreover, the afferents were mostly sensitive to a particular strain orientation. This property might be explained by the shape of the afferent receptive fields, which is elongated (rather than circular) and contains numerous zones of high sensitivity (‘hot spots’, Johansson, 1978; Phillips et al., 1990; Jarocka et al., 2021). A particular arrangement of those hot spots might lead to increased sensitivity to a particular strain orientation. Indeed, it has been demonstrated that particular arrangements enable each afferent to differentiate the orientation of edges moving across the receptive fields from the strength or the fine timing of their response (Pruszynski and Johansson, 2014; Suresh et al., 2016). In fact, the moving strain wave related to partial slip observed in this study can be seen as an edge moving across the contact area.

Limitations

Our methods generate high-resolution representations of the distribution of surface strain in the contact area of the fingertip (i.e., 2D, x and y axes) but do not allow measurements of the distribution of the deformations perpendicular to the fingertip surface inside the contact area (i.e., along the z-axis), nor outside the contact area. Given that the SA-I afferents did not respond to the surface deformation measured in this study, they could be sensitive to two other non-measured aspects of the local deformations. First, since SA-I afferents are known to be exquisitely sensitive to contact and contact pressure (Goodwin and Wheat, 1999; Wheat et al., 2010; Khamis et al., 2015), they may be more sensitive to the strains perpendicular to the surface, rather than the tangential (i.e., planar) strains. Indeed, the tangential loading can cause a progressive redistribution of the pressure inside the contact area (even if the global normal force is maintained constant) that will consequently generate ‘local’ vertical strains (not measured, which would be denoted ‘ezz’). The SA-I shows slow firing rate variations that might be well attributed to a relatively slow movement of the center of pressure during tangential loading that results in local pressure change. Another possibility is that the SA-I afferents are sensitive to local shear (which would be denoted ‘exz’ and ‘eyz’). The fact that their firing rates correlate with the tangential forces supports this idea, and more measurements are needed to disentangle the two possibilities. Importantly, these two other aspects of skin deformations are likely much less informative about the partial slip state of the finger; therefore, SA-I afferents are less likely to contribute to the detection of the slip events. When considering the subdermal location of SA-II and FA-II afferent terminals, it is not surprising to observe a poor coupling between their responses and the measured surface strains.

The tangential speed chosen for practical reasons in our experiments is certainly lower than that during normal manipulation actions, during which much higher strain rates are probably generated. In fact, it is unknown what the typical amplitudes of surface skin strain during manipulation are. Moreover, during active manipulation, humans typically also vary the amount of grip (normal) force according to the tangential force, while our experiment kept the normal force constant. The development of a manipulandum equipped with embedded imaging of skin deformation in the contact area will make it possible to quantify deformation in an ideal environment with both normal and tangential force fluctuations (Barrea et al., 2017). Nevertheless, it is reasonable to expect that FA-I afferents respond even more vigorously during actual manipulation. It remains also to be demonstrated that in a typical manipulation task the timing of this signal enables online control of grip, that is, that there is enough time to react. In this study, the partial slip phase lasts several hundreds of milliseconds, which would definitely leave enough time for cutaneous signals to contribute to online control.

Most surfaces encountered during natural object manipulation have frictional properties that cause stick-and-slip when the tangential load is sufficiently large, that is, local breaking of the bond between the surface and even a single fingerprint ridge. However, stick-and-slip events were largely absent in our experiment because we had to use a flat and transparent material to be able to measure skin surface deformations. With natural material, partial slip might therefore occur in a stepwise manner and result in readily measurable accelerations (Johansson and Westling, 1984; Johansson and Westling, 1987), as opposed to a very smooth progressive wave observed in our experiments. Nevertheless, while the timing might be affected, the patterns of deformation observed in their purest form in this study will generalize in some form to other surfaces and therefore the observations made in our study are likely to hold with natural materials.

There are several reasons why the models' R2 are relatively low. First, the models are linear, whereas the response of tactile afferents to skin deformations is known to be far more complex than accounted for by simple linear relationships (Dong et al., 2013; Saal et al., 2017). Second, the inputs of the model were obtained from a single point inside the receptive field of the afferent. Yet, we know that human tactile afferents have complex receptive fields with multiple zones of sensitivity that they owe to the branching of the afferent terminals in the skin (Johansson, 1976; Phillips et al., 1992; Pruszynski and Johansson, 2014). A richer stimulus using more stimulation conditions (tangential speeds, normal forces, more orientations) to reduce the effect of the correlations inherently present in the strain pattern might provide sufficient data to identify multiple zones of sensitivity, using the STA approach developed in this study. However, combining our stimulus with another one dedicated to identifying such receptive field topography (Jarocka et al., 2021) would be needed to establish a causal relationship between the two. Finally, our setup does not provide the ability to measure all aspects of local skin deformation as discussed above; we were unable to measure deformations along the axis normal to the glass surface. Analyses of strain normal to the contact surface and deformations outside the contact area would require other approaches, for example, finite element analysis.

Conclusion

Our sense of touch enables us to grasp and manipulate objects with great dexterity. The ability to capture and extract very subtle mechanical phenomena arising continuously during fine manipulation is probably a key to its success. We have demonstrated in this study that subtle skin compressions taking place in the contact area with the object’s surface before the slippage provide essential feedback for grasp stability. Given that estimating friction is a complex problem influenced by many factors of the object itself (texture, adhesion, hydrophilic or hydrophobic properties) and by properties of the skin (for instance, stiffness), some of which are constantly changing (e.g., humidity), the nature of this feedback and its independence from friction makes it particularly well suited for grasp stability.

Materials and methods

Participants

Sixteen healthy human subjects (seven females; ages 19–24 years) participated in the experiment. Each subject provided written informed consent to the procedures, and the study was approved by the local ethics committee at the host institution (Université catholique de Louvain, Brussels, Belgium). Subjects were seated in a comfortable dentist chair with the forearm slightly pronated and abducted. The forearm was resting on a horizontal cushioned support. The right hand, with the volar side facing the ground, was fixed to a custom-made support (Figure 1A), which enabled to lock one finger while keeping the rest of the fingers away in a safe position. The participant’s fingers were stabilized by gluing the nail of digit II and III to aluminum bars using cyanoacrylate and connecting the bars rigidly with the custom-made support. Such fixation hindered finger movement during the stimulation but allowed fingertip deformation due to the compliance of the fingerpad.

Apparatus

The stimulations were applied by a robotic platform based on an industrial robot, as already described in previous studies (Delhaye et al., 2014; Delhaye et al., 2016; Barrea et al., 2018). Briefly, a transparent plate was mounted horizontally on the end effector of an industrial robot (Denso Robotics, Japan) and its movement in three orthogonal directions could be controlled precisely. Two force transducers were mounted, one on each side of the plate, and measured the forces applied to the subject’s fingertip (ATI force sensors, ATI Industrial Automation, USA). The measured RMS error of the force measurement was low (ranging 0.01–0.02 N along the tangential axes and 0.03–0.06 N along the vertical axis). The stimulus was applied to the finger where the receptive field of the recorded afferent was located. The relative angle between the stimulus and the finger was around 30°. The transparent plate consisted of smooth glass and was either plain or covered with a hydrophobic coating, RainOff (Arexons, Italy) reducing friction. The robot followed a programmed tangential (horizontal) trajectory with the help of the manufactory position controller. A custom PID controller was developed to servo-control the normal (vertical) force applied to the fingerpad by feeding back the force measurements (average RMS error is 5% during the whole tangential loading, and average peak error is 12% or 0.5 N).

The robot was combined with a custom-made fingerprint imaging apparatus composed of a high-speed (50 fps) and high-resolution (1200 dpi) camera (Mikrotron MC1362, 1280 × 1024 pixels, Mikrotron GmbH, Germany) and a coaxial light source (White LED Backlight, Phlox, France). This optical apparatus enabled imaging of fingerprints in contact with the transparent stimuli to compute strains in the fingertip contact, as described in Delhaye et al., 2016.

Experimental procedures

Request a detailed protocol

We used the microneurographic technique (Vallbo and Hagbarth, 1968) to record single skin afferent activity (‘unit’). An insulated tungsten needle electrode was percutaneously inserted into the right median nerve ~10 cm proximal to the elbow joint (Figure 1A). Once the recording electrode was in an intraneural position, it was manipulated in minute steps until single-unit activity clearly stood up from the background noise. To evoke the responses of the afferents, the relevant skin areas of the fingertips were stimulated. Once a single afferent activity was identified and a corresponding receptive field located, the threshold force for the receptor was determined using von Frey hairs. According to well-known criteria (Vallbo and Johansson, 1984), receptors were identified as slowly-adapting (SA) if the discharge was sustained while pressing with a glass probe for at least 2 s; otherwise, they were considered fast-adapting (FA). The type I units have small, easily located receptive fields, whereas type II units are characterized by large and poorly defined receptive fields. Furthermore, SA-II could be distinguished from SA-I by the regularity of their inter-spike intervals, and FA-II units from FA-I by their response to remote light taps. Once the spot with the highest sensitivity within the receptive field had been marked with a permanent marker, the experimental procedure was initiated. The afferent search was voluntarily biased toward type I afferents because of their relatively small receptive field, which makes them more likely to respond precisely to the local strain patterns recorded within the fingertip contact.

While recording impulses from single tactile afferents innervating the right index or middle finger of subjects, the stimulus was moved vertically toward the fingerpad in single trials, made contact, and reached a preset normal force of 4 N (‘contact’ phase). That normal force was kept constant during the entire trial. After a delay of 1.2 s, which was necessary to induce occlusion and make the contact visible on the camera (Dzidek et al., 2017), the surface was moved tangentially (horizontally) with a constant speed (5.5 mm/s) first in one direction (ulnar, radial, distal, or proximal) for 8 mm (‘forward’ movement) and then in the opposite direction for 12 mm (‘backward’ movement) (Figure 1B). The relatively slow tangential speed is explained by two reasons. First, since our optical system has a limited frame rate (50 Hz), we wanted to have a relatively slow stick-to-slip transition such that it was accurately measured by the imaging system. Second, the normal force servo-control was more effective at low speed. The movement amplitudes were thus large enough to ensure eventually full sliding between the fingertip and the surface in both directions. The surface was then retracted from the finger, and the trial was ended. The procedure was the same for all trials. Each protocol consisted of five repetitions of the four stimulation directions for each of the two friction conditions, totalizing 40 trials (2 frictions × 4 directions × 5 repetitions). For practical reasons, the same sequential order was used for all protocols: all high friction trials (on plain glass) were run first, followed by all low friction trials (on coated glass). Moreover, the direction sequence was always the following: ulnar, distal, radial, and proximal, with all five repetitions made in a row.

Plate position and forces exerted on the finger were sampled at 1 kHz along the three axes (PCIe-1433, National Instruments, and LabVIEW). Fingerprint deformations were monitored through the transparent plate during the tangential movement of the platform, allowing the derivation of surface strains at the fingertip contact. The neural signals were sampled at 12.8 kHz after amplification close to the recording site. The identification of single action potentials was made semiautomatically under visual control (Edin et al., 1988). WINSC/WINZOOM software (Umeå University, Sweden) was used for recording and analyzing the neural data. The instantaneous firing rate (as shown in Figures 2 and 4) was defined as the inverse of the time interval between two consecutive spikes for the interval duration. This rate was resampled at 1 kHz to obtain convenient time series for data analysis.

Image processing

Request a detailed protocol

For each trial, strains in the contact area were computed from the images as described in Delhaye et al., 2016 (Figure 1C). Briefly, the contact region was first extracted semiautomatically from each image of the sequence. Second, equally spaced features were sampled in the contact area of the initial frame and then tracked from frame to frame to measure the displacement field in the contact using the optical flow technique (Lucas and Kanade, 1981) implemented in the OpenCV online computer vision toolbox (Bradski, 2008). Third, the tracked features were triangulated (Delaunay triangulation), and Green-Lagrange strains were computed for each triangle by calculating the gradient of the displacement field. This operation yielded a 2-by-2 symmetric strain matrix for each triangle and each pair of consecutive frames. Axial strains, that is, the diagonal elements of the strain matrix, were denoted exx and eyy and shear strain (off-diagonal) was denoted exy. The x-axis was aligned to the radial-ulnar orientation (Figure 1B). Strains were filtered, first spatially (using Smooth Triangulated mesh from Matlab FEX, https://mathworks.com/matlabcentral/fileexchange/26710-smooth-triangulated-mesh), and then temporally (median filtering over three strain values). By using the term 'strain' throughout the article, we refer to the elements of the strain tensor computed between two consecutive images and expressed in percent per second, that is, strain rates.

The principal strain components denoted e1 and e2 were then obtained by an eigenvalue/eigenvector decomposition of the strain matrix. The principal strains e1 and e2 correspond to the maximum compressive and tensile strains at the location where the strain tensor is measured. The eigenvalue decomposition is equivalent to a rotation of the reference frame so that the shear strain is canceled and only axial strains remain, which thus corresponds to the maximum local compression and/or dilation. Therefore, the principal strains are not necessarily aligned with the axes of the x-y reference frame defined above and their orientation is potentially different for each local strain tensor. Therefore, the matrix components of the local strain tensor e are solely given by the axial and shear strains along the x-y plane (exx, eyy, and exy). Besides, given the high stiffness of the outer layer of the skin (Wang and Hayward, 2007), it is reasonable to assume that the area of local patches of skin is conserved after deformation. Under this hypothesis, it can be shown that e1 and e2 have opposite signs. This means that if one of the principal strain is compressive, the other one is dilative and vice versa. Moreover, we follow the convention to sort the eigenvalues by ascending order. Therefore, e1 will always be smaller than e2. Under the hypothesis of area conservation, this implies that e1 will be negative and thus compressive, whereas e2 will be positive and thus tensile. This hypothesis has been verified in practice on the data.

To assess the strains taking place at each given point of the contact area across different trials, we interpolated the strains on an arbitrary rectangular grid, the ‘strain grid’. First, an easily identifiable feature of the fingerprint located near the center of contact was manually spotted on the first image of each trial and used to precisely align all trials. Given the precise repeatability of the trials, the spotted feature only moved by a few pixels from trial to trial. Then, the contact area was centered on the grid using its geometric center on the first image. The feature coordinates on the first image were then used to interpolate the strains over the entire trial on the given grid using MATLAB’s scatteredInterpolant function. The ‘strain grid’ spacing was set to 10 pixels (~200 um) and was composed of 120-by-90 elements (1200 × 900 pixels).

The local fingerprint orientation was also estimated as described previously (Delhaye et al., 2016) by computing the image gray-level gradients (Bazen and Gerez, 2000) over a 60-by-60 pixels window.

Data analyses

Receptive field location and size

Request a detailed protocol

The spot of highest sensitivity in the receptive field was marked on the finger, and a close view picture of the finger was taken at the end of the experiment. For the afferents inside or on the border of the contact area, the location and local fingerprint pattern were visually identified on the picture and matched with the pattern recorded using the bottom view camera (see insets in Figure 2B, D for instance), to obtain the coordinates of the receptive field center in the image recordings. The receptive field of a type I afferent was defined as a circle around this central coordinate with a radius of 150 pixels (~3 mm, Johansson and Vallbo, 1979).

Phases of tangential loading

Request a detailed protocol

Forward and backward movements of the glass surface loading the fingerpad were divided into three sequential phases. The onset started with the tangential (horizontal) movement of the platform and lasted 100 ms. This value was set arbitrarily as short as possible but also ensuring that the transient response due to the start of the motion observed in some afferents did not affect the partial slip phase. The partial slip phase followed and lasted until the tangential force plateau was reached, namely, when the tangential force derivative returned close to zeros. At this point, the finger is fully slipping. The plateau phase ended when the plate movement stopped. Afferent responses were related to the local strains only during the partial slip phase.

Afferents’ recording quality

Request a detailed protocol

The quality of the electrophysiological recordings was assessed by computing the spikes' signal-to-noise ratio (SNR). The SNR was defined as the ratio of the mean peak-to-peak amplitude of spikes to the mean of peak-to-peak noise (measured from a 1-ms-long window before each spike). The noise level was subtracted from the spike signal before calculating the ratio. The mean SNR across all units is 4.3 (range 2.1–7.9).

Afferents’ effective sampling frequency

Request a detailed protocol

We checked that the frequency content of the afferent response matched the optical recording system by estimating their ‘effective sampling frequency’ (Dawdy and Matalas, 1964; Dimitriou and Edin, 2008). We found that, indeed, the stimulus did not generate effective information above 50 Hz; across afferents, median effective sampling frequency was 15 Hz (Q1-3 = 3.5–38.5 Hz). Such low-frequency content is explained by the slow nature of the stimulus.

Spike-triggered average

Request a detailed protocol

 The STA matrices were computed only during the partial slip phase, and data from all conditions and repetitions were combined. The spikes were first binned into 20 ms bins to match the image/strain frame rates (given the afferents' frequency content, such time-bin resolution is adequate). The strains consisted of a matrix of 90-by-120 elements covering the contact area for each time bin. First, strain frames from all the corresponding bins containing at least one spike were selected, later referred to as the 'spike selection'. Frames corresponding to bins containing more than one spike were repeated (n times for n spikes per bin). The median number of spike used to build the STA was 373. Second, the same number of strain frames was randomly chosen from bins that did not contain spikes (referred to as 'no-spike selection'). Then, the average of the no-spike selection was subtracted from the average of the spike selection. The result was a 90-by-120 matrix defined as the STA. The sum of the standard deviation of the two selections was also obtained. The STA elements that were lower than four times the sum of the standard deviations were set to 'NaN' (not a number). The STA elements that were not inside the contact area for at least 50% of the time were also set to 'NaN'. It is possible to compute an STA for any strain component. In Figure 5, we chose to compute the STA for the strain norm, which is defined as the square root of the sum of the square of all elements of the strain tensor (|e|=exx2+eyy2+2·exy2). The strain norm has no physical meaning but provides an estimate of the intensity of the deformation at any point and is rotation invariant, and therefore not dependent on the choice of a given reference frame. When expressed as the strain norm, we expect the stimulus to take the shape of an annulus of deformation moving from the periphery to the center. Therefore, the STA is expected to tend toward the shape of an annulus.

Linear regression model and cross-validation

Request a detailed protocol

For each element of the 90-by-120 'strain grid' covering the contact area, we fitted a multiple regression model to predict the afferent firing rate. Only the time bins in the partial slip phase were used, and only the elements of the matrix that were inside the contact area for more than 50% of the time were used. We used regression with six distinct predictors and an intercept. Each of the three strain components (exx, eyy, and exy) were half-wave rectified using both the positive (related to stretch and positive shear, exx+, eyy+, and exy+) and negative part (related to compression and negative shear, exx-, eyy-, and exy-). The firing rate was obtained as follows. The vector of the spike times, sampled at 1 kHz and made of 1’s at the time of a spike and 0’s elsewhere, was first convolved by a Gaussian window (using MATLAB’s gausswin function, with 480 points and an alpha value of 6, normed) and multiplied by 1000. The windows' width was empirically found to maximize the prediction performance across afferents (suggesting that our stimulus frequency content is centered around 4 Hz). Second, the result was resampled at the image frame rate (50 Hz). The result was defined as the firing rate and used to fit our models. We used three cross-validation schemes. In the first, the data was split in two (twofold CV), with the high friction trials in one fold and the low friction trials in the other fold. In the second twofold scheme, the first fold consisted of the data during the forward movement and the second fold the backward movement. In the third, fourfold, scheme, each fold consisted of all trials in a given direction (ulnar [U], distal [D], radial [R], or proximal [P]). For all schemes, the model was fitted on the data that excluded one fold and tested for prediction on that fold. The procedure was repeated for each fold such that a prediction was made for all data points. The R2 was obtained for each scheme by combining prediction from each fold.

Component selection and rotation

Request a detailed protocol

The same cross-validation procedure was followed to fit linear regressions with different predictors. All models included an intercept. For that procedure, only FA-I afferents inside the contact area were used, and the strains coming from a single location were used: the location inside the afferent receptive field yielding the maximal performance for the full model (with six predictors). The strain tensor was recomputed in a rotated reference frame following this transformation:

exx'exy'exy'eyy'=cosθsinθ-sinθcosθexxexyexyeyycosθ-sinθsinθcosθ

The reference frame was rotated from 0 to 90° with steps of 5°. When the frame is rotated by 90°, the rotated x-axis corresponds to the initial y-axis (Figure 6B). Four different models were tested using this procedure, with a single predictor each, using half-wave rectified strain along the two axial components.

Statistical analyses

Request a detailed protocol

All statistical analyses were performed in MATLAB using the functions corr (for Pearson correlation), ttest (for paired t-tests), and ranova (for one-way analysis of variance with repeated measures). For the ranova, the normality and sphericity were verified and the Greenhouse–Geisser adjustment was applied if needed. The test performed, degrees of freedom, T or F statistic is always mentioned with the p-value.

Data availability

All the data used to create the figures in the manuscript are available for download following this Zenodo link: https://zenodo.org/record/4818439.

The following data sets were generated
    1. Delhaye BP
    2. Jarocka E
    3. Barrea A
    4. Thonnard J
    5. Edin B
    6. Lefèvre P
    (2021) Zenodo
    ID 4818439. High-resolution imaging of skin deformation shows that afferents from human fingertips signal slip onset.

References

  1. Conference
    1. Barrea A
    2. Delhaye BP
    3. Lefevre P
    4. Thonnard J-L
    (2017)
    Finger pad mechanics during dexterous object manipulation
    WorldHaptics Conference.
  2. Conference
    1. Bazen A
    2. Gerez S
    (2000)
    Directional field computation for fingerprints based on the principal component analysis of local gradients
    ProRISC 2000 Workshop on Circuits, Systems and Signal Processing.
    1. Bradski G
    (2008)
    The OpenCV library
    Dr Dobb’s J Softw Tools 25:120–125.
  3. Book
    1. Dawdy D
    2. Matalas N
    (1964)
    Statistical and Probability Analysis of Hydrologic Data, Part III: Analysis of Variance, Covariance and Time seriesHandbook of Applied Hydrology: A Compendium of Water-Resources Technology
    McGraw-Hill.
  4. Conference
    1. Levesque V
    2. Hayward V
    (2003)
    Experimental evidence of lateral skin strain during tactile exploration
    Proc Eurohaptics.
    1. Lucas B
    2. Kanade T
    (1981)
    An iterative image registration technique with an application to stereo vision
    IJCAI 130:121–130.
  5. Conference
    1. Tada M
    2. Mochimaru M
    3. Kanade T
    (2006)
    How does a fingertip slip? Visualizing partial slippage for modeling of contact mechanics
    Eurohaptics. pp. 2–7.
    1. Vallbo AB
    2. Johansson RS
    (1984)
    Properties of cutaneous mechanoreceptors in the human hand related to touch sensation
    Human Neurobiology 3:3–14.

Decision letter

  1. Cornelius Schwarz
    Reviewing Editor
  2. Richard B Ivry
    Senior Editor; University of California, Berkeley, United States
  3. Cornelius Schwarz
    Reviewer
  4. Rochelle Ackerley
    Reviewer

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

Delhaye et al., study the role of local tangential skin strain for tactile neuronal encoding in humans, variables that were not accessed by classic studies. They visualized the fingerprint when moving across a smooth surface together with extracellular recordings of primary afferents with a receptive field on or close to the fingerprint. Focusing on the period, in which the fingertip partially loses contact with the surface and starts to move (partial slip), they found that fast-adapting primary afferents type 1 (FA1) consistently respond to local strain, with predominance of compression over stretch, while the slowly adapting type 1 (SA1) do not. The FA1 responses during partial slip show selectivity to directional/orientation of the compression wave with some initial insights about the contribution of papillary ridges at the site of the neuron's receptive field.

Decision letter after peer review:

Thank you for submitting your article "High-resolution imaging of skin deformation shows that afferents from human fingertips signal slip onset" for consideration by eLife. Your article has been reviewed by 2 peer reviewers, including Cornelius Schwarz as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Richard Ivry as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Rochelle Ackerley (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

As the editors have judged that your manuscript is of interest, but as described below that additional experiments are required before it is published, we would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). First, because many researchers have temporarily lost access to the labs, we will give authors as much time as they need to submit revised manuscripts. We are also offering, if you choose, to post the manuscript to bioRxiv (if it is not already there) along with this decision letter and a formal designation that the manuscript is "in revision at eLife". Please let us know if you would like to pursue this option. (If your work is more suitable for medRxiv, you will need to post the preprint yourself, as the mechanisms for us to do so are still in development.)

Summary:

Delhaye et al., study the role of tangential skin deformation for tactile encoding in humans, variables that were not accessed by classic studies. They use visualizing the fingerprint when moving across a smooth surface together with extracellular recordings of primary afferents with a receptive field on or close to the fingerprint. 'Fast-adapting type 1, one out four classes of human primary afferents, are shown to respond to strain, rather than stretch when the fingerprint's adherence to the surfaces goes from a fixed state to a partial attachment to full slippage.

Essential revisions:

The reviews both see considerable merit in your work. They agree that it contains considerable advances in the field of human tactile coding by combining, for the first time, the visualization of skin strain patterns and microneurography. They state the results on FA1 will be important to students of perception as well prehension. Both generally support future publication. However, they state major comments as well, which I will summarize below. I consider the major points all important enough to expect you to address them either with new analyses or through improvement your treatment of them in the text.

1. The most important criticism was raised unanimously by the reviews. It should be addressed by adequate new analyses. It states that while making excellent points about FA1 afferents, there is a deficiency in covering SA1 primary afferents. (The other two classes are considered to be not sufficiently sampled and should be clearly labeled as circumstantial findings). This deficiency was considered most remarkable as SA1 do fire vigorously to presentation of the stimuli. The authors argue that the unexplained SA1 activity was mainly due to vertical skin deformation that they cannot monitor. In apparent contrast, however, the experiment was designed to abolish or minimize normal, i.e. vertical forces. The authors need to clarify this issue. They should explain what the purpose of robotic normal force control was, and importantly, in how far it worked (or did not work). What is the precision of their normal force measurement? Is the measurement suited to relate spike trains to time series of measured normal force? If yes, the reviewers recommend to use those measurements. In the discussion the authors need to clarify how their speculation about vertical forces activating SA1, relates to the robotic normal force control and measurements and Figure S1.

The reviewers feel that figure S1 is a good start, but more detailed analysis in this direction would be helpful. A possible temporal relationship of tangential deformations and SA1 firing is not sufficiently analyzed. The analyses used (STA and regression) are not systematically focused to bring out temporal relationships between SA1 and skin deformation. The implemented STA analysis seems to consider just one time-bin of 20 ms length for spiking and strain (the camera frame rate), while, in unexplained discrepancy, the regression analysis uses convolution of the spike train with a Gaussian window of 480 ms length – strongly diluting the temporal relationship of spike and strain parameters. The authors need to explain the rationale using the two extreme temporal settings in the two analyses. And they should apply a systematic approach to analyze temporal relationships of spikes and strain maps.

Maybe the authors even have runs/sessions without normal force control? If yes, these could be analyzed and presented.

Toward SA1 coding the reviewers specifically recommended to study/analyze the following:

Is it feasible that SA1 spiking dynamics are different for different directions of skin deformations (indentation vs. horizontal), e.g. could SA1 be slow for vertical but fast for horizontal stimulation, or vice versa, etc.?

The authors should go beyond their present attempts (mainly Figure S1), and build a model that is focused to find out the temporal relationship of strain/stretch/shear as well as normal forces and SA1 spikes.

Could the STA and regression analysis be helped by fitting responses to past skin deformations at longer delays than 20 ms (STA) or 12 ms (Figure S1)?

The authors use 'correlations' of deformation and spikes as basis for their arguments. Maybe it would be more appropriate to calculate delayed correlation, i.e. e.g cross correlation patterns to help to explain SA1 responses to the strain dynamics?

A further observation was that SA1 code for movement without slipping (in apparent contrast to FA1). A related phenomenon observed in the firing rate traces is a conspicuous silencing of SAI firing during movement (especially in the low friction condition). Is it feasible that SA1 signal touch well – albeit through the absence of firing?

It is interesting to note that the only afferents that have any response during movement, but not slipping, are SA units. This would be expected, but it is interesting that the non-slip moving phase show overall less firing than in the initial stationary load phase. Do the authors think that this could be an effect of the general decrease in firing frequency (i.e. adaptation) from SA units (with a re-increase in firing from slip) or is it more specific and somehow related to a difference in fingertip forces? This second idea about a real difference in firing between the stationary and moving-without-slip phases could imply that SA units do encode such specific aspects of touch, which FAs do not. From Suppl. Figure 2, it certainly seems that there are such differences.

2. Another unanimous point of the reviewers was the perceived insufficiency of the analysis of preferred direction. First of all, the definition (line 167) was not clear. Was it medial/lateral/radial/ulnar or in degrees (what does 0 deg refer to)?. The authors are recommended to calculate a standard directionality index (e.g. (pref-unpref)/(pref+unpref), vector sum, etc.). Further, the reviewers see the need to differentiate preference for 'directionality' from that for 'orientation'. In addition, it is recommended to clarify the following parameters related to the issue of directionality/orientation tuning:

a. Receptive field location

b. Orientation of papillary ridges inside the RF

c. Directionality relative to fingertip or papillary ridges?

In 8 of 13 FA1s the direction preference was not detailed. Was it like the other units or did they have no preference? Figure 3C and legend should be improved to clarify these questions.

3. From figure 4 it appears that responses to the same movement direction performed as 'forward' and 'backward' are very different. The reviewers felt that this phenomenon deserves a proper treatment. The authors are asked to clarify how consistent the phenomenon is, and whether there is a possible explanation in terms of different contexts that leads to it? It is considered worth to report whether SA1 show the same.

In Figure 4 please label the top and bottom series of strain maps – what exactly do they represent? High/low friction? The touch does not cover the receptive field in the bottom maps, what does that mean, which effect may it have on firing?

4. The reviews also noted that STA analysis was only done with ||e||. As a main conclusion of the paper is about responses to strain versus stress, shouldn't STA be performed with exx, eyy, exy to differentiate those parameters? In this respect the question was raised why the authors focus on the annular pattern. Given the annular form of the partial slip and the high-correlation of deformations across the fingertip such a pattern is trivially expected (especially using ||e||). The authors are recommended to use the mentioned more specific parameters exx, eyy, exy, and focus first on the RF, and second on deformations elsewhere (maybe using some decorrelation techniques).

5. It was further suggested to attempt using GLMs or similar to capture non-linearities between strain and spikes. The rigid linear model will fail to capture expected non-linear relationships between spike rates and other parameters.

6. Both reviewers found the part of the discussion stimulating, in which the authors touch upon orientation selectivity (line 351-354). It seems an interesting point to consider the differences between slips and edges – both in forces and neural information. Do the authors have evidence to think the brain could tell the difference between a slip and an edge? Alternatively, the speculation starts to explain orientation selectivity of primate tactile neurons in terms of prehension (rather than perception of edge orientation, as is typically assumed). This point is considered to be an outcome of their novel measurements. It should be worked out and presented more prominently.

https://doi.org/10.7554/eLife.64679.sa1

Author response

Essential revisions:

The reviews both see considerable merit in your work. They agree that it contains considerable advances in the field of human tactile coding by combining, for the first time, the visualization of skin strain patterns and microneurography. They state the results on FA1 will be important to students of perception as well prehension. Both generally support future publication. However, they state major comments as well, which I will summarize below. I consider the major points all important enough to expect you to address them either with new analyses or through improvement your treatment of them in the text.

Thank you for this very deep and thorough review. We greatly appreciate the reviewers’ effort to help us to improve the paper. We have tried to address each comment adequately and this resulted in significant revisions of the manuscript. The manuscript has been much improved thanks to the reviewers’ feedback.

Overall, we believe that several major comments made by the reviewers at least partly originate from a failure on our side to make a clear distinction between “global” and “local” aspects of the stimulation and the measurements used in our study. On the one hand, the afferent responses were briefly studied with respect to “global” parameters, such as the different stimulus phases (tangential movement: onset, partial slip and plateau; Figure 3), the stimulus direction (i.e., ulnar, distal, radial and proximal; Suppl. Figure 1) or the force measurements (Suppl. Figure 2). This short analysis was aimed at providing a general context and relating this study to previous works covering those aspects in more details (e.g. Birznieks et al., 2001 JNeuroSci, Johansson and Birznieks 2004 Nat NeuroSci, Birznieks et al. 2009 JNeuroSci, Khamis et al. 2015 JNeuroPhysiol). On the other hand, however, our main focus, which is the main originality of our work, was to relate the afferent responses to “local” skin deformations which occur within the afferent’s receptive field and which can be measured with our optical system (Figure 2, 4, 5, 6 and Supp Figure 3, 4, 5 and 6). We have made numerous changes at several places in the manuscript (detailed below) to make this crucial difference clearer. Moreover, we added several important details in the Methods to better explain our choices. Finally, to address the reviewers’ major concerns, we provide new analyses that are reported in the Methods and also in the new supplementary figures 3 and 6.

Note that the order of the Supplementary Figures 1 and 2 has been flipped to accommodate their order of appearance in the main text.

1. The most important criticism was raised unanimously by the reviews. It should be addressed by adequate new analyses. It states that while making excellent points about FA1 afferents, there is a deficiency in covering SA1 primary afferents. (The other two classes are considered to be not sufficiently sampled and should be clearly labeled as circumstantial findings). This deficiency was considered most remarkable as SA1 do fire vigorously to presentation of the stimuli.

Yes, indeed, the SA-I afferents do fire vigorously to the tangential loading. In Supplementary figure 2 (former Supplementary figure 1), we show that these vigorous responses are at least partially related to the fluctuations of the tangential forces, a global measurement. Nevertheless, we failed to find any correlate of those discharges in our “local” measurements of the skin strain. The lack of such relationship may suggest that the SA-I afferents encode other, non-measured aspects of the skin deformation. This is the only conclusion that we can draw with respect to the SA-I afferent responses to the “local” measured skin deformation which we aimed to explore in this study.

Following this comment, in the Discussion, we re-wrote the paragraph about the lack of relationship between SA-I responses and local skin strain hypothesizing two possible aspects of the deformations that might be encoded by these afferents.

“Our methods generate high-resolution representations of the distribution of surface strain in the contact area of the fingertip (i.e. 2D, x and y axes) but do not allow measurements of the distribution of the deformations perpendicular to the fingertip surface inside the contact area (i.e. along the z-axis), nor outside the contact area. […] When considering the subdermal location of SA-II and FA-II afferents terminals, it is not surprising to observe a poor coupling between their responses and the measured surface strains.”

We would also like to stress that due to the complexity of the microneurography experiment, i.e. finding an afferent with a receptive field inside the contact area and maintaining the recording sufficiently long to finish a full protocol, we were able to obtain data only from two SA-I afferents fulfilling these criteria (six in total, but two inside the contact area). This is another reason why the main focus of the analyses is on the responses from FA-I afferents.

The authors argue that the unexplained SA1 activity was mainly due to vertical skin deformation that they cannot monitor. In apparent contrast, however, the experiment was designed to abolish or minimize normal, i.e. vertical forces. The authors need to clarify this issue. They should explain what the purpose of robotic normal force control was, and importantly, in how far it worked (or did not work). What is the precision of their normal force measurement? Is the measurement suited to relate spike trains to time series of measured normal force? If yes, the reviewers recommend to use those measurements. In the discussion the authors need to clarify how their speculation about vertical forces activating SA1, relates to the robotic normal force control and measurements and Figure S1.

We believe that there was a confusion in the Discussion about the SA-I afferents with respect to their response to global parameters, i.e. normal and tangential force (measured), and to the local skin deformations, normal (not measured) and tangential (measured) to the surface. What is proposed in the Discussion is that SA-I afferents may be sensitive to the strains normal to the skin surface (not measured) rather than to the tangential strains we measured, which failed to explain SA-I firing rates. In the revised manuscript under Limitations of the study we have described this in detail and also throughout the manuscript we have now attempted to make the distinction between global and local parameters very clear (see response to comment #1).

The robot controller was intended to maintain the normal force constant and was successful with a RMS error of only 5% (average peak error is 12%) and our force measurements were precise. The normal force control is necessary because if the surface is maintained at a constant vertical position, the lateral movement and increase of tangential force generate an increase in normal force (i.e. “poynting effect”). In order to dissociate the two forces, it is therefore necessary to lower the vertical position as the tangential movement is launched.

We have modified the Methods and the last paragraph of the Results to accommodate this information. Given our sampling frequency (1 kHz), the measurements are indeed well suited to relate the time series to the spike trains.

Methods:

“A custom PID controller was developed to servo-control the normal (vertical) force applied to the fingerpad by feeding back the force measurements (average RMS error is 5% during the whole tangential loading and average peak error is 12% or 0.5N).”

Results:

“Finally, we asked how much the response of all types of afferents was related to the “global” external 3D force vector. […] Their firing rates were however not related to the small vertical (normal) force fluctuations.”

The reviewers feel that figure S1 is a good start, but more detailed analysis in this direction would be helpful. A possible temporal relationship of tangential deformations and SA1 firing is not sufficiently analyzed. The analyses used (STA and regression) are not systematically focused to bring out temporal relationships between SA1 and skin deformation. The implemented STA analysis seems to consider just one time-bin of 20 ms length for spiking and strain (the camera frame rate), while, in unexplained discrepancy, the regression analysis uses convolution of the spike train with a Gaussian window of 480 ms length – strongly diluting the temporal relationship of spike and strain parameters. The authors need to explain the rationale using the two extreme temporal settings in the two analyses. And they should apply a systematic approach to analyze temporal relationships of spikes and strain maps.

First, it is important to stress that we chose a relatively slow tangential speed for the robot (5 mm/s), for two limiting reasons. 1) so that the stick to slip transition could be accurately measured at the frame rate (50 fps) and 2) so that the normal force control could be effective.

Second, the image system time resolution is indeed 20 ms (50 Hz), which is why our STA analyses were performed at that resolution.

Third, we checked that the afferent response frequency content matched the optical recording system by estimating their “effective sampling frequency” (see references below). We found that indeed, the stimulus did not generate effective information above 50 Hz in the afferent responses (median effective frequency 15 Hz, quartiles 3.5 – 38.5 Hz). The STA analysis with a 20 ms time-bin resolution is therefore well suited to capture all the frequency spectra of the afferent responses.

Finally, the Gaussian window width (effective cutoff 4 Hz) used to smooth the spike trains was found such that it empirically maximized the prediction performance across afferents. Such a low cutoff is not surprising given the elements above, and suggests that our stimulus actual frequency content was centered at 4 Hz.

We have amended the Methods to accommodate the above-mentioned information.

“Afferents’ effective sampling frequency. […] We found that indeed, the stimulus did not generate effective information above 50 Hz; across afferents, median effective sampling frequency was 15 Hz (Q1-3 = 3.5 – 38.5 Hz). Such low frequency content is explained by the slow nature of the stimulus.”

“The vector of the spike times, sampled at 1 kHz and made of 1’s at the time of a spike and 0’s elsewhere, was first convolved by a Gaussian window (using MATLAB’s gausswin function, with 480 points and an alpha value of 6, normed) and multiplied by 1000. The windows width was empirically found to maximize the prediction performance across afferents (suggesting that our stimulus frequency content is centered around 4 Hz).”

Dawdy DR and Matalas NC (1964). Statistical and probability analysis of hydrologic data, part III: Analysis of variance, covariance and time series. In Handbook of Applied Hydrology, a Compendium of Water-Resources Technology, ed. Chow VT, pp. 68–90. McGraw-Hill Co., New York.

Dimitriou M, Edin BB (2008) Discharges in human muscle spindle afferents during a key-pressing task. J Physiol 586:5455–5470;

Maybe the authors even have runs/sessions without normal force control? If yes, these could be analyzed and presented.

No, we did not run such sessions.

Toward SA1 coding the reviewers specifically recommended to study/analyze the following:

Is it feasible that SA1 spiking dynamics are different for different directions of skin deformations (indentation vs. horizontal), e.g. could SA1 be slow for vertical but fast for horizontal stimulation, or vice versa, etc.?

Unfortunately, we do not have the data for such analysis, nor the methods to obtain such data (our imaging technique does not allow to study perpendicular skin deformation). We were specifically interested in the tangential loading of the fingerpad in the context of grasp stability thus we did not strive to study skin indentation.

The authors should go beyond their present attempts (mainly Figure S1), and build a model that is focused to find out the temporal relationship of strain/stretch/shear as well as normal forces and SA1 spikes.

Our local measurements provide a mechanistic description of the deformations in the receptive field of an afferent, while the “global” force measurements are a consequence of the global deformation of the fingertip. We believe that combining “global” and “local” parameters into a single model will not help to explain the relationship between the skin afferent responses and local skin deformations. Adding the forces to the model might increase the amount of variance that can be explained by the model but will not inform us about the nature of the local skin deformations that generated spiking activity (the aim of our study).

Could the STA and regression analysis be helped by fitting responses to past skin deformations at longer delays than 20 ms (STA) or 12 ms (Figure S1)?

The authors use 'correlations' of deformation and spikes as basis for their arguments. Maybe it would be more appropriate to calculate delayed correlation, i.e. e.g cross correlation patterns to help to explain SA1 responses to the strain dynamics?

Thank you for the suggestion. Unfortunately, given the slow nature of our stimuli (see above), the frequency content of the response, the short pathway and the firing properties of the primary afferent and the image sampling frequency, we do not believe that this approach is appropriate.

A further observation was that SA1 code for movement without slipping (in apparent contrast to FA1). A related phenomenon observed in the firing rate traces is a conspicuous silencing of SAI firing during movement (especially in the low friction condition). Is it feasible that SA1 signal touch well – albeit through the absence of firing?

We are not sure that we understood the reviewers first point and to which figure it referred. As mentioned above and now in the Discussion paragraph about the SA-I afferents responses, the pressure distribution can change during tangential loading and might be strongly reduced under certain circumstances for afferents close to the border of the contact, which might explain the silencing. Again, we did not measure the pressure distribution so this is only speculation. Moreover, the rolling of the finger could make the afferents lose contact with the glass plate. However, following deep data inspection, we did not find any systematic silencing when the contact was lost.

It is interesting to note that the only afferents that have any response during movement, but not slipping, are SA units. This would be expected, but it is interesting that the non-slip moving phase show overall less firing than in the initial stationary load phase. Do the authors think that this could be an effect of the general decrease in firing frequency (i.e. adaptation) from SA units (with a re-increase in firing from slip) or is it more specific and somehow related to a difference in fingertip forces? This second idea about a real difference in firing between the stationary and moving-without-slip phases could imply that SA units do encode such specific aspects of touch, which FAs do not. From Suppl. Figure 2, it certainly seems that there are such differences.

We are not sure how to understand this comment, mainly because there is no “non-slip moving” phase. The main focus of our analyses was on the afferent responses during tangential movement of the plate against the fingerpad i.e., from the moment the fingerpad started slipping until it was in a full slip and the three moving phases were identified as the onset, partial slip and plateau. We monitored skin strains during a transition from a fully stuck contact to a fully slipping contact and tried to relate them to the afferents’ responses. We do not report any results (just display the firing rates from whole trials on the Figure 2 and Supplementary figure 1) from the periods before the tangential movement started or after it fully slipped.

2. Another unanimous point of the reviewers was the perceived insufficiency of the analysis of preferred direction. First of all, the definition (line 167) was not clear. Was it medial/lateral/radial/ulnar or in degrees (what does 0 deg refer to)?. The authors are recommended to calculate a standard directionality index (e.g. (pref-unpref)/(pref+unpref), vector sum, etc.). Further, the reviewers see the need to differentiate preference for 'directionality' from that for 'orientation'.

We believe that this comment is also related to the confusion between global and local direction/orientation preference. On the one hand, we describe a “global” preferred direction related to the robot movement (P/U/R/D). On the other hand, we show that the afferents preferentially encoded “local” compressive strain rates along a given orientation (in degrees).

We have made the following changes to avoid this ambiguity:

In the paragraph related to Figure 3, we gave more details about how we defined the “global” preferred direction, that is in relation with the plate movement and which can only be one of the four directions (ulnar, distal radial or proximal). We changed the Figure 3B to remove angles in degrees, and report direction with respect to the preferred (North) as pointing to cardinal direction (East, South and West). When possible, we also added a word “global” to the preferred directions to avoid confusion, i.e. “preferred global directions”.

“First, we describe how the tactile afferent responded with respect to “global” stimulus parameters such as the movement phase or direction. […] While the firing rates were strongly correlated across friction levels (Figure 3E; correlation r = 0.95, n = 176, p < 0.001), the firing rates across the whole population tended to be slightly lower for low friction (paired t-test, t(171) = 7.57, p < 0.001) and the ratio of the mean firing rate during low and high friction condition was 0.82 ± 0.55 (mean ± standard deviation).”

In the paragraph related to Figure 6, we added two sentences to clarify the concept of local strain orientation preference. We also always use “preferred strain orientation” to avoid confusion.

“It is important to avoid the confusion between the units’ preferred direction described in Figure 3, which relates to the robot movement direction and the maximal firing rate of the afferent, and the strain orientation preference described here, which is related to the orientation of the local deformation in the afferent receptive field.”

Finally, we realized that we had been using the terms “orientation” and “direction” interchangeably at several places in the text, which is very unfortunate given the importance of the distinction between them. We verified that all instances of the use of those words were corrected.

In addition, it is recommended to clarify the following parameters related to the issue of directionality/orientation tuning:

a. Receptive field location

b. Orientation of papillary ridges inside the RF

c. Directionality relative to fingertip or papillary ridges?

Thank you for this important point. The receptive field location for all neurons is shown in Supp Figure 1. We now also documented the orientation of the papillary ridges for the six FA-I afferents that we could study in details with respect to the strain in a new Supp. Figure 6, which shows the strain orientation preference (obtained from Supp Figure 5) vs the orientation across the fingerprint ridges around their RFs. We also added the local fingerprint pattern for clarity and transparency.

In 8 of 13 FA1s the direction preference was not detailed. Was it like the other units or did they have no preference? Figure 3C and legend should be improved to clarify these questions.

The movement direction preference (Figure 3) was computed for all afferents, and all of them are shown in Figure 3C.

The strain orientation preference was computed only for the afferents for which we had strain data, i.e. for the 6 afferents with their receptive fields inside the contact area. We clarified this point in the paragraph related to the “Aspects of the skin strain rates encoded by the FA-I afferents”.

“Having demonstrated that the recorded FA-I afferents respond to local strain patterns, we then sought to uncover what aspects of the strains were responsible for these responses. […] The analysis was performed only on the FA-I afferents for which we had optical measurements of the strains, that is, those having their receptive fields inside the contact area (n = 6, all shown in the top line of Figure 5A).”

3. From figure 4 it appears that responses to the same movement direction performed as 'forward' and 'backward' are very different. The reviewers felt that this phenomenon deserves a proper treatment. The authors are asked to clarify how consistent the phenomenon is, and whether there is a possible explanation in terms of different contexts that leads to it? It is considered worth to report whether SA1 show the same.

We believe that this comment might stem from a misunderstanding of the figure layout. Forward movement (Figure 4A left), should be compared with backward in the same direction (Figure 4B, right).

As shown in Figure 3D, all the afferents had very similar firing rates in the forward and the backward movements. The afferent shown in Figure 4 follows the same trend: the firing rates elicited in the proximal direction (Figure 4A left and Figure 4B right) are very similar. The timing is different because the timing of the strain wave is different.

One difference is the short burst elicited at the movement onset in the distal backward condition. Unfortunately, the local strains do not explain that burst. Such burst at the very onset of the movement only occurred in a small fraction of the afferents (cfr Figure 3A). We added a sentence to clarify this point.

“Note that a short burst was elicited at the onset of the movement in the distal direction in the backward case (Figure 4A, right). Such transient burst cannot be explained by our strain measurements and occurred in a small fraction of the trials and only in few afferents (Figure 3A).”

In Figure 4 please label the top and bottom series of strain maps – what exactly do they represent? High/low friction? The touch does not cover the receptive field in the bottom maps, what does that mean, which effect may it have on firing?

The two rows show the maps for the two friction conditions. We added labels. And yes, due to rolling, part of the RF could lose contact with the plate. This is part of the unexplained variance of our models. We mentioned this element at the end of the first paragraph of the “FA-I afferents respond to local skin strain rates” section.

“Also note that part of the unit receptive field lost contact during the partial slip phase in the high friction case.”

4. The reviews also noted that STA analysis was only done with ||e||. As a main conclusion of the paper is about responses to strain versus stress, shouldn't STA be performed with exx, eyy, exy to differentiate those parameters? In this respect the question was raised why the authors focus on the annular pattern. Given the annular form of the partial slip and the high-correlation of deformations across the fingertip such a pattern is trivially expected (especially using ||e||). The authors are recommended to use the mentioned more specific parameters exx, eyy, exy, and focus first on the RF, and second on deformations elsewhere (maybe using some decorrelation techniques).

We thank the reviewers for this suggestion and we agree that it is valuable to look at the STA for more specific parameters. Instead of looking at exx, eyy and exy which are specific to a choice of reference frame (and are further analyzed with the linear regression model in Figure 6), we choose to use e1 and e2, the principal strains. The results are shown in a new Supplementary Figure 3, and are consistent with the linear model results. Indeed, while the STA of the compressive component matches the location of the RF, the STA of the tensile component is weaker and does not match the RF location.

We added a paragraph in the Results to cover this part. Given that the two approaches (STA and linear model) provide complementary but converging evidence, we did not estimate necessary to put this figure in the main text.

“The same STA analysis was repeated using the two principal strain components, one compressive and one tensile, separately to build two STA maps (see Materials and methods). […] Moreover, we found that the compressive STA peaks were generally larger and more often found in the afferent receptive field than their tensile counterpart, suggesting that FA-I afferents are more sensitive to compression. This finding will be further supported in the next section.”

Note, we do not mention at any point stresses, as we do not measure them. We only measure and discuss the strains. We suppose this comment is about stretch vs compression. Yes, indeed, an annular pattern is trivially expected IF the afferents encode something about the strains, as already mentioned in the manuscript. As the first goal of this STA analysis is to confirm the link between strain and spiking, it is reasonable to use it, and it explains the choice of the norm of the strain rates as well, which summarizes the strains in a single variable.

5. It was further suggested to attempt using GLMs or similar to capture non-linearities between strain and spikes. The rigid linear model will fail to capture expected non-linear relationships between spike rates and other parameters.

The modeling approach is not aimed at perfectly predicting the firing activity of the afferents, but rather uncover which aspects of the skin mechanics are encoded (i.e. using model selection). Building a more accurate firing model is a complex problem, given the afferent branching for instance but also the precise spike timing of the afferents; that would require a lot more data with more experimental conditions and a better temporal resolution, as already mentioned in the Discussion. We are actually currently trying this with new experiments.

6. Both reviewers found the part of the discussion stimulating, in which the authors touch upon orientation selectivity (line 351-354). It seems an interesting point to consider the differences between slips and edges – both in forces and neural information. Do the authors have evidence to think the brain could tell the difference between a slip and an edge? Alternatively, the speculation starts to explain orientation selectivity of primate tactile neurons in terms of prehension (rather than perception of edge orientation, as is typically assumed). This point is considered to be an outcome of their novel measurements. It should be worked out and presented more prominently.

An actual edge would be a more complete stimulus, given that it would also add an important local vertical compression (probably explaining SA-I strong sensitivity to edges), and its symmetrical nature (step up and then step down) would generate tangential compression and then stretch. Therefore, the comparison is only limited, and we do not want to speculate about it too much in the Discussion.

https://doi.org/10.7554/eLife.64679.sa2

Article and author information

Author details

  1. Benoit P Delhaye

    1. Institute of Information and Communication Technologies, Electronics and Applied Mathematics, Université catholique de Louvain, Louvain-la-Neuve, Belgium
    2. Institute of Neuroscience, Université catholique de Louvain, Brussels, Belgium
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    For correspondence
    delhayeben@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3974-7921
  2. Ewa Jarocka

    Department of Integrative Medical Biology, Umeå University, Umeå, Sweden
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Allan Barrea

    1. Institute of Information and Communication Technologies, Electronics and Applied Mathematics, Université catholique de Louvain, Louvain-la-Neuve, Belgium
    2. Institute of Neuroscience, Université catholique de Louvain, Brussels, Belgium
    Contribution
    Data curation, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1094-4596
  4. Jean-Louis Thonnard

    1. Institute of Information and Communication Technologies, Electronics and Applied Mathematics, Université catholique de Louvain, Louvain-la-Neuve, Belgium
    2. Institute of Neuroscience, Université catholique de Louvain, Brussels, Belgium
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - review and editing
    Competing interests
    No competing interests declared
  5. Benoni Edin

    Department of Integrative Medical Biology, Umeå University, Umeå, Sweden
    Contribution
    Conceptualization, Supervision, Investigation, Writing - review and editing
    Competing interests
    No competing interests declared
  6. Philippe Lefèvre

    1. Institute of Information and Communication Technologies, Electronics and Applied Mathematics, Université catholique de Louvain, Louvain-la-Neuve, Belgium
    2. Institute of Neuroscience, Université catholique de Louvain, Brussels, Belgium
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2032-3635

Funding

European Space Agency

  • Jean-Louis Thonnard
  • Philippe Lefèvre

PRODEX (BELSPO)

  • Jean-Louis Thonnard
  • Philippe Lefèvre

Swedish Research Council (VR 2016-01635)

  • Benoni Edin

Fonds De La Recherche Scientifique - FNRS (FNRS)

  • Benoit P Delhaye

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Hannes Saal, Vincent Hayward, and the members of the DEX group at UCLouvain for useful comments on a previous version of the manuscript. We also thank Julien Lambert for invaluable logistical support. This work was supported by a grant from the European Space Agency, Prodex (BELSPO, Belgian Federal Government), and the Swedish Research Council (VR 2016-01635). BPD is a postdoctoral researcher of the Fonds de la Recherche Scientifique – FNRS (Belgium).

Ethics

Human subjects: Each subject provided written informed consent to the procedures, and the study was approved by the local ethics committee at the host institution (Institute of Neuroscience, Université catholique de Louvain, Brussels, Belgium).

Senior Editor

  1. Richard B Ivry, University of California, Berkeley, United States

Reviewing Editor

  1. Cornelius Schwarz

Reviewers

  1. Cornelius Schwarz
  2. Rochelle Ackerley

Publication history

  1. Received: November 6, 2020
  2. Accepted: April 13, 2021
  3. Accepted Manuscript published: April 22, 2021 (version 1)
  4. Version of Record published: June 1, 2021 (version 2)
  5. Version of Record updated: June 7, 2021 (version 3)

Copyright

© 2021, Delhaye et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 778
    Page views
  • 113
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    Kristen Lee, Chris Q Doe
    Research Advance Updated

    Individual neurons can undergo drastic structural changes, known as neuronal remodeling or structural plasticity. One example of this is in response to hormones, such as during puberty in mammals or metamorphosis in insects. However, in each of these examples, it remains unclear whether the remodeled neuron resumes prior patterns of connectivity, and if so, whether the persistent circuits drive similar behaviors. Here, we utilize a well-characterized neural circuit in the Drosophila larva: the moonwalker descending neuron (MDN) circuit. We previously showed that larval MDN induces backward crawling, and synapses onto the Pair1 interneuron to inhibit forward crawling (Carreira-Rosario et al., 2018). MDN is remodeled during metamorphosis and regulates backward walking in the adult fly. We investigated whether Pair1 is remodeled during metamorphosis and functions within the MDN circuit during adulthood. We assayed morphology and molecular markers to demonstrate that Pair1 is remodeled during metamorphosis and persists in the adult fly. MDN-Pair1 connectivity is lost during early pupal stages, when both neurons are severely pruned back, but connectivity is re-established at mid-pupal stages and persist into the adult. In the adult, optogenetic activation of Pair1 resulted in arrest of forward locomotion, similar to what is observed in larvae. Thus, the MDN-Pair1 neurons are an interneuronal circuit – a pair of synaptically connected interneurons – that is re-established during metamorphosis, yet generates similar locomotor behavior at both larval and adult stages.

    1. Neuroscience
    Philipp Schlegel et al.
    Research Article Updated

    The hemibrain connectome provides large-scale connectivity and morphology information for the majority of the central brain of Drosophila melanogaster. Using this data set, we provide a complete description of the Drosophila olfactory system, covering all first, second and lateral horn-associated third-order neurons. We develop a generally applicable strategy to extract information flow and layered organisation from connectome graphs, mapping olfactory input to descending interneurons. This identifies a range of motifs including highly lateralised circuits in the antennal lobe and patterns of convergence downstream of the mushroom body and lateral horn. Leveraging a second data set we provide a first quantitative assessment of inter- versus intra-individual stereotypy. Comparing neurons across two brains (three hemispheres) reveals striking similarity in neuronal morphology across brains. Connectivity correlates with morphology and neurons of the same morphological type show similar connection variability within the same brain as across two brains.