1. Neuroscience
Download icon

Perception of microstimulation frequency in human somatosensory cortex

  1. Christopher L Hughes  Is a corresponding author
  2. Sharlene N Flesher
  3. Jeffrey M Weiss
  4. Michael Boninger
  5. Jennifer L Collinger
  6. Robert A Gaunt  Is a corresponding author
  1. Rehab Neural Engineering Labs, University of Pittsburgh, United States
  2. Department of Bioengineering, University of Pittsburgh, United States
  3. Center for the Neural Basis of Cognition, University of Pittsburgh, United States
  4. Department of Neurosurgery, Stanford University, United States
  5. Department of Electrical Engineering, Stanford University, United States
  6. Department of Physical Medicine and Rehabilitation, University of Pittsburgh, United States
  7. Human Engineering Research Laboratories, VA Center of Excellence, Department of Veterans Affairs, United States
Research Article
  • Cited 1
  • Views 834
  • Annotations
Cite this article as: eLife 2021;10:e65128 doi: 10.7554/eLife.65128

Abstract

Microstimulation in the somatosensory cortex can evoke artificial tactile percepts and can be incorporated into bidirectional brain–computer interfaces (BCIs) to restore function after injury or disease. However, little is known about how stimulation parameters themselves affect perception. Here, we stimulated through microelectrode arrays implanted in the somatosensory cortex of two human participants with cervical spinal cord injury and varied the stimulus amplitude, frequency, and train duration. Increasing the amplitude and train duration increased the perceived intensity on all tested electrodes. Surprisingly, we found that increasing the frequency evoked more intense percepts on some electrodes but evoked less-intense percepts on other electrodes. These different frequency–intensity relationships were divided into three groups, which also evoked distinct percept qualities at different stimulus frequencies. Neighboring electrode sites were more likely to belong to the same group. These results support the idea that stimulation frequency directly controls tactile perception and that these different percepts may be related to the organization of somatosensory cortex, which will facilitate principled development of stimulation strategies for bidirectional BCIs.

Introduction

Bidirectional brain–computer interfaces (BCI) can restore lost function to people living with damage to the brain, spine, and limbs (Collinger et al., 2018; Fetz, 2015; Flesher et al., 2021; Hughes et al., 2020). BCI users can control an end effector using neural activity recorded from motor cortex and receive sensory feedback through intracortical microstimulation (ICMS) in somatosensory cortex (Flesher et al., 2021). Beyond the practical aim of restoring sensation to improve motor function, existing bidirectional BCIs in human participants provide an unprecedented ability to investigate the nature of sensory perception.

The behavioral effects of ICMS in somatosensory cortex have been studied in detail in non-human primates (NHPs) (Dadarlat et al., 2015; Kim et al., 2015a; Kim et al., 2015b; Romo et al., 2000; Romo et al., 1998). However, animals are limited in their ability to perform certain psychophysical tasks. NHPs can learn to discriminate between two or more stimuli, and their ability to perform these tasks can provide insight into how stimulus parameters affect sensory perception. However, they can never describe the qualitative nature of the sensory percepts, nor can they be trained to perform more complex psychophysical tasks such as magnitude estimation. NHP studies can therefore lead to hypotheses about how stimulus parameters affect qualitative aspects of perception, but only human studies can investigate these directly.

Limited work has been conducted in humans using ICMS of somatosensory cortex to restore sensation (Armenta Salas et al., 2018; Fifer et al., 2020; Flesher et al., 2016). From these studies we know that ICMS can evoke tactile sensations that are perceived to originate from the hands (Fifer et al., 2020; Flesher et al., 2016) and arms (Armenta Salas et al., 2018) and that the stimulation locations in the cortex that elicit these percepts agree with known cortical somatotopy (Penfield and Boldrey, 1937). Participants reported naturalistic sensations such as ‘pressure’ and ‘touch’ (Flesher et al., 2016) as well as ‘squeeze’ and ‘tap’ (Armenta Salas et al., 2018), but the quality and naturalness varied between stimulated electrodes within each participant. Additionally, all studies found that increasing the stimulus current amplitude consistently increased the perceived intensity of the tactile percepts. The effect of stimulus pulse frequency has been less studied, although low frequencies may require higher amplitudes to evoke a detectable percept (Armenta Salas et al., 2018).

More is known about the perceptual effects of stimulating the human thalamus (Davis et al., 1996; Dostrovsky et al., 1993Heming et al., 2010; Ohara et al., 2004; Swan et al., 2018; Willsey et al., 2020). Similar to ICMS, increasing the stimulation amplitude increased percept intensity (Dostrovsky et al., 1993Swan et al., 2018). However, changing stimulation frequency and temporal patterns have had different effects on perception. In some cases high-frequency stimulation (333 Hz) evoked the most natural percepts (Heming et al., 2010), while in others it evoked primarily paresthesias and low-frequencies produced "tapping" sensations (Dostrovsky et al., 1993). In other cases burst stimulation evoked more natural percepts than tonic stimulation (Willsey et al., 2020) and two-pulse bursts evoked less natural sensations (Swan et al., 2018). Ultimately, temporal factors have clear effects on the sensations evoked through thalamic stimulation, but it remains unclear how to optimally control these parameters to manipulate percept quality.

It has often been suggested that increasing the stimulus frequency increases the perceived intensity of a stimulus train. Increasing the pulse frequency of ICMS reduced the current amplitude required to evoke a detectable percept in NHPs (Kim et al., 2015a; Romo et al., 2000; Romo et al., 1998) and rats (Butovas and Schwarz, 2007; Semprini et al., 2012). This was thought to indicate that increasing pulse frequency increased perceived intensity. Additionally, in a frequency discrimination task, increasing amplitude biased NHPs (Callier et al., 2020) and rats (Fridman et al., 2010) toward selecting stimulus trains as having higher frequencies, providing further evidence that increasing pulse frequency increases perceived intensity. Perceived intensity also increases as stimulation amplitude and frequency are increased in human peripheral nerves (Graczyk et al., 2016) and human visual cortex (Schmidt et al., 1996). This is also true for mechanical stimuli where perceived intensity increased with increasing vibration frequency in able-bodied human participants using tactile input to the hand (Hollins and Roy, 1996; Muniak et al., 2007; Verrillo et al., 1969). Overall, these results imply that both electrical and mechanical stimulation with higher frequency components are perceived as being more intense. Our goal here was to understand whether this same principle applies to ICMS of human somatosensory cortex and to evaluate whether perceptual qualities were affected by changes in stimulus pulse frequency.

In ongoing experiments, we implanted microelectrode arrays into the motor and somatosensory cortices of two participants (referred to as P2 and P3) with cervical spinal cord injuries to evaluate the safety and efficacy of bidirectional BCIs and to study sensorimotor control in humans. In P2, ICMS of somatosensory cortex evoked tactile percepts that felt like they originated from the paralyzed hand (Flesher et al., 2016). However, the percepts themselves varied considerably, from more natural sensations, such as touch and pressure, to less natural sensations, such as vibration and tingle. In order to represent more complex and intuitive tactile inputs with ICMS, it is critical that we understand how stimulus parameters directly affect sensation.

We are particularly interested in how stimulus parameters, such as current amplitude, pulse frequency, and train duration, change the perceived intensity of tactile percepts. The ability to control perceived intensity in a bidirectional BCI will be essential, as modulated sensory feedback is crucial for object interaction (Johansson and Flanagan, 2009; Nowak et al., 2013). While grasp contact could be relayed by simple on–off stimulation, conveying grip force, which is essential for grasp stability, efficiency, and precision (Godfrey et al., 2016; Nowak et al., 2004; Nowak and Hermsdörfer, 2006), requires the ability to modulate the perceived intensity of a stimulus. We sought to assess the effects of changing the stimulus pulse frequency on several perceptual metrics in two participants, P2 and P3, and expected to see increases in the perceived intensity as the stimulus pulse frequency increased.

Results

Effects of frequency on perceived intensity are electrode-dependent

In participant P2, we delivered ICMS trains through individual electrodes and asked him to report the perceived intensity on a self-selected scale, which typically ranged from 0 to 4. We found that increasing the stimulus current amplitude and train duration consistently increased the perceived intensity of the evoked sensations on all tested electrodes (Figure 1—figure supplement 1). However, the relationship between stimulus frequency and perceived intensity was electrode dependent (Figure 1). We delivered a 60 μA stimulus train for 1 s at pulse frequencies ranging from 20 to 300 Hz. On some electrodes, percept intensity increased with stimulus pulse frequency (Figure 1B). However, on over half of the tested electrodes, the opposite effect occurred; stimulus trains with low pulse frequencies (20–100 Hz) were perceived as being the most intense and the intensity decreased as the stimulus pulse frequency increased (Figure 1C,D). We used k-means clustering to separate electrodes into three categories based on the reported percept intensity at 20, 100, and 300 Hz (Figure 1—figure supplement 2): electrodes with the highest intensity response at 20 Hz (Figure 2A), electrodes with the highest intensity responses at 100 Hz (Figure 2B), and electrodes with the highest intensity response at 300 Hz (Figure 2C). For simplicity, we refer to these groups based on the pulse frequency range at which the maximal intensity occurred: high-frequency preferring (HFP), intermediate-frequency preferring (IFP), and low-frequency preferring (LFP) electrodes. These electrode groups varied in both the median-reported intensity across all frequencies as well as the frequency at which the maximum intensity occurred.

Figure 1 with 4 supplements see all
Pulse frequency drives electrode-specific changes in intensity which can be grouped into three categories.

(A) Perceived intensity for each aggregated frequency preference group. Different colors represent different categories. Each data point shows the mean intensity response of all of the electrodes in a given category. (B) Perceived intensity for two examples of high-frequency preferring electrodes that evoked the most intense percepts at the highest pulse frequencies and that generated the least intense percepts overall. (C) Perceived intensity for two examples of intermediate-frequency preferring electrodes that generated the most intense overall percepts, which occurred between 40 Hz and 100 Hz. (D) Perceived intensity for two examples of low-frequency preferring electrodes, which generated intermediate overall intensities that were maximized between 20 and 100 Hz. Error bars represent the standard error. The points are connected with piecewise fits. Axes are scaled differently between panels for clarity.

Figure 1—source data 1

This file contains all the magnitude estimation data from participant P2 using an amplitude of 60 μA and frequencies of 20, 40, 60, 80, 100, 150, 200, 250, and 300 Hz.

Each sheet contains data for one of the three frequency preference groups (LFP, IFP, or HFP). Each sheet contains information about the electrode number, post-implant day on which the testing was performed, stimulation frequency, the participant’s response, and the block for each data point.

https://cdn.elifesciences.org/articles/65128/elife-65128-fig1-data1-v2.xlsx
Figure 2 with 1 supplement see all
Stimulus current amplitude does not change the relationship between pulse frequency and intensity at suprathreshold amplitudes.

Magnitude estimation data for different current amplitudes and pulse frequencies. Data were aggregated across electrodes by their category, where each plot shows a different category of electrodes. Perceived intensity values for (A) LFP electrodes, (B) IFP electrodes, and (C) HFP electrodes at different current amplitudes and pulse frequencies. Different colored bars represent different current amplitudes. Error bars indicate the standard error across electrodes. We tested two LFP electrodes, three IFP electrodes, and two HFP electrodes which were each tested twice in different sessions.

Figure 2—source data 1

This file contains all the magnitude estimation data from participant P2 using amplitudes of 20, 50, and 80 μA and frequencies of 20, 100, and 300 Hz.

Each sheet contains data for a for one of the three frequency preference groups (LFP, IFP, or HFP). Each sheet contains information about the electrode number, post-implant day, stimulation amplitude, stimulation frequency, the participant’s response, and the block for each data point.

https://cdn.elifesciences.org/articles/65128/elife-65128-fig2-data1-v2.xlsx

Seven electrodes were tested multiple times (three to six per electrode) to determine whether the relationships between pulse frequency and perceived intensity were consistent across sessions. The perceived intensity on these electrodes changed by statistically significant amounts as the stimulus pulse frequency changed (p<0.001, Friedman test). The reported intensities at each frequency on these electrodes did not change significantly across test days (p>0.05, Friedman test) (Figure 1—figure supplement 3). An additional 22 electrodes were tested in one or two sessions. Of the 29 electrodes tested in total, 20 electrodes exhibited perceived intensities that changed by statistically significant amounts as the stimulus frequency changed (p<0.02, Friedman test). Of these 20 electrodes, five were classified as LFP, seven were classified as IFP, and eight were classified as HFP.

The three different electrode groups had significantly different median intensities (p<0.001, Kruskal–Wallis). Electrodes categorized as IFP had the highest median intensity, while electrodes categorized as HFP had the lowest median intensity (Figure 1A).

In participant P3, we tested 23 electrodes at 80 μA and three different frequencies (20, 100, and 300 Hz). The perceived intensity changed by statistically significant amounts on 22 electrodes as the stimulus frequency changed (p<0.05, Friedman test). There were similar electrode-specific effects, where some electrodes evoked the highest intensity percepts at the highest frequencies and others had the highest intensity at the lowest frequencies (Figure 1—figure supplement 4). Using the same clustering approach, the data divided into two clusters, which were most similar to the LFP and HFP categories. Fifteen electrodes were classified as HFP, and seven were classified as LFP.

Frequency-intensity relationships are preserved across suprathreshold amplitudes

We measured whether the frequency–intensity relationships were affected by stimulus current amplitude. If the frequency–intensity relationships were dependent on the current amplitude, this result might reflect idiosyncratic recruitment effects of ICMS. Therefore, in P2, we presented stimulus trains at three current amplitudes (20, 50, and 80 µA) and three pulse frequencies (20, 100, and 300 Hz), which spanned the range of detectable and safe parameters, and asked the participant to report the perceived intensity of the evoked percepts. There were no significant differences in the shape of the frequency–intensity relationships for the three electrode groups at 50 and 80 μA after controlling for changes in median intensity caused by increasing current amplitude (p=0.21–0.99, Friedman’s test, Figure 2). The reported intensity on LFP electrodes peaked at 20 Hz at both current amplitudes (p=0.02, Kruskal–Wallis, Figure 2A), whereas the reported intensities of IFP electrodes peaked at 100 Hz for both current amplitudes (p<0.001, Kruskal–Wallis, Figure 2B) and the reported intensity on HFP electrodes peaked at 300 Hz for both current amplitudes (p<0.001, Kruskal–Wallis, Figure 2C). Interestingly, when we decreased the current amplitude to 20 μA, which was close to the detection threshold for most electrodes, increasing the pulse frequency from 20 to 100 Hz evoked more intense percepts for all electrode groups (p<0.05, Kruskal–Wallis, Figures 2A–C, 20 µA). There were highly significant differences between the shape of the frequency–intensity relationships for all groups at 20 μA versus 50 or 80 μA (p<0.001, Friedman’s test) even after controlling for changes in the median intensity caused by increasing current amplitude. At 20 μA, the percept intensity was very low, making magnitude estimation akin to a detection task.

High-frequency stimuli are detected more reliably at perithreshold amplitudes

Our observation that higher stimulus pulse frequencies can evoke less-intense percepts at suprathreshold stimulus current amplitudes differs from predictions made from non-human primate studies; higher frequencies evoked detectable percepts at lower amplitudes in NHPs, which led to predictions that higher frequency always results in higher perceived intensities (Kim et al., 2015a; Romo et al., 2000; Romo et al., 1998). However, the effect of changing ICMS parameters on perceived intensity cannot be tested directly in NHPs. Indeed, we found that the perceived intensity at the lowest tested currents always increased when the frequency increased from 20 to 100 Hz (Figures 2A–C, 20 µA), but that this effect was not always maintained at higher current amplitudes (Figures 2A,B, 50 and 80 µA). To explicitly compare our results to NHP work, we performed a detection task in P2 in which the current amplitude was set to perithreshold levels and the pulse frequency was varied between 20, 100, and 300 Hz. We found that at 300 Hz, the interval containing the stimulus train was correctly identified 80% of the time across all tested electrodes (Figure 2—figure supplement 1). Similarly, when the pulse frequency was set to 100 Hz, the mean detection accuracy was 72%. In contrast, when the pulse frequency was set to 20 Hz, the mean detection accuracy was just 42%, which was not significantly different than chance levels of 50% (p=0.14, one-sample t-test). Detection accuracies at 100 Hz and 300 Hz were significantly higher than the detection accuracy at 20 Hz (p<0.05, ANOVA) but were not significantly different from each other (p=0.66, ANOVA).

Frequency-intensity relationships are associated with different perceptual qualities

One advantage of studying somatosensation in humans is the ability to document the sensory qualities evoked by stimulation (Figure 3—figure supplement 1). We found that there were significant differences in the qualities evoked on electrodes belonging to different categories defined by the effect of pulse frequency on intensity in P2 (Figure 3A). Additionally, the sensory qualities for electrodes in each group were differentially modulated by pulse frequency (Figure 3B).

Figure 3 with 2 supplements see all
Perceptual qualities are associated with specific electrode categories and stimulus pulse frequencies.

Radar plots showing the distribution of reported qualities at different pulse frequencies for each electrode category. (A) Percepts sorted by pulse frequency. Electrode categories are indicated with different colors. (B) Percepts sorted by electrode categories. Pulse frequencies are indicated with different colors. In each plot, qualities on which there was a significant difference between categories, as determined with Fisher’s exact test, are marked with an asterisk.

Figure 3—source data 1

This file contains the total number of reports of each percept quality in participant P2 across each frequency preference group (LFP, IFP, and HFP).

It also contains the number of times an electrode was stimulated in each group. Each sheet has data for a different stimulus frequency.

https://cdn.elifesciences.org/articles/65128/elife-65128-fig3-data1-v2.xlsx

At 20 Hz, LFP and IFP electrodes evoked percepts with pressure, tapping, sparkle, and touch qualities. These qualities were not evoked on HFP electrodes at any frequency. At this low stimulation frequency, HFP electrodes were generally not detectable, resulting in few reports of any percepts. At 100 Hz, IFP electrodes evoked percepts with buzzing, vibration, and sharp qualities. LFP and HFP electrodes never evoked these qualities when stimulated at 100 Hz. HFP electrodes also evoked sensations of touch and prick at 100 Hz that never occurred on LFP or IFP electrodes at any frequency. However, these qualities occurred on less than 30% of trials on HFP electrodes. At 300 Hz, the responses were similar to those at 100 Hz except that all electrode categories evoked less pressure.

We also clustered electrodes based on the verbal reports of percept quality at all frequencies. Interestingly, these clusters were remarkably similar to those based on intensity responses at different frequencies (Figure 3—figure supplement 2). That these electrode categories were nearly identical when created using completely different data sets – perceptual qualities and perceived intensities – strongly suggests that these two features are measures of the same underlying properties of the neurons recruited by stimulation.

Perceptual responses are spatially clustered in cortex

Finally, we asked whether the categorization of an electrode, which corresponds to its frequency–intensity responses and evoked perceptual qualities, was related to its location in cortex. We compared the observed spatial occurrence of the different electrode categories with a simulation that randomly assigned each category to one of the tested electrode locations while maintaining the same number of electrodes in each category. In P2, there was significant clustering of electrodes in the same category (Figure 4A) across arrays (pseudo-p=0.00017). This was particularly apparent on the lateral array. In P3, LFP electrodes only occurred on one of the arrays (Figure 4—figure supplement 1), which resulted in clustering across the arrays (pseudo-p=0.0045, local indicators of spatial association [LISA]).

Figure 4 with 1 supplement see all
Electrode location is significantly related to electrode categorization.

(A) Map of the medial electrode array (top) and lateral electrode array (bottom) implanted in somatosensory cortex and the distribution of the frequency preference categorizations. The electrode arrays were implanted close to the central sulcus with the left edge of the medial array being approximately parallel to the central sulcus. The arrays are oriented to reflect the implant orientation. Colored squares represent different types of electrodes as indicated by the color bar. (B) The projected field locations for each tested electrode. The label for each electrode corresponds to the most reported projected field for each electrode on all 100 Hz surveys taken in the same year as the magnitude estimation data.

Figure 4—source data 1

This file contains the spatial mapping of each electrode and the frequency preference group for each electrode for participant P2.

‘Nan’ values represent unwired electrodes.

https://cdn.elifesciences.org/articles/65128/elife-65128-fig4-data1-v2.xlsx

While there was some overlap between the projected field location and frequency preference, in some cases, electrodes with different frequency preferences evoked percepts from the same region of the hand (Figure 4B). For example, LFP, IFP, and HFP electrodes elicited sensations on the palmar region beneath the middle and ring fingers. As a result, percepts from a single region of the hand could be evoked by electrodes that generated multiple response types.

Discussion

We found that ICMS frequency alters the perceived intensity (Figure 1, Figure 1—figure supplement 4) and quality (Figure 3) in an electrode-specific manner. Furthermore, we found that electrodes with similar intensity responses and qualities clustered spatially in somatosensory cortex (Figure 4, Figure 4—figure supplement 1). This implies that the observed electrode-specific relationships between frequency and perception are not caused by random factors and are instead related to the underlying structure of the cortex.

Neural populations preferentially respond to different stimulus frequencies in somatosensory cortex

Intracortical microstimulation at the maximum amplitudes used in this study can directly activate neurons up to 2 mm away from the electrode tip, but most activation occurs less than 500 µm from the electrode tip (Overstreet et al., 2013; Stoney et al., 1968). At intermediate amplitudes (e.g. 50–60 µA), direct activation primarily occurs within 200–300 µm of the electrode tip. Stimulation can also recruit passing axons which can project to far away areas, resulting in sparse, distributed activation of the cortex (Histed et al., 2009; Michelson et al., 2019). Using optical imaging, clusters of neurons with similar responses extend from 0.2 to 1 mm in squirrel monkeys (Friedman et al., 2004) while electrophysiological recordings have measured similar effects over 0.5 to 1 mm in the mediolateral direction and multiple millimeters in the rostrocaudal direction in macaque monkeys (Sur et al., 1984). These spatial scales over which function varies are similar to the expected recruitment distances from ICMS, supporting the idea that different perceptual or frequency responses may be linked to activating different functional groups of neurons.

Electrophysiological (Mountcastle et al., 1969; Sur et al., 1981; Sur et al., 1984) and optical (Chen et al., 2001; Friedman et al., 2004) recordings have shown organized neural populations in the somatosensory cortex that are sensitive to tactile input with specific frequency content. These experiments promoted the idea of submodality separation in the cortex in which the activity of cortical neurons is primarily driven by input from either rapidly adapting Meissner corpuscles (RAs), slowly adapting Merkel cells (SAs), or Pacinian corpuscles (PCs). However, many cortical neurons receive heterogeneous input from multiple classes of mechanoreceptors (Pei et al., 2009; Reed et al., 2010; Saal and Bensmaia, 2014), resulting in neurons that can exhibit both sustained and transient responses. Therefore, the different effects of stimulus frequency on intensity and perception are unlikely to arise from activating inputs representing specific tactile subpopulations (e.g. SAs, RAs, or PCs), but instead from how a local region of somatosensory cortex can respond to different stimulus frequencies, consistent with the concept of the cortex encoding different frequency features (Prsa et al., 2019).

The idea that somatosensory cortex is organized for feature encoding is supported by human psychophysics experiments where frequency perception was dependent on specific spiking patterns and not on the types of mechanoreceptor that were activated (Birznieks et al., 2019). Similarly, individual cells in mouse cortex are preferentially activated by different mechanical stimulation frequencies (Prsa et al., 2019). In those same experiments, the frequency preference of the neural population tended toward higher frequencies when the indentation depth decreased, similar to our results that higher frequencies were perceived as being more intense when the ICMS amplitude was decreased (Figure 2). Together, these results suggest that the somatosensory cortex receives convergent input from different mechanoreceptors and is organized for feature-selective encoding, which results in different preferential responses to ICMS frequency and different evoked qualities.

Mechanisms for heterogeneous perceptual responses to stimulus frequencies in cortex

The effects described here must be related to different cellular responses to stimulation in different regions of the cortex. In fact, different stimulation frequencies in mouse somatosensory cortex can alter the activation of neurons far away from the stimulation electrode (Michelson et al., 2019). Specifically, high pulse frequencies lead to rapid habituation of neurons far away from the electrode, while low pulse frequencies can maintain the activation of these same neurons. This reduced activity in neurons far away from the electrode could lead to decreases in perceived intensity and changes in percept quality in a way that depends on electrode location and local neural populations.

A potential mechanism to explain electrode-dependent responses are varying distributions of inhibitory and excitatory neurons. The presence of more inhibitory neurons in a local region could result in stronger inhibitory drive at higher frequencies, resulting in more robust responses to low-frequency stimuli. Indeed, recruitment of inhibitory Martinotti cells in the somatosensory cortex of rats increases as the duration and frequency of presynaptic action potentials increase (Kapfer et al., 2007; Silberberg and Markram, 2007). Furthermore, rostrocaudal heterogeneity of inhibition has been documented in rat olfactory cortex (Large et al., 2018; Luna and Pettit, 2010). Whether such organization exists in human somatosensory cortex remains to be seen.

Short-term plasticity (Tsodyks and Markram, 1997) at synapses driven by stimulation may also explain the observed effects. If a synapse is unable to resupply neurotransmitter at a rate faster than the stimulus frequency, transmitter release at the synapse could become depressed. In this scenario, neurons would be unable to recruit other neurons in synchrony with stimulation, which could result in lower recruitment and lower perceived intensity. If cells in somatosensory cortex have different time constants for transmitter recovery, this could serve as a mechanism for frequency filtering (Rosenbaum et al., 2012). Elucidating the precise mechanisms underlying observed frequency responses in cortex will require further studies in animal models.

ICMS in humans directly evaluates intensity and perception

Higher stimulus pulse frequencies decreased the current amplitude required to detect a stimulus train in NHPs (Kim et al., 2015a). This suggested that higher stimulus frequencies would increase the perceived intensity of a stimulus train. Similar to these animal studies, we found that higher frequencies improved the detectability of stimulus trains at perithreshold amplitudes. However, at suprathreshold current amplitudes, increasing the frequency did not always produce higher perceived intensities. A question that emerges then is whether the prediction of increasing intensities at higher frequencies can be reconciled with our observations of decreased intensities at higher frequencies on a subset of the electrodes.

To determine whether changes in frequency could be perceived independently of changes in amplitude, animals were trained to identify which of two intervals contained the higher frequency stimulus train, regardless of current amplitude (Callier et al., 2020). Increasing the amplitude always biased the animals toward selecting a stimulus train as having a higher frequency, suggesting that both amplitude and frequency have similar perceptual effects. However, animals were still able to distinguish between changes in amplitude and frequency on some electrodes. In our experiments, LFP and IFP electrodes, which generated high-intensity percepts at low frequencies, often evoked percepts with highly salient qualities, such as tapping or buzzing. The presence of these qualities at some frequencies and not others (Figure 3) would allow the participant to distinguish between increases in amplitude, which only increases the percept intensity (Figure 1—figure supplement 1), and increases in frequency, which changes the percept quality and intensity (Figure 3). On electrodes without highly salient frequency-dependent qualities, such as the HFP electrodes, it would be difficult to disambiguate changes in amplitude and frequency.

However, an important difference between these experiments is that many electrodes in our study evoked less-intense percepts as the pulse frequency increased, which was not observed in NHPs. The reason for this is unclear, and it may be related to the larger frequency range explored in this study or the electrode location in the cortex. Another interpretation is that since frequency can change percept quality (Figure 3), different qualities are understood to have different intensities. Animals cannot directly report perceived intensity on an open scale as is simply done in humans. Rather, perceived intensity, as well as other subjective aspects of perception such as quality and naturalness, must be inferred from other experimental paradigms, which makes it difficult to assess how ICMS affects subjective aspects of perception in animals. This demonstrates that human experiments are crucial to understanding how ICMS modulates tactile perception, particularly for subjective evaluation of experience.

Limitations of study

There are several limitations associated with this work. First, most of these experiments were conducted in a single participant with a chronic implant. Different participants, with different timelines of injury preceding implant, could potentially respond differently to stimulation, particularly if the electrodes are implanted in a different part of the somatosensory cortex. However, the repeatability of our findings suggests that these effects were at least not due to day-to-day variations. Additionally, we found electrode-specific frequency effects, including LFP electrodes, that were spatially clustered in a second participant. This suggests that changing frequency will affect intensity and perception similarly in other participants. One important difference in the second participant was that we only observed LFP relationships on a single array.

The participants had limited residual sensation in their hands, which made it difficult to measure responses in cortex to tactile indentation. Comparing perceptual responses to ICMS with cortical responses to tactile indentation could help better relate these findings to previous studies in monkeys. Additionally, it is notable that due to spinal cord injury there may be reorganization of cortex (Chen et al., 2002; Freund et al., 2011; Henderson et al., 2011; Wrigley et al., 2009). However, recent work has argued that measured remapping may be simply driven by the uncovering of pre-existing latent activity, corresponding instead to homeostasis (Makin and Bensmaia, 2017; Muret and Makin, 2021). The ability to elicit sensations with ICMS years after injury is supportive of this idea (Armenta Salas et al., 2018; Fifer et al., 2020; Flesher et al., 2016).

Another potential confound is that perceived intensity can change throughout a session. Because we used pseudo-randomized presentations of different stimulus parameters to ensure that electrodes were not tested in the same order, and excluded the first block of trials from analysis for each set for magnitude estimation, we believe that this phenomenon did not affect our results.

Our results are consistent with the idea that somatosensory cortex is organized in a way that represents different features in different locations; however, there are several limitations that should be considered. First, the electrodes covered just a small region of somatosensory cortex, and with a limited spatial resolution, limiting the ability to create detailed maps. Second, we divided electrodes into three groups for participant P2 and two groups for participant P3. This categorical division arose from considering the frequency–intensity relationships and the unique perceptual qualities reported for the electrodes in each group. Categorical divisions are commonly used to describe neural responses in the cortex, including somatosensory cortex (Friedman et al., 2004; Sur et al., 1981; Sur et al., 1984). However, neurons receive convergent input from multiple sub-type modalities (DiCarlo et al., 1998; Saal and Bensmaia, 2014), and it is possible that the responses to stimulation may divide into more than three groups or fall on a spectrum of different frequency preferences with no discrete categories. More data will need to be collected across additional participants and regions of somatosensory cortex to see whether these patterns persist. Third, we do not know if electrodes across the array are in different layers of cortex. Different layers of cortex may drive different perceptual responses with the same input. However, if this were the case, this would still reflect important functional differences in cortex, which need to be understood for bidirectional BCIs.

Finally, a challenge for developing mechanistic explanations of these observations is that there are few neuroscientific tools that we can use to further probe these effects in a human. Because of this, addressing the neurophysiological mechanisms of these frequency responses is difficult in a human participant, and further investigation of these properties in animal models is needed.

Implications for prostheses

Stimulus amplitude linearly modulates intensity, while frequency has non-monotonic and electrode-specific effects on intensity and percept quality. To signal changing the intensity of a tactile input, amplitude should be used and not frequency. Other potential options also exist to modulate intensity that were not explored in this paper, including pulse width modulation and multielectrode stimulation. Future studies should assess the efficacy of these parameters.

Knowing that different electrodes encode different perceptual features can inform our approach to creating a functional bidirectional BCI in two primary ways. First, these results may help identify electrodes that have perceptual or intensive properties that are relevant to the task being performed. Certain electrodes are more likely to represent specific perceptual qualities, and these electrodes could be selectively used depending on the type of tactile input to the prosthetic device.

Second, these results suggest that electrode-specific stimulation encoding schemes would be particularly useful. In the peripheral nervous system, biomimetic approaches to stimulation using models such as TouchMime have been used (George et al., 2019; Okorokova et al., 2018; Valle et al., 2018). In the cortex, combining these biomimetic models with electrode selection based on measured feature-preferences may yield more natural percepts. For example, electrodes that represent ‘tapping’ sensations could receive large amplitude transients, signaling the onset and offset transients, while electrodes that do not evoke this sensation could receive low-amplitude, tonic stimulation, signaling maintained contact. Another promising future direction is to use machine learning methods to categorize the feature-preferences of different electrodes more quickly and accurately. These methods could ultimately improve the usefulness of somatosensory feedback, in turn improving the performance of bidirectional BCIs and ultimately improving the quality of life for people living with paralysis.

Materials and methods

Regulatory and subject details

Request a detailed protocol

This study was conducted under an Investigational Device Exemption from the U.S. Food and Drug administration, approved by the Institutional Review Boards at the University of Pittsburgh (Pittsburgh, PA) and the Space and Naval Warfare Systems Center Pacific (San Diego, CA), and registered at ClinicalTrials.gov (NCT0189-4802). Informed consent was obtained before any study procedures were conducted. The purpose of this trial is to collect preliminary safety information and demonstrate that intracortical electrode arrays can be used by people with tetraplegia to both control external devices and generate tactile percepts from the paralyzed limbs; this manuscript presents the analysis of data that were collected during participation in the trial but does not report clinical trial outcomes. All data included in this paper (including magnitude estimation, surveys, detection thresholds, etc.) were limited to 1 year of data collection in P2 to minimize the impact of changes in perception that can occur over long time periods. Data in P3 were collected over 2 months.

Participant P2 was 28 years old at the time of implant and had a C5 motor/C6 sensory ASIA B spinal cord injury. Two microelectrode arrays (Blackrock Microsystems, Salt Lake City, UT) were implanted into the somatosensory cortex. Results from this participant have been reported previously (Flesher et al., 2016; Flesher et al., 2021; Hughes et al., 2021a). Each electrode array consisted of 32 wired electrodes arranged on a 6 × 10 grid with a 400 μm interelectrode spacing resulting in a device with an overall footprint of 2.4 × 4 mm. The remaining 28 electrodes were not wired due to technical constraints related to the total available number of electrical contacts on the percutaneous connector. Electrode tips were coated with a sputtered iridium oxide film. The stimulation return electrode was the titanium pedestal that was fixed to the skull.

Participant P3 was 28 years old at the time of implant and had a C6 ASIA B spinal cord injury. He received the same type of microelectrode arrays in the somatosensory cortex. Data from this participant have not been published previously. The electrodes were also targeted to the hand region of area 1 of the somatosensory cortex using preoperative imaging and evoked sensations that he described as originating from his hand.

Stimulation protocol

Request a detailed protocol

Stimulation was delivered using a CereStim C96 multichannel microstimulation system (Blackrock Microsystems, Salt Lake City, UT). Pulse trains consisted of cathodal phase first, current-controlled, charge-balanced pulses, which could be delivered at frequencies from 20 to 300 Hz and at amplitudes from 2 to 100 μA. The cathodal phase was 200 μs long, the anodal phase was 400 μs long, and the anodal phase was set to half the amplitude of the cathodal phase. The phases were separated by a 100 μs interphase period. At the beginning of each test session involving stimulation, we sequentially stimulated each electrode first at 10 μA and 100 Hz for 0.5 s and then at 20 μA and 100 Hz for 0.5 s. During these trials, the interphase voltage on each electrode was measured at the end of the interphase period, immediately before the anodal phase (Cogan, 2008). If an electrode’s measured interphase voltage was less than −1.5 V, the electrode was disabled for the day (Flesher et al., 2016). This step was performed to minimize stimulation on electrodes that might potentially experience high voltages, which could result in irreversible damage.

Magnitude estimation

Request a detailed protocol

We assessed the effect of stimulus parameters on perceived intensity using a magnitude estimation task. To test the potential effect of pulse frequency on intensity in P2, pulse trains were delivered for 1 s at 60 µA with frequencies of 20, 40, 60, 80, 100, 150, 200, 250, and 300 Hz. Following each pulse train, P2 was asked to report the magnitude of the perceived intensity on a self-selected scale. P2 was instructed to use values such that a value twice as large as a previous value was twice as intense, and a value half as large was half as intense. These values typically ranged from zero to six. Each set of stimulus pulse frequencies was presented six times, with the presentation order randomized in each block. The responses from the first block were not used in the analysis to allow the participant to establish a baseline for reporting for the session. Data collected on the same electrode over multiple sessions were aggregated for analysis. We tested 29 total electrodes using this paradigm. Seven electrodes were tested in three to six sessions, while 22 electrodes were tested in one to two sessions.

To increase the number of trials and decrease the time for data collection, we presented 20, 100, and 300 Hz stimulus trains at 80 μA to participant P3. We presented each frequency 21 times and removed the first trial from the analysis. Twenty-two of the 23 tested electrodes showed a significant difference between intensities across tested frequencies (Friedman’s test, p<0.05). Data for each electrode were only collected once.

We also assessed the effect of changing the stimulus current amplitude on perceived intensity, while the stimulus pulse frequency was held constant in P2. The pulse frequency was set to 100 Hz, the train duration to 1 s, and the current amplitude ranged from 20 to 80 μA in 10 μA increments. Data were fit with a linear function. We tested nine electrodes for this paradigm. Finally, we assessed the effect of changing the stimulus train duration on perceived intensity in P2. The stimulus pulse frequency and current amplitude were set to 100 Hz and 60 μA, respectively, and the train duration was set to 0.1, 0.2, 0.3, 0.4, 0.5, 0.75, 1, 1.5, and 2 s. Data were fit with a logistic function. We tested four electrodes for this paradigm. For current amplitude and train duration plots, the data were normalized to the median intensities of the set in which it was collected for visualization purposes.

To investigate the interaction between current amplitude and pulse frequency, we additionally tested frequency and amplitude pairs in P2. The train duration was set to 1 s, the current amplitude was set to 20, 50, or 80 μA, and the pulse frequency was set to 20, 100, or 300 Hz. All frequency and amplitude combinations were tested for each tested electrode six times, and the first trial was excluded from analysis. Each tested electrode was tested twice on two different test sessions, resulting in 10 total trials for each frequency and amplitude pair. For analysis and plotting, we divided electrodes into the categories defined in the frequency magnitude estimation described previously. We tested two LFP electrodes, three IFP electrodes, and two HFP electrodes. We tested six electrodes for this paradigm, each measured twice.

Detection thresholds

Request a detailed protocol

Detection thresholds were determined using a two-alternative forced choice task in P2. P2 was instructed to focus on a fixation cross on a screen located in front of him. Two 1-s-long windows, separated by a variable delay period, which averaged 1 s in length, were presented and indicated by a change in the color of the fixation cross. Stimulation was randomly assigned to one of the two windows. After the last window, the fixation cross disappeared, and the participant was asked to report which window contained the stimulus.

A one-up three-down staircase method was used, so that if the participant correctly identified the window containing the stimulus in three consecutive trials, the current amplitude was decreased for the next trial (Leek, 2001; Levitt, 1971). If the participant incorrectly identified the window containing the stimulus, the current amplitude was increased for the next trial. The current amplitude started at 10 μA and was increased or decreased by a factor of 2 dB. The pulse frequency was held constant at 100 Hz. This method reduced the time spent testing uninformative values but does not guarantee that all current amplitudes will be tested the same number of times. After five changes in the direction of the stimulus current amplitude (increasing to decreasing, or decreasing to increasing), the trial was stopped. The detection threshold was calculated as the average of the last 10 values tested before the fifth direction change.

We also conducted standard detection trials where the stimulus pulse frequency was changed while the stimulus current amplitude was held constant in P2. The current amplitude was set to 1.2× the detection threshold for each electrode measured at 100 Hz. The tested frequencies were 20, 100, and 300 Hz, and each pulse frequency was presented 30 times. Pulse frequencies were interleaved randomly resulting in 90 trials per tested electrode. We tested four electrodes with this paradigm.

Surveys

Request a detailed protocol

Surveys were conducted once every month from the time the arrays were implanted in P2. During a survey, each enabled electrode was stimulated sequentially using a 1 s pulse train at 60 μA. These parameters were selected because they were typically able to evoke sensations consistently while remaining well below our maximum stimulus current amplitude of 100 μA. In participant P2, surveys were conducted once a month at 100 Hz, but we collected additional surveys at 20 and 300 Hz. This resulted in 152 samples at 20 Hz, 621 samples at 100 Hz, and 85 samples at 300 Hz. Surveys were conducted to quantify stimulus-evoked tactile percepts. No visual or auditory cue was provided to the participant to indicate when stimulation was occurring. The participant was instructed to indicate when a sensation was detected, at which point progression through the trial was paused. The participant verbally reported when he detected a sensation, and the pulse train was repeated as many times as necessary for the participant to be able to accurately describe the location and quality of the sensation. A drawing of the hand was partitioned into different segments and the participant reported on which segments the sensation was felt. The participant also used a tablet and stylus to circumscribe the precise areas where sensation was felt on a map of the hand.

After the location of the percept was established, the participant reported the quality of the sensation using the descriptors in Figure 3—figure supplement 1. The participant’s response was documented by the experimenter, and video recordings were also taken during all responses. If the participant felt that the sensation was not accurately described by the provided descriptors, his response was recorded, and the best approximation using the descriptors was used. The descriptors included a five-point scale for naturalness ranging from totally unnatural to totally natural, the location of the sensation on or below the skin surface, and an assessment of pain ranging from 0 to 10. The quality of the sensation was further assessed using the following descriptors: mechanical (touch, pressure, or sharp), movement (vibration or movement across the skin), temperature (warm or cool), and tingle (electrical, tickle, or itch). These descriptors were based on a previously described questionnaire (Heming et al., 2010). The participant could report multiple qualities for a single stimulus, and in some cases, the subcategories (e.g. electrical, tickle, or itch) could not be described. P2 also reported qualities that deviated from the descriptors. P2 developed four new descriptors that were not originally included, which often were combinations of the other descriptors. We attempted to reidentify these percepts in the context of a new questionnaire, which was published during this study in consultation with the participant (Kim et al., 2018). Three of these sensations were reidentified as ‘tapping’, ‘buzzing’, and ‘prick’. One descriptor P2 reported, ‘sparkle,’ could not be reidentified with the new questionnaire. P2 described this percept as feeling like tapping that varied in intensity and moved around the projected field in a random manner. It should be noted that all percepts in our study were identified as tactile percepts and no proprioceptive sensations were evoked.

The survey data collected in P2 included in these analyses were collected during the same year as the frequency magnitude estimation data to ensure the evoked sensations were consistent across paradigms, which included data from post-implant days 630–962.

K-means clustering

Request a detailed protocol

Electrodes were divided into three categories using k-means clustering using the reported intensity at 20, 100, and 300 Hz. Both silhouette and elbow analysis were used to validate that k = 3 was a suitable parameter choice for P2. We labeled the categories as LFP, IFP, and HFP based on the frequency at which the maximum intensity occurred. Based on silhouette analysis, we found that data from P3 divided best into two clusters. We labeled these clusters as LFP and HFP in line with the classification from the first participant.

Electrodes were additionally clustered based on the reported perceptual qualities at 20, 100, and 300 Hz in P2. Each reported quality (of which there were 10) was summed across sessions and pulse frequencies for each electrode. The total number of reports for each quality was then divided by the maximum number of reports for any electrode, so that each quality was represented by number between zero and one and contributed equally to the clustering of each electrode. No dimensionality reduction was used and electrodes were clustered within the 10 dimensions of reported qualities.

Statistics

All quantification and statistical analyses were done in MATLAB (Mathworks, Natick, MA). Sample sizes are listed in the methods for each experiment. A power analysis was not conducted to determine the number of replicates for each experiment. The number of repetitions for psychophysics experiments were based on commonly used values. Electrodes that elicited clearly perceptible sensations and showed a significant change in perception with a change in a parameter were collected across multiple sessions to determine whether effects were consistent over time.

For all statistical tests, we determined whether to use parametric or non-parametric tests based on the normality of the data as assessed with an Anderson–Darling test. If the data were significantly different than normal, then we used non-parametric tests. Any time multiple comparisons were made, we used the Benjamini–Hochberg procedure to correct for multiple comparisons, which resulted in a critical p-value that was used as a cut-off. If no values were significant, then the critical p-value returned is 0 and not reported and no values are considered significant. For any tests that required post-hoc comparisons, we used Tukey’s HSD test.

For magnitude estimation data, we used Friedman’s test to assess significant differences between the intensity responses at different pulse frequencies as well as differences between electrode responses across days. Friedman’s test also allowed us to compare significant effects of pulse frequency on intensity across multiple sessions by excluding experimental day as a cofactor. When comparing the same electrode across sessions, we compared intensity responses with the same tested pulse frequency and corrected for multiple comparisons. We compared differences in the median intensity of electrodes within each category using a Kruskal–Wallis test.

For detection data, we used an ANOVA to assess significant differences in the detection accuracy at different pulse frequencies.

For quality data obtained from surveys, we used Fisher’s exact test to evaluate whether there was a relationship between the categorization of each electrode and the perceptual qualities evoked on the electrode. Contingency tables were developed for each descriptor and responses were row-divided by the three categories (LFP, IFP, and HFP) and column-divided by the presence or absence of the quality. Each category was compared pairwise. Fisher’s exact test was used instead of a chi-squared test because the sample sizes for each group were relatively small.

To test whether there was spatial clustering of the effects of frequency on perceived intensity across the array, we adopted a technique used in geographic information systems, where they are described as LISA (Anselin, 1995). We quantified the number of electrodes that had an adjacent electrode with the same frequency response category and divided this by the total number of adjacent electrodes to obtain a fraction. We then randomly distributed the categorized electrodes on two simulated arrays with the same tested electrode locations. We conducted this simulation 100,000 times and compared the output values of this random simulation to the observed values. A pseudo p-value was obtained by comparing the total number of simulations that had a fraction greater than or equal to the observed fraction, which indicates the probability of obtaining our observed value by chance.

For all statistical tests, we considered p<0.05 to be significant.

Data and code availability

Request a detailed protocol

Data and code for this paper are available at GitHub (https://github.com/chughes003r/FrequencyPaperHughes et al., 2021b; copy archived at swh:1:rev:96f81aa826f68b9f509a3d73b7765a68ce0193e4).

Data availability

Data and code for this paper are available at GitHub (https://github.com/chughes003r/FrequencyPaper, copy archived at https://archive.softwareheritage.org/swh:1:rev:96f81aa826f68b9f509a3d73b7765a68ce0193e4).

The following data sets were generated
    1. Hughes CL
    2. Flesher SN
    3. Weiss JM
    4. Boninger M
    5. Collinger JL
    6. Gaunt RA
    (2021) GitHub
    ID github.com/chughes003r/FrequencyPaper. Code and data for "Perception of microstimulation frequency in human somatosensory cortex".

References

    1. Dostrovsky JO
    2. Davis KD
    3. Lee L
    4. Sher GD
    5. Tasker RR
    (1993)
    Electrical stimulation-induced effects in the human thalamus
    Advances in Neurology 63:219–248.
  1. Conference
    1. Godfrey SB
    2. Bianchi M
    3. Bicchi A
    4. Santello M
    (2016) Influence of force feedback on grasp force modulation in prosthetic applications: a preliminary study
    Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. pp. 5439–5442.
    https://doi.org/10.1109/EMBC.2016.7591957
  2. Book
    1. Hughes CL
    2. Herrera A
    3. Gaunt R
    4. Collinger J
    (2020) Bidirectional brain-computer interfaces
    In: Ramsey NF, Millán JR, editors. Handbook of Clinical Neurology, 168. Elsevier. pp. 163–181.
    https://doi.org/10.1016/B978-0-444-63934-9.00013-5
    1. Levitt H
    (1971)
    Transformed Up‐Down Methods in Psychoacoustics
    The Journal of the Acoustical Society of America 49(2B:467–477.
  3. Conference
    1. Semprini M
    2. Bennicelli L
    3. Vato A
    (2012) A parametric study of intracortical microstimulation in behaving rats for the development of artificial sensory channels
    Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society. pp. 799–802.
    https://doi.org/10.1109/EMBC.2012.6346052

Decision letter

  1. J Andrew Pruszynski
    Reviewing Editor; Western University, Canada
  2. Barbara G Shinn-Cunningham
    Senior Editor; Carnegie Mellon University, United States
  3. Tobias Heed
    Reviewer; Bielefeld University, Germany
  4. Silvestro Micera
    Reviewer; Scuola Superiore Sant'Anna, Italy

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Acceptance summary:

This paper characterizes percepts evoked by micro-stimulating the human somatosensory cortex. The study provides new insight into the organization of the human somatosensory cortex and represents an important step in providing more effective somatosensory feedback for brain-machine interface users.

Decision letter after peer review:

Thank you for submitting your article "Perceptual responses to microstimulation frequency are spatially clustered in human somatosensory cortex" for consideration by eLife. Your article has been reviewed by 4 peer reviewers, and the evaluation has been overseen by Andrew Pruszynski as the Reviewing Editor and Barbara Shinn-Cunningham as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Tobias Heed (Reviewer #2); Silvestro Micera (Reviewer #3).

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

Essential revisions:

1) The observed spatial clustering requires more evidence to accept it as a physiologically significant finding. To ultimately do so, the authors likely need additional analyses and more data (see specific comments by Reviewers #1, 2, 4). While more analysis can help note that we are not suggesting more data is needed for this paper. However, this finding needs to be toned down and approached more cautiously to avoid over-interpretation. Changing the title needs to be a part of such changes.

2) The authors should provide a more cohesive high level take-away message from the study. Although all reviewers felt the work was important and exciting from a technical perspective, the paper lacks a strong cohesive message that synthesizes the results either for BMIs or for S1 organization.

3) As laid out by Reviewer #4, it is important that the authors place their efforts in context of previous work that systematically examined effects of related stimulus parameters in human somatosensory thalamus. In general, several reviewers provide very useful literature that should be cited appropriately and discussed as needed.

Reviewer #1 (Recommendations for the authors):

Thank you for a really interesting manuscript.

I suggest to have a closer look at Prsa et al. Nature 2019 which shows stimulus-response curves for cortical neurons in mice showing similar dependency of stimulation frequency as observed in the current study.

Title: I am not convinced at all that there are spatially organised clusters as the title suggests. The number of electrodes is too small and there are many factors (including technical) which might explain why the electrode properties are not at random. I would be very careful making such a statement.

The electrodes are spaced 400 μm apart, but is it known within how large an area around the electrode neurons are activated?

Figures: All figures report SEMs, however SDs or CI would be more appropriate to evaluate variability.

Lines 149-153: I don't see contradiction here as HFP electrodes in general had very low median intensity. Figure 2A also shows that for HFP electrode at 20 μm the increased frequency leads to more intense percept.

Line 167: see Callier et al. PNAS 2020 who developed paradigms to address this.

Paragraph 208-225: The logic and writing is not very clear. Can you please clarify your views and refine the text?

217-219: Do you mean that convergent inputs from different sub-modalities would necessarily imply uniformity of distribution? This is not what studies mentioned in that context would suggest. Please clarify this statement.

Line 225: consider ref Birznieks et al. eLife 2019.

Paragraph 226-233: consider reference to Prsa et al. Nature 2019.

Lines 235-241: out of context it is not clear what are distal neurons.

Section 258-286: I disagree that there is a contradiction with non-human primate studies as I indicated in the public review. To play devil's advocate, the term "intensity" could be used to describe stimuli of different qualities and even modalities, but does it mean that, for example, intensity of olfactory stimuli could be compared with intensity of pain or any other modality? What about intensity of skin stretch vs vibrotactile intensity? Would you expect the same neural code in all those instances? Thus the sense of intensity due to electrical stimulation of cortical neurons may not be compatible with intensity of mechanical stimuli which have indentation depth, certain spatial pattern of afferent activation including size of the area, it's shape and type of borders (sharp/diffuse). All of those features may interact between each other creating a complex integrated percept of intensity which is absent with electrical stimulation.

Page 8: statements in two sections in regard to frequency filtering in cortex and human psychophysics: One possibility is that the function of some of these electrically stimulated circuits might be to detect presence of a given stimulus quality (specific qualitative feature) which might be associated with specific discharge rate in those neurons – like, for example, discharge rate <60Hz (>15ms inter-spike intervals) might mediate sense of flutter. There is experimental evidence which might support such arrangements preferentially detecting inter-spike intervals of certain length – in case of presence of short (<15ms) and long inter-spike intervals in afferent activity, the short inter-spike intervals (corresponding to high frequency) are simply ignored and perceived frequency corresponds exclusively to the longest (flutter range) inter-spike interval and not periodicity or mean discharge rate. When shorter inter-spike intervals became longer, their weighted contribution to perceived frequency increases. This has been demonstrated with mechanical (Birznieks and Vickery Current Biology 2017) and electrical stimulation (Ng et al. PlosOne 2020) and might fit with the current observations. Vibrotactile stimuli are very simplistic, but they can demonstrate principles how neural activity translates into perception.

Methods section: Please mention where the reference electrode was.

Reviewer #2 (Recommendations for the authors):

I am wary about the fitting and clustering approaches.

– For instance, in Figure 1, the purple curve fit of panel A seems inappropriate – the data would be better fit with a monotonically decreasing function.

– Similarly, the data in C seem to differ from D only in the very first data point of each data series. So the difference between C and D appears to be at a very low frequency only (?). I cannot assess whether this is generally true as only 2 curves are illustrated, and the use of 100 Hz in the other parts of the manuscript suggest otherwise. As is, this is an unclear point for me.

– If there is actually more spread of the preferred frequency, then I wonder how adequate it is to cluster, rather than assume, say, a continuously changing gradient across a larger cortical region.

– Related, I understand that 3 comes out as adequate from the clustering analysis, but this is based on a small number of sites. This is inherent in using electrode arrays. If more sites were available, do the authors think a 3 cluster interpretation would still be supported? What speaks against a more fine-grained, e.g. gradient-like, distribution? In other words, is there evidence against such an organization, or is the 3 cluster solution possibly just due to the small amount of testable regions/electrodes? Do the authors prefer a categorical account because they think there is a direct link to distinct perceptual qualities?

– I find it difficult to take away a "higher-order" result. To me, the presented work is clearly illuminating in that the different tested manipulations must be acquired to know in what kind of percepts they result. Also, the finding that supra-threshold stimulation results in different conclusions than near-threshold stimulation seems to me an important point. However, in many passages, the paper reads mainly like a report of all the different detailed tested aspects. I am missing some more "visionary" ideas about what the results might mean.

Reviewer #3 (Recommendations for the authors):

The paper is interesting and well-written. I think it could be improved by addressing the following issues:

1. The interplay between frequency and amplitude is of course very important: it could be interesting to see whether the authors could find a unifying model/equation as done in Graczyk et al., 2016 for PNS stimulation

2. Figure 1 seems to show that many electrodes have a small "intensity dynamics": could this be a problem for closed-loop control? The authors should elaborate a bit more on this

3. Figure 4 is very interesting but it could stronger and more convincing by testing more electrodes

4. the last section of the discussion on personalization is very interesting and the authors should elaborate a bit more also in this case. What kind of solution do they have in mind? Biomimetic? or Machine learning based?

5. it is not clear the duration of the overall testing. This is important to gather more information about the stability of the sensations over time.

Reviewer #4 (Recommendations for the authors):

A few questions and suggestions:

The major one being that the authors need to put their work in context of the original work that systematically examined effects of stimulus parameters in human somatosensory thalamus (Dostrovsky et al. 1993 Adv Neurol). Of course more publications exist on cortex recently but several of the concepts presented here as new were already studied and understood to be true in human thalamus. Also the major limitation is that all these experiments were performed in deafferented cortex; it is not just that it is one subject, but that their cortex must be reorganized. Of course that is what makes the prosthetic side of the story stronger (rather than the comparisons to normal non-human primate experiments), in that it remains amazing that sensations can be elicited from cortex that has not received afferent input in years.

1. To improve the wording of second last sentence in abstract. It does summarize the results but is somewhat confusing: is it the electrodes that were of 3 types or the brain sites stimulated.

2. While I understand how difficult it is to plot the different sensory modalities on a graph and I appreciated the radar plots in Figure 3 as the best way to do this, I was wondering how much altering different frequencies and intensities could alter the percepts evoked through a single microelectrode. Because it is unlikely we will have multiple opportunities to move cortical arrays in humans, to make this a practical application we need to know how much we can use electrical parameters to modulate percepts.

3. I did not follow the pseudo-p LISA statistical analysis shown in Figure 4. I see what the authors are trying to say in Figure 4A, but Figure B may either be simplified, better explained in the legend or perhaps moved to supplemental.

4. Why did the authors not test the effects of pulse width. It appears they were trying to test the entire parameter space that can be applied with electrical microstimulation, why not this one?

5. What about applying stimulation through multiple electrodes simultaneously? In fact this may help answer the question about whether short-term plasticity is involved (from discussion).

6. Please clarify if this patient is the same one reported in previous papers. The introduction suggested that it was and perhaps this was why the projected fields evoked with microstimulation were not described. If this is the case then how did Figure 4 compare to the somatotopy described in the previous paper? And what part of S1 do they believe the array is located (i.e. what Brodmann area 1, 2, 3a,b)? Adding a statement about what cortical layer the authors believe the microelectrodes are located before the limitations section of the discussion would be helpful. There are a few words about somatotopy at the end of the Results section indicating that all three types of responses: low, med, and high frequency preferring regions can subserve the same body region. However because the statement says "in some cases", it makes it equivocal, yet is a major component of the discussion.

7. The authors describe that because the subject is deafferented they could not identify receptive fields from the recordings. Were recordings performed at all in the deafferented cortex? While that is not the subject of this paper, it would be a welcome addition to the literature.

8. Please clarify the consistency of results longitudinally. Figure S3 does not seem to demonstrate all electrode showing same results over time: 3 of 7 electrodes (2, 3, 36) seem different over time.

https://doi.org/10.7554/eLife.65128.sa1

Author response

Essential revisions:

1) The observed spatial clustering requires more evidence to accept it as a physiologically significant finding. To ultimately do so, the authors likely need additional analyses and more data (see specific comments by Reviewers #1, 2, 4). While more analysis can help note that we are not suggesting more data is needed for this paper. However, this finding needs to be toned down and approached more cautiously to avoid over-interpretation. Changing the title needs to be a part of such changes.

Thank you for these helpful comments. We will address the individual concerns throughout this document relevant to each of these points. However, here we summarize how we plan to address these essential revisions.

Based on the robust statistical testing, we are confident that the responses were not randomly distributed across the cortex in these experiments and that they do indeed cluster. However, the analysis itself may need to be explained better. We have addressed this point later in the response and in the main text. However, we agree that with just a single participant and without more extensive datasets, making strong claims about structural organization may not be appropriate. However, since receiving the responses we have collected a smaller data set in a second participant and observed similar clustering which is now included as figure supplements (Figure 1—figure supplement 4, Figure 4—figure supplement 1). Information about these data have been added to the manuscript. While we believe this demonstrates even more strongly that these relationships are spatially clustered, the reviewers are correct to note there are other possible explanations for this observed clustering. Alternative explanations, including layer specificity and more continuous variation in these relationships, were considered, but ultimately removed from the original paper to make discussion more concise. These points have been added back into the discussion. Nevertheless, we will also deemphasize spatial clustering as a major point of the paper by changing the title as follows:

Title: Perception of microstimulation frequency in human somatosensory cortex

2) The authors should provide a more cohesive high level take-away message from the study. Although all reviewers felt the work was important and exciting from a technical perspective, the paper lacks a strong cohesive message that synthesizes the results either for BMIs or for S1 organization.

This study has important implications for both BCI development and for basic studies of human brain structure and function. Based on point 1 above, we will de-emphasize claims related to the latter point, although we still feel that it is important to discuss these. The critical conclusions of this study are, (1) stimulation frequency does not have consistent linear effects on percept intensity across electrodes as we have seen with amplitude, (2) stimulation frequency changes the perceptual qualities in different ways on different electrodes, (3) these intensity and quality effects are coupled, and (4) bidirectional BCIs will need to consider these electrode-specific effects to implement effective feedback algorithms. Importantly, these disparate effects of frequency on perception were not predicted from prior animal work and have implications for the structure and function of the brain that need to be studied further. The revised manuscript now emphasizes these points clearly to present a cohesive story.

3) As laid out by Reviewer #4, it is important that the authors place their efforts in context of previous work that systematically examined effects of related stimulus parameters in human somatosensory thalamus. In general, several reviewers provide very useful literature that should be cited appropriately and discussed as needed.

We would like to thank the reviewers for pointing out the literature on thalamic stimulation. This is an important body of work we have discussed in other work. However, we originally chose not to reference this here because it is likely that the effects we observed are limited to cortical stimulation. However, we will reference this work and draw links to the cortical stimulation experiments that we conducted.

Reviewer #1 (Recommendations for the authors):

Thank you for a really interesting manuscript.

I suggest to have a closer look at Prsa et al. Nature 2019 which shows stimulus-response curves for cortical neurons in mice showing similar dependency of stimulation frequency as observed in the current study.

Thank for the positive feedback. Prsa et al. 2019 is indeed a very relevant paper and we find the parallels between our own work and the Prsa study very intriguing. Specifically they found that different neurons in S1 responded very differently to mechanical stimulation at different frequencies (similar to our finding that different ICMS frequencies evokes different perceptual and intensity effects). They also found that decreasing the mechanical stimulus amplitude biased the preferred frequency higher (conceptually similar to our finding that as ICMS current decreases, higher stimulation frequencies are more easily detected). However, it is important to note that there are several key differences between the studies that complicate any direct comparison. Specifically, the most substantial effects in our study occurred in a frequency range from 20-100 Hz, which incorporates the ‘flutter’ range where cortical neuron firing is phase-locked to the stimuli. Conversely, Prsa studied frequencies from 100-800 Hz where phase-locking does not occur. The following text has been added to the discussion (Line 243):

“Similarly, individual cells in mouse cortex are preferentially activated by different mechanical stimulation frequencies (Prsa et al., 2019). In those same experiments, the frequency preference of the neural population tended towards higher frequencies when the indentation depth decreased, similar to our results that higher frequencies were perceived as being more intense when the ICMS amplitude was decreased (Figure 2).”

Title: I am not convinced at all that there are spatially organised clusters as the title suggests. The number of electrodes is too small and there are many factors (including technical) which might explain why the electrode properties are not at random. I would be very careful making such a statement.

Thank you for the feedback. This was a common concern amongst reviewers. We agree that the number of samples on each array was small, however we are confident that the statistical analysis support the idea that the responses were clustered in these experiments. However, as you and others noted, this could be due to a variety of factors that were not previously emphasized, including layer specificity and potentially electrode-tissue interactions. Further, with data from a single subject, it remains to be shown whether such a finding is broadly true. However, we now have data from a second participant in which similar clustering was observed. We have updated the discussion to emphasize alternative explanations and the point that while the relationships are clustered, this may or may not be due to cortical organization. Further, we have revised the title to not include this issue about spatial clustering.

The title is now:

“Perception of microstimulation frequency in human somatosensory cortex”.

The following text was added to the discussion (Line 328):

“Our results are consistent with the idea that somatosensory cortex is organized in a way that represents different features in different locations, however, there are several limitations that should be considered. […] However, if this were the case, this would still reflect important functional differences in cortex which need to be understood for bidirectional BCIs.”

The electrodes are spaced 400 μm apart, but is it known within how large an area around the electrode neurons are activated?

This is a great question that unfortunately doesn’t have a simple answer. While there is literature measuring and simulating how stimulation activates elements as a function of stimulus parameters and distance from the electrode (see (Overstreet et al., 2013; Stoney et al., 1968)), this can depend on several factors that are difficult to measure including any tissue encapsulation. Activation volumes have been estimated to be anywhere from a few hundred microns to a few millimeters for 100 μA currents (Overstreet et al., 2013). Perhaps the more complicated issue is that local stimulation can activate passing axons that project to distant cortical areas. Studies in mice have shown that stimulation can sparsely activate distant neurons hundreds of microns from the stimulating electrode (Histed et al., 2009; Michelson et al., 2019) but this itself may be frequency dependent. Ultimately, it is difficult for us to know exactly what volume of tissue is activated by ICMS. Nevertheless, passing axons close to electrodes may themselves be organized (by layer or column) and could contribute to clustering we observed. With our setup, the precise neural mechanisms that result in “frequency preference” is elusive, and will likely require further studies in animal models. We have added the following text to the discussion about the effects of ICMS on local and distant neural populations (Line 216):

“Intracortical microstimulation at the maximum amplitudes used in this study can directly activate tissue up to 2 mm away from the electrode tip, but most activation occurs less than 500 µm from the electrode tip (Overstreet et al., 2013; Stoney et al., 1968). […] These spatial scales over which function varies are similar to the expected recruitment distances from ICMS, supporting the idea that different perceptual or frequency responses may be due to activation of different functional groups of neurons.”

Figures: All figures report SEMs, however SDs or CI would be more appropriate to evaluate variability.

We agree that SD or CI would be most appropriate for assessing variability within a group. However, here, we are primarily interested in whether the mean values at different frequencies are different for the three different categories of electrodes. In this case, standard error shows how well we are estimating the mean, and we feel this is the most appropriate metric. Psychophysical studies of electrical stimulation typically use SEM in similar situations (Callier et al., 2019; Graczyk et al., 2016; Kim et al., 2015b, 2015a). Standard deviation would instead be informative about how much the participants’ intensity reports varied on different electrodes within a group. While the overall magnitude of the intensity is related to the group with which an electrode is divided into, the specific numbers used can vary from one electrode to another and even from day to day (although we found the day-to-day variance was not significant). How the intensity measures vary on electrodes within a group is not of interest and the standard deviation then doesn’t really illustrate what we are interested in. Therefore, we believe that the standard error is the more appropriate statistical metric to report.

Lines 149-153: I don't see contradiction here as HFP electrodes in general had very low median intensity. Figure 2A also shows that for HFP electrode at 20 μm the increased frequency leads to more intense percept.

Thank you for this comment. We have revised this statement to try and better capture our thoughts. Briefly, we don’t believe that our data are contradictory with the findings in NHPs. Rather, the predictions that were made from NHPs (included in the discussion of both Callier et al. 2020 – see Discussion section titled “Increased ICMS Frequency Leads to Increased Perceived Magnitude” – and Kim et al. 2015 –see Discussion section titled “Effects of Frequency.”) may be inaccurate. We work closely with the group responsible for this work (the Bensmaia group) and have spoken at length about how our results work together. We believe that the detection results from the human participant confirm that the directly-comparable data (detection thresholds) are not contradictory as the results from stimulation at 20 µA (Figure 2) is more in line with what is expected based on the primate work. However, the frequency effects we saw at higher amplitudes could not be predicted from monkeys because these percepts would always be detectable and monkeys can’t reasonably report intensity. Therefore, we believe these results do not contradict NHP findings, but rather give us new insight that the monkey experiments could not. To clarify our thoughts, here are the updated passages (Line 160):

“Our observation that higher stimulus pulse frequencies can evoke less intense percepts at suprathreshold stimulus current amplitudes differs from predictions made from non-human primate studies; higher frequencies evoked detectable percepts at lower amplitudes in NHPs, which led to predictions that higher frequency always results in higher perceived intensities ((S. Kim, Callier, Tabot, Gaunt, et al., 2015; Romo et al., 2000, 1998).”

The remainder of this paragraph presents the idea that increases in perceived intensity only occur consistently when the stimulus amplitudes are very low. It is only at higher amplitudes where the differences begin to emerge (Figure 2).

Additionally, we changed the title of the Discussion section related to this point (Line 275):

“ICMS in humans directly evaluates intensity and perception”.

This revised Discussion section expands upon this point.

Line 167: see Callier et al. PNAS 2020 who developed paradigms to address this.

In this paper, the authors vary amplitude and frequency together to separate out the effects of amplitude and frequency on perception. They found that amplitude biases discrimination of frequency on some electrodes more than others. And still on other electrodes, the monkeys are not able to perform the task. We believe that non-linear effects of frequency on perception could result in monkeys being unable to perform the discrimination task. The following text in the Discussion directly addresses this point (Line 284):

“To determine if changes in frequency could be perceived independently of changes in amplitude, animals were trained to identify which of two intervals contained the higher frequency stimulus train, regardless of current amplitude (Callier et al., 2020). […] On electrodes without highly salient frequency-dependent qualities, such as the HFP electrodes, it would be difficult to disambiguate changes in amplitude and frequency.”

Paragraph 208-225: The logic and writing is not very clear. Can you please clarify your views and refine the text?

We apologize for the confusion that this text created. Briefly, the major point of this text is that previous literature emphasized the importance that specific types of peripheral afferents (RAs, SAs, PCs), had on the response in cortex. More specifically, the thought was that cortical neurons basically just reflected the way that different peripheral afferents responded to mechanical input. If this were true, direct cortical input through ICMS should have had the same effects everywhere. Since we saw that there were differences in intensity and percept quality at the same frequencies in different locations, this suggests that different regions of the cortex may be tuned to process ‘features’ of the input, and not simply reflect peripheral afferent activity. This is consistent with recent work in the mouse (Prsa et al., 2019)Click or tap here to enter text. and the fact that different peripheral submodaliti es (SAs, RAs, PCs) actually all converge on many cortical neurons (Pei et al., 2009). We have significantly revised this section and hope that it is now more clear (Line 228):

“Electrophysiological (Mountcastle, Talbot, Sakata, and Hyvärinen, 1969; Sur, Wall, and Kaas, 1981; Sur et al., 1984) and optical (L. M. Chen, Friedman, Ramsden, LaMotte, and Roe, 2001; Friedman et al., 2004) recordings have shown organized neural populations in the somatosensory cortex that are sensitive to tactile input with specific frequency content. […] Therefore, the different effects of stimulus frequency on intensity and perception are unlikely to arise from activating inputs representing specific tactile subpopulations (e.g. SAs, RAs, or PCs), but by how a local region of somatosensory cortex can respond to different stimulus frequencies, consistent with the concept of the cortex encoding different frequency features (Prsa, Morandell, Cuenu, and Huber, 2019).”

217-219: Do you mean that convergent inputs from different sub-modalities would necessarily imply uniformity of distribution? This is not what studies mentioned in that context would suggest. Please clarify this statement.

We apologize that this statement did not accurately convey our thoughts. We did not mean to imply that all cortical neurons receive equal input from all tactile submodalities. In the dorsal column nuclei and thalamus there is a clearer preservation of submodality specific activity. However, the study by Pei (Pei et al., 2009)Click or tap here to enter text. shows that this concept breaks down in the cortex with many cortical neurons representing activity that is associated with both rapidly and slowly adapting responses. In area 1 (where our arrays are implanted), about 40% of the neurons received these mixed inputs. The papers we referenced were presented as a counterpoint to papers (e.g. (Friedman et al., 2004; Sur et al., 1984)) that suggest that cortical neurons can be classified simply as SA or RA neurons. We believe that our work provides further evidence that the somatosensory cortex does not appear to be organized by segregated input modalities consistent with labelled line inputs, but rather the cortex itself is organized in a manner to preferentially encode different features of perception. This text has been revised as follows (Line 231):

“These experiments promoted the idea of submodality separation in the cortex in which the activity of cortical neurons is primarily driven by input from either rapidly adapting Meissner corpuscles (RAs), slowly adapting Merkel cells (SAs), or Pacinian Corpuscles (PCs). However, many cortical neurons receive heterogeneous input from multiple classes of mechanoreceptors (Pei, Denchev, Hsiao, Craig, and Bensmaia, 2009; Reed et al., 2010; Saal and Bensmaia, 2014), leading to neurons that exhibit both sustained and transient responses.”

Line 225: consider ref Birznieks et al. eLife 2019.

Thanks for the suggestion. We have added the following sentence to our discussion (Line 241):

“The idea that somatosensory cortex is organized for feature encoding is supported by human psychophysics experiments where frequency perception was dependent on specific spiking patterns and not on the types of mechanoreceptor that were activated (Birznieks et al., 2019).”

Paragraph 226-233: consider reference to Prsa et al. Nature 2019.

Thanks for this suggestion. This paragraph was revised as follows (Line 243):

“Similarly, individual cells in mouse cortex are preferentially activated by different mechanical stimulation frequencies (Prsa et al., 2019). In those same experiments, the frequency preference of the neural population tended towards higher frequencies when the indentation depth decreased, similar to our results that higher frequencies were perceived as being more intense when the ICMS amplitude was decreased (Figure 2).”

Lines 235-241: out of context it is not clear what are distal neurons.

We agree that this term may have been unclear and have revised this paragraph to use a more accurate and descriptive phrase. Essentially, we meant “neurons that are far away from the electrode”. The revised paragraph now reads (Line 252):

“The effects described here must be related to different cellular responses to stimulation in different regions of the cortex. […] This reduced activity in neurons far away from the electrode could lead to decreases in perceived intensity and changes in percept quality in a way that depends on electrode location and local neural populations.”

Section 258-286: I disagree that there is a contradiction with non-human primate studies as I indicated in the public review. To play devil's advocate, the term "intensity" could be used to describe stimuli of different qualities and even modalities, but does it mean that, for example, intensity of olfactory stimuli could be compared with intensity of pain or any other modality? What about intensity of skin stretch vs vibrotactile intensity? Would you expect the same neural code in all those instances? Thus the sense of intensity due to electrical stimulation of cortical neurons may not be compatible with intensity of mechanical stimuli which have indentation depth, certain spatial pattern of afferent activation including size of the area, it's shape and type of borders (sharp/diffuse). All of those features may interact between each other creating a complex integrated percept of intensity which is absent with electrical stimulation.

We have modified this section to be clearer about our thoughts and think that our use of the word ‘contradicts’ was misleading. We have therefore removed this term and focus on the idea that our data appear to be inconsistent with the predictions about the effects of frequency on intensity that have been made from non-human primate studies. With good reason, several papers suggested that increasing ICMS frequency in somatosensory cortex will always result in increases in perceived intensity (see discussions from (Callier et al., 2019; Kim et al., 2015b)). However, in our results we see changes in perceived intensity that are clearly different from these predictions. There are several factors that could lead to these differences. First, in this study we directly measure perceived intensity whereas this can only be inferred indirectly in monkeys. Second, there was a relationship between the way that ICMS frequency modulated intensity and percept quality. When the quality changes, it is possible that certain percepts are under associated with lower intensities. However, there is literature to support the idea that cross-modal intensity matching can in fact be done (Marks et al., 1986). Nevertheless we more clearly point out that intensity and quality both change as frequency changes and specifically address the possibility that quality changes could be a factor in intensity perception. Abstract, Line 27:

“These results support the idea that stimulation frequency directly controls tactile perception and that these different percepts may be related to the organization of somatosensory cortex, which will facilitate principled development of stimulation strategies for bidirectional BCIs.”

Discussion, Line 299:

“Another interpretation is that since frequency can change percept quality (Figure 3), different qualities are understood to have different intensities.”

Page 8: statements in two sections in regard to frequency filtering in cortex and human psychophysics: One possibility is that the function of some of these electrically stimulated circuits might be to detect presence of a given stimulus quality (specific qualitative feature) which might be associated with specific discharge rate in those neurons – like, for example, discharge rate <60Hz (>15ms inter-spike intervals) might mediate sense of flutter. There is experimental evidence which might support such arrangements preferentially detecting inter-spike intervals of certain length – in case of presence of short (<15ms) and long inter-spike intervals in afferent activity, the short inter-spike intervals (corresponding to high frequency) are simply ignored and perceived frequency corresponds exclusively to the longest (flutter range) inter-spike interval and not periodicity or mean discharge rate. When shorter inter-spike intervals became longer, their weighted contribution to perceived frequency increases. This has been demonstrated with mechanical (Birznieks and Vickery Current Biology 2017) and electrical stimulation (Ng et al. PlosOne 2020) and might fit with the current observations. Vibrotactile stimuli are very simplistic, but they can demonstrate principles how neural activity translates into perception.

Thanks for the interesting thoughts. We certainly agree that the observed frequency responses might be a consequence of detecting specific qualitative features. This was a point we tried to make in the discussion and make more clearly now (see below). We also agree that somatosensory cortex may encode certain ‘features’, and may be more sensitive to large interpulse spacings. These could be the electrodes we identified as “low” or “intermediate” preferring. This would imply that the other areas of cortex, the “high frequency preferring” are sensitive not to the interpulse spacing, but rather the overall rate (or number of pulses). The results from the Birznieks lab are definitely relevant to our work. In fact, we have conducted another set of experiments in which we observed if the changes in the interpulse timings affected perception of ICMS in a way that was consistent with the Birznieks papers cited here. For the small data set collected, we found this to be true: increasing pulse spacing resulted in a decrease in perceived frequency. This was presented as a 4-page paper at the IEEE NER conference titled “Changes in interpulse spacing changes perception of microstimulation in human somatosensory cortex.”

Line 241:

“The idea that somatosensory cortex is organized for feature encoding is supported by human psychophysics experiments where frequency perception was dependent on specific spiking patterns and not on the types of mechanoreceptor that were activated (Birznieks et al., 2019).”

Methods section: Please mention where the reference electrode was.

Apologies for this omission. The ‘return’ electrode or ‘anode’ for stimulation is actually the titanium pedestal that is screwed to the skull. As a result, ICMS is effectively monopolar (distant return electrode). We have added this detail to the methods (Line 389):

“The stimulation return electrode was the titanium pedestal that was fixed to the skull.”

Reviewer #2 (Recommendations for the authors):

I am wary about the fitting and clustering approaches.

– For instance, in Figure 1, the purple curve fit of panel A seems inappropriate – the data would be better fit with a monotonically decreasing function.

We agree that the purple curve here should not be used to make assumptions about how these electrodes behave. We intended that the curves be understood simply for illustrative purposes to emphasize the differences between the group trends. However, given the potential confusion and the point raised here, we have revised all the figures to use simple piecewise fits between each data point. In the specific case mentioned (purple data in Figure 1A), there actually is a low intensity data point at the lowest frequency, which would make a monotonically decreasing function inappropriate. We believe that replacing these curve fits with simple connecting lines will help illustrate this more clearly and reduce confusion. Additionally, to the figure legends we have added:

“The points are connected with piecewise fits.”

– Similarly, the data in C seem to differ from D only in the very first data point of each data series. So the difference between C and D appears to be at a very low frequency only (?). I cannot assess whether this is generally true as only 2 curves are illustrated, and the use of 100 Hz in the other parts of the manuscript suggest otherwise. As is, this is an unclear point for me.

We agree that most of the changes appear in the lowest frequency ranges. Above about 150 Hz, there were only small relative changes in intensity within a group of electrodes. However, in Figure 1 B,C,D, we did not keep the y-axis range consistent in order to emphasize the shape of the relationships on individual electrodes, which may lead to some misinterpretation. Panel A shows the dominant effects more clearly. The low and intermediate groups have similar intensity responses at 20 Hz, but the intermediate group has higher intensities at all other frequencies. Specifically, both low and intermediate groups have intensities of 1.5-2 around 20 Hz. Low groups then have intensities of 1-1.5 for 40+ Hz and intermediate groups have intensities of 2-2.5 for 40+ Hz. This is what results in the different shapes of the curves.

– If there is actually more spread of the preferred frequency, then I wonder how adequate it is to cluster, rather than assume, say, a continuously changing gradient across a larger cortical region.

This is an interesting suggestion, and we think that it is possible that these effects could be more continuous rather than being divided into the three groups as we proposed here. To determine these groups we used both silhouette and elbow analysis of our k-means clustering results to confirm that separating into three groups was reasonable. And in fact, in our second participant, clustering into two groups was more reasonable based on this same approach. However, with additional data from new electrodes distributed across larger regions of the cortex, it is certainly possible that more continuous distributions could emerge. In our specific case however, three groups captured the data well. We have added the following text to the discussion to address this point (Line 328):

“Our results are consistent with the idea that somatosensory cortex is organized in a way that represents different features in different locations, however, there are several limitations that should be considered. […] However, if this were the case, this would still reflect important functional differences in cortex which need to be understood for bidirectional BCIs.”

– Related, I understand that 3 comes out as adequate from the clustering analysis, but this is based on a small number of sites. This is inherent in using electrode arrays. If more sites were available, do the authors think a 3 cluster interpretation would still be supported? What speaks against a more fine-grained, e.g. gradient-like, distribution? In other words, is there evidence against such an organization, or is the 3 cluster solution possibly just due to the small amount of testable regions/electrodes? Do the authors prefer a categorical account because they think there is a direct link to distinct perceptual qualities?

Thank you for these interesting comments. This response is related to several of the previous responses. We believe that it is possible that the divisions that we described are in fact more continuous. We simply do not have the data to conclusively rule out either of these two interpretations. However, the three groups were parsimonious explanations of the data and there are physiological reasons to think that the responses may be more effectively considered to divide into groups, rather than be distributed across a continuum. While it might be tempting to think that the three divisions align with the frequency response ranges of SA1, RA and PC mechanoreceptors, this had nothing to do with how we approached it. Neurons in somatosensory cortex can receive converging input from different classes of mechanoreceptors, and as such, these frequency response ranges probably do not directly reflect mechanoreceptor responses. An alternative view from optical imaging experiments showed these relationships as being more continuous (Friedman et al. 2004). Nevertheless, the categorical account given here made sense from our observation and clustering analysis, particularly when considering the quality reports. We have added the following text to the Limitations section of the Discussion to address this issue (same text as previous question and response).

– I find it difficult to take away a "higher-order" result. To me, the presented work is clearly illuminating in that the different tested manipulations must be acquired to know in what kind of percepts they result. Also, the finding that supra-threshold stimulation results in different conclusions than near-threshold stimulation seems to me an important point. However, in many passages, the paper reads mainly like a report of all the different detailed tested aspects. I am missing some more "visionary" ideas about what the results might mean.

We hope that this revised manuscript more clearly addresses this point and we apologize that our broader vision of the importance of these experiments did not come through clearly. The reviewer is correct in that these data provide new information about relevant factors that must be considered to construct bidirectional brain-computer interfaces. However, more importantly, we believe that this work provides direct insight into the organization of perception in somatosensory cortex that can only be inferred from animal experiments. While we knew that different regions of the cortex could respond to different types of mechanical input, it was not clear whether this itself was causally linked to different percepts. Stimulation experiments, as we did here, provides this causal link. The text has been revised throughout, but a specific point to this effect is now the conclusion of the abstract (Abstract, Line 27):

“These results support the idea that stimulation frequency directly controls tactile perception and that these different percepts may be related to the organization of somatosensory cortex, which will facilitate principled development of stimulation strategies for bidirectional BCIs.”

Reviewer #3 (Recommendations for the authors):

The paper is interesting and well-written. I think it could be improved by addressing the following issues:

1. The interplay between frequency and amplitude is of course very important: it could be interesting to see whether the authors could find a unifying model/equation as done in Graczyk et al., 2016 for PNS stimulation

Thank you for this suggestion. This would be ideal. However, the key to building a function as the Graczyk paper did is the consistent effect of frequency on intensity in the periphery. Because the relationships are electrode specific here, it is not possible to develop a unifying equation that could make predictions about intensity. We could potentially create functions for each preference group, but it is not clear how useful this would be.

2. Figure 1 seems to show that many electrodes have a small "intensity dynamics": could this be a problem for closed-loop control? The authors should elaborate a bit more on this

This would be an issue if we were attempting to use frequency to modify intensity. However, because of the non-linear and electrode-to-electrode effects of frequency, it would not be a first choice to modulate intensity. Rather, modifying amplitude provides a reliable way to modulate intensity, and also allows much better dynamics (see Figure 1—figure supplement 1, although the intensity is normalized here). Amplitude allows better modulation, but we have also found multielectrode stimulation can be used for adjacent electrodes to increase intensity, and this could be another potential means to modulate intensity dynamically (but requires further investigation). Based on these results, we believe frequency could be modified based on the periodicity of the input and the electrode being stimulated, but not specifically to modulate intensity. We have tried to clarify these thoughts in the Discussion section “Implications for prostheses.” (Line 348):

“Stimulus amplitude linearly modulates intensity, while frequency has non-monotonic and electrode specific effects on intensity and percept quality. […] Future studies should assess the efficacy of these parameters.”

3. Figure 4 is very interesting but it could stronger and more convincing by testing more electrodes.

Thank you for the positive feedback. This general issue was raised by several other reviewers as well. This figure has been modified and the general results and discussion surrounding this have been changed. We agree that to fully test this basic idea more experimentation will need to be done. Please see the responses to Reviewer 1, Questions 2 and 3 for more information.

4. the last section of the discussion on personalization is very interesting and the authors should elaborate a bit more also in this case. What kind of solution do they have in mind? Biomimetic? or Machine learning based?

This is an incredibly interesting and relevant question as we are currently investigating both approaches in our lab. We are using biomimetic stimulation (based on neural recordings and TouchMime models; see references below)) to understand if biomimetic stimulation can change the perception of stimulation in a desirable way. We are also testing methods of Bayesian inference to allow parameter optimization to select desirable parameters. How these methods may be useful and how they can potentially be combined remains to be seen. We have elaborated on this in the “Implications for prostheses” section (Line 358).

“Second, these results suggest that electrode-specific stimulation encoding schemes would be particularly useful. […] These methods could ultimately improve the usefulness of somatosensory feedback, in turn improving the performance of bidirectional BCIs and ultimately improving the quality of life for people living with paralysis.”

5. it is not clear the duration of the overall testing. This is important to gather more information about the stability of the sensations over time.

We apologize for this oversight and have made this more clear in the methods section. We tried to limit all data collection to one year to avoid results being inconsistent with each other over longer periods of time. Added in Methods (Line 378):

“All data included in this paper (including magnitude estimation, surveys, detection thresholds, etc.) were limited to one year of data collection in P2 to minimize the impact of changes in perception that can occur over long time periods. Data in P3 were collected over two months.”

Reviewer #4 (Recommendations for the authors):

A few questions and suggestions:

The major one being that the authors need to put their work in context of the original work that systematically examined effects of stimulus parameters in human somatosensory thalamus (Dostrovsky et al. 1993 Adv Neurol). Of course more publications exist on cortex recently but several of the concepts presented here as new were already studied and understood to be true in human thalamus. Also the major limitation is that all these experiments were performed in deafferented cortex; it is not just that it is one subject, but that their cortex must be reorganized. Of course that is what makes the prosthetic side of the story stronger (rather than the comparisons to normal non-human primate experiments), in that it remains amazing that sensations can be elicited from cortex that has not received afferent input in years.

We thank the reviewer for the suggestions. We tried to find the specific article referenced here titled “Electrical stimulation-induced effects in the human thalamus” and while we were able to find the citation, we could not find the original article anywhere. We did go through all of the papers that cited this paper, of which there were 25, and looked for relevant articles. Additionally, we reviewed more recent papers on thalamic stimulation. We have attempted to include the relevant information in the introduction and discussion. We have currently added the following text to the introduction (Line 59):

“More is known about the perceptual effects of stimulating the human thalamus (Davis, Kiss, Tasker, and Dostrovsky, 1996; Heming, Sanden, and Kiss, 2010; Ohara, Weiss, and Lenz, 2004; Swan, Gasperson, Krucoff, Grill, and Turner, 2018; Willsey et al., 2020). […] Ultimately, temporal factors have clear effects on the sensations evoked through thalamic stimulation, but it remains unclear how to optimally control these parameters to manipulate percept quality.”

We agree that the cortex being deafferented is likely a major factor to consider in this work, and as such we have included more information about this in the discussion. Although much recent work has promoted the idea that cortex doesn’t actually reorganize after injury in ways previously thought which may explain why we are still able to induce naturalistic sensations years after injury. We added the following text to the Limitations section (Line 315):

“The participants had limited residual sensation in their hands, which made it difficult to measure responses in cortex to tactile indentation. […] The ability to elicit sensations with ICMS years after injury is supportive of this idea (Armenta Salas et al., 2018; Fifer et al., 2020; Flesher et al., 2016)”.

1. To improve the wording of second last sentence in abstract. It does summarize the results but is somewhat confusing: is it the electrodes that were of 3 types or the brain sites stimulated.

We apologize for the confusion. We believe that the three groups are a result of the structure and function of the different brain regions activated by stimulation. We have rewritten the end of the abstract to make this point clearer (Line 24).

“These different frequency-intensity relationships were divided into three groups which also evoked distinct percept qualities at different stimulus frequencies. […] These results support the idea that stimulation frequency directly controls tactile perception and that these different percepts may be related to the organization of somatosensory cortex, which will facilitate principled development of stimulation strategies for bidirectional BCIs.”

2. While I understand how difficult it is to plot the different sensory modalities on a graph and I appreciated the radar plots in Figure 3 as the best way to do this, I was wondering how much altering different frequencies and intensities could alter the percepts evoked through a single microelectrode. Because it is unlikely we will have multiple opportunities to move cortical arrays in humans, to make this a practical application we need to know how much we can use electrical parameters to modulate percepts.

The perceptual quality and intensity can vary significantly across frequencies for individual electrodes. The group trends are very representative of the effects on individual electrodes within the group: a low frequency preferring electrode will likely have very different percepts at 20 Hz vs 100+Hz. Stimulation amplitude on the other hand tends to only modulate the intensity and not the quality. Figure 3 summarizes the effects of changing frequency on perception. Certainly more work needs to be done in this area, but we believe that this type of analysis is representative, although it will need to be validated in more participants.

3. I did not follow the pseudo-p LISA statistical analysis shown in Figure 4. I see what the authors are trying to say in Figure 4A, but Figure B may either be simplified, better explained in the legend or perhaps moved to supplemental.

We apologize for the confusion. The intention of this figure was to demonstrate the results of the random simulations that showed the clustering measured experimentally was significantly greater than chance. However, we agreed that Figure 4B is not a critical finding that needs to be included in the main figure. In fact, we have replaced Figure 4B with the somatotopic organization across the arrays in participant P2, which we feel is a more illustrative and useful figure. The figure legend has been appropriately updated. Information about how the p-values were calculated will remain in the methods section.

4. Why did the authors not test the effects of pulse width. It appears they were trying to test the entire parameter space that can be applied with electrical microstimulation, why not this one?

This is a great suggestion, and one that we wish that we could easily manipulate. However, our clinical protocol limits pulse width because the preclinical safety studies that were performed did not vary pulse width. We expect that as new studies are conducted, and the general effects of electrical stimulation safety become more well understood, pulse width modulation could be a useful may to uniquely modulate perception.

5. What about applying stimulation through multiple electrodes simultaneously? In fact this may help answer the question about whether short-term plasticity is involved (from discussion).

This is an interesting suggestion. Multielectrode stimulation may be desirable for many reasons, and we hope to investigate using this feature in detail to understand its usefulness. We have performed limited studies showing that multielectrode stimulation can expand the dynamic range for intensity. Further, multielectrode stimulation studies are the focus of some ongoing work and we hope to share those results in the coming years.

6. Please clarify if this patient is the same one reported in previous papers. The introduction suggested that it was and perhaps this was why the projected fields evoked with microstimulation were not described. If this is the case then how did Figure 4 compare to the somatotopy described in the previous paper? And what part of S1 do they believe the array is located (i.e. what Brodmann area 1, 2, 3a,b)? Adding a statement about what cortical layer the authors believe the microelectrodes are located before the limitations section of the discussion would be helpful. There are a few words about somatotopy at the end of the Results section indicating that all three types of responses: low, med, and high frequency preferring regions can subserve the same body region. However because the statement says "in some cases", it makes it equivocal, yet is a major component of the discussion.

We apologize for the lack of clarity here. Most of the data in this manuscript are from the same participant as previous papers. We have added details to clarify this (Line 384).

“Results from this participant have been reported previously (Flesher et al., 2016, 2021; Hughes, Flesher et al., 2020).”

Because these results have been reported before, we originally chose to not report somatotopic maps. However, we have now modified Figure 4 to include this information. We believe the electrodes are in Layer IV based on the length of the electrodes, but this is impossible to confirm. It is possible that electrode arrays with multiple electrodes along the shank could more accurately determine depth, but with these electrodes, this is not possible. We have added text to the discussion for this (Line 339):

“Third, we do not know if electrodes across the array are in different layers of cortex. Different layers of cortex may drive different perceptual responses with the same input. However, if this were the case, this would still reflect important functional differences in cortex which need to be understood for bidirectional BCIs.”

The reason we did not strongly emphasize the result that the frequency preference groups could span multiple projected fields is because our sample of the cortex is very small and we really only have one example of a single projected field with more than one frequency preference. Because of this, we did not want to emphasize this as a finding of the work.

7. The authors describe that because the subject is deafferented they could not identify receptive fields from the recordings. Were recordings performed at all in the deafferented cortex? While that is not the subject of this paper, it would be a welcome addition to the literature.

Thank you for this comment. We agree that this type of information would be very useful, but as a result of the spinal cord injury, this is very difficult. We are developing a better apparatus to more explicitly probe this issue in ongoing work.

8. Please clarify the consistency of results longitudinally. Figure S3 does not seem to demonstrate all electrode showing same results over time: 3 of 7 electrodes (2, 3, 36) seem different over time.

We agree that the figure might look like this. However, there was no statistically significant change over time for any of these tested electrodes according to Friedman’s test. However, two electrodes, 3 and 36, were more variable for individual frequencies due to the low intensity of the percepts. So, while these electrodes always showed a general increase in intensity across frequencies, there was often variability in the precise shape of the curve. A contributing factor here is that each day only contains 5 trials for each tested frequency. More trials would have resulted in less variability, but would have been an issue for other reasons (e.g. participant engagement in task). This makes individual days of data contain more variability, but the statistics support that these relationships do not change significantly over time.

Callier, T., Brantly, N.W., Caravelli, A., and Bensmaia, S.J. (2019). The frequency of cortical microstimulation shapes artificial touch. P Natl Acad Sci Usa 117, 1191–1200.

Friedman, R.M., Chen, L.M., and Roe, A.W. (2004). Modality maps within primate somatosensory cortex. P Natl Acad Sci Usa 101, 12724–12729.

Graczyk, E.L., Schiefer, M.A., Saal, H.P., Delhaye, B.P., Bensmaia, S.J., and Tyler, D.J. (2016). The neural basis of perceived intensity in natural and artificial touch. Sci Transl Med 8, 362ra142 362ra142.

Histed, M.H., Bonin, V., and Reid, R.C. (2009). Direct activation of sparse, distributed populations of cortical neurons by electrical microstimulation. Neuron 63, 508 522.

Hughes, C., Herrera, A., Gaunt, R., and Collinger, J. (2020). Bidirectional brain-computer interfaces. Handb Clin Neurology 168, 163–181.

Kim, S., Callier, T., Tabot, G.A., Tenore, F.V., and Bensmaia, S.J. (2015a). Sensitivity to microstimulation of somatosensory cortex distributed over multiple electrodes. Frontiers Syst Neurosci 9, 47.

Kim, S., Callier, T., Tabot, G.A., Gaunt, R.A., Tenore, F.V., and Bensmaia, S.J. (2015b). Behavioral assessment of sensitivity to intracortical microstimulation of primate somatosensory cortex. Proc National Acad Sci 112, 201509265.

Marks, L.E., Szczesiul, R., and Ohlott, P. (1986). On the Cross-Modal Perception of Intensity. J Exp Psychology Hum Percept Perform 12, 517–534.

Michelson, N.J., Eles, J.R., Vazquez, A.L., Ludwig, K.A., and Kozai, T.D.Y. (2019). Calcium activation of cortical neurons by continuous electrical stimulation: Frequency dependence, temporal fidelity, and activation density. J Neurosci Res 97, 620–638.

Overstreet, C.K., Klein, J.D., and Tillery, S.I.H. (2013). Computational modeling of direct neuronal recruitment during intracortical microstimulation in somatosensory cortex. J Neural Eng 10, 066016.

Pei, Y.-C., Denchev, P.V., Hsiao, S.S., Craig, J.C., and Bensmaia, S.J. (2009). Convergence of submodality-specific input onto neurons in primary somatosensory cortex. J Neurophysiol 102, 1843 1853.

Prsa, M., Morandell, K., Cuenu, G., and Huber, D. (2019). Feature-selective encoding of substrate vibrations in the forelimb somatosensory cortex. Nature 567, 384–388.

Stoney, S.D., Thompson, W.D., and Asanuma, H. (1968). Excitation of pyramidal tract cells by intracortical microstimulation: effective extent of stimulating current. Journal of Neurophysiology 31, 659 669.

Sur, M., Wall, J.T., and Kaas, J.H. (1984). Modular distribution of neurons with slowly adapting and rapidly adapting responses in area 3b of somatosensory cortex in monkeys. Journal of Neurophysiology 51, 724 744.

https://doi.org/10.7554/eLife.65128.sa2

Article and author information

Author details

  1. Christopher L Hughes

    1. Rehab Neural Engineering Labs, University of Pittsburgh, Pittsburgh, United States
    2. Department of Bioengineering, University of Pittsburgh, Pittsburgh, United States
    3. Center for the Neural Basis of Cognition, University of Pittsburgh, Pittsburgh, United States
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    For correspondence
    clh180@pitt.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9257-8659
  2. Sharlene N Flesher

    1. Rehab Neural Engineering Labs, University of Pittsburgh, Pittsburgh, United States
    2. Department of Bioengineering, University of Pittsburgh, Pittsburgh, United States
    3. Center for the Neural Basis of Cognition, University of Pittsburgh, Pittsburgh, United States
    4. Department of Neurosurgery, Stanford University, Stanford, United States
    5. Department of Electrical Engineering, Stanford University, Stanford, United States
    Contribution
    Conceptualization, Data curation, Software, Investigation, Visualization, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
  3. Jeffrey M Weiss

    1. Rehab Neural Engineering Labs, University of Pittsburgh, Pittsburgh, United States
    2. Department of Physical Medicine and Rehabilitation, University of Pittsburgh, Pittsburgh, United States
    Contribution
    Data curation, Software, Methodology, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1332-674X
  4. Michael Boninger

    1. Rehab Neural Engineering Labs, University of Pittsburgh, Pittsburgh, United States
    2. Department of Bioengineering, University of Pittsburgh, Pittsburgh, United States
    3. Department of Physical Medicine and Rehabilitation, University of Pittsburgh, Pittsburgh, United States
    4. Human Engineering Research Laboratories, VA Center of Excellence, Department of Veterans Affairs, Pittsburgh, United States
    Contribution
    Resources, Funding acquisition, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6966-919X
  5. Jennifer L Collinger

    1. Rehab Neural Engineering Labs, University of Pittsburgh, Pittsburgh, United States
    2. Department of Bioengineering, University of Pittsburgh, Pittsburgh, United States
    3. Center for the Neural Basis of Cognition, University of Pittsburgh, Pittsburgh, United States
    4. Department of Physical Medicine and Rehabilitation, University of Pittsburgh, Pittsburgh, United States
    5. Human Engineering Research Laboratories, VA Center of Excellence, Department of Veterans Affairs, Pittsburgh, United States
    Contribution
    Resources, Supervision, Funding acquisition, Project administration, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-4517-5395
  6. Robert A Gaunt

    1. Rehab Neural Engineering Labs, University of Pittsburgh, Pittsburgh, United States
    2. Department of Bioengineering, University of Pittsburgh, Pittsburgh, United States
    3. Center for the Neural Basis of Cognition, University of Pittsburgh, Pittsburgh, United States
    4. Department of Physical Medicine and Rehabilitation, University of Pittsburgh, Pittsburgh, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Validation, Methodology, Project administration, Writing - review and editing
    For correspondence
    rag53@pitt.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-6202-5818

Funding

Defense Advanced Research Projects Agency (N66001-16-C4051)

  • Michael Boninger
  • Jennifer L Collinger
  • Robert A Gaunt

National Institutes of Health (UH3NS107714)

  • Michael Boninger
  • Jennifer L Collinger
  • Robert A Gaunt

National Institutes of Health (U01NS108922)

  • Michael Boninger
  • Jennifer L Collinger
  • Robert A Gaunt

National Science Foundation (DGE-1247842)

  • Sharlene N Flesher

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

Acknowledgements

We would like to acknowledge N Copeland and Mr. Dom for their extraordinary commitment to this study and insightful discussions with the study team, as well as Debbie Harrington (Physical Medicine and Rehabilitation) for regulatory management of the study. This work was supported by the Defense Advanced Research Projects Agency (DARPA) and Space and Naval Warfare Systems Center Pacific (SSC Pacific) under Contract N66001-16-C4051 and the National Institute of Neurological Disorders and Stroke of the National Institutes of Health under Award Numbers UH3NS107714 and U01NS108922. SNF was supported by an NSF Graduate Research Fellowship under grant number DGE-1247842. Any opinions, findings, and conclusions or recommendations expressed here are those of the authors and do not necessarily reflect the views of DARPA, SSC Pacific, or the National Institutes of Health. The funders had no role in the study design, data collection, interpretation of the results, or the decision to submit this work for publication.

Ethics

Clinical trial registration ClinicalTrials.gov (NCT0189-4802).

Human subjects: This study was conducted under an Investigational Device Exemption from the U.S. Food and Drug administration, approved by the Institutional Review Boards at the University of Pittsburgh (Pittsburgh, PA) and the Space and Naval Warfare Systems Center Pacific (San Diego, CA), and registered at ClinicalTrials.gov (NCT0189-4802). Informed consent was obtained before any study procedures were conducted.

Senior Editor

  1. Barbara G Shinn-Cunningham, Carnegie Mellon University, United States

Reviewing Editor

  1. J Andrew Pruszynski, Western University, Canada

Reviewers

  1. Tobias Heed, Bielefeld University, Germany
  2. Silvestro Micera, Scuola Superiore Sant'Anna, Italy

Publication history

  1. Preprint posted: July 17, 2020 (view preprint)
  2. Received: November 24, 2020
  3. Accepted: July 22, 2021
  4. Accepted Manuscript published: July 27, 2021 (version 1)
  5. Version of Record published: August 19, 2021 (version 2)

Copyright

This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Metrics

  • 834
    Page views
  • 121
    Downloads
  • 1
    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
    Debora Fusca, Peter Kloppenburg
    Research Article

    Local interneurons (LNs) mediate complex interactions within the antennal lobe, the primary olfactory system of insects, and the functional analog of the vertebrate olfactory bulb. In the cockroach Periplaneta americana, as in other insects, several types of LNs with distinctive physiological and morphological properties can be defined. Here, we combined whole-cell patch-clamp recordings and Ca2+ imaging of individual LNs to analyze the role of spiking and nonspiking LNs in inter- and intraglomerular signaling during olfactory information processing. Spiking GABAergic LNs reacted to odorant stimulation with a uniform rise in [Ca2+]i in the ramifications of all innervated glomeruli. In contrast, in nonspiking LNs, glomerular Ca2+ signals were odorant specific and varied between glomeruli, resulting in distinct, glomerulus-specific tuning curves. The cell type-specific differences in Ca2+ dynamics support the idea that spiking LNs play a primary role in interglomerular signaling, while they assign nonspiking LNs an essential role in intraglomerular signaling.

    1. Neuroscience
    Wanhui Sheng et al.
    Research Article Updated

    Hypothalamic oxytocinergic magnocellular neurons have a fascinating ability to release peptide from both their axon terminals and from their dendrites. Existing data indicates that the relationship between somatic activity and dendritic release is not constant, but the mechanisms through which this relationship can be modulated are not completely understood. Here, we use a combination of electrical and optical recording techniques to quantify activity-induced calcium influx in proximal vs. distal dendrites of oxytocinergic magnocellular neurons located in the paraventricular nucleus of the hypothalamus (OT-MCNs). Results reveal that the dendrites of OT-MCNs are weak conductors of somatic voltage changes; however, activity-induced dendritic calcium influx can be robustly regulated by both osmosensitive and non-osmosensitive ion channels located along the dendritic membrane. Overall, this study reveals that dendritic conductivity is a dynamic and endogenously regulated feature of OT-MCNs that is likely to have substantial functional impact on central oxytocin release.