Our overall aim in this line of work is to find a method that would allow us to use microstimulation (MiSt) or other neural stimulation modalities and emulate natural neural responses in the somatosensory cortices (S1) and other somatosensory regions as determined necessary for the perception of touch. We hypothesize that similarity of neural responses following MiSt and tactile stimulation will translate into similarity of perceptions. On the other hand, MiSt that produces “unnatural” neural patterns will not result in “natural” touch perception. Therefore, if we can determine the best locations and patterns to produce such naturalistic neural responses, we should create more natural sensations. We may need to consider the neural response in a more extensive set of structures to fine-tune this approach to achieve this goal. Here, we present our findings in a non-human primate (macaque) model. We started with responses in S1 to natural touch as our template in which to optimize our VPL MiSt-induced responses as shown in our previous rat and simulation-based studies (Brockmeier et al., 2011; Choi et al., 2012, 2016; Li et al., 2013, 2014, 2015; Choi and Francis, 2018). We note that the results presented in this paper were recorded circa 2008 and the above optimization methods had not been developed or implemented. However, we feel these results from the macaques should be shared as we move toward human implementation of these systems, where we can directly interrogate our underlying hypothesis that more naturalistic S1 responses lead to more naturalistic sensations. The rodent somatosensory system is significantly different from humans and macaques we utilized (Francis et al., 2008). Here, we used simple non-optimized VPL-MiSt in the acute macaque preparation and show that the somatotopy between VPL and S1 is generally well-maintained with VPL-MiSt, comparable to natural touch in the macaque, as in the rodent (see section “Discussion”).

It has been demonstrated that neuronal activity in the motor cortex can be used to directly control computer cursors and robotic systems via a brain-machine interface (BMI) (Chapin et al., 1999; Serruya et al., 2002; Taylor et al., 2002; Hochberg et al., 2006; Ganguly and Carmena, 2009; Ajiboye et al., 2017; Degenhart et al., 2020). Recently, interest in BMIs has exploded as it has become clear that such systems are likely to restore motor function lost due to spinal cord injury (SCI), neurological disease, or amputation. Such BMIs require a closed-loop configuration that uses not only real-time neural data to move an actuator, such as a robotic arm, but also delivers sensory feedback to the user (Flesher et al., 2021). To date, this feedback has come mainly through the intact visual system of the user who is viewing their performance with the BMI. However, it is known that natural reaching and dexterous tasks require somatosensory feedback for high levels of performance and control. Therefore, somatosensory feedback from sensors on a neurally controlled prosthetic arm/hand presented directly to the user via MiSt of the neural substrate should lead to a better controlled prosthetic. This somatosensory feedback, along with visual feedback, helps control such devices and allows them to become one with the user.

The use of cortical MiSt to directly introduce information into the brain has been demonstrated with some success (Talwar et al., 2002; Fitzsimmons et al., 2007; London et al., 2008; O’Doherty et al., 2011; Flesher et al., 2016, 2021). Several investigators have shown that macro- and microelectrode stimulation in the human somatosensory thalamus (ventral caudal nucleus Vc and nearby thalamus) can produce a variety of somatosensations, including both natural and artificial, ranging from perceptions of touch or movement to sensations of hot or cold, tingling, or the sense of pressure (Lenz et al., 1995; Davis et al., 1998, 1999; Kiss et al., 2003; Ohara et al., 2004; Chien et al., 2017). In many cases, the elicited sensation depended on the stimulus frequency and its amplitude (Patel et al., 2006). Proprioceptive and cutaneous sensory modalities were found to segregate between different regions of the thalamus as described in the literature (Sacco et al., 1987; Kaas, 2007; Francis et al., 2008). This separation should help produce separate touch and proprioceptive channels for sensory input via MiSt or other stimulation modalities.

Although human thalamic studies have been beneficial in demonstrating conscious perceptions induced by (Vc/VPL) electrical stimulation, we still lack a method for producing reliably “normal” sensations. When conducting intraoperative experiments on humans, there are several constraints, such as the amount of time one must rigorously explore the stimulation state space and ethical concerns that limit the areas from which one can sample. Much work on humans has utilized large (>1 mm) macroelectrodes, which are often intended for deep brain stimulation (DBS) to alleviate movement disorders such as tremor or to alleviate chronic pain. This stimulation is generally at high frequencies (100–300 Hz). Wu et al. have suggested that such macrostimulation may activate a neural area 4000-fold greater than MiSt (Wu et al., 2001). This more extensive activation may cause tingling or other paresthetic sensations through the widespread recruitment of axons and neurons.

With these limitations in mind, as an initial step toward developing an optimized somatosensory neuroprosthesis in humans, we have utilized multielectrode neurophysiological techniques in macaques (Macaca radiata) to determine how thalamic MiSt might be used to evoke neural responses in the somatosensory cortex (S1). As the macaque has a similar somatosensory stream to humans, it is a more suitable animal model for such work as compared to the rodent model (Kaas, 2007; Francis et al., 2008). Our protocol involved implantation of multi-electrode arrays in the hand representations of the VPL (VPL; 4 electrodes) and primary somatosensory (S1) cortex (32 electrodes). These simultaneous recordings allowed us to record the response patterns of hundreds of single and multi-units in the S1 cortex during computer-controlled natural touch stimulation and VPL-MiSt in equivalent somatotopic areas. These acute experiments allowed us to directly and quantitatively compare the post-stimulus responses evoked by natural touch and VPL-MiSt in S1 in anesthetized subjects intraoperatively. We used MiSt in the low-frequency range (≤5 Hz), with the work presented here held to just one biphasic MiSt pulse in the VPL. Here, we show that simple MiSt in the VPL elicits S1 cortical responses with similar properties, such as amplitude and duration, to those induced via natural touch, at least in the anesthetized state. Thus, we add evidence that utilizing such VPL-MiSt may be suitable for a somatosensory neuroprosthesis.

All work adhered to NIH Guide for the Care and Use of Laboratory Animals and was approved by SUNY Downstate’s IACUC and followed the recommendations of the Weatherall report, “The use of non-human primates in research.” All efforts were made to minimize animal suffering, including anesthetics for all surgical procedures (see below). For this work, animals were given ad lib food and water.

A total of 357 recording experiments were conducted on three macaques. All utilized simultaneous recordings from the S1 cortical hand area using spaced electrode arrays consisting of two rows of electrodes (2 × 16 electrodes). These experiments yielded neural activity from stimulus fields driven by a natural touch of the hand, or MiSt in the VPL at varying stimulus intensities, where a stimulus field is defined as the brain region responsive to touch at a single point on the periphery, or MiSt at a single point in the VPL. Each experiment typically involved approximately 450 stimulus presentations (at 5 Hz) of touch or VPL MiSt. All macaques also received a 2 × 2 electrode array implanted in the VPL thalamus.

In this paper, we have presented results that indicate VPL-MiSt in macaques can induce neural responses in S1 similar to those produced via natural touching of the hand, as we have previously demonstrated in the rat model (Choi et al., 2016; Choi and Francis, 2018). In Figure 6 we showed the 25 μA VPL-MiSt S1 induced peak activity lagged the higher amplitude VPL-MiSt; further experimentation will be needed to elucidate this temporal shift. However, we expect that this difference between 25 μA and the higher amplitude stimulations (≥ 50 μA) would be due to differences in the population of neural elements exited within the high current density zone of the VPL-MiSt electrodes. It has been shown that both the density of stimulated neural elements and the radius of stimulation increase with increased stimulation amplitude (Histed et al., 2009). Further work will be needed to determine how differences in such local responses propagate to other regions, such as S1, either by VPL neurons or axons going up to the cortex, antidromic activation of corticothalamic fibers and fibers going to the inhibitory thalamic Reticular Nucleus. Despite the worry that VPL stimulation might provoke widespread non-specific neural responses, we measured precise matches between the SFs in the VPL stimulus sites and the SFs in the activated regions of the S1 cortex. Small areas of single digits were readily discerned. This may be because the VPL thalamocortical fibers rapidly conduct somatotopically congruent axon bundles to circumscribed koniocortical target zones in S1. In contrast, antidromically activated corticothalamic fibers are relatively slow and dispersed. We observed that the sizes of the VPL-MiSt SFs in S1 cortex were tightly correlated with stimulus current, suggesting that the current spread approximately spherically in the thalamus before transmitting directly to the 2D surface of S1. We have conducted modeling of this spread in the rat (Choi et al., 2012), which improves our ability to model VPL-MiSt and S1 activation patterns. Bipolar stimulation between two VPL electrodes spaced more than 1 mm apart produced two separate response areas in the S1 cortex. This suggests that the stimulation mainly occurred in the high-current density regions around the electrodes, which has been shown using optical techniques in other sensory areas (Histed et al., 2009).

The ability of the brief VPL-MiSt to emulate the spatiotemporal characteristics of S1 cortical responses to simple natural touch implicates the thalamocortical path as the major determinant in at least initiating these responses, as would be expected. On the other hand, direct cortical MiSt leads to widespread inhibition that typically lasts for 100 ms in the rat with only a very brief initial excitatory response lasting about 2 ms (Butovas and Schwarz, 2003). This could be why it generally takes trains of MiSt in the cortex for perception. Recently, it has been shown that by stimulating over several electrodes simultaneously in S1 that one can induce perception quicker while injecting more total current spread over the array (Sombeck and Miller, 2019). However, the authors stated the following, “RT consistently decreased when we stimulated with more current for the same number of electrodes….” “Interestingly, the RT increased as the number of electrodes increased…, an effect that was more pronounced at smaller total currents.” This may not be the case for VPL-MiSt as in the current paper and our rat work, we see S1 responses like those induced by touch for single VPL-MiSt. However, this needs to be tested in the awake behaving macaque to determine if, like in humans, perceptions can be induced with brief stimuli, such as four pulses at 10 Hz (Patel et al., 2006). In addition, our optimized VPL-MiSt in the rat always spread the MiSt over a set of many VPL electrodes, which was likely due to our setting limits on the current the optimization could use per electrode; apparently leading to a similar conclusion reached from behavioral experiments as used by Sombeck and Miller, which indicate stimulating over several electrodes may lead to faster perception.

We found that strong VPL stimulation in macaques produced five or more oscillatory responses in the S1 cortex, while natural stimulation generally only produced 2–3 oscillatory responses. These oscillations occur at approximately 600 Hz and have been discussed previously (Baker et al., 2003). These results suggest that a possible mechanism for paresthesias is the highly synchronous nature of the VPL stimulation. Our conjecture, therefore, is that the ideal VPL-MiSt is one that closely approximates the response patterns produced by natural stimuli not only in the spatial extent, which has been the focus of this report but also concerning the fine temporal structure of the responses, as accomplished in our rodent work (Choi et al., 2016). However, we ultimately need to move such work into humans to address questions on the qualia of the evoked sensations.

The work we have presented was conducted with an anesthetized macaque preparation. However, we have obtained similar results in the awake restful state in rats indicating these ideas will at least hold in that neural state (Brockmeier et al., 2011). It seems prudent to replicate this in the awake macaque before moving to humans. Indeed, this is just the beginning of this work, as we expect that the awake, actively sensing state would be more complicated. We may need to consider information about other brain regions that feed into S1, such as motor related cortices, in addition to the current state of the S1 cortex and VPL while utilizing VPL-MiSt. Recently it has been shown that S1 cortical activity is modulated by reward expectation (Pantoja et al., 2007; Ramakrishnan et al., 2017; Atique and Francis, 2021; Yao et al., 2021), punishment expectation (Yao et al., 2021) and their delivery. Indicating such affective modulation should also be tested with somatosensory neuroprostheses as this information (reward/punishment) changes neural activity within these regions and this may change perception of induced sensory neural activity.

Recently much work has been conducted using MiSt of S1 cortex directly to evoke percepts (Talwar et al., 2002; London et al., 2008; O’Doherty et al., 2011; Flesher et al., 2021). Work such as that conducted by Romo et al. (1998, 2000) has shown that S1 MiSt can be used by non-human primates as somatosensory information in the flutter vibration domain, while others have shown artificial proprioception via learning (Dadarlat et al., 2015). In addition, our previous work utilizing the Roborat rat paradigm has allowed us to demonstrate somatosensory neural prosthetic capabilities in a rat model. We used MiSt of the primary somatosensory cortex in this previous work with clear results (Talwar et al., 2002). Our preliminary results using this same paradigm with VPL-MiSt have been successful, indicating that at least in the rat, such VPL-MiSt with single biphasic stimuli, like those presented here, are perceivable by the animal (data not shown).

Our neuroprosthetic techniques utilizes the production of a neural template generated via the natural peripheral sensory organ, such as the skin for touch in our case, and working toward minimizing the difference between that template and cortical responses induced via MiSt, such as VPL-MiSt in the present case (Brockmeier et al., 2011; Li et al., 2013; Choi et al., 2016). Our model-based-methodologies (Li et al., 2013, 2015; Choi et al., 2016; Choi and Francis, 2018) work when utilizing a simulation of the periphery and neural substrate as well. This model based approach, which allows us to directly minimize S1 neurophysiological characteristics to VPL-MiSt vs. natural touch, or simulated cortical responses (Song et al., 2013; Choi et al., 2015), complements neurosurgical efforts (Hanajima et al., 2004; Ohara et al., 2004; Patel et al., 2006) that have been conducted utilizing electrical stimulation of the Vc thalamus and peripheral nerves. Significant work on Vc-MiSt has been conducted during surgical implantation of deep brain stimulators into the VIM thalamus for the treatment of tremor. Stimulation of the Vc in humans produces various paresthesias, especially tactile sensations in the core Vc area. Many MiSt-evoked sensations were reported as tingling, which could be related to the use of high frequency stimulation (about 150 Hz). These stimulus trains were found essential for evoking somatosensory perceptions in many, but not all cases. It has been shown that just a few pulses of stimulation can induce perceptions in humans (Patel et al., 2006).

The possibility of future optogenetic implementations of the basic ideas put forth in this paper have already been shown in the retina (Nirenberg and Pandarinath, 2012). Others have described some of the difficulties with current light sensitive ion channels injected into the somatosensory thalamus (Cruikshank et al., 2010) as well as targeting these into the appropriate areas along with possible solutions to such problems (Yizhar et al., 2011). A very attractive aspect of these techniques is the fact that the VPL thalamus is a small deep brain structure that can be inject with a relatively small volume of channelrhodopsins, as compared to the cortex, and have them transported to the thalamocortical synaptic terminals (Cruikshank et al., 2010). One can then then use an optical array at the larger, somatosensory cortex, at least for areas 1 and 2. However, as much of the S1 cortical region related to fine touch and proprioception is buried in the central sulcus in humans and macaques, it may still be difficult to access without causing some damage. In addition, obtaining fine spatial and temporal optical stimulation at depth below the cortical surface without causing damage is still a challenge. As high-density arrays of micro- and even nano- electrodes are evolving, it is very likely that electrical stimulation will remain the chosen intervention for clinical applications.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The animal study was reviewed and approved by the State University of New York Downstate Medical Center IACUC.

JF, AR, LK, SX, PC, MS, and JC conducted the experiments. JF and JC conducted the analysis and wrote the manuscript. EH produced the stimulator. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.