(C) PLOS One [1]. This unaltered content originally appeared in journals.plosone.org.
Licensed under Creative Commons Attribution (CC BY) license.
url:https://journals.plos.org/plosone/s/licenses-and-copyright

------------



Corollary discharge enables proprioception from lateral line sensory feedback

['Dimitri A. Skandalis', 'Department Of Biology', 'Whitney Laboratory For Marine Bioscience', 'University Of Florida', 'St. Augustine', 'Florida', 'United States Of America', 'Department Of Psychological', 'Brain Sciences', 'Johns Hopkins University']

Date: 2021-10

Animals modulate sensory processing in concert with motor actions. Parallel copies of motor signals, called corollary discharge (CD), prepare the nervous system to process the mixture of externally and self-generated (reafferent) feedback that arises during locomotion. Commonly, CD in the peripheral nervous system cancels reafference to protect sensors and the central nervous system from being fatigued and overwhelmed by self-generated feedback. However, cancellation also limits the feedback that contributes to an animal’s awareness of its body position and motion within the environment, the sense of proprioception. We propose that, rather than cancellation, CD to the fish lateral line organ restructures reafference to maximize proprioceptive information content. Fishes’ undulatory body motions induce reafferent feedback that can encode the body’s instantaneous configuration with respect to fluid flows. We combined experimental and computational analyses of swimming biomechanics and hair cell physiology to develop a neuromechanical model of how fish can track peak body curvature, a key signature of axial undulatory locomotion. Without CD, this computation would be challenged by sensory adaptation, typified by decaying sensitivity and phase distortions with respect to an input stimulus. We find that CD interacts synergistically with sensor polarization to sharpen sensitivity along sensors’ preferred axes. The sharpening of sensitivity regulates spiking to a narrow interval coinciding with peak reafferent stimulation, which prevents adaptation and homogenizes the otherwise variable sensor output. Our integrative model reveals a vital role of CD for ensuring precise proprioceptive feedback during undulatory locomotion, which we term external proprioception.

We hypothesized that CD serves a critical role in regulating sensor adaptation from heterogeneous sensors ( Fig 1Bii ). Hair cell activity is autoregulated by activity-dependent pre- and postsynaptic mechanisms, including synaptic depression [ 10 , 11 , 39 , 59 , 60 ], which is graphically represented by vesicle depletion in Fig 1Aiii and is schematically represented by negative feedback in Fig 1Bii . We predicted that because CD acts upstream of hair cell signaling to suppress evoked vesicle release, the CD also suppresses the nonlinear effects of adaptation. Our goals in this work were 2-fold. First, we develop a neuromechanical model of proprioceptive signals and their transduction in the fish body continuum joint. Second, we build on previous examinations of the independent effects of heterogeneity, adaptation, and CD in the lateral line [ 10 , 49 ] by examining the interactions between these factors. These interactions are critical for understanding how sensory processing is modulated during the behavioral shifts between rest and motor activity. The proposed proprioceptive function of the lateral line is enabled by the central role of CD in regulating the magnitude and temporal characteristics of feedback from populations of heterogeneous sensors. The capacity of CD to regulate spike temporal codes as well as rate may have broad significance throughout the nervous system.

The utility of peripheral cancellation of reafference in other animals has led to the view that CD to the lateral line likewise cancels reafference [ 3 ]. This view is based on the principle that cancelling reafference could increase the robustness of behaviors that depend on tracking hydrodynamic stimuli through the lateral line, like schooling, prey detection, and rheotaxis [ 29 , 44 – 47 ]. Cholinergic projections from the hindbrain hyperpolarize hair cells and reduce spontaneous and evoked afferent spike rates (efferent neurons in Fig 1Aii ) [ 13 , 48 , 49 ]. The CD is strictly activated during swim bouts [ 35 , 48 , 49 ], and laser ablation of the cholinergic cell bodies prevents inhibition [ 48 ]. Cancellation could be produced if the strength of sensor inhibition induced by the CD matches the magnitude of reafference ( Fig 1B ). Indeed, hair cells that are polarized to respond to the anterograde (rostrocaudal) currents encountered during forward swimming are the most heavily suppressed [ 49 ]. Nonetheless, the CD-induced hyperpolarization is insufficient to completely suppress spontaneous spiking [ 48 – 50 ] let alone evoked activity [ 49 , 50 ]. Reafferent spikes will likely always be transmitted to the brain, depending on motor effort [ 51 , 52 ], which is consistent with the suggestion that CD represents a sensory goal rather than sensory prediction [ 53 ]. In this way, the spikes can encode the mismatch between goal and execution, such that the resulting error signal helps train the sensorimotor loop ( Fig 1B ). This implies that spikes transmitted despite CD inhibition do in fact carry information to the central nervous system, even in the absence of exafference. The value of these spikes is that reafference is the error signal on which motor learning and sensor planning are based ( Fig 1B ) [ 3 , 27 , 54 , 55 ]. Elimination of reafferent feedback in the phylogenetically related auditory system, as occurs in the hearing impaired, in fact compromises voice control [ 56 ]. We argue that reafference from the lateral line likewise encodes feedback that is essential for guiding motor patterns, including the animal’s relative orientation, speed, and acceleration with respect to the flow [ 29 , 57 , 58 ].

The biophysics of lateral line sensors might limit the potential for proprioception. As a whole, lateral line sensors are highly variable, which is thought to benefit exteroception (sensing of exafference) through spatial integration and by imparting a wide sensitivity range and frequency bandwidth [ 9 , 10 , 23 , 34 ]. However, because the lateral line is a near-field sensor, reafference will typically be a greater stimulus than exafference [ 12 , 13 , 35 ]. Highly sensitive and heterogeneous sensors that are beneficial in a stationary fish thus risk becoming detrimental in a swimming fish if they transmit an overwhelming volume of feedback. The salience of both proprioceptive and exteroceptive signals will be reduced if significant computation is required to discriminate reafference from exafference and then to distinguish specific features of the reafference. Behavioral decisions would therefore be more difficult to execute while swimming, which could lead to the animal mistaking vigorous reafference for an impending attack or to it missing a real attack altogether [ 3 , 36 , 37 ]. Besides behavioral selection, hair cell systems like the lateral line exhibit high stimulus gain (up to nanometer sensitivity [ 10 , 38 ]) combined with finite vesicle populations [ 11 , 39 ]. Sustained reafference fatigues sensors at different rates [ 10 ], reducing sensitivity and distorting the timing of neural signals [ 40 – 42 ]; in the extreme, overstimulated sensory circuitry may lead to excitotoxic cell death [ 43 ]. Proprioceptive capacities of the lateral line for controlling the body as a continuum joint would therefore seem to be intrinsically limited by time-varying changes in the gain and temporal faithfulness to the reafference.

(A) (i) Fish sense water flows that originate in the environment (exafference) or are self-induced by undulatory locomotor mechanics (reafference). Both sources generate alternating currents across a wide band of frequencies, which are transduced by neuromasts along the body length. (ii) Afferent neuron signaling of neuromast deflections is mediated by receptor potentials in hair cells, as modulated by inhibitory efferent neuron activity. (iii) Depolarization of hair cells results in vesicle exocytosis onto postsynaptic afferent neuron boutons. Sustained neuromast deflection results in adaptation, here depicted as synaptic depression. (B) Hypothetical architecture of closed-loop motor control in fishes. (i) Many individual sensors in the lateral line transduce the summed exafferent and reafferent inputs. Each sensor transmits a filtered signal to the central nervous system. Brain computations extract key flow features that guide motor actions. Motor signals drive behavior, and motor signal copies called CD modulate lateral line feedback in anticipation of self-induced feedback from those behaviors. (ii) Elaboration of the sensor block to understand the effects of CD modulation on lateral line feedback. Cupula deflection drives vesicle exocytosis from hair cells, which leads to synaptic depression and adaptation of afferent neuron spike rates. Inhibition by CD reduces exocytosis and therefore afferent spike rates, but also prevents adaptation. This is replicated throughout the lateral line, so the interactions between adaptation, CD, and sensor heterogeneity are central to understanding the nature of feedback during swim bouts. CD, corollary discharge.

To fully understand the contribution of proprioceptive feedback within motor control, we must know which motor and mechanical signals are being transduced by sensors. This is an especially challenging problem for continuum joints in which shape varies continuously [ 14 ], such as that of the fish body during undulatory locomotion. While swimming, the fish’s body curvature varies from head to tail according to the phase of the motor cycle ( Fig 1A ) [ 15 ]. A fish must therefore sense and control body angles over its axial length through successive muscle layers that may span up to 10 vertebrae, rather than one or a few dedicated muscle joints [ 16 ]. In robotics, feedback from continuum joints is typically provided by sensing changes in length and curvature [ 14 ]. However, it has recently been demonstrated that stable swimming rhythms can be generated through sensing of hydrodynamic pressures on the body [ 17 ]. Fish possess few, if any, length sensors like muscle spindles [ 18 , 19 ], but can sense body curvature in the spinal cord and through stretching of the skin [ 20 – 22 ]. There is a long-standing question of whether fish can also receive proprioceptive feedback through flow sensing in the lateral line ( Fig 1 ) (reviewed by [ 12 , 13 ]). The lateral line transduces fluid motions within the boundary layer through the bending of neuromast cupulae, depolarization of hair cells, and, ultimately, the transmission to the brain of afferent fiber spike activity [ 10 , 13 , 23 – 25 ] (henceforth collectively lateral line sensors). Basic rhythmic motions like respiration are well known to be represented by spiking in the brain [ 26 , 27 ]. In swimming sea lamprey, the sister clade of jawed fishes and a comparative model of vertebrate evolution, unmodulated lateral line feedback in the brain robustly encodes periodic body motions [ 12 ]. Ablation of the sea lamprey lateral line results in increasingly distorted kinematics as swim speeds increase, pointing to a need for body awareness as motor effort increases [ 12 ]. In jawed fishes, ablation of the lateral line likewise compromises efficient navigation of turbulent flows [ 28 , 29 ], and anterior lateral line feedback is necessary for coordinating head and trunk motions to optimize sensing and respiration during forward swimming [ 30 ]. These roles highlight the specific need for the animal to be aware of its body configuration within the invisible eddies of the fluid environment [ 31 – 33 ]. Because this feedback depends on sensing body motions through coupling with the boundary layer, we refer to it as a sense of external proprioception.

Maneuvering through the environment requires an awareness of the body and its movements, a sense called proprioception. Proprioception enables smooth actions like reaching and grasping for a doorknob in the dark or rapid righting responses to perturbing effects like walking on an unstable surface [ 1 ]. Importantly, proprioceptive feedback is not simply a readout of body and limb positions and angles but a dynamic sense that depends on motor context. For example, the gain of type 1a (muscle spindle) and 1b (Golgi tendon organ) afferent fibers is modulated according to gait cycle phase [ 1 , 2 ]. The neural signals that modulate sensor gain during motor activity are called corollary discharge (CD) and are widespread across the animal kingdom [ 3 , 4 ]. CD is a signal transmitted in parallel to motor commands that modulates sensor feedback according to the predicted magnitude of self-induced feedback (reafference) [ 3 , 4 ]. Often, sensor gain is reduced because sensors respond equally to self-induced as to environmental stimuli (exafference). Consequently, vigorous body movements will be translated to a large volume of reafference that risks overwhelming the nervous system and masking both the proprioceptive signal and environmental stimuli. How CD modulates proprioception should be especially important for locomotion in fluids. The need to operate with millisecond precision [ 5 – 8 ] may be challenged by nonlinearities introduced by the fluid itself and by heterogeneous sensor gain and adaptation rates within sensor and afferent populations [ 1 , 7 , 9 – 11 ]. System variability fundamentally limits the accuracy of processing in the central nervous system [ 7 , 8 ]. We examine the role of CD in regulating sensor adaptation and heterogeneity to maximize spikes’ proprioceptive information content in an aquatic environment. We develop this role in the context of fishes’ lateral line system, the proprioceptive capacity of which has long been suspected but difficult to verify [ 12 , 13 ].

Results

A biomechanical theory of external proprioception We first consider how feedback may signal the current configuration of a body. This is a challenging problem in a continuum joint like the fish body because of the high degrees of freedom and deformations under applied loads. We note that in the absence of CD, reafference encodes periodicity of fishes’ body motions through spiking in the lateral line [12,13,26,27,35]. On the basis that lateral line ablation consistently leads to dysregulation of the mechanical body wave [12,28], sensing the body wave itself can be the proprioceptive signal. Specifically, we propose that fish detect the peak of the body wave as it passes each neuromast (Fig 2A). Peak detection within a noisy signal is fundamentally challenging, which can be exacerbated by highly heterogeneous sensors with nonlinear response characteristics. When subject to stimulation, the responses of hair cells and afferent neurons readily adapt [10,39,59,60] (Fig 1). A major cause is vesicle depletion, resulting in decreased sensitivity to stimuli and distortions of afferent spike timing [40,42]. Ostensibly, nonlinear operations in the brain might counter this effect. However, the lateral line is composed of many highly variable elements. Factors like variation in cupula length and the number and maturity of hair cells [23,34,61] underlie variation in adaptation rates even within a single neuromast [10,23,34]. Adaptation-associated distortions thus occur not just once but are repeated at every synapse between hair cells and afferent neurons (sensor heterogeneity in Fig 1). The combinatorial complexity of this variability would pose an enormous computational challenge for the brain. The capacity to learn these heterogeneous nonlinearities is moreover restricted by continuous turnover and maturation of hair cells throughout a fish’s life [61], so that the nature of reafference is unlikely to be constant. We hypothesized that CD is a key element of the lateral line circuit that modulates hair cell gain to prevent adaptation and its associated nonlinearities. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 2. Effects of undulation, adaptation, and CD on signaling of reafference in the lateral line. (A) (i) Larval zebrafish swimming is characterized by body undulations that induce periodic reafference in the lateral line. The instantaneous body angle at each neuromast (ϕ) depends on body wavelength and frequency. In the quasi-steady approximation, the body wave passing a neuromast dictates the timing of maximum cupular deflection (denoted by intense blue). (ii) The timing of the body wave peak from the first to last tail beats is shown at the locations of the most rostral (green) and most caudal (blue) neuromasts. (B) Neuromasts were stimulated by a dipole piezo stimulus to examine afferent neural responses from the first to last stimulus period. In hair cells’ sense direction, the peak of feedback is expected to coincide with maximum stimulation at phase π/2 (denoted by intense blue). (C) Polarized hair cells have a maximum signaling phase of a half stimulus cycle. (i) An idealized sensor maintains a proportional output between deflection and output throughout the stimulus bout. The response interval is therefore constant, as reported by the quantiles (q) of afferent spike phase with respect to the stimulus (ϕ). Thick vertical lines denote the median spike phase. (ii) In real sensors, we observe adaptation from the first to last stimulus intervals. We predict this leads to decreased sensitivity and thus distorted response intervals and increased VS. Peak reafference precedes π/2 and changes over time, so cannot reliably signal body wave progression. (iii) CD (ON) reduces hair cell sensitivity during motor bouts, so hair cells provide feedback only to the strongest cupular deflections coinciding with the passage of the body wave. CD thus restructures afference to reliably signal motor phase throughout the motor bout. CD, corollary discharge; VS, vector strength. https://doi.org/10.1371/journal.pbio.3001420.g002 We propose that fish determine the joint’s configuration by sensing the location of the peak of the body wave. In the quasi-steady approximation, the curvature of the body causes flow acceleration like around a bluff body. Therefore, as the peak passes each cupula, the body curvature will cause a transient elevation in flow, resulting in a maximum of cupula deflection (Fig 2Aii). Accurate feedback on peak timing between sequential neuromasts can be used to calculate parameters of the body wave, adding to the diversity of flow calculations performed in fish brains [32,62,63]. Ultimately, the computations that occur in the brain will depend on how the peripheral nervous system handles nonlinearities in the lateral line feedback. We examined 4 parameters of lateral line afferent neuron spiking to study the conditions in which lateral line feedback can provide faithful feedback on the progression of the body wave. We first examined the number of evoked spikes per stimulus, a basic measure of feedback. However, the number of spikes alone is unlikely to provide the capacity for peak detection, since it cannot discriminate between the rising edge, peak, and falling edge of the stimulus. Motor systems in fluid environments can depend on spike timing with millisecond precision [5], and the lateral line is thought to encode stimuli with similar precision [8]. We therefore examined spike timing by the location and variability of spike phases with respect to the stimulus (0 to 2π; simplified sinusoidal stimulus depicted in Fig 2B). The median spike phase (0.5 quantile, q) was used to determine spikes’ tendencies to lead the stimulus (median < π/2) or to be distributed around its peak (median = π/2). Spikes with highly variable timing convey less information about the stimulus. The response variability was examined by the width of the phase response interval, which is related to the vector strength. The response interval width was determined by the timings of early and late arriving spikes (0.1 and 0.9 q; the earliest spikes are also called the spike latency, e.g., [42]). Both the response interval width defined by the quantiles and the commonly reported vector strength are functions of the circular variance [64]. A narrow interval has high vector strength and reflects low variance in spike timing, meaning that spikes are reliably evoked at a certain phase of the stimulus (phase locked). Finally, we report the spike gain, which reflects the salience of a distinct feature of the stimulus, such as its maximum. The spike gain was calculated by the product of spike count and the vector strength [24]. This allowed us to discriminate between scenarios in which multiple evoked spikes were uniformly distributed over the stimulus cycle (low gain) and a single, highly phase-locked spike (high gain). We first consider the output of a hypothetical, idealized sensor that linearly transduces all cupular bending to an output spike rate code (Fig 2Ci). In the idealized sensor, the response interval encompasses bending in the sense direction [10,49,65]. The median phase coincides with maximum cupular deflection at π/2 because this is the peak of stimulation (stimulus phase depicted in Fig 2B). Although the spike rate encodes the peak location, the sensor’s wide response interval (low vector strength) means that individual spikes within single swim bouts occur probabilistically at any phase of bending in the sense direction (Fig 2B). Accordingly, the sensor has low gain because individual spikes within any single cycle cannot reliably signal peak deflection. To provide high-quality proprioceptive feedback, the idealized sensor should therefore also have a narrow response interval. In contrast to the idealized sensor, real hair cells exhibit substantial adaptation that induces nonlinearities in afferent feedback [10,42,60]. Hair cells’ vesicles are primed to rapidly release in response to stimulation, resulting in an initially strong and phase leading response. Subsequently, reduced vesicle availability means that a larger cupular deflection angle is required to evoke the same response. For the same stimulus, the response interval therefore becomes delayed and narrowed (Fig 2Cii). The quality of the proprioceptive feedback depends on the location of the response interval with respect to the stimulus peak. Moreover, because lateral line sensors are highly heterogeneous [10,66], the individual responses of adapted sensors will in general be unpredictable. Adaptation is an auto feedback response to evoked stimulation and therefore minimizing evoked responses by CD will likewise minimize adaptation (Fig 1B). While swimming, CD carried by cholinergic efferent neurons hyperpolarizes hair cells so that a larger cupular deflection and depolarizing current is required to achieve the same evoked spike rate (Figs 1B and 2Ciii) [35,48,49,59]. When the cupula is deflected from rest along its preferred axis, spiking will be reliably evoked only when the deflection passes the higher threshold set by CD, i.e., the hyperpolarization offset. This results in a delay of early spikes. During the falling phase of the stimulus after the peak, spiking can be evoked until the cupular deflection again passes the CD-induced threshold. As the level of hyperpolarization increases, the delay of early spikes and advance of late spikes will shrink the response interval, as depicted in Fig 2Ciii. Eventually, sensory responses will be restricted to only the strongest phase of the input stimulus and thus will occur in a narrow band (response interval) centered on π/2. The change in threshold due to CD is therefore equivalent to sharpening lateral line sensors’ polarization by rejecting subthreshold stimuli. Even though the CD may reduce the number of spikes per stimulus, their gain is high because the fish can reliably detect the location of the peak from single spikes (note that in our formulation, the gain is always equal to or less than the spike count). In principle, adaptation could have the same effect if it strongly reduces the spike count and delays and narrows the response interval. As noted above, this would be unpredictable given lateral line sensors’ heterogeneity. Importantly though, the function of CD to inhibit vesicle release will protect the vesicle pool from depletion so that it is primed to robustly respond to flow perturbations after the swim. On the other hand, as discussed below, excessive inhibition by the CD can also introduce post-swim artifacts. We therefore paid special attention to the effect of CD, stimulation, and sensor heterogeneity on spike properties through the swim bout and into the post-swim period. In sum, the location (median phase) and gain (product of spike count and vector strength) of evoked responses are key to understanding their ability to facilitate proprioceptive content. Changes in these properties over the course of the swim or due to CD will reveal how sensor nonlinearities are regulated.

Heterogeneity in adaptation and responses to corollary discharge We found considerable heterogeneity across all units at all frequencies, which prompted us to examine how heterogeneity among sensors resulted in differing responses to CD inhibition. We hypothesized that heterogeneity in adaptation rates [10] results in more complex responses to CD than revealed by average responses alone. We used cluster analysis to categorize types of lateral line responses according to the observed effects of CD on spontaneous and evoked spike rates (R S and R E , respectively, depicted in Fig 3). The use of R S provided an independent measurement of inhibition when responses were not influenced by adaptation (as outlined in Fig 1Bii). We found 3 distinct response types (Fig 6Ai). The first 2 groups exhibited consistent levels of inhibition between spontaneous and evoked spike rates (R S = R E , Fig 6Aii). In the third group, although spontaneous spike rates were strongly suppressed (low R S ), we observed only weak inhibition of evoked spike rates (high R E ; Fig 6Aii). A linear regression model indicated that variation in R E could be explained just by the interaction between R S and the level of adaptation (R S : p = 0.20, CI: −0.08 to 0.36; adaptation = Last/First response: p = 0.31, CI: −0.06 to 0.19; R S × adaptation: p < 0.001, CI: 0.33 to 0.94). Adaptation was weakest in group 1 and strongest in group 3, which respectively corresponded to high and low R E despite the same R S (Fig 6Aiii). Adaptation was intermediate in group 2. We infer that group 2 responses are composed of a mixture of weakly and strongly adapting responses, which is supported by the importance of the interaction term in the linear model. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 6. Heterogeneous adaptation rates and CD strengths result in distinct afferent neuron response types. (A) (i) The ratios (CD ON/CD OFF) of evoked and spontaneous activities (R E and R S ) revealed 3 distinct clusters of response types. (ii) R E and R S were equal in groups 1 and 2 (Δ Inhibition = 0) but diverged in group 3. (iii) The extent of adaptation increased from group 1 to group 3, shown as First/Last responses to highlight the range of adaptation. (B) Evoked activities and responses to CD differ among response types at 20 Hz. (i) Only groups 1 and 3 exhibited significant CD inhibition, as followed from the cluster analysis. (ii) Group 1 and 2 responses had wider response intervals than group 3, but all groups’ response intervals were narrowed by CD. Additionally, CD ON resulted in a delayed median phase of groups 1 and 3, compared with the median phase of CD OFF. (iii) The larger response interval widths of groups 1 and 2 resulted in lower vector strength that was more impacted by CD, compared with group 3. (iv) Gain was substantially reduced in group 1 responses reduced and constant or slightly increased in groups 2 and 3, although still approximately unity. Data and code underlying this figure and details of statistics and group comparisons may be found at DOI: 10.6084/m9.figshare.13034012. CD, corollary discharge; VS, vector strength. https://doi.org/10.1371/journal.pbio.3001420.g006 The afferent response types differed in their magnitude and phase characteristics of spiking. Groups 1 and 2 exhibited lower spike counts per stimulus compared with group 3. As followed from the clustering procedure based on the activity ratios, CD ON greatly depressed spike responses only in group 1, moderately in group 3, and very little in group 2 (Fig 6Bi). With CD OFF, groups 1 and 2 exhibited wider response intervals compared with group 3 (Fig 6Bii). With CD ON, all response intervals were narrowed, although to different extents. Additionally, the phases of early (0.1 q) and median spikes in groups 1 and 3 were delayed and substantially so in group 1 (Fig 6Bii). This resulted in increased vector strength in all groups, particularly group 1 (Fig 6Biii). Altogether, CD substantially reduced the gain of group 1, reduced group 3 to approximately unity gain and did not change, or even slightly increased, the gain of group 2 (Fig 6Biv). Thus, the heterogeneity among cells leads to classifiable differences in responses to stimulation and CD ON.

[END]

[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3001420

(C) Plos One. "Accelerating the publication of peer-reviewed science."
Licensed under Creative Commons Attribution (CC BY 4.0)
URL: https://creativecommons.org/licenses/by/4.0/


via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.