Artificial neural networks estimate the contribution of taste neurons to coding

Taste qualities are believed to be coded in the activity of populations of taste neurons. However, it is not clear whether all neurons are equally responsible for coding. To clarify the point the relative contribution of each taste neuron to coding was assessed by constructing simple three-layer neural networks with input neurons that represent cortical taste neurons of the rat. The networks were trained by the back-propagation learning algorithm to classify the neural response patterns to the basic taste stimuli (sucrose, HCl, quinine-hydrochloride, and NaCl). The networks had four output neurons representing the basic taste qualities, the values of which provide a measure for similarity of test stimuli to the basic taste stimuli. We estimated relative contributions of input neurons to the taste discrimination of the network by examining their significance S(j), which is defined as the sum of the absolute values of the connection weights from the jth input neuron to the hidden layer. When the input neurons with a smaller S(j) (e.g., 15 out of 39 input units) were "pruned" from the trained network, the ability of the network to discriminate the basic taste qualities was not greatly affected. On the other hand, the taste discrimination of the network progressively deteriorated much more rapidly with pruning of input neurons with a larger S(j). These results suggest that cortical taste neurons differentially contribute to the coding of taste qualities. Input neurons with a larger S(j) tended to be with a larger variation of neural discharge rates to the basic taste stimuli. The variation of neural discharges may be important in the coding of taste qualities.     

Significance of distributed representation in the output layer of a neural network in a pattern recognition task

In the cerebral cortex, it is assumed that information is represented by the activity pattern of an assembly of neurons and the synaptic efficacies among them. A distributed representation of pattern is incorporated in the output layer of a neural network with an error back-propagation algorithm, in order to study its technological merits. The network has three layers, which consist of a 32×32 array of units (1024) for the input layer, 6–25 units for the hidden layer and 12 units for the output layer. 12 triangular patterns with a variety of parameters are presented to the input layer. Three output-layer units are assigned to each input figure. After initial learning, the network responds to the learned pattern with high accuracy. In addition, it responds with high accuracy to similar but unpresented patterns, showing a generalisation for patterns. The network shows resistance to unit de-activation procedures. When the input layer is exposed to the learned pattern, the hidden-layer units show associative activation pattern. These results indicate that the organisation of information representation in the output layer in a neural network strongly influences both the performance of the whole network and information representation in the hidden layer.     

Responsiveness of dog thalamic neurons to taste stimulation of various tongue regions

Receptive fields of single thalamic taste neurons in dogs were studied by examining responses to stimuli representing the four basic taste qualities applied to five different regions on the tongue surface. Twenty single thalamic units responded to taste stimulation of the tongue ipsilateral to the recording site were classified into three types according to characteristics of their receptive fields. In 11 out of 20 units, the location and the size of the receptive field and the most sensitive region in the receptive field did not vary among their effective taste stimuli (type A units). In 5 units, the location and the size of the receptive field varied with taste stimulus, although the most sensitive region was common to the stimuli (type B units). In the remaining 4 units, the most sensitive region varied with taste stimulus (type C units).     

Variability in responses and temporal coding of tastants of similar quality in the nucleus of the solitary tract of the rat

In the nucleus of the solitary tract (NTS), electrophysiological responses to taste stimuli representing four basic taste qualities (sweet, sour, salty, or bitter) can often be discriminated by spike count, although in units for which the number of spikes is variable across identical stimulus presentations, spike timing (i.e., temporal coding) can also support reliable discrimination. The present study examined the contribution of spike count and spike timing to the discrimination of stimuli that evoke the same taste quality but are of different chemical composition. Responses to between 3 and 21 repeated presentations of two pairs of quality-matched tastants were recorded from 38 single cells in the NTS of urethane-anesthetized rats. Temporal coding was assessed in 24 cells, most of which were tested with salty and sour tastants, using an information-theoretic approach. Within a given cell, responses to tastants of similar quality were generally closer in magnitude than responses to dissimilar tastants; however, tastants of similar quality often reversed their order of effectiveness across replicate sets of trials. Typically, discrimination between tastants of dissimilar qualities could be made by spike count. Responses to tastants of similar quality typically evoked more similar response magnitudes but were more frequently, and to a proportionally greater degree, distinguishable based on temporal information. Results showed that nearly every taste-responsive NTS cell has the capacity to generate temporal features in evoked spike trains that can be used to distinguish between stimuli of different qualities and chemical compositions.     

Neural network-based system for early keratoconus detection from corneal topography

Some automatic methods have been proposed to identify keratoconus from corneal maps; among these methods, neural networks have proved to be useful. However, the identification of the early cases of this ocular disease remains a problem from both a diagnostic and a screening point of view. Another problem is whether a keratoconus screening must be performed taking into account both eyes of the same subject or each eye separately; hitherto, neural networks have only been used in the second alternative. In order to examine the differences of the two screening alternatives in terms of discriminative capability, several combinations of the number of input, hidden and output nodes and of learning rates have been examined in this study. The best results have been achieved by using as input the parameters of both eyes of the same subject and as output the three categories of clinical classification (normal, keratoconus, other alterations) for each subject, a low number of neurons in the hidden layer (lower than 10) and a learning rate of 0.1. In this case a global sensitivity of 94.1% (with a keratoconus sensitivity of 100%) in the test set as well as a global specificity of 97.6% (98.6% for keratoconus alone) have been reached.     

Single neurons in the nucleus of the solitary tract respond selectively to bitter taste stimuli

Molecular data suggest that receptors for all bitter ligands are coexpressed in the same taste receptor cells (TRCs), whereas physiological results indicate that individual TRCs respond to only a subset of bitter stimuli. It is also unclear to what extent bitter-responsive neurons are stimulated by nonbitter stimuli. To explore these issues, single neuron responses were recorded from the rat nucleus of the solitary tract (NST) during whole mouth stimulation with a variety of bitter compounds: 10 microM cycloheximide, 7 mM propylthiouracil, 10 mM denatonium benzoate, and 3 mM quinine hydrochloride at intensities matched for behavioral effectiveness. Stimuli representing the remaining putative taste qualities were also tested. Particular emphasis was given to activating taste receptors in the foliate papillae innervated by the quinine-sensitive glossopharyngeal nerve. This method revealed a novel population of bitter-best (B-best) cells with foliate receptive fields and significant selectivity for bitter tastants. Across all neurons, multidimensional scaling depicted bitter stimuli as loosely clustered yet clearly distinct from nonbitter tastants. When neurons with posterior receptive fields were analyzed alone, bitter stimuli formed a tighter cluster. Nevertheless, responses to bitter stimuli were variable across B-best neurons, with cycloheximide the most, and quinine the least frequent optimal stimulus. These results indicate heterogeneity for the processing of ionic and nonionic bitter tastants, which is dependent on receptive field. Further, they suggest that neurons selective for bitter substances could contribute to taste coding.     

Behavioral support for a neural taste theory

The time required for rats to make a behavioral taste discrimination was predicted from neural discharge rates and tested using conditioned aversion. Predictions were based on the hypothesis that the responses evoked from a neural population by two different chemicals must diverge by a certain critical total number of spikes before the chemicals are discriminable. This total could be derived from the responses of all neurons in the population. Behavioral discrimination times generally supported predictions made from second-order (bulbar) neural responses, but were ambiguous concerning predictions based on fourth-order (thalamic) responses. The implications of these results for the possible functions of bulbar and thalamic taste neurons is discussed.     

Responses of neurons in the primate taste cortex to glutamate.

In order to investigate the neural encoding of glutamate in the primate, recordings were made from 190 taste responsive neurons in the primary taste cortex and adjoining orbitofrontal cortex taste area in macaques. Single neurons were found that were tuned to respond best to glutamate (umami taste), just as other cells were found with best responses to glucose (sweet), sodium chloride (salty), HCl (sour), and quinine HCl (bitter). Across the population of neurons, the responsiveness to glutamate was poorly correlated with the responsiveness to NaCl, so that the representation of glutamate was clearly different from that of NaCl. Further, the representation of glutamate was shown to be approximately as different from each of the other four tastants as they are from each other, as shown by multidimensional scaling and cluster analysis. Moreover, it was found that glutamate is approximately as well represented in terms of mean evoked neural activity and the number of cells with best responses to it as the other four stimuli, glucose, NaCl, HCl and quinine. It is concluded that in primate taste cortical areas, glutamate, which produces umami taste in humans, is approximately as well represented as are the tastes produced by: glucose (sweet), NaCl (salty), HCl (sour) and quinine HCl (sour).     

Neural representation of bitter taste in the nucleus of the solitary tract

Based on the molecular findings that many bitter taste receptors (T2Rs) are expressed within the same receptor cells, it has been proposed that bitter taste is encoded by the activation of discrete neural elements. Here we examined how a variety of bitter stimuli are represented by neural activity in central gustatory neurons. Taste responses (spikes/s) evoked by bathing the tongue and palate with intensity-matched concentrations (in M) of 2 sugars (0.32 sucrose and 0.5 D-fructose), ethanol (40%), 4 salts (0.01 NaCl, 0.008 NaNO(3), 0.01 MgCl(2), and 0.05 KCl), 2 acids (0.003 HCl and 0.005 citric acid), and 10 bitter ligands (0.007 quinine-HCl, 0.015 denatonium benzoate, 0.003 l-cysteine, 0.001 nicotine, 0.005 strychnine-HCl, 0.04 tetraethylammonium chloride, 0.03 atropine-SO(4), 0.005 brucine-SO(4), 0.03 papaverine-HCl, and 0.009 sparteine) were recorded from 51 neurons in the nucleus of the solitary tract of anesthetized rats. Cluster analysis was used to categorize neurons into types based on responses to sucrose, NaCl, HCl, and quinine-HCl. Three groupings emerged: type S (responded optimally to sweets), type N (sodium-optimal), and type H/Q (responded robustly to bitters, acids, and salts). Multivariate analyses revealed that across-neuron patterns of response among bitter stimuli were strongly correlated. However, neural type H/Q, which was most responsive to bitter tastants, was not differentially sensitive to bitter stimuli and Na(+) salts, which rats perceive as distinct. Thus central neurons most responsive to bitter substances receive significant input from receptors that mediate other tastes, indicating that bitter stimuli are not represented by activity in specifically tuned neurons.     

Local connections to specific types of layer 6 neurons in the rat visual cortex

Because layer 6 of the cerebral cortex receives direct thalamic input and provides projections back to the thalamus, it is in a unique position to influence thalamocortical interactions. Different types of layer 6 pyramidal neurons provide output to different thalamic nuclei, and it is therefore of interest to understand the sources of functional input to these neurons. We studied the morphologies and local excitatory input to individual layer 6 neurons in rat visual cortex by combining intracellular labeling and recording with laser-scanning photostimulation. As in previous photostimulation studies, we found significant differences in the sources of local excitatory input to different cell types. Most notably, there were differences in local input to neurons that, based on analogy to barrel cortex, are likely to project only to the lateral geniculate nucleus of the thalamus versus those that are likely to also project to the lateral posterior nucleus. The more striking finding, however, was the paucity of superficial layer input to layer 6 neurons in the rat visual cortex, contrasting sharply with layer 6 neurons in the primate visual cortex. These observations provide insight into differences in function between cortical projections to first-order versus higher-order thalamic nuclei and also show that these circuits can be organized differently in different species.     

Distributed population mechanism for the 3-D oculomotor reference frame transformation

Human saccades require a nonlinear, eye orientation-dependent reference frame transformation to transform visual codes to the motor commands for eye muscles. Primate neurophysiology suggests that this transformation is performed between the superior colliculus and brain stem burst neurons, but provides little clues as to how this is done. To understand how the brain might accomplish this, we trained a 3-layer neural net to generate accurate commands for kinematically correct 3-D saccades. The inputs to the network were a 2-D, eye-centered, topographic map of Gaussian visual receptive fields and an efference copy of eye position in 6-dimensional, push-pull "neural integrator" coordinates. The output was an eye orientation displacement command in similar coordinates appropriate to drive brain stem burst neurons. The network learned to generate accurate, kinematically correct saccades, including the eye orientation-dependent tilts in saccade motor error commands required to match saccade trajectories to their visual input. Our analysis showed that the hidden units developed complex, eye-centered visual receptive fields, widely distributed fixed-vector motor commands, and "gain field"-like eye position sensitivities. The latter evoked subtle adjustments in the relative motor contributions of each hidden unit, thereby rotating the population motor vector into the correct correspondence with the visual target input for each eye orientation: a distributed population mechanism for the visuomotor reference frame transformation. These findings were robust; there was little variation across networks with between 9 and 49 hidden units. Because essentially the same observations have been reported in the visuomotor transformations of the real oculomotor system, as well as other visuomotor systems (although interpreted elsewhere in terms of other models) we suggest that the mechanism for visuomotor reference frame transformations identified here is the same solution used in the real brain.     

Bitter-responsive brainstem neurons: Characteristics and functions

The sensation that humans describe as “bitter” is evoked by a large group of chemically diverse ligands. Bitter stimuli are avoided by a range of species and elicit reflex rejection, behaviors considered adaptations to the toxicity of many of these compounds. We review novel evidence for neurons that are narrowly tuned to bitter ligands at the initial stages of central processing. These “B-best” neurons in the nucleus of the solitary tract (NST) and parabrachial nucleus (PBN) respond to multiple types of bitter stimuli and exhibit average responses to bitter tastants that are 6-8 times larger than to moderate concentrations of compounds representing other qualities. However, in the PBN B-best units are appreciably activated by intense salt and acid. Neurons broadly sensitive to salts and acids (“AN” neurons) also responded to bitter stimuli. This sensitivity appeared restricted to stronger intensities of ionic bitters, as cycloheximide remained ineffective across concentrations. In addition to chemosensitive profile, B-best neurons were also distinctive with regard to their posterior receptive fields, long latencies, slow firing rates and projection status. Compared to B-best NST cells, those in the PBN received increased convergence from anterior and posterior receptive fields and responded to a greater number of bitter stimuli. We conclude that B-best neurons likely contribute to pathways underlying gaping, aversive hedonic quality and taste coding. The differential responsiveness of B-best and AN neurons to ionic and nonionic bitter ligands also suggests a potential substrate for discrimination within this quality.     

Gustatory coding in the precentral extension of area 3 in Japanese macaque monkeys; comparison with area G

The precentral extension of area 3 as well as the transition between the frontal operculum and insula (area G) comprises the primary gustatory cortex in the subhuman primate, receiving projections from the thalamic taste relay. However, in contrast to the extensive studies that have been carried out on the latter area, only a few taste units in the former area have been recorded. To clarify gustatory coding in area 3, we investigated the taste response properties of neurons in area 3 compared with those in area G in alert monkeys by infiltrating into their mouths seven taste stimuli [0.3 M sucrose (S), 0.1 M NaCl (N), 0.01 N HCl (H), 0.003 M quinine-HCl (Q), 0.1 M monosodium glutamate (MSG), distilled water (W), and orange juice (OR)] and artificial saliva (SA). A larger number of HCl-best units and a smaller number of quinine-best units were found in area 3 than in area G. The onset latency and response duration were significantly shorter in area 3 than in area G. Weighted multi-dimensional scaling showed that area G divided eight stimulants into four classes, i.e. two groups (H-Q-W and S-MSG-OR), N and SA, whereas area 3 divided them into three classes (N-MSG-W-OR, S-Q, and H-SA). This suggested that tastants not separated in area G were separated in area 3, and vice versa. This indicates that both areas complement each other in the representation of taste stimuli, each contributing to taste information processing in a different manner.     

Neural coding mechanisms for flow rate in taste-responsive cells in the nucleus of the solitary tract of the rat

When a taste stimulus enters the mouth, intentional movement of the stimulus within the oropharyngeal cavity affects the rate at which taste receptors are exposed to the stimulus and may ultimately affect taste perception. Early studies have shown that stimulus flow rate, the experimental equivalent of the effects of these investigative movements, modulates the portion of the peripheral nerve response that occurs when behavioral assessments of tastants are made. The present experiment studied the neural coding mechanisms for flow rate in the nucleus of the solitary tract (NTS), the first central relay in the taste pathway. Responses to NaCl (0.1 M) presented at high (5 ml/s) and low (3 ml/s) flow rates, sucrose (0.5 M), quinine HCl (0.01 M), and HCl (0.01 M) were recorded extracellularly from single NTS units in multiple replications. Information conveyed by evoked responses was analyzed with a family of metrics that quantify the similarity of two spike trains in terms of spike count and spike timing. Information about flow rate was conveyed by spike timing and spike count in approximately equal proportions of units (each approximately 1/3), whereas information about taste quality was conveyed by spike timing in about half of the units. Different subsets of units contributed information for discrimination of flow rate and taste quality.     

Effects of acetylcholine on coding of taste information in the primary gustatory cortex in rats

Acetylcholine (ACh) receptors are widely distributed throughout the cerebral cortex in rats. Recently, cholinergic innervation of the gustatory cortex (GC) was reported to be involved in certain taste learning in rats. Here, the effects of iontophoretic application of ACh on the response properties of GC neurons were studied in urethane-anesthetized rats. ACh affected spontaneous discharges in a small fraction of taste neurons (11 of 86 neurons tested), but influenced taste responses in 27 of 43 neurons tested. No correlations with ACh susceptibility were noted for spontaneous discharges and taste responses. Among the 27 neurons, ACh facilitated taste responses in 13, inhibited taste responses in 13 and either facilitated or inhibited taste responses depending on the stimuli in 1. Furthermore, ACh affected the responses to best stimuli that produced the largest responses among four basic tastants (best responses) in 7 of 27 taste neurons, to non-best responses in 9, and to both best and non-best responses in 11. ACh mostly inhibited the best responses (13 of 18 neurons). Thus, ACh often decreased the response selectivity to the four basic tastants and changed the response profile. Atropine, a general antagonist of muscarinic receptors, antagonized ACh actions on taste responses or displayed the opposite effects on taste responses to ACh actions in two-thirds of the neurons tested. These findings indicate that ACh mostly modulates taste responses through muscarinic receptors, and suggest that ACh shifts the state of the neuron network in the GC, in terms of the response selectivities and response profiles.     

Gustatory responses of single neurons in the caudolateral orbitofrontal cortex of the macaque monkey

1. In recordings made from 3,120 single neurons, a secondary cortical taste area was found in the caudolateral part of the orbitofrontal cortex of the cynomolgus macaque monkey, Macaca fascicularis. The area is part of the dysgranular field of the orbitofrontal cortex and is situated anterior to the primary cortical taste areas in the frontal opercular and adjoining insular cortices. 2. The responses of 49 single neurons with gustatory responses in the caudolateral orbitofrontal taste cortex were analyzed using the taste stimuli glucose, NaCl, HCl, quinine HCl, water, and blackcurrant juice. 3. A breadth-of-tuning coefficient was calculated for each neuron. This is a metric that can range from 0.0 for a neuron that responds specifically to only one of the four basic taste stimuli to 1.0 for one that responds equally to all four stimuli. The mean coefficient for 49 cells in the caudolateral orbitofrontal cortex was 0.39. This tuning is much sharper than that of neurons in the nucleus of the solitary tract of the monkey, and sharper than that of neurons in the primary frontal opercular and insular taste cortices. 4. A cluster analysis showed that at least seven different groups of neurons were present. For each of the taste stimuli glucose, blackcurrant juice, NaCl, and water, there was one group of neurons that responded much more to that tastant than to the other tastants. The other groups of neurons responded to two or more of these tastants, such as glucose and blackcurrant juice. In this particular region neurons were not found with large responses to HCl or quinine HCl, although such neurons could be present in other parts of the orbitofrontal cortex. 5. On the basis of this and other evidence it is concluded that in the caudolateral orbitofrontal cortex there is a secondary cortical taste area in which the tuning of neurons has become finer than in early areas of taste processing, in which foods, water, and NaCl are strongly represented and where motivation dependence first becomes manifest in the taste system.     

Arousal-related associative response characteristics of dorsal lateral geniculate nucleus neurons during acoustic Pavlovian fear conditioning

This research determined whether fear-conditioned, acoustic stimuli induce thalamic arousal reflected in associative responses in dorsal lateral geniculate nucleus (dLGN) neurons. Rabbits received a Pavlovian discriminative fear conditioning procedure in which one tone conditioned stimulus (CS +) was always paired with an aversive unconditioned stimulus (US) and another tone (CS-) was never paired with the US. Responses of single dLGN neurons to random CS+ and CS- presentations were then recorded. Nine of 15 recorded neurons demonstrated significantly greater firing during the CS+ versus the CS-. Their spontaneous activity demonstrated tonic firing during increased neocortical arousal and burst firing during decreased neocortical arousal. The results demonstrate that dLGN neurons show associative responses to fear-conditioned, acoustic stimuli and present a model for investigating the neural circuits by which such stimuli affect sensory processing at the thalamic level.     

Synaptic physiology of the flow of information in the cat's visual cortex in vivo

Each stage of the striate cortical circuit extracts novel information about the visual environment. We asked if this analytic process reflected laminar variations in synaptic physiology by making whole-cell recording with dye-filled electrodes from the cat's visual cortex and thalamus; the stimuli were flashed spots. Thalamic afferents terminate in layer 4, which contains two types of cell, simple and complex, distinguished by the spatial structure of the receptive field. Previously, we had found that the postsynaptic and spike responses of simple cells reliably followed the time course of flash-evoked thalamic activity. Here we report that complex cells in layer 4 (or cells intermediate between simple and complex) similarly reprised thalamic activity (response/trial, 99 ± 1.9 %; response duration 159 ± 57 ms; latency 25 ± 4 ms; average ± standard deviation;   n = 7  ). Thus, all cells in layer 4 share a common synaptic physiology that allows secure integration of thalamic input. By contrast, at the second cortical stage (layer 2+3), where layer 4 directs its output, postsynaptic responses did not track simple patterns of antecedent activity. Typical responses to the static stimulus were intermittent and brief (response/trial, 31 ± 40 %; response duration 72 ± 60 ms, latency 39 ± 7 ms;   n = 11  ). Only richer stimuli like those including motion evoked reliable responses. All told, the second level of cortical processing differs markedly from the first. At that later stage, ascending information seems strongly gated by connections between cortical neurons. Inputs must be combined in newly specified patterns to influence intracortical stages of processing.     

An approach to estimate EEG power spectrum as an index of heat stress using backpropagation artificial neural network

A method has been presented for an effective application of backpropagation artificial neural network (ANN) in establishment of electro-encephalogram (EEG) power spectra as an index of stress in hot environment. The power spectrum data for slow wave sleep (SWS), rapid eye movement (REM) sleep and awake (AWA) states in three groups of rats (acute heat stress, chronic heat stress and the normal) were tested by an ANN, containing 60 nodes in input layer, weighted from power spectrum data from 0 to 30 Hz, 18 nodes in hidden layer and an output node. The target output values for this network were determined with another five-layered neural network (with the structure of 3-12-1-12-3). The input and output of this network was assigned with the three well-established heat stress indices (body temperature, body weight and plasma corticosterone). The most important feature for acute stress, chronic stress and normal conditions were extracted from the third layer single neuron and used for the target value for the three-layered neural network. The ANN was found effective in recognising the EEG power spectra with an average of 96.67% for acute heat stress, 97.17% for chronic heat stress and 98.5% for normal subjects.     

Adaptation in thalamic barreloid and cortical barrel neurons to periodic whisker deflections varying in frequency and velocity

Layer IV circuitry in the rodent whisker-to-barrel pathway transforms the thalamic input signal spatially and temporally. Excitatory and inhibitory barrel neurons display response properties that differ from each other and from their common thalamic inputs. Here we further examine thalamocortical response transformations by characterizing the responses of individual thalamic barreloid neurons and presumed excitatory and inhibitory cortical barrel neurons to periodic whisker deflections varying in frequency from 1 to 40 Hz. Both pulsatile and sinusoidal periodic stimulation of fixed deflection amplitude were used to assess stimulus-evoked adaptation of thalamocortical units (TCUs), fast-spike barrel units (FSUs: presumed inhibitory neurons), and regular-spike barrel units (RSUs: presumed excitatory neurons). Monotonic, frequency-dependent reductions in firing were observed in thalamic and cortical neurons to the second and subsequent stimuli in trains of high (pulsatile)- and low (sinusoidal)-velocity deflections. RSUs and FSUs adapted substantially more than their thalamic input neurons, and at all frequencies, FSUs fired at higher rates than the other two cell types. For example at 40 Hz, response magnitudes of TCUs decreased by 34%, FSUs by 72%, and RSUs by 78%. Across frequencies, RSUs and FSUs displayed more cycle-by-cycle entrainment and phase-locked responses for (high velocity) pulsatile than (lower velocity) sinusoidal deflections; for TCUs, phase-locking was equivalent for both stimuli, but entrainment was higher for sinusoidal deflections. Strong feed-forward inhibition, in conjunction with synaptic depression, renders the firing of barrel neurons sparse but temporally faithful to the occurrence of repetitive whisker deflections, especially when they are of high velocity.