Neural codes: Firing rates and beyond

Computational neuroscience has contributed significantly to our understanding of higher brain function by combining experimental neurobiology, psychophysics, modeling, and mathematical analysis. This article reviews recent advances in a key area: neural coding and information processing. It is shown that synapses are capable of supporting computations based on highly structured temporal codes. Such codes could provide a substrate for unambiguous representations of complex stimuli and be used to solve difficult cognitive tasks, such as the binding problem. Unsupervised learning rules could generate the circuitry required for precise temporal codes. Together, these results indicate that neural systems perform a rich repertoire of computations based on action potential timing.

Detection of synchrony in the activity of auditory nerve fibers by octopus cells of the mammalian cochlear nucleus

The anatomical and biophysical specializations of octopus cells allow them to detect the coincident firing of groups of auditory nerve fibers and to convey the precise timing of that coincidence to their targets. Octopus cells occupy a sharply defined region of the most caudal and dorsal part of the mammalian ventral cochlear nucleus. The dendrites of octopus cells cross the bundle of auditory nerve fibers just proximal to where the fibers leave the ventral and enter the dorsal cochlear nucleus, each octopus cell spanning about one-third of the tonotopic array. Octopus cells are excited by auditory nerve fibers through the activation of rapid, calcium-permeable, α-amino-3-hydroxy-5-methyl-4-isoxazole-propionate receptors. Synaptic responses are shaped by the unusual biophysical characteristics of octopus cells. Octopus cells have very low input resistances (about 7 MΩ), and short time constants (about 200 μsec) as a consequence of the activation at rest of a hyperpolarization-activated mixed-cation conductance and a low-threshold, depolarization-activated potassium conductance. The low input resistance causes rapid synaptic currents to generate rapid and small synaptic potentials. Summation of small synaptic potentials from many fibers is required to bring an octopus cell to threshold. Not only does the low input resistance make individual excitatory postsynaptic potentials brief so that they must be generated within 1 msec to sum but also the voltage-sensitive conductances of octopus cells prevent firing if the activation of auditory nerve inputs is not sufficiently synchronous and depolarization is not sufficiently rapid. In vivo in cats, octopus cells can fire rapidly and respond with exceptionally well-timed action potentials to periodic, broadband sounds such as clicks. Thus both the anatomical specializations and the biophysical specializations make octopus cells detectors of the coincident firing of their auditory nerve fiber inputs.

Linear and nonlinear pathways of spectral information transmission in the cochlear nucleus

At the level of the cochlear nucleus (CN), the auditory pathway divides into several parallel circuits, each of which provides a different representation of the acoustic signal. Here, the representation of the power spectrum of an acoustic signal is analyzed for two CN principal cells—chopper neurons of the ventral CN and type IV neurons of the dorsal CN. The analysis is based on a weighting function model that relates the discharge rate of a neuron to first- and second-order transformations of the power spectrum. In chopper neurons, the transformation of spectral level into rate is a linear (i.e., first-order) or nearly linear function. This transformation is a predominantly excitatory process involving multiple frequency components, centered in a narrow frequency range about best frequency, that usually are processed independently of each other. In contrast, type IV neurons encode spectral information linearly only near threshold. At higher stimulus levels, these neurons are strongly inhibited by spectral notches, a behavior that cannot be explained by level transformations of first- or second-order. Type IV weighting functions reveal complex excitatory and inhibitory interactions that involve frequency components spanning a wider range than that seen in choppers. These findings suggest that chopper and type IV neurons form parallel pathways of spectral information transmission that are governed by two different mechanisms. Although choppers use a predominantly linear mechanism to transmit tonotopic representations of spectra, type IV neurons use highly nonlinear processes to signal the presence of wide-band spectral features.

Cellular mechanisms for resolving phase ambiguity in the owl's inferior colliculus

Both mammals and birds use the interaural time difference (ITD) for localization of sound in the horizontal plane. They may localize either real or phantom sound sources, when the signal consists of a narrow frequency band. This ambiguity does not occur with broadband signals. A plot of impulse rates or amplitude of excitatory postsynaptic potentials against ITDs (ITD curve) consists of peaks and troughs. In the external nucleus (ICX) of the owl's inferior colliculus, ITD curves show multiple peaks when the signal is narrow-band, such as tones. Of these peaks, one occurs at ITDi, which is independent of frequency, and others at ITDi ± T, where T is the tonal period. The ITD curve of the same neuron shows a large peak (main peak) at ITDi and no or small peaks (side peaks) at ITDi ± T, when the signal is broadband. ITD curves for postsynaptic potentials indicate that ICX neurons integrate the results of binaural cross-correlation in different frequency bands. However, the difference between the main and side peaks is small. ICX neurons further enhance this difference in the process of converting membrane potentials to impulse rates. Inhibition also appears to augment the difference between the main and side peaks.

The corticofugal system for hearing: Recent progress

Peripheral auditory neurons are tuned to single frequencies of sound. In the central auditory system, excitatory (or facilitatory) and inhibitory neural interactions take place at multiple levels and produce neurons with sharp level-tolerant frequency-tuning curves, neurons tuned to parameters other than frequency, cochleotopic (frequency) maps, which are different from the peripheral cochleotopic map, and computational maps. The mechanisms to create the response properties of these neurons have been considered to be solely caused by divergent and convergent projections of neurons in the ascending auditory system. The recent research on the corticofugal (descending) auditory system, however, indicates that the corticofugal system adjusts and improves auditory signal processing by modulating neural responses and maps. The corticofugal function consists of at least the following subfunctions. (i) Egocentric selection for short-term modulation of auditory signal processing according to auditory experience. Egocentric selection, based on focused positive feedback associated with widespread lateral inhibition, is mediated by the cortical neural net working together with the corticofugal system. (ii) Reorganization for long-term modulation of the processing of behaviorally relevant auditory signals. Reorganization is based on egocentric selection working together with nonauditory systems. (iii) Gain control based on overall excitatory, facilitatory, or inhibitory corticofugal modulation. Egocentric selection can be viewed as selective gain control. (iv) Shaping (or even creation) of response properties of neurons. Filter properties of neurons in the frequency, amplitude, time, and spatial domains can be sharpened by the corticofugal system. Sharpening of tuning is one of the functions of egocentric selection.

A cellular mechanism for targeting newly synthesized mRNAs to synaptic sites on dendrites

Long-lasting forms of activity-dependent synaptic plasticity involve molecular modifications that require gene expression. Here, we describe a cellular mechanism that mediates the targeting newly synthesized gene transcripts to individual synapses where they are locally translated. The features of this mechanism have been revealed through studies of the intracellular transport and synaptic targeting of the mRNA for a recently identified immediate early gene called activity-regulated cytoskeleton-associated protein Arc. Arc is strongly induced by patterns of synaptic activity that also induce long-term potentiation, and Arc mRNA is then rapidly delivered into dendrites after episodes of neuronal activation. The newly synthesized Arc mRNA localizes selectively at synapses that recently have been activated, and the encoded protein is assembled into the synaptic junctional complex. The dynamics of trafficking of Arc mRNA reveal key features of the mechanism through which synaptic activity can both induce gene expression and target particular mRNA transcripts to the active synapses.

Synaptic regulation of protein synthesis and the fragile X protein

Protein synthesis occurs in neuronal dendrites, often near synapses. Polyribosomal aggregates often appear in dendritic spines, particularly during development. Polyribosomal aggregates in spines increase during experience-dependent synaptogenesis, e.g., in rats in a complex environment. Some protein synthesis appears to be regulated directly by synaptic activity. We use “synaptoneurosomes,” a preparation highly enriched in pinched-off, resealed presynaptic processes attached to resealed postsynaptic processes that retain normal functions of neurotransmitter release, receptor activation, and various postsynaptic responses including signaling pathways and protein synthesis. We have found that, when synaptoneurosomes are stimulated with glutamate or group I metabotropic glutamate receptor agonists such as dihydroxyphenylglycine, mRNA is rapidly taken up into polyribosomal aggregates, and labeled methionine is incorporated into protein. One of the proteins synthesized is FMRP, the protein that is reduced or absent in fragile X mental retardation syndrome. FMRP has three RNA-binding domains and reportedly binds to a significant number of mRNAs. We have found that dihydroxyphenylglycine-activated protein synthesis in synaptoneurosomes is dramatically reduced in a knockout mouse model of fragile X syndrome, which cannot produce full-length FMRP, suggesting that FMRP is involved in or required for this process. Studies of autopsy samples from patients with fragile X syndrome have indicated that dendritic spines may fail to assume a normal mature size and shape and that there are more spines per unit dendrite length in the patient samples. Similar findings on spine size and shape have come from studies of the knockout mouse. Study of the development of the somatosensory cortical region containing the barrel-like cell arrangements that process whisker information suggests that normal dendritic regression is impaired in the knockout mouse. This finding suggests that FMRP may be required for the normal processes of maturation and elimination to occur in cerebral cortical development.

Heteromeric association creates a P2Y-like adenosine receptor

Adenosine and its endogenous precursor ATP are main components of the purinergic system that modulates cellular and tissue functions via specific adenosine and ATP receptors (P1 and P2 receptors), respectively. Although adenosine inhibits excitability and ATP functions as an excitatory transmitter in the central nervous system, little is known about the ability of P1 and P2 receptors to form new functional structures such as a heteromer to control the complex purinergic cascade. Here we have shown that Gi/o protein-coupled A1 adenosine receptor (A1R) and Gq protein-coupled P2Y1 receptor (P2Y1R) coimmunoprecipitate in cotransfected HEK293T cells, suggesting the oligomeric association between distinct G protein-coupled P1 and P2 receptors. A1R and P2Y2 receptor, but not A1R and dopamine D2 receptor, also were found to coimmunoprecipitate in cotransfected cells. A1R agonist and antagonist binding to cell membranes were reduced by coexpression of A1R and P2Y1R, whereas a potent P2Y1R agonist adenosine 5′-O-(2-thiotriphosphate) (ADPβS) revealed a significant potency to A1R binding only in the cotransfected cell membranes. Moreover, the A1R/P2Y1R coexpressed cells showed an ADPβS-dependent reduction of forskolin-evoked cAMP accumulation that was sensitive to pertussis toxin and A1R antagonist, indicating that ADPβS binds A1R and inhibits adenylyl cyclase activity via Gi/o proteins. Also, a high degree of A1R and P2Y1R colocalization was demonstrated in cotransfected cells by double immunofluorescence experiments with confocal laser microscopy. These results suggest that oligomeric association of A1R with P2Y1R generates A1R with P2Y1R-like agonistic pharmacology and provides a molecular mechanism for an increased diversity of purine signaling.

GTP-binding protein βγ subunits mediate presynaptic calcium current inhibition by GABAB receptor

A variety of GTP-binding protein (G protein)-coupled receptors are expressed at the nerve terminals of central synapses and play modulatory roles in transmitter release. At the calyx of Held, a rat auditory brainstem synapse, activation of presynaptic γ-aminobutyric acid type B receptors (GABAB receptors) or metabotropic glutamate receptors inhibits presynaptic P/Q-type Ca2+ channel currents via activation of G proteins, thereby attenuating transmitter release. To identify the heterotrimeric G protein subunits involved in this presynaptic inhibition, we loaded G protein βγ subunits (Gβγ) directly into the calyceal nerve terminal through whole-cell patch pipettes. Gβγ slowed the activation of presynaptic Ca2+ currents (IpCa) and attenuated its amplitude in a manner similar to the externally applied baclofen, a GABAB receptor agonist. The effects of both Gβγ and baclofen were relieved after strong depolarization of the nerve terminal. In addition, Gβγ partially occluded the inhibitory effect of baclofen on IpCa. In contrast, guanosine 5′-O-(3-thiotriphosphate)-bound Goα loaded into the calyx had no effect. Immunocytochemical examination revealed that the subtype of G proteins Go, but not the Gi, subtype, is expressed in the calyceal nerve terminal. These results suggest that presynaptic inhibition mediated by G protein-coupled receptors occurs primarily by means of the direct interaction of Go βγ subunits with presynaptic Ca2+ channels.

Different modes of hippocampal plasticity in response to estrogen in young and aged female rats

Estrogen regulates hippocampal dendritic spine density and synapse number in an N-methyl-d-aspartate (NMDA) receptor-dependent manner, and these effects may be of particular importance in the context of age-related changes in endocrine status. We investigated estrogen's effects on axospinous synapse density and the synaptic distribution of the NMDA receptor subunit, NR1, within the context of aging. Although estrogen induced an increase in axospinous synapse density in young animals, it did not alter the synaptic representation of NR1, in that the amount of NR1 per synapse was equivalent across groups. Estrogen replacement in aged female rats failed to increase axospinous synapse density; however, estrogen up-regulated synaptic NR1 compared with aged animals with no estrogen. Therefore, the young and aged hippocampi react differently to estrogen replacement, with the aged animals unable to mount a plasticity response generating additional synapses, yet responsive to estrogen with respect to additional NMDA receptor content per synapse. These findings have important implications for estrogen replacement therapy in the context of aging.

BDNF but not NT-4 is required for normal flexion reflex plasticity and function

Neurotrophins can directly modulate the function of diverse types of central nervous system synapses. Brain-derived neurotrophic factor (BDNF) might be released by nociceptors onto spinal neurons and mediate central sensitization associated with chronic pain. We have studied the role of BDNF and neurotrophin-4 (NT-4), both ligands of the trkB tyrosine kinase receptor, in synaptic transmission and reflex plasticity in the mouse spinal cord. We used an in vitro spinal cord preparation to measure monosynaptic and polysynaptic reflexes evoked by primary afferents in BDNF- and NT-4-deficient mice. In situ hybridization studies show that both these neurotrophins are synthesized by sensory neurons, and NT-4, but not BDNF, also is expressed by spinal neurons. BDNF null mutants display selective deficits in the ventral root potential (VRP) evoked by stimulating nociceptive primary afferents whereas the non-nociceptive portion of the VRP remained unaltered. In addition, activity-dependent plasticity of the VRP evoked by repetitive (1 Hz) stimulation of nociceptive primary afferents (termed wind-up) was substantially reduced in BDNF-deficient mice. This plasticity also was reduced in a reversible manner by the protein kinase inhibitor K252a. Although the trkB ligand NT-4 is normally present, reflex properties in NT-4 null mutant mice were normal. Pharmacological studies also indicated that spinal N-methyl-d-aspartate receptor function was unaltered in BDNF-deficient mice. Using immunocytochemistry for markers of nociceptive neurons we found no evidence that their number or connectivity was substantially altered in BDNF-deficient mice. Our data therefore are consistent with a direct role for presynaptic BDNF release from sensory neurons in the modulation of pain-related neurotransmission.

Long-term plasticity in interneurons of the dentate gyrus

Single interneurons influence thousands of postsynaptic principal cells, and the control of interneuronal excitability is an important regulator of the computational properties of the hippocampus. However, the mechanisms underlying long-term alterations in the input–output functions of interneurons are not fully understood. We report a mechanism of interneuronal plasticity that leads to the functional enhancement of the gain of glutamatergic inputs in the absence of long-term potentiation of the excitatory synaptic currents. Interneurons in the dentate gyrus exhibit a characteristic, limited (≈8 mV) depolarization of their resting membrane potential after high-frequency stimulation of the perforant path. The depolarization can be observed with either whole-cell or perforated patch electrodes, and it lasts in excess of 3 h. The long-term depolarization is specific to interneurons, because granule cells do not show it. The depolarization requires the activation of Ca2+-permeable α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors and the rise of intracellular Ca2+, but not N-methyl-d-aspartate (NMDA) receptor activation. Data on the maintenance of the depolarization point to a major role for a long-term change in the rate of electrogenic Na+/K+-ATPase pump function in interneurons. As a result of the depolarization, interneurons after the tetanus respond with action potential discharges to previously subthreshold excitatory postsynaptic potentials (EPSPs), even though the EPSPs are not potentiated. These results demonstrate that the plastic nature of the interneuronal resting membrane potential underlies a unique form of long-term regulation of the gain of excitatory inputs to γ-aminobutyric acid (GABA)ergic neurons.

A genetic selection for Caenorhabditis elegans synaptic transmission mutants.

We have isolated 165 Caenorhabditis elegans mutants, representing 21 genes, that are resistant to inhibitors of cholinesterase (Ric mutants). Since mutations in 20 of the genes appear not to affect acetylcholine reception, we suggest that reduced acetylcholine release contributes to the Ric phenotype of most Ric mutants. Mutations in 15 of the genes lead to defects in a gamma-aminobutyric acid-dependent behavior; these genes are likely to encode proteins with general, rather than cholinergic-specific, roles in synaptic transmission. Ten of the genes have been cloned. Seven encode homologs of proteins that function in the synaptic vesicle cycle: two encode cholinergic-specific proteins, while five encode general presynaptic proteins. Two other Ric genes encode homologs of G-protein signaling molecules. Our assessment of synaptic function in Ric mutants, combined with the homologies of some Ric mutants to presynaptic proteins, suggests that the analysis of Ric genes will continue to yield insights into the regulation and functioning of synapses.

Differential synaptic localization of two major gamma-aminobutyric acid type A receptor alpha subunits on hippocampal pyramidal cells.

Hippocampal pyramidal cells, receiving domain specific GABAergic inputs, express up to 10 different subunits of the gamma-aminobutyric acid type A (GABAA) receptor, but only 3 different subunits are needed to form a functional pentameric channel. We have tested the hypothesis that some subunits are selectively located at subsets of GABAergic synapses. The alpha 1 subunit has been found in most GABAergic synapses on all postsynaptic domains of pyramidal cells. In contrast, the alpha 2 subunit was located only in a subset of synapses on the somata and dendrites, but in most synapses on axon initial segments innervated by axo-axonic cells. The results demonstrate that molecular specialization in the composition of postsynaptic GABAA receptor subunits parallels GABAergic cell specialization in targeting synapses to a specific domain of postsynaptic cortical neurons.

A model of multiplicative neural responses in parietal cortex.

Visual responses of neurons in parietal area 7a are modulated by a combined eye and head position signal in a multiplicative manner. Neurons with multiplicative responses can act as powerful computational elements in neural networks. In the case of parietal cortex, multiplicative gain modulation appears to play a crucial role in the transformation of object locations from retinal to body-centered coordinates. It has proven difficult to uncover single-neuron mechanisms that account for neuronal multiplication. Here we show that multiplicative responses can arise in a network model through population effects. Specifically, neurons in a recurrently connected network with excitatory connections between similarly tuned neurons and inhibitory connections between differently tuned neurons can perform a product operation on additive synaptic inputs. The results suggest that parietal responses may be based on this architecture.