Feedforward inhibition​--论文代写范文精选

PGN神经元之间的连接和集选区内，轴突和突触延迟是2.0毫秒。中间神经元和PGN细胞拥有一个复杂的混合离子电流，计算资源的高效使用和时间。下面的essay代写范文进行详述。

We have found typical lagged responses for strong feedforward inhibition with weak feedforward excitation, in agreement with Mastronarde (1987b), Humphrey & Weller (1988b), and Heggelund & Hartveit (1990). On the other hand, typical nonlagged responses are produced by weak feedforward inhibition with strong feedforward excitation; see Fig. 2. We have therefore implemented lagged and nonlagged relay cells in the model by varying the relative strengths of feedforward excitation and feedforward inhibition. For an explanation of lagged and nonlagged responses see the results below. The model system comprises 100 lagged and 100 nonlagged relay neurons. Their RF centers are 0.5 degrees in diameter (Cleland et al., 1979) and are spatially arranged in a lagged and a nonlagged clusters subtending 0.7 degrees each and displaced by 0.45 degrees; see Fig. 1b. 

This layout matches the basic structure of an on or off region in a RF of a directional simple cell in cortical layer 4B onto which the GRCs are envisaged to project (Saul & Humphrey, 1992a; Saul & Humphrey, 1992b; DeAngelis et al., 1995; Jagadeesh et al., 1997; Murthy et al., 1998). To complete the geniculate input to a RF of this type, this lagged-nonlagged unit would have to be repeated with alternating on-off polarity and a spatial offset that would determine the simple cell’s preference for some spatial frequency. Since we are not concerned here with effects of spatial frequency (see previous footnote 1), omission of the other on/off regions does not affect our conclusions. Results for rescaled RF geometries can be derived straightforwardly; see the results below. 

The number of geniculate cells contributing to the simple cell’s RF has been estimated roughly from Ahmed et al. (1994). Only its order of magnitude matters. We have also taken into account feedback inhibition via the PGN (Lo & Sherman, 1994; Sherman & Guillery, 1996); see Fig. 1a. Connections between PGN neurons and GRCs are all to all within, and separate for the lagged and nonlagged populations. Axonal plus synaptic delays are 2.0 ms in both directions. Intrageniculate interneurons and PGN cells, like GRCs, posses a complex blend of ionic currents. They are, however, thought to be active mainly in a tonic spiking mode during the awake state (Contreras et al., 1993; Pape et al., 1994). For an efficient usage of computational resources and time we have therefore modeled these neurons by the spike-response model (Gerstner & van Hemmen, 1992), which gives a reasonable approximation to tonic spiking (Kistler et al., 1997). 

Results for geniculate input to the cortex For different values of the resting membrane potential, Fig. 3 shows in the columns from left to right the velocity tuning of the lagged population (Rl), of the nonlagged population (Rnl), the peak-time differences (tnl − tl) of their responses for the preferred direction, and the tuning of the total geniculate input (R) to a cortical cell for the preferred and nonpreferred direction of motion; see above for details. As in vivo, the lagged cells prefer lower velocities and have lower peak firing rates than the nonlagged cells (Mastronarde, 1987a; Humphrey & Weller, 1988a; Saul & Humphrey, 1990). The key observation, however, is that the maximum of the total geniculate input rate to a cortical neuron shifts to lower velocities as the membrane potential hyperpolarizes; see Fig. 3 right column. 

The total geniculate input rate R assumes its maximum at a velocity of bar motion where the peak discharges of the lagged and nonlagged neurons coincide, i.e., where tnl − tl ≈ 0. The shift of the maximum with hyperpolarization to lower velocities is produced by a corresponding shift of the peak-time differences tnl − tl and of the lagged tuning Rl , while the peak of the nonlagged tuning Rnl remains essentially unchanged. As is demonstrated in Fig. 2, the change in peak-time difference is due to (i) a shift of the lagged response peak to later times, and (ii) a shift of the nonlagged response peak to earlier times. 

levels of inhibition received by lagged and nonlagged neurons. With only weak feedforward inhibition, nonlagged neurons respond to retinal input with immediate depolarization, eventually reaching the activation threshold for the Ca2+ current. If the Ca2+ current is in the de-inactivated state, it will boost depolarization and give rise to an early burst component of the visual response (Lu et al., 1992; Guido et al., 1992). 

The lower the resting membrane potential, the more de-inactivated the Ca2+ current will be, and the stronger the early response component relative to the late tonic component. In an ensemble of neurons that receive retinal input at slightly different times, like the 100 nonlagged neurons with spatially scattered RFs (cf. Fig. 1b), this leads to a gradual shift of the ensemble-response maximum with membrane polarization. Lagged neurons, on the other hand, receive 10 strong feedforward inhibition and, hence, initially respond to retinal input with hyperpolarization. Repolarization occurs when inhibition gets weaker due to declining retinal input rate or adaptation of the inhibitory input. With the Ca2+ current being de-inactivated by the excursion of the membrane potential to low values, lagged spiking starts with burst spikes as soon as the voltage reaches the Ca2+-activation threshold. 

This will take longer, if the resting membrane potential is lower, leading to the observed shift in response timing with membrane polarization. Not surprisingly, the total geniculate input rate R is higher for the direction of bar motion where tnl − tl assumes lower values. In other words, the direction preferred is the one where the lagged cells receive their retinal input before the nonlagged cells; cf. Figs. 1b and 3 right column. We have investigated the geniculate input to simple cells, which clearly cannot be compared with their output directly. Nonetheless it is interesting to note that, much like velocity tuning in areas 17 and 18 (Orban et al., 1981), the modeled geniculate input at the optimal velocity decreases and the tuning width increases with decreasing optimal velocity; see Fig. 3 right column. Because of scaling properties of the retinal ganglion cells’ velocity tuning (Cleland & Harding, 1983), rescaled versions of the RF geometry shown in Fig. 1b produce accordingly shifted tuning curves (on a logarithmic speed scale). In particular, we retrieve the positive correlation between RF size and preferred speed found in areas 17 and 18 (Orban et al., 1981) from the geniculate input.

The corticogeniculate loop 
Once a dynamic gating mechanism for thalamocortical information transfer has been identified, one has to face the key question of how gating can be controlled. The computational goal of this control should certainly be an enhanced cortical representation of behaviorally relevant stimuli, generally referred to as objects, and suppression of less significant bits, such as neuronal noise and incoherent background motion. The general idea of object segmentation by adaptive velocity tuning is illustrated in Fig. 4. Since many details of corticothalamic circuits are not known yet and since we aim at a thorough analytical treatment of the closed-loop system for general stimulus statistics we have kept the modeling at this point at a more abstract level than in the above case. Moreover, the underlying principles are best exposed by a simple model.

Model of the corticogeniculate loop 
The tuning curves shown at the top of the figure schematically represent the peak response rates of (four) cortical neurons as a function of the speed of a local feature passing their receptive field in an unadapted state (top left) and an adapted state (top right). Within some region of the visual field a natural stimulus consists of a collection of local features (depicted as dots, center) moving from left to right and from right to left at various speeds (depicted as arrows). A subset of them is moving at a common speed from left to right. The velocity density (bottom) of this type of stimulus consists of two components: one symmetrical with respect to the two directions of motion and one asymmetrical. The former derives from the incoherent, the latter from the coherent motion and is the statistical signature of a moving object. The adaptive motion system has to detect and tune in to the asymmetrical component of the velocity density. After adaptation of cortical velocity tuning (top right), object features are prominent in cortical representation, whereas other features are suppressed (center right). The given stimulus scenario generalizes straightforwardly to motion in two dimensions.

that is, an alpha function. We are interested here specifically in the slow, modulatory effect of cortical input, mediated by NMDA and metabotropic glutamate receptors in the case of depolarization, and by GABAB receptors for hyperpolarization. The rise time τ for combinations of NMDA and metabotropic glutamate receptor responses and for GABAB receptor responses may be several 100 ms (von Krosigk et al., 1999), but is kept as a free parameter in the model, i.e., we do not specify a numerical value for τ throughout the analysis. We neglect corticothalamic delays, which are at least one order of magnitude smaller than τ . In particular, for layer 6 projection neurons that are visually responsive and thus relevant to our model, they are mostly below 10 ms (Tsumoto et al., 1978; Tsumoto & Suda, 1980). One can show that the inclusion of an adequate distribution of delays does not alter the general dynamic behavior of the system, but merely adds small corrections to some characteristic quantities.

The local circuit, shown here only for one speed-tuned layer 4 neuron, is replicated for different retinal positions and two opposite directions of motion (arrows). Local circuits with identical speed tuning in layer 4 are globally interconnected by divergence in the feedback pathway. Excitatory and inhibitory inputs to the GRCs are interchanged for feedback from cortical neurons with opposite direction preferences, as indicated once for each type of feedback connection by the solid arrows. Because of the long-range connections, each GRC receives modulatory input from cortical cells jointly representing an extended visual field, with antagonism between the two direction populations. Note that, rather than by divergence in the corticogeniculate projection, the same function could alternatively be implemented by lateral intracortical connections.

51Due网站原创范文除特殊说明外一切图文著作权归51Due所有；未经51Due官方授权谢绝任何用途转载或刊发于媒体。如发生侵犯著作权现象，51Due保留一切法律追诉权。（essay代写)
更多essay代写范文欢迎访问我们主页 www.51due.com 当然有essay代写需求可以和我们24小时在线客服 QQ:800020041 联系交流。-X（essay代写)