Presynaptic release probability versus postsynaptic efficacy as the carrier of simple-cell receptive-field structure in visual cortex: an untested experimental discrimination and its consequences for spatiotemporal dynamics and coding under short-term synaptic depression

A 1998 modeling study (Artun, Shouval and Cooper) showed that a simple-cell receptive field (RF) in primary visual cortex, built from thalamocortical synapses exhibiting short-term depression, behaves very differently over time depending on whether the RF's orientation-selective structure is functionally encoded in the presynaptic probability of release (PR model) or in the postsynaptic efficacy (SE model): under flashed bars the SE model produces uniformly scaling unimodal tuning curves while the PR model produces time-varying, eventually bimodal tuning, and in orientation-domain reverse correlation the PR model predicts kernel zero-crossings that vary with orientation while the SE model predicts orientation-invariant zero-crossings. The authors proposed the zero-crossing criterion as an experimental test and asked which mechanism operates in vivo, noting that input-layer data available at the time favored the SE model. Later literature measured thalamocortical depression in vivo and concluded that depression contributes minimally to orientation tuning, and reattributed orientation-tuning dynamics largely to cortical suppression; however, the central discrimination — whether thalamocortical RF structure is presynaptic or postsynaptic — has never been directly tested, and the consequences of realistic depression for temporal coding of stimuli matched in mean rate (orientation/contrast) and for population codes remain unresolved.

Background

In the primary visual cortex (V1) of mammals, 'simple cells' have receptive fields (RFs) — the map of how visual stimulation at each retinal position excites or suppresses the neuron — organized into oriented ON and OFF subregions, so that the cell fires selectively for bars or edges of a particular orientation ('orientation selectivity'). A long-standing feedforward account holds that this selectivity arises from the spatial arrangement of synaptic weights of inputs relayed from the lateral geniculate nucleus (LGN) of the thalamus through 'thalamocortical' synapses onto a cortical cell. Short-term synaptic depression — the frequency-dependent weakening of synaptic transmission during repeated or sustained activation, with recovery over hundreds of milliseconds — was measured in neocortex in vitro by Markram and Tsodyks (1996) and Abbott, Varela, Sen and Nelson (1997). The standard phenomenological description (Tsodyks and Markram, 1997) characterizes each synapse by a pool of neurotransmitter 'resources' that are consumed on each presynaptic action potential and recover with a time constant of roughly 600–1000 ms; the steady-state postsynaptic current is then approximately inversely proportional to stimulation frequency. A synapse is described by its presynaptic probability of release, $p_r$ (the fraction of resources used per presynaptic spike), and its postsynaptic efficacy, $e$ (the current generated per unit of released resource).

Artun, Shouval and Cooper (1998) asked what such 'dynamic synapses' do to the spatiotemporal RF of a V1 simple cell. They modeled a single leaky integrate-and-fire neuron (a standard point-neuron model that integrates synaptic current and fires when the membrane potential crosses a threshold) receiving Poisson spike trains from ON- and OFF-center LGN cells (relay cells whose receptive-field centers respond to luminance increments and decrements, respectively), through depressing synapses, with the simple-cell RF created either by spatial modulation of $p_r$ at constant $e$ (the PR model) or by spatial modulation of $e$ at constant $p_r$ (the SE model). These two implementations correspond to distinct cellular loci for the RF structure — presynaptic release machinery versus postsynaptic receptor content — a question tied to the contemporary debate about whether long-term potentiation acts pre- or postsynaptically. Because depression is driven by presynaptic firing rate and by $p_r$, the PR model's preferred orientation depresses fastest: flashed-bar responses at the preferred orientation decay more quickly than at nonpreferred orientations, producing tuning curves that change shape over time and can become bimodal (two sideband peaks) hundreds of milliseconds after stimulus onset, whereas the SE model yields tuning curves that scale uniformly in time and remain unimodal. Moving gratings, a standard orientation stimulus, do not distinguish the two models.

As a sharper experimental discriminator, Artun et al. turned to reverse correlation in the orientation domain, the method Ringach, Hawken and Shapley (1997) used in macaque V1: a rapid random sequence of oriented grating stimuli is presented, and the spike train is cross-correlated with the stimulus history to estimate a first-order Wiener kernel — the time-dependent transfer function from stimulus orientation to response. In the model, the PR model yields a kernel whose zero-crossings (the times at which the kernel value passes through zero) vary with the orientation at which the kernel is probed, while the SE model yields orientation-invariant zero-crossings; the authors noted this criterion survives random fluctuations in $p_r$ and is valid even if both $e$ and $p_r$ are structured. Comparing with the dynamics reported by Ringach et al. (1997) — input-layer (layer 4C) neurons showed a single stable orientation preference, while output-layer neurons often showed time-varying, multi-peaked tuning — Artun et al. suggested that input-layer properties, for which their thalamocortical model is most appropriate, agree better with the SE model, 'but further experiments are required to settle this point.' The study also noted two methodological and biological caveats: the orientation-domain kernels were extracted from stimuli that changed only every 20 ms (so the whiteness, i.e. temporal-decorrelation, assumption is violated below that scale and faster dynamics are inaccessible), and the depression parameters were taken from in vitro measurements on the assumption that they are not substantially altered in vivo. Finally, the model illustrated a coding consequence: two stimuli that produce the same mean synaptic conductance over ~150 ms (a high-contrast bar at a nonpreferred orientation and a low-contrast bar at the preferred orientation) can be distinguished by the time course of the conductance — a temporal code generated purely by synaptic depression — which a pure firing-rate code, single-neuron or ensemble (population), could not resolve.

Subsequent work transformed parts of this landscape. In vivo measurements established that thalamocortical synapses do depress in vivo (Chung, Li and Nelson, 2002, in somatosensory cortex; Boudreau and Ferster, 2005, in cat V1, who also found the synapses tonically depressed by spontaneous activity, constraining depression's functional role; Sedigh-Sarvestani, Palmer and Contreras, 2019, from paired in vivo LGN–V1 recordings, found weak, unreliable, variable synapses with short-term plasticity). Modeling with in-vivo-fit parameters concluded that thalamocortical depression does not appreciably contribute to the orientation tuning of simple cells, acting mainly to saturate the contrast-response curve (Cimenser and Miller, 2014; similarly Fortier, 2017), while depression was instead invoked to explain cross-orientation suppression (the reduction of a neuron's response to an optimally oriented stimulus by a superimposed, orthogonally oriented stimulus) and contrast saturation (the leveling-off of responses as stimulus contrast increases) (Carandini, Heeger and Senn, 2002; Freeman, Durand, Kiper and Carandini, 2002). The dynamics of orientation tuning were reinterpreted as dominated by cortical mechanisms — untuned and tuned suppression, largely inhibitory and recurrent — rather than thalamocortical depression (Ringach, Hawken and Shapley, 2003; Shapley, Hawken and Ringach, 2003; Xing, Shapley, Hawken and Ringach, 2005; Schummers et al., 2007), and high-speed reverse correlation in awake macaques found orientation tuning largely stable over the response (Mazer et al., 2002). What no later study has done, however, is directly test the original discrimination — whether the orientation-selective structure of thalamocortical input to a simple cell is carried by structured presynaptic release probability or by structured postsynaptic efficacy — nor settle the coding question in the specific form posed (which stimuli matched in mean rate are distinguishable through depression-generated temporal structure, in single cells or populations, once realistic in vivo parameters and recurrent cortical circuitry are included).

Problem Statement

Determine which synaptic parameter carries the orientation-selective structure of the thalamocortical input to V1 simple cells: the presynaptic probability of release ($p_r$, the fraction of synaptic resources consumed per presynaptic action potential; the PR model) or the postsynaptic efficacy ($e$, the postsynaptic current generated per unit of released resource; the SE model). Whether the orientation-selective structure is carried presynaptically or postsynaptically has not been established in vivo; the authors' own proposed zero-crossing criterion and the later in vivo measurements of thalamocortical depression both bear on the discrimination without settling it. The two models make distinguishing predictions: under flashed-bar stimulation the SE model's tuning scales uniformly and remains unimodal, whereas the PR model's preferred orientation depresses fastest, so its tuning changes shape over time and can become bimodal; and in orientation-domain reverse correlation the PR model's kernel zero-crossings (the times at which the kernel passes through zero) vary with stimulus orientation (indicating structured $p_r$) while the SE model's are orientation-invariant (indicating structured $e$). That zero-crossing difference holds even if both $e$ and $p_r$ are structured and is robust to random fluctuations in $p_r$, provided the stimuli are decorrelated in time on a timescale fine enough to resolve the dynamics of interest. Noting that properties of input-layer neurons — for which their feedforward thalamocortical model is most appropriate — agree better with the SE model, the authors concluded that 'further experiments are required to settle this point.'

The same model leaves a second question open: which stimuli that are matched in mean rate — for example, stimuli differing in orientation and contrast — are distinguishable through the temporal structure generated by depression, once realistic in vivo depression parameters and recurrent cortical circuitry are included, and what such a distinction implies for a single-neuron or population code that a purely rate-based code could not resolve.

Two auxiliary issues raised in the source paper — whether in vitro depression parameters hold in vivo, and the 20 ms temporal-resolution limit of the original reverse-correlation protocol — have been addressed by later work and are settled background rather than targets.

References

1. Ömer B. Artun, Harel Z. Shouval, Leon N. Cooper (1998), "The effect of dynamic synapses on spatiotemporal receptive fields in visual cortex", Proceedings of the National Academy of Sciences USA 95(20):11999–12003. DOI: 10.1073/pnas.95.20.11999. https://doi.org/10.1073/pnas.95.20.11999 (open-access full text: https://europepmc.org/article/PMC/21754)
2. D. L. Ringach, M. J. Hawken, R. Shapley (1997), "Dynamics of orientation tuning in macaque primary visual cortex", Nature 387(6630):281–284. DOI: 10.1038/387281a0. https://doi.org/10.1038/387281a0
3. H. Markram, M. Tsodyks (1996), "Redistribution of synaptic efficacy between neocortical pyramidal neurons", Nature 382(6594):807–810. DOI: 10.1038/382807a0. https://doi.org/10.1038/382807a0
4. M. V. Tsodyks, H. Markram (1997), "The neural code between neocortical pyramidal neurons depends on neurotransmitter release probability", Proceedings of the National Academy of Sciences USA 94(2):719–723. DOI: 10.1073/pnas.94.2.719. https://doi.org/10.1073/pnas.94.2.719
5. L. F. Abbott, J. A. Varela, K. Sen, S. B. Nelson (1997), "Synaptic depression and cortical gain control", Science 275(5297):220–224. DOI: 10.1126/science.275.5297.221. https://doi.org/10.1126/science.275.5297.221
6. S. Chung, X. Li, S. B. Nelson (2002), "Short-term depression at thalamocortical synapses contributes to rapid adaptation of cortical sensory responses in vivo", Neuron 34(3):437–446. DOI: 10.1016/s0896-6273(02)00659-1. https://doi.org/10.1016/S0896-6273(02)00659-1
7. T. C. Freeman, F. Durand, D. C. Kiper, M. Carandini (2002), "Suppression without inhibition in visual cortex", Neuron 35(4):759–771. DOI: 10.1016/s0896-6273(02)00819-x. https://doi.org/10.1016/S0896-6273(02)00819-X
8. M. Carandini, D. J. Heeger, W. Senn (2002), "A synaptic explanation of suppression in visual cortex", Journal of Neuroscience 22(22):10053–10065. DOI: 10.1523/JNEUROSCI.22-22-10053.2002. https://doi.org/10.1523/JNEUROSCI.22-22-10053.2002
9. J. A. Mazer, W. E. Vinje, J. McDermott, P. H. Schiller, J. L. Gallant (2002), "Spatial frequency and orientation tuning dynamics in area V1", Proceedings of the National Academy of Sciences USA 99(3):1645–1650. DOI: 10.1073/pnas.022638499. https://doi.org/10.1073/pnas.022638499
10. D. L. Ringach, M. J. Hawken, R. Shapley (2003), "Dynamics of orientation tuning in macaque V1: the role of global and tuned suppression", Journal of Neurophysiology 90(1):342–352. DOI: 10.1152/jn.01018.2002. https://doi.org/10.1152/jn.01018.2002
11. R. Shapley, M. Hawken, D. L. Ringach (2003), "Dynamics of orientation selectivity in the primary visual cortex and the importance of cortical inhibition", Neuron 38(5):689–699. DOI: 10.1016/s0896-6273(03)00332-5. https://doi.org/10.1016/S0896-6273(03)00332-5
12. C. E. Boudreau, D. Ferster (2005), "Short-term depression in thalamocortical synapses of cat primary visual cortex", Journal of Neuroscience 25(31):7179–7190. DOI: 10.1523/JNEUROSCI.1445-05.2005. https://doi.org/10.1523/JNEUROSCI.1445-05.2005
13. D. Xing, R. M. Shapley, M. J. Hawken, D. L. Ringach (2005), "Effect of stimulus size on the dynamics of orientation selectivity in Macaque V1", Journal of Neurophysiology 94(1):799–812. DOI: 10.1152/jn.01139.2004. https://doi.org/10.1152/jn.01139.2004
14. J. Schummers, B. Cronin, K. Wimmer, M. Stimberg, R. Martin, K. Obermayer, K. Koerding, M. Sur (2007), "Dynamics of orientation tuning in cat V1 neurons depend on location within layers and orientation maps", Frontiers in Neuroscience 1(1):145–159. DOI: 10.3389/neuro.01.1.1.011.2007. https://doi.org/10.3389/neuro.01.1.1.011.2007
15. A. Cimenser, K. D. Miller (2014), "The effects of short-term synaptic depression at thalamocortical synapses on orientation tuning in cat V1", PLoS ONE 9(8):e106046. DOI: 10.1371/journal.pone.0106046. https://doi.org/10.1371/journal.pone.0106046
16. P. A. Fortier (2017), "Comparison of mechanisms for contrast-invariance of orientation selectivity in simple cells", Neuroscience 348:41–62. DOI: 10.1016/j.neuroscience.2017.01.052. https://doi.org/10.1016/j.neuroscience.2017.01.052
17. M. Sedigh-Sarvestani, L. A. Palmer, D. Contreras (2019), "Thalamocortical synapses in the cat visual system in vivo are weak and unreliable", eLife 8:e41925. DOI: 10.7554/eLife.41925. https://doi.org/10.7554/eLife.41925
18. D. Ferster, S. Chung, H. Wheat (1996), "Orientation selectivity of thalamic input to simple cells of cat visual cortex", Nature 380(6571):249–252. DOI: 10.1038/380249a0. https://doi.org/10.1038/380249a0