Resumen de (cómo el tálamo cambia) lo que el ojo del gato le dice al cerebro del gato
Visual information reaches the brain through the activity of thousands of neurons distributed in non-random arrays across the innermost layer of the retina (Wässle et al., 1981a). In cats, high-resolution information is first encoded by X retinal ganglion cells (RGCs) which provide the main sensory input to the primary visual cortex (V1) via a relay in the Lateral Geniculate Nucleus of the thalamus (LGN). Anatomical and physiological studies have shown that both the dendritic arbors and the receptive fields of X retinal ganglion cells form mosaics that are coordinated to approach the theoretical resolution limit of a hexagonal lattice (Wässle et al., 1981b; Eglen et al., 2005; Gauthier et al., 2009; Liu et al., 2009).In addition, it has recently been shown that these receptive field mosaics are established without the need of visual experience (Anishchenko et al., 2010). The stereotyped, optimized and hardwired nature of the retinal arrays has led to two main predictions. First, that due to an extensive pooling from neighboring photoreceptors and bipolar cells, visual acuity in the cat should match the Nyquist limit for spatial resolution induced by the tiling of the X retinal ganglion cell population. And second, that the spatial arrangement of the mosaics should set a strong constraint on the emergence of different receptive field structures, local circuits and functional properties in downstream visual areas of the brain.
The first prediction has not been confirmed experimentally. Cat visual acuity turned out to be higher than it would be expected from the density of On or Off X retinal ganglion cells alone and lower than predicted by the minimum intercone distance (Hall and Mitchell, 1991), suggesting that other factors could contribute to high-resolution visual processing.
The second prediction has received new experimental and theoretical support. It has been proposed that oriented receptive fields, columns and maps in primary visual cortex could simply emerge by random, statistical sampling of inputs from a small patch of the retinal ganglion cell mosaics (Wässle et al., 1981a, Soodak, 1987; Ringach, 2004, 2007; Paik and Ringach, 2011); even without the need to invoke any other developmental wiring rules (Miller, 1994; Swindale, 1996; Weliky and Katz, 1999). In favor of the random-wiring model, the organization of orientation columns and maps can be established without normal visual experience (Hubel and Wiesel, 1963; Crair et al., 1998; Chapman et al., 1996; Gödecke et al., 1997; White et al., 2001). In addition, it was recently found that the orientation preference of a cortical column could be predicted from the linear combination of its On and Off thalamic inputs (Jin et al., 2011), further suggesting that the structure of the cortical map can be already encoded at least at the level of the LGN.
What is the role the thalamus plays in these two phenomena? The random-wiring hypothesis assumes that thalamic afferents to the cortex faithfully reflect the spatial statistics of the retinal ganglion cell mosaics. This assumption would require that thalamic neurons get input from only one retinal afferent. However, over the years, anatomical and physiological studies have consistently shown much larger average values of retino-thalamic convergence than those supported by random-wiring models (Cleland et al., 1971a,b; Freund et al., 1985; Mastronarde, 1992; Peters and Payne, 1993; Usrey et al., 1999; Alonso et al., 2001; Yeh et al., 2009). These results suggest that, in principle, the thalamic relay could completely transform the potential influence that the spatial structure of the retinal mosaics could exert on subsequent visual processing.
In the present work we have addressed this question including in the random-wiring models imprecise, i.e. probabilistic, convergent connectivity from retina to LGN. Based on experimental data previously obtained by our laboratory, our three-step model demonstrates that the retino-thalamic projection transforms significantly the retinal input. When geniculate cells pool inputs from an average of 3 retinal ganglion cells, the convergence value that best fits our experimental data, the stereotyped retinal dipoles generate partially segregated thalamic domains of On- and Off-center cells. Importantly, this new clustering does not compromise retinotopy. Considering that relay cells in the thalamus are twice as many as retinal ganglion cells (Madarasz, Gerle et al. 1978; Stone and Campion 1978; Illing and Wassle 1981; Williams, Cavada et al. 1993), probabilistic retino-thalamic convergence leads to receptive field spatial decorrelation, improving visual space coverage at an even finer scale than in the retina. In addition, the version of the random-wiring model we propose further revealed that this thalamic clustering correlates with the spatial structure of cortical receptive fields and orientation maps. Remarkably, it also showed that imprecise convergent connectivity from retina to LGN, rather than detrimental, might actually be essential to maintain the periodicity and stability of the cortical orientation map in the presence of experimentally derived, large values of thalamocortical convergence.
In summary, our results suggest that a simple developmental rule based on statistical connectivity and wiring optimization could have the potential to profoundly shape not only information coding strategies but also the emergence of diverse receptive fields, precise local circuits and maps anywhere in the nervous system.
