Neurons in primary visual cortex (V1) exhibit orientation selectivity (OS). This is true for animals like cats or primates in which V1 has an orientation map (OM) as well as for those without such a map (salt-and-pepper organization), e.g. rodents. Whether OS is primarily due to feedforward (FF) connectivity or to recurrent interactions has not been settled. In the former case the presence or absence of an OM hardly matters, but in the latter the spatial organization of preferred orientations (PO) could affect the mechanism. The connectivity in V1 without map is hotly debated. Unclear is the extent to which connections in adult mice are stimulus feature specific. With such a specificity, the distribution of the POs of cells projecting to a neuron would be similar to that in cortices with a map and the same mechanism could operate in both cases. In contrast, when connectivity is independent of POs, this distribution is flat. Importantly, it has been demonstrated that in mice V1 neural responses are already highly selective at eye opening although neurons responding to similar visual features are not yet preferentially connected [1]. How can OS arise in this case? In a recent theoretical paper [2] we showed that strong OS emerges naturally in layer 2/3 in V1 if it operates in the balanced excitation/inhibition regime even if the recurrent connectivity is random and FF inputs from Layer 4 (L4) result from orientation selective neurons with random preferred orientations. A question then arises: How do L4 neurons exhibit OS in rodents? This is the issue we address here. We consider a model comprising two networks, one for L4 and one for the LGN. We assume first circular LGN receptive fields and purely random FF connections. We show that despite the absence of OS in LGN: 1) the FF connectivity carries a very weak information on the stimulus orientation 2) It can be extracted and amplified by the L4 network if it operates in the balanced regime. We show that the orientation selectivity index (OSI) which quantifies the degree of OS of the neurons depends only weakly on the model parameter with typical mean OSI=0.1-0.2 and 0.15-0.3 for the F0 and F1 components of the response. We also find that for realistic parameter values, the F0 component is substantially less tuned for inhibitory neurons than for excitatory neurons but for F1 components tuning is similar in both populations. Additional contributions to the OS in L4 with salt and paper organization can come from LGN neurons if they have elongated center-surround receptive fields, as recently described in experimental studies or in specificity in the LGN projections as in the Hubel & Wiesel mechanism. We show that these three contributions can be unambiguously disentangled experimentally by characterizing how the PO of the neurons varies with the spatial frequency of a drifting grating stimulus.

Work conducted in the framework of the France Israel Laboratory of Neuroscience.
[1] H. Ko, L. Cossell, C. Baragli, J. Antolik, C. Clopath, S.B. Hofer, T.D. Mrsic-Flogel Nature 496:96-100 (2013).
[2] D. Hansel and C. van Vreeswijk, J. Neurosci. 32:4049-4064 (2012).