Introduction

Motion detection is essential for many animals for survival, and studies show that many cell types of the retina  \citep{Barlow:1964we, Olveczky:2007cy, Vaney:2012jl} and a considerable portion of the cerebral cortex  \citep{Burr:2011gc, Nishida:2011ji} play a role in related functions. However, those studies have mainly focused on artificial stimuli detection, such as random dots, lines or moving gratings, but ignored but ignored naturalistic stimuli in the environment in whuch animals have evolved and to which their visual system would be attuned to \citep{Simoncelli:2001dn,Geisler:2008gu,Webster:1997tn,Dyakova:2015dy,Yu:2005jh,OLSHAUSEN:2004fw}. Moreover, computational models based on the response to simple, artificial stimuli fail to predict the response to naturalistic images  \citep{Wu:2006gs,David:2004gl,Carandini:2005kw}. Nevertheless, working with natural images directly is not always appropiate due to their complexity regarding critical parameters of signals processing, including visual content and its variability, what makes it hard to control, such as the motion information in a sequence of images  \citep{Rust:2005bq}.
In the visual cortex, motion is detected at different levels and by populations of cells attuned to different parameters of the stimulus (see \cite{Troy:2002tq} for a review). It has been reported that visual neurons stimulated with natural images change their tuning curves compared to the response obtained with artificial stimulus, becoming narrower and thus more selective, e.g. to speed \citep{Priebe:2006jk} compared to the response obtained with artificial stimulus. In turn, they generate sparser code \citep{Vinje:2000fk,Vinje:2002vg,Haider:2010eo} and eye tracking responses get more precise \citep{Simoncini:2012ik}. While it has been shown that the retina is capable of advanced computations beyond standard processing \citep{Gollisch:2010kv}, many capabilities remain unexplored \citep{Masland:2007wi}, for example the fine tuning of retinal responses to motion in natural images.
In this study we focus on the properties of the retinal ganglion cells (RGC) responses to motion in several scenarios. In particular, we introduced an artificially generated yet complex naturalistic stimulus called Motion Clouds (MC), described by \citet{Leon:2012dh}, and for which key properties of natural stimuli are preserved (see details in \BOXmotionclouds). The recorded activity of a population of RGC was compared using simple moving stimuli (drifting gratings) and MC through same speed and two levels of naturalness, in terms of the spatial frequency spectrum (narrow and broad bandwidth). We found that for a large fraction of RGC responding to motion the cells’ tuning bandwidths become narrower when the spatio-temporal frequency content of the stimulus increases. This result shows for the first time that adaptation (or fine-tuning) to the statistics of natural images emerges already in the retina.