Referee 3's Review (Gonzalo de Polavieja):

**********
N.B. These are the comments made by the referee when reviewing an earlier version of this paper. Prior to publication, the manuscript has been revised in light of these comments and to address other editorial requirements.
*********

Faivre and Juusola have made long recordings from locust photoreceptors and analyzed their different responses depending on background level, temperature and type of stimuli (naturalistic vs. white noise). They use a complete suite of methods, including information theory and bump analysis. Altogether, their approach is both experimentally and theoretically very elegant. Their analysis is long and with many figures. This will allow interested readers to know all details. However, I miss some originality. Their most interesting results are not analysed further and I cannot see they take the advantages of long recordings to their full potential.

In any case, in my opinion the paper fulfils the criteria set by PloS ONE and therefore recommend the paper for publication. However, I wish the authors could take into account my comments below.

Page 12. Third paragrah. The authors mention that for both NS and WN the signal is the average response and the noise the difference of response and signal. This is true for WN when the response is linear, but not for NS. The basic object in the noise term for NS is the probability of a response given the stimulus. And nothing more. I believe this has led the authors to lot of confusion in their analysis of responses to NS.

Page 26. Last paragraph. Comparing responses to NS and WN, the authors conclude that there might be evidence for a preference for the long-term correlations in the stimulus. This is one example where the authors might have found something of more relevance, but the question is not pursued experimentally. For example, simple stimuli with long-term correlations, or stimuli that interpolate between WN and NS could have been use for this and many other questions raised tangentially by the authors.

Page 37. First paragraph. The authors mention that the coupling of slow and fast dynamics will exhibit some for of memory. I think it would be nice if they cited references for a proof of the statement, at least for the mechanisms of adaptation to statistics.

Page 41. First paragraph. The authors mention that they always record at same time of the day. This is also a relevant question not pursued by the authors: how does coding vary during the day? Is it circadian? The authors had the tools for a quantitative answer.

Page 42. Second paragraph. The authors mention that the variability could be due to position in the eye. This is again another interesting problem they pursue no further.

Another issue they do not even mention (or I missed it) is that locusts have a gregarious and a solitaire phase. Does the encoding of photoreceptors change when changing phase? Clearly, if the photoreceptors are matched to stimulus statistics, a difference should be observed. A story around this issue would have probably made an interesting paper. If observed, I think that recordings might even be long enough to capture the transition from one phase to the other.

Page 45. Equation (4). In the denominator, why not to write modulus square instead of product of S (or C) with its complex conjugate.