Completed on 11 Jan 2017 by Stephen D Van Hooser . Sourced from http://biorxiv.org/content/early/2017/01/03/097923.
Login to endorse this review.
Author response is in blue.
Suggestion: plot Fig 2f on a log-log-scale, or somehow allow us to see the more numerous points in the middle of the graph.
Good point. We should generate a supplementary figure for that. We'd like to keep both presentations, though, since one may also argue that large absolute changes in response amplitude of individual cells are more 'meaningful' than smaller changes.
Question: what are the typical "error bars" around the points in Fig. 2f. For example, if you calculated these responses by bootstrapping across trials, would the error bars be tight or large? That is, I am asking, are the points that greatly deviate from 0 on the Y axis "real", or are those values derived from noise? How many neurons exhibit an amplitude change that is significant, by such a "within bouton" analysis? (Right now I'm looking at Fig 2F, wondering if I believe that the changes from 0 on the y axis are "real".)
This is, again, a good point. Bootstrapping across trials is something we did in Rose et al 2016 (2H). As you can see in the high within-session ODI correlation (I know that this is not exactly what you asked for - but it comes close), within-session variability is not the main source of variability here. It's across-session variability that has additional sources of both 'real' and artefactual sources (we're going to have a review on this out ~1 weeks). I'd be very much open for suggestions on how to integrate a measure of confidence in Fig 2f that would not be overly confusing.
Our approach to this issue was to use 2f as raw-data display and quantify the data by ODI-binning in 2g.
Question: how certain can you be that the signal reflects individual boutons and is not occasionally contaminated by nearby boutons. If this situation occurred, it could produce the data that is observed as an artifact, without an actual "conversion" of an LGN cell.
Again - fair point in principle. However, this is precisely one of the reasons that we record volumetric data of sparse axons in L1 (Fig 1/S1). We have clear structural information of our imaged boutons and can therefore be rather certain that such 'neuropil' issues are comparably unproblematic in our case. The multiple levels allow us to take the optical section that allows the most clear separation of individual axons and also allows us to assess if crossing axons are an issue or not. On top of this we acquire structural stacks during visual stimulation (different experiment) that allow us to cross-check regions for overlapping axons.
Most importantly: This is, of course, exactly the reason for Fig 2a. The crisp functional maps of individual boutons of a single animal should be convincing enough to show the robustness of our longitudinal single bouton data - and should illustrate that the surprisingly robust change in single bouton ODI is _not_ caused by ROI contaminations.
Question: is it certain the calcium activity in the bouton reflects only the presynaptic signals? If you patched a cortical neuron and fired it, is truly no signal observed in the presynaptic boutons? (One might imagine responses derived via presynaptic NMDA receptors, etc.)
I would be very much surprised if acute local postsynaptic activity would have an influence on global presynaptic Ca2+ levels. I could envision _highly_ local effects on individual boutons that may be the result of long-term changes at the entire synapse that may affect the level of presynaptic Ca2+-entry per bouton in a retrograde fashion.
But: How would that explain eye-specific changes? Initially contralateral boutons gain ipsilateral eye responsiveness (2g). Changes at individual synapses would be eye-agnostic and either lead to overall increases or overall decreases in evoked Ca2+ changes that would _not_ change bouton ODI (but they do. See 2c).
Importantly: the changes we observe are relatively global and uniform for boutons putatively belonging to the same axon (not shown).
Juliane, Mark, and Tobiases: Interesting work! I am new to this, and I hope my "single issue posting" of questions and suggestions is acceptable. Best wishes, Steve
Hi Steve, Thanks - also for going through the data so thoroughly! We are also rather new to this. It's exciting to see that we get so much feedback already at this stage of the MS! I am certain that I will do this more often in the future.
(I have to say, though, that the commenting system is a bit confusing. I am not sure that all of my replies to your questions have been posted - they disappeared after a refresh. To avoid double-posting I'm going to wait a day to see if they actually appear on the site after they have been moderated...)
Suggestion: Put data in Fig 2F on a log-log scale, so we can see the more numerous points near the origin?
Question: what are the "error bars" around the individual points in Fig. 2F. For example, if one used bootstrap (across trials) to create a distribution of the likely change in response to contra or ipsi eye for each cell, we could see if the values were noisy or robust. Right now, as I look at Fig. 2F, I can't decide whether the points that deviate from 0 on the Y axis are "real" or just "noisy". Similarly, if you did the bootstrap, how many cells exhibit "significant" deviations from 0 on the Y axis, vs. those that exhibit significant deviations from 0 on the X axis?
Question: is it possible that there is contamination of the signal from nearby boutons, that would give the appearance of binocular changes when really they are monocular?
Question: is it for certain that the bouton responses are purely presynaptic? One might imagine that there could be presynaptic NMDA receptors that might show responses that are actually derived from activity in the post-synaptic cell. A control: patching a nearby cortical cell, and demonstrating that nearby boutons do not exhibit responses when the cortical cell is activated at similar firing rates as it is during in vivo visual stimulation.