Saltar a PREreview

PREreview del Odour representations supporting ethology-relevant categorisation and discrimination in the Drosophila mushroom body

por Brown Owl del HHMI Transparent and Accountable Peer Review Training Program
Publicado
DOI
10.5281/zenodo.17925524
Licencia
CC BY 4.0

This study effectively combines connectomic analysis, in vivo calcium imaging, and computational modeling to investigate how the connectivity between the 2nd-order olfactory projection neurons (PNs) and 3rd-order Kenyon cell (KC) of the mushroom body supports odor categorization and discrimination in Drosophila. While the prevailing view has long held that mushroom body input is random, recent connectomics data indicates that PN-to-KC connectivity may actually deviate from chance. Bridging the gap between these anatomical findings and their functional roles, the authors provide evidence that αβ and α’β’ KCs have biased connectivity toward food-related projection neurons (PNs), facilitating the categorization and potential discrimination of food odors. In contrast, γ KCs process odors more randomly without category preferences. Below are suggestions to further strengthen the conclusions.

Major Points

1. The central conclusion is that αβ and α’β’ KCs segregate odors based on “ethological relevance” (e.g., food vs. non-food). However, the “food” odor panel consists primarily of esters, whereas the “reproduction” and “danger” panels are chemically diverse. It is difficult to determine if the observed clustering in αβ KCs results from ethological categorization or simply reflects the ability to distinguish esters. Including a more diverse range of “food” odors (e.g., ethanol, acetic acid) would strengthen the conclusion regarding ethological relevance.

2. For Figure 4, the model parameters (spiking threshold, excitability) were optimized so that vector lengths matched experimental data. There is a risk of overfitting, which might force the model to replicate the separation observed biologically regardless of the underlying connectivity structure. Consider testing a variety of parameters to demonstrate that the connection structure, rather than parameter tuning alone, drives category separation.

3. The network models make strong predictions about discrimination and categorization performance, but these currently lack behavioral validation. Although feasibility may be a concern, consider performing an assay where αβ/α’β’ silencing is expected to reduce such categorization between food vs non-food odors, whereas γ silencing should have smaller effects. This would further strengthen the conclusion that αβ/α’β’ KC connectivity is important for food odor discrimination.

Minor Points

1. In Figure 1B, consider including 3D orientation axes to clearly indicate the dorsal-ventral and anterior-posterior directions.

2. Although explained in the Methods, it would be beneficial for the authors to briefly elaborate in the main text on why Euclidean distance and Cosine distance were used in specific contexts and how they should be interpreted to improve reader comprehension.

Competing interests

The authors declare that they have no competing interests.

Use of Artificial Intelligence (AI)

The authors declare that they did not use generative AI to come up with new ideas for their review.

Comentarios

Escribir un comentario
  1. por Gaia Tavosanis
    Publicado
    Licencia
    CC BY 4.0

    Thank you for your valuable feedback. While we are preparing a fully revised version of this manuscript, we respond to each of your comments below:

    1. For this work, we started from the anatomical observation that the likelihood of ab Kenyon cells to connect to a specific group of PNs (FPNs in the manuscript) was higher than expected from a random distribution of connections. We thus concentrated on those FPNs and the odours that they respond to- and contrasted them with other types of PNs that do not belong to this particular group. Those FPNs respond in fact to odorants released through yeast fermentation of sugar: These esters products are the indication of ripe fruits and yeast presence, the main sugar and protein source for flies (Mansourian and Stensmyr 2015).

    To make this logic clearer (starting from the anatomy and then investigating the responses of the KCs based on the types of connections they form), we have worked on the presentation of the main text in our revised manuscript.

    Interestingly, while the chemical similarity among these odorants suggests that discriminating between them is a difficult task, we found that αβ KCs show larger Euclidean distances between these food odours than between other odours. This suggests a specific transformation to improve the discrimination performance of these esters by αβ KCs, specifically.

    2. We found that connectivity, but not physiological cell properties (threshold, excitability) are primary drivers of the enhanced categorization. The cell properties were fixed for all the connectivity models (Hemibrain, random uniform, random non-uniform), so the observed biases were the results of connectivity patterns and not the cell properties.

    3. We carefully considered a potential way of addressing experimentally the behavioral outcome of the categorization of representation of food odours that we report. We would test the capacity of flies to generalize learned information to odours that are in the same or different categories, using odour generalization paradigms established in the literature. As the Reviewer points to, though, in the context of the presented findings, we would need to separate the role of the individual KC types in these experiments (silence or activate individual KC types in behaviour experiments). A confounding factor in these types of tests is that flies need to first learn an association to be able to display generalization - but this step relies differently on the different KC types. A second important factor complicates the experimental planning: given that not many odours are clearly assigned to specific categories, it is unclear whether we will be able to define odours that are at similar chemical distances, but either in the same or in different categories to provide for appropriate controls for the test. The preparation of this test and the control of all the variables involved represents a long-term project that we feel exceeds the scope of this manuscript.

    Minor points:

    1. Thank you, we will consider adding these in the updated version.

    2. To evaluate the nature of the differences between food and non-food odour representations in the different KC types, we utilized two different measures of distance (Euclidean distance and cosine distance). The cosine distance between two odour representation vectors measures the angular distance between the two vectors. We found that the cosine distances between the food-odour representations were not significantly different from those between the other-odour representations, while the Euclidean distances were. These results suggested that the observed enhanced categorization might rely primarily upon different response amplitudes of KCs, Figure 3D. We agree that the original manuscript lacked an explicit explanation in the main text (but see lines 648-678 in the methods explaining how we analysed similarity among the functional responses of the different KC types). To improve the clarity, we explain the rationale for using the cosine distance in the results section and discuss more explicitly the implications in the discussion of our revised manuscript.

    Competing interests

    I am one of the authors and I am responding on behalf of all authors.