Rachel Potter, Anne Colgan, Chuck Flaherty, Lauren Buckley, Grace Leuchtenberger, Julia Smith, Kindall Murie, Olivia Cattau, Sanford Leake, Zachary Bengtsson

We are a pre-review journal club made up of ecologists, primarily graduate students. We chose this manuscript because it fit nicely with our interests in lepidoptera and network analysis. 

Summary

This paper investigates the relationship between environmental factors and the structure of plant-lepidoptera potential (inferred) interaction networks. The analysis reveals differences in diet breadth, dietary niche overlap, and modularity across lepidoptera clades, life stages, and environmental factors. The authors compare the observed network metrics with null models to test for contributions of (1) species/environmental reliance and (2) lepidoptera-plant interdependencies on the observed network structure. Finally, the authors provide ecological explanations for the patterns they observe and briefly discuss potential conservation implications. Overall, the paper presents a well explained, data-rich analysis of patterns in potential interactions between lepidoptera and plants. It could benefit from more explicit hypotheses that clearly account for the limitations of the data, including that the networks are based on inferred interactions over relatively large spatial scales. 

Strengths

• The paper is well written and well explained. The authors do a particularly nice job summarizing their findings in layman’s terms (e.g. line 102 “Here, we integrate biogeographic (i.e., who occurs where) and ecological (i.e., who interacts with whom and how” ) and translating the quantitative analyses into ecological terms (e.g. line 233 “Notably, for the hereinafter network analyses we excluded small local networks with either plant or Lepidoptera richness fewer than ten. Small networks tend to generate unreliable and artifactual structural-metric readings”).
• The literature review is thorough and flows well. It is accessible both to audiences with expertise on lepidoptera but little knowledge of network theory and vice versa. For instance, It does a good job of introducing features of lepidoptera that make them ecologically interesting, explaining how Lepidopterans’ two distinct life stages contribute to “dual roles” in their interaction with plants (mutualistic and antagonistic) so both stages are important to consider in network analyses. It also does a good job of explaining network related terms such as modularity. 
• Figure 1 is both visually pleasing and very effective; visualizing the data spatially helps with interpretation of the study design. Figure 3 is very helpful in understanding the null model comparisons and interpreting the rest of the figures.
• The inclusion of null models strengthens this analysis. The null models are well explained both in terms of how they were constructed and their ecological interpretation. 
• The inclusion of separate networks for different life stages and clades aids in ecological interpretation and increases the conservation relevance of the analysis especially since the nature of the relationships between plant-larval lepidoptera and plant-adult lepidoptera are so different. 
• The scale of this analysis is novel and impressive. The authors compare community metrics at a regional scale across over 300 grid-cells/communities at two life stages and four different clades. We appreciate the ambitious effort to link environmental and biotic factors and integrate biogeographical and ecological knowledge at a community level. 

Major Concerns

• In this paper, the authors use species occurrence data to create inferred interaction networks. If two species co-occur within a grid cell and are known to interact, they are assumed to be interacting in the network. Our major concerns with this approach include the following:
• The spatial scale at which co-occurrence is defined is large (10 x 10 km^2). It seems like a major assumption that species that co-occur within a 10 x 10 km^2 area are interacting. 
• Similarly, the analysis could benefit from some validation that species that co-occur spatially also co-occur temporally. As we understand it, the analysis currently assumes that species that have been observed to interact at one location in a given year are able to interact at any other location, where they both occur, in any year. Due to phenological variation and resulting temporal mismatches between interacting species, this may not be a valid assumption. The authors could incorporate iNaturalist or other citizen science data to control for phenology and see which individuals are likely to interact given temporal and spatial overlap. 
• The approach does not account for the influence of intra-guild competition on niche breadth and the dynamic nature of interaction networks.
• It does not account for spatial autocorrelation.We would expect the composition and characteristics of neighboring grid cells to influence species observed near the edge of grid cells. 
•  This approach may be appropriate for creating networks of potential interactions; however, it is not clear whether such “potential interaction networks” are sufficient or appropriate for addressing the authors’ hypotheses. We suggest that the paper would benefit from a clearer articulation of the research questions/hypotheses and a justification for why it is appropriate to use inferred interaction data to answer the research question(s) and/or an explicit discussion of how relying on inferred rather than observed interactions might affect the results. 

• Lines 289-290: The R^2 values for the regressions on the two PCs are very small at 0.04-0.05. The highest value was 0.09 (line 336). If under 10% of variation in the Z-scores can be explained by these PCs, what else accounts for variation? 
• It would be good to use a dataset using known rather than inferred networks to validate the statistical methods used here. It feels like this paper is trying to accomplish two goals: (1) introduce statistical workflows to understand networks (presented in figure 3), and (2) examine network dynamics in a system over a large geographic scale. It might benefit from being split into two papers, one which examines the statistical tools in a smaller dataset, and another that deals with this particular system with more validation (for instance, collecting observations in the field in 5-10 cells and see if predictions match what is observed).
• Data is from 30 years, but interactions most likely changed in 30 years due to ecological and evolutionary factors, especially in light of climate change. From figure S2, it looks like there is relatively even sampling over the past 30 years, so perhaps you could subset the occurrence into 10 year periods. 

Minor Concerns

• Lines 166-167: It would be helpful to mention more specifics of the Ebert (1991–2005) datasets. How many observations does this dataset include? Might be helpful to include a summary table, at least in the supplementary section of observations for each clade and life stage.
• The information presented in figures 4-6 is removed from the underlying data making interpretation difficult. The authors provide a helpful real world interpretation of what the PCs mean: they interpret PC1 as geoclimate and PC2 as land cover. If land cover and geoclimate are in fact the main drivers it may be stronger and  more interpretable to use them as predictor variables rather than the PCs. This would allow the reader to more easily draw biologically relevant conclusions from the figures.
• Line 224-225: In the supplemental section, the authors looked at connectance and nestedness. It would be good to elaborate on the findings in the main text, why the authors excluded these, and why they chose the three that appeared in the main text.
• It would be helpful to provide definitions of the terms “species-environment reliance” and “lepidoptera-plant interdependence”. Introducing these terms in the introduction and providing background on why they were chosen could help set up the motivation for the hypotheses being tested. 
• Line 351: The authors state that they looked at abiotic factors such as “climate and land cover.” However, the authors looked at elevation, not climate directly, and we would classify land cover as biotic, since the composition of plants are changing (trees vs farmland vs urban). While there are strictly abiotic factors to consider, such as the amount of shade, the authors didn’t measure these. Therefore, saying the authors looked at abiotic factors seems overstated. We understand the attempt was to draw the distinction between the primary biotic interactions (pollination and herbivory) that they are studying versus the rest of the environmental conditions, so this could be clarified using different terms. They could more clearly explain why they think the variables they selected are appropriate for characterizing the landscape scale differences in habitat between grid cells and discuss the limitations of those variables.
• The authors could also look at climatic variables for each of the gridded cells such as moisture and sunlight from local weather stations and incorporate that. However, due to phenology, they would need to know the time the larvae and adults are active to find corresponding climatic variables that would be biologically realistic.
• Line 366: Might be good to define “geometric Rapoport effect”
• In the methods section the authors do a good job communicating that the interaction networks and derived network metrics represent potential rather than observed interactions; however, in the discussion they appear to draw conclusions about the actual interaction structure. For example, in the paragraph starting on line 469, the authors write: “By addressing the geo-climate and land-cover relevance of landscape-scale Lepidoptera-plant networks, we revealed across Lepidoptera clades and life stages how their (co-)evolved traits (e.g., biological diet breadth) and ecological relationships with others (e.g., resource competition, local food availability) collectively drive different interaction structures in response to environmental variations.”