In this paper, the authors use an impressive, global-scale sampling design on native and non-native house sparrows to test the relationship between microbial surveillance and cytokine responsiveness. The results are interesting, figures are clear, and the manuscript is well-written. I had some general comments about the data analysis, especially pertaining to potential effects of spatial autocorrelation, as well as whether inclusion of leukocyte profile data would enhance the interpretation of results. However, most of my comments are relatively minor.

L72: The key results statement here would be improved by making explicit mention of differences (or lack thereof) between native and non-native birds, as this stems back to the primary research objective and hypothesis (L66). The text as-is only describes what is happening in non-native birds.

L84: Include “hypothesis” after EICA as well.

L99: Comma after “but”

L127: It would help to make the comparison to what you would expect in native HOSP explicit here.

L147: It would help to briefly describe why the non-breeding season was prioritized here.

L151: Replace “DNA/RNA shield” with “DNA/RNA Shield”

L152: Can the authors briefly justify (or just provide a citation) for using this particular dose of LPS?

L159, 173, etc: See above RE DNA/RNA Shield

L195: To clarify model structure, did each LMM include population, baseline TLR expression, and their interaction as fixed effects?

L196: The data stem from 5 native populations and 4 non-native populations, across a global scale. While this is impressive, it also introduces potential issues of spatial dependence among sites (e.g., you would expect Australia and New Zealand data to look more similar based on proximity). The country-level random effect likely captures (and adjusts for) this effect, but the authors should test model residuals for spatial autocorrelation (and if present, consider alternative models that allow you to better account for spatial dependence).

L198: You could also make the argument that wing chord is much more repeatable than tarsus length (see Subasinghe et al. 2021).

L203: Were LMMs fit with ML or REML? The latter is required to obtain unbiased estimates of variance components.

L204: The authors should consider reporting marginal and conditional R2 for all models.

L207: Because the authors find a significant interaction among TLR surveillance and population size, it is unnecessary (and potentially confusing) to describe the effects of population alone. The interaction is describing how the relationship between TLR surveillance and cytokine response differs by population.

L224 and/or L233: The wording here is nice (highlighting the key results in the context of the native/non-native comparison) and could be adopted in the abstract.

L260: Comma after non-native

L275-300: The discussion here of leukocyte profiles and HL ratios is appropriate and interesting, but it also raises a question of whether the authors collected blood smears and could simply test for %H, %L, and HL ratio differences among native and non-native populations here (the same LMM), as well as if leukocyte composition could explain the negative relationships observed.

The authors have done a very nice job in their revision, and I thank them for addressing the vast majority of previous concerns.

The one outstanding comment pertains to spatial autocorrelation. I am aware that house sparrows are non-migratory and typically have small home ranges, thereby reducing the possibility that birds could actually move between any of these field sites. However, you could still expect birds in geographically closer populations to have more similar responses to LPS based on evolutionary history (e.g., within the native range, Spain and Netherlands are likely more similar than Spain and Vietnam) or the spatiotemporal trajectory of the invasion (e.g., you would expect Australia and New Zealand to be more similar than Australia and Canada). While I do think your site-level random effect most likely accounts for any residual spatial autocorrelation, I think the authors would only need to test their model residuals for global Moran's I or plot the correlogram of local Moran's I against inter-site distance.