Linking Systemic Endotoxin Exposure to Retinal Microglia Migration Through Mathematical Modeling
Curation statements for this article:-
Curated by eLife
eLife Assessment
This study presents a useful mathematical model for investigating microglia migration in the retina across different anatomical layers during acute inflammation. It provides a solid methodological and technically rigorous analysis of the modelling framework, which is applied to experimental data from three different species. The model can be used as a reference point for future studies to investigate inflammatory responses in the retina, with some current conclusions concerning the relevance of pharmacokinetic aspects potentially benefiting from more detailed analyses or future datasets.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Microglia are the resident immune cells of the central nervous system (CNS), including the brain, spinal cord, and retina, where they serve as the first line of defense against infection and inflammation. Dys-regulated microglia activity has been implicated in vision-threatening diseases, highlighting the need to understand how retinal microglia respond to inflammatory stimuli. Importantly, acute inflammation induces substantial redistribution of microglia across retinal layers, yet the mechanisms governing this migration remain poorly understood. Here, we develop the first mathematical model of retinal microglia migration during inflammation to determine how inflammatory exposure, administration route, and species-specific pharmacokinetics shape redistribution dynamics across the retina. The model couples lipopolysaccharide (LPS) pharmacokinetics with microglia migration between the outer plexiform layer (OPL), inner plexiform layer (IPL), and ganglion cell layer/nerve fiber layer (GCL/NFL). Model parameters are calibrated to retinal microglia density measurements from mice following LPS (bacterial endotoxin) challenge, before extending the framework to rats and rhesus macaques to investigate species-specific responses. Simulations also compare how administration route, i.e. intravenous or intraperitoneal injections, alter retinal LPS exposure and subsequent microglia redistribution. Our results suggest that redistribution patterns are driven primarily by LPS delivery route and species-specific pharmacokinetics, rather than the initial microglia distribution across retinal layers. Together, these findings provide new insight into immune cell reorganization in the inflamed retina and demonstrate how mechanistic mathematical modeling can be adapted across experimental designs, administration routes, and animal species.
Article activity feed
-
eLife Assessment
This study presents a useful mathematical model for investigating microglia migration in the retina across different anatomical layers during acute inflammation. It provides a solid methodological and technically rigorous analysis of the modelling framework, which is applied to experimental data from three different species. The model can be used as a reference point for future studies to investigate inflammatory responses in the retina, with some current conclusions concerning the relevance of pharmacokinetic aspects potentially benefiting from more detailed analyses or future datasets.
-
Reviewer #1 (Public review):
Summary:
In this paper, Jackson et al. propose a mathematical model - to be precise, an ordinary differential equation (ODE) model - of microglia migration in the retina following inflammation, which in the data used in this paper, is induced by injection of bacterial lipopolysaccharide (LPS). The model consists of a component describing the LPS absorption (1-2 equations) and a component describing microglia migration through three tissue layers (3 equations). The model is fitted to data from mice, rats, and rhesus macaques. The results suggest that microglia migration kinetics mattered the most for shaping the outcome, whereas LPS distribution kinetics and the initial distribution of the microglia appeared to matter less.
Strengths:
(1) The authors state that this is the first model of microglia migration …
Reviewer #1 (Public review):
Summary:
In this paper, Jackson et al. propose a mathematical model - to be precise, an ordinary differential equation (ODE) model - of microglia migration in the retina following inflammation, which in the data used in this paper, is induced by injection of bacterial lipopolysaccharide (LPS). The model consists of a component describing the LPS absorption (1-2 equations) and a component describing microglia migration through three tissue layers (3 equations). The model is fitted to data from mice, rats, and rhesus macaques. The results suggest that microglia migration kinetics mattered the most for shaping the outcome, whereas LPS distribution kinetics and the initial distribution of the microglia appeared to matter less.
Strengths:
(1) The authors state that this is the first model of microglia migration in the retina following inflammation. While I did not check this claim in detail, I am indeed not aware of any earlier migration modeling efforts done in this specific field. There are a number of adjacent fields in immunology where the migration of other immune cells has been studied in great detail; examples include T cell migration in the lymph nodes, skin, liver, and brain, B cell migration in lymph nodes, tumor-infiltrating lymphocytes, and many others. This paper therefore adds to an already rich body of literature and could be potentially interesting for a diverse group of modelers in immunology. Several related migration models are cited in the paper already.
(2) The paper is clearly written and easy to follow.
(3) The model has a simple and intuitive structure and is easy to understand.
(4) The model is analyzed in great detail, including several parameter sweeps and a range of sensitivity analyses that were performed to understand the uncertainty of model predictions.
(5) The approach to calibrate the model primarily to data from one species and then perform a range of predictions of the impact of species-specific differences, given available data, is interesting and is not frequently done in the modeling literature, as far as I am aware. This could serve as an inspiration for other modeling studies.
Weaknesses:
(1) Related to the first strength, the links to existing models are currently not made explicit (although the papers are cited, the models are not explicitly compared to the paper's model). A more detailed comparison could make the work even more relevant for others.
(2) The main research question of the paper - how do the different kinetic parameters shape the outcome - is rather generic and of a somewhat exploratory nature; it does not appear to be linked to a current specific question in the field of retinal neuroinflammation. This potentially limits the impact of the work; while mathematical modelers could be interested in the first model of this specific system, biologists may find this paper more difficult to digest and may be less motivated to work through this body of work.
(3) The ratio of model parameters to amount of data appears to be on the high side. The authors are aware of and credit this limitation, and conduct multiple sensitivity analyses and practical identifiability analyses. This is laudable, but ultimately these analyses do show signs that the model may be insufficiently constrained by the data, at least in some aspects. In this regard, it could be important to clearly state the number of raw data points used (see suggestions below).
(4) The ODE formulation of the model could overlook important spatial variation of microglia density. Since the cells appear to be quite dispersed (from looking at the figures in reference 15), an agent-based model or other spatially explicit model could add additional important insight. Indeed, the model includes a few terms such as the saturation terms in Equations 5 and 6 that appear to be included to account for such spatial effects indirectly.
(5) The model predicts how microglia distribution across the retina should be affected by changing administration routes and/or LPS kinetics. LPS data from rats and rhesus macaques are used to re-fit the LPS part of the model, and other parameters are kept fixed. However, it seems to be at least conceivable that the other parameters could also be species-dependent. There is no data available for checking these predictions for other species, making this part of the work more speculative.
-
Reviewer #2 (Public review):
Summary:
The authors present a novel, first-of-its-kind mathematical framework aimed at investigating the spatial redistribution of retinal microglia across specific anatomical layers-the outer plexiform layer (OPL), inner plexiform layer (IPL), and ganglion cell layer/nerve fiber layer (GCL/NFL)-during acute inflammation induced by exposure to lipopolysaccharide (LPS; a bacterial endotoxin). The model couples an LPS pharmacokinetic (PK) model with a system of ODEs describing cell migration between layers. The authors used experimental data from mice subject to an LPS challenge to select and calibrate a final model among different candidates. The authors then extended the framework to rats and rhesus macaques using blood LPS PK data, and simulated how different administration routes (intravenous vs. …
Reviewer #2 (Public review):
Summary:
The authors present a novel, first-of-its-kind mathematical framework aimed at investigating the spatial redistribution of retinal microglia across specific anatomical layers-the outer plexiform layer (OPL), inner plexiform layer (IPL), and ganglion cell layer/nerve fiber layer (GCL/NFL)-during acute inflammation induced by exposure to lipopolysaccharide (LPS; a bacterial endotoxin). The model couples an LPS pharmacokinetic (PK) model with a system of ODEs describing cell migration between layers. The authors used experimental data from mice subject to an LPS challenge to select and calibrate a final model among different candidates. The authors then extended the framework to rats and rhesus macaques using blood LPS PK data, and simulated how different administration routes (intravenous vs. intraperitoneal) and species contexts might influence retinal microglia layout. To draw core conclusions, the authors also executed a careful global sensitivity analysis. Their primary conclusion is that redistribution patterns observed in mouse retinal microglia, and hypothesized in rats and macaques, are dictated predominantly by LPS pharmacokinetics and delivery routes rather than initial cellular distribution states.
Strengths:
(1) The manuscript establishes a first-of-its-kind, baseline mechanistic model for retinal immune cell movement between specific layers. This constitutes a solid contribution to theoretical studies of opthamology.
(2) The authors perform thoughtful model selection and include in the supporting information a suite of alternative models. By testing and comparing alternative modeling hypotheses, the authors support their final choice of model as a deliberate and data-driven process.
(3) High technical quality: The authors present a clear, thorough, and transparent technical description of the modeling and subsequent analyses. The use of profile likelihood to estimate parameter intervals is strong as it reveals any structural identifiability issues. The inclusion of a meticulous global sensitivity analysis adds rigor to the investigation.
Weaknesses:
I do, however, have some concerns about this study.
(1) The authors are clear that this is a first-of-its-kind study, and while that's exciting, the authors never explicitly explain why mapping the precise spatial redistribution of microglia across the OPL, IPL, and GCL/NFL matters to the broader scientific community. This is likely low-hanging fruit: what disorders are associated with inflammation in these layers, for example?
(2) The model relies entirely on blood concentration of LPS. That is, with one exception, every forward and backward migration rate in the ODE system (eqs 4-6) is directly multiplied by the circulating blood LPS fraction B/(B+eta_G, O). Because blood LPS concentration enters nearly every migration term, the model structurally assigns PK a central role in determining redistribution dynamics. It is difficult to disentangle whether the dominance of PK in the sensitivity analysis reflects a biological insight inferred from the data or a property imposed by the model architecture.
(3) In equations (5) and (6), the return pathway from the GCL/NFL back to the IPL (G -> I) is modeled as a homeostatic density-dependent term (k_G*G*(G/G_0-1)) that is completely independent of B. The authors explicitly justify this choice by stating it enforces a "once inflammatory signals begin to weaken, microglia in the GCL/NFL are expected to return to their homeostatic density on a shorter time scale" (lines 190-192). However, because the k_G term lacks any mathematical coupling to B(t), it does not "wait" for the stimulus to subside, even during peak inflammation. The term appears to represent density-dependent regulation rather than inflammation-dependent resolution. The biological interpretation given in the text does not seem to follow directly from the mathematics.
(4) It is also unclear why no LPS-dependent return pathway from the GCL/NFL to the IPL is considered. Such a mechanism could potentially provide a biologically plausible alternative to the purely density-dependent k_G term discussed above. Notably, no such term appears in any candidate model presented in the Supporting Information (the mention of k_GI in line 426 is likely a typo meant to be k_G).
(5) The 95% Confidence Intervals (CIs) for the primary migration parameters (k_OI and k_IO) are unbounded above (line 284, Figure S3). This practical non-identifiability is rather unsatisfying and concerning: it indicates that the available mouse experimental data is insufficient to put a ceiling on cell migration speeds. The data do not appear to constrain an upper limit on migration speed (potentially because of the use of Hill functions).
(6) Because the true statistical ranges stretch to infinity, the authors have to choose a range for PRCC. They choose to substitute an arbitrary Delta chi^2 <= 1 cutoff ("practical ranges") to run the PRCC sensitivity analysis and shift parameters through an arbitrary 600% blowout window. While the authors' transparency is commendable, the sensitivity analysis appears to depend on investigator-defined parameter bounds rather than ranges supported by the available data.
(7) A key study prediction is that species-specific PK is a major factor affecting redistribution. However, the model uses systemic blood LPS concentration as the driving variable for microglial migration. Ocular exposure is not explicitly modeled. This omission becomes particularly important in the cross-species simulations, where anatomical differences between species could alter the relationship between systemic and retinal exposure. Because species comparisons are performed primarily by changing the blood PK profile (and, in the case of macaques, changing initial cellular distributions) while keeping retinal migration dynamics fixed, it is unclear whether the reported differences represent species-specific retinal biology or simply the expected consequence of imposing different LPS exposure patterns.
(8) If I understand the framework correctly, the rat and rhesus macaque simulations utilize mouse-calibrated migration parameters and cellular sensitivity thresholds. Parameters governing responsiveness to LPS exposure (e.g., eta_O) are estimated entirely from mouse data and assumed to remain unchanged across species. The cross-species simulations primarily vary the blood PK profile while holding retinal response dynamics fixed. While this approach is useful for isolating the effect of systemic exposure, it limits the extent to which the resulting predictions can be interpreted as species-specific retinal biology.
-
Author response:
We sincerely thank the reviewers for their feedback. In the revised manuscript, we will expand our discussion on the significance of microglia redistribution, particularly as it pertains to retinal inflammation and degeneration. Microglia play a key role in retina development, maintenance and protection and their distribution across the layers of the retina is important [1,2,3]. Most notably, changes in microglia location can alter their interactions with the neuronal, glial and vascular cells associated with different retinal layers, potentially affecting functional processes such as synaptic maintenance [4,5].
This link makes microglia redistribution relevant to basic research into how the retina maintains homeostasis and responds to inflammation. We acknowledge that results obtained from parameterization of the model …
Author response:
We sincerely thank the reviewers for their feedback. In the revised manuscript, we will expand our discussion on the significance of microglia redistribution, particularly as it pertains to retinal inflammation and degeneration. Microglia play a key role in retina development, maintenance and protection and their distribution across the layers of the retina is important [1,2,3]. Most notably, changes in microglia location can alter their interactions with the neuronal, glial and vascular cells associated with different retinal layers, potentially affecting functional processes such as synaptic maintenance [4,5].
This link makes microglia redistribution relevant to basic research into how the retina maintains homeostasis and responds to inflammation. We acknowledge that results obtained from parameterization of the model to rat and rhesus macaque pharmacokinetics rely on a few assumptions recognized by both ourselves and the reviewers. Most notably, these assumptions include that microglia migration parameters and cellular sensitivity to LPS are transferable across species. To our knowledge, comparable time-resolved measurements of layer-specific microglia densities following systemic LPS challenge are unavailable for rats and macaques, making this a necessary assumption that we will elaborate further on in the revision. We hope that our results might motivate such experiments, enabling further confirmation of our modeling approach and predictions.
We also agree that the contribution of LPS pharmacokinetics must be interpreted in relation to the model’s structure and accompanying parameter uncertainty. In the revision, we will investigate the following.
- Compare candidate models fitted to the figure 3 data to test whether direct LPS effects confined to the outer plexiform layer, with density-dependent migration in the other layers, can reproduce the observed trajectories.
- Assess how rat and rhesus macaque predictions change when mouse microglial migration rates are scaled allometrically by body mass.
- Test a model in which migration from the ganglion cell and nerve fiber layers (GCL/NFL) to the inner plexiform layer (IPL) depends on circulating LPS.
- Examine how peak blood LPS concentration (Cmax) and cumulative exposure over a finite time window affect the predicted maximum and minimum microglial densities in the outer and inner plexiform layers.
- Use the model to generate synthetic datasets and test which types and amounts of data improve our ability to estimate the microglial migration rates.
We remain mindful that the available data render some migration parameters difficult to identify. In light of this, we will offer an improved discussion on the trade-offs between reproducing the observed trajectories, model complexity and parameter identifiability. Finally, we will clarify the uncertainty quantification methods, more clearly report sample sizes and the meaning of plotted summary statistics, and discuss the ideal data that could help constrain the model.
(1) Silverman, Sean M., and Wai T. Wong. "Microglia in the retina: roles in development, maturity, and disease." Annual review of vision science 4.1 (2018): 45-77.
(2) Singaravelu, Janani, et al. "Microglia in the primate macula: specializations in microglial distribution and morphology with retinal position and with aging." Brain Structure and Function 222.6 (2017): 2759-2771.
(3) Chen, Ling, Peizeng Yang, and Aize Kijlstra. "Distribution, markers, and functions of retinal microglia." Ocular immunology and inflammation 10.1 (2002): 27-39.
(4) Mills, Samuel A., et al. "Fractalkine-induced microglial vasoregulation occurs within the retina and is altered early in diabetic retinopathy." Proceedings of the National Academy of Sciences 118.51 (2021): e2112561118.
(5) Wang, Xu, et al. "Requirement for Microglia for the Maintenance of Synaptic Function and Integrity in the Mature Retina." Journal of Neuroscience 36.9 (2016): 2827–2842.
-
-
-
Authors used C57BL/6J from #InternationalMouseSt in their study. Including #RRIDs will make this less ambiguous. SciScore made a table with this resource, see “Automated Services” module (download as csv, xml or #jats) #STMpublishing #reproducibility
-
Linking Systemic Endotoxin Exposure to Retinal Microglia Migration Through Mathematical Modeling https://www.biorxiv.org/content/10.64898/2026.06.30.735687v1
-
Linking Systemic Endotoxin Exposure to Retinal Microglia Migration Through Mathematical Modeling https://www.biorxiv.org/content/10.64898/2026.06.30.735687v1
-