Mathematical analysis of photoreceptor changes in conditions of separation from the underlying retinal pigment epithelium Authors: Erika T Camacho, Miriam A Goldman, Lucia Gonzalez-Buendia, Daniel E Maidana, Demetrios G Vavvas, Stephen Wirkus Published: 2026 ABSTRACT AbstractPhotoreceptors (PR) are responsible for absorbing and converting light into electrical signals necessary to create vision. To accomplish that, they are in an intimate relationship and attached to the underlying retinal pigment epithelium (RPE). Separation of the PR from the underlying RPE is seen in many retina disorders and leads to PR cell death and subsequent vision loss. Separation of PR from the RPE (detachment) interferes with the normal phagocytosis and recycling of PR outer segments by the RPE, and disrupts nutrient and metabolite delivery, including glucose. All these factors contribute to cell dysfunction and eventual cell death. In this work, we sought to understand the importance of different factors in PR cell loss after detachment using mathematical modeling. We used known information of photoreceptor interactions in the healthy retina from the literature, and datasets for rod and cone degeneration after detachment. We also included three additional published datasets of PR cell death kinetics. A mathematical sensitivity analysis examined the impact of the parameters on the system over a detachment of 150 days and found that the parameters that significantly impact the rod and the cone population at the stages where “early intervention” happens (3–7 days after detachment) are not always the same ones that significantly impact the populations at 150 days. Additionally, some of the parameters negatively affected one population while having the opposite effect on the other. An increase in nutrient availability and efficiency of rod energy uptake were the only parameters that did not have a negative effect on either population. Similar results were obtained for reattachment. The interplay between these variables indicates that effective photoreceptor neuroprotection in retinal detachment may require multiple strategies. The prediction of these impactful parameters over time can be further assessed in experimental models and may provide guidance on the most effective ways to improve PR survival after their separation from RPE.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-53417-3.Subject terms: Cell biology, Diseases, Neuroscience Introduction The retina is the part of the eye responsible for phototransduction. It utilizes millions of photopigment molecules to absorb photons of different wavelengths and convert them into electrical signals that are processed and sent to the brain to form what we call vision. There are two distinct types of photoreceptors involved in phototransduction: rod photoreceptors, responsible for night vision, and cone photoreceptors, responsible for color and daylight vision. Rods and cones are in direct apposition to each other, and both lie in a close relationship with the retinal pigment epithelium (RPE), which regulates nutrient flow as well as recycling of PR essential products via phagocytosis (see Fig. 1). Photoreceptors are supplied with oxygen from the choriocapillaris layer underneath the RPE. Rods are more numerous in mammals and make up about 97% of the total photoreceptors in the mouse eye, with cones making up about 3%1,2. The cell bodies of both are located in the outer nuclear layer (ONL)3. Layers of the outer retina. Representative cryosection of a murine retina stained with peanut agglutinin (PNA) (red) and DAPI (blue). Nuclei of cones and rods are stained blue. The nuclei of the photoreceptors form the outer nuclear layer (ONL); Cone inner and outer segments are stained in red. The RPE is not stained in this photograph, besides the blue nuclei in the monolayer abating the PR outer segments (OS). Under the RPE is the choroid and choriocapillaris (not stained, besides the nuclear blue stain). The renewal of the photoreceptors occurs at the basal portion; shedding occurs at the distal tip that interfaces with the RPE. Rods and cones are close in proximity and uptake nutrients, such as oxygen, and metabolites from the same resource pool. In a healthy eye, the rods and cones are able to share these resources and thrive in the same nuclear layer. However, it is poorly understood how resources are utilized in different states of retinal degeneration, and more importantly, how this differential utilization can further affect the survival of the remaining photoreceptors or neighboring cells. Moreover, it is important to acknowledge that excessive delivery can also be harmful. For example, excess oxygen availability can contribute to harmful free radical formation4,5. Over the course of a day, normal vision processes cause rods and cones to undergo great changes. Each day, new disks are formed, and previously formed disks are shed towards the RPE, and ultimately phagocytosed by the latter. Although phagocytosis follows a circadian rhythm, the overall rate of formation and shedding is kept approximately equal over the course of a day6. Disruptions in shedding and renewal of the photoreceptors are important since they can affect the photoreceptors’ morphology and functionality and, in turn, sight3. The RPE is a monolayer of pigmented cells that form the outer barrier between the blood and the retina. The RPE transports ions, water, and metabolic products from the sub-retinal space (a potential space) back into the blood. It absorbs nutrients, including glucose, retinol, fatty acids, and other essentials from the blood, and delivers them to the photoreceptors. In our model this is referred to as the trophic pool7. The rods, cones, and RPE have been modeled as a functional unit because of the dependence of the photoreceptors on the RPE8–11,. Retinal detachment is the separation of the photoreceptors from the RPE. This occurs in many retinal diseases, such as rhegmatogenous retinal detachment, exudative age-related macular degeneration (AMD), central serous chorioretinopathy, diabetic retinopathy, and retinopathy of prematurity, among others. If the PR stay detached for prolonged time, significant cell death will ensue at least partly due to the dysregulation of the PR-RPE dynamics: nutrient delivery, trophic factor composition, and visual cycle homeostasis. Without the support and sufficient metabolic supply for survival, the photoreceptors undergo cell death through mainly caspase regulated apoptosis, and RIPK regulated necrosis. The longer time it takes to reattach the retina the more photoreceptor cell death occurs, with associated vision loss, thought large variability exists12–16. A recent manuscript proposed a math model to study rod dynamics during retinal detachment and our work is complementary to this as we consider cones as well17. In this manuscript, we present our experimental and mathematical work on retinal detachment and also reattachment in the “Results” section. We consider two different approaches to study retinal detachment: (i) an equilibrium approach where we assume the system is at equilibrium and then determine the parameters to match the detachment data; and (ii) a time series approach where we assume that some parameters have changed as a result of the detachment and use inverse problem methodology to determine these new parameters. Both of these approaches are based on the mathematical model discussed in the “Materials and methods” section. In addition, the same mathematical model can be applied to reattachment of the retina. Thus, we also present (iii) a time series approach for reattachment of the retina, where we compare the resulting reattachment parameters with the corresponding situation of detachment. We present a sensitivity analysis for both time series approaches (detachment and reattachment). This work aims to understand the importance of different factors and key mechanisms contributing to photoreceptor degeneration, survival, and function after their separation from the RPE through mathematical modeling, utilizing three previously published and one new dataset. Our modeling approach offers a valuable lens to interpret the intricate and interdependent metabolic, structural, and temporal changes that occur after retinal detachment; factors that are difficult, if not impossible, to isolate through experimental methods alone. Our mathematical model focuses on processes needed for the maintenance of photoreceptors. While many cells and processes are connected to their health, including Müller cells and ganglion cells, we focus on the renewal and shedding processes of the photoreceptors, nutrients, and trophic factors that are facilitated by RPE, as described above. Recent work has suggested that the rods, cones, and RPE work as a functional unit18. The results and their implications, as well as the assumptions and simplifications in our model, are important for future research and are further explored in the “Discussion” section. Results Retinal detachment: equilibrium approach We quantified the most influential parameters on cell death and survival so that through improved understanding, better targeted treatments to save photoreceptors may be tested. Three previously published datasets (C57BL/6, BALB/c, and B6129SF2) give the ONL thickness from which we can deduce the approximate number of rods and cones alive where we assume rod and cone death occurs proportional to their respective ratios in the mice. The fourth dataset, generated as part of this work, will give the additional understanding of any differences between the death kinetics of rods and cones after retinal detachment. Following the procedure introduced in9, we assume that the system begins at equilibrium and then observe the effect that detachment has on the parameters. As discussed in the “Materials and methods” section, our variables are Number of rods Number of cones Trophic pool / contribution of RPE/ Choriocapillaris We let (R*, T*, C*) denote the equilibrium values of the mathematical model, Eqs. (1–3). In Eqs. (1–3), R, C, and T represent the number of rods, number of cones, and the trophic pool, respectively, where the trophic pool encompasses all nutrients, including glucose, oxygen, and metabolites, delivered to the photoreceptors by the RPE and choriocapillaris. Equation (1) describes rod dynamics: rods continuously renew their outer segments (OS), consuming energy in the process (captured by an), and periodically shed the tips of their OS to the RPE, which also requires energy (captured by µn). Equation (2) governs the trophic pool: in the absence of photoreceptors, nutrient inflow from the RPE would reach a maximum level of Γ/κ, where Γ describes nutrient inflow and κ represents the biological constraints that limit how much the RPE can deliver. When rods and cones are present, they draw down this pool at rates β and γ respectively. Equation (3) describes cone dynamics: cones take up glucose at a baseline rate p, which represents the intrinsic glucose demand of cones for OS renewal when no rod influence is present. This baseline uptake is augmented by RdCVF, a protein secreted by rods that stimulates cone glucose uptake by increasing the abundance of the glucose transporter GLUT1; the Holling type III term captures this accelerating, saturable effect, with Vmax representing the maximum achievable rate of GLUT1 formation, km the RdCVF concentration at which half that maximum is reached, and dn the amount of free RdCVF available. Finally, µc represents cone energy consumption associated with OS shedding. Full parameter definitions and values are provided in the “Materials and methods” section. At equilibrium, Eqs. (1–3) can be solved for the parameters an, κ, and p to obtain Equations (4–6) thus tell us how three key parameters change over the period of detachment, with all other parameters remaining constant: an (the rate at which rods take up energy to sustain their outer segment renewal), κ (the biological constraint that limits the maximum nutrient supply the RPE can deliver to the photoreceptors), and p (the intrinsic rate at which cones take up glucose independently of any rod influence; that is, the baseline cone glucose demand when rods are absent or non-functional) change over the period of detachment with all other parameters remaining constant. The remaining parameters are defined in the “Materials and methods” section and remain constant for this discussion. Data from induced retinal detachment in C57BL/6, BALB/c, and B6129SF2 mouse models by subretinal injection of sodium hyaluronate from14 is used and we then fit cubic splines to the data points representing TUNEL positive cells. These data points can be used to obtain the number of rods and cones still alive. We considered the following key quantities: Equations (7–10) that were previously identified in8,9,19,20 for analyzing the stability of the equilibria. In words, Dn is the ratio of energy consumption to uptake in the rods, Dt is the overall nutrient carrying capacity provided via the RPE that includes metabolites, oxygen, and glucose, ac is the energy uptake in the cones due to both natural glucose uptake and RdCVF, and Dc is the overall metabolism in the cones. From Eqs. (4)-(6) together with the R*, T*, and C* values derived from the TUNEL curves, we allowed κ, an, and p to change under the equilibrium assumption. Substituting into (7–10), we observed minimal change involving an, and p but a pronounced change involving κ. Figure 2 shows the decrease in Dt over time as the rods and cones degenerated for the four datasets. The percentage decrease in the four datasets (from smallest decrease to largest) are 15.9%, 28.3%, 38.2%, and 46.5%. Plot of Dt = Γ/κ over time for each of the four datasets, assuming equilibrium conditions. The diamonds are the points where the greatest change in photoreceptors take place. The percentage decrease in the four datasets (from smallest decrease to largest) are 15.9%, 28.3%, 38.2%, and 46.5%. The dark blue dataset is previously unpublished from this group; the other three datasets are from14. Experimental studies have shown that, for example, short-term overexposure to oxygen in the time immediately following detachment helps prevent degeneration. Although the actual oxygen delivered may not have changed, the PR may consume more O2 after detachment. We can also study which mechanisms represented by the parameters have the greatest effect on minimizing death of the photoreceptors. Retinal detachment: time series approach Figure 3A-B examines photoreceptor loss during detachment; a related paper examined local cell death differences21. Because our mathematical model described in the “Materials and methods” section and given in (1–3) assumes differences in the rod and cone populations, our experimental data also considered rods and cones separately in the detachment data and this is shown in Fig. 4 (data not published elsewhere). Total photoreceptor loss on days 1, 3, 7, and 10 days after detachment. The detached part of the retina was compared with the attached part of the retina of the same eye. The number of ONL cells were plotted over the observed time points. As expected, a progressive reduction of the mean number of ONL cells was found in the detached retina compared to attached retina. The error bars represent standard deviation for n = 4; previously unpublished data. Rod and cone loss on days 1, 3, 7, and 10 days after detachment. For rods (A, left panel), the greater loss of rods was found 1 day after retinal detachment. For cones (B, right panel), we observed a significant reduction and this loss showed a linear trend that did not plateau within the first 10 days after retinal detachment. The error bars represent standard deviation for n = 4; previously unpublished data. For every eye, two regions 230 μm wide of attached retina and two of detached retina were selected, excluding peripheral and paracentral area. In all cases, the detached part of the retina was at its maximum high (reaching the lens) to reduce variability. Five sections with a separation of 100 μm were quantified for every eye. Rods make up 97% of the mouse photoreceptors; while it might be reasonable to think that rods and cones are affected the same by retinal detachment, these data show that this is not the case. That is, our results suggest that it is important to separately consider rods and cones in looking at photoreceptor degeneration, in addition to needing to more closely investigate rod and cone death. We derived the number of rods and cones as follows: In examining the number of rods at different time points, we observed a decreased number of these photoreceptors after retinal detachment. The rod and cone count data shown in Fig. 4 are new, unpublished data generated as part of this study using the C57BL/6J detachment model described in the “Materials and methods” section. These data are distinct from the ONL thickness measurements from Maidana et al.21, which are also used in Fig. 5. The Matsumoto et al.14 paper provides the three previously published datasets (C57BL/6, BALB/c, B6129SF2) that use TUNEL counts rather than separate rod/cone quantification. The greater loss of rods was found 1 day after retinal detachment, and this finding is consistent with results previously published by this group14; see Fig. 4A. For the number of cones over time, our cone count revealed a significant reduction in the mean number of these photoreceptors after retinal detachment. This loss showed a linear trend and, to our belief, we have not yet seen a plateau for this loss within the first 10 days after retinal detachment; see Fig. 4B. Degeneration during detachment: Time series comparison of normalized math model output with normalized experimental data for rod count and cell count data as well as ONL thickness. Seven parameter values are the same for detached and reattachment models: µn = 0.103, β = 2.50 × 10− 8, γ = 7.5400 × 10− 6, and µc = 0.119, km =0.0614, dn =7.5 × 10− 12, and Vmax =0.375; these are explained in more detail in the text of the manuscript. The remaining parameter values used in the decay model are Γ = 2.35, an =1.1 × 10− 6, κ = 0.00023, and p = 5.824 × 10− 6. The ONL thickness is given in21; the rod and cell count data is previously unpublished. Figure 5 compares the model output with the time series data of Fig. 4 normalized to an initial maximum of 1. The parameter values used to generate these plots are the same as for regrowth except that parameter estimation is used to obtain values for Γ, an, κ, and p, which are given in the caption of Fig. 5. While the model output is not a great fit for all time points, we note that the initial decrease of the time-series data was only matched through a significant decrease in Γ; however, leaving Γ at such a low level quickly brought the photoreceptor numbers to near zero and caused a poor fit for the data values at 10 and 14 days. We suppose that this lack of fit over the full span of 14 days is likely caused by some parameter values changing over time. Nevertheless, for the sake of comparison with the equilibrium approach, we use these plots for the basis of the sensitivity analysis given in Fig. 6. Sensitivity analysis for rods and cones during detachment for baseline decay values given in Fig. 5. (a) Sensitivity for cones for the first 150 days. Any parameter that exceeds an R-value of 0.4 in absolute value is shown in the legend. (b) Zoomed in version of (a) for “early” intervention of 3 and 7 days, for medium-term intervention of 14 and 30 days and for “delayed” intervention of 150 days. (c) Sensitivity for rods for the first 150 days. Any parameter that exceeds an R-value of 0.4 in absolute value is shown in the legend. (d) Zoomed in version of (b) for “early” intervention of 3 and 7 days, for medium-term intervention of 14 and 30 days and for “delayed” intervention of 150 days. In order to determine the parameters for which a 10% change in value have the greatest effect on the model output, we perform a global sensitivity analysis. The baseline values for the estimated parameters are described in the previous section and are used in this sensitivity analysis. Based on Marino et al.22, we require PRCC-values greater than 0.4 in absolute value in order to be significant. Figure 6 shows the plots over 150 days with zoomed in plots at the time periods for “early” intervention (3 and 7 days), medium-term intervention (14 and 30 days), and “delayed” intervention (150 days). Cones Considering the key quantities (7–10), we see that both Γ and κ, parameters in the overall nutrient carrying capacity Dt = Γ /κ, are significant throughout in Fig. 6a-b. Both µc (the energy consumption in the cones) and p (natural glucose uptake in cones), which is present in ac and thus in the key quantity Dc = µc /ac, are significant throughout. γ is significant and negatively correlated at 10 days but loses significance around day 38. Rods Considering the key quantities (7–10), we see that Dt = Γ /κ is again significant as well as Dn = µn /an (the ratio of energy consumption to uptake in the rods) in Fig. 6c-d. The parameters µc (the energy consumption in the cones) and γ are present but not as significant as the others and losing their significance by around days 92 and 53, respectively. Retina reattachment: time series approach In the case of reattachment of the retina, we quantified the most influential parameters on cell survival and outer segment (OS) renewal so that through improved understanding, better targeted treatments to encourage photoreceptor OS regrowth may be tested. As discussed in the “Materials and methods” section, the definition of our variable changes slightly since new rods and cones are not created but rather the OS length increases. Thus, when we discuss “regrowth”, we are referring to OS height, knowing that the overall number of OS is not changing but their heights are, and assuming that OS regrowth is approximately the same for each OS. We again consider a 10% change in value of the parameters to determine which parameters have the greatest effect on the model output through a global sensitivity analysis; see Figs. 7 and 8. The baseline values for the estimated parameters are described in the previous section and are used in this sensitivity analysis. Cone growth after reattachment: Parameters that have a significant effect on cone OS length (with significant PRCC-values, greater than 0.4 in absolute value) over the regrowth period of 150 days after reattachment. The leftmost figure shows the full time period and the five figures that make of the right panel are zoomed in from this left panel and focus on the first 7 days after reattachment, then days 14, 30, and 90 after reattachment, followed by days 148–150 after reattachment. Rod growth after reattachment: Parameters that have a significant effect on rod OS length (with significant PRCC-values, greater than 0.4 in absolute value) over the regrowth period of 150 days after reattachment. The leftmost figure shows the full time period and the five figures that make of the right panel are zoomed in from this left panel and focus on the first 7 days after reattachment, then days 14, 30, and 90 after reattachment, followed by days 148–150 after reattachment. Cones As with detachment, considering the key quantities (7–10), we see that both Γ and κ, parameters in the overall nutrient carrying capacity Dt = Γ /κ, are significant throughout in Fig. 7. Both µc (the energy consumption in the cones) and p (natural glucose uptake in cones), which is present in ac and thus in the key quantity Dc = µc /ac, are significant throughout. The parameters µn and anboth appear for part of the time with an positively correlated and appearing significant around day 86 and µn negatively correlated and appearing significant around day 114. Rods Considering the key quantities (7–10), we see that Dt = Γ /κ is again significant as well as Dn = µn /an (the ratio of energy consumption to uptake in the rods) in Figs. 8 and 9. The parameters µc (the energy consumption in the cones) and γ are present but not as significant as the others and losing their significance by around days 92 and 53, respectively. Induced retinal detachment in C57BL/6J mouse by subretinal injection of sodium hyaluronate. Three consecutive areas (white rectangles) of the detached (D) and attached retina (A), beginning from the pars plana (pp) are shown. The first region, closest to the pars plana, was considered peripheral retina, and the two consecutive regions were considered central retina. The ROI is 230 μm wide. Comparison of time series sensitivity results: detachment vs. reattachment Discussion Our mathematical model of photoreceptor interactions allows us to gain quantitative insight into key mechanisms that are at play in photoreceptor degeneration in the case of retinal detachment and reattachment. We used existing data to calibrate the model so that it can extend from qualitative predictions to quantitative ones using inverse problem methodology9,23. These datasets allowed us to examine the key parameters that separately affect rod OS and cone OS degeneration in the case of detachment and their growth after reattachment. For retinal detachment, our model requires knowledge of the rods and cones separately. While we were able to provide estimates of nutrient decrease based on the three datasets from the literature, the results of our mathematical model required lumping the populations of rods and cones together. To test the importance of separately counting rod and cone populations, we conducted an experiment utilizing the same detachment procedure as the previous three datasets14; see also21. Our results suggest that rods and cones do not share identical death kinetics and thus should be considered separately in any experiments. In particular, our results show that cone death progresses significantly more than rod death and suggests that the cones are not able to adapt to the detachment in the same way that the rods adapt. The overall nutrient carrying capacity, Dt = Γ/κ,that includes metabolites, oxygen, and glucose, is a key quantity whose change we considered over the course of the observed photoreceptor degeneration period following a retinal detachment. In all four curves of the death kinetics for combined photoreceptor death, see Fig. 2, we observed a decrease in Dt with a maximum change occurring when the photoreceptor death was at its peak. The percentage decrease in Dt ranged from 15.9% to 46.5%. With our estimated parameters, we conducted a sensitivity and uncertainty analysis for rods and cones separately at various time points during detachment, corresponding to early, mid-range, and late-stage interventions. These results showed that the rate of nutrient inflow into the system, Γ, and the limiting effect for the nutrient carrying capacity κ, are significant with the former positively correlated and the latter negatively correlated. The effect of these parameters is seen in the degeneration of both rods and cones. In addition, for cones the parameters µc and p (seen in µc /p: the ratio of cone energy consumption to natural uptake of glucose) are important with consider µc being positively correlated and p being negatively correlated. For rods specifically, the parameters µn and an (seen in Dn = µn /an: the ratio of energy consumption to uptake in the rods), are important with µn being positively correlated and an being negatively correlated. While Γ/κ represents the overall levels of metabolites, oxygen, and glucose going to the system, the cone-specific ratio µc /p and the rod-specific ratio µn /an can be thought of as representing the efficiency to utilize this influx. This suggests that besides ensuring that sufficient amounts of nutrients are available, that therapies can also target the efficiency by which these are utilized. An examination of the equilibrium of our mathematical model confirms results that are recurrent here: affecting nutrient carrying capacity and glucose levels has the greatest effect on the rods and cones during their decay. In terms of the ratios we can consider µc /p (ratio of cone energy consumption to natural uptake of glucose), Dn = µn /an (the ratio of energy consumption to uptake in the rods), and Dt = Γ /κ (the nutrient saturation level in the RPE), at least one parameter of each of those quantities is seen to affect either the rods or cones or both at every time point. As both rods and cones utilize the same nutrient source and are part of an interrelated system, creating targeted strategies to help one type of photoreceptor may well negatively affect the other type. Increasing Γ and an (energy uptake in rods) are the only parameters that never have a negative effect on rods or cones and thus one could increase the nutrient levels and additionally target the energy uptake in rods without adversely affecting either photoreceptor population. Taken together, our results suggest the importance of experimental work that accounts for rods and cones separately, and for potential targeted therapies that may differ at key points in time. We considered two modeling approaches: equilibrium and time-series. The fit to data was best for the equilibrium approach. We note that the time-series approach assumes that the initial detachment alters parameter values of the model and these remain unchanged thereafter. As the retina adapts over the course of the detachment, these parameters may in fact change. In particular, there was an initial decrease of the time-series data that was only matched through a significant decrease in Γ; however, leaving Γ at such a low level quickly brought the photoreceptor numbers to near zero and caused a poor fit for the data values at 10 and 14 days. Even with this observation, without specific knowledge of how these may change, we only considered them as an initial permanent change. For reattachment of the retina, we then examined cone and rod growth and looked at various time points, which could correspond to check-ups after reattachment. For cone OS growth, the results predicted that increases in the parameters p and Γ, corresponding to the natural glucose intake of cones and the influx of trophic factors, respectively, result in the biggest increase in the length of the cone OS. The results also predicted that increases in the parameters κ and µc, corresponding with the limiting effects of available trophic factors and cone energy consumption and metabolism associated with shedding, respectively, result in the biggest decrease in the length of the cone OS. These results apply for a regrowth model, which can be particularly useful in the case of reattachment and suggests that targeting nutrient supplies including availability of metabolites and oxygen, together with glucose availability will have the greatest impact on the growth of the OS. While numerous experimental studies have also reached similar conclusions, our model provides researchers with a method to quantify the level of these contributions. Additionally, our results suggest that focusing some efforts on the energy and metabolism associated with cone OS shedding may be another way to best ensure OS growth after reattachment. We note that our model was that of an eye with rods and cones present in near normal numbers and thus small changes in RdCVF do not have a significant effect. This can be seen in Eq. (3) since the term dn quantifies the RdCVF and, for near normal or comparable levels of R and C, small changes in dn do not have a significant effect. Of the 16 combinations of significance of parameters presented in Table 1, only one parameter that showed up did so less than four times (γ, corresponding to removal of nutrients, trophic factors, and glucose from the trophic pool). In other words, these results predict a consistent set of parameters that researchers can focus on for detachment and reattachment of rods and cones. Table 1Summary of significant parameters at time values of 14 days and 150 days for growth and decay parameter values.Parametervalues TimeC, Significant posC, Significant negR, Significant posR, Significant negReattachment14Γ, pµc, κan, Γµn, κReattachment150Γ, p, anµc, κ, µnµc, anµn, pDetachment14Γ, pµc, γ, κµc, an, Γµn, κ , γDetachment150Γ, pµc, κan, Γµn, κ Summary of significant parameters at time values of 14 days and 150 days for growth and decay parameter values. The results of this mathematical modeling study highlight that enhancing nutrient availability (Γ) and improving energy uptake efficiency in photoreceptors—specifically rod energy uptake (an) and cone glucose uptake (p)—are critical therapeutic targets for managing retinal detachment. Practically, this suggests that adjunctive therapies boosting nutrient delivery, such as increasing nutrients through elevated inflow (Γ), and optimizing photoreceptor metabolism, such as inducing metabolic dormancy during detachment to conserve energy and enhancing reparative metabolism after attachment, could significantly improve outcomes alongside surgical re-attachment. Immediate next steps involve developing drugs or supplements targeting these factors. Before re-attachment surgery, minimizing cell death could target glucose transporters and metabolic enzymes to enhance rod energy uptake (an) and cone glucose uptake (p), countering nutrient deprivation. Our sensitivity analysis identifies the nutrient carrying capacity (Γ/κ) and glucose uptake parameters (p, an) as significant determinants of photoreceptor survival. While the physical separation from the RPE almost certainly increases diffusion distance and thus reduces nutrient flux, our model captures this primarily through Γ. The significance of κ, p, and an suggests that even given a fixed reduction in nutrient supply, the efficiency of uptake and transport, which involves glucose transporter activity, may represent an additional modifiable factor. We acknowledge, however, that our model does not explicitly separate diffusion distance from transporter-mediated uptake, and experimental work directly measuring transporter expression and activity in detached retinas would be needed to validate this interpretation. After re-attachment, supporting recovery involves increasing nutrient inflow (Γ) and cone glucose uptake (p) by targeting retinal pigment epithelium (RPE) function and systemic nutrient delivery pathways, such as those regulating blood flow or metabolite transport, to promote photoreceptor regrowth and survival. This study employs a mathematical model that, while providing valuable insight into the nutrient- and metabolism-driven mechanisms of photoreceptor loss and recovery, is inherently limited by several factors. First, the model omits key biological contributors such as Müller glia, inflammatory responses, and spatial heterogeneity, focusing solely on the dynamics of photoreceptors and nutrient pools. As such, the quantitative predictions represent upper-bound effects under a nutrient-limited scenario and may not be fully generalizable to clinical or experimental settings where additional cell types and inflammatory processes are influential. Second, parameter estimation draws from heterogeneous sources—combining data from different species and experimental designs—which introduces significant uncertainty. We have addressed this by using broad parameter ranges in sensitivity analyses; nonetheless, results should be interpreted qualitatively rather than as precise quantitative forecasts. Third, measurement artifacts (e.g., retinal stretching post-detachment) and the lack of Day 0 normalization in some datasets may result in overestimation of early cell loss, potentially biasing parameter inference. Additionally, the model extrapolates outcomes far beyond far beyond the most dense available experimental time window (up to 150 days), and predictions in these later phases should be viewed as speculative. Lastly, the model operates at the population level, without accounting for individual cell fate variability or local microenvironmental differences. These limitations collectively underscore the need for caution in translating these findings to clinical interventions and highlight directions for future work, including the integration of additional biological factors, spatial modeling, and more refined parameter calibration using consistent, species-specific data. Mathematical models are always meant to be a starting point and numerous aspects of our model could be extended to explicitly include various factors, additional mechanisms of interaction, and even additional variables such as bipolar cells or Müller cells. A strategy that incorporates analytical and computational models together with experimental results that inform each other provide the best hope of understanding photoreceptor degeneration. Materials and methods Retinal detachment data We induced retinal detachment in 8-week old C57BL/6J mice by subretinal injection of sodium hyaluronate at different time points14. We selected 3 consecutive areas (white rectangles) of the detached (D) and attached retina (A), beginning from the pars plana (pp); see Fig. 9. The first region, closest to the pars plana, was considered peripheral retina, and the two consecutive regions were considered central retina, according to the organizational characteristics in photoreceptor population previously described24. Recent results suggest that the local cell death differences may occur in the detached murine retina21. For this reason, we include in our analysis the apex region of retinal detachment, where photoreceptor cell death occurs more markedly. A limitation of our approach in this paper could be the stretching of the retina after detachment. Comparing mean number of ONL cells between attached and detached retina can make us overestimate the death of PR due to the stretching of the detached part of the retina, and this can also affect the dramatic loss found at day 1 after detachment. This limitation was overcome in the 2021 paper by analyzing day 0 (immediately after detachment) but this current study didn’t include day 025. Additional data from14 that described TUNEL positive cells was also used. The death kinetics curves for retinal detachment via subretinal injection of sodium hyaluronate for these datasets from the mouse models C57BL/6, BALB/c, and B6129SF2 are presented in Fig. 10. For these three mouse models, the greatest loss of photoreceptors is seen at 24 h with data points taken 0, 12, 24, 48, 72, 120, and 168 h after detachment. A second limitation of this study is that only male animals were included; therefore, the findings may not fully capture sex-dependent differences in photoreceptor cell death and should be validated in females in future studies. Induced retinal detachment in C57BL/6, BALB/c, and B6129SF2 mouse models by subretinal injection of sodium hyaluronate14. The greatest loss of photoreceptors is seen on the first day. Data was taken and is shown at 0, 12, 24, 48, 72, 120, and 168 h after detachment. Animals. All animals used in experiments and breeding adhered to the statement of the Association for Research in Vision and Ophthalmology (ARVO). Animal protocols were reviewed and approved by the Animal Care Committee of the Massachusetts Eye and Ear Infirmary. All experiments were performed in accordance with institutional regulations. This study was carried out in compliance with the ARRIVE guidelines. C57BL/6J mice were purchased from The Jackson Laboratories (Bar Harbor, ME). Mice were maintained in a standard 12-hour light/dark cycle. Mice were euthanized by CO2 inhalation in an approved chamber, followed by the secondary physical method of cervical dislocation to ensure death, in accordance with institutional animal care and welfare guidelines. They were euthanized at predetermined experimental timepoints (1, 3, 7, and 10 days post-treatment). Humane endpoints were defined according to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. Animals were monitored daily for signs of distress (e.g., significant weight loss, lethargy, impaired mobility). None of the animals exhibited clinical signs requiring early euthanasia. The experiment lasted 10 days in total. Animals were euthanized at defined timepoints corresponding to their group assignment: 1, 3, 7, or 10 days after retinal detachment. A total of 32 C57BL/6J mice were used in the study. All animals were euthanized at their scheduled timepoints. No animals died spontaneously, and no unexpected deaths occurred during the course of the experiment. Animals were monitored daily by trained personnel to assess their general health status, behavior, and potential signs of pain or distress. All procedures were performed under general anesthesia using an intraperitoneal injection of 2,2,2-tribromoethanol and 2-methyl-2-butanol (Avertin; 125 mg/kg). Mice were housed in individually ventilated cages under a 12-hour light/dark cycle, with ad libitum access to food and water. Environmental enrichment was provided. No signs of pain or distress were observed during the experiment. Retinal detachment model. Eight-week-old male C57BL/6J mice were anesthetized with an intraperitoneal injection of a mixture of 2,2,2-tribromoethanol and 2-methyl-2-butanol at a dose of 125 mg/kg, as previously described26. Briefly, a Hamilton injector (Hamilton Company, Reno, NV) with a 34-gauge needle was introduced through a nasal sclerotomy into the subretinal space. 4 µL of sodium hyaluronate were gently injected to detach the neurosensory retina from the underlying RPE. The nasal sclerotomy was sealed with surgical glue. Eyes with hemorrhage, leakage, or cataract were excluded from further analysis. A topical ointment containing bacitracin was applied after the procedure26. Tissue processing and immunofluorescence. Mice were euthanized 1, 3, 7 and 10 days after retinal detachment. Eyes were enucleated immediately after euthanasia, embedded in O.C.T. (Optimal Cutting Temperature) compound and cryopreserved at – 20 °C. Cryosections of 10 μm thickness at 1000 μm distance from the first section of the specimen of each sample were cut, mounted, and dried at room temperature. Slides were frozen at − 80 °C for further processing. Sections were fixed in 1% paraformaldehyde incubated in blocking buffer containing 2% bovine serum albumin and 0.3% Triton X-100 in phosphate buffered saline (PBS) and then subsequently incubated with peanut agglutinin (PNA) conjugated with AlexaFluor-647 (1:50). Then, slides were washed with 1% bovine serum albumin and 0.3 M sodium chloride in PBS. Nuclei were stained with DAPI and samples were mounted with Fluoromount-G. Pictures were acquired with ZEISS Axio Imager M1 epifluorescence microscope. Cell death quantitation method. Eyes with less than 50% of detached retina and those with subretinal bleeding were excluded. For every eye, two regions 230 μm wide of attached retina and two of detached retina were selected, excluding peripheral and paracentral area. In all cases, the detached part of the retina was at its maximum high (reaching the lens) to reduce variability. Five sections with a separation of 100 μm were quantified for every eye. PNA positive cells were counted on a manually by two different masked observers, and ONL nuclei were counted in a semi-automated fashion with a plugin for ImageJ with minor modifications 27. Figure 11 illustrates the cone and rod counts with the method of counting having been described in our previous work21,28. Examples of counts for ONL and PNA. The method of counting has been described in previous work25. Mathematical model In order to investigate key components that can be targeted to prevent or slow down cell death in retinal detachment, we first consider the model of8 that was proposed and analyzed mathematically, but never calibrated to a dataset, in order to obtain realistic parameter values for a healthy eye. The governing equations of this system are given as (11–13): These three equations represent: R, the number of rods, T, the trophic pool/contribution of RPE/Choriocapillaris, and C the number of cones. The equations were described fully in8 but we briefly discuss them here. Parameters and variables A table for the variables and parameters in Eqs. (11–13) is given below with more detailed explanation following. Number of rods Number of cones Trophic pool / contribution of RPE/ Choriocapillaris Metabolism in cones associated with shedding Metabolism in rods associated with shedding Uptake of nutrients from RPE by rods Uptake of nutrients from RPE by cones Cellular metabolism in rods associate with their renewal Rate of unbound RdCVF synthesis Measure of carrying capacity Γ/κ Metabolism associated with the cone renewal in the absence of RdCVF Maximum rate of formation of GLUT1 Dissociation constant to permit ½ Vmax to be achieved Equation (11) describes the change in the total number of rods. The parameter an in this equation is the energy uptake and metabolism of the rods associated their outer segment (OS) renewal. The parameter µn gives the rate of energy consumption and metabolism associate with OS shedding of the rods. Equation (12) is the equation for the trophic pool, represented by RPE cell function and transport of nutrients from the choriocapillaris through the RPE, T. In this equation β and γ represent the uptake of nutrients from RPE by rods and cones, respectively. With this modeling approach representing in-flow and out-flow for a given variable, these same parameters also incorporate the role of phagocytosis of shed photoreceptor outer segments. In the absence of rods and cones, nutrient flow reaches a carrying capacity, Γ/κ. Thus, we can view Γ as the natural nutrient flow facilitated by the RPE and κ as the term that limits this inflow. Equation (13) describes the change in the total number of cones. The cone cells intake glucose at a natural rate, denoted by p; this rate of glucose intake is increased due to the effects of rod-derived factor RdCVF as described by29,30. To model this allosteric process, a Holling type III equation gives us an estimate for the glucose intake in the cones that is dependent on RdCVF. Since RdCVF production is directly proportional to how many rods are present in the eye, there is a functional relationship with glucose dependent on RdCVF because there is some point where saturation occurs in high density and at low volumes of glucose there is an accelerating use by the cones29,30. In Eq. (13), Vmax is the rate of formation of glucose transporter 1 (GLUT1) and dnR represents the total free unbound RdCVF content. Finally, the parameter km is the dissociation constant so that the RdCVF concentration is at a point that permits ½ Vmax or one half of the maximum rate of formation of GLUT1, while µc is the energy consumption and associated metabolism of the cone shedding8. The qualitative long-term behavior of Eqs. (11–13) were studied in8 with a stability analysis via steady state solutions and bifurcations by examining mathematical relationships between the various parameters without considering any specific parameter values. However, to use these equations to model the effects of retinal detachment on the parameters in the overall death kinetics of the photoreceptors, we need to have an accurate set of parameters. We can use inverse problem methodology to obtain estimates on parameters not available in the literature31. We note that the model focuses on rods, cones, and the trophic pool but excludes other retinal cells (e.g., Müller glia, immune cells) that may influence photoreceptor survival. Müller glia, for instance, provide metabolic support and mediate inflammation, which are significant in detachment-induced degeneration. The model emphasizes nutrient dynamics but we omit other detachment-induced processes, such as inflammation and oxidative stress. Additionally, the model assumes a homogeneous distribution, such as is present in the murine mouse model, but which is different than a human where the distribution is not homogeneous. Extending the model in any of these directions would be a significant contribution to the field but is beyond the scope of this current work. Incorporation of these and other factors that significantly contribute to photoreceptor loss and should be integrated in future work for biological fidelity. Equilibrium approach for detachment The mathematical model, Eqs. (11–13) describe the photoreceptor interactions in a non-diseased retina. In the case of retinal detachment, the same model can be used as it is the parameters of the model that have undergone a change. The dataset (C57BL/6J) from our experiment separately counts the rods and cones alive after detachment as previously described and can thus be directly used by our model Eqs. (11–13). However, we also examine three previous datasets on retinal detachment even though these combine the results as total photoreceptors and mark cells as TUNEL positive14. TUNEL staining identifies cells actively undergoing apoptosis at a given moment, not the cumulative number of dead cells. To convert this death-rate signal into an estimate of surviving cells, we integrate the TUNEL-positive cell counts over time and subtract from the initial population, as given in Eq. (14). This approach assumes that the TUNEL signal captures the instantaneous death rate, which when integrated yields total cell loss. To use this TUNEL data, the death kinetics curve is first fit to a sigmoidal curve so that points on the sigmoidal curve are the estimation of the number of photoreceptors that are TUNEL positive at any point in time in the experimentally measured area. The number of photoreceptors alive at any given time, P(t), can thus be calculated as where U(t) is the number of TUNEL positive cells at time t in the given area, A is the total area of the retina, P(0) is the initial number of photoreceptors alive, and tf represents the end of the experimental observations. This allows the first three datasets to be compared with the fourth by adding the outputs of our model as R(t) + C(t). It is well-known that early detection of retinal detachment is important for maximizing the chance of saving the photoreceptors; however, early detection is not always possible. Our math model output thus examines parameters that most significantly affect the rod and cone values from 0 to 150 days after detachment. While our data used to fit the model was taken for 10 days and 14 days, we extrapolate the math model to 150 days in order to give potential guidance, acknowledging that this might bias late-stage dynamics and knowing that additional modifications to parameters may be needed if and when additional data is available. Parameter estimation and the inverse problem The inverse problem methodology considers the measured data written in the form (tj, yj) for j = 1,…,n. The solution of the mathematical model is written as f(tj ,q), where the input parameters are represented by the vector q. We can then write where εj denotes the errors between the model output and observed data31. Our goal is to minimize the vector of parameters q using Generalized Least Squares (GLS) instead of ordinary least squares because the first round of analysis with ordinary least squares showed highly correlated residuals over time. Thus the GLS is solving the following equation A re-scaled version of Eqs. (11–13), whose derivation is given in the next section, are implemented in MATLAB and solved making use of the built-in ordinary differential equation solver ode23s to obtain numerical solutions and the constrained minimization scheme fmincon to solve the inverse problem. Re-scaling the R, T, C equations In numerically solving Eqs. (11–13), the system of parameters and variables values can differ by several orders of magnitude (e.g., µn ≈ 0.1 rods/day and R≈ 3.6 × 106 rods in a mouse retina). To minimize the effect of numerical errors, we rescale the equations to obtain a more balanced system. The mathematical rescaling of (11–13) is straightforward although not unique. We set where r, s, and c are the re-scaled rods, trophic, and cones, respectively. Next, the system of equations is rewritten with respect to the new variable: where we note that time is not re-scaled. Then the new parameters θ, φ, and δ satisfy Finally, we combine parameters as and substitute into (18–22), thus obtaining (23–25): where r, s, and c are the respective re-scaled R, T, C variables. In Eqs. (11–13), the parameter estimates of µn = 0.103, β = 2.50 × 10− 8, γ = 7.5400 × 10− 6, and µc = 0.119 were previously obtained9,19. The remaining parameters are obtained with inverse problem methodology by minimizing the maximum likelihood estimator (MLE) of the model output and the data31. The parameter estimates for detachment and reattachment (below) were the same for µn, β, γ, and µc as well km =0.0614, dn =7.5 × 10− 12, and Vmax = 0.375. For the decay curves in Fig. 10, the additional parameters are Γ = 2.35, an =1.1 × 10− 6, κ = 0.00023, and p = 5.824 × 10− 6. These are substituted into Eq. (22) when solved in MATLAB. We note that there are limitations to our parameter estimation as no data sources exist that quantify these parameters for the same species. We draw from both mice and primates in obtaining some of our estimates, such as Γ, an, κ, p, etc. Biological differences (e.g., rod/cone ratios and metabolic rates) between species introduce additional uncertainty. The sensitivity of the results to such uncertainty is partially addressed in the methods below. Sensitivity analysis and uncertainty analysis The sensitivity analysis and uncertainty analysis were performed on the model to determine how sensitive the dynamics of the cones and rods are to changes in the parameters. For the method of uncertainty and sensitivity analysis we used a Latin Hypercube Sampling (LHS) and Partial Rank Correlation Coefficient (PRCC), respectively, from22. We also used a uniform distribution for each parameter that is sampled, then ranked and put back into the model. Then the parameters to which the model output (i.e., cone OS, rod OS, and trophic factor) are most sensitive to were identified. These findings help us understand what can be changed to influence the model output the most. The steps of sensitivity analysis, as developed in22, are as follows: input the initial guesses of the parameters into the model to achieve an output in terms of the dependent variables in this case rods, trophic factor, and cones; input the initial guesses of the parameters into the model to achieve an output in terms of the dependent variables in this case rods, trophic factor, and cones; pick approximate sampling distribution for all parameters within the model. In this case a uniform distribution was used, but a normal or other distribution can also be used; pick approximate sampling distribution for all parameters within the model. In this case a uniform distribution was used, but a normal or other distribution can also be used; perform a Monte Carlo simulation to generate the LHS, which allows for an unbiased estimate of the mean of the model output. For this to happen the random parameter distributions are divided into N equal probability intervals. The samples are randomly chosen with at least one value chosen for each interval without replacement. This means that the full range for each parameter is sampled. N is also equal to the amount sampled and should always be at least one more than the number of parameters. This method ensures that the sampling of all parameters are independent of one another. The samples of each parameter are the components of the LHS matrix and these are put into the model with exactly N outputs from the N parameter samplings. perform a Monte Carlo simulation to generate the LHS, which allows for an unbiased estimate of the mean of the model output. For this to happen the random parameter distributions are divided into N equal probability intervals. The samples are randomly chosen with at least one value chosen for each interval without replacement. This means that the full range for each parameter is sampled. N is also equal to the amount sampled and should always be at least one more than the number of parameters. This method ensures that the sampling of all parameters are independent of one another. The samples of each parameter are the components of the LHS matrix and these are put into the model with exactly N outputs from the N parameter samplings. The PRCC of the different outputs are calculated to measure the strength of the linear association between the input and the output. The parameters that result in the most varied outcomes of the outputs are those parameters to which the model is most sensitive. Interpretation of variables: detachment vs. reattachment In retinal detachment or other conditions in which photoreceptors degenerate, such as retinitis pigmentosa (RP), the experimental data usually estimates dying or surviving cells, e.g., TUNEL assay, ONL count, or ONL thickness. In the death of a photoreceptor be it via apoptosis or necrosis it is not usually the case that the OS shortens to zero length and then the photoreceptor dies (although some shortening may occur); rather, a photoreceptor dies through one of those processes even though the OS is still present at some length (likely shorter than “normal” but also not likely of zero length). Thus, for cell death, the experimental data suggests interpreting the mathematical variables as the number of rods and cones remaining, understanding that they may not be full height. Experimental results also confirm the correlation between ONL thickness and cone cell density32. Thus, a decrease in the model variables R, C corresponds to a decrease in the overall number of rods and cones, respectively, and to a lesser extent, shorter OS. When this model is applied to detachment data a normalized output of 0.5, for example, is interpreted as half of the photoreceptor have died. For reattachment, we re-define our variables R and C to be proportions of the rod OS and cone OS at full length, where the continuous renewal and periodic shedding of the OS allows for a continuous increase and decrease in the variables. The differential equation model is an “averaged model” that gives overall numbers but does not focus on individual photoreceptors. Thus, the mathematics presents an average over the variables scaled to overall number of OS or normalized number for easier comparison. When this model is applied to reattachment data a normalized output of 0.5, for example, is interpreted as all the OS are at half their full height. These interpretations allow for the death or regrowth of an entire photoreceptor to be represented by the same model. In retinal detachment and degenerative diseases such as RP, the goal of the mathematical models can be to determine what processes can be altered to stop degeneration. As before, the data suggests that if R, C degenerate more slowly or remain constant, this will allow sight to persist for a longer time. The exact numbers of OS present vs. their exact height is ambiguous in the answer but the conclusion of “sight persists” holds regardless of interpretation. Thus, when we discuss “regrowth”, we are referring to OS height, knowing that the overall number of OS is not changing but their heights are, and assuming that OS regrowth is approximately the same for each OS. When we use the term “degeneration”, we consider cells and OS to be approximately equal and do not consider shortening of the OS to significantly affect this number. Parameter estimation for reattachment In order to calibrate our model given in Eqs. (11–13), time series data measuring the growth of OS for multiple time points is needed. In addition, our model requires separation of photoreceptors into rods and cones and not just an examination of, for example, total number of PR cell bodies in the ONL over time that is not able to demonstrate OS length of the photoreceptors. Given that our model was developed for adult organisms, we did not take into account developmental factors and thus excluded early post-natal datasets. Moreover, we chose data from reattachment experiments to estimate some of the parameters in our model, understanding that OS regrowth after reattachment is also not ideal33. As time series data is required and no mouse datasets with the above characteristics exist in the literature, we needed to use data from a different animal model (primate). However, because our model is scalable, any dataset of OS lengths is immediately scaled, so that an OS at full length has a value of 1. This re-scaling helps mitigate differences that may arise by using a non-mouse dataset. The times series data we use is from Guérin and gives reattachment data for the mean of 12 rhesus monkeys23. In this work, the retina was first surgically detached so that most of the rod and cone OS had been eliminated but not the inner segments. Following the retina was reattached and monitored over time to where the rod and cone OS had completely grown back. Data points for rods and cones were taken at 7, 14, 30 and 150 days. In order to have a sufficient number of points to use for the estimation of parameters in our current work, we fit a smooth sigmoidal curve through the experimental data points to generate 8 additional points. See Fig. 12. Model output of Eqs. (23–25) using optimal parameters plotted together with the Guerin data to see the accuracy of the fit. The orange dots are from the experimental dataset and a smooth sigmoidal curve is fit to pass through these data points23. The black crosses are interpolated points on this sigmoidal curve that are used in our parameter estimation. All but four of the parameters are the same as for the detachment data. The parameter values in common for both reattachment and detachment data are µn = 0.103, β = 2.50 × 10− 8, γ = 7.5400 × 10− 6, and µc = 0.119, km =0.0614, dn =7.5 × 10− 12, and Vmax =0.375; the additional four parameters for this regrowth data after reattachment are Γ = 5.89, an =4.909 × 10− 6, κ = 0.0002093, and p = 5.424 × 10− 6. Parameter estimation for detachment Results from the equilibrium approach, given in (4–6) suggested that those parameters are the ones that affect the system the most. For comparison, the time series approach only considered the modification of those three parameters with additional inclusion of Γ since this parameter would be impacted by a physical separation of the retina from the RPE. The same inverse problem methodology used for reattachment was done for detachment, allowing for a change in those 4 parameters. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors thank the two anonymous reviewers for their suggestions to help improve the manuscript. Author contributions EC, DV, and SW conceived of the project and modeling. EC, SW created math model. EC, SW, MG analyzed math model including parameter estimation, sensitivity analysis, and simulations. DV, LGB, DM performed experiments. All authors wrote the manuscript and interpreted the results. Funding ETC received funding from the Retina Research Foundation. LGB received financial support from the Alfonso Martin Escudero Foundation. DM received financial support from the National Institutes of Health (NIH) K08EY034935 and the Research to Prevent Blindness (RPB) Career Development Award. DV received funding from the National Institutes of Health (NIH) R01EY025362 and R21EY023079, the Yeatts Family Foundation, and the MSL Renewed Hope Foundation. Data availability All data are fully available without restriction. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement All data are fully available without restriction.