1. Evolutionary Biology
  2. Immunology and Inflammation

Male rat leukocyte population dynamics predict a window for intervention in aging

  1. Hagai Yanai  Is a corresponding author
  2. Christopher Dunn
  3. Bongsoo Park
  4. Christopher Coletta
  5. Ross A McDevitt
  6. Taylor McNeely
  7. Michael Leone
  8. Robert P Wersto
  9. Kathy A Perdue
  10. Isabel Beerman  Is a corresponding author
  1. Epigenetics and Stem Cell Unit, Translational Gerontology Branch, National Institute on Aging, United States
  2. Flow Cytometry Core, National Institute on Aging, United States
  3. Computational Biology and Genomics Core, Laboratory of Genetics & Genomics, National Institute on Aging, United States
  4. Comparative Medicine Section, National Institute on Aging, United States
https://doi.org/10.7554/eLife.76808
Share this article
Cite this article
  1. Hagai Yanai
  2. Christopher Dunn
  3. Bongsoo Park
  4. Christopher Coletta
  5. Ross A McDevitt
  6. Taylor McNeely
  7. Michael Leone
  8. Robert P Wersto
  9. Kathy A Perdue
  10. Isabel Beerman
(2022)
Male rat leukocyte population dynamics predict a window for intervention in aging
eLife 11:e76808.
https://doi.org/10.7554/eLife.76808
  2. Download BibTeX
  3. Download .RIS

Abstract

Many age-associated changes in the human hematopoietic system have been reproduced in murine models; however, such changes have not been as robustly explored in rats despite the fact these larger rodents are more physiologically similar to humans. We examined peripheral blood of male F344 rats ranging from 3 to 27 months of age and found significant age-associated changes with distinct leukocyte population shifts. We report CD25+ CD4+ population frequency is a strong predictor of healthy aging, generate a model using blood parameters, and find rats with blood profiles that diverge from chronologic age indicate debility; thus, assessments of blood composition may be useful for non-lethal disease profiling or as a surrogate measure for efficacy of aging interventions. Importantly, blood parameters and DNA methylation alterations, defined distinct juncture points during aging, supporting a non-linear aging process. Our results suggest these inflection points are important considerations for aging interventions. Overall, we present rat blood aging metrics that can serve as a resource to evaluate health and the effects of interventions in a model system physiologically more reflective of humans.

Editor's evaluation

This paper uses flow cytometry to characterize the changes in immune cell composition of the peripheral blood, as well as DNA methylation, during aging in male rats. Using this data, the authors were able to observe distinct cell composition and DNA methylation profiles with age, providing predictive measures of aging. Additionally, the authors identified a novel marker, CD25+ T cell frequency, as a predictor of age in rats. This resource will be useful to the community as a well-controlled dataset dissecting changes in circulating blood cells with aging in an important mammalian model.

https://doi.org/10.7554/eLife.76808.sa0

eLife digest

Our blood contains many types of white blood cells, which play important roles in defending the body against infections and other threats to our health. The number of these cells changes with age, and this in turn contributes to many other alterations that happen in the body as we get older. For example, the immune system generally gets weaker at fighting infections and preventing other cells from developing into cancer. On top of that, the white blood cells themselves can become cancerous, resulting in several types of blood cancer that are more likely to happen in older people.

Many previous studies have examined how the number of white blood cells changes with age in humans and mice. However, our understanding of this process in rats is still poor, despite the fact that the way the human body works has more in common with the rat body than the mouse body.

Here, Yanai, Dunn et al. have studied samples of blood from rats between three to 27 months old. The experiments found that it is possible to accurately predict the age of healthy rats by measuring the frequency of populations of white blood cells, especially a certain type known as CD25+ CD4+ cells. If the animals had any form of illness, their predicted age deviated from their actual age. Furthermore, while some changes in the blood were gradual and continuous, others displayed distinct shifts when the rats reached specific ages.

In the future, these findings may be used as a tool to help researchers diagnose illnesses in rats before the animals develop symptoms, or to more easily establish if a treatment is having a positive effect on the rats’ health. The work of Yanai, Dunn et al. also provides new insights into aging that could potentially aid the design of new screening methods to predict cancer and intervene using a model system that is more similar to humans.

Introduction

Among all mammalian tissues, blood is perhaps the easiest to collect in relatively large quantities for various advanced analyses, with modest discomfort to the donor. This has made blood the prevailing tissue for a wide variety of applications requiring a sizable amount of material, such as: clinical diagnosis, immune surveillance, and molecular investigations, including DNA methylation (Fahy et al., 2019; Levine et al., 2020).

To date, peripheral blood aging has been largely characterized in humans (Márquez et al., 2020; Mahlknecht and Kaiser, 2010; Wayne et al., 1990) and mice (Pinchuk and Filipov, 2008). Studies in mouse models have provided important insights as this model system allows investigation of the impact of a variety of genetic interventions and hematopoietic stem cell (HSC) transplant assays. Many murine aging phenotypes accurately reflect reports from human studies including a decline in immunogenic response of lymphoid cells (Pinchuk and Filipov, 2008), decreased CD4/CD8 T cell ratio (Nikolich-Zugich, 2008), and an increase in myeloid cells at the expense of lymphoid cells (Pang et al., 2011). These phenotypes have been shown to be influenced, at least in part, by intrinsic HSC aging (Beerman et al., 2010).

However, the mouse has drawbacks as a model of human hematopoiesis including the fact that the common laboratory strains do not reliably develop spontaneous blood cancers, which is a common aging phenotype in humans (Zjablovskaja and Florian, 2019). The differences between the two species are further exemplified by the fact that mouse blood is dominated by lymphocytes whereas human blood is rich in neutrophils (Mestas and Hughes, 2004); which in the murine model may dampen the impact of the age-associated decrease in lymphoid potential.

An attractive alternative system is the rat model. Rat physiology is more analogous to humans (Blais et al., 2017; Gibbs et al., 2004), yet these rodents still have relatively short lifespans (Turturro et al., 1999) in which to study aging phenotypes. Furthermore, the rat and human genomes share genes involved in immunity and hematopoiesis absent in mice (Gibbs et al., 2004). Importantly, rats display age-related incidence of leukemia, especially the Fischer 344 CDF (F344) strain (Ward and Reynolds, 1983). All the above suggest the rat may be better suited to study the aging of the blood and immune systems. Despite these advantages, use of the rat model is less common for hematopoietic studies as mice are smaller, generally cost less to maintain, and transgenic tools to manipulate the murine genomes have been extensively developed. As a result, the impact of aging on the blood compartment in rats is not well characterized.

We sought to describe changes in the composition of the peripheral blood during aging in the F344 male rat using flow cytometry. Additionally, we found that CD25+ T cell frequency is a novel marker for predicting aging, and rats have defined age-associated inflection points linked to altered methylation profiles and cell composition. These results indicate that age-associated blood phenotypes in rats provide relevant insight into human blood aging and point to a critical time period of hematopoiesis which may best be targeted for anti-aging interventions.

Results

Aging patterns in rat peripheral blood leukocyte populations

We analyzed the peripheral blood composition of 146 male F344 rats ranging in age from 3 to 27 months, using a panel of seven monoclonal antibodies to evaluate major leukocyte populations. As cell-specific gating strategies for rat blood populations are less defined than in humans or mice, we initially evaluated the flow cytometry results in an unbiased manner using t-distributed stochastic neighbor embedding (t-SNE) and self-organizing maps analyses. Cells generally clustered in a manner that allowed for identification of defined populations (Figure 1a and Figure 1—figure supplement 1) and clustering analysis helped inform decisions for gating strategies we devised for cell-type classification (Figure 1—figure supplement 2). Age-related changes were also profiled from single-cell analyses and revealed an accumulation of cells in clusters corresponding to myeloid lineage cells and a reduction of cells in clusters expressing markers associated with B and CD8+ T cells (Figure 1b). Globally CD4+ T cells decreased with age, but we identified two distinct CD4+ populations with different trajectories during aging. The CD4+ T cell cluster expressing high levels of CD25 did not show an age-associated decline in cell number and instead increased in frequency over time. t-SNE-defined clusters 5 and 9 (brown and cyan, Figure 1a) also show age-associated increases in frequency (same top left location as Figure 1a, but blue in Figure 1b); however, these clusters could not be fully identified from this combination of antibodies (Figure 1—figure supplement 2). Cluster 5, due to a high expression of CD11b/c, is most likely a type of myeloid cell, whereas cluster 9 most likely consists of debris (Figure 1—figure supplement 1) as it stained equally positive for all antibodies. Most identified cell populations that were categorized encompassed two or more clusters, indicating potential subpopulations we were unable to define with the panel of antibodies in this study.

Figure 1 with 2 supplements see all
Download asset Open asset
Unsupervised clustering of all F344 rat peripheral blood cells.

Clustering of rat peripheral blood using a self-organizing map algorithm. For detailed clustering also see Figure 1—figure supplement 1. (a) Overall clustering of all leukocytes from 146 rats overlaid on a t-SNE map; each cluster is denoted as a different color (see legend). (b) Specific blood cell types illustrated on the same t-SNE map for three age groups (n = 10,000 cells from each age group; total 30,000 cells). For visualization purposes, rats were divided into age groups: young (3–4 months), middle-aged (15–23 months), and old (24+ months).

We next sought to determine if age-specific blood parameter characteristics could be defined using frequencies of the distinct cell populations identified (Figure 1—figure supplement 2). Using unsupervised clustering of all rats based on leukocyte population frequencies, animals clustered into three discrete groups (Figure 2), indicating the overall fingerprint of blood composition is age dependent. The cluster classified as ‘young’ is characterized by a high proportion of lymphocytes in general, and T cells in particular, but with a low frequency of CD25+ CD4+ T cells.

Figure 2
Download asset Open asset
Distinct age clustering of rats based on white blood cell frequencies.

Hierarchical agglomerative clustering with distance matrix by Ward’s method. The top lines represent the distance matrix for the individual rats, and the three major branches are annotated by green, dark gray, and orange. The heatmap represents the normalized frequency of each population as indicated by the legend. The lines on the left indicate the distance matrix of the frequencies. The bottom histogram represents the age distribution for each of the major clusters in the same colors as above.

In contrast, the ‘old’ cluster, on the far right, is characterized by a low overall lymphocyte frequency, and elevated proportions of both myeloid and CD25+ CD4+ T cells (Figure 2). This increase in CD25+ CD4+ T cells is similar to what is seen in human peripheral blood aging (Gregg et al., 2005), while in the murine model CD25+ CD4+ T cells only accumulate in secondary lymphoid organs (spleen and lymph nodes) but not in the peripheral blood (Chiu et al., 2007). This analysis also defines a specific ‘middle age’ profile that is distinct from both young and old phenotypes, but clusters more closely to the young rat profile than the old.

Peripheral blood parameters predict age and pathology in rats

To assess the blood cell types most predictive of aging, we fit all leukocyte frequencies in a standard least square model to predict aging in animals with no detected pathology and then eliminated the least significant parameters in a backstep fashion (Figure 3a, Figure 3—figure supplement 1). The health status of all rats was recorded throughout the study, and necropsy was performed on all animals, less than a month post blood collection (for full health/pathology report see Source data 1). This enabled testing of how the health status of each rat fit into the model. One caveat was that moribund animals were excluded and euthanized per vet recommendation, so the health analysis captures mostly underlying illnesses only discernable during necropsy. The resulting blood parameter model predicted age with an approximate 2-month margin of error (as indicated by the root mean square error value). The single most predictive parameter of rat age was the proportion of CD25 expression in CD4+ T cells (Figure 3a, right panel). Animals with histological indication of disease, including indications of large granular lymphocytic leukemia (LGLL) in spleen or liver, described in Finesso et al., 2021, are generally predicted by our model as more variable than their actual age (i.e. have stronger residual age values) (Figure 3b). When constructing an age prediction model only for ill animals, we found the myeloid/lymphoid ratio was the strongest contributor to age prediction (Figure 3d). Additionally, we constructed a nominal logistic regression model to predict illness and found that myeloid/lymphoid ratio is the strongest contributor, even more so than age (Figure 3—figure supplement 1). While the myeloid/lymphoid ratio increases with aging, this phenomenon is more pronounced in sick animals. However, this ratio alone is not sufficient to predict illness (Figure 3d, left) and must be combined with the other blood parameters that are characteristic of age and how variable the predicted age is from the chronologic age (Figure 3b and c). We also observed that myeloid/lymphoid ratio increases concurrently with severity of LGLL (Finesso et al., 2021; Figure 3d, right).

Figure 3 with 2 supplements see all
Download asset Open asset
Age and illness prediction by blood leukocyte composition.

All blood cell population frequencies were used to create a standard least square model of age prediction. Variables that did not contribute to the model were removed in a backstep fashion. (a) Predicted age vs. actual age, the red line indicating the mean and fit. Animals excluded due to illness are depicted in light gray. The effect summary for the variables used is depicted in the right panel (ffect Summary) with the Logworth cutoff depicted by a blue line. (b) Residual age (i.e. the difference between actual age and predicted age) plotted as absolute mean values against assessment of health as determined by necropsy. Yes = no health issues; Mild = mild health problems (e.g. minor foot lesions); No = clear health issues found during necropsy (Source data 1). (c) The correlation between myeloid/lymphoid ratio and age in healthy (blue) and ill (red) animals. (d) Myeloid/lymphoid ratio in the peripheral blood plotted against health status (left panel) and large granular lymphocytic leukemia (LGLL) pathology score as detected in the liver and spleen. Each dot represents a single rat. Dot color denotes age.

As blood samples from the 146 rats were analyzed post-fixation (see Methods), we were interested to determine whether the model generated is also applicable for freshly isolated blood. We thus generated data for both fresh and fixed rat blood, taken from the same donors. We found only small differences (<10%) in population frequencies associated with fixation, with changes in the T cell and B cell population frequencies post-fixation showing the biggest variation (Figure 3—figure supplement 1a). Despite these differences, applying the age-prediction model, generated from fixed blood samples, on the blood profiles generated on unfixed samples of varying ages (Figure 3—figure supplement 1b) demonstrated that the existing model (from fixed blood) could be used in a similar fashion to predict age with only minor mathematical adjustments.

As this study was primarily focused on male rats, we wanted to evaluate if the model could be used in a sex-independent manner. We collected a small dataset of 5 young (5 months), 5 middle-aged (15 months), and 5 old (23 months) females and compared the blood parameters to the males of the same age. We observed at the three age groups evaluated, females present sex-specific differences in leukocyte populations with notably lowered B cell and T cell percentages, and a higher frequency of neutrophils (Figure 3—figure supplement 1c). Interestingly, we also observed sex-specific aging trends that exacerbated the T cell and neutrophil differences. Surprisingly, despite the sex-specific phenotypes, the male generated model still robustly predicted female age (p<0.0001) (Figure 3—figure supplement 1d).

Regression analysis reveals potential aging intervention points

This study involved a minimum of five rats from each month of age starting from 3 months to 27 months (Source data 2) which enabled us to cross-sectionally examine the dynamics of blood changes during aging. Overall, the changes in leukocyte frequencies of age-matched rats, reflected described changes in aging human blood, including a reduction in B cells and T cells, and an increase in monocytes, neutrophils, and CD25+ CD4+ T cells (Figure 4—figure supplement 1). However, while the ratio of CD4/CD8 in T cells initially declined, it increased after 15 months of age. To better understand the dynamics of these blood composition shifts, we performed a multiple regression analysis for each cell population and identified timepoints in which there was a major trajectory shift (Figure 4a and Figure 4—figure supplement 1b). Most of these shifts appear to converge at approximately 15 months of age (Figure 4a and b). A strong age-associated increase in the overall variance of the different parameters was also observed, which was especially pronounced at around 24 months of age (Figure 4c). Interestingly, the first hematopoietic inflections appeared prior to the decline in survival of the classic Gompertz curve for this rat strain (Turturro et al., 1999).

Figure 4 with 1 supplement see all
Download asset Open asset
Dynamics of leukocyte frequencies during aging.

(a) Aging trends of variables with inflection points depicted by a cubic spline regression (λ = 0.521). Shaded colored areas indicate the fit. Inflection/breakpoints are depicted as dotted vertical lines with SE(Standard error) in shaded gray. (b) Summary of leukocyte critical aging points (color coded) in relation to reproductive capacity (orange) and survival curve points (gray) (Turturro et al., 1999). (c) Variance (S2) and coefficient of variation for each parameter as a function of age (in a sliding window of 3 months).

Peripheral rat blood DNA methylation changes with age

DNA methylation data was generated from the peripheral blood of these rats and was previously analyzed to define a rat aging ‘methylation clock’ (Levine et al., 2020). We analyzed the epigenic landscapes for hematopoietic-specific changes in methylation associated with aging using the Bismark pipeline. Global, age-dependent hypomethylation (Figure 5a) occurred, located mostly in intergenic and intronic regions. Promoter regions had a more balanced ratio of sites that either gained or lost methylation (Figure 5b). Interestingly, overall decreases in methylation were more pronounced in the transition between middle and old age compared to the global methylation changes seen in the shift from young to middle age (Figure 5a). To gain insight into the difference between the early-aging and late-aging changes, we performed differentially methylated region (DMR) analysis on locations near transcription start sites (TSS). Most regions with significant DNA methylation changes were progressively altered during aging, which is slightly incongruous with the phase shifts seen in blood composition. However, we also found non-linear changes in methylation using a differential methylation analysis of three major age groups (Figure 5). We identified 140 DMRs unique to the young vs. middle age comparison and 94 DMRs unique to the middle-aged vs old (Figure 5c). 456 DMRs appeared as significantly differential only between the young and old animals, which we attribute to a slow progressive change that only passed the cutoff of significance and fold change in the young vs old comparison (Figure 5c). Enrichment analysis of DMRs showed age-associated hypermethylation for several processes and included some overlaps between the early (Y-M) and late (M-O) aging phenotypes, such as regulation of T cells and differentiation (Figure 5d). However, some enrichments were unique to the middle age transition (regulation of apoptosis) and the late age (notch signaling pathway), again indicating a difference between the early and late aging phenotypes. Of note, while hypomethylated DMRs between young and middle age were identified for several processes (Figure 5d, right panel), hypomethylated DMRs in the transition from middle-aged to old were not enriched for any specific process (Source data 3); perhaps suggesting that these changes are more reflective of drift or random alterations and not necessarily concerted DNA methylation changes. To assess if the observed DNA methylation aging changes are specific to rats or might be more conserved, we compared our results to previously published data from mice (Sziráki et al., 2018) and humans (Hannum et al., 2013). Significant age-associated methylation changes in whole blood from both human and rat were predominantly hypomethylation; but in mice the distribution between hypomethylation and hypermethylation is more balanced (Figure 5a and f). When comparing the DMRs observed in rats to other species (Figure 5g) we find a high level of concordance, suggesting that many age-associated methylation changes may be common to several mammalian species. However, it is important to note that the sequencing was performed with different techniques for each study, resulting in a difference of coverage of orthologous loci (Figure 5g, right panel).

Figure 5 with 1 supplement see all
Download asset Open asset
Differentially methylated regions (DMR) with rat aging.

(a) Global differential methylation histograms per age group comparison. A cutoff of 5% methylation difference was used. (b) Proportion of hypermethylated and hypomethylated regions (DMR) compared to distance from the closest transcription start site (TSS). (c) Venn diagram for the number of DMR within 20 k bp distance of the nearest TSS based on age group comparisons. A cutoff of false discovery rate (FDR) <0.05, >5% CpG mean change, and at least 3 CpG differential DMR site per block was used. Hypermethylated and hypomethylated changes are indicated by the noted colors. (d) Summary of gene set enrichment analysis for DMRs in proximity of TSSs (<20 k bp). Analysis was performed for three age groups: young (3–8 months), middle-aged (13–17 months), and old (22–27 months). Combined score was calculated as −log10(FDR) × fold enrichment. YM - young vs. middle-aged; MO - middle-aged vs. old; YO - young vs. old. The list of hypomethylated DMRs for the MO comparison had statistically significant enrichments. For a full enrichment analysis see Table S4. (e) Individual methylation levels of regions near the Runx1 and Dnmt3a genes. Left panel - Runx1; middle panel - methylation map in the Runx1 TSS; right panel - Dnmt3a. (f) Global differential methylation histograms of mouse (Sziráki et al., 2018) and human (Hannum et al., 2013). A cutoff of 5% methylation difference was used. (g) Heatmap depicting concordance of DMRs near orthologous gene sites between rat (left), mouse (center), and human (right) datasets. Only genes identified as differentially methylated in old rats are shown, the rightmost panel depicts the coverage for each dataset, where missing data is black.

To eliminate a potential bias of predetermining age groups, we also analyzed DNA methylation at all TSS to determine if changes occurred as a function of age using the response screening platform (see Experimental Procedures) (Figure 5—figure supplement 1a). The most significantly changed sites were all located near the TSS of RUNX family transcription factor 1 (Runx1), and these sites all show strong hypermethylation with age (Figure 5e, left panel). The hypermethylation of Runx1 can be mapped immediately distal to the TSS (Figure 5e, middle panel). The second most significantly changed DMR was located near the TSS of Dnmt3a (Figure 5e, right panel). A principal component analysis based only on the 2311 DMRs that significantly changed as a function of age (Figure 5—figure supplement 1b) demonstrates a marked increase in variation around the 2-year mark, analogous to the increased variation observed for population frequencies (Figure 4c).

Early and late life DNA methylation switches

As the blood population frequencies indicated a mid-life inflection, we explored whether DNA methylation patterns exhibited similar signature transitions. First, we performed both linear and logistic regressions on significant age-associated DMRs near TSSs (Figure 5—figure supplement 1a and Figure 6—figure supplement 1). We next selected the regression with the best fit (based on Akaike Information Criterion, Shavlakadze et al., 2019). 62.9% of the sites changed linearly with age while 37.1% showed a better fit to a logistic regression (Figure 6—figure supplement 1). The breakpoint was calculated for sites that did not change linearly with age (Figure 6a), and a module enrichment analysis was performed for genes with switches before and after 15 months of age (Figure 6b). The methylation switch points defined converged at around 22 months of age and mildly at close to 12 months of age. Genes that had a breakpoint earlier than 15 months of age were enriched for cytokine production, regulation of differentiation and hematopoiesis, and include Apolipoprotein E (Apoe) and Runx1. Genes identified to have a switch later in life were enriched for myeloid differentiation, leukocyte activation, ion transport, and epithelial development. Similar to the changes in blood parameter composition, these DNA methylation results highlight differences come in waves, notably at around 15 months, rather than occurring in a linear fashion and suggest a key remodeling timepoint of the hematopoietic system.

Figure 6 with 1 supplement see all
Download asset Open asset
Early and late life methylation breakpoints.

(a) Distribution of breakpoints by age. Each dot represents the breakpoint of a single gene transcription start site; color scales by age from young (blue) to old (red). Contour violin plot of the distribution is displayed in gray. (b) Functional module networks of the genes with breakpoints before (bottom - blue) and after (top - red) 15 months of age. Each dot represents a single gene with size corresponding to its connectivity in the network. The lines indicate interactions (edges). The networks were built for blood tissue, with human orthologs as input. The interaction network is built using the closest gene neighbors and then clustered based on enrichment in GO(Gene Ontology) categories. Networks were generated using HumanBase (https://hb.flatironinstitute.org).

Discussion

Although the rat peripheral blood compartment has been analyzed in regard to aging (Flaherty et al., 1997), newer panels of monoclonal antibodies offer improved insight into changes of the hematopoietic system. Here, a relatively simple blood profiling technique using eight antibodies led us to construct a model that predicts both age and, potentially, pathology. The model can be used to predict the risk of illness as a function of the deviation from predicted age to actual age, and we propose such a model could also be used to determine if intervention experiments mitigate aging phenotypes. While our dataset and analyses were for a specific set of flow cytometric parameters generated on fixed cells, we found we were able to adapt these analyses to also predict the age of freshly isolated blood (using similar flow cytometry markers). Thus, with inclusion of proper controls to account for experimental design variation, this resource of aging rat blood parameters could be adapted for studies of rat aging or pathology without requiring euthanasia of the animals. We believe this would be especially useful for longitudinal or intervention studies given the ease of blood collection at multiple timepoints. It would also be interesting to see how our reported blood parameter changes synergize with other non-invasive measurements such as the frailty index, gait speed, etc.

This experiment was performed in male rats. In humans, clear sex aging differences have been defined, both before and after menopause (Márquez et al., 2020; Gubbels Bupp, 2015; Huang et al., 2021). In this study, we were surprised to observe a good fit of females to the male generated age prediction model. However, we only tested a relatively small sample of 15 females of varying ages. Additionally, we found sex-specific leukocyte frequency differences that were accentuated by age. These differences highlight potential sex-specific aging patterns that remain to be fully explored. This comparison would be especially intriguing if viewed in the context of aging of the reproductive system as in the males we observed an alignment between hematopoietic aging and the end of the late reproductive period (Figure 4b).

A strong predictor of age was the accumulation of CD25 on CD4+ T cells in the F344 rat model. These results support previous findings that report an age-dependent accumulation of regulatory T cells (Tregs) (Chiu et al., 2007; Lages et al., 2008; Sharma et al., 2006). This stable increase in the F344 rat is more similar to that observed in humans (Gregg et al., 2005) than in mice (Chiu et al., 2007) and highlights the relevance of the rat model to study hematopoietic aging. Garg et al., 2014 have previously suggested that this accumulation is driven by hypomethylation of forkhead box P3 (Foxp3) in CD25+CD4+ T cells; however, in our dataset analyzing total blood cell methylation, we do not see differential methylation of Foxp3. This is likely because CD25+CD4+ T cells are a rare subpopulation of the total blood, and DNA methylation changes in this population would not significantly affect the global blood methylation profiles.

We observed an age-dependent myeloid bias with rat aging, similar to mice (Beerman et al., 2010) and humans (Pang et al., 2011). The observed myeloid bias was a strong predictor of pathology in general, and of LGLL in particular. Importantly, strong myeloid bias was correlated with illness that was otherwise only noticeable during autopsy and undetectable during routine veterinary observation. A caveat to this observation is that myeloid bias increases with age, even in healthy animals, meaning that the myeloid/lymphoid ratio can only be useful as a diagnostic parameter when plotted against the expected ‘normal’ age-dependent increase. In this reference dataset, we found a significant association between severity of LGLL and increased myeloid cell numbers. The myeloid bias we report in rats has previously been documented in mice; however, most mouse models do not develop age-associated blood diseases seen in humans. Thus, we find the correlation between increased myeloid cells and LGLL severity in rats to be of particular relevance for potential modeling of human blood aging phenotypes.

Our analysis of the age-related changes in leukocyte profiles indicates inflection points in the trajectories occurring at early middle age (~15 months) and at ~24 months of age. The first inflection point is characterized by several shifts in leukocyte frequencies, including a stabilization of T cell decline, peak in CD8+ T cells, rapid drop of B cells, deceleration in the accumulation of CD25+ CD4+ T cells, and an acceleration in the increase of stimulated monocytes. The second shift point we observed is defined by a marked increase in variance between individual rats which alludes to a pan-hematopoietic loss of homeostasis (Figure 4c). Alternatively, this increased variability could be the result of a divergence in biological aging rates between individual rats. It would be interesting to investigate this further in a longitudinal study that can make a clear distinction between these postulations. We hypothesize that interventions aiming at mitigating aging phenotypes would be significantly more effective if performed prior to these defined shift points. The inflection points in the blood are similar to the transition pattens (linear, early logistic, and mid-logistic) of gene expression on tissues isolated from aging Sprague-Dawley rats (Shavlakadze et al., 2019). Furthermore, supporting a claim for interventions predating the observed switch point are studies of caloric restriction (CR). It is quite clear that early onset CR has more benefits than late onset, but it is yet unclear what the ideal starting age should be (Ingram and de Cabo, 2017). In fact, late onset CR has even been shown to be potentially detrimental to cognitive function (Todorovic et al., 2018). However, Chen et al. (Chen et al., 2015) reported a beneficial effect of late onset CR to muscle mass and metabolism in Sprague-Dawley rats, indicating that the ideal points of intervention could vary greatly between different systems. In fact, interventions that aim to prevent or delay aging could be fundamentally different than those that aim to reverse it, even to the extent where harm could be caused when one strategy is applied instead of the other (Hadley et al., 2005). It is therefore critical to understand which systems display aging switch points, and when, in order to design the appropriate intervention regimen. This is especially true when translating experimental paradigms to human interventions due to the great heterogeneity between the aging rates of both individuals and different systems within the individual (Gonzalez-Freire et al., 2020).

The combined results of changes in population frequencies and the DNA methylome indicate that hematopoietic aging of rats occurs in phases as opposed to a continuous process. However, with the current data, we cannot ascertain the drivers underlying the shift between these aging phases, whether they are triggered by a specific event, or simply manifest when a critical mass of small changes reaches a threshold. Whichever the case, investigating these relatively early timepoints in which blood profiles begin to irreversibly shift, indicating critical trajectory alterations, is of great interest.

The age-related DNA methylation changes in all leukocytes highlight the potential involvement of epigenetic changes that have occurred in HSCs and/or early progenitors, and are transmitted to all the differentiated progeny, as indicated by the hypermethylation of Runx1 and the gene enrichment analysis. In addition, the lack of enrichment for hypomethylated TSS in the transition from middle age to old indicates that age-related gains of methylation may be a directed process, whereas loss of methylation later in life is more non-specific and perhaps attributed to drift. If so, mitigating the seemingly random loss of methylation would be an interesting potential aging intervention. Intriguingly, there is a surprising concordance between the rat and human differential methylation patterns, both in terms of a predominant global loss of methylation with age and in specific DMRs that are near orthologous genes. However, a more comprehensive coverage of overlapping methylation sites is needed to validate these initial findings. As we observed a change of leukocyte frequencies with age consistent between species, and different cell types present different DNA methylation profiles, it would be interesting to ascertain in future studies to what extent the observed methylation change rely on distinct changes in blood composition. However, we observed different breakpoints for the population frequencies and the DNA methylation, which indicate a slight disconnect between the two. This supports the notion that at least some of the methylation changes are derived from aging alterations in stem/progenitor cells that are inherited by their progeny (Beerman et al., 2013).

In summary, we present this work as a resource for investigators studying aging in the rat model, which appears to be a robust system for modeling the aging human hematopoietic system. Our research emphasizes the importance of studying changes in homeostasis during aging as we present key trajectory shifts in blood populations that occur at relatively early age (15 months). Finally, our research model describes aging ‘juncture points’. Given these specific breakpoints in composition of the hematopoietic system, we posit that strategic aging interventions would likely be more robust and beneficial if performed earlier in life to mitigate or stall the major changes that occur under homeostatic conditions. While our data support that the male-derived model can be used for female predictions, it was evident there are sex-specific differences in the blood of rats. We hope these data will be expanded upon in a female longitudinal study, in which fraility measurements would also be included. It would be exciting to further explore if an illness prediction model based off of these defined blood parameters could be extrapolated for a broad range of diseases, and ultimately if similar age-associated shifts occur in other organs/tissue systems, animal models, and specifically in humans.

Methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
Strain, strain background (Rattus Norvegicus)Fischer 344 male and female (F344-cdf)NIA aging colonyF344-cdfHoused at the Charles River Laboratories (Frederick, MD)
Antibody(Mouse monoclonal) Anti-rat CD3-FITC, Clone 1F4BioLegendCat#201,403(1:200)
Antibody(Mouse monoclonal) anti-rat CD25-PE, clone OX-39BioLegendCat#202,105(1:200)
Antibody(Mouse monoclonal) anti-rat CD8a-PerCP, clone OX-8BioLegendCat#201,712(1:200)
Antibody(Mouse monoclonal) anti-rat CD11b/c-PE-Cy7, clone OX-42BioLegendCat#201,818(1:200)
Antibody(Mouse monoclonal) anti-rat CD4-APC-Cy7, clone W3/25BioLegendCat#201,518(1:200)
Antibody(Mouse monoclonal) anti-rat RT1B- AF647, clone OX-6BD BiosciencesCat#562,223(1:200)
Antibody(Mouse monoclonal) anti-rat CD45RA-BV421, clone OX-33BD BiosciencesCat#740,043(1:200)
Antibody(Mouse monoclonal) Anti-rat CD45-BV605, clone OX-1BD BiosciencesCat#740,515(1:200)
commercial assay or kitIMMUNOPREP Reagent SystemBeckman CoulterCat# 7546999Compatible with the
COULTER TQ-Prep workstation
software, algorithmTrim GaloreBabraham Bioinformatics, Krueger et al., 2021https://github.com/FelixKrueger/TrimGalore
software, algorithmCutAdaptNational Bioinformatics Infrastructure Sweden, Martin, 2022https://github.com/marcelm/cutadapt
software, algorithmBismarkBabraham Bioinformaticshttps://www.bioinformatics.babraham.ac.uk/projects/bismark/
software, algorithmMethylKitBioconductorhttps://www.bioconductor.org/packages/release/bioc/html/methylKit.html
software, algorithmDMRseqBioconductorhttps://www.bioconductor.org/packages/release/bioc/html/dmrseq.html
software, algorithmClusterProfilerBioconductorhttps://bioconductor.org/packages/release/bioc/html/clusterProfiler.html
software, algorithmHomer packageUCSDhttp://homer.ucsd.edu/homer/
software, algorithmJMPJMP-SASVer16https://www.jmp.com
software, algorithmFlowJoBD Life SciencesVer10.8https://www.flowjo.com/
software, algorithmCytobankBeckman-Coulterhttps://www.beckman.com/flow-cytometry/software/cytobank-premium

Animals

All experimental procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals and approved by the NIA(National Institute on Aging) Animal Care and Use Committee (467-CMS-2018 and 469-TGB-2022). Male and female Fischer 344 CDF (F344) rats (were obtained from the NIA Aged Rodent Colony housed at the Charles River Laboratories [Frederick, MD]). The animals were housed with Nylabone supplementation and ad libitum access to food (Envigo 2018åSX) and water. Rats younger than 3 months were housed in groups of three; all other rats were single housed. All rats were maintained on a 12/12 lighting schedule, with all procedures carried out during the light cycle. Rats were habituated to the facility for at least 3 days before sample collection. Some of rats reported here were also analyzed in previous studies (Levine et al., 2020; Finesso et al., 2021).

Flow cytometry and sample collection

500 µl of whole blood was collected via retro-orbital bleedings for DNA and FACS analysis. Blood for DNA for DNA methylation was collected in heparinized tubes, spun, and the plasma removed; buffy coat and red blood cells were frozen at −80°C until DNA extraction. Blood for FACS analysis was collected in EDTA-treated tubes, chilled on ice, and 100 µL was stained and then processed using a Beckman Coulter TQ-prep (fixation step) and the Beckman Coulter immunoprep reagent system. For the fresh sample analysis, blood was drawn as above, ACK treated, and immediately stained with antibodies. For all samples, staining was performed on ice for 30 min in a 100 µl staining cocktail containing the following antibodies at a dilution of 1:200: FITC-conjugated anti-rat CD3 (clone 1F4, Cat#201403), PE-conjugated anti-rat CD25 (clone OX-39, Cat#202105), PerCP-conjugated anti-rat CD8a (clone OX-8, Cat#201712), PE-Cy7 conjugated anti-rat CD11b/c (clone OX-42, Cat#201818), APC-Cy7 conjugated anti-rat CD4 (clone W3/25, Cat#201518) from Biolegend (San Diego, USA), and AF647-conjugated anti-rat RT1B (clone OX-6, Cat#562223), BV421-conjugated anti-rat CD45RA (clone OX-33, Cat#740043), and BV605-conjugated anti-rat CD45 (clone OX-1, Cat#740515) from BD Biosciences (Franklin Lakes, USA). Immunophenotyping data was acquired on a BD FACSCanto II and analyzed using FlowJo (https://www.flowjo.com/). Original FCS files are available on http://flowrepository.org/id/FR-FCM-Z59K.

DNA methylation analysis

Samples were treated and sequenced for DNA methylation as described in Levine et al., 2020. The raw sequencing datasets are available from GEO (GSE161141). To identify differentially methylated loci: we trimmed reads of adapter dimers using Trim Galore (0.4.3) and quality trimmed with a minimum quality score above >25 (--rrbs -q 25). The attached adapter dimers were trimmed using cutadapter. First, bisulfite-converted index (GA and CT conversion) was generated using F344 rat genome with bismark build option and trimmed reads were aligned with bismark (Krueger and Andrews, 2011). Once we created aligned reads and corresponding locations, we used the bismark_methylation_extrator tool to summarize the level of methylation in CpG sites (bismark_methylation_extractor -p –comprehensive –no_overlap –bedGraph –counts –buffer_size 16 G [$Aligned read bam file]). Approximately a total 1–2 million sites per sample were predicted with DNA methylated sites (or unmethylated). DMR and blocks of differentially methylated sites were identified with a minimum of 3 CpG sites per block and at least >5% methylation difference (FDR <0.05) using MethylKit (PMID: 23034086 Akalin et al., 2012) and DMRseq (Korthauer et al., 2019). Functional annotation was performed using ClusterProfiler (Yu et al., 2012).

To generate RRBS(Reduced representation bisulfite sequencing) DNA methylation block: RRBS datasets were processed in a uniform way and DNA methylation levels for each sample were extracted. Then, we generated 200 base-pair binned DNA methylation levels across the genome. Each DNA methylation block contains 1–20 CpG sites. We first calculated the average DNAm level per DNAm block for each rat age then used this average DNAm level for mean CpG imputation. We calculated the DNAm level using all methylated and unmethylated sites from the binned CpG sites when the minimum coverage was more than 5. The F344 build of the rat reference genome was used for bisulfite sequencing alignment, and the rn6 genome feature was used to extract genomic annotation information using the Homer package (Heinz et al., 2010). Genes, exons, introns, and UTRs were taken from Homer annotation tools (annotatePeaks.pl DMR rn6). TSS-promoter sites were considered the identified DMRs close to TSS (less than 1 kb). TSS-proximal sites were considered the identified DMRs close to TSS (less than 20 kb), and TSS-distal sites were considered the DNAm block whose distance to TSS is above 20 kb. For downstream analysis to find inflection points, we used the TSS-promoter DNAm block for searching inflection point analysis (e.g. Runx1, Dnmt3a).

Calculation of DNA methylation breakpoints

Genes for DNA methylation breakpoint analysis were selected by first filtering for DMRs that changed as a function of age, determined by a response screening analysis (JMP platform) with an FDR p-value<0.05. The resulting 2,311 sites were then fitted for a linear and a logistic 4 P regression, and the better fit was selected based on Akaike Information Criterion (Shavlakadze et al., 2019). Next, we calculated the aging change breakpoint of DMRs that were more likely to be non-linear using the formula in Figure 6—figure supplement 1.

Statistical analysis and software used

Population frequencies were determined with FlowJo v10.8 Software (BD Life Sciences) ViSNE and self-organizing maps analyses were performed using Cytobank (Kotecha et al., 2010). All statistical analyses except those specifically detailed otherwise were performed using JMP (JMP, Version 16. SAS Institute Inc, Cary, NC, 1989–2021). The predictive model of age based on leukocyte frequencies was constructed using the least square method with backstep elimination of insignificant variables. Analysis of pathology prediction was performed using the JMP Decision Tree Platform. Determination of regression formulas and breakpoints was performed in JMP according to the formulas indicated in Figure 6—figure supplement 1.

Data availability

The data that support the findings of this study are available in the supplementary material and supplementary tables of this article. The DNA methylation raw data is available via GEO Accession (GSE161141). FCS files have been uploaded to FlowRepository (FR-FCM-Z59K). Further information and requests for resources and reagents should be directed to and will be replied to by the corresponding authors.

The following data sets were generated
    1. Yanai H
    (2022) FlowRepository
    ID FR-FCM-Z59K. Rat Cross Sectional Aging PB.
    http://flowrepository.org/id/FR-FCM-Z59K
The following previously published data sets were used
    1. Levine M
    (2020) NCBI Gene Expression Omnibus
    ID GSE161141. A rat epigenetic clock recapitulates phenotypic aging and co-localizes with heterochromatin-associated histone modifications.
    https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE161141

References

    1. Gibbs RA
    2. Weinstock GM
    3. Metzker ML
    4. Muzny DM
    5. Sodergren EJ
    6. Scherer S
    7. Scott G
    8. Steffen D
    9. Worley KC
    10. Burch PE
    11. Okwuonu G
    12. Hines S
    13. Lewis L
    14. DeRamo C
    15. Delgado O
    16. Dugan-Rocha S
    17. Miner G
    18. Morgan M
    19. Hawes A
    20. Gill R
    21. Holt RA
    22. Adams MD
    23. Amanatides PG
    24. Baden-Tillson H
    25. Barnstead M
    26. Chin S
    27. Evans CA
    28. Ferriera S
    29. Fosler C
    30. Glodek A
    31. Gu Z
    32. Jennings D
    33. Kraft CL
    34. Nguyen T
    35. Pfannkoch CM
    36. Sitter C
    37. Sutton GG
    38. Venter JC
    39. Woodage T
    40. Smith D
    41. Lee HM
    42. Gustafson E
    43. Cahill P
    44. Kana A
    45. Doucette-Stamm L
    46. Weinstock K
    47. Fechtel K
    48. Weiss RB
    49. Dunn DM
    50. Green ED
    51. Blakesley RW
    52. Bouffard GG
    53. De Jong PJ
    54. Osoegawa K
    55. Zhu B
    56. Marra M
    57. Schein J
    58. Bosdet I
    59. Fjell C
    60. Jones S
    61. Krzywinski M
    62. Mathewson C
    63. Siddiqui A
    64. Wye N
    65. McPherson J
    66. Zhao S
    67. Fraser CM
    68. Shetty J
    69. Shatsman S
    70. Geer K
    71. Chen Y
    72. Abramzon S
    73. Nierman WC
    74. Havlak PH
    75. Chen R
    76. Durbin KJ
    77. Egan A
    78. Ren Y
    79. Song XZ
    80. Li B
    81. Liu Y
    82. Qin X
    83. Cawley S
    84. Worley KC
    85. Cooney AJ
    86. D’Souza LM
    87. Martin K
    88. Wu JQ
    89. Gonzalez-Garay ML
    90. Jackson AR
    91. Kalafus KJ
    92. McLeod MP
    93. Milosavljevic A
    94. Virk D
    95. Volkov A
    96. Wheeler DA
    97. Zhang Z
    98. Bailey JA
    99. Eichler EE
    100. Tuzun E
    101. Birney E
    102. Mongin E
    103. Ureta-Vidal A
    104. Woodwark C
    105. Zdobnov E
    106. Bork P
    107. Suyama M
    108. Torrents D
    109. Alexandersson M
    110. Trask BJ
    111. Young JM
    112. Huang H
    113. Wang H
    114. Xing H
    115. Daniels S
    116. Gietzen D
    117. Schmidt J
    118. Stevens K
    119. Vitt U
    120. Wingrove J
    121. Camara F
    122. Mar Albà M
    123. Abril JF
    124. Guigo R
    125. Smit A
    126. Dubchak I
    127. Rubin EM
    128. Couronne O
    129. Poliakov A
    130. Hübner N
    131. Ganten D
    132. Goesele C
    133. Hummel O
    134. Kreitler T
    135. Lee YA
    136. Monti J
    137. Schulz H
    138. Zimdahl H
    139. Himmelbauer H
    140. Lehrach H
    141. Jacob HJ
    142. Bromberg S
    143. Gullings-Handley J
    144. Jensen-Seaman MI
    145. Kwitek AE
    146. Lazar J
    147. Pasko D
    148. Tonellato PJ
    149. Twigger S
    150. Ponting CP
    151. Duarte JM
    152. Rice S
    153. Goodstadt L
    154. Beatson SA
    155. Emes RD
    156. Winter EE
    157. Webber C
    158. Brandt P
    159. Nyakatura G
    160. Adetobi M
    161. Chiaromonte F
    162. Elnitski L
    163. Eswara P
    164. Hardison RC
    165. Hou M
    166. Kolbe D
    167. Makova K
    168. Miller W
    169. Nekrutenko A
    170. Riemer C
    171. Schwartz S
    172. Taylor J
    173. Yang S
    174. Zhang Y
    175. Lindpaintner K
    176. Andrews TD
    177. Caccamo M
    178. Clamp M
    179. Clarke L
    180. Curwen V
    181. Durbin R
    182. Eyras E
    183. Searle SM
    184. Cooper GM
    185. Batzoglou S
    186. Brudno M
    187. Sidow A
    188. Stone EA
    189. Venter JC
    190. Payseur BA
    191. Bourque G
    192. López-Otín C
    193. Puente XS
    194. Chakrabarti K
    195. Chatterji S
    196. Dewey C
    197. Pachter L
    198. Bray N
    199. Yap VB
    200. Caspi A
    201. Tesler G
    202. Pevzner PA
    203. Haussler D
    204. Roskin KM
    205. Baertsch R
    206. Clawson H
    207. Furey TS
    208. Hinrichs AS
    209. Karolchik D
    210. Kent WJ
    211. Rosenbloom KR
    212. Trumbower H
    213. Weirauch M
    214. Cooper DN
    215. Stenson PD
    216. Ma B
    217. Brent M
    218. Arumugam M
    219. Shteynberg D
    220. Copley RR
    221. Taylor MS
    222. Riethman H
    223. Mudunuri U
    224. Peterson J
    225. Guyer M
    226. Felsenfeld A
    227. Old S
    228. Mockrin S
    229. Collins F
    230. Rat Genome Sequencing Project Consortium
    (2004) Genome sequence of the Brown Norway rat yields insights into mammalian evolution
    Nature 428:493–521.
    https://doi.org/10.1038/nature02426
    1. Ward JM
    2. Reynolds CW
    (1983)
    Large granular lymphocyte leukemia. A heterogeneous lymphocytic leukemia in F344 rats
    The American Journal of Pathology 111:1–10.

Decision letter

  1. Bérénice A Benayoun
    Reviewing Editor; University of Southern California, United States
  2. Carlos Isales
    Senior Editor; Medical College of Georgia at Augusta University, United States

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Rat leukocyte population dynamics predicts a window for intervention in aging." for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Carlos Isales as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

Based on our 3 assessments, the most salient points the authors need to address for a revision are:

1. Add additional important context and discussion for choices that may have a significant impact on results (all 3 reviewers):

– Using only males (the expectations if females were included should be discussed);

– Censoring of overtly ill animals (would the results differ if included?), which may bias against frail animals; (If these data points can be included, that would be ideal, if not, the caveat must be discussed explicitly);

– Choice of time breakpoint for the analysis;

– Mentioning the cell composition shifts as caveats for the methylation data.

2. Methodologies need to be reported in more detail and in a more transparent manner for long-term reproducibility (i.e. staining for flow cytometry, availability of the model for other people to reuse, per animal data needs to be reported in a way that allows re-analysis [with identical animal identifiers in all data forms], use of fresh/fixed protocol for the flow cytometry, etc.) (reviewers 1 and 2).

3. Clarifications need to be brought to the relative importance/results linked to myeloid bias (reviewers 1 and 2).

4. If possible, providing more contextualization as compared to studies in other species (reviewer 2).

Finally, we would suggest careful copyediting, including checking the numbers of animals, making sure all figures are referenced and clarifying the study goals in the abstract clarification, may be needed (reviewer 1 and 3).

Reviewer #1 (Recommendations for the authors):

1. Since the authors were able to obtain data from 146 animals (Figure 1 data), it would be important to include information on a per animal basis (e.g. average/median number of cells per individual, etc) to have an idea of inter-individual variability.

2. It will be interesting to characterize the age-associated changes with CD4/CD8 T cell ratio in Figure 2 (heatmap), since a decrease in the CD4/CD8 T cell ratio is one of the key characteristics found in human peripheral blood (line 47).

3. Authors should recheck animal numbers in line 76 (146 animals) and line 165 (159 animals). Which number is correct? It will be important to make sure all animals are accounted for.

4. A key characteristic observed in aged peripheral blood is increased myeloid bias. Also, it is described that illness is also associated with increased myeloid bias. Thus, it is slightly puzzling that in Figure 3a more than half of animals with illness (gray dots) present with younger predicted age relative to actual age.

5. Authors discuss possible implications one may infer from the DNA methylation profiles of ALL leukocytes in the peripheral blood (lines 351-359). However, since it is very clear that the cell composition of peripheral blood undergoes significant changes with age, it will be highly relevant to understand the changes in DNA methylation levels along with changes in leukocyte composition. This should be mentioned in the Discussion session.

6. It will be important to describe the staining method used for the flow cytometry (e.g. staining time, antibody concentrations, etc.) should anyone wish to use the protocol from this manuscript.

Reviewer #2 (Recommendations for the authors):

1. The authors state: "The second shift point we observed is defined by a marked increase in variance between individual rats which alludes to a pan-hematopoietic loss of homeostasis." This increase in variance can also be explained by the difference in the rate of aging between different individuals. The same goes for the methylation analysis.

2. It is unclear why the authors decided to split the age at 15 months to study the breakpoints of sites. Using this threshold obtained from the cell composition data can introduce a bias into the methylation analysis. The manuscript says: "The methylation switch points defined converged at around 22 months of age and mildly at close to 12 months of age" but it's unclear how these points were obtained. There is no table or figure demonstrating this observation provided in the paper.

3. Genes breakpoint calculations from the data on individual CpGs require a more detailed explanation. Also, are the ApoE and Runx1, mentioned as the genes having a breakpoint at earlier than 15 months, the top hits?

4. The myeloid/lymphoid ratio seems to be a very good predictor by itself. How does it perform in comparison to the proposed model?

5. Description of the differentially methylated regions and the overall change to the methylation landscape would be more insightful if compared to what has been identified in the other species – for example, in another rodent – mouse (Sziráki et al., Aging Cell, 2018).

Reviewer #3 (Recommendations for the authors):

1) It is understandable that the authors have not yet completed studies on females, but this should be acknowledged as a major limitation of the work. This should be added to the discussion and the word "male" added to the title and abstract. Also, in the discussion the authors should consider this when they are speaking of similarities and differences between their findings in male rats and comparisons to studies done in humans (potentially in both sexes). As the authors argue that they are looking for a model that is highly translatable to humans, it is important that they also consider translation for both sexes.

2) The authors should completely re-write their abstract. At present is far too vague and speculative and does not give the reader a clear idea of the work they have done.

3) The word fragility is used in the abstract but nowhere else. I suspect they mean frailty, but this should be defined. They should discuss a potential role for frailty in the context of their work in a modified discussion.

4) The authors have not analyzed blood from rats with overt illness. This would have the effect of removing data from the frailest rats from their study. Do the authors have these data to add to the study? This would be very interesting. The absence of these measures should at least be discussed as a limitation in the revised discussion.

https://doi.org/10.7554/eLife.76808.sa1

Author response

Essential revisions:

Based on our 3 assessments, the most salient points the authors need to address for a revision are:

1. Add additional important context and discussion for choices that may have a significant impact on results (all 3 reviewers):

– Using only males (the expectations if females were included should be discussed);

This study was indeed designed with only males initially to avoid experimental variation at later ages which is a recognized challenge in aging studies. Though we are unable to comprehensively compare female rats at all ages, we too appreciated the relevance of comparing female data. To attempt to address this concern, we analyzed leukocyte frequencies in the peripheral blood of female F344 rats taken from 3 age groups: 5 young (5 months old), 5 middle-aged (15 months old), and 5 old (23 months old) and have included these results (please see Figure 3 —figure supplement 2, Results, and Discussion). To briefly address the finding here, we were pleasantly surprised to see the age prediction model generated from the male rats seems to work robustly on predictions for female data (prediction p <0.0001). This was true even though we were able to observe sex-specific differences, that appear to become exacerbated with age. The blood parameters with sex-specific differences though, do not seem to be dominant parameters driving the prediction model. Though these data would need to be more comprehensively validated, we show both sex-specific aging differences in male and female rats, but also demonstrate the current prediction model robustly predicts the female age as well. We did not have any sick or overtly ill females from this small data set, so we were cautious in the overall conclusions drawn from the additional female samples; however, were still pleased with the data supporting the utility of the male-derived model for female rats as well.

– Censoring of overtly ill animals (would the results differ if included?), which may bias against frail animals; (If these data points can be included, that would be ideal, if not, the caveat must be discussed explicitly);

Our apologies for not making the analyses statement more clear. The excluded animals were part of the originally planned cohort, but these rats did not have blood measurements taken as the veterinarian staff identified them as moribund and the rodents were culled for humane reasons. Their blood was not collected as the rest of the age-matched cohorts were not being assessed at the time of the sick animals’ humane euthanasia. Instead, we attempted to determine how the blood parameters were presented in illness based on veterinary assessment and necropsy on non-overtly ill rodents. We acknowledge that, in retrospect, it would have been beneficial to collect blood from these animals for analysis but unfortunately, we did not. As our study was cross-sectional and blood collections were grouped, we can only get a “snapshot” of the state of the animals during collection. We believe that tracing individual rats in a longitudinal study, ideally using our blood parameters examined in this manuscript, would be an excellent topic to tackle to pre-identify animals that are projected to become moribund.

– Choice of time breakpoint for the analysis;

Please see the specific response to reviewer #2, but briefly: the breakpoints for the analysis were data-driven. In the case of blood parameters, we performed a multiple regression analysis for each cell population and identified time points in which there was a major trajectory shift and most of these shifts appear to converge at approximately 15 months of age. For the methylation data, we performed both linear and logistic regressions on significant age-associated DMRs near TSSs and for each DMR selected the regression with the best fit (based on Akaike Information Criterion). Most sites changed linearly with age, while ~40% showed a better fit to a logistic regression. We used the DMRs that better fit the logistic regression to determine the breakpoint. This was calculated using the breakpoint analysis formula described in Supp Figure 7C. The methylation switch points defined from these analyses, converged at around 22 months of age and mildly at close to 12 months of age.

– Mentioning the cell composition shifts as caveats for the methylation data.

This is an excellent point that we overlooked mentioning and we now raise this caveat in the discussion. As this analysis is a correlative analysis, it is naturally difficult to distinguish between the effects of co-factors that change together. To fully answer this question, we would need to obtain the DNA methylation profiles of individual cell types. This would also require the generation of profiles from each cell type from aged animals, as we hypothesize that specific populations will likely display unique age-associated methylation changes. We have attempted to provide some insight into this question bioinformatically with our existing data by comparing the aging methylation changes to the population frequency changes. While we found some interesting correlations, we do not believe the evidence is strong enough to be included in the main paper and will only be described here.

Briefly, we compared the correlations between the top 2311 DMRs that changed as a function of age with leukocyte population frequencies. To do that, we ran a least-square model that predicts the DNA methylation levels, with or without age as a predictor:

As seen in Author response image 1, Age is a stronger predictor for DNA methylation levels than leukocyte populations on their own. Myeloid/lymphoid ratio is also a strong coefficient, suggesting that cell composition does indeed contribute to the observed DNA methylation changes. However, as seen in Author response image 2, overall R-square values (where the higher the value, the more accurately the parameters describe the observed changes in the DMR) are higher when age is included as a variable than without. This suggests that age, on its own, is a contributor to the DNA methylation changes. Further, the breakpoints between changes in the blood parameters do not overlap directly with the breakpoints in the methylation changes, suggesting the methylation and blood parameter changes may not be completely dependent. While we think this analysis is indicative of a cell composition independent methylation change, we do not believe this evidence is strong enough to make that claim definitively. Therefore, we have added a caveat describing that we cannot exclude the possibility that the methylation changes are driven by changes in the cell composition in the Discussion.

Author response image 1
Download asset Open asset
Heatmap summarizing the impact of each leukocyte population on predicting DNA methylation by least-square model.

A least square model was constructed for each DMR of the 2311 top hits using age and leukocyte frequencies as predictive variables. Each row in the heatmap indicates a single DMR. Each column indicates the parameter as indicated. The color corresponds to the P value where blue is < 0.05.

Author response image 2
Download asset Open asset
Summary of R-square values for the models constructed with and without age as a predictor.

The R-square values are summarized as a violin plot overlayed with a box blot. Left (blue) are the R-square values with age and right (red) without age.

2. Methodologies need to be reported in more detail and in a more transparent manner for long-term reproducibility (i.e. staining for flow cytometry, availability of the model for other people to reuse, per animal data needs to be reported in a way that allows re-analysis [with identical animal identifiers in all data forms], use of fresh/fixed protocol for the flow cytometry, etc.) (reviewers 1 and 2).

We have expanded the Materials and methods to include all data needed to reproduce the experiment. We have reconfigured the Suppl. Files to be in a more user-friendly format and ensured all identifiers correspond correctly.

3. Clarifications need to be brought to the relative importance/results linked to myeloid bias (reviewers 1 and 2).

Please see specific responses to reviewers. Briefly, we should reiterate that the model was generated on healthy rats, and as such myeloid/lymphoid ratio was not among the strongest predictors. The other parameters (such as CD25+ CD4 cells, NK frequency, CD8+ frequency of T-cells, etc.) are the drivers of the healthy model. The myeloid bias has a small dynamic range in the young rats and therefore is not a strong predictor of “healthy” aging. If we had included sick rats in the model, myeloid/lymphoid ratio would indeed be a robust predictor of illness. We believe that a model to predict age should be based on animals that are healthy (as measurable) to reduce the confounding effects of pathologies.

4. If possible, providing more contextualization as compared to studies in other species (reviewer 2).

We were excited to see the analyses when we performed this suggested comparison as we found that the rat blood had quite similar trends to what was seen in human blood with aging, with the dominant changes being loss of methylation in the older cells. Interestingly, the mouse data show similar amounts of both loss and gains of methylation. When we looked for concordance between the three species, we see that the majority of DMRs that gain or lose methylation during aging in rats, also show the same trends in both mouse and human data. Our one large caveat to this analysis is each of these data sets was generated using different platforms with different levels of coverage. While the human global analysis looks similar to what we report in the rats, when examining concordance, there were excessive amounts of missing orthologous data points. We have included these analyses in Figure 4.

Finally, we would suggest careful copyediting, including checking the numbers of animals, making sure all figures are referenced and clarifying the study goals in the abstract clarification, may be needed (reviewer 1 and 3).

We sincerely apologize for the oversights and likely annoyance that they caused. We have done a better job editing and have revised the abstract significantly.

Reviewer #1 (Recommendations for the authors):

1. Since the authors were able to obtain data from 146 animals (Figure 1 data), it would be important to include information on a per animal basis (e.g. average/median number of cells per individual, etc) to have an idea of inter-individual variability.

We apologize for the confusion partially due to how we uploaded the Supplemental Tables. Suppl. Table 2 is now presented in a more approachable form and includes all the population frequencies on an individual animal basis to reflect inter-individual variability and allow others to re-analyze the results. As we record a specific number of events in the flow cytometry data collection, the cell number recorded does not accurately reflect interindividual variability on the total cellularity, so we present the data as frequencies. We have also detailed in the Methods that the original flow files are available upon request, and our gating strategy is included in Figure 1 —figure supplement 2.

2. It will be interesting to characterize the age-associated changes with CD4/CD8 T cell ratio in Figure 2 (heatmap), since a decrease in the CD4/CD8 T cell ratio is one of the key characteristics found in human peripheral blood (line 47).

We agree, thank you! We have now included the CD4/CD8 ratio in Figure 4 —figure supplement 1 and discuss it in the results. We do find a modest decline in CD4/CD8 ratio in the first half of life (as seen in humans). Surprisingly, we also observed that in the later stages of aging, there is a strong dysregulation of this ratio. In part, this may be explained by an accumulation of CD25+/CD4+ T cells that we described.

3. Authors should recheck animal numbers in line 76 (146 animals) and line 165 (159 animals). Which number is correct? It will be important to make sure all animals are accounted for.

We apologize for this oversight. We have data on a total of 159 male animals. However, we chose to exclude animals younger than 3 months for this study, as we wanted to examine changes in adult hematopoiesis only and not early development, leading to a total number of 146. The numbers are now consistent in the text.

4. A key characteristic observed in aged peripheral blood is increased myeloid bias. Also, it is described that illness is also associated with increased myeloid bias. Thus, it is slightly puzzling that in Figure 3a more than half of animals with illness (gray dots) present with younger predicted age relative to actual age.

This is an excellent point. It is important to stress here that the model was generated using only healthy animals. As such, myeloid/lymphoid ratio was not among the strongest predictors (the top predictors are shown in Figure 3). While the ratio does significantly increase with age, it is not a strong predictor of “healthy” aging. We explored including all rats (healthy and ill) into the prediction model and found that myeloid/lymphoid becomes a significant contributor (p = 0.000232). However, due to the low dynamic range during the early ages, it was not amongst the strongest predictors when excluding sick rats. It seems that on its own, myeloid lymphoid ratio is more predictive of “ailment” than actual age but is a useful predictor of aging and pathology combined.

5. Authors discuss possible implications one may infer from the DNA methylation profiles of ALL leukocytes in the peripheral blood (lines 351-359). However, since it is very clear that the cell composition of peripheral blood undergoes significant changes with age, it will be highly relevant to understand the changes in DNA methylation levels along with changes in leukocyte composition. This should be mentioned in the Discussion session.

We agree. Please see the above response for the related analysis. The discussion now mentions this important caveat.

6. It will be important to describe the staining method used for the flow cytometry (e.g. staining time, antibody concentrations, etc.) should anyone wish to use the protocol from this manuscript.

Done- we have now included timing, volume, and concentration required for the staining. The same staining protocol was used on fixed and fresh blood. Please see the methods section to verify if these modified methods are clear enough for the reproducibility of the protocol.

Reviewer #2 (Recommendations for the authors):

1. The authors state: "The second shift point we observed is defined by a marked increase in variance between individual rats which alludes to a pan-hematopoietic loss of homeostasis." This increase in variance can also be explained by the difference in the rate of aging between different individuals. The same goes for the methylation analysis.

That is an excellent point that we overlooked. Thank you. It is now mentioned in the discussion.

2. It is unclear why the authors decided to split the age at 15 months to study the breakpoints of sites. Using this threshold obtained from the cell composition data can introduce a bias into the methylation analysis. The manuscript says: "The methylation switch points defined converged at around 22 months of age and mildly at close to 12 months of age" but it's unclear how these points were obtained. There is no table or figure demonstrating this observation provided in the paper.

We chose the 15-month split based on the shift we saw in the blood population frequencies and indeed it is somewhat arbitrary. However, 16 months of age is also where a drop in the survival curve begins and is also where the male rat reproductive capacity dramatically drops. While we acknowledge that this time point may present a bias, we believe it still accurately reflects a differentiation between early adulthood events and late ones. The convergence of switch points is meant to be descriptive rather than quantitative and is shown in Figure 6a (where the bulges are).

3. Genes breakpoint calculations from the data on individual CpGs require a more detailed explanation. Also, are the ApoE and Runx1, mentioned as the genes having a breakpoint at earlier than 15 months, the top hits?

We have expanded the explanation of breakpoint analysis in an added section of the Materials and methods:

“Genes for DNA methylation breakpoint analysis were selected by first filtering only for DMRs that changed as a function of age, determined by a Response Screening analysis (JMP platform) with an FDR p value < 0.05. The resulting 2,311 sites were then fitted for a linear and a logistic 4P regression and the better fit was selected based on Akaike Information Criterion (Shavlakadze, et al., 19). Next, we calculated the breakpoint on sites that were selected to be less likely to be linear using the formula in Figure 6 —figure supplement 1.”

ApoE and Runx1 were highlighted due to their relevance in the regulation of hematopoiesis. Runx1 is indeed the top hit in the differential methylation analysis and ApoE is more modestly altered.

4. The myeloid/lymphoid ratio seems to be a very good predictor by itself. How does it perform in comparison to the proposed model?

This is indeed an important point brought up by multiple reviewers. As mentioned in the response to reviewer #1, myeloid/lymphoid ratio did correlate strongly with age. When including sick animals, myeloid lymphoid ratio is indeed a very strong predictor of age (p = 0.000232). However, due to the low dynamic range during the early ages, it was not amongst the strongest predictors when excluding sick rats. It seems that on its own, myeloid lymphoid ratio is more predictive of “ailment” than actual age but is a useful predictor of aging and pathology combined.

5. Description of the differentially methylated regions and the overall change to the methylation landscape would be more insightful if compared to what has been identified in the other species – for example, in another rodent – mouse (Sziráki et al., Aging Cell, 2018).

This is an excellent suggestion, thank you. We have performed the comparison between our dataset and publicly available DNA methylation aging datasets from both humans and mice, including the one suggested. We find the results interesting and have expanded main Figure 5 to include these results showing the overall trends towards hypomethylation in the rat and humans, and how mice have a more balanced change- with similar numbers of both gains and loss of methylation associated with aging. However, it is important to note that due to differences in how the DNA methylation data was generated and ultimately sequenced, the comparison is not completely straightforward. We believe that this question deserves a deeper meta-analysis beyond the data we present here, but we feel that analysis may warrant a separate paper with additional data generation required.

Reviewer #3 (Recommendations for the authors):

1) It is understandable that the authors have not yet completed studies on females, but this should be acknowledged as a major limitation of the work. This should be added to the discussion and the word "male" added to the title and abstract. Also, in the discussion the authors should consider this when they are speaking of similarities and differences between their findings in male rats and comparisons to studies done in humans (potentially in both sexes). As the authors argue that they are looking for a model that is highly translatable to humans, it is important that they also consider translation for both sexes.

This is an important point, and we have tried to provide a bit of insight into the differences that might be seen in female and male rats. We have edited the title to more explicitly state this paper is describing male rat blood phenotypes. We were pleased to see that the predictive model also robustly predicted the female age albeit in a small sample set. Even given the limited number of female blood donors we could acquire; we saw significant sex-specific blood parameter differences and agree a more in-depth analysis of female rats would be an important follow-up study.

2) The authors should completely re-write their abstract. At present is far too vague and speculative and does not give the reader a clear idea of the work they have done.

We have taken the advice of the reviewer and significantly edited the abstract to more accurately reflect the key findings.

3) The word fragility is used in the abstract but nowhere else. I suspect they mean frailty, but this should be defined. They should discuss a potential role for frailty in the context of their work in a modified discussion.

In the abstract revision we have rephrased the statement that had included fragility. In retrospect, we agree that frailty measurements would have of great interest to measure and include a statement in the discussion about future inclusion of frailty measurements as parameters to determine if such a blood composition analysis could provide insight not only into illness, but frailty as well.

4) The authors have not analyzed blood from rats with overt illness. This would have the effect of removing data from the frailest rats from their study. Do the authors have these data to add to the study? This would be very interesting. The absence of these measures should at least be discussed as a limitation in the revised discussion.

We unfortunately do not have any of the blood parameters from the moribund animals (as described above). We hope to have addressed the major concerns raised and bring up in the discussion how a longitudinal study would be relevant to investigate if the model presented could pre-identify those rats heading towards overt-illness before they became moribund, preventing future studies from losing information by allowing for identification of rodents that will need to be collected / analyzed perhaps before scheduled collection.

https://doi.org/10.7554/eLife.76808.sa2

Article and author information

Author details

  1. Hagai Yanai

    Epigenetics and Stem Cell Unit, Translational Gerontology Branch, National Institute on Aging, Baltimore, United States
    Contribution
    Data curation, Formal analysis, Investigation, Methodology, Project administration, Writing – original draft, Writing – review and editing
    Contributed equally with
    Christopher Dunn
    For correspondence
    hagai.yanai@nih.gov
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1742-5411

  2. Christopher Dunn

    Flow Cytometry Core, National Institute on Aging, Baltimore, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Methodology
    Contributed equally with
    Hagai Yanai
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-7899-0110

  3. Bongsoo Park

    Epigenetics and Stem Cell Unit, Translational Gerontology Branch, National Institute on Aging, Baltimore, United States
    Contribution
    Formal analysis, Methodology
    Competing interests
    No competing interests declared

  4. Christopher Coletta

    Computational Biology and Genomics Core, Laboratory of Genetics & Genomics, National Institute on Aging, Baltimore, United States
    Contribution
    Formal analysis, Investigation, Methodology
    Competing interests
    No competing interests declared

  5. Ross A McDevitt

    Comparative Medicine Section, National Institute on Aging, Baltimore, United States
    Contribution
    Data curation, Formal analysis, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3722-9047

  6. Taylor McNeely

    Epigenetics and Stem Cell Unit, Translational Gerontology Branch, National Institute on Aging, Baltimore, United States
    Contribution
    Formal analysis, Investigation
    Competing interests
    No competing interests declared

  7. Michael Leone

    Epigenetics and Stem Cell Unit, Translational Gerontology Branch, National Institute on Aging, Baltimore, United States
    Contribution
    Formal analysis, Investigation
    Competing interests
    No competing interests declared

  8. Robert P Wersto

    Flow Cytometry Core, National Institute on Aging, Baltimore, United States
    Contribution
    Conceptualization, Resources
    Competing interests
    No competing interests declared

  9. Kathy A Perdue

    Comparative Medicine Section, National Institute on Aging, Baltimore, United States
    Contribution
    Conceptualization, Data curation, Resources
    Competing interests
    No competing interests declared

  10. Isabel Beerman

    Epigenetics and Stem Cell Unit, Translational Gerontology Branch, National Institute on Aging, Baltimore, United States
    Contribution
    Formal analysis, Funding acquisition, Investigation, Supervision, Writing – review and editing
    For correspondence
    isabel.beerman@nih.gov
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7758-8231

Funding

National Institutes of Health (Intramural NIA)

  • Isabel Beerman

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

Many thanks to Dr. Rafael de Cabo, Dr. Luigi Ferrucci, Dr. Michel Bernier, and all members of TGB for invaluable support. We would like to thank all the members of the NIA Comparative Medicine Section for their consistent efforts and high standards of animal care. A special thank you to Daniel Ariad for invaluable advice. This research was supported entirely by the Intramural Research Program of the NIH, National institute on Aging.

Ethics

All experimental procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals and approved by the NIA Animal Care and Use Committee (ASP 467-CMS-2018 and 469-TGB-2022).

Senior Editor

  1. Carlos Isales, Medical College of Georgia at Augusta University, United States

Reviewing Editor

  1. Bérénice A Benayoun, University of Southern California, United States

Publication history

  1. Received: January 5, 2022
  2. Preprint posted: January 14, 2022 (view preprint)
  3. Accepted: May 3, 2022
  4. Accepted Manuscript published: May 4, 2022 (version 1)
  5. Version of Record published: May 30, 2022 (version 2)

Copyright

This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Hagai Yanai
  2. Christopher Dunn
  3. Bongsoo Park
  4. Christopher Coletta
  5. Ross A McDevitt
  6. Taylor McNeely
  7. Michael Leone
  8. Robert P Wersto
  9. Kathy A Perdue
  10. Isabel Beerman
(2022)
Male rat leukocyte population dynamics predict a window for intervention in aging
eLife 11:e76808.
https://doi.org/10.7554/eLife.76808

Categories and tags

Research organism