Macrophages on the run: Exercise balances macrophage polarization for improved health

Abstract

Objective: Exercise plays a crucial role in maintaining and improving human health. However, the precise molecular mechanisms that govern the body's response to exercise or/compared to periods of inactivity remain elusive. Current evidence appears to suggest that exercise exerts a seemingly dual influence on macrophage polarization states, inducing both pro-immune response M1 activation and cell-repair-focused M2 activation. To reconcile this apparent paradox, we leveraged a comprehensive meta-analysis of 75 diverse exercise and immobilization published datasets (7000+ samples), encompassing various exercise modalities, sampling techniques, and species.

Methods: 75 exercise and immobilization expression datasets were identified and processed for analysis. The data was analyzed using boolean relationships which uses binary gene expression relationships in order to increase the signal to noise achieved from the data, allowing for the use of comparison across such a diverse set of datasets. We utilized a boolean relationship-aided macrophage gene model [1], to model the macrophage polarization state in pre and post exercise samples in both immediate exercise and long term training.

Results: Our modeling uncovered a key temporal dynamic: exercise triggers an immediate M1 surge, while long term training transitions to sustained M2 activation. These patterns were consistent across different species (human vs mouse), sampling methods (blood vs muscle biopsy), and exercise type (resistance vs endurance), and routinely showed statistically significant results. Immobilization was shown to have the opposite effect of exercise by triggering an immediate M2 activation. Individual characteristics like gender, exercise intensity and age were found to impact the degree of polarization without changing the overall patterns. To model macrophages within the specific context of muscle tissue, we identified a focused gene set signature of muscle resident macrophage polarization, allowing for the precise measurement of macrophage activity in response to exercise within the muscle.

Conclusions: These consistent patterns across all 75 examined studies suggest that the long term health benefits of exercise stem from its ability to orchestrate a balanced and temporally-regulated interplay between pro-immune response (M1) and reparative macrophage activity (M2). Similarly, it suggests that an imbalance between pro-immune and cell repair responses could facilitate disease development. Our findings shed light on the intricate molecular choreography behind exercise-induced health benefits with a particular insight on its effect on the macrophages within the muscle.

Keywords: Exercise biology; Immune modulation; Machine learning models; Macrophage polarization; Muscle; Tissue regeneration.

Figure 1

Boolean workflow for analyzing macrophage polarization changes in exercise datasets. A) Project Outline, with initial selection of datasets for meta-analysis, applying the macrophage polarization boolean model to each dataset to calculate a “macrophage score” for each sample, with lower scores representing M1 activation and higher scores representing M2 activation, and finally the comparing of the “macrophage scores” in pre and post exercise samples to measure the effect of exercise on macrophage polarization. B) StepMiner algorithm, in which samples are ranked by the expression of the selected genes, and a step function is fitted to the ranked data to minimize the sum of square error of the fitted data. The two sides of this threshold and margin allow for the binarization of the gene’s expression into “low” and “high” values. C) Boolean relationship scatter plots of the six possible boolean relationships between two genes. A Boolean implication relationship is identified if any of the scatter plot quadrants, which are split based on each gene’s StepMiner threshold, or two diagonally opposite ones, are sparsely populated. D) Formation of Boolean network. Using equivalent gene clusters as nodes and the boolean relationships between the clusters as the edges and network can be created from the boolean relationships. This network can then be oriented by condition and using machine learning to find a path from one condition to the other that can accurately separate between samples from each condition.

Figure 2

PRISMA flowchart showing the selection of datasets for the meta-analysis, with initial screening of the terms: exercise or training, alongside either aerobic endurance resistance or strength across PubMed, Embase, Web of Science, and the Cochrane Library leading to 819 non-duplicate studies. Number of databases is listed (nb) alongside the total number of samples (n). Of these 93 were retrieved for full text, 8 studies were filtered for containing only control non-exercised subjects, 5 were removed for not containing gene expression data, 3 for having no correlated published paper and 2 for belonging to a species for which no other dataset could be found.

Figure 3

Immediate exercise triggers the activation of M1 macrophages, while long term exercise yields the converse effect of M2 activation. A) Immediate exercise results in a statistically significant reduction in “macrophage scores,” indicating the activation of M1 macrophages and an enhanced pro-immune response. This pattern is evident in selected datasets involving 6-week treadmill-trained mice (GSE104079 [52]), COPD human patients (GSE27536 [85]), male humans undergoing knee extension exercise (GSE120862 [65]), and male and female humans post–marathon (GSE83578 [21]). B) Prolonged exercise leads to an elevation in “macrophage scores,” indicative of the activation of M2 macrophages and a conducive environment for wound healing. Chosen datasets exhibiting this opposing pattern include young male human subjects engaged in aerobic training (GSE111555 [120]), resistance training alongside a high fat diet (GSE99963 [43]), lifelong exercise in both male and female young (20–30 yr) and old (75+ yr) subjects (GSE144304 [111]) and a 12 week resistance training regime for young male and female subjects (GSE28998 [34]). C) Immobilization studies illustrate a distinct contrast to the immediate exercise response, wherein higher post-exercise scores signify the effects of inactivity in direct opposition to those induced by exercise. The datasets presented encompass 60 days of bed rest in all female subjects (GSE14798 [27]) as well as observations at 2 and 14 days following quad brace in both male and female young patients (GSE14901 [20]). The scores for GSE27536 and GSE14798 were calculated with only the M1 associated genes.

Figure 4

Exercise Macrophage Activations are Consistent across diverse datasets. A) Resistance and Endurance exercise, B) Blood and muscle biopsy sampling, C) Human and Mouse species, and D) RNA-Seq vs Microarray sequencing. In each section, dot plots depict exercise datasets on the y-axis, identified by their GSE accession numbers, while the x-axis represents the ROC-AUC scores obtained when attempting to differentiate pre and post-exercise data based solely on the composite “macrophage score”. In this context, a ROC-AUC value of 0 indicates that all post-exercise samples exhibit lower scores compared to pre-exercise samples, indicating a higher degree of M1 macrophage activation. Conversely, a ROC-AUC of 1 signifies that all post-exercise samples possess higher scores than pre-exercise samples, indicating a heightened level of M2 macrophage activation. The dot sizes are proportional to the -Log10(P Value) derived from a t-test comparing the pre and post-exercise groups, with statistically significant values displayed in solid shading, while non-significant datasets are slightly faded. Notably, all immediate exercise datasets exhibit ROC-AUC values below 0.5, aligning with M1 macrophage activation and an immune response. Conversely, regardless of dataset segmentation, all long term training exercise datasets showcase ROC-AUC scores surpassing 0.5.

Figure 5

Gender, Intensity and Age Influence Macrophage Polarization in Humans, for datasets that included exercise data of multiple subgroups the pre/post or sedentary/trained macrophage scores were calculated for each group separately and compared on the same y-axis. A) Male and Female human subjects B) Endurance and Resistance exercise regimes C) Higher and lower intensity human subjects (based on VO2 max percentage) D) Younger and older human subjects. In each section, the dot plots depict exercise datasets on the y-axis, identified by their GSE accession numbers, while the x-axis represents the ROC-AUC scores obtained when attempting to differentiate pre/post or sedentary/trained exercise data based solely on the composite “macrophage score”. In this context, a ROC-AUC value of 0 indicates that all post-exercise samples exhibit lower scores compared to pre-exercise samples, indicating a higher degree of M1 macrophage activation. Conversely, a ROC-AUC of 1 signifies that all post-exercise samples possess higher scores than pre-exercise samples, indicating a heightened level of M2 macrophage activation. The dot sizes are proportional to the -Log10(P Value) derived from a t-test comparing the pre and post-exercise groups, with statistically significant values displayed in solid shading, while non-significant datasets are slightly faded. Notably exercise’s impact on M1 and M2 macrophage activation was stronger in female subjects than males, not noticeably different in endurance and resistance exercise types, stronger in higher intensity subjects, and stronger in younger subjects when compared to older ones.

Figure 6

Identification new Signature of Muscle Resident Macrophage Polarization. A) Both Group 1 and Group 2 model gene sets are able to accurately separate and predict polarized macrophages. B) Pre and Post immediate exercise (within 6 h of exercise) C) Sedentary and Long term Trained (at least 10 weeks) individuals. Barplots show the Pre/Sendtary (in green) and Post/Exercised (in brown) group macrophage scores of both the Group 1 and 2 model. Higher Roc-Auc values represent a higher “M1” score while lower values represent a lower “M1” and higher M2 score. The two groups show opposite patterns to one another, with Group 1 having lower scores in the immediate exercise and higher scores in the long term training, and Group 2 having the exact opposite patterns. D) The expression of each of the gene model groups in both Spleen and Heart or Skeletal Muscle tissue. While the expression of Group 1 is not differentially distinct between the two tissue types, Group 2 shows a consistently higher expression in the macrophage rich spleen tissue as opposed to the muscle tissues identifying Group 2 as the Signature of Muscle Resident Macrophage Polarization.