Abstract
The structure and function of the brain and cardiovascular system change over the lifespan. In this study, we aim to establish the extent to which age-related changes in these two vital organs are linked. Utilizing normative models and data from the UK Biobank, we estimate biological ages for the brain and heart for 2,904 middle-aged and older healthy adults, including both males and females. Biological ages were based on multiple structural, morphological, and functional features derived from brain and cardiovascular imaging modalities. We find that cardiovascular aging, particularly aging of its functional capacity and physiology, is selectively associated with the aging of specific brain networks, including the salience, default mode, and somatomotor networks as well as the subcortex. Our work provides unique insight into brain–heart relationships and may facilitate an improved understanding of the increased co-occurrence of brain and heart diseases in aging.
Significance Statement
As individuals age, both the brain and cardiovascular systems undergo significant changes, making them susceptible to various neurodegenerative and cardiovascular conditions. Despite the physiological interplay between the brain and the cardiovascular system, brain networks and circuits that may drive this brain–heart aging axis remain largely unknown. Using multimodal multiorgan imaging and advanced analytic approaches, we elucidate age-related structural and functional brain changes that selectively associate with cardiovascular aging. Our findings may pave the way toward an improved understanding of the increased co-occurrence of brain and cardiovascular diseases in aging, thereby facilitating the development of more effective and synergistic intervention strategies in disease management, prognosis, and prevention.
Introduction
The brain and the heart are linked via extensive biophysiological pathways (McCraty, 2016; Tahsili-Fahadan and Geocadin, 2017). The brain is involved in the regulation and coordination of cardiovascular activity through the sympathetic and parasympathetic branches of the autonomic nervous system (McCraty, 2016), while adequate blood supply from the heart and proper blood pressure levels are essential for maintaining cerebral perfusion and normal brain functioning (Tzeng and Ainslie, 2014).
Using magnetic resonance imaging (MRI), researchers have found that the morphology and function of the aorta and cardiac chambers are associated with various brain characteristics across tissue types and functional systems, including gray matter, white matter, and functional connectivity (FC). Across different brain networks, widespread associations between heart characteristics and functional brain connectivity were found, specifically between the left ventricle morphology, cardiac function, and somatomotor and default mode resting-state networks (Zhao et al., 2023). In another study, MRI-derived free water (a putative marker of neuroinflammation), in the gray matter of these brain networks, was found to be associated with nonimaging blood-based cardiovascular biomarkers commonly used to inform cardiovascular dysfunction in myocardial injury and heart failure (Ji et al., 2023). This suggests a possible shared pathological pathway linking the dysfunction of the cardiovascular system and certain brain networks. Further work focusing on brain network-specific characteristics may help disentangle these complex brain–heart interactions.
Biophysiological pathways linking the brain and cardiovascular system may underlie the increased incidence of comorbidity between cardiac and brain diseases as people age (Abete et al., 2014; Saeed et al., 2023). Indeed, aging significantly impacts both cardiovascular and brain health and is associated with an increased risk of a number of cardiovascular and neurodegenerative conditions (Abete et al., 2014; van der Velpen et al., 2017). More specifically, aging impacts the structural integrity and functional capacity of the cardiovascular system, including increased stiffness of the aorta and other vessels (Wong et al., 2015), heightened fibrosis, thickening of the left ventricle walls, and reduced end-systolic and end-diastolic volumes in both sides of the ventricles (Kitzman and Edwards, 1990; Strait and Lakatta, 2012; Agrawal and Nagueh, 2022). Aging also results in a prolonged contraction and relaxation phase in the systole and diastole, as well as thickening of the heart valves (Kitzman and Edwards, 1990; Strait and Lakatta, 2012; Agrawal and Nagueh, 2022), which leads to impaired cardiac efficiency and an increased risk of cardiovascular diseases in the aging population. Moreover, several neuroimaging studies have demonstrated distinct age-related changes in brain structure and function across the adult lifespan (Bethlehem et al., 2022). For example, significant age-related decline in gray matter volume, cortical thickness, and morphological changes in subcortical nuclei is evident (Raz et al., 2005; Walhovd et al., 2011; Fjell et al., 2014). Aging also impacts white matter, with reduced volume, increased incidence of white matter hyperintensities, and altered microstructure (Song et al., 2005; Sun et al., 2005). Similarly, older adults tend to have reduced strength of functional brain connections compared with younger people (Ferreira and Busatto, 2013; Marstaller et al., 2015; Sala-Llonch et al., 2015; Hughes et al., 2020). A comprehensive characterization of the relationship between age-related changes in the cardiovascular system and the brain may facilitate a fundamental understanding of the increased prevalence of comorbidity between cardiac and brain diseases in aging.
The provision of openly available multiorgan and multimodal imaging data in large-scale population-based biobanks provides new opportunities to investigate these links. The concept of biological age is useful in this regard and provides an estimate of the extent to which an organ's structure, health, and functional capabilities deviate from the norm for individuals of the same chronological age (Cole and Franke, 2017; Gialluisi et al., 2019; Smith et al., 2020; Baecker et al., 2021). Associations between the biological age of the brain and heart have been reported (Goallec et al., 2021; Tian et al., 2023), and it has been suggested that advanced cardiovascular age may explain accelerated brain aging (Tian et al., 2023). Further work is needed to determine whether specific brain networks and circuits underlie joint aging of the brain and cardiovascular system. Cardiovascular activity is regulated by specific cortical and subcortical brain regions, including the prefrontal cortex, cingulate cortex, insula, somatosensory cortex, amygdala, hippocampus, thalamus, hypothalamus, cerebellum, and medulla (Benarroch, 2012; Chang et al., 2016; Shivkumar et al., 2016; Tahsili-Fahadan and Geocadin, 2017). Interestingly, these regions are also key regions comprising several spatially distributed brain networks, particularly the salience, default mode, and somatomotor networks. As such, we hypothesize that cardiovascular aging is uniquely linked to brain aging in these specific networks.
In this study, we utilized a cohort of middle-aged and older individuals to estimate the biological age of the heart and the brain, using both whole-brain (i.e., considering all brain regions) and network-specific approaches. Brain age was estimated using MRI-derived morphology and connectivity features, while heart age was estimated from cardiac MRI, carotid ultrasound, and pulse wave analysis (PWA; Fig. 1). We show that cardiovascular aging, specifically the decline in functional capacity and physiological changes, is linked with the aging of particular brain networks.
Overview of study design. a, Seven cortical networks, subcortex, and cardiac chambers and vessels were investigated. b, Different modalities for the brain and the cardiovascular system are listed. c, Brain age was estimated for the seven canonical functional brain networks and the subcortex, and heart age was predicted for the structural and functional aspects of the cardiovascular system using multimodal phenotypes from healthy individuals in the UK Biobank dataset. d, The accuracy of each model was evaluated based on Pearson’s correlation coefficient (r) and mean absolute error (MAE). e, Associations between brain age gap and heart age gap were investigated using Pearson’s partial correlation, controlling for the effect of chronological age and sex. SC, structural connectivity; FC, functional connectivity.
Materials and Methods
Dataset
The UK Biobank is an extensive biomedical dataset comprising over half a million individuals aged between 40 and 69 years at study recruitment in the United Kingdom (2006–2010). It involves a comprehensive baseline evaluation of participant demographics, medical history, lifestyle, genomic data, and various physical measures and blood sampling. Brain and heart MRI were acquired between 2014 and 2020 (Miller et al., 2016).
Participants
In the present study, we focused on a cohort of healthy individuals from the UK Biobank. These individuals had neither self-reported nor healthcare-documented chronic medical conditions. We included brain data from 8,398 individuals and cardiovascular data from 3,484 individuals for model training. Among these, 2,904 individuals had both brain and cardiovascular data collected at the first imaging visit (1,537 females; age range, 46–80 years; mean, 60.48 ± 7.24). Detailed demographic information is provided in Table 1.
Demographic characteristics
Brain phenotypes
In our analysis, we considered both morphological and connectivity metrics. The Destrieux cortical atlas with 148 regions was used for cortical parcellation (Destrieux et al., 2010), and the Melbourne Subcortical Atlas (Scale I) with 16 regions was used for subcortical connectivity (Tian et al., 2020). Details of brain regions are provided in Extended Data Table 1-1.
Table 1-1
Details of brain regions. Download Multimedia/Extended Data, XLSX file.
Table 1-2
Cardiovascular features details. Download Multimedia/Extended Data, XLSX file.
Morphology features consisted of gray matter volume, cortical thickness, and surface area estimated for each region. These features were derived from T1-weighted MRI and sourced from the UK Biobank (Miller et al., 2016).
Connectivity features, including both structural connectivity (SC) and functional connectivity (FC), were examined using data derived from diffusion-weighted imaging (DWI) and resting-state fMRI. Brain structural connectivity can provide a network characterization of white matter structure whereas functional connectivity characterizes statistical dependencies between the blood oxygen level-dependent (BOLD) time series recorded from different gray matter brain regions. Structural connectivity strength was assessed utilizing two measures: mean fractional anisotropy (FA) and streamline count. FA is a measure of the water diffusion within white matter tracts, while the streamline count offers a proxy for the strength of anatomical connections between brain regions (Smith et al., 2015; Zhang et al., 2022). In our study, streamline counts were used without normalization. Regarding functional connectivity, we utilized a pairwise Pearson’s correlation analysis of regional BOLD signals. Given the connectivity matrices were undirected, we utilized the upper triangle of both SC and FC matrices to avoid redundancy. Comprehensive details on the methods employed for computing SC and FC are provided elsewhere (Mansour et al., 2023).
To facilitate network-level inference, the cerebral cortex was partitioned into seven canonical networks: the frontoparietal/executive control network, dorsal attention network, default mode network, somatomotor network, ventral attention/salience network, visual network, and limbic network (Yeo et al., 2011). Utilizing surface geometry data, cortical regions from the Destrieux atlas were mapped into the seven networks. For each region, we identified the corresponding vertices from the Destrieux atlas within the network parcellation; the network label that appeared most frequently among the associated vertices was assigned as the region's network label (Extended Data Fig. 1-1). While this commonly used approach assumes that each cortical region contributes exclusively to a single network, some cortical areas may have overlapping functions across multiple networks (Yeo et al., 2014). We chose this approach for its simplicity and effectiveness (Yeo et al., 2011) in facilitating meaningful network-level analysis in our study. The total number of features corresponding to each network and modality (morphology, SC, FC) is shown in Extended Data Figure 1-2.
Figure 1-1
Mapping the Destrieux atlas regions into the Yeo 7-network. Each colour represents a specific network rendered on cortical surface. Download Figure 1-1, TIF file.
Figure 1-2
Number of features and regions corresponding to each network, including the count of features categorized under morphology, SC, and FC. Download Figure 1-2, TIF file.
Cardiovascular phenotypes
Cardiovascular features comprised cardiac magnetic resonance imaging (CMR), carotid ultrasound, and physiological data collected from pulse wave analysis.
Imaging features of the heart and aorta were extracted from CMR images through an automated machine learning-based analysis pipeline, as described elsewhere (Petersen et al., 2016; Bai et al., 2018). These features include measurements of aortic sections and the cardiac chambers: the ascending and descending aorta, left and right ventricle, left and right atrium, and regional phenotypes of the LV myocardial wall thickness and strain (Bai et al., 2020). Ultrasound imaging of the carotid vessels which deliver blood to the brain can provide information on the thickness of the carotid vessels, indicated as the carotid intima–medial thickness (IMT). We employed IMT at four angles, including 120°, 150°, 210°, and 240°. We also included pulse wave analysis (PWA) measurements obtained from a Vicorder device and an infrared sensor. The Vicorder, a brachial cuff-based device, was used during the cardiac MRI scan to estimate blood pressure measurements. The infrared sensor (PulseTrace PCA2), clipped to the finger, measured arterial stiffness or vascular reactivity. The shape of the pulse waveform reflects how long it takes for pulse waves to travel through the arteries in the body and return to the finger. Details about the PWA measurement procedure are available on the UK Biobank website.
We grouped the cardiovascular phenotypes into two categories based on their structural and functional relevance. Structural features encompassed image-derived phenotypes related to the structure of cardiovascular system such as chamber volume, myocardial wall thickness, vessel area, and thickness. Functional features index functional dynamics of the cardiovascular system, including the ventricle's ability to efficiently perfuse blood throughout the entire body including the brain, the elasticity of arteries, and the measurement of blood pressure. This categorization allowed separate examinations of the structural and functional aspects of the cardiovascular system. In total, we included 38 structural and 80 functional features. Details of the cardiovascular features and their categories are provided in Extended Data Table 1-2.
Average measures were calculated if measurements, such as heart rate and blood pressure, were repeated at the same visit. Participants with missing values for any of the heart and brain phenotypes were excluded from our analysis. Table 1 shows the sample size after these exclusions. An overview of different feature categories for the brain and cardiovascular system is provided in Table 2.
An overview of different modalities and features
Calculating the biological age of the brain and cardiovascular system
A CatBoost regressor model was trained to predict each individual's chronological age using multiple combinations of brain and cardiovascular phenotypes. CatBoost is an open-source machine learning library that employs gradient boosting on decision trees (Prokhorenkova et al., 2018).
Employing 10-fold cross-validation, separate predictive models were developed for the cardiovascular system (structure, function, and all features) and the brain (the whole brain, the seven brain networks, and the subcortex). Within each iteration of the 10-fold cross-validation, a CatBoost regressor was trained to estimate chronological age. The training dataset comprised nine folds, and chronological age was predicted for individuals comprising the remaining fold. Tuning hyperparameters did not lead to substantial enhancements in model performance; thus, CatBoost was trained using its default hyperparameter set. Individuals were stratified by sex to train dedicated models, and each model was trained exclusively using different feature categories [brain: SC, FC, morphology, and all (combination of SC, FC, and morphology); cardiovascular: structure, function, and all (combination of structure and function)]. As a result, we trained 36 [(the whole brain, seven brain networks, and the subcortex) × (SC, FC, morphology and all)] × 2 (female and male) = 72 models for brain age prediction and 3 (structure, function, and all features) × 2 (female and male) = 6 models for cardiovascular age prediction. Connectivity-based brain network age was estimated based on connections between regions within the same network only and internetwork connections were excluded. We evaluated prediction performance using the mean absolute error (MAE) and Pearson’s correlation coefficient (r) between predicted and actual chronological age.
The age gap index, defined as the difference between predicted and actual chronological age (predicted age minus chronological age) was calculated for the whole brain, seven networks, the subcortex, and cardiovascular system exclusively for females and males using different categories of features. Chronological age was regressed from the age gap to correct for the regression to the mean bias (Beheshti et al., 2019).
To evaluate the association between the brain age gap and cardiovascular age gap, the Pearson’s partial correlation was computed, adjusting for the effects of age and sex for all pairs. Correction for multiple comparisons was undertaken with the false discovery rate (FDR) level, using the Benjamini–Hochberg procedure to control the FDR at 5% (Benjamini and Hochberg, 1995) across 67 comparisons (64 brain network tests plus 3 whole-brain tests; Extended Data Fig. 4-1).
Figure 4-1
Brain-heart association analysis. Brain age gap calculated for each network exclusively for morphology, SC, FC and all features. Partial correlation was used to test associations between brain age gap and heart age gap, adjusting for chronological age and sex. Significant correlations (P < 0.05, FDR corrected across 67 tests) are depicted using solid lines, with line widths proportional to the absolute values of correlation coefficients. Dashed lines indicate non-significant correlations. Download Figure 4-1, TIF file.
Results
Biological age was estimated for the whole brain, as well as for regions belonging to seven canonical cortical networks (i.e., executive control, dorsal attention, default mode, somatomotor, salience, visual, limbic network) and the subcortex. Cardiovascular phenotypes were used to predict the biological age of cardiovascular structure and function.
Chronological age prediction accuracy
Model accuracy was assessed separately in predicting chronological age for the whole brain [females: Pearson’s correlation coefficient (r) = 0.78, MAE = 3.64; males: r = 0.80, MAE = 3.76], each of the seven cortical networks, and the subcortex using different categories of features (Fig. 2a, Extended Data Table 2-1). We found variation in model accuracy across brain networks, with the whole-brain model being the most accurate, while FC provided the lowest accuracy compared with other imaging modalities. This suggests that the FC features were less related to chronological age, potentially due to the susceptibility of FC measures to noise, including physiological fluctuations and motion artifacts (Mahadevan et al., 2021). For the cardiovascular system, model accuracy was evaluated when utilizing all features (females, r = 0.78, MAE = 3.57; males, r = 0.76, MAE = 3.91) as well as structural and functional features separately (Fig. 2b, Extended Data Table 2-2).
Table 2-1
Performance of brain age prediction model. Download Multimedia/Extended Data, XLSX file.
Table 2-2
Performance of heart age prediction model. Download Multimedia/Extended Data, XLSX file.
Model accuracy. a, Dot plots show the accuracy distribution of different models utilizing four groups of features to predict age from seven brain networks, subcortical regions, and the whole brain. Dot points are presented separately for females and males, and colors represent different networks. Large dot points with error bars (95% CI) show the average accuracy within each feature group for females and males. r, Pearson’s correlation coefficient; MAE, mean absolute error. b, Bar charts illustrate the accuracy of the model for predicting the age of the cardiovascular system based on three categories of features, stratified by sex. SC, structural connectivity; FC, functional connectivity.
Linking brain and cardiovascular aging
Having predicted age for each individual enabled the computation of an “age gap,” a measure of biological age. We investigated associations between brain and heart age gaps in the 2,904 individuals with brain and cardiovascular data. We examined associations across three scales: (1) a global association between brain and heart age gap estimated based on multimodal imaging phenotypes; (2) network-specific associations between brain and heart age gap estimated based on multimodal imaging phenotypes; and (3) a network-specific and modality-specific association between brain and heart age gap estimated based on specific imaging modalities.
We found a significant positive association between the overall brain (whole-brain) age gap and the overall cardiovascular age gap estimated based on the combination of all features (r = 0.07, p = 7 × 10−5). Associations were also evident between the brain and specific heart feature categories. Specifically, the whole-brain age gap was associated with cardiovascular age gaps estimated based on structural (r = 0.05, p = 0.01) and functional (r = 0.07, p = 6 × 10−5) cardiovascular feature categories adjusting for chronological age and sex (Fig. 3a).
Associations between brain and heart age gaps. a, Scatter plots display the association between the whole-brain age gap derived from all features and heart age gap derived from three categories of heart features, including function, structure, and all. Each data point represents the brain and cardiovascular age gap value for each individual, and solid lines indicate the best regression fit. b, Significant correlations between the functional cardiovascular age gap and cortical networks, as well as the subcortical brain age gap, rendered on cortical and subcortical surfaces. Color intensity indicates the correlation strength. The radar chart shows the correlation coefficient value between the functional cardiovascular age gap and brain age gap. An asterisk (*) indicates a significant correlation (p < 0.05, FDR corrected across 67 tests). All statistics are at the level of networks, not a single region.
Next, we assessed associations between cardiovascular aging and the aging of different cortical brain networks, as well as subcortical regions. Our analyses revealed significant correlations between the cardiovascular age gap calculated using functional phenotypes and age gaps for the somatomotor (r = 0.08, p = 2 × 10−5), salience (r = 0.07, p = 0.0002), and default mode (r = 0.06, p = 0.001) networks and subcortex (r = 0.07, p = 0.0001; Fig. 3b). In contrast, the structural cardiovascular age gap was only significantly correlated with the subcortical age gap (r = 0.05, p = 0.009), but not with other networks.
We then systematically re-evaluated these associations based on age gaps computed for specific brain imaging modalities (Fig. 4, Extended Data Fig. 4-1). For morphology, we found a widespread association with the functional cardiovascular age gap extending across different networks: the salience network (r = 0.07, p = 0.0003), limbic network (r = 0.07, p = 0.0001), dorsal attention network (r = 0.06, p = 0.001), somatomotor network (r = 0.05, p = 0.004), default mode network (r = 0.05, p = 0.005), and subcortical regions (r = 0.07, p = 0.0003; Fig. 4a). In terms of FC, significant associations were observed for the somatomotor (r = 0.05, p = 0.005) and salience networks (r = 0.05, p = 0.007), as well as subcortex (r = 0.07, p = 0.0002; Fig. 4b), despite the relatively lower age prediction accuracy of the FC models for these networks compared with other modalities. Regarding SC, we found a significant association between the functional cardiovascular age gap and the somatomotor network age gap (r = 0.05, p = 0.009), but not with other brain networks (Fig. 4c). The structural cardiovascular age gap exhibited a significant correlation with only one modality-specific brain network age, namely, the visual network calculated using FC (r = 0.05, p = 0.01).
Association between brain and heart age gaps for different imaging modalities. Significant correlations between the functional cardiovascular age gap and cortical networks, as well as the subcortical brain age gap, rendered on cortical surfaces and subcortical slices. Color intensity indicates the correlation strength. In the last column, radar charts show the correlation coefficient value between the functional cardiovascular age gap and brain age gap. a, Brain age gap calculated using morphology features. b, Brain age gap calculated using FC features. c, Brain age gap calculated using SC features. An asterisk (*) indicates a significant correlation (p < 0.05, FDR corrected across 67 tests). SC, structural connectivity; FC, functional connectivity. All statistics are at the level of networks, not a single region.
Discussion
Our work provides new insight into the links between brain and cardiovascular aging in the general population. We established separate models for several brain imaging modalities, including morphology and connectivity, as well as two groups of cardiovascular phenotypes, encompassing cardiac structure and functional capacity. This enabled us to determine the brain and cardiovascular systems that are most strongly linked to aging effects. We demonstrated that the aging of multiple cortical networks and subcortical regions is selectively associated with the aging of the cardiovascular system, with somatomotor, salience, and default mode networks being most predominantly implicated. Notably, our observations also revealed a stronger link between the functional characteristics of the cardiovascular system and the brain compared with their structural aspects. Together, this work suggests an interplay between brain and cardiovascular aging, highlighting specific associations between the aging of distinct brain networks and subcortical regions, and corresponding cardiovascular aging, with functional capacity and physiology showing the strongest effects.
Significant associations between age-related variations in cardiovascular function and the aging of the brain (Jefferson et al., 2010) may be attributed to the interplay between heart function and blood flow to the brain. Given that the brain receives 15–20% of cardiac output in healthy adults (Xing et al., 2017) due to its substantial metabolic demand, the brain is particularly susceptible to altered blood flow delivery. While alterations in blood flow can be temporary and reversible, long-term chronic alterations in blood flow supply due to aging may lead to or exacerbate brain aging. Indeed, a recent study suggests that subtle age-related changes in cardiac hemodynamics may be a risk factor for accelerated cognitive decline and abnormal brain aging (Moore and Jefferson, 2021). In addition, as people age, the structure and function of blood vessels change, resulting in reduced regulation of cerebral blood flow (CBF) and increased vascular resistance (De Silva and Faraci, 2020). These changes may affect oxygenation and nutrient delivery (Benedictus et al., 2017; Ware et al., 2020) and lead to subsequent cerebrovascular diseases and cognitive impairment (Lefferts et al., 2020; Wahl and Clayton, 2024). Future work investigating the role of cerebrovascular aging in mediating the relationship between cardiac and brain aging is needed.
We identified significant associations between cardiovascular aging and subcortical aging. Our findings align with previous studies highlighting key subcortical regions such as the thalamus, hypothalamus, amygdala, and hippocampus, involved in the control of cardiovascular activity (Chang et al., 2016; Tahsili-Fahadan and Geocadin, 2017). Physiological signals originating from the intrinsic cardiac nervous system are transmitted to the brain via ascending pathways in both the spinal column and the vagus nerve. These signals then proceed to the medulla, hypothalamus, thalamus, and amygdala before reaching the cerebral cortex (McCraty, 2016). In line with this circuitry, our results demonstrating links between brain and heart aging were localized to subcortical brain regions that play an important role in regulating heart function (Shivkumar et al., 2016; Tahsili-Fahadan and Geocadin, 2017).
Confirming previously reported connections between whole-brain aging and cardiovascular aging (Goallec et al., 2021; Tian et al., 2023), our study extends these results by showing that the brain–heart link is most prominent within particular brain networks: namely, the salience, default mode, and somatomotor networks. Interestingly, these brain networks are consistent with brain regions and connections implicated in cerebral small vessel disease (CSVD), an age-related disease affecting the function and structure of small vessels in the brain. For example, a study indicates that individuals aged 50 and older with covert CSVD exhibit advanced global and regional brain ages, particularly within the somatomotor and default mode networks (Lee et al., 2022). Furthermore, CSVD is recognized as a slowly progressing condition that primarily disrupts frontosubcortical networks over time (Du and Xu, 2019; Petersen et al., 2020). This suggests that cerebrovascular health plays an important role in the aging of different brain networks.
The specific involvement of the salience network in cardiovascular aging may be explained by the extensive involvement of the anterior cingulate and the insular cortices in modulating cardiovascular autonomic responses via the sympathetic branches (Beissner et al., 2013; Oppenheimer and Cechetto, 2016). For example, studies show that the anterior cingulate cortex regulates blood pressure in reaction to stress (Gianaros et al., 2005) and that damage in the insula is associated with a range of cardiovascular conditions, such as arrhythmia, diurnal blood pressure variation disruption, and myocardial injury (Nagai et al., 2010).
In contrast, the significant association between cardiovascular aging and aging of the default mode network may be related to the default mode network's role in regulating parasympathetic function. For example, a meta-analysis (Beissner et al., 2013) shows that the posterior cingulate cortex, a central node of the default mode network, is associated with parasympathetic outflow, as indexed by high-frequency heart rate variability, which is a metric strongly affected by vagal nerve stimulation of the heart. Our results suggest a link between declining parasympathetic modulation of heart function and aging of the default mode network. We note that the brain sympathetic and parasympathetic networks can be examined by focusing on specific brain regions and gyri that are uniquely associated with them (Lohman et al., 2024). For instance, the parasympathetic network also includes several subcortical regions including the amygdala, ventral tegmental area, and hypothalamus, which are not considered part of the default mode network (Beissner et al., 2013). Future research focusing on sympathetic and parasympathetic networks will enable a better understanding of how autonomic pathways contribute to aging processes in both the brain and cardiovascular system.
Associations between cardiovascular aging and the aging of the somatomotor network were evident across all imaging modalities. The somatosensory cortex, a key region comprising the somatomotor network, is involved in regulating heart function by providing sensory feedback that can trigger autonomic responses (Tahsili-Fahadan and Geocadin, 2017). The significant synchrony in aging between the somatomotor network and cardiovascular system may thus primarily be driven by the somatosensory cortex. Moreover, recent evidence suggests that neuroinflammation (as indexed by free water imaging) in gray matter within the somatomotor network is associated with circulatory cardiovascular biomarkers that are associated with myocardial infarction and heart failure (Ji et al., 2023). This suggests a possibly shared immune aging pathway linking the brain and the heart. Future neuroinflammation studies can provide insights into the immune aging pathway linking the brain and the heart by elucidating shared inflammatory mechanisms.
We found that brain aging is more strongly associated with the aging of the heart's functional capacity and physiology, compared with its structure (e.g., chamber size, myocardial wall thickness). We also found a widespread association between the aging of the heart and the aging of the brain's gray matter, compared with white matter. This suggests that the brain's gray matter system may be more susceptible to inadequate blood flow and insufficient oxygen and nutrition. Given that gray matter neuronal cells are consistently active, a high oxygen supply is necessary for them to function efficiently (Mercadante and Tadi, 2024). Indeed, cerebral blood flow (CBF) to the gray matter of the brain is higher than in white matter. Furthermore, while CBF decreases with age in both gray and white matter, the association between CBF and age is more pronounced in gray matter compared with white matter (Juttukonda et al., 2021). Importantly, cerebrovascular reactivity, which plays a critical role in maintaining cerebral autoregulation to ensure stable CBF over a large range of arterial pressures, declines with age (Peng et al., 2018) and varies between brain regions, with gray matter exhibiting greater vascular reactivity than white matter (Zimmerman et al., 2021). As such, the effect of cardiac aging on the brain may start from the gray matter before spreading to other brain tissues.
Furthermore, CBF can also be influenced by age-related cardiovascular risk factors (Jennings et al., 2013). For example, high blood pressure can cause arteriosclerosis (arterial stiffness), which restricts blood flow and consequently reduces the supply of oxygenated blood to neural tissue (Alexander, 1995; Fabiani et al., 2022). Nevertheless, it is worth noting that white matter aging was investigated using structural connectivity mapped from diffusion MRI utilizing both FA and streamline count in our study. Given the known increased prevalence of white matter lesions (i.e., white matter hyperintensities) in older individuals and those with cardiovascular risk factors, such as hypertension (Launer, 2004; Debette and Markus, 2010), further work utilizing alternative brain imaging modalities (i.e., T2-weighted and FLAIR) that are sensitive to white matter hyperintensities along with structural connectivity measurements is needed to compare the vulnerability of gray matter versus white matter to cardiovascular aging.
Several limitations should be noted. First, we focused on brain–heart connections in healthy individuals. Further exploration within the context of specific cardiovascular diseases or mental health disorders might offer valuable insights. For example, future work should investigate how links between brain and cardiovascular aging manifest in conditions like heart failure, hypertension, depression, or neurodegenerative disorders. Secondly, this study used cross-sectional data. Future investigations, leveraging additional data resources such as long-term longitudinal data, have the potential to monitor changes in both the brain and cardiovascular system as well as their connection over an extended period. Thirdly, at the network level, our focus was on the connectivity within networks, without investigating internetwork connections. Future studies may consider exploring associations between internetwork brain connectivity and heart aging. Fourthly, our analysis relied on the Yeo-7-network parcellation, which assumes each cortical region contributes exclusively to a single network. This simplification may not fully capture the complex, overlapping nature of brain networks. Future studies should consider more flexible parcellation schemes that account for the possibility of regions contributing to multiple networks (Yeo et al., 2014), which could enhance our understanding of the interactions between different networks and cardiovascular aging. Finally, our study was based on a single dataset predominantly comprising participants of white ethnicity. Further work is needed to assess the generalizability of our current findings in other cohorts with a diversity of ethnicity and socioeconomic backgrounds.
In summary, we investigated age-related changes in the brain and cardiovascular system using multiorgan and multimodal imaging. We showed that aging of the brain–heart axis can be localized more strongly to specific brain networks, particularly the salience, default mode and somatomotor networks and subcortical regions, and the functional capacity of the cardiovascular system. Our work may provide a deeper understanding of the intricate relationship between the aging of the brain and the cardiovascular system, opening avenues for comprehending the comorbidity of brain and heart diseases and exploring potential clinical implications.
Footnotes
This research has been conducted using the UK Biobank Resource under Application Number 60698. We thank the UK Biobank for making the data available and to all study participants, who generously donated their time to make this resource possible. This work was supported by The University of Melbourne's Research Computing Services and the Petascale Campus Initiative. Y.A. was supported by Melbourne Research Scholarship. V.C. was supported by a National Health and Medical Research Council (NHMRC) Grant (APP1177370). S.M.L. was supported by an Early Career Researcher Grant from the University of Melbourne. C.S. was supported by the Australian Research Council Grant (DP170101815). A.Z. was supported by an NHMRC Grant (APP118153). Y.E.T. was supported by the Mary Lugton Postdoctoral Fellowship and an NHMRC Investigator Grant (APP2026413).
The authors declare no competing financial interests.
- Correspondence should be addressed to Yalda Amirmoezzi at yamirmoezzij{at}student.unimelb.edu.au.
