Given the aging of the worldwide population, the rising prevalence of both obesity and neurocognitive disturbances among older adults, and considering the economic and social burden of these conditions, unveiling the complex link between obesity and brain health in the mid- to late-life is of utmost importance. However, the pathophysiological pathways linking obesity to brain damage and/or impaired cognition are far from being fully understood, and a number of aspects remain to be clarified. Firstly, obesity is often comorbid with several cardio-metabolic abnormalities, namely hypertension, dyslipidemia, and poor glycemic control, the latter manifesting as either insulin resistance, impaired fasting glucose and/or tolerance, or manifest diabetes (García-García et al., 2022). Moreover, systemic inflammation is often characteristic of this condition (García-García et al., 2022). While neuroanatomical abnormalities in obese subjects with cardio-metabolic conditions have been documented using different brain MRI-derived biomarkers (Alfaro et al., 2018; Bokura et al., 2010; Yaffe et al., 2004), neuroimaging traits are notably under-investigated in metabolically healthy obesity (Angoff et al., 2022; Beyer et al., 2019b; Medic et al., 2016). Therefore, the neuroimaging literature is still inconclusive on whether patterns of aberrant structural/functional findings in brain MRI data are directly attributable to obesity itself, to physiological age-related neurodegeneration, or are possibly mediated through cardio-metabolic and inflammatory dysregulation (Beyer et al., 2019b; García-García et al., 2022). Secondly, conflicting findings have been reported in the literature, especially in the mid- to late- adulthood, on the association between obesity and cognitive disorders, with obesity being linked to both deleterious and protective effects on cognitive functions and dementia risk, a phenomenon often referred to as the “obesity paradox” (Buchman et al., 2005; Hughes et al., 2009; Pedditzi et al., 2016; Qizilbash et al., 2015; Xu et al., 2011).

Thirdly, the employed definition of metabolically healthy obesity and sample demographics typically differs among studies, thus preventing their direct comparison (Eckel et al., 2016).

The purpose of this narrative review is to summarize the current brain MRI literature on metabolically (un)healthy adults with obesity, and to examine the link between obesity and cognitive impairment as assessed through test-based evaluations. Particularly, we aim to elucidate how obesity can coexist with apparently preserved metabolic health and cognitive function despite being widely recognized as a major risk factor for cardiovascular, metabolic, and neurodegenerative diseases. Furthermore, we attempt to shed light on the “obesity paradox” by clarifying whether obese individuals with preserved metabolic control are either at increased or reduced risk for impaired brain health and cognition as compared to their lean counterparts. In addition, we aim to highlight the biological mechanisms that possibly link obesity, its comorbidities, and brain health, with special attention to obesity-induced neuroinflammation and alterations in the blood–brain barrier. Recognizing these early indicators of neurodegeneration holds the potential to improve both disease diagnosis and treatment outcomes.

To reach these goals, we first define the metabolically healthy obese phenotype and describe the most common MRI-derived markers and neurocognitive traits tested for association with obesity (Section “2 Metabolically healthy obesity (MHO)”). We focus on evidence from structural MRI alterations typical of brain aging (including brain atrophy, vascular pathology, and loss of white matter integrity) and on functional MRI literature to examine how obesity relates to alterations in brain networks and task-related activation patterns. We then provide a narrative review of the neuroimaging findings in metabolically healthy obesity, linking them to neurocognitive findings, while speculating on the possible pathophysiological mechanisms underlying them (Section “3 Neuroimaging and neurocognitive findings in obesity and MHO”). Particularly, we highlight the role of inflammatory, hormonal and cerebrovascular mechanisms on the pathogenesis of impaired brain health as described in the neuroimaging and cognitive findings. Finally, we discuss current and future challenges in the field and the potential implications of the collected evidence for personalized public health strategies (Section “4 Discussions and conclusions”).

This review emphasizes that, although metabolically healthy obesity has been hypothesized to represent a relatively benign condition, it might still confer an increased risk for accelerated brain aging compared to metabolically healthy individuals with a healthy weight. This suggests a potential role of adipose tissue as an active endocrine organ damaging the central nervous system even in the absence of overt cardio-metabolic comorbidities. However, the available literature is sparse and partially contradictory, especially with respect to late adulthood, and more studies are needed to elucidate the complex interplay between obesity, metabolic health, brain health, and cognition.

The MRI literature of neuroanatomical changes in metabolically healthy obesity, especially in late adulthood, is relatively scarce, see Table 6. Among the few direct brain MRI studies (Angoff et al., 2022; Nam et al., 2020; Ma et al., 2019), the differences in the definition of MHO and sample selection complicate direct comparisons among studies (García-García et al., 2022; Eckel et al., 2016).

In Angoff et al. (2022), associations of metabolic health and obesity with brain health were investigated in a direct study of early- to middle-aged adults free from prevalent diabetes or brain disorder. The sample consisted of 2170 individuals from the Framingham Heart Study Exam 2 cohort (mean age: 46 ± 9 years, 54% women, MHO/MUO/MHL/MUL: 423/198/1,385/164, predominantly Western European) and included T1-weighted and DTI along with neurocognitive data spanning executive function, memory, abstract reasoning, and visual processing. The MHO group was defined using the National Cholesterol Education Program Adult treatment panel III (NCEP-ATP III) criterion, as in Ma et al. (2019). They reported a lower total cerebral volume (TCV) in MUO and MUL, but not MHO, as compared to the MHL referent group in a multivariable model adjusted for confounders (age, squared age, sex, and interval between risk factors assessments). This finding suggests that poor metabolic health, rather than obesity status, may be the main determinant for global brain atrophy, in line with other studies (Tiehuis et al., 2014). Furthermore, while (Angoff et al., 2022) reported higher Free water (FW) content and lower FA in several cerebral regions on MHO relative to MHL, suggesting a reduced WM integrity, these differences were more circumscribed than those observed when comparing MUL against MHL. This suggests a more limited impact of obesity status compared to metabolic status, or possibly an earlier stage of brain damage (see also Section “4.1 Methodological limitations in the literature”). While the authors detected no evidence of higher WMH burden or silent brain infarcts in MHO, they observed an impaired global cognitive score, verbal memory, and abstract reasoning in MHO but not in MUL, relative to the referent MHL group. This last finding suggests that obesity, more than unhealthy metabolic profile, was associated with poor cognitive performances even in relatively young adults.

Another direct study investigated the effects of the MHO phenotype on cerebrovascular disease markers in 3,165 neurologically healthy Asian adults (age-range: 50–69 years, 46% women, MHO/MUO/MHL/MUL: 63/1099/588/1415) (Nam et al., 2020). This study examined the prevalence of silent brain infarcts in each group after adjusting for demographic, vascular, and inflammatory confounders -age, sex, current smoking, current alcohol use, anti-platelet agents, high-sensitivity, C-reactive protein (CRP), and white blood cell counts, and found that metabolic status, but not obesity status, was associated with prevalence and burden of silent brain infarcts (i.e., higher burden in MUO/MUL relative to MHL, with no difference between MHO and MHL). It should be noted that this study adopted an Asian-specific definition of MHO, using a BMI cut-off of ≥25 kg/m² to define the obesity status, and a stricter definition of metabolically healthy status that notably included the absence of enlarged waist circumference (WC) with a cut-off of 90 cm in men and 85 cm in women. This definition might thus represent a subclass of the obese population with less risky fat distribution: greater subcutaneous adipose tissue (SAT) than visceral adiposity tissue (VAT) proportions (Nam et al., 2020). Thus, the negative findings on the prevalence of silent brain infarcts could be partially attributed to the healthier profile of MHO in the study.

In elderly samples, results have been more mixed. For example, (Ma et al., 2019) examined T1-w MRI data in a sample of 1,199 non-demented elderly individuals (mean age: 73.5 ± 7.1 years, MHO/MUO/MHL/MUL: 415/363/289/132) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a largely North American cohort at the time of the study. Using the NCEP-ATP III definition of MHO, they reported that the MHO subjects (defined using the NCEP-ATP III criterion) showed a decreased AD risk compared with their lean counterpart (MHL), after adjustment for age, sex, genetic (APOE ϵ4) predisposition to dementia, cognitive diagnosis, education, tobacco use, alcohol consumption and LDL-cholesterol (Ma et al., 2019). A further analysis on brain T1-w MRI data reported that whole brain and hippocampal volume were significantly higher in late-life MHO than in MHL groups, thus confirming a protective tendency of the MHO phenotype against the MHL. These findings suggest that, in late life, MHO may paradoxically confer neuroprotection against neurodegenerative processes, even after adjusting for confounders (genetic, clinical, and lifestyle risk factors). The protective tendency of MHO was also corroborated by a higher load of CSF-Amyloid β (Aβ), whose reduction characterizes AD, even after controlling for several possible confounders (Ma et al., 2019).

Of note, existing MHO neuroimaging literature typically does not consider the fact that this status might be a transient phenotype (Ler et al., 2024). To the best of our knowledge, the only longitudinal study on MHO is the one described by Kouvari et al. (2023) only including neuropsychological testing. In Kouvari et al. (2023), 2⋅892 participants from the Framingham Offspring Study (mean age: 60.7 ± 9.4 years, MHO/MUO/MHL/MUL: 234/1152/271/333) were followed longitudinally (12.9 ± 3.5 years) with repeated neurocognitive testing. This study highlighted the transient nature of MHO status, with approximately 70% of participants classified as MHO at baseline developing at least one trait of metabolic disturbance (among dyslipidemia, hypertension and glycemic control) during the follow-up period, and were classified as non-resilient MHO. The remaining participants retained a healthy metabolic status and were thus classified as resilient MHO. The MHO group (defined at baseline) as a whole did not show an accelerated cognitive decline as compared with the metabolically unhealthy groups (MUO/MUL). However, non-resilient MHO showed greater decline in processing speed and executive functions over time compared to resilient MHO participants (Kouvari et al., 2023). Taken together, their findings suggest that obesity or overweight status per se are less harmful than the metabolic status for cognitive functioning, in contradiction with evidence reported in Angoff et al. (2022). In other words, the study suggests that long-term maintenance of metabolic health, not obesity status alone, determines resilience against cognitive decline.

The literature on functional neuroimaging in MHO versus MUO is extremely limited. Most fMRI studies to date have focused on obesity as a whole, without stratifying participants by metabolic status. To the best of our knowledge, a single direct study (Dunn et al., 2023) specifically tagged brain activity in MHO, as compared to either MUO or MHL groups. More specifically, (Dunn et al., 2023) carried out a task-fMRI cross-sectional study on 78 participants (mean age: 39 ± 1 years, MHO/MUO/MHL: 29/34/15) stratified using a combination of criteria which included obesity and updated homeostatic model assessment of insulin resistance (HOMA2). They found that insulin resistance and obesity alter dorsal striatal responses (BOLD signals) in an activation task induced by palatable taste. Aberrant neuronal responses to taste in MHO were intermediate between MHL (primarily positive BOLD response) and MUO (primarily negative BOLD response), suggesting that functional brain alterations may parallel the structural and cognitive differences observed between MHO and MUO. Interestingly, lower palatable taste-induced neuronal activation predicted higher food craving and food intake, in line with previous reports of reduced neural response to pleasant food taste in obese versus lean individuals in dorsal striatum (Babbs et al., 2013; Stice et al., 2008). Dorsal striatum integrates cognitive control signals from the prefrontal cortex and homeostatic inputs from hypothalamic regions, and it is thus central for weight control.

No longitudinal fMRI studies have yet determined whether MHO functional brain profiles predict resilience or eventual progression to MUO. Moreover, few studies combine structural and functional modalities, leaving unanswered the question of whether preserved GM/WM integrity in MHO is mirrored by preserved connectivity patterns.

Overall, the few existing direct comparisons of MHO group against MHL do not provide strong evidence for the harmful effects of obesity status in metabolically healthy individuals on MRI-based markers or cognition, with one study reporting protective effects of MHO relative to MHL (Ma et al., 2019), and one reporting harmful effects of MHO relative to MHL (Angoff et al., 2022). Particularly, the lower WM integrity reported by Angoff et al. (2022) in MHO as compared to MHL group was milder in intensity and less widespread than when comparing MUL versus MHL. This suggests the MHO status as a potential intermediate risk phase, as compared to the MUO status. In terms of cognitive functions, while lower global cognitive score in MHO was reported (Angoff et al., 2022), the longitudinal examination of cognitive function in Kouvari et al. (2023) suggests that those who retain a metabolically healthy status have similar cognitive trajectories as MHL subjects, suggesting a limited impact of overweight/obesity status in the absence of other metabolic impairment. However, it should also be noted that only a minority of participants classified as MHO in Kouvari et al. (2023) remained metabolically healthy in the following decade, suggesting MHO as an often-transient status and that the overweight/obesity status as often leading to metabolic disturbances later in life. Their observation is corroborated by several studies following the weight and metabolic status of individuals over time and demonstrating the higher conversion rate of individuals with overweight and/or obesity from metabolically healthy to unhealthy status compared to normal weight individuals (Bogdanov et al., 2020).

Indirect studies on the MHO are by far more numerous. Among them is a VBM analysis presented by Kharabian Masouleh et al. (2016), which investigated the effects of BMI on GM volume and its cognitive implications on a sample of cognitively healthy subjects with no history of stroke or use of medication affecting the CNS (N = 617, age-range: 60–80 years, 42% women, BMI range: 17–41 kg/m²) from the Life Adult Study, a predominantly European cohort in Germany (Loeffler et al., 2015). After controlling for prevalent cardio-metabolic conditions, medications, APOE, education status, and WMH traits, their analysis revealed a negative association between BMI and GM volume in multiple cortical areas, including the prefrontal, temporal, and occipital cortices, as well as in subcortical structures such as the thalamus, putamen, amygdala and cerebellum (Kharabian Masouleh et al., 2016). In addition, a mediation analysis revealed that lower GM volume partially explained the association between higher BMI and poorer memory performance and impaired attentional processes (Kharabian Masouleh et al., 2016).

The same group carried out a follow-up analysis (Beyer et al., 2019b) on a partly overlapping sample (N = 748, mean age: 68.4 (±4.8), 44% women, BMI range: 17 – 42 kg/m²) of cognitively healthy participants with no major brain pathology from the LIFE-Adult Study cohort. VBM-based GM volume, markers of overall (BMI) and visceral (WHR) adiposity, and metabolic features including markers of glucose metabolism (glycated hemoglobin), lipid metabolism (total cholesterol, HDL), systemic inflammation (CRP, interleukin-6), and adipose-tissue derived hormones (leptin and adiponectin) were tested for covariation while controlling for the confounding effects of age, sex, and total intracranial volume (Beyer et al., 2019b). Rather than investigating the independent effects of highly correlated measures of obesity and metabolic status, they applied a multivariate statistical approach to examine the associations between latent factors underlying different measured variables and the GM volumetry. They demonstrated that the higher values of BMI, WHR, leptin, glycated hemoglobin, CRP and lower levels of adiponectin were jointly associated with widespread patterns of decreased GM volume in temporal, frontal and occipital lobe, sub-cortical regions, and cerebellum, suggesting a shared basis among obesity, inflammation, endocrine regulation of appetite and brain alterations (Beyer et al., 2019b). Particularly, BMI, WHR, and systemic inflammation had the highest influence on GM reduction. This study, however, did not control for genetics and behavioral traits, despite a reduced GM volume and impaired executive functions (particularly impulsive behavior leading to disinhibited eating behaviors) might be both secondary to obesity or act as a risk factor for developing obesity later in life (Beyer et al., 2019b; Chuang et al., 2015; Opel et al., 2015).

In the same year, (Lampe et al., 2019) investigated the contribution of obesity to WMH loads using a whole-brain voxel-based approach on a larger sample extracted from the same LIFE-adult cohort (N = 1,825, age range: 20–82 years, 44% women, BMI range: 18.4–55.4 kg/m²). They found WHR to be robustly associated with deep WMH, after controlling for age, sex, and cardiovascular risk factors including hypertension diagnosis, systolic (SBP) and diastolic (DBP) blood pressure, and smoking. These cardiovascular factors, in turn, were significantly associated with more periventricular WMH. Although BMI was significantly associated with the total WMH volume and the deep to periventricular WMH ratio, the voxel-based analysis did not identify a significant impact of BMI on regional WMH probability, and the effects on the overall WMH volume and regional ratio was uncontrolled for cardiovascular risk factors. The authors in Lampe et al. (2019) also performed mediation analyses and showed that while both BMI and WHR were associated with inflammatory markers like CRP and interleukin-6 (IL-6), only IL-6 significantly mediated the effects of obesity on the WMH ratio. The latter finding aligns with previous studies showing that visceral fat is more linked to inflammation and to microstructural brain damage in deep WM, compared to subcutaneous adiposity (Kim et al., 2017; Widya et al., 2015).

More recently, a comprehensive study conducted on a middle- to older-aged participants of the UK Biobank (N = 23,676, mean age: 62.8 ± 7.5 years, 52.5% women, mean WC: 87.9 ± 12.5 cm) examined the impact of obesity in relation to neurodegenerative diseases and metabolic brain aging (Shen et al., 2023). Obesity, defined by elevated waist circumference (WC ≥ 102 cm in males and ≥88 cm in females), showed the strongest association with structural changes in the brain – namely the higher cortical surface area, lower cortical thickness, and lower subcortical volumes-, after adjusting for demographic factors –such as age, sex, ethnicity, and handedness, as well as brain size, socioeconomic and lifestyle factors-. These associations exceeded the influence of other metabolic syndrome components such as hyperglycemia, hypertension, hypertriglyceridemia, high cholesterol and low HDL levels. This result supports the view that obesity defined by WC has an impact on brain aging, beyond other cardiometabolic conditions. The authors also reported significant intercorrelations among all five metabolic components. Importantly, the overall burden of metabolic dysfunction was associated with the magnitude of brain morphological alterations, characterized by reduced basal ganglia volume, increased cortical surface area (CSA), and decreased cortical thickness (CT) in key regions such as the frontal, temporal, and sensorimotor cortices.

Another UK Biobank study (Dekkers et al., 2019) investigated total body fat (TBF), an adiposity measure assessed by impedance, and its effects on brain structural and microstructural integrity (N = 12,087, mean age: 62 ± 7.3, 53% women, mean BMI: 26.6 (±4.4) kg/m²), stratifying on BMI groups (normal weight, overweight, and obese group). They reported sex-specific effects of TBF, showing a positive association with global cortical volume in women but negative association in men, although there were some specific regional volumes with concordant negative associations in both sexes (in the temporal fusiform anterior cortex and the ventral striatum). Men also exhibited a negative association between TBF and subcortical GM volumes, particularly among the obese group, while in women this association was much weaker and significant only for globus pallidus. In line with this evidence, TBF-by-sex interaction revealed that obese men were more vulnerable to mild cognitive impairment compared to women (Dekkers et al., 2019).

In Caunca et al. (2019a), the link between several indexes of obesity and global MRI-derived markers of brain health was studied in a racially and ethnically diverse North American urban cohort (N = 1,289, mean age: 64 ± 8 years, 60% women) including Hispanic/Latino (66%), non-Hispanic Black (17%), non-Hispanic White (15%), and other races/ethnicities (2%). This study revealed significant associations between greater BMI and WC and a reduced CT after adjustment for sociodemographic, lifestyle, cognitive, vascular and metabolic risk factors. Similarly, CT was reduced in obese (BMI > 30) compared to normal-weight participants (BMI < 25), particularly in those younger than 65 years. Concordantly, weaker negative associations were observed for BMI and WC with total cerebral volume, but not with markers of WMH burden. Interestingly, these negative associations between obesity metrics (BMI and WC) and brain atrophy did not change significantly after adjusting for vascular risk factors, suggesting that these factors did not strongly mediate this association (Caunca et al., 2019a).

Taken together, the evidence from indirect studies suggests that both general obesity, as assessed via BMI (Caunca et al., 2019a,b; Kharabian Masouleh et al., 2016), and abdominal (e.g., visceral) adiposity, as assessed via WC or WHR (Caunca et al., 2019a,b; Shen et al., 2023; Suzuki et al., 2019), are negatively associated with markers or regional and global GM volume. Nonetheless, stronger and more robust associations are reported with single or composite measures of visceral adipose tissue [WC, WHR, and multivariate composite measures as in Beyer et al. (2019b)] rather than BMI alone. This may be due to visceral adipose tissue potentially playing a key role in initiating neuroinflammatory pathways that contribute to neuronal loss, as suggested by Lampe et al. (2019) and discussed in Section “3.4 Biological pathways of brain damage.”

Adipose tissue in the abdominal region can be considered as an active endocrine organ secreting pro-inflammatory cytokines, such as interleukin-6 (IL-6), and inflammation-related proteins, such as CRP (Morys et al., 2021). These cytokines promote a status of chronic low-grade inflammation, endothelial dysfunction, and disrupted fibrinolysis (Hakim, 2019; Kivimäki et al., 2022). These have been shown to increase the permeability of the blood-brain-barrier to circulating cytokines and immune cells, paving the way for neuroinflammation (Chiefari et al., 2021; Yaffe et al., 2004). In turn, neuroinflammation can impair cognition, learning-related skills, and social behavior, potentially leading to neuropsychiatric disorders (Biessels and Despa, 2018). Interestingly, elevated markers of systemic inflammation (such as IL-6 and CRP) have been linked to increased dementia risk (Yaffe, 2007; Yaffe et al., 2004), even decades prior to its occurrence (Schmidt et al., 2002), possibly due to accelerated accumulation of amyloid-β (Abdallah et al., 2021; Keaney et al., 2015; Ma et al., 2018). In accordance with this, markers of systemic inflammation have been also linked to MRI-derived markers of global brain atrophy in middle-aged adults (Gustafson et al., 2003), to GM reduction particularly in the hypothalamus, hippocampus and prefrontal cortex (Marsland et al., 2008), and to macroscopic WM markers of cerebral vessel disease (Hakim, 2019). This is in line with, previous animal studies showing that highly fatty diets induce blood-brain-barrier breakdown, hypothalamic inflammation and subsequent atrophy, whereas calorie restriction reduces systemic inflammation and brain atrophy (Hargrave et al., 2016; Kanoski et al., 2010; Kanoski and Davidson, 2011; Thaler et al., 2012; Willette et al., 2010).

Leptin, adiponectin and ghrelin are signaling hormones playing a key role in energy homeostasis that also influence brain structure and function (Angoff et al., 2022; Beyer et al., 2019a; Han et al., 2021). In the hypothalamus, a group of nuclei respond to these hormonal signals to either promote or suppress feeding (arcuate nucleus), to regulate energy expenditure (paraventricular nucleus), to activate food-seeking and rewards-related behaviors (lateral hypothalamus), and to activate satiety signals (ventromedial hypothalamus) (Purnell and le Roux, 2024). Dysregulation of these nuclei has been consistently observed in both rodent-models of diet-induced obesity and in humans, and linked to abnormal feeding behaviors and energy imbalance (Briggs et al., 2010; Enriori et al., 2007).

Besides being involved in the hypothalamic control of satiety signals and suppression of food intake, leptin also acts in the hippocampus to support memory functions (Beyer et al., 2017; Klok et al., 2007). Increased levels of leptin have been associated with reduced dementia incidence in non-obese adults and larger brain volume (Lieb, 2009). On the other hand, chronically elevated levels of leptin in obesity induce resistance to the effects of the molecule (Myers et al., 2010) and have been linked to impaired executive functions in older adults (Angoff et al., 2022; Spitznagel et al., 2010). In line with this, because of the higher expression of leptin in subcutaneous fat tissues, this hormone has been hypothesized to be a less important neurodegenerative risk factor compared to visceral adiposity (Debette et al., 2010).

Adiponectin has positive effects on the brain due to its insulin sensitizing (Cisternas et al., 2019) and glucose-regulating properties (Lihn et al., 2005). However, existing neuroimaging studies have failed to find evidence of a link between adiponectin and GM volume (Beyer et al., 2019b; García-Casares et al., 2016).

Ghrelin stimulates meal-initiation and thus food intake, but also affects brain regions controlling rewards and mood regulation such as prefrontal cortex and striatum (Bogdanov et al., 2020). Beyond feeding, ghrelin has also been found to have a neuroprotective effect after ischemic brain injury (Spencer et al., 2013) and brain traumatic injury in mice (Shao et al., 2018), possibly due to prevention of blood–brain barrier breakdown and neuronal death. Ghrelin is inversely correlated with BMI (Klok et al., 2007), and its metabolic action appears dysregulated in obesity, possibly due to ghrelin resistance.

Finally, high levels of cortisol, a steroid hormone usually secreted under stress conditions and oversecreted in subjects with obesity (Björntorp, 2001) might lead to premature brain atrophy (Simmons et al., 2000).

Obesity increases oxidative stress - an imbalance between generation and clearance of reactive oxygen species (ROS) and reactive nitrogen species (RNS) – through biochemical mechanisms such as superoxide generation and oxidative glyceraldehyde auto-oxidation (Čolak and Pap, 2021; Wang, 2010). These in turn trigger further deposition of adipose tissue by promoting preadipocyte proliferation, adipocyte differentiation and growth (Čolak and Pap, 2021; Dandona et al., 2010; Higuchi et al., 2013). Oxidative stress also impacts insulin secretion and glucose transport in adipose tissue and muscles, being therefore involved in the development of metabolic disturbances (Čolak and Pap, 2021; Hopps et al., 2010), and has also been linked with increased permeability of the blood-brain-barrier (Čolak and Pap, 2021).

Due to its high metabolic activity, the brain is particularly vulnerable to oxidative damage (Reynolds et al., 2007; Singh et al., 2019). Cerebral oxidative stress causes insults to neurons, such as hypoxia and hypoglycemia, leading to cell injury and neuronal dysfunction in specific population of neurons more vulnerable to age-associated decline -particularly in the hippocampus, amygdala, and prefrontal cortex-, with consequent behavioral and cognitive impairment (Salim, 2017; Wang, 2010). This ultimately leads to neuropsychiatric diseases such as depression (Correia et al., 2023), and to neurodegenerative disorders such as Alzheimer’s, Parkinson’s, and Huntington diseases, as well as amyotrophic lateral sclerosis (Reynolds et al., 2007; Singh et al., 2019; Wang, 2010).

Central obesity – prominently hypertrophic adipocytes - has been linked to the disruption of insulin signaling and insulin resistance, possibly via diet-induced inflammation (Medic et al., 2016; Vachharajani and Granger, 2009; Velloso, 2009). Notably, neuronal cells influenced by insulin can affect critical CNS functions, including neurotransmission, synaptic plasticity, and neuroprotection (Fanelli et al., 2022; Klinedinst et al., 2019). Disturbed insulin signaling pathway and consequent glucose dysregulation have been also linked to impaired memory and to GMV reduction in key memory regions such as the hippocampus and the temporal lobe (Beyer et al., 2019a; Cherbuin et al., 2012; Kerti et al., 2013). Concordantly, disturbed insulin signaling has been thought to be implicated in the pathogenesis of neurodegeneration (Wrighten et al., 2009), particularly of AD (Kroner, 2009).

Overall, this evidence suggests a pathway linking obesity and cognitive impairments, through cardio-metabolic dysfunction and neuroanatomical alterations.

Obesity has also been linked to maladaptive changes of the vasculature (O’Rourke and Safar, 2005), including arterial stiffness, thickening of the carotid wall, ventricular hypertrophy, vascular endothelial dysfunction, hypertension, and ischemia (Taki et al., 2008). These alterations can compromise cerebral perfusion and damage brain tissues. In particular, cerebral vascular pathology has been associated with decline in GM tissues, especially in the hippocampus and lateral temporal lobe (de Toledo Ferraz Alves et al., 2010; Tini et al., 2020). Each standard deviation increase in carotid-femoral pulse wave velocity corresponds to a reduction in total cerebral volume (TCV) equivalent to ∼1.2 years of accelerated brain aging (Tsao et al., 2013).

In Morys et al. (2021), cerebrovascular alterations were also associated with cognitive impairments and the link between BMI and working memory was partly mediated by WMH load (9%), as were the associations of WHR with working memory (7%) and fluid intelligence (21%), and the association between BF% and working memory (9%). In WM, vascular dysfunction might lead to demyelination, loss of oligodendrocytes, and gliosis (Morys et al., 2021), which might accumulate in the years and result into increased WMH load and/or into a sensible change in other markers of WM pathology, such as silent cerebral infarcts (de Toledo Ferraz Alves et al., 2010; Tini et al., 2020; Tsao et al., 2013).

Vascular stiffening might also lead to hypoperfusion and hypoxia, and cause cerebral tissue damage (Lampe et al., 2019). A significant clinical consequence of cerebrovascular pathology is the onset of cognitive impairment and increased risk of dementia, even in otherwise healthy individuals (Friedman et al., 2014; Furlano et al., 2021; Wrighten et al., 2009). A recent study in the field (Siedlinski et al., 2023), combining genetic and observational evidence, led to similar conclusions, supporting a negative influence of elevated blood pressure on cognitive performance also through genetic causal inference methods (Mendelian randomization). Notably, the latter study also identified deleterious effects of systolic blood pressure on t neuroimaging markers of cognitive function, especially in the anterior thalamic radiation, the anterior corona radiata, and the external capsule (Siedlinski et al., 2023).

Among the instruments summarized in Table 5, the Montreal Cognitive Assessment (MoCA) stands out for its greater sensitivity compared with the Mini-Mental State Examination (MMSE) in detecting subtle deficits, particularly in fronto-executive domains (Freitas et al., 2013; Nasreddine et al., 2005). Because obesity and metabolic dysregulation are thought to preferentially affect prefrontal circuits involved in executive control (Rösch et al., 2020), the use of the MoCA or other sensitive instruments is essential to capture early brain–behavior changes.

The lack of standardized cognitive batteries, combined with the frequent use of screening tools with different cut-offs and follow-up periods, further complicates comparisons across studies and may partly explain the variability of neurocognitive findings. Therefore, when interpreting cognitive outcomes in this literature, it is important to consider both the psychometric properties of the instruments and the specific domains they assess, as these factors strongly influence the detection of subtle deficits and the characterization of longitudinal cognitive trajectories in individuals with obesity.

Despite its wide use, the definition of obesity via BMI has some obvious limitations since it cannot discriminate lean from fat mass (body composition), nor subcutaneous adiposity from either visceral or ectopic fat deposition (body-fat distribution) (De Lorenzo et al., 2019; Hainer and Aldhoon-Hainerová, 2013; Kassir et al., 2023; Macek et al., 2020; Smith et al., 2019; Tchernof and Després, 2013). This is especially important when studying the elderly (Huxley et al., 2008; Kim et al., 2017), since visceral adipose tissue deposition tends to increase significantly with age (Smith et al., 2011; Tchernof and Després, 2013) while skeletal muscle mass is progressively reduced (i.e., sarcopenia) (Guglielmi et al., 2016). Moreover, there is a difference between males versus females, due to body composition, when computing BMI (Pray and Riskin, 2023). Consequently, other obesity indexes like the accumulation of visceral adipose tissue (VAT) measured from abdominal CT images and the % BF computed from Dual-Energy X-ray Absorptiometry, have been proved to be better predictors of clinical and neurocognitive health outcomes (Kim et al., 2017), especially in the case of late-onset obesity (Hainer and Aldhoon-Hainerová, 2013).

Previous reviews have highlighted that fat distribution cannot be overlooked, as excess visceral and ectopic adiposity (central obesity), more than subcutaneous fat accumulation (peripheral obesity), are linked to dyslipidemia, proinflammatory and prothrombotic activity, increased risk for cerebrovascular disease, more severe microstructural brain damage and more adverse cognitive outcomes (Hainer and Aldhoon-Hainerová, 2013; Iacobini et al., 2019; Kim et al., 2017; Lampe et al., 2019; Tchernof and Després, 2013; Widya et al., 2015). Interestingly, (Debette et al., 2010) reported a negative association between VAT and total brain volume, independent of BMI in middle-aged community participants.

The prevalence of MHO has been reported to range from approximately 6% to 75% of the total adult obese population (Eckel et al., 2016; Ma et al., 2019; Rey-López et al., 2014; Smith et al., 2011), depending on the employed criteria applied. In other words, the definition of the MHO phenotype is not unique, mostly due to differences in the number and severity of metabolic abnormalities included across the different definitions. Therefore, this represents an aspect which might have contributed to the high heterogeneity of findings reported in the obesity literature (Eckel et al., 2016; Gómez-Zorita et al., 2021; Ma et al., 2019; Rey-López et al., 2014; Smith et al., 2011).

Furthermore, the most commonly used definitions of MHO phenotype suffer from the inherent limitations of dichotomous classifications: (1) a small variation in one parameter can lead to classify a subject into an opposite class/category; (2) subjects within the “metabolically unhealthy” class are considered as equally affected (same level of “sickness”), without taking into account the number or severity of metabolic traits; (3) subjects within the “obese” class are considered equivalent in terms of adiposity (same level of “fatness”), without taking into account its severity and distribution; (4) subjects which do not meet yet a clinical diagnosis for any metabolic trait but express sub-clinical disturbances are all classified as “metabolically healthy.”

Finally, the two most common MHO definitions in the neuroimaging literature (namely the NCEP-ATP III and the IDF criteria) do not incorporate markers of systemic inflammation and/or hormonal imbalance, neglecting the importance of these molecular mechanisms as possible pathways underlying brain damage in metabolically healthy and unhealthy obesity.

Adipose tissue tends to distribute differently between sexes, with visceral fat more common in men and subcutaneous fat more common in women (Nakamura et al., 1994; Taki et al., 2008). Furthermore, women tend to be more insulin-sensitive than men, possibly due to their higher levels of adiponectin, and/or to the effect of estrogens onto insulin and glucose homeostasis (Cavalieri et al., 2010; Geer and Shen, 2009; Kim, 2024). Although prevalence estimates vary depending on the diagnostic criteria employed, the MHO phenotype is more common in women than in men (Smith et al., 2019). As a further confirmation, previous neuroimaging studies have suggested that the relationship between GM atrophy and obesity might be modulated by gender (Kurth et al., 2013; Taki et al., 2008). This calls for a more careful treatment of sex, which may likely act as an effect modifier.

Different MRI markers are sensitive to different anatomical/functional changes which might happen at different times and at a different spatial (microscopic versus macroscopic) scale. As an example, DTI-derived markers of WM integrity can detect earlier and more spatially subtle alterations than cSVD markers from T2w-MRI, which typically indicate macroscopic WM anomalies. More so, cSVD induces protracted stenosis of small arteries, that leads to hypoperfusion of the brain, resulting first in small cerebral infarcts and WMH loads, and only later into cerebral atrophy due to the accumulation in time of brain damage (Hakim, 2019). It follows that inconsistent neuroimaging findings might partially be due to the timing that subjects have been imaged during their lifespan, and to the MRI-modalities used to examine anomalies.

Most of the reviewed studies are cross-sectional (Table 6). This approach underlines the assumption that MHO is a stable rather than transient state, despite previous epidemiological evidence indicating the transitions for many individuals either in metabolic health or obesity status over time (Bobbioni-Harsch et al., 2012; Eckel et al., 2016; Janssen et al., 2004; Ler et al., 2024; Soriguer et al., 2013). Also, subjects with a later onset of metabolic dysfunction and shorter exposition to the obesity insult might be representative of an earlier stage of the trajectory, and this in turn might explain different neuroimaging (and neurocognitive) findings (Hainer and Aldhoon-Hainerová, 2013). Furthermore, cross-sectional studies implicitly assume that the observed neuroanatomical, neurofunctional, or cognitive alterations are consequences of the obesity insult. However, emerging evidence suggests that abnormalities in rewards- and control-related regions (e.g., such as atrophy and/or altered activation in the prefrontal, orbitofrontal cortex, striatum, and anterior cingulate) might actually precede obesity. These abnormalities might then predispose subjects to overeating behavior and impaired self-regulation, and thus to weight gain (Lowe et al., 2019; Stice et al., 2008; Yokum et al., 2011). This could further exacerbate brain damages in these same regions (Lowe et al., 2019; Stice et al., 2008; Yokum et al., 2011).

In this view, longitudinal studies may help to untangle the temporal direction of effects and clarify if executive dysfunction might be a cause of obesity, a consequence, or both. Statistic approaches that take advantage of genetic instruments, such as Mendelian randomization, may allow us to clarify the causal relationship between obesity sub-types and brain trait variability, in a way free of typical biases of observational studies like residual confounding and reverse causality (e.g., increased mortality among subjects with more risky fat distributions, and weight loss associated with a preclinical dementia status - see Section “4.3.2 Recommendations for future research: toward a precision obesity approach”).

It has been postulated that subjects with a less favorable obesity, both in terms of degree and fat distribution, tend to have a shorter life expectancy than obese subjects with less risky (subcutaneous lower body) obesity patterns (Beyer et al., 2017, 2019a). Furthermore, unintentional weight loss has been reported to precede AD diagnosis in older adults (Johnson et al., 2006), and to be exacerbated as dementia progresses (Franx et al., 2017). Age and other differences in the sample demographics, as well as limited sample sizes, might therefore partly account for the discrepancies in the literature (Kharabian Masouleh et al., 2016). In addition, since obesity is associated with increased head motion during scanning and thus image artifacts, special attention should be paid to control for head micro-movements during image acquisition (Medawar and Witte, 2022). Education, smoking habit, and cardiorespiratory fitness status are further factors that might confound the study of possible links among adiposity metrics, MRI markers and cognitive performances, as are the possible concurrent presence of other clinical/subclinical conditions (e.g., depression) and related medication intake, which are not accounted for by the most used MHO definitions (Kharabian Masouleh et al., 2016). Finally, collider stratification bias, survivorship bias and other potential methodological aspects may explain the obesity paradox and definitely deserve careful consideration (Banack and Kaufman, 2013).

Most neuroimaging studies on obesity and brain health have been conducted on Caucasian cohorts from high-income countries, mainly in North America and Europe. Compared to Caucasian and European populations, Asian populations typically exhibit higher central obesity for each stratum of BMI (Huxley et al., 2008), which may lead to different susceptibility to metabolic complications and brain tissue damage (Kim et al., 2017). Similar ethnic differences in body composition, fat distribution, and metabolic profiles exist in African, Hispanic/Latino, and Indigenous populations (Martos-Moreno et al., 2020; Justice et al., 2021). Besides minority ethnic groups, individuals from lower socioeconomic backgrounds face higher risks of obesity, metabolic disturbances, and cognitive aging possibly due to dietary, cultural, and/or environmental factors (García-García et al., 2022; Tchernof and Després, 2013), yet they are typically under-represented in neuroimaging studies. This ethnic, geographic, and socioeconomic imbalance limits the generalizability of findings to these populations. Beyond methodological limitations, this lack of diversity also raises ethical concerns related to health equity.

Weight-loss interventions, whether behavioral, pharmacological, or surgical, are often followed by compensatory “rebound” responses that tend to restore body weight (Contreras et al., 2019; Strohacker et al., 2014). These include reduced resting energy expenditure, increased appetite mediated by ghrelin, and decreased satiety hormones such as leptin and GLP-1, as well as heightened activity of hypothalamic–pituitary–adrenal stress pathways (Aschbacher et al., 2014; van Loenen et al., 2022). Such homeostatic adaptations favor rapid weight regain and recurrent metabolic stress, which in turn may exacerbate systemic inflammation, insulin resistance, and cerebrovascular dysfunction (Phuong-Nguyen et al., 2024). Repeated weight cycling (“yo-yo” dieting) has been associated with greater visceral fat accumulation and impaired WM integrity, potentially accelerating age-related brain atrophy and cognitive decline. Considering rebound physiology is therefore critical when interpreting neuroimaging findings in obese individuals, particularly in those undergoing weight-loss interventions. For these reasons, our review does not also cover samples of individuals who have undergone significant weight-lost interventions.

While it is acknowledged that obesity, with or without other cardio-metabolic comorbidities, might cause measurable neuroanatomical and neurocognitive damage, the possible influence of neuroanatomical and neurocognitive alterations on incident obesity remains an open and intriguing question. Indeed, it has been hypothesized that anatomical changes in the prefrontal cortex, a brain region supporting executive functions and under genetic control (Chouinard-Decorte et al., 2014), might precede the development of overeating - through deficits of self-regulatory control over food intake - and thus obesity (García-García et al., 2019). These same neuroanatomical alterations might then be further exacerbated by the obesity insult in a vicious cycle of progressive obesity severity, brain dam-age and cognitive decline (Hargrave et al., 2016). Similar positive feedback has been hypothesized for the hippocampus (Hakim, 2019), possibly initiated by hippocampal dysfunctions due to congenital deficits, early exposition to highly-caloric fatty diets, and/or to environmental toxin exposition (Hargrave et al., 2016; Parent et al., 2014; Shefer et al., 2013). In turn, hippocampal dysfunctions might cause impaired memory inhibition leading to unhealthy responses to food cues, consumption of energy dense food, and in turn weight gain (Dye et al., 2017).

Most of the current knowledge of neuronal control on eating behaviors is coming from animal studies. In humans, task-based functional MRI studies have reported differences in the activation patterns of individuals with obesity when presented with food stimuli, that possibly would make them more inclined to high-caloric food craving and thus to obesity (Rothemund et al., 2007). Aberrant brain networks and related cognitive functions - such as rewards evaluation and eating behaviors - are also important aspects in obesity research. Indeed, network connectivity studies from resting-state MRI have found that BMI is negatively associated to posterior default mode functional connectivity in older adults - independently of possible cardio-metabolic comorbidities, gray matter volume and APOE genotype - and with worse executive functions and outcomes (Beyer et al., 2017). Interestingly, decreased posterior default mode connectivity has been also reported in cognitively normal adults at increased genetic AD risk (APOE-e4 carriers) (Sheline et al., 2010), and in subjects presenting mild cognitive impairment (Sorg et al., 2007), suggesting that obesity might be associated to patterns of functional connectivity abnormality similar to those observed in individuals at AD risk (Beyer et al., 2017; Sheline et al., 2010).

To avoid the aforementioned shortcomings of current neuroimaging literature on the MHO phenotype (see Sections “4.1 Methodological limitations in the literature” and “4.2 Strengths and limitations”), we recommend that future studies should be carried out in large multicenter cohorts and in a longitudinal setting, following in time both MHO subjects who will convert into metabolic unhealth (MHO converters) and those who will not (MHO resilients) (García-García et al., 2019). The use of multi-modal (including both structural and functional) and multi-scale (both microscopic and macroscopic) brain imaging, might allow a more comprehensive understanding of the shared underpinnings of obesity, metabolic regulation, and brain health (Petersen et al., 2024). In particular, more functional MR imaging studies (such as rs-fMRI and task-MRI), which remain underutilized in this field, might fill critical gaps in understanding network-level brain changes in obesity. Also, systematic meta-analyses might provide complementary information on the effect sizes and direction of effects across different cohorts and imaging modalities. Furthermore, we recommend the use of a standardized criterion to define the MHO phenotype that would capture the truly multifaceted nature of obesity, especially of visceral and ectopic adiposity, and related metabolic comorbidities, in each of its facets/dimensions: severity and distribution of fatness, vascular health, glycemic control, dyslipidemia, low-grade systemic inflammation, hormonal dysregulation and resistance, ethnicity, cardio-metabolic fitness, lifestyles. Given the limitations of BMI as a proxy for adiposity (see Section “2 Metabolically healthy obesity (MHO)”), we also recommend that future neuroimaging research on obesity adopt more biologically informative measures of obesity, such as waist circumference, waist-to-hip ratio, or imaging-based assessments of visceral and subcutaneous fat distribution. Stratifying future analyses by sex/gender or other potential modifiers (e.g., polygenic influences on obesity, neuroimaging traits, eating behaviors, and socioeconomic status) might also improve the sensitivity of future studies. Similarly, integrating genetic information through the use of polygenic scores or, even better, instrumental variables in two-sample Mendelian randomization studies, might help uncover possible causal links in this complex relationship.

Studying obesity status through these lenses and its interplay with brain imaging endophenotypes may contribute to furthering our understanding of the pathological processes associated with neurodegenerative and neuropsychiatric diseases. Most importantly, it may represent one step forward in establishing features for the early diagnosis of neurodegenerative conditions, which currently suffer from the lack of risk-predictive biomarkers. If MHO converters, a subset of the adult MHO population at high risk of developing brain health complications, could be identified early enough, this information could be used in clinical settings. Also, targeted public health strategies could be planned for weight loss in these subjects, before brain (and cardio-metabolic) health is irreversibly impaired. The use of risk-stratified treatment for obesity may possibly lead to a consequent reduction of the prevalence and burden of both obesity and of its neurological and neuropsychiatric sequelae, which have currently reached epidemic diffusion in most populations, especially among elderlies, with unprecedented costs for most welfare and healthcare systems.