All participants were part of the IMAP+ cohort (“Imagerie Multimodale de la Maladie d’Alzheimer à un stade Précoce,” Caen, France, ClinicalTrials.gov: NCT01638949), a multimodal neuroimaging study of early Alzheimer’s disease. In the longitudinal study protocol, over up to three timepoints (baseline/T1; 18-months follow-up/T2; and 36-months follow-up/T3) VRF, MRI, PET imaging and neuropsychology are assessed in close temporal proximity (within 3 months). Given the longitudinal and multifactorial approach of the current analyses, we selected 95 IMAP+ participants, who had at least two valid structural and functional MRI sessions to assess changes in functional connectivity over time and at least one assessment of each VRF measure of interest, WMH and florbetapir-PET across the study duration. Figure 1 provides an overview of the current study.

The present study sample comprised 43 cognitively unimpaired older adults (OA), 18 participants with subjective cognitive decline (SCD) and 34 participates with mild cognitive impairment (MCI). By including participants along the cognitive continuum with normal cognition, SCD, and MCI, we ensured that there was enough variance due to VRF and brain pathology, including of Aβ and WMH, in our sample (Wirth et al., 2018; Gaubert et al., 2021) to assess the proposed associations. Other important reasons to include the asymptomatic or preclinical stages (i.e., SCD and MCI) are that these are clinically relevant in terms of early treatment and some mechanisms may specifically appear in stages before the onset of clinical symptoms.

All IMAP+ participants were French-speaking, educated for >7 years, right-handed, aged between 51 and 85 years and had no history or clinical evidence of major neurological or psychiatric disorder, alcoholism, drug abuse, or head trauma. Uncontrolled diabetes mellitus was leading to exclusion. Participants underwent a clinical and neuropsychological examination, assessing individual’s health and multiple domains of cognition (see further details below) by a multidisciplinary team of senior neurologists and neuropsychologists. Based on this evaluation and according to internationally agreed criteria, the inclusion and group classification of the participants were conducted (McKhann et al., 1984; André et al., 2019).

The OA were recruited from the community and had cognitive performance in the normal range (i.e., within –1.65 standard deviation [SD] of the normal mean) given their age, sex, and education level in all screening tests. Participants with SCD and MCI were recruited from the local memory clinics with self-reported cognitive concerns, which were not linked to current medication, psychiatric or neurological diseases (including anxiety or depression), or other medical conditions. The patient’s independence in daily life needed to be preserved and they should not meet NINCDS-ADRDA criteria for probable AD (McKhann et al., 1984). The SCD patients reported subjective cognitive concerns in the clinical interview and in a cognitive complaint questionnaire (McNair and Kahn, 1983), in the absence of objective cognitive impairment on the neuropsychological screening as described previously (Perrotin et al., 2017; Kuhn et al., 2019). The MCI patients showed objective cognitive deficits, predominantly in episodic memory, fulfilling the clinical criteria for single or multiple domain amnestic MCI (Petersen and Morris, 2005).

The IMAP+ study was approved by the local ethics committee (Comité de Protection des Personnes Nord-Ouest III) and all participants received detailed study instructions and gave written informed consent prior to study participation.

We assessed VRF of blood pressure, body-mass-index (BMI) and levels of hemoglobin A1c (HbA1c, long-term glucose marker). Systolic and diastolic blood pressure were measured three times consecutively at MRI or/and PET visit in a seated position. In the first and latter case, blood pressure measures were averaged over 3 (n = 27) or 6 (n = 68) assessments, respectively. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared (kg/m²). The HbA1c level were measured after participants underwent a fasting blood draw. The measures were reported in our previous studies (Garnier-Crussard et al., 2020, 2021) and corresponds to the 3-month averaged blood sugar level and is related to diabetes mellitus risk. Vascular risk factors were all assessed at baseline.

Participants underwent a MRI up to three times across the study duration (T1-T3) on a 3T Achieva scanner (Philips, Netherlands) at the Cyceron Center (Caen, France). Structural high-resolution T1-weighted images were acquired first using a 3D fast field echo (FFE) sequence [3D-T1-FFE sagittal; 180 slices, no gap, slice thickness = 1 mm, in-plane resolution = 1 × 1 mm², field of view = 256 × 256 mm², matrix = 256 × 256, repetition time (TR) = 20 ms, echo time (TE) = 4.6 ms, flip angle = 20°]. High-resolution T2-weighted FLAIR images were obtained subsequently (3D-IR sagittal sequence, 90 slices, no gap, slice thickness = 2 mm, in-plane resolution = 0.78 × 0.78 mm², TR = 8000 ms, TE = 348 ms, flip angle = 90°, field of view = 250 × 250 mm², matrix = 320 × 320).

Resting-state functional MRI (rs-fMRI) scans were obtained using an interleaved 2D T2* SENSE EPI sequence designed to reduce geometric distortions using parallel imaging, short echo time, and small voxels (2D-T2*- FFE-EPI axial, SENSE = 2; 42 slices, no gap, slice thickness = 2.8 mm, in-plane resolution = 2.8 × 2.8 mm², field of view = 224 × 224 mm², 280 volumes, TR = 2382 ms, TE = 30 ms, flip angle = 80°, acquisition time = 11.5 min). The participants were asked to keep their eyes closed without falling asleep and to remain as still as possible during scanning. Further, participants were equipped with earplugs, their head motion was minimized by stabilizing foam pads and the scanner room was darkened. Using a subsequent debriefing questionnaire, we assured that the participants has no difficulty staying awake throughout the rs-fMRI scan and that nothing particular had disturbed their attention during the scanning.

The PET scans with florbetapir ([¹⁸F]AV45) were acquired on a Discovery RX VCT 64 PET-CT scanner (General Electric Healthcare, United States) at the Cyceron Center (Caen, France) with a resolution of 3.76 × 3.76 × 4.9 mm³ (field of view = 157 mm) to measure Aβ deposition within the brain. Approximately 4 MBq/kg of AV45 were injected intravenously and 47 planes with a voxel size of 1.95 × 1.95 × 3.2 mm³ were obtained at 50–70 min post-injection. A transmission scan was performed for attenuation correction before the PET acquisition.

The AV45-PET images were linearly co-registered to the corresponding T1-weighted MRI image and warped onto the MNI template, using the deformation matrices derived from the “Segment” procedure (described under MRI analyses). Standardized uptake value ratios (SUVR) were computed by dividing the cerebral tracer uptake by the tracer uptake of the whole cerebellum serving as a reference region. Mean SUVR values were extracted from brain regions typically affected in AD, using a modality-specific AD meta-signature mask, including all neocortical regions, but excluding para-hippocampi, pre- and post-central gyri, and occipital cortices (Besson et al., 2015). For the current analyses, the first available timepoint with a valid measurement of Aβ-PET SUVR was selected: n = 77 at T1 and n = 18 at T2.

A comprehensive neuropsychological test battery, including tests of episodic, semantic and working memory, language skills, and executive and visuospatial functions as previously described (Besson et al., 2015), was administrated to all participants longitudinally. Participants included in the current study, underwent the cognitive assessment at least two times and up to three times across 3 years of study duration (Table 1).

To obtain robust proxies of domain-sensitive cognitive performance, four individual cognitive composite scores were calculated, assessing executive function, episodic memory, processing speed, and working memory, as described in our previous study (Garnier-Crussard et al., 2020). To create composite scores, the individual cognitive scores were z-transformed (i.e., using OA of the IMAP+ sample as the reference group) and averaged. Composite scores were computed in participants with at least two corresponding individual scores. Z-scores derived from reaction times (i.e., TMT and Stroop) were reversed before being averaged, such that higher values always indicated better performance.

In brief, the executive function composite included the phonemic verbal fluency test score (Letter P), a flexibility score calculated on the Trail Making Test (TMT) (time difference between parts B and A divided by time of part A), and a score of inhibition computed from the Stroop test (time difference between interference and naming tasks) (Godefroy, 2008). The episodic memory composite included immediate free recalls and delayed free recall of the Free and Cued Selective Reminding Test (FCSRT) (Grober et al., 1988), memory subscore of the Mattis Dementia Rating Scale (Mattis, 1976), and free recalls of the “encoding, storage, retrieval” paradigm (Eustache et al., 2015). The processing speed composite included the TMT part A score (times), and color reading and color naming of the Stroop test (Godefroy, 2008). The working memory composite included scores of forward and backward digit span (Petermann and Wechsler, 2012).

Statistical analyses were performed with R 4.0.3 and R-Studio 1.3.1093 (The R Foundation). Two-sided P values less than 0.05 were considered to be statistically significant. The statistical analysis procedure was carried out in close correspondence with the previous study (Köbe et al., 2021). Overall, effect sizes were expected to be subtle and the previous findings (Köbe et al., 2021) clearly supported a priori hypotheses with respect to higher-order brain networks that are susceptible to VRF, aging and AD, thus no corrections for multiple comparisons were applied.

Our total sample was first characterized by demographic factors, VRF levels and biomarkers of brain pathology. Baseline differences between diagnostic groups (OA, SCD, and MCI) were investigated using χ² (Chi-Square) tests for categorical variables. One-way analyses of variance (ANOVA) or Kruskal-Wallis test were used for comparisons of parametric and non-parametric continuous variables, respectively, followed by Tukey’s multiple comparisons tests and Wilcoxon rank sum tests for corresponding pairwise comparisons. Visual inspection for approximate symmetric distribution of the data was done on QQ-plots. For descriptive purpose, abnormal levels of VRF were indicated for all participants using established cut-off thresholds (World Health Organization, 2000, 2006; Whelton et al., 2018). For all statistical analyses, the measures of BMI, HbA1c and WMH volumes were log-transformed to reduce skewness and to achieve a near normal distribution. Subsequent analyses were conducted across all diagnostic groups to map the anticipated trajectories along the cognitive continuum with range of older people free of dementia, but some of whom are at increased risk.

We ran linear mixed-effects models (LME, lmer function of the lme4 package) (Bates et al., 2015) to characterize the overall change in network/global RSFC over up to three years, and its relation to peripheral VRF (i.e., systolic and diastolic blood pressure, BMI, HbA1c), WMH volume and brain Aβ deposition. All of these independent variables of interest showed no multi-collinearity (variance inflation factor < 5) and were therefore entered in the same statistical model. Next to these variables of interest, baseline age, sex, gray matter volume (adjusted for TIV), diagnostic status and mean FD at each visit were included as potential confounding factors (covariates). Follow-up time of rs-fMRI was operationalized individually as years from baseline measurement (below referred to as: time). The following equation was used for these models: longitudinal network/global RSFC ∼ systolic blood pressure × time + diastolic blood pressure × time + BMI × time + HbA1c × time + WMH × time + Aβ-PET SUVR × time + covariates × time. Selected interaction effects were graphed to evaluate the directionality of the associations. Continuous measures were thereby divided into tertiles, only for the purpose of improved visualization.

We also investigated, if significant associations between the VRF of interest and changes in RSFC over time were moderated by Aβ burden. Therefore, we dichotomized participants with an Aβ-PET SUVR ≥ 1.22 into Aβ-positive individuals and < 1.22 into Aβ-negative individuals, using thresholds that were established and described in our previous study (Vanhoutte et al., 2021). Then the models described above were rerun, except that the term “Aβ-PET SUVR × time” was replaced by a 3-way interaction between the vascular risk factor of interest (i.e., that was significantly associated with RSFC in the first models), Aβ positivity status and time.

Finally, an exploratory analysis was conducted to assess whether or not changes in RSFC that were significantly related to our VRF measures, were linked to changes in individual cognitive performance over time. We thus extracted the predicted slopes of RSFC (change over time) from the first LME models (see above) and included these as dependent variables, as well as their interaction with time, in subsequent LME models. Longitudinal cognitive performance was operationalized by four composite scores (described above). Age, sex, education and their interaction with time were included as covariates. The following base equation was applied: longitudinal cognition ∼ RSFC slope × time + covariates × time. Follow-up time of cognition was operationalized individually as years from baseline cognitive assessment.

In all LME models, we included individual RSFC or cognition intercepts and slopes as random effects long with the fixed effects of variables of interest (i.e., VRF, WMH, Aβ-PET SUVR, and RSFC slopes) and covariates. Models included both all main effects as well as their interactions with time. All continuous variables were z-transformed prior to model estimation. All models used the restricted maximum likelihood method and were fit with an unstructured variance-covariance and type III sum of squares. Denominator degrees of freedom were calculated with the Satterthwaite approximations.

Participant characteristics are summarized in Table 1. Diagnostic groups were comparable regarding sex distribution, blood pressure (systolic and diastolic), BMI, HbA1c, WMH volumes and assessment timepoints of rs-fMRI (all p’s ≥ 0.1). There were significant differences in age and education across diagnostic groups (ANOVA: p’s < 0.05). The MCI patients showed a higher mean Aβ-PET SUVR (Kruskal-Wallis: p = 0.003; OA vs. MCI p = 0.003; SCD vs. MCI p = 0.036) and lower GM volume (ANOVA: p = 0.032; SCD vs. MCI p = 0.024). Across the total sample, abnormal levels of VRF were recorded for systolic blood pressure (≥130 mmHg, n = 70 [74%]), diastolic blood pressure (≥80 mmHg, n = 45 [47%]), body mass index (≥30 kg/m², n = 8 [6.5%]) and HbA1c (≥ 6.5%, n = 3 [3%]).

Across the entire sample and time, we found an overall increase in RSFC over time within the DMN (β = 0.060, standard error [SE] = 0.026, t = 2.31, p = 0.024), the SAL/VAN (β = 0.069, SE = 0.024, t = 2.86, p = 0.005), FPN (β = 0.095, SE = 0.030, t = 3.13, p = 0.002), VIS (β = 0.090, SE = 0.037, t = 2.42, p = 0.018) networks and in global RSFC (β = 0.047, SE = 0.023, t = 2.05, p = 0.043), adjusted for age, sex, diagnosis and mean FD (Table 2, extended table is provided in the Supplementary Material in Supplementary Table S1). No significant overall increase in RSFC over time was observed within the LIM, DAN and SM networks (all p’s ≥ 0.05).

Next, we assessed how longitudinal change in RSFC was related to VRF. We found negative interactions for diastolic blood pressure (dBP × time) and for that HbA1c (HbA1c × time). Visualization of these significant interactions (Figure 2) indicated that higher levels of diastolic blood pressure were more likely to be associated with a decrease of RSFC over time (i.e., a reduction of slope) within the DMN (β = –0.034, SE = 0.016, t = –2.05, p = 0.044), the SAL/VAN (β = –0.040, SE = 0.015, t = –2.63, p = 0.010), and the FPN (β = –0.041, SE = 0.019, t = –2.14, p = 0.036) (Figure 2 and Table 2). Higher HbA1c levels were rather associated with a reduction of the observed RSFC increase over time within the SAL/VAN (β = –0.029, SE = 0.014, t = –2.12, p = 0.037; Figure 2 and Table 2). Longitudinal changes in RSFC were not significantly associated with systolic blood pressure, BMI, or WMH volume in our sample (all p’s ≥ 0.05, Table 2).

Regarding Aβ pathology, there was no significant association between global Aβ deposition and change in RSFC over time in the specific networks and in global RSFC (all p’s ≥ 0.05, Table 2). Also, the Aβ positivity status did not significantly moderate the observed significant associations between diastolic blood pressure and RSFC changes over time (interaction: Aβ positivity × diastolic blood pressure × time on longitudinal RSFC in the DMN, SAL/VAN, and FPN: all t’s ≤ 1.606 and p’s ≥ 0.112) or HbA1c and RSFC changes over time (interaction: Aβ positivity × HbA1c × time on longitudinal RSFC in SAL/VAN: t = –1.129 and p = 0.262) (data not shown).

We further explored whether or not the VRF-related changes in RSFC trajectories (modeled by slopes of significant predicted LMEs) were associated with longitudinal changes in cognitive performance. No significant associations were found between the RSFC slopes of the DMN, SAL/VAN and FPN and cognitive changes in the four cognitive composites over time, adjusted for age, sex and education (p’s ≥ 0.2, Supplementary Table S2).

The main goal of this study was to replicate and expand earlier findings on the impact of VRF on RSFC trajectories within intrinsic brain networks (Köbe et al., 2021). Maintaining the data analysis strategy, the current results were derived from the independent IMAP+ cohort (e.g., Mutlu et al., 2017; Gaubert et al., 2021). (1) Our results show that VRF were independently associated with RSFC changes over time within higher-order brain networks. This finding was replicated for diastolic blood pressure. (2) The VRF-related associations were independent of pathological brain burden and no comparable effects of brain pathology were observed. This finding was replicated for Aβ load. (3) Our study expands earlier findings by three important novel aspects: We document that higher HbA1c levels were independently associated with RSFC changes over time. We clarify that the VRF-related associations were independent of WMH volumes, which had no measurable effect on RSFC changes. Finally, there were no detectable associations between the VRF-related changes in RSFC and longitudinal changes in cognitive performance. Altogether, our findings substantiate a presumed causal role of VRF in RSFC changes over time within cognitive networks, namely the DMN, SAL/VAN and FPN, known to be affected by aging and AD.

The present study largely confirms that VRF can be independently associated with RSFC changes over time, involving higher-order brain networks pertinent to cognitive health. First, we replicated earlier findings showing that there is an overall increase in RSFC over time in our sample of non-demented older adults – both within global RSFC and within the DMN, SAL/VAN and FPN networks. The effect has been interpreted to reflect some form of hyper-connectivity that mirrors earlier observations by Köbe et al. (2021) and other longitudinal findings, proposing that an initial increase in RSFC occurs in older adults until about ∼74 years of age, which is followed by a decline of RSFC as age progresses (Ng et al., 2016; Staffaroni et al., 2018). Similar to the previous study (Köbe et al., 2021), our current sample of older adults is relatively young, thus potentially explaining the observed increase of RSFC over time. Moreover, it appears that the cascading network failure may be hastened in the presence of VRF, as specifically indicated by the negative interactions between specific VRF and changes in RSCF over time. Namely, higher diastolic blood pressure levels and higher HbA1c levels were independently associated with RSFC changes over time in the DMN, the FPN, and/or the SAL/VAN, with the latter brain network showing an overlapping effect of the two VRF. Interestingly, out of the seven intrinsic brain networks, those displaying a functional vulnerability to VRF are also targeted by aging and AD (Buckner et al., 2008; Betzel et al., 2014; Elman et al., 2016; Grady et al., 2016; Badhwar et al., 2017). In contrast, we did not find such effects in the non-cognitive visual (VIS) and somatomotor (SM) networks.

As an important outcome, we thus reproduced the previous findings (Köbe et al., 2021) in the PREVENT-AD cohort (Breitner et al., 2016) using a highly similar methodological approach in the independent IMAP+ cohort. Both cohorts are comparable in the sense that they provide monocentric, longitudinal, and multimodal (i.e., clinical, behavioral and neuroimaging) data in older participants at increased risk for developing AD. Previously, it has been shown that higher levels of blood cholesterol and diastolic blood pressure were associated with a reduction of RSFC over time in the DMN and in global RSFC, respectively (Köbe et al., 2021). Together, the earlier and present longitudinal findings converge very well with the existing (mostly cross-sectional) studies, indicating a greater vulnerability of higher-order brain networks to a variety of VRF including higher levels of blood pressure, cholesterol, and HbA1c as well as the metabolic syndrome (Li et al., 2015; Zhang et al., 2016; Rashid et al., 2019; Feng et al., 2020; Veldsman et al., 2020). As in the previous study (Köbe et al., 2021), a similar association was not found for BMI. The BMI is proposed to influence VRF related to diabetes and hypertension (Gill et al., 2021), which may be more closely associated with RSFC alterations in older adults (see below).

Our current results add important novel aspects hinting at a causal role of VRF, namely diastolic blood pressure and HbA1c, in RSFC changes over time. More specifically, we document that higher diastolic blood pressure levels were more likely to be associated with a decrease of RSFC over time in the DMN, FPN, and SAL/VAN. In addition, higher HbA1c levels were rather associated with a reduction of the observed RSFC increase in the SAL/VAN, thus pinpointing the latter network as a common target point. These findings mirror cross-sectional findings, showing that hypertension and higher HbA1c levels are associated with altered RSFC in distributed brain networks at the population level (Feng et al., 2020) and in patients with diabetes (Zhang et al., 2020), respectively. The broad pattern of RSFC disturbances may reflect an overall disruption of microvascular structure as previously linked to diastolic hypertension (Caunca et al., 2020) and long-term glycemic exposure (Samanta, 2021).

Importantly, it is shown here that the impact of VRF on RSFC changes over time was largely independent of concomitant pathology, as measured using Aβ load and WMH. While this was formerly shown for Aβ deposition (Köbe et al., 2021), here we extend these findings to WMH, as a manifestation of cerebral small vessel disease (Wardlaw et al., 2013; Prins and Scheltens, 2015). It is likely that VRF work through various mechanisms that adversely impact brain and cognitive health (Villeneuve and Jagust, 2015; Vemuri et al., 2022) and are not fully captured by the present neuroimaging biomarkers of brain pathology. The observed effects of VRF may involve other microvascular lesions such as lacunes, microbleeds and microinfarcts (Wardlaw et al., 2013; Van Veluw et al., 2017) or other cerebrovascular mechanisms such as changes in cerebral blood flow or blood-brain barrier permeability (Zlokovic, 2011; de la Torre, 2012). The latter mechanisms are proposed to promote neural damage that is not dependent on Aβ pathology and vascular brain lesions (Villeneuve and Jagust, 2015), warranting further investigation.

The fact that poor vascular health may hasten RSFC changes within cognitive networks is important, because such functional brain alterations are implicated in AD pathogenesis – even at preclinical stages. While this was mainly shown for the DMN (Buckner et al., 2009), recent observations in normal older adults clearly demonstrate that lower RSFC in the DMN, SAL and FPN networks are predictive of cognitive decline over time in synergy with Aβ deposition (Buckley et al., 2017). By promoting early RSFC alterations within the very same brain networks, VRF may perhaps deplete cognitive reserve capacities (Franzmeier et al., 2017; Benson et al., 2018) and thereby accelerate cognitive disturbances in the setting of Aβ. In a recent framework functional connectivity is proposed as a candidate substrate for cognitive reserve, meaning that adaptive functional brain processes may help counteract brain pathology and damage in older age (Stern et al., 2020). In this line, RSFC alterations within higher-order brain networks, as associated with VRF, could perhaps contribute to limited cognitive reserve in late life and increase the vulnerability to aging and AD. Higher levels of diastolic blood pressure and HbA1c have also been linked to Aβ pathology, AD-typical neurodegeneration, and/or cerebrovascular disease (Byun et al., 2017; McNeil et al., 2017; Caunca et al., 2020; Garfield et al., 2021). Together, the findings converge on the view that good vascular health may help preserve brain health and cognitive resilience in older adults (de la Torre, 2010).

We did not detect any significant associations between pathological brain burden and changes in RSFC trajectories within the intrinsic brain networks. This observation replicates the previous study (Köbe et al., 2021) and is in part convergent with other reports (Tahmi et al., 2020). Cross-sectional studies have, however, documented RSFC alterations in the setting of Aβ burden or vascular brain injury in non-clinical and clinical cohorts (Hedden et al., 2009; Mormino et al., 2009; Elman et al., 2016; Schulz et al., 2021), with some of the Aβ-related RSFC alterations involving more focal hub regions (Hedden et al., 2009; Mormino et al., 2011). In addition, we did not detect a cognitive relevance of the VRF-related changes in RSFC alterations over time, albeit this was shown in a recent cross-sectional study on hypertension (Feng et al., 2020). It might be possible that pathological levels of VRF are more closely associated with RSFC alterations of cognitive relevance. Alternatively, VRF may foster vascular brain injury (van Dijk et al., 2008), which is reliably associated with lower cognitive functioning in older adults of the IMAP+ cohort (Garnier-Crussard et al., 2020) and other studies (Kloppenborg et al., 2014; Vemuri et al., 2015; Benson et al., 2018). Given our current null results, the cognitive relevance of the VRF-related changes in RSFC trajectories over time remains to be established.

The present study has strengths and limitations. Our current findings provide supporting evidence for a critical role of vascular health on RSFC within higher-order brain networks in older adults. One particular strength of our study is the longitudinal study design, allowing to investigate, replicate and extend earlier findings on the impact of VRF on changes in RSFC over time (Köbe et al., 2021) using an independent monocentric cohort of well-described older participants. Thanks to this unique and high-quality data pool, we additionally adjusted for measures of brain pathological burden and incorporated longitudinal cognitive assessments in our analyses. We feel that the present combination of well-characterized data outweighs some of the limitations of our study design including the limited sample size. Knowledge about the role of vascular health in brain aging is a key factor in the development of preventative strategies.

Several limitations need to be considered, when interpreting our results. (1) The majority of VRF were assessed at baseline with only one measurement available, potentially limiting the reliability of our markers. (2) We had a limited amount of VRF available. Therefore, potential effects of e.g., cholesterol levels on RSFC changes over time seen in the previous study (Köbe et al., 2021) could not be replicated. (3) It would be important to consider the effects of cardiovascular medication in the observed associations. There was no detailed information on medication intake in the present cohort. However, similar results as shown by the present study were found with respective adjustments for vascular medication in Köbe et al. (2021). (4) We merged participants along the cognitive continuum with normal cognition, SCD, and MCI in the attempt to enhance statistical power and variance. On the downside, this approach may have obscured group-specific associations, necessitating further investigation with larger sample sizes. (5) The current sample of older adults was mildly affected by vascular risk, especially for the BMI and HbA1c measures, comparable to the previous study (Köbe et al., 2021). It is nonetheless important to document the impact of VRF on RSFC in higher-order brain networks, even if subtle. (6) Having replicated previous findings assessing the association between VRF and within-network RSFC changes over time, there is a need to assess the link between VRF and changes in between-network RSFC in prospective studies. Overall, it could be assumed that functional brain alterations associated with VRF may become more relevant to cognitive health as pathological burden accumulates.