Type 2 Diabetes Mellitus (T2DM) is a multifactorial metabolic disorder, representing the sixth leading cause of disability, with an ever-increasing incidence (Stumvoll et al., 2005; McCarthy, 2010; Chatterjee et al., 2017; Yuan and Wang, 2017). T2DM causes macro and microvascular alterations (McCarthy, 2010) which cause long-term damage, dysfunction, and deterioration in several tissues and organs, such as the brain, causing cerebrovascular disease and changes in cerebral hemodynamics (McCarthy, 2010; American Diabetes Association, 2012; American Diabetes Association, 2017; Chatterjee et al., 2017; Yuan and Wang, 2017). Due to its association with an increased risk for neural functional loss, long-term cognitive impairment, and dementia (Biessels and Reijmer, 2014; Ryan et al., 2014; Duarte et al., 2015; Moran et al., 2015; Callisaya et al., 2018; Moran et al., 2019; Biessels et al., 2020), it becomes crucial to understand the neurobiological correlates of early brain dysfunction in T2DM, in particular, at early stages of the disease when there are no evident structural lesions (Espeland et al., 2013; Moran et al., 2013; Biessels and Reijmer, 2014).

Several authors have used functional magnetic resonance imaging (fMRI) blood oxygenation level-dependent (BOLD) signal (Ogawa et al., 1990a; Ogawa et al., 1990b) to indirectly measure neuronal activity and investigate neurophysiological impairment in T2DM. However, results in the clinical setting have been inconsistent, namely regarding the presence or absence of functional alterations and its association with cognition (Zhou et al., 2010; Chhatwal and Sperling, 2012; Musen et al., 2012; Xia et al., 2013; Cui et al., 2014; Wang et al., 2014; Liu et al., 2018, 2019; Xiong et al., 2020). The BOLD signal is an intrinsic hemodynamic signal which depends on the neurovascular coupling (NVC) between neuronal activity and cerebral blood flow (Attwell et al., 2010; Phillips et al., 2015). On the other hand, neurovascular coupling reflects the close temporal and regional linkage between neural activity and cerebral blood flow. This close matching of local blood flow to neuronal activation is related to the high energetic demands of brain cells. Thus, even slight changes to the energy supply require precise autoregulation mechanisms. Therefore, an impaired NVC can result in brain dysfunction and neuronal atrophy. Abnormal NVC has been implicated in Alzheimer’s disease, multiple sclerosis, traumatic brain injury, spinal cord injury, and stroke (Fletcher et al., 2022).

Notably, in diabetes, pathophysiological vascular changes can influence the blood flow regulation in cerebral microvasculature, possibly impairing the NVC. Therefore, the interaction of neurons with the surrounding vasculature in terms of autoregulatory mechanisms is critical for proper function of the central nervous system. Their deregulation may contribute to early cognitive impairments, namely in the memory domain. Indeed, several longitudinal studies have shown that patients with T2DM have an increased risk of developing dementia (Hardigan et al., 2016). However, alterations in the BOLD signal may reflect abnormal neuronal activity or inefficient neurovascular coupling, generating an ambiguity in interpretation. Thus, underlying mechanistic changes might be indistinguishable. They could be neurodegenerative, vascular, or a combination of both.

The vascular response triggered by neuronal activation and measured with fMRI BOLD is described by a hemodynamic response function (HRF) (Carusone et al., 2002), which has been shown to be altered, for instance, in patients with stroke (Bonakdarpour et al., 2007; Altamura et al., 2009), resulting in misinterpretation or under-estimation of the fMRI signal. In a previous fMRI study (Duarte et al., 2015), we have used a performance-matched visual stimulation task, thereby recruiting similar neural and cognitive resources in T2DM and control participants, to examine the BOLD response specifically in three task-related regions-of-interest (ROIs) involved in visual processing, perceptual decision mechanisms, and executive functioning (Duarte et al., 2017). This earlier study suggested an overal change in HRF at early stages of T2DM, which might reflect an impaired neurovascular coupling (Duarte et al., 2015), raising the question of which components of the HRF might be changed in diabetes.

This previous study highlighted the need to obtain a complete understanding of the components of hemodynamic model parameters and the underlying physiology of the BOLD signal to improve the utility of functional brain mapping in the context of diabetes.

In this new study, we investigated whether the HRF in its distinct phases would be compromised in individuals with T2DM depending on the brain region or would instead represent a general cortical phenomenon, in a large cohort of patients with T2DM and healthy controls, in response to the same visual motion discrimination task presented with both block and event-related paradigms. The analysis of the data measured during block stimulation allowed to localize activated brain regions, in which we extracted the HRF during the event-related stimulation task by a deconvolution GLM analysis. Overall, and as expected, patients with T2DM revealed significantly different HRF profiles. Furthermore, as a novel approach, several HRF parameters corresponding to distinct phases were extracted from response profiles and compared between groups in each region to understand what sort of physiological changes arise with T2DM and what consequences they can cause.

Since alterations in the HRF may reflect changes in the NVC and, consequently, pathophysiological brain changes even at early stages, the use of the HRF as a functional imaging biomarker has been investigated in the context of cerebral pathology (Miezin et al., 2000; Taylor et al., 2018). While previous studies investigating HRF parameters as biomarkers have focused on peak amplitude (Peck et al., 2004), there has been interest in the evaluation of other parameters, such as the activation duration and the peak latency (Miezin et al., 2000; Bonakdarpour et al., 2007; Lindquist and Wager, 2007). This is important to sort out, prior to the investigation of HRF as a potential biomarker of clinical/cognitive complications in T2DM. As T2DM induces a decrease in cerebral blood flow (Kelly-Cobbs et al., 2012), reducing neural efficiency (Abdelkarim et al., 2019), we hypothesized that T2DM leads to neurovascular decoupling (Duarte et al., 2015) and yields consequences in the HRF morphology components (Hillman, 2014). We hypothesize that T2DM is characterized by a lower overall amplitude of the peak, initial dip and undershoot, and a higher peak latency and that the source of these differences may be possibly driven by the disruption of the NVC.

In this study, we found that the HRF in the brain of patients with T2DM is altered relative to healthy controls, as shown by a novel quantitative analysis of the HRF morphology. Overall, in all ROIs, T2DM patients presented a more sluggish (time-delayed) HRF, diverging from the canonical HRF, with a higher peak latency and an absent or less intense but lengthier undershoot (curve below the baseline at the end of the response). Furthermore, the HRF in T2DM patients exhibited a smaller peak amplitude and relative slope to peak. These differences were found using age as a covariate, thus suggesting a specific effect of T2DM independent of healthy aging.

Although the area under the curve, including the positive and negative sections, appears to be relatively similar between the two populations, there is still a trend to be smaller in T2DM patients. Besides, the variability of HRF parameters was, on average, higher in T2DM patients. Lastly, the HRF of T2DM patients often included an initial dip (the classic initial reduction due to the mismatch between oxygen consumption and blood supply) with clear enhancement, unlike what is observed in the control group. This even led to some regional dominantly negative responses in diabetic patients in contrast with controls.

These results are consistent with our hypothesis that the underlying reason for these changes in the HRFs of T2DM patients might be an early compromised neurovascular coupling. Possible explanations for the observed differences include vascular damage (such as endothelial dysfunction), impairment of the vasodilation regulatory mechanisms, or excessive O₂ consumption. It is not likely that different levels of neuronal activity alterations could have a significant effect on these curves because, as mentioned, the stimuli were individually fitted, so that the task difficulty was similar for all participants. Furthermore, the task performance was identical between groups, revealing a comparable perceptual discrimination ability.

In T2DM patients, there is an imbalance between relative O₂ consumption and blood supply. In fact, given similar neuronal activation conditions in both groups, with NVC changes in T2DM, the oxygenated blood supply becomes smaller, and, thus the O₂ decrease induced by neuronal activity will not be so swiftly compensated. This contributes to the overall HRF delay observed. Consequently, the deoxyhemoglobin concentration will further increase initially, hence decreasing the BOLD signal, which explains the large initial dip seen in its average and median HRFs, which in turn was most often absent or less perceptible in the controls.

When the blood supply finally starts to balance the O₂ consumption, the BOLD signal increases, but more gradually and slower than in healthy conditions, until the peak is reached, which elucidates, in turn, the higher peak latency and smaller peak amplitude and relative slope to peak overall seen in the average and median HRFs of T2DM patients. On the opposite, the controls, with a more efficient NVC, show a higher and faster BOLD signal increase, producing a higher peak amplitude, relative slope to peak, AUC, PCSA, and NCSA, but a lower peak latency, which matches with the overall results. In T2DM patients, after reaching the peak, the return to the baseline, due to the aforementioned reasons, will be delayed and will be slower than in controls, which is mainly in line with the achieved results. On the other hand, the undershoot may not be discerned, which may be due to a masking effect prompted by the lower efficiency of the first phase of the hemodynamic response.

The components of the NVC may present intraindividual, inter-region, and inter-condition variability (Elbau et al., 2018). Its characterization to understand how the hemodynamic response is altered in T2DM, is therefore quite relevant. Our results show that the hemodynamic response, did indeed display a large inter-subject variability in T2DM, which is consistent with the notion of damage induced differences in NVC. Conversely, control participants might have a more efficient NVC, and the hemodynamic response will be more similar between them and will show less variability, also taking into account the relatively restricted age of the sample (40–76 years). Furthermore, the results also demonstrate a robust replication of the HRF effect between ROIs and between the type of stimuli. This consistency suggests an overall effect of T2DM on the HRF which is similar between regions and conditions. To further test the roles of variables such as age, sex, or HbA1c levels as possible causes for this variability, we investigated the correlation of the HRF parameters with these factors in each group. We did not find any significant correlation of HRF with age or HbA1c, nor a differential effect of sex, which suggests that these are not the cause of the variability in this data set, but rather T2DM.

In the ROIs with negative signal change, notably, their particular morphology justifies the reason why we designated these ROIs as negative signal change ROIs (may be corresponding and exaggerated negative dip) instead of deactivation regions. The HRFs of these ROIs were distinct from the classical deactivation pattern, which expectedly would be like those existing in the positive change ROIs, yet inverted. That is, there would be a large signal decrease against the baseline, a peak (which would correspond to the minimum of the function) around 4–6 s, and a return to the baseline at 16–20 s. Similarly to the canonical HRF, there could also be an initial dip (positive) before the peak (Havlicek et al., 2017). Overall, it could be captured as a deactivation by the GLM. As the term deactivation is commonly associated with a physiological meaning - a region which decreases its activity during a condition - and as the meaning of the HRFs of these ROIs is not understood, it was considered wiser to designate them as negative signal change ROIs. To maintain the coherence, the remaining ROIs, on which the discussion was focused, were designated as positive signal change ROIs, in line with their typical activation responses.

A very interesting observation is that the same brain region (Insula BA13) showed very distinct HRFs in the two hemispheres. This intriguing finding will need to be investigated in the future, in line with the hemispheric functional asymmetries that are known to exist in this region (Eckert et al., 2009; Uddin, 2015; Intaitė et al., 2016; Zhang et al., 2019; Sayal et al., 2020).

Finally, as a limitation of the study, we should mention that despite the strong statistical power of these results, besides T2DM, there are several other sources of NVC changes, which, in turn, can introduce variability. For instance, the BOLD signal, which measures indirectly the NVC, is also sensitive to other hemodynamic processes, even from non-pathological sources (e.g., atypical brain physiology) (Rossini et al., 2004; Elbau et al., 2018). Thus, when the NVC is investigated, its interpretation and comparison between groups become more complicated by additional factors (Whittaker et al., 2016). Since the NVC is a rather complex phenomenon whose components change between regions and conditions, conditioning how the coupling takes place (Drew, 2019), further research, including in animal models, should be done on this topic.

This study provides robust evidence that T2DM patients have an early compromised neurovascular coupling, as shown by the changes in the hemodynamic response function overall in the brain. The novel demonstration of a distinct HRF profile is expressed by a more sluggish response (higher peak latency), with a smaller peak amplitude, a pronounced initial dip, a delayed or absent undershoot, a decreased relative slope to peak, and lower positive and negative areas under the curve. These changes in the HRF morphology may provide novel markers of neurovascular uncoupling caused by T2DM.

Therefore, this study reinforces the role of fMRI as a tool to evaluate brain function and neurovascular coupling in T2DM. Duarte et al. (2015) had previously suggested that BOLD signal differences in diabetic patients could potentially identify NVC disruption, which was shown here. On the other hand, the novel analysis of the HRF parameters may prompt further studies that may involve classification and/or its use as a biomarker which might be related to the pathophysiological evolution and, consequently, play a potentially important role in the assessment of disease progression and effects of therapy.