The neurovascular unit coming of age: a journey through neurovascular coupling in health and disease

Abstract

The concept of neurovascular unit (NVU), formalized at the 2001 Stroke Progress Review Group meeting of the National Institute of Neurological Disorders and Stroke, emphasizes the intimate relationship between the brain and its vessels. Since then, the NVU has attracted the interest of the neuroscience community resulting in considerable advances in the field. Here the current state-of-knowledge of the NVU will be assessed, focusing on one of its most vital roles: the coupling between neural activity and blood flow. The evidence supports a conceptual shift in the mechanisms of neurovascular coupling, from a unidimensional process involving neuronal-astrocytic signaling to local blood vessels, to a multidimensional one in which mediators released from multiple cells engage distinct signaling pathways and effector systems across the entire cerebrovascular network in a highly orchestrated manner. The recently appreciated NVU dysfunction in neurodegenerative diseases, although still poorly understood, supports emerging concepts that maintaining neurovascular health promotes brain health.

Introduction

The brain, an organ of unparalleled sophistication, seems to have a fundamental design glitch: it consumes a large amount of energy, but lacks a reservoir to store fuel for use when needed. Therefore, the brain receives energy substrates, primarily oxygen and glucose, “on the fly” through its blood supply. Considering the dynamic and regionally diverse energy requirements imposed by brain activity, blood flow needs to reach the brain at the right time and place, and in the right amount. Failure to do so has disastrous consequences. Complete interruption of the cerebral blood supply for more than a few minutes, for example when a cerebral artery is occluded in stroke or after pump failure in cardiac arrest, causes irretrievable brain damage and death. On the other hand, if flow is not completely stopped, but is reduced or not well matched to the energy demands of the tissue, more subtle brain alterations ensue, leading to chronic brain injury in vulnerable areas often associated with cognitive impairment (Iadecola, 2013).

There has been a long-standing interest in how the brain regulates its own blood supply, driven not only by the desire to gain a better understanding of the harmful effects of cerebrovascular insufficiency, but also by the early realization that regional changes in cerebral blood flow (CBF) may provide a window on brain function. This large body of work spanning over almost two centuries revealed a remarkable complexity in the interaction between the brain and its vessels, unmatched by the vasculature in other organs.

In this context, the concept of neurovascular unit (NVU) emerged from the first Stroke Progress Review Group meeting of the National Institute of Neurological Disorders and Stroke of the National Institutes of Health (July 2001) to emphasize the unique relationship between brain cells and the cerebral vasculature (https://www.ninds.nih.gov/About-NINDS/Strategic-Plans-Evaluations/Strategic-Plans/Stroke-Progress-Review-Group). Although the vital importance of cerebral blood vessels in brain health had long been appreciated, hard-core neuroscientists considered brain cells and cerebral blood vessels as distinct entities. Such dichotomy led to the tacit assumption that, unless the delivery of blood flow to the brain was critically compromised, neurons had little to do with the vasculature, and vice-versa. Similarly, a rigid distinction was placed between “neurodegenerative diseases”, e.g., Alzheimer’s disease (AD), and cerebrovascular diseases, e.g., stroke, such that these conditions were considered mutually exclusive and mechanistically unrelated. The NVU concept challenged these assumptions and emphasized the symbiotic relationship between brain cells and cerebral blood vessels, calling attention to their developmental, structural and functional interdependence in health and disease.

The neuroscience community embraced the concept enthusiastically, and the NVU has drawn increasing interest, as attested by the dramatic rise in the number of yearly citations over the past decade (Figure 1). New technologies have enabled investigators to delve deeper into the functions of the NVU in health and disease. Consequently, the NVU has taken center stage in all facets of normal brain function, and in the pathobiology of a wide variety of brain diseases.

Figure 1. Citations of the neurovascular unit in the literature from 1997 to 2016.

The search term “neurovascular unit” was used in the Web of Science database (Thompson Reuters). Since the SPRG meeting in 2001, a dramatic rise in the number of citations was observed. Prior to 2001, the term was often used for critical care units combining neurology and neurosurgery.

Here we will assess the current state-of-knowledge of the NVU, focusing on one of its prototypical functions: the coupling between neural activity and CBF (neurovascular coupling). To this end, we will review and integrate observations related to the contribution of different NVU cells to neurovascular coupling, in an attempt to provide a comprehensive picture of how activation of restricted groups of neurons is able to generate hemodynamic responses engaging the entire cerebrovascular tree, from capillaries deep in the substance of the brain to pial arteries on the brain surface. The contribution of NVU dysfunction to brain pathologies will also be examined, focusing on neurodegenerative diseases, highly prevalent conditions previously considered unrelated to vascular factors. Finally, a critical assessment of the evidence will be provided in an effort to flesh out unresolved issues and outstanding questions.

Neurovascular interactions through the ages

Throughout the 18^th and 19^th century the prevailing view was that the brain was not involved in regulating its own blood supply, thought to be controlled exclusively by the systemic circulation (Friedland and Iadecola, 1991). Hints of change can be appreciated at the end of the 19^th century in the work of Roy and Sherrington in animals and Angelo Mosso in humans (Figure 2). Roy and Sherrington, based on the effect of injecting brain extracts in the carotid arteries of animals equipped with a device to monitor intracranial pressure, proposed that metabolites produced by neuronal activity diffuse to local blood vessels to increase blood flow to the activated brain (Roy and Sherrington, 1890). Even earlier, Mosso, studying individuals with skull defects, had discovered that somatic, mental or emotional stimuli induce changes in brain volume attributable to hemodynamic factors (Mosso, 1880). This body of work, albeit based on indirect indices of CBF, not measurable at the time, raised the possibility that brain activity was able to elicit changes in cerebral perfusion. These prescient observations were dismissed and quickly forgotten, and the idea that the brain could regulate its own blood supply did not reemerge until decades later (Friedland and Iadecola, 1991). In the late 1930s, Schmidt and Hendrix (Schmidt and Hendrix, 1938) demonstrated in the cat that illuminating the eye increases visual cortex temperature, a proxy for cerebral perfusion (Figure 2). Remarkably, the work of Roy and Sherrington was not mentioned in this and related publications of this period, resurfacing in the literature only in the 1960s (Friedland and Iadecola, 1991). A separate line of investigations, in the early 20^th century, revealed that vascular density differs across the brain, and suggested that the density of vessels correlated with regional levels of energy consumption and functional activity (Craigie, 1945). Consistent with this hypothesis, it was found that sustained increases in neural activity lead to increases in local vascular density (Craigie, 1945). These early observations led to the idea that the vascular changes induced by neural activity were needed to fulfill the increased metabolic needs of active brain regions, providing initial evidence of the critical relationship linking brain function to the cerebral blood supply.

Figure 2. Historical overview of the concept of neurovascular coupling.

Pioneers in establishing the concept of neurovascular coupling are shown on top and examples of their work presented below the timeline. Studies in the late 1800s, by Angelo Mosso and Charles Roy and Charles Sherrington, hinted at the possibility that brain activity increases cerebral blood flow. The lower panels illustrate the apparatus used by Mosso to record changes in brain volume in patients with skull defects (Mosso, 1880), and Roy and Sherrington’s recordings of brain expansion in response to intracarotid infusion of a brain extract (Roy and Sherrington, 1890). In the 1930s Carl Schmidt recorded increases in temperature in the cat visual cortex by shining light into the eye. The original recording is shown in the lower panel (Schmidt and Hendrix, 1938). In the 1950–60s Seymour Kety and Lou Sokoloff developed autoradiographic methods to image regional CBF during neural activation. The lower panel shown the increase in CBF produced in the calcarine cortex and superior colliculus by visual stimulation (Freygang and Sokoloff, 1958). In the 1960-70s David Ingvar and Niels Lassen developed methods to measure regional CBF in the human brain using radioactive tracers and external detection. The lower panel illustrates the increase in CBF produced by hand movement in the contralateral sensory-motor cortex and supplementary motor area (Lassen et al., 1978), marking the birth of functional brain imaging.

Kety and Schmidt pioneered a method to measure CBF quantitatively using nitrous oxide as a tracer, and demonstrated that CBF increases with global brain hyperactivity, for example, in anxiety or hyperthyroidism, and is reduced when brain activity is suppressed, e.g., in coma (Kety, 1950). However, this method provided an average of CBF for the whole brain and could not assess regional cerebral perfusion and its relationship to local neural activity. The development of autoradiographic techniques using diffusible tracers to measure CBF quantitatively and in multiple brain regions, provided evidence that neural activity evokes CBF increases highly restricted to the activated network (Figure 2). Building on this work, Ingvar and Lassen developed methods to measure regional CBF in the human brain using intracarotid injection of radioactive tracers and external detection by a γ-camera (Figure 2). The subsequent development of positron emission tomography and MRI-based methods allowed investigators to monitor CBF in humans with great spatial resolution (Raichle and Mintun, 2006). In particular, the discovery of the blood oxygenation-level dependent (BOLD) effect, reflecting excess CBF delivery relative to local oxygen consumption, enabled the non-invasive detection of activity-dependent hemodynamic signals across the behaving human brain (Raichle and Mintun, 2006). While advancing our understanding of the neurobiology of human behavior, MRI-based functional brain imaging has also firmly established the concept that cerebrovascular function is intimately related to brain activity.

Structural diversity of the NVU along the cerebrovascular tree

The association between neurons and vessels varies significantly across the cerebrovascular network. Branching off the circle of Willis at the base of the brain, cerebral vessels run on the brain surface within the subarachnoid space (pial arteries and arterioles) forming a highly collateralized network (Blinder et al., 2013; Cipolla, 2010) (Figures 3, 4). Pial arterioles have multiple layers of smooth muscle cells (SMC) separated from the endothelium by a prominent elastic lamina (Roggendorf and Cervos-Navarro, 1977) (Figure 4). Although not in direct contact with brain cells, pial vessels are richly innervated by nerve fibers originating from peripheral autonomic and sensory ganglia, including the superior cervical, the sphenopalatine and the trigeminal ganglia (Hamel, 2006) (Figure 4). These nerve fibers, which express a wide variety of neurotransmitters and neuropeptides, run at the adventitia-media border forming an intricate mesh enveloping the vessels (Iadecola et al., 1993). Pial arteries dive into the substance of the brain surrounded by an extension of the subarachnoid space, the perivascular space, a virtual space delimited by the vascular basement membrane and the glia limitans (Virchow-Robin space)(Jones, 1970; E. T. Zhang et al., 1990) (Figure 4). Penetrating arterioles are endowed with a thinner smooth muscle cell layer, which eventually becomes a single layer (Dahl, 1973; Roggendorf and Cervos-Navarro, 1977). At this level, perivascular nerves are more sparse and the elastic lamina becomes less prominent (Roggendorf and Cervos-Navarro, 1977). The perivascular compartment contains several cell types, including perivascular macrophages (PVM), Mato cells, pial cells, and mast cells, among others, as well as nerve and collagen fibers (E. T. Zhang et al., 1990) (Figure 4). As arterioles penetrate deeper into the brain (intraparenchymal arterioles), the glial membrane and the vascular basement membrane fuse together obliterating the perivascular space (Jones, 1970; E. T. Zhang et al., 1990) (Figure 5). At this level, arterioles have a single or discontinuous layer of SMC, lack perivascular nerves and are encased in astrocytic end-feet (Roggendorf and Cervos-Navarro, 1977) (Figure 5). Occasionally axonal terminals or dendrites are seen in close apposition to the vascular basement membrane, often with an intervening glial leaflet (Iadecola et al., 1993; H. Wang et al., 2005). These neural processes originate from interneurons or from subcortical and brainstem nuclei projecting to the cortex, such as the basal forebrain cholinergic nuclei, locus coeruleus, and raphe magnus (Cohen et al., 1996; Hamel, 2006; Iadecola, 1993). Endothelial cells extend protrusions through the basal lamina into SMC (myoendothelial projections) enriched with gap junctions (Dahl, 1973; Longden et al., 2016). As arterioles transition into cerebral capillaries (Figure 5), SMC are replaced by pericytes, mural cells embedded into the endothelial basement membrane, covering approximately 30% of the vascular surface, and occasionally extending “peg and socket” protrusions into endothelial cells (Armulik et al., 2011; Dahl, 1973; Damisah et al., 2017). Like in arterioles, the circumference of the capillaries is covered by astrocytic end-feet, although neural processes can be at times observed next to the capillary basal lamina.

Figure 3. Anatomy of the cerebrovascular tree and segmental vascular resistance.

The internal carotid artery (ICA) enters the skull and merges with branches of the vertebral arteries to form the circle of Willis at the base of the brain. The middle cerebral artery (MCA) takes off from the circle of Willis and supplies a large territory of the cerebral cortex. The MCA gives rise to pial arteries and arterioles that run on the surface of the brain forming a heavily interconnected network from which arterioles penetrating into the substance of the brain originate (penetrating arterioles). Penetrating arterioles give rise to the capillary network, which feeds into the venous system returning the blood to the heart. The component (%) of the total vascular resistance that each cerebrovascular segment offers to blood flow, which reflects their potential for flow control, is indicated. Therefore, vessels outside the brain are responsible for 60% of the resistance, and vessels within the substance of the brain for 40% (based on data from (Stromberg and Fox, 1972) and (De Silva and Faraci, 2016)).

Figure 4. Neurovascular associations along the cerebrovascular tree: Pial arteries and penetrating arterioles.

A pial arteriole on the cortical surface, giving rise to a penetrating arteriole is shown on the right. Pial arterioles have thick coat of SMC, are surrounded by the subarachnoid space, and are densely innervated by nerve fibers originating form cranial autonomic and sensory ganglia, such as the sympathetic, parasympathetic and trigeminal ganglia. Penetrating arterioles enter the substance of the brain and are surrounded by a perivascular space containing several cell types including perivascular macrophages (PVM). In Figures 3 and 4, the vasculature (green) was visualized with the lipophilic dye DiO injected into the circulation. PVM and meningeal macrophages (blue), located in the perivascular and subarachnoid space, respectively, are visualized by CD206 immunocytochemistry.

Figure 5. Neurovascular associations along the cerebrovascular tree: Intraparenchymal arterioles and capillaries.

As the arterioles penetrate deeper into the brain the perivascular space disappears and the vessels become encased in astrocytic end feet (intraparenchymal arterioles). Endowed with a single layer of SMC, intraparenchymal arterioles lack perivascular nerves. Occasionally, neural processes originating from local neurons or subcortical pathways projecting to the cerebral cortex terminate near the vessel. Capillaries are devoid of SMC but are endowed with pericytes, which are fully enclosed by the basement membrane of the endothelium. Occasional neurovascular contacts similar to those seen in intraparenchymal arterioles are also observed.

Neurovascular coupling

One of the better-appreciated and most studied roles of the NVU pertains to the link between neural activity and CBF. In the “resting” brain, CBF varies in proportion to the energy consumption of each brain region. Thus, flow is higher in regions with higher energy utilization, e.g., inferior colliculus, and lower in regions with lower energy use, like the white matter (Sokoloff, 2013). Furthermore, increases in neural activity lead to increases in CBF highly restricted to the activated areas (functional hyperemia) (Chaigneau et al., 2003; Cox et al., 1993; Freygang and Sokoloff, 1958; Iadecola, 1993; LeDoux et al., 1983). The spatial and temporal correspondence between neural activity and the associated hemodynamic response is sufficiently precise, at least at the regional level, that the flow response can be used to map brain function (functional brain imaging) (Raichle and Mintun, 2006). However, at the microvascular level, the CBF increase in some regions exceeds the area of activation. For example, in the auditory, visual and cerebellar cortex the vascular response does not faithfully match the activated area (Harrison et al., 2002; Iadecola et al., 1997; O’Herron et al., 2016), while in the olfactory bulb there is close overlap between the two (Chaigneau et al., 2003). Non-overlapping vascular and neuronal topology in neocortex (Blinder et al., 2013) and retrograde vasodilatation (discussed later in this review) are likely responsible for such lack of fidelity (Chen et al., 2014; Iadecola et al., 1997; Longden et al., 2017). Since hemodynamic signals are the most powerful tool at our disposal for functional brain mapping, their spatial alignment with neural activity is becoming increasingly relevant as the resolution of fMRI increases.

The close relationship between neuronal and vascular function is also illustrated by the profound changes in neurovascular coupling that take place in the developing brain. In rodents before post-natal day 11, brain activation is not associated with sustained CBF increases, leading to an absent or negative BOLD signal (Colonnese et al., 2008; Kozberg et al., 2013). The lack of a flow response may be important for vascular development since the resulting hypoxia is a critical stimulus for cerebral angiogenesis (Lacoste and Gu, 2015). However, in the second and third week of life neural activity leads to increasingly larger hemodynamic responses resulting in progressively larger BOLD signals (Colonnese et al., 2008). The transition between the second and third post-natal week corresponds to dramatic neurovascular and systemic changes, including increases in vascular density, synaptogenesis, myelination and connectivity, energy metabolism, resting CBF and sensitivity of the cerebral microcirculation to vasoactive stimuli (Colonnese et al., 2008; Engl et al., 2017; Goyal et al., 2014; Nehlig et al., 1989), as well as systemic blood pressure (Kozberg et al., 2013). Although the factors responsible for the developmental shift in neurovascular coupling remain to be elucidated, the findings attest to the profound influence of key structural, metabolic and functional changes in the developing brain on shaping cerebral perfusion.

Why does CBF increase during brain activity?

The tight coupling between neural activity and CBF is thought to reflect the lack of energy reserves in the brain requiring a well-timed delivery of oxygen and glucose restricted to the activated areas. The flow increase may also be needed to clear the brain of potentially toxic by-products of brain activity, for example, lactate, CO₂, the amyloid-β peptide (Aβ), and tau, as well as for brain temperature regulation (Tarasoff-Conway et al., 2015; Zhu et al., 2006). These considerations have led to a “feed-back” model in which the metabolic and clearance needs of the tissue drive the delivery of blood flow. Indeed, metabolic by-products of brain activity include potent vasodilators, such as adenosine, CO₂, H⁺, and lactate, which could potentially initiate the flow response (Freeman and Li, 2016; Ko et al., 1990). However, other evidence supports a “feed-forward” model in which the increase in CBF is not driven by the tissue metabolic state. As revealed by the transformative work of Fox and Raichle, the increase in CBF is greater than the need of the tissue oxygen (Raichle and Mintun, 2006), resulting in an excess delivery of O₂. In addition, increases in CBF occur also in conditions of excess oxygen and glucose, suggesting that activation-induced depletion of these energy substrates does not drive the flow increase (reviewed in (Attwell and Iadecola, 2002)). Based on this evidence, it was suggested that the CBF delivery is regulated by a feed-forward mechanism driven by neurovascular signaling pathways resulting in the release of vasoactive by-products of synaptic activity, such as K⁺, nitric oxide (NO) and prostanoids (Attwell et al., 2010; Attwell and Iadecola, 2002; Drake and Iadecola, 2007).

In the end, these two models may not be mutually exclusive (Figure 6). Microvascular studies showing that at the onset of neural activity there is a reduction in O₂ and/or glucose preceding the CBF increase have rekindled the view that metabolic factors may also play a role in neurovascular coupling (Freeman and Li, 2016), but there might be regional diversity in their involvement. Basal oxygen levels vary widely in different brain regions (Lyons et al., 2016), and, depending on local vascular topology and intensity of the activating stimulus, regional hypoxia may develop, promoting vasodilatation of local vessels. Another hint that intravascular hypoxia may contribute to the flow response is provided by experiments demonstrating that at the beginning of activation a dip in intravascular O₂ leads to increased deformability of red blood cells which improves microvascular rheology and increases capillary flow (Wei et al., 2016). These lines of evidence, collectively, suggest that a feed-forward mechanism may trigger an exaggerated flow response driven by neurovascular signaling, but feedback mechanisms may also be in place to adjust the CBF delivery more closely to the metabolic needs of the tissue. In support of this hypothesis, the vascular response during sustained neural activation tends to peak at the onset, subsequently readjusting toward a lower level (Drew et al., 2011; Freeman and Li, 2016; Ngai et al., 1988). Therefore, both metabolism-dependent (feed-back) and independent (feed-forward) mechanisms may be involved in functional hyperemia, depending on the timing, intensity and duration of the activation, as well as the brain region and the brain’s developmental stage. Although the mechanistic details remain incomplete, the evidence suggests that the main function of neurovascular coupling is to maintain the homeostasis of the cerebral microenvironment by delivering the energy substrates needed to initiate and sustain neural activity, while clearing potentially toxic by-products of brain metabolism including heat.

Figure 6. Potential feed-forward and feed-back mechanisms driving the local vascular response evoked by synaptic activity.

Glutamate released by synaptic activity activates postsynaptic glutamate receptors (GluR) leading to activation of Ca²⁺-dependent signaling pathways resulting in the release of vasoactive factors that may drive the initial “feed-forward” component (metabolism independent) of the local vascular response in arterioles and capillaries. At the same time a reduction in tissue O₂ caused by the increase energy consumption induced by activation leads to the accumulation of vasoactive metabolic byproducts that may drive a secondary feed-back component (metabolism dependent) to better match the flow response to the metabolic needs of the tissue.

It has also been suggested that, in a reversal of roles, the hemodynamic response could influence the neural response via mechanical, thermal or chemical effects on astrocytes (hemo-neural hypothesis or vasculo-neuronal coupling) (Kim et al., 2016; Moore and Cao, 2008). This concept received support from a study in brain slices in which increases or decreases in transmural pressure/flow in penetrating arterioles was found to suppress or enhance, respectively, pyramidal cell activity through mechanosensitive transient receptor potential vanilloid receptor type 4 (TRPV4) channels in astrocytes (Kim et al., 2016). These data suggest an autoregulatory mechanism whereby intravascular pressure, and possibly tissue pressure, would directly modulate resting neural activity via astrocytic mechanosensors. The mechanisms and implications of these intriguing observations need further exploration.

Cellular bases of neurovascular coupling

During the previous decades, most studies focused on the mediators responsible for neurovascular coupling. This work suggested that multiple agents released by neural activity are involved. However, it was also recognized that local release of vasoactive mediators could not account for hemodynamic changes in upstream vascular segments remote from the activated site. The availability of tools to investigate neurovascular interactions at the cellular level has provided greater insight into the cellular processes initiating, transmitting, propagating and implementing the vascular response (Table 1). These will be examined next.

Table 1.

Key cellular steps underlying neurovascular coupling in neocortex

Step	Cell type	Mediator/Mechanism	Cellular targets	Vascular segment
Initiation	Principal neuron Interneuron	NO Neurotransmitters Neuropeptides Prostanoids Adenosine ATP K+ Hypoxia	Astrocytes Endothelium Pericytes SMC Red blood cells	Capillaries Arterioles
Modulation/ Spatial shaping	Subcortical projection neurons (LC, BF, RM) Perivascular innervation (?)	Acetylcholine Serotonin Noradrenalin Neuropeptides	Neurons Astrocytes Endothelium Pericytes SMC	Capillaries Arterioles Pial arteries
Neurovascular transmission	Astrocytes	K+ Prostanoids (?) Adenosine (?)	Endothelium Pericytes SMC	Capillaries Arterioles
Retrograde propagation	Endothelium Pericytes (?) Astrocytes (?)	Gap junctions Myoendothelial junctions Myogenic response/FMV Hyperpolarizing factor (?)	Endothelium Pericytes (?) Astrocytes (?) SMC	Capillaries Arterioles
Implementation	SMC Contractile mural cells (pericytes?)	Hyperpolarization ↓Ca2+ or Ca2+ sensitivity Contractile proteins	–	Capillaries (?) Arterioles Pial arteries

Abbreviations: BF: basal forebrain; FMV: flow-mediated vasodilatation; LC: locus coeruleus; MLC: myosin light chain; RM: raphe magnus; SMC: smooth muscle cells

Neurons: initiation of the local vascular response

Neurons regulate CBF by generating signals, which, either directly or through interposed cells, act on local blood vessels to initiate the vascular response. Glutamatergic synaptic activity activates post-synaptic NMDA and AMPA receptors, leading to increases in intracellular Ca²⁺ and activation of Ca²⁺ dependent enzymes such as neuronal NO synthase (nNOS) and cyclooxygenase-2 (COX-2), which produce potent vasodilators, respectively, NO and prostanoid (Attwell et al., 2010; Lecrux and Hamel, 2016) (Figure 6). At the same time, glutamate acts on metabotropic glutamate receptors in astrocytes, triggering Ca²⁺ increases in these cells and production of vasoactive agents (see below). Adenosine, a potent vasodilator produced from ADP by ecto-nucleotidases, also contributes to the increase in CBF (Iadecola, 1993; Ko et al., 1990). In addition, a role for ATP acting on purinergic receptors has been postulated both in retina and neocortex (Biesecker et al., 2016; Mishra et al., 2016). Finally, ionic changes evoked by neural activity, in particular the increase in extracellular K⁺, a vasodilator at concentrations <12mM (Filosa et al., 2006), could also participate (see section on astrocytes and endothelium).

Recent studies have begun to shed light on the role of specific neuronal types and their mediators in neurovascular coupling. In whisker barrel cortex hyperemia induced by stimulation of the contralateral facial whiskers, pyramidal cells expressing the prostanoid-synthesizing enzyme COX-2 are activated from thalamic afferents (Lecrux et al., 2011) and contribute to the vascular response through prostaglandin E₂ (PGE₂) acting on EP2 and EP4 receptors (Lecrux and Hamel, 2016). These pharmacological observations are consistent with earlier findings in transgenic mice that COX-2, but not COX1, is required for the full expression of the vascular response (Niwa et al., 2000a; 2001a). Somatosensory stimuli also activate GABA interneurons (Lecrux et al., 2011). Increasing evidence suggests that interneurons may be critical for neurovascular coupling. Interneurons are enriched with vasoactive neurotransmitters and neuromodulators and have close associations with cerebral microvessels (Cauli and Hamel, 2010). In cerebral cortex and cerebellum, inhibitory interneurons are able to modify vascular diameter in brain slices through the vasoconstrictor NPY or the vasodilator NO (Cauli et al., 2004; Rancillac et al., 2006). In vivo, optogenetic stimulation of GABAergic interneurons induces biphasic hemodynamic responses, dilatation followed by constriction, closely resembling those induced by somatosensory activation in this preparation (Uhlirova et al., 2016b). While the constrictor response was attributed to NPY, the mediator(s) responsible for the dilatation could not be identified, but a role of NO or VIP was suggested (Uhlirova et al., 2016b). Interestingly, optogenetic activation of GABAergic interneurons increases CBF even if glutamatergic or GABAergic activity is pharmacologically blocked (Anenberg et al., 2015), suggesting that these cells can influence flow independently of the activity of their neuronal network, but the mechanisms of the vasodilatation remain to be established. While interneurons have the potential to modulate CBF, evidence for their participation in neurovascular coupling to physiological stimuli is limited. In mice deficient in the cell-cycle enzyme D2 cyclin, which lack stellate interneurons in the cerebellar molecular layer, the increase in blood flow produced by somatosensory activation of the cerebellar crus II is suppressed (Yang et al., 2000). Considering that stellate interneurons are enriched with nNOS and that nNOS-derived NO is the major determinant of functional hyperemia in cerebellum (Rancillac et al., 2006; Yang et al., 1999; Yang and Iadecola, 1998; Yang et al., 2003), these findings are consistent with the hypothesis that stellate interneurons mediate the response by releasing NO. However, more direct evidence in cerebellum or elsewhere in brain is lacking.

Another aspect of the involvement of neurons in neurovascular coupling pertains to neurovascular projections arising from subcortical nuclei. Locus coeruleus, raphe, and basal forebrain send diffuse adrenergic, serotoninergic and cholinergic projection, respectively, to the neocortex (Cohen et al., 1996; Hamel, 2006). These projections often terminate on astrocytic end-feet close to cerebral microvessels, and are able to powerfully modulate cerebral perfusion (Cohen et al., 1996; Toussay et al., 2013; F. Zhang et al., 1995). In particular, activation of the basal forebrain cholinergic system increases neocortical CBF diffusely, an effect attributable to acetylcholine release and endothelial NO production, and to activation of COX-2 expressing pyramidal cells and GABAergic interneurons (Lecrux and Hamel, 2016; F. Zhang et al., 1995). Although its lesion does not alter the changes in CBF induced by neural activity (Ibayashi et al., 1991), this pathway may contribute to maintain resting CBF and modulate the increases in CBF produced by somatosensory activation (Iadecola et al., 1983; Lecrux and Hamel, 2016; Lecrux et al., 2017). On the other hand, cortical norepinephrine levels, which depend in large part on the noradrenergic innervation from the locus coeruleus, have been shown to increase the vasoconstrictor tone in cerebral cortex and help focus oxygen delivery to the activated areas (Bekar et al., 2012). Therefore, these neurovascular pathways do not directly mediate the vasodilatation in response to activation, but, consistent with their broader role in neocortical neuromodulation (Lee and Dan, 2012), may contribute to shape the spatial and temporal features of the hemodynamic response (Table 1).

Little is known about the participation of perivascular nerves arising from the cranial ganglia in neurovascular coupling, but surgical ablation or pharmacological inactivation of the sympathetic and parasympathetic innervation does not affect the magnitude of the pial arterial dilatation evoked by somatosensory activation (Ibayashi et al., 1991). However, these nerves may protect cerebral blood vessels from forced vasodilatation during seizures or acute hypertension (Bill and Linder, 1976; Mueller et al., 1979). Furthermore, they may also contribute to the development and integrity of the vascular wall (Eichmann and Brunet, 2014), a process requiring molecular signals distinct from those in peripheral arteries (Brunet et al., 2014). Interestingly, the perivascular innervation has been implicated in generating resting state fMRI connectivity signals in surgically disconnected brain regions in humans (Warren et al., 2017). This finding, if confirmed, would suggest a role of perivascular nerves in the synchronization of hemodynamic signals across the brain and require a re-evaluation of the significance of fMRI-based assessments of functional neural connectivity.

Astrocytes: signal transmission to the local vasculature

Owing to their intimate association with both synapses and local microvessels, astrocytes are well positioned to link neural activity to microvascular function. Paulson and Newman proposed that astrocytes could influence blood flow by “siphoning” extracellular K⁺ released by neural activity into end-feet abutting cerebral microvessels (Paulson and Newman, 1987). However, subsequent studies in the retina of K_IR4.1 channel null mice, in which glia K⁺ siphoning is suppressed, showed that neurovascular coupling was still present suggesting that this mechanism may not be involved in retinal arteriolar dilatation (Metea et al., 2007).

Zonta et al. provided evidence that Ca²⁺ increases in astrocytes, evoked by electrical stimulation, could influence the diameter of adjacent microvessels in brain slices through metabotropic glutamate receptors (mGluR) and COX reaction products (Zonta et al., 2003). Subsequent studies using pharmacological approaches also in brain slices extended these initial observations demonstrating that increases in astrocytic Ca²⁺, triggered by neural activity, photolysis of caged Ca²⁺ or activation of mGluR5, have the potential to modulate microvascular diameter, producing constriction or dilatation depending on the oxygen levels (Gordon et al., 2008), resting vascular tone (Blanco et al., 2008) or extracellular K⁺ (Girouard et al., 2010). The postulated sequence of events leading to vasodilation is that glutamate released by synaptic activity activates mGluR on astrocytes leading to inositol 3-phosphate (IP3) mobilization and Ca²⁺ release from intracellular stores, phospholipase-A₂ (PLA₂) activation, release of arachidonic acid and production of PGE₂ and epoxyeicosatrienoic acids (EETs) via the COX and cytochrome p450 epoxygenase pathways, respectively (Attwell et al., 2010; Mishra, 2016). An alternative mechanism is that Ca²⁺ transients in astrocytes activate large conductance Ca²⁺ activated K⁺ (BK) channels in astrocytic end-feet leading to K⁺ release and vascular SMC relaxation through K_IR channels (Filosa et al., 2006). However, the relevance of this mechanism remains unclear because mice lacking a subunit of the BK channel have normal neurovascular coupling (Girouard et al., 2010), probably reflecting intervening compensatory mechanisms. Similarly, experiments in brain slices suggest that activity-induced release of D-serine by astrocytes may induce endothelial NO production (Stobart et al., 2013), a mechanism that also needs further study since neurovascular coupling is preserved in mice lacking endothelial NO synthase (Girouard et al., 2007).

Subsequent in vivo studies challenged the role of astrocytic Ca²⁺ transients and attendant upstream and downstream signaling in neurovascular coupling (reviewed in (Mishra, 2016; Petzold and Murthy, 2011; Uhlirova et al., 2016a)). The increases in Ca²⁺ may be too slow (Takano et al., 2006), often occurring after the onset of the hemodynamic response (Bonder and McCarthy, 2014; Nizar et al., 2013). Faster responses have been described (Lind et al., 2013), but questions have been raised about spillover of signals from neurons due to bulk labeling with organic Ca²⁺ dyes (Otsu et al., 2015; Petzold and Murthy, 2011; Uhlirova et al., 2016a). In experiments using genetically encoded astrocytic Ca²⁺ sensors in the visual cortex, no arteriolar blood flow changes were observed when astrocytic Ca²⁺ was modulated by chemogenetic approaches (Bonder and McCarthy, 2014). In contrast, in the activated olfactory bulb rapid increases in astrocytic Ca²⁺ preceding the vascular response were observed even with genetically encoded sensors (Otsu et al., 2015), suggesting regional differences in the role of astrocytic Ca²⁺ surge in neurovascular coupling. Questions have also been raised about the role of mGluR5 and IP3 receptors. mGluR5 expression is developmentally regulated and suppressed after the third week of life in mice (Sun et al., 2013), challenging the hypothesized mechanism that the Ca²⁺ increase in astrocytes in adult mice depends on these receptors. Indeed, in the olfactory bulb, mGluR5 were responsible for the Ca²⁺ increase in astrocytes of juvenile, but not adult mice (Otsu et al., 2015). In mice lacking IP3 type 2 receptors, the main regulators of Ca²⁺ release from intracellular stores in astrocytes, neurovascular coupling occurs in the absence of astrocytic Ca²⁺ elevations (Bonder and McCarthy, 2014; Jego et al., 2014; Nizar et al., 2013), although persistence of Ca²⁺ transients in astrocytic processes has been described in these mice (Srinivasan et al., 2015). The role of COX-1, a rate-limiting enzyme for PGE₂ synthesis, in astrocytic neurovascular coupling also remains unclear. The expression of COX-1 in astrocytes has been questioned (Anrather et al., 2011; Lau et al., 2014; Lecrux et al., 2011), and neurovascular coupling is not altered by COX-1 inhibitors or in COX-1 null mice (Liu et al., 2012; Niwa et al., 2001a; Rosenegger et al., 2015). Therefore, whatever the source and timing of the Ca²⁺ signal in astrocytes, the downstream signaling pathways leading to the vasodilatation remain unclear.

Recent data may help reconcile, at least in part, these discordant observations. The role of astrocytes in neurovascular coupling may be restricted to capillaries and may not involve arterioles, and the signaling pathways regulating the response of these vascular segments may be distinct (Biesecker et al., 2016; Mishra et al., 2016). In brain slices, electrical stimulation increases the diameter of cerebral capillaries and astrocytic Ca²⁺, an effect suppressed by AMPA or P2X1 receptor inhibition, suggesting that ATP released from neurons acts on astrocytic P2X1 receptors leading to the Ca²⁺ rise (Mishra et al., 2016). Contradicting previous evidence implicating PLA₂ and cytochrome p450 epoxygenase pathways (Attwell et al., 2010; Mishra, 2016), pharmacological studies indicated that the Ca²⁺ rise leads to activation of phospholipase-D2, arachidonic acid production, COX-1 activation and PGE₂ release, which, in turn mediates the vasodilatation by acting on EP4 receptors, presumably on pericytes (Mishra et al., 2016). In contrast, neurovascular coupling in arterioles does not involve astrocytes, but is mediated by NMDA receptors and neuronal NO production (Mishra et al., 2016). Accordingly, the arteriolar response to activation is blocked by NOS inhibition or NMDA receptor antagonism, which does not affect the response of capillaries (Mishra et al., 2016). Findings supporting a dual neurovascular regulation of capillaries and arterioles were also obtained in the retina. Ca²⁺ transients in Muller cells, assessed with genetically encoded sensors, were linked to dilatation of capillaries, although the timing between Ca²⁺ rise and vascular response could not be consistently resolved (Biesecker et al., 2016). The Ca²⁺ transients were not linked to dilatation of arterioles, and conditional deletion of IP3 type 2 receptors in Muller cells abolished the dilatation of capillaries but not arterioles. Although there is still considerable controversy about the role of capillaries and pericytes (discussed in subsequent sections) and COX-1 in neurovascular coupling (discussed above), the data may help explain some of the negative studies on the role of Ca²⁺ transients in neurovascular coupling, which focused on arteriolar responses. Taken together, these findings support the concept that astrocytic signals may be a critical link between neural activity and the microvasculature, at least in certain segments of the cerebrovascular network.

Endothelial cells: retrograde propagation of vasomotor responses

Endowed with powerful vasoactive agents (NO, prostanoids, endothelin, etc.), endothelial cells have long been known to regulate CBF in response to chemical and mechanical signals (Andresen et al., 2006; Seals et al., 2011), but their role in neurovascular coupling has not received attention until recently. Emerging evidence suggests a crucial role of the endothelium in the retrograde propagation of activity-induced neurovascular signals.

Hemodynamic considerations dictate that in a vascular network there has to be a coordinated dilatation of downstream and upstream vessels in order to increase flow while avoiding a “flow steal” from interconnected vascular territories (Segal, 2015). Pial arterioles are an important site of flow control (Figure 3), and signals generated by neural activity deep in the brain tissue have to be conveyed to upstream arterioles remote from the area of activation to increase flow efficiently. Indeed, visualization of pial arterioles in the somatosensory cortex revealed branch-selective vasodilatation supplying the activated site during somatosensory stimuli activating layer IV neurons, e.g., (Ngai et al., 1988). Since diffusion of vasoactive agents released by active neurons could not explain these remote vascular adjustments, it was proposed that, in analogy with the retrograde vasodilation observed in hamster cheek pouch by Duling (Segal, 2015), the vascular response generated in or near the activated area is conducted retrogradely along blood vessels engaging larger pial arteries upstream (Iadecola, 1993; Woolsey and Rovainen, 1991). Evidence was subsequently provided that isolated cerebral arterioles exhibit conducted vasodilation (Dietrich et al., 1996), and neural activity-induced retrograde propagation of vasodilatation was described in cerebellar cortex pial arterioles in vivo (Iadecola et al., 1997). A role for brain-derived vasoactive factors released in the subarachnoid space and acting on pial arterioles was also excluded (Ngai and Winn, 2002). At the same time, vascular mapping and fMRI studies revealed that during somatosensory activation vascular responses are first observed in deep cortical laminae, where the thalamic afferents terminate, and then more superficially, suggesting retrograde propagation of the vascular response (Silva and Koretsky, 2002; Uhlirova et al., 2016b). These observations supported the idea that activation restricted to a specific site generates a vascular response that propagates retrogradely to the upstream vascular segment supplying the activated region, but the cellular bases of the effect remained unclear.

In systemic vessels the endothelium is well known to participate in the retrograde propagation of vascular signals (Segal, 2015). In brain, however, damage of the pial vessel endothelium using the light-and-dye technique did not attenuate the peak pial dilatation produced by somatosensory activation (Xu et al., 2008), but the temporal profile of the vasodilatation was not resolved. More recently, Chen et al. used a similar approach to produce a highly localized lesion of the endothelium of a single somatosensory cortex pial arteriole, and found that the conducted vasodilation induced by somatosensory activation failed to propagate beyond the lesion site (Chen et al., 2014). Furthermore, wide-field endothelial lesions markedly altered the amplitude and temporal dynamics of the hemodynamic response, with a slower rise and no initial peak (Chen et al., 2014), implicating endothelial cells in the full expression and temporal coordination of the vasomotor response induced by somatosensory activation.

What drives the endothelial propagation of vasomotor signals? In peripheral vessels, conducted vasomotor responses have two components: a fast component mediated by Ca²⁺-activated K⁺ channels (K_Ca) and spreading through electrical coupling between endothelial cells, and a slow component mediated by Ca²⁺ waves triggering the endothelial release of NO and prostanoids (Segal, 2015; Tallini et al., 2007). Although much less in know about the mechanisms of retrograde vasodilatation in the cerebral vasculature, a recent study implicates endothelial K_IR channels, rather than K_Ca, in the mechanisms of the fast component. Enriched with K_IR channels, but not K_Ca channels, capillary cerebral endothelial cells are highly sensitive to K⁺ generated during neural activity, either from the abutting astrocytic end-feet or diffusion form nearby synapses (Longden et al., 2017). Owing to a unique capillary-parenchymal arteriole preparation, this study showed that micropipette application of 6–10mM K⁺ to capillaries generated a robust hyperpolarization in endothelial cells, transmitted retrogradely to penetrating arterioles at an estimated speed of 2mm/sec, and leading to hyperpolarization and relaxation of SMC (Longden et al., 2017). K⁺ application did not induce capillary dilatation, suggesting that the capillary was the K⁺ sensor and the upstream penetrating arteriole the effector of the vasodilation. This conclusion was supported by in vivo experiments demonstrating that the increase in capillary flow produced by K⁺ application is not due to local effects on capillary diameter but is linked to conducted dilatation of upstream arterioles (Longden et al., 2017). The propagation of the vasodilatation was blocked by K_IR inhibition with barium or endothelial deletion of K_IR1.2 channels, pointing to K_IR channels as key mediators of the conducted hyperpolarization (Longden et al., 2017). This rapid propagation mechanism could be mediated by ionic currents traveling through the endothelium via gap junctions and from endothelium to SMC via myoendothelial junctions (Segal, 2015; Tallini et al., 2007). Consistent with their role in neurovascular coupling, endothelial deletion of K_IR1.2 suppressed the increase in CBF produced by somatosensory activation (Longden et al., 2017). However, in contrast with the endothelial injury model in which the time course of the hemodynamic response was altered (Chen et al., 2014), in K_IR1.2 null mice the CBF increase evoked by whisker stimulation was uniformly reduced by ≈50% without major effects on its temporal profile (Longden et al., 2017). Although methodological differences cannot be excluded, the suppression of the flow response may suggest functions of endothelial K_IR1.2 channels beyond conducted vasodilatation. Conducted vasomotor responses, both vasodilation and vasoconstriction, can be generated also by other agents, for example acetylcholine and catecholamines, respectively (Bagher and Segal, 2011). In brain arterioles, ATP and prostaglandin F2α generate constriction followed by conducted vasodilatation (Dietrich et al., 1996), but their role in neurovascular coupling has not been explored.

Smooth muscle cells and pericytes: the vasomotor apparatus

Signals generated by neurons, astrocytes and endothelial cells ultimately engage the vasomotor apparatus to induce vasodilatation, reduce vascular resistance and increase blood flow. In brain as in other organs, SMC wrapping around resistance vessels are considered the major effectors of vasomotor responses and flow regulation (Cipolla, 2010), and neural activity leads to SMC relaxation in pial and penetrating arterioles (Drew et al., 2011; Ngai et al., 1988; Uhlirova et al., 2016b). SMC constrict and relax in response to a wide variety of vasoactive agents and to conducted vasodilatatory stimuli originating from downstream vessels (see previous section). Another key physiological characteristic of SMC is their ability to constrict or relax in response to increases or decreases in intravascular pressure, a phenomenon termed myogenic response (Cipolla, 2010; Longden et al., 2016). Cerebral arteries, especially penetrating arterioles, have a strong propensity to generate myogenic tone, a property that is essential for cerebral blood vessels to keep CBF stable during changes in arterial pressure within a certain range (cerebrovascular autoregulation) (Koller and Toth, 2012). Thus, blood pressure increases lead to vasoconstriction and decreases to vasodilatation. The resulting changes in vascular resistance keep CBF relatively stable. Cerebrovascular autoregulation and myogenic tone set the level of resting CBF and may contribute to neurovascular coupling (see next section). The changes in the contractile state of SMC are mediated by the interplay between, membrane potential and intracellular Ca²⁺, and Ca²⁺ sensitivity of the contractile apparatus, which ultimately controls the assembly of contractile proteins (Cipolla, 2010; Longden et al., 2016).

In capillaries, SMC are replaced by pericytes (Figure 5). Pericyte have an important role in establishing and maintaining vascular structure and the blood-brain barrier (BBB) (Armulik et al., 2011; 2010; Daneman et al., 2010), but have long been implicated also in flow regulation (Krueger and Bechmann, 2010). Some studies have shown that a proportion of pericytes are contractile, respond to brain-generated vasoactive signals, and are able to influence capillary diameter both in vitro and in vivo (Hall et al., 2014; Mishra et al., 2016; Peppiatt et al., 2006). In addition, a 30% pericyte loss in mice with haploinsufficiency of platelet-derived growth factor receptor-β (Pdgfr-β) is associated with reduced neurovascular coupling and brain oxygenation levels (Kisler et al., 2017b). However, other studies have failed to demonstrate a role of pericytes in flow regulation (Cudmore et al., 2016; Fernández-Klett et al., 2010; Hill et al., 2015; Wei et al., 2016). A major problem is that, due to the lack of specific markers, it has been difficult to distinguish pericytes from SMC at the arteriolar-capillary transition. Vessel size is not a reliable indicator of capillaries and the commonly used Pdgfr-β and proteoglycan NG2 are not specific pericyte markers (Armulik et al., 2011). Consequently, due to different criteria for pericyte identification, there has been considerable confusion about their functional roles (Attwell et al., 2016; Krueger and Bechmann, 2010). Pericytes are morphologically heterogeneous, but the functional significance of such diversity remains unclear (Armulik et al., 2011). One interpretation is that mural cells next to feeding arterioles, which encircle the vessel like SMC and contain SMC actin, are “contractile pericytes”, whereas cells running along the major axis of the capillary with short stubby processes projecting sideways are “non-contractile pericytes” (Attwell et al., 2016; Hartmann et al., 2015). Another point of view is that “contractile pericytes” are SMC located at the arteriole-capillary transition, and that “true” pericytes have no contractile function (Hill et al., 2015). A limitation of these studies is that the ultrastructural features of these cells, for example their relationship to the endothelial basement membrane which envelops the pericyte, a key feature for their identification (Armulik et al., 2011; Park et al., 2013), was not determined. A tracer recently introduced may provide a new tool to study pericytes (Damisah et al., 2017), but, as discussed in the final section of this review, other approaches would also be needed to reliably phenotype mural cells. Irrespective of whether these contractile cells are pericytes or SMC, the larger issue concerns the role of cerebral capillaries in the regulation of CBF during neural activity, which will be addressed next.

Which segment of the cerebrovascular network mediates neurovascular coupling?

Neural activity evokes coordinated vasomotor responses distributed over the vascular network supplying the activated area. The structural diversity of the NVU along the cerebrovascular tree (Figures 4,5) suggests differences in the role of each vascular segment in CBF regulation. In the cat neocortex, pooled measurements of pressure gradients between vascular segments indicated that 39% of the total resistance to flow is attributable to vessels upstream of pial arteries 300μm in diameter, 21% across the pial microcirculation (300 to 50μm pial arterioles), and 40% downstream of these vessels (De Silva and Faraci, 2016; Stromberg and Fox, 1972) (Figure 3). This distribution of vascular resistance seems to be conserved across species (De Silva and Faraci, 2016). Therefore, while the pial microcirculation is responsible for a sizable fraction of the total resistance, hence potential for flow control, an even larger component is attributable to downstream intracerebral vessels. However, the breakdown of the vascular resistance among penetrating arterioles, precapillary arterioles, capillaries and venules has been difficult to assess because intravascular pressure measurements in smaller vessels embedded deep in the brain is not feasible. Capillaries are closer to neurons than arteries and, due to their large surface area, minimal changes in their diameter could produce large changes in flow (Attwell et al., 2010). Consequently, capillaries are well suited to regulate CBF in response to the metabolic needs of the brain. Computational studies attempting to assess segmental cerebrovascular resistance have given varying results due to difficulties in setting boundary parameters in idealized networks (Schmid et al., 2017). In two recent studies, in which realistic vascular networks of the mouse somatosensory cortex were used, one attributed the bulk of resistance to arterioles in deep cortical layers and capillaries in superficial layers (Schmid et al., 2017), and the other to capillaries “adjacent to feeding arterioles” (Gould et al., 2017), highlighting the uncertainty of estimating segmental vascular resistance.

Experiments investigating capillary vasoactivity have also led to inconsistent results. In brain slices neural activity was found to induce capillary dilatation due to relaxation of pericytes (Hall et al., 2014). Related in vivo experiments demonstrated that the capillary hemodynamic response precedes arteriolar dilatation, suggesting that the increase in capillary flow is not a passive consequence of the increased flow in upstream arterioles (Hall et al., 2014; Kisler et al., 2017b). Based on these findings and on the mouse cerebrovascular tree topology (Blinder et al., 2013) it was estimated that capillaries are responsible for 84% of the flow increase produced by neural activity (Hall et al., 2014). However, other studies demonstrated that somatosensory activation does not change capillary diameter, and that relaxation of upstream arterioles accounts for the bulk of the flow response (Drew et al., 2011; Hill et al., 2015; Wei et al., 2016). A lack of capillary vasoactivity was also shown with local application of K⁺ (Longden et al., 2017), seizures (Fernández-Klett et al., 2010), ischemia due to carotid artery occlusion or carotid stenosis (Damisah et al., 2017; Hill et al., 2015), and isoflurane anesthesia (Cudmore et al., 2016). One possible explanation for these discordant results is that studies reporting capillary vasoactivity may have focused on transitional microvessels closer to arterioles and endowed with α-actin-containing mural cells (SMC?), whereas negative studies on capillary segments with “non-contractile” mural cells (pericytes?). Furthermore, considering the small activity-induced increases in capillary diameter, e.g., 1% in (Kisler et al., 2017b), the resolution of the imaging approach used could also be a factor (Drew et al., 2011). However, as discussed in the next section, whether or not capillaries are major contributors to cerebrovascular resistance, does not rule out their participation in neurovascular coupling.

Local and conducted vasomotor responses in neurovascular coupling

Capillaries are uniquely positioned to detect neuronal and astroglial signals, which led to the hypothesis that neurovascular coupling could be initiated at the microvascular level and conducted upstream (Cox et al., 1993; Iadecola, 2004; 1993; Woolsey and Rovainen, 1991). As reviewed in the previous section, the evidence indicates that, irrespective of their ability to directly regulate vascular resistance, capillaries are equipped with the signaling mechanisms to detect neural activity and transmit the signal to α-actin-containing mural cells and SMC in vessels upstream. A likely scenario for the transmission and coordination of the vascular response across the cerebrovascular tree is the following. Activation-induced increases in extracellular K⁺ triggers hyperpolarization of capillary endothelial cells and, possibly, pericytes via K_IR channels (Longden et al., 2017) (Figure 7). The hyperpolarization propagates upstream through inter-endothelial gap junctions and reaches SMC in penetrating arterioles, most likely through myoendothelial junctions, producing their relaxation (Longden et al., 2017) (Figure 7). At the same time, capillary hypoxia increases the deformability of red blood cells, reducing blood viscosity, and increasing capillary flow independently of their diameter (Wei et al., 2016) (Figure 7). The resulting shear stress on the endothelium of feeding arterioles produces SMC relaxation by releasing endothelial vasorelaxing factors (flow-mediated vasodilation) (Koller and Toth, 2012) (Figure 8). Owing to their strong myogenic response (Longden et al., 2016), penetrating arterioles may also relax further in response to the pressure drop produced by the retrograde vasodilatation. The SMC relaxation produced by conducted hyperpolarization and myogenic tone could be complemented by the direct action on SMC of vasoactive agents released by neurons and glia during neural activity (NO, adenosine, ATP, prostanoids, etc.) (Table 1) (Figure 8). Finally, in upstream pial arterioles remote from the site of activation, the vasodilatation is likely to result from two mechanisms: the conducted vasodilatation propagating from arterioles downstream and flow-mediated vasodilatation/myogenic response locally (Figure 9). Accordingly, the pial arteriolar relaxation observed on the brain surface may not be driven directly by neural signals, but by intrinsic vascular signals traveling retrogradely from capillaries and arterioles at the activated site deep in the substance of the brain. It must be noted that lesion of the glia limitans near pial arterioles with L-2-aminoadipidic acid attenuates the vasodilatation produced by neural activity, implicating astrocytes in the transmission of the vascular response (Xu et al., 2008). However, this observation has been difficult to interpret since damaging effects of this harsh treatment on other cell types cannot be ruled out. Concerning the role of conducted vasodilatation, a caveat is that there are likely to be regional variations in this mechanism related to vascular topology, modality of activation, and location of the site of activation relative to resistance vessels. Furthermore, as mentioned above, there are regional differences in the cellular bases neurovascular response, e.g., olfactory bulb, neocortex, cerebellum, etc., which may also play a role. Regional variations not withstanding, the available evidence suggests that neurovascular coupling reflects a coordinated multicellular response acting at different levels of the cerebrovascular network through segment-specific mechanisms.

Figure 7. Neurovascular coupling at the capillary level.

Synaptic activity leads to release of extracellular K⁺ and increase O₂ utilization. In turn, K⁺ activates K_IR channels on endothelial cells, and pericytes leading to endothelial cell hyperpolarization, which is conducted retrogradely via gap junctions linking adjacent endothelial cells. At the same time the reduction in O₂ leads to increase deformability of red blood cells (RBC), reducing their viscosity and increasing capillary flow. The resulting shear stress on capillary endothelium may also contribute to their hyperpolarization. The role of astrocytes is less clear, but activation of K⁺ channels in astrocytes may also contribute to increase K⁺ in the vicinity of endothelial cells, and due to their link to adjacent astrocytes could also play a role in the retrograde propagation of the hyperpolarization. ATP released during neural activity could increase astrocytic Ca²⁺ via P2X1 receptors and relax contractile mural cells via reaction products of the COX pathways (Mishra et al., 2016), but as discussed in the text there is no consensus on the ability of capillaries to dilate. Neurovascular contacts from interneurons or subcortical pathways are not depicted (see figure 5).

Figure 8. Neurovascular coupling at the level of intraparenchymal arterioles.

The propagated endothelial hyperpolarization arising from capillaries is transferred to smooth muscle cells (SMC) via myoendothelial junctions, resulting in their relaxation and vasodilatation. The signal continues to be transferred retrogradely to vessels upstream through gap junctions linking SMC as well as endothelial cells. The vasodilatation is complemented by vasoactive factors released by nearby activated neurons and astrocytes, which contributes to sustain the vasodilatation in its propagation to upstream vessels. In addition, the drop in intravascular pressure and increased flow velocity caused by the vasodilatation downstream may contribute to smooth muscle hyperpolarization and relaxation by activating the myogenic response/flow-mediated vasodilatation (FMV). Therefore, at the arteriolar level the vasodilatation has two components: propagated response from capillaries and local response from activated neurons and astrocytes. Neurovascular contacts from interneurons or subcortical pathways are not depicted (see figure 5). Ado: adenosine; Glu: glutamate; Prost: prostanoids.

Figure 9. Neurovascular coupling at the level of pial arteriole.

At this level the vasodilation depends on the propagated endothelial and SMC hyperpolarization, and, possibly, activation of the myogenic response/flow mediated vasodilation (FMV). Therefore, at the level of pial arterioles the hemodynamic response is not directly related to synaptic activity, but depends on the retrograde propagation of vasodilatation arising from smaller arterioles and capillaries surrounding the activated area. Perivascular nerves are not depicted (see figure 4).

The neurovascular unit: beyond blood flow

Neurovascular coupling provides a striking example of the close interaction between the brain and its vessels. However, the NVU has also other structural components and functional attributes not directly related to cerebral perfusion that are critically important for brain health. These are briefly reviewed in the following sections.

Brain development

Despite their different embryonic origin, the proliferation, fate determination, migration and terminal differentiation of neural progenitors is intertwined closely with that of vascular cells (Wälchli et al., 2015). A remarkably common array of signaling molecules is involved in both brain and vascular development (Raab and Plate, 2007). Early in brain development, chemoattractant signals, e.g., VEGF-A, secreted by neural progenitors in the subventricular zone guide the in-growth of blood vessels from the perineural vascular plexus, the primitive vascular network surrounding the brain (Raab et al., 2004). In turn, endothelial cells in the subventricular zone and hippocampal subgranular zone provide instructive cues critical for adult neurogenesis, which continues throughout life (Raab and Plate, 2007). The vasculature acts as scaffold for the migration of neuronal progenitors through BDNF produced by endothelial cells (Snapyan et al., 2009), a process that also depends on astrocytic VEGF signaling (Bozoyan et al., 2012). Blood vessels also guide the migration of oligodendrocyte precursors (Tsai et al., 2016), and, in turn, oligodendrocyte precursors are essential for postnatal white matter vascularization through HIF1-dependent transcription (Yuen et al., 2014). The vital trophic and metabolic interaction between brain cells and vessels continues into adulthood, when endothelial metabolism and growth factors contribute to the survival of neurons, astrocytes and oligodendrocytes (Brix et al., 2012; Carmeliet and Ruiz de Almodovar, 2013), and in repair processes following acute brain injury (Hayakawa et al., 2012).

Blood-brain barrier

Neurovascular interactions are also critical for the formation and maintenance of the BBB (Andreone et al., 2015; Zhao et al., 2015a). Cerebral endothelial cells are joined by impermeable tight junctions, express unique molecular transporters and are endowed with endocytotic systems for transcytosis, the transfer of molecules through cells in membrane-bound vesicles (Zhao et al., 2015a). While Wnt/β-catenin signaling from neuronal precursors is critical for vascular patterning and BBB development prenatally (Andreone et al., 2015), after birth astrocytes contribute to the maintenance of the BBB through sonic hedgehog as well as β-catenin, which promote tight junction expression and integrity (Alvarez et al., 2011; Tran et al., 2016). Pericytes are also required for normal BBB development, and pericyte deficiency increases BBB permeability by promoting endothelial transcytosis (Armulik et al., 2010; Daneman et al., 2010). Suppression of transcytosis may depend on the endothelial expression of the omega-3 fatty acid transporter mfsd2a, a member of the major facilitator superfamily of secondarily-active transporters (Andreone et al., 2015). Mfsd2a establishes a unique lipid environment in cerebral endothelial cells that suppresses caveolae-mediated transcytosis (Andreone et al., 2017). These data suggest that the two key features of the BBB, transcytosis and tight junctions, are regulated independently, which may have disease relevance. For example in ischemic stroke, a condition association with BBB disruption, the early opening of the BBB, which is pathogenically relevant, is mediated by an increases in transcytosis and followed by tight junction disruption days later after tissue damage sets in (Knowland et al., 2014).

Cerebrovascular matrix: from structural support to signal transduction

An integral part of the NVU, the matrix embeds the neurovascular cellular assembly and is comprised of: (a) collagen subunits, constituting the structural framework, (b) proteoglycans, which bind water and secreted regulatory factors, such as growth factors, and (c) glycoproteins, which subserve a wide variety of functions (Haffner et al., 2016; Joutel et al., 2015). Matrix proteins are involved in signal transduction by binding to cell-surface receptors, such as integrins, and regulating the distribution secreted factors with growth promoting and regulatory function. Initial studies suggest that the matrix of cerebral vessels varies in composition depending on the specific segment of the cerebrovascular tree. For example, collagen and glycoproteins are more abundant in pial arteries, possibly reflecting their ability to withstand higher intravascular pressures, whereas proteoglycans, rich in regulatory factors, predominate in microvessels (Badhwar et al., 2014; Chun et al., 2011). A significant proportion of matrix proteins is located in the basal lamina that separates the different cellular compartment of the NVU (endothelial, smooth muscle, astrocytes), collagen IV being most abundant (Joutel et al., 2015). Highlighting the critical role of the matrix in cerebrovascular structure and function, mutations of matrix and matrix related proteins are associated with hereditary cerebral arteriopathies affecting small vessels, including collagen IV-related small vessel diseases, cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL), and cerebral autosomal recessive arteriopathy with subcortical infarcts and leukoencephalopathy (CARASIL) (Haffner et al., 2016; Joutel et al., 2015). These conditions, although uncommon, have provided valuable insight into sporadic small vessels disease, a major cause of cognitive impairment on vascular basis and a contributor to AD (Iadecola, 2013).

Perivascular compartment: clearance pathways and immune surveillance

The perivascular compartment has attracted much interest as a major avenue for the removal of potentially toxic byproducts of brain activity released in the extracellular space, including Aβ and tau (Tarasoff-Conway et al., 2015) (Figure 10). Several vascular clearance pathways have been proposed, with particular reference to factors involved in neurodegenerative diseases. A transvascular pathway has been described through which BBB transporters transfer Aβ from the perivascular space to the circulating blood and vice-versa, involving the low density lipoprotein receptor-related protein-1 (LRP1) and the receptor for advanced glycation end-products (RAGE), respectively (Deane et al., 2012; 2004; Zhao et al., 2015b) (Figure 10). Pathological levels of Aβ lead to LRP1 degradation via the proteasome, promoting vascular Aβ accumulation (Deane et al., 2004; Park et al., 2013). Pericytes may also be involved since brain Aβ clearance is markedly reduced Pdgfr-β-deficient mice crossed with mice overexpressing APP, resulting in Aβ accumulation in brain and blood vessels (Sagare et al., 2013). Interestingly, tight junction proteins at the BBB may hinder Aβ clearance, because their disruption enhances the disposal of brain Aβ into the blood (Keaney et al., 2015).

Figure 10. The perivascular space and its clearance pathways.

The perivascular space surrounding penetrating arterioles is involved in the clearance of unwanted molecules form the brain, including, but not limited to, Aβ and tau. Products of cerebral activity and metabolism reach the perivascular space by diffusion and can be disposed of through different pathways. A transvascular pathway is thought to rely on scavenger receptors and molecular transporters that carry the unwanted molecules across the vascular wall. A perivascular pathway has long been proposed in which the molecules are transported retrogradely along vascular basement membranes reaching the subarachnoid space and eventually discharging into the cervical nodes. A paravascular (glymphatic) pathway relies on aquaporin-4 channels (AQP4) on astrocytic end-feet to allow the CSF to enter the interstitial space, creating a convective flow that clears unwanted molecules from the brain and feeds into the perivascular venous side eventually draining into dural lymphatics or the cribriform plate. Abbreviations: CD36: cluster of differentiation 36; LRP1: low-density lipoprotein receptor-related protein-1; PICALM: Phosphatidylinositol Binding Clathrin Assembly Protein; SR-B: scavenger receptor B.

A perivascular pathway has long been proposed in which solutes leave the brain parenchyma by traveling retrogradely along the vascular basal lamina, eventually reaching cerebral arteries on the brain surface and draining into cervical lymphnodes (Hladky and Barrand, 2014; Morris et al., 2016). The routes linking the arterial wall to cervical lymphnodes have not been clearly defined, but they may involve the dural lymphatic system and the CSF outflow pathway through the nasal cribriform plate (Hladky and Barrand, 2014; Louveau et al., 2016).

A paravascular pathway (glymphatic system) runs in opposite direction to the perivascular pathway and involves the aquaporin-4 water channel enriched in astrocytic end-feet. By imaging the movement of fluorescent dyes injected into the CSF of the cisterna magna, an expansion of the subarachnoid space between the cerebellum and the dorsal medulla, the dye is seen first around arteries on the brain surface, then in the parenchyma, and, later, in draining veins (Iliff et al., 2012). Lack of aquaporin-4 channels slows the movement of the dye dramatically (Iliff et al., 2012). On these bases, it was proposed that the CSF penetrates into the brain substance by tracking around arteries, enters the interstitial space through aquaporin-4 channels in perivascular astrocytic end-feet, and, by exchanging with the interstitial fluid, carries the waste away from the brain parenchyma, exiting the interstitial space through aquaporin-4 channels surrounding the venous perivascular space (Nedergaard, 2013). Eventually, the CSF loaded with waste collected from the interstitial space drains into the dural lymphatics or other CSF exit pathways (Nedergaard, 2013). This system may be involved in the clearance of brain Aβ and tau (Iliff et al., 2012), and is modulated by the brain’s physiological state, particularly sleep (Iliff et al., 2014). Aging, brain trauma, and brain microinfarcts impede this clearance pathway (Kress et al., 2014; Plog et al., 2015; M. Wang et al., 2017), raising the possibility that failure of the glymphatic system is involved in the brain dysfunction associated with these conditions.

These studies raise several questions. In the paravascular pathway, it is unclear how protein waste enters the perivenous compartment and how it reaches the lymphatics. The perivascular and paravascular clearance models are based on different methodological approaches, i.e., injection of markers either in the brain parenchyma (perivascular) or CSF (paravascular), which may introduce artifacts (Hladky and Barrand, 2014). Furthermore, the anatomical relationships between the perivascular and paravascular pathways, which run in opposite directions, and how they share the periarterial compartment need further exploration (Bedussi et al., 2017; Morris et al., 2016). Finally, the relative contribution of these pathways vs. endogenous clearance mechanisms, e.g., enzymatic degradation, in the human brain remain to be established (J. S. Miners et al., 2014).

The perivascular space also houses innate immune cells, which are involved in immune homeostasis and have a profound influence on the response of the brain to infectious or immune challenges. PVM are a subgroup of resident brain macrophages located in the perivascular space, juxtaposed to the outer wall of penetrating arteries and veins (Prinz et al., 2017) (Figure 4). PVM originate from erithromyeloid precursors in the yolk sac and, like microglia, enter the brain early in development (Prinz et al., 2017). PVM promote hypothalamic-adrenal axis activation and autonomic responses in models of cardiac ischemia, stress, and inflammation, may harbor the HIV virus infecting the brain, and participate in tumor angiogenesis and cerebrovascular repair (Prinz et al., 2017). Capable of producing large amounts of free radicals in close proximity to the vessel wall, PVM are a major source of vascular oxidative stress and key contributors to neurovascular and/or cognitive dysfunction in models of chronic hypertension (Faraco et al., 2016b) and Aβ overproduction (Park et al., 2017). On the other hand, owing to their scavenging function, PVM may prevent cerebrovascular accumulation of amyloid in models of cerebral amyloid angiopathy (CAA) (Hawkes and McLaurin, 2009; Thanopoulou et al., 2010). Overall, PVM are emerging as a novel constituent of the NVU that awaits further exploration.

The NVU in neurodegenerative disease

Due to their age dependence and lack of effective treatments, neurodegenerative diseases are predicted to grow to epidemic proportions over the next several decades (Gammon, 2014). Since alterations of the NVU are anticipated to lead to brain dysfunction and damage (Figure 11), there has been a growing interest in exploring the potential contribution of neurovascular dysfunction to neurodegeneration (De-la-Torre, 2017). The resulting body of work has provided evidence for a wide variety of neurovascular alterations in neurodegenerative diseases (Table 2), reviewed in the next sections.

Figure 11. Potential pathogenic mechanisms by which neurovascular dysfunction can cause brain dysfunction.

Alterations of the NVU may lead to reductions in CBF below the threshold required for normal brain oxygenation leading to hypoxia. BBB dysfunction may alter the homeostasis of the brain internal milieu by limiting the delivery of glucose and other nutrients and impairing the clearance of unwanted metabolites through efflux transporters. Reduction in trophic factor production by NVU cells may increase neuronal and glial vulnerability and susceptibility to disease. Alterations of clearance pathways may promote the accumulation of molecules, such as Aβ and tau, leading to proteinopathies.

Table 2.

Evidence of neurovascular dysfunction in major neurodegenerative diseases

Disease	Vascular morphology	Blood-brain barrier	Cerebral perfusion	Selected references
Alzheimer’s disease	Microvascular damage Microbleeds WM lesions ATS in cerebral arteries	Early increase in BBB perm. to Gd Late small increase in Qalb	Reduced CBF before symptoms Reduced neurovascular coupling	Janelidze et al, 2017 Toledo et al., 2013 Roher et al., 2004 Montagne et al, 2015 Kisler et al., 2017a
Frontotemporal dementia	Microvascular damage Microbleeds WM lesions Perivascular astrocyte damage	Increase in Qalb	Reduced CBF before symptoms	Sudre et al., 2017 Janelidze et al., 2017 Dopper et al., 2016 Martin et al., 2001 Thal et al., 2015
Amyotrophic lateral sclerosis	Microvascular damage RBC extravasation in some cases	Increase in Qalb Reduced TJ proteins Plasma protein extravasation	Reduced CBF correlates with disease progression Reduced neurovascular coupling	Murphy et al., 2012 Rule et al., 2010 Abrahams et al.,1996 Henkel et al., 2009 Winkler et al., 2013
Idiopathic Parkinson disease	Microvascular damage late in disease Increased endothelial cell nuclei in SN	Increase in Qalb in late phase BBB Pgp dysfunction Increase in VEGF in CSF	Early CBF reductions Preserved neurovascular coupling	Janelidze et al., 2015 Kortekaas et al., 2005 Pisani et al., 2012 Rosengarten et al., 2010 Al-Bachari et al., 2014 Guan et al., 2013 Faucheux et al., 1999
Dementia with Lewy bodies	Microvascular damage ATS in cerebral arteries	Increase in Qalb	Early CBF reductions	Janelidze et al., 2017 Roquet et al., 2016 Ghebremedhin et al., 2010 Miners et al., 2014

Abbreviations: ATS: atherosclerosis; Qalb: CSF/Plasma albumin ratio (index of BBB permeability); Gd: gadolinium; Pgp: P-glycoprotein efflux transporter; SN: substantia nigra; TJ: tight junctions; WM: white matter.

Alzheimer’s disease

The most common cause of dementia in the elderly, AD is characterized pathologically by the accumulation of the Aβ in extracellular amyloid plaques and hyperphosphorylated tau in intracellular neurofibrillary tangles, which are thought to induce neuronal dysfunction and damage (De Strooper and Karran, 2016). It has long been known that cerebrovascular alterations are associated with AD (De-la-Torre, 2017), but only recently the contribution of vascular factors has received direct attention. AD brains have microvascular alterations (increased capillary tortuosity, capillary rarefaction, thickened basement membrane, “string vessels” etc.) (Love and J. S. Miners, 2016) and more atherosclerosis in intracranial vessels than age matched controls (Roher et al., 2004). Community-based pathological studies of clinically diagnosed AD have revealed that AD pathology (plaques and tangles) and ischemic lesions frequently coexist in the same brain (Schneider et al., 2007; Toledo et al., 2013), and that ischemic lesions lower the threshold for clinical dementia (Kapasi et al., 2017). Resting CBF is reduced and hemodynamic responses to neural activation are attenuated in the presymptomatic phase of AD (reviewed in (Kisler et al., 2017a)), suggesting early involvement in the disease process. Similarly, neurovascular dysfunction is observed in patients with CAA (Charidimou et al., 2017). BBB permeability is increased in patients with prodromal AD, an effect associated with increased soluble Pdgfr-β in the CSF, suggesting pericyte degeneration (Montagne et al., 2015). A pathogenic role for vascular factors in AD has also been suggested by epidemiological studies indicating that cerebrovascular diseases and AD have similar risk factors (hypertension, diabetes, obesity, etc.), and that improved cardiovascular health may have contributed to the recently reported reduction in AD incidence (Pase et al., 2017). Indeed, vascular risk factors, such as hypertension, promote amyloid deposition (Gottesman et al., 2017; Petrovitch et al., 2000; Rodrigue et al., 2013), attributable to reduced Aβ clearance and/or enhanced amyloid precursor protein (APP) processing (Carnevale et al., 2012; Faraco et al., 2016a).

Most studies on the mechanisms of neurovascular dysfunction in AD have focused on the role of Aβ (Kisler et al., 2017a). Thomas et al. first reported that Aβ, in addition to causing neuronal dysfunction, also impairs the ability of endothelial cells to relax systemic vessels in vitro (Thomas et al., 1996). Using mice overexpressing mutated APP (Tg2576 mice), it was subsequently established that Aβ impairs all the major factors regulating the cerebral circulation: neurovascular coupling, endothelial function and cerebrovascular autoregulation (Iadecola et al., 1999; Niwa et al., 2002a; 2000b). Aβ was also found to reduce resting CBF and to promote vasoconstriction (Niwa et al., 2002b; 2001b), resulting in increased susceptibility to ischemic brain injury (F. Zhang et al., 1997). These neurovascular alterations were observed before amyloid plaques and behavioral deficits, suggesting an early pathogenic role, and were subsequently confirmed in other APP overexpression models (Beckmann et al., 2003; Tong et al., 2005). Mechanistically, Aβ may act on the innate immunity receptor CD36 in PVM, inducing vascular oxidative stress through the superoxide synthesizing enzyme NADPH oxidase (Han et al., 2008; Park et al., 2017; 2013; 2008; Tong et al., 2005). Radicals, in turn, cause endothelial dysfunction via DNA damage, poly-ADP ribose activation, ADP-ribose-mediated TRPM2 channel opening and Ca²⁺ overload (Park et al., 2014b). In addition to failure in LRP1-mediated transvascular Aβ clearance (Deane et al., 2004), innate immunity receptors, such as CD36, also promote CAA (Park et al., 2013), in which Aβ accumulation leads to vascular oxidative stress, neurovascular dysfunction, and, with aging, irreversible damage to the vasomotor apparatus (Park et al., 2014a).

Frontotemporal dementia

Frontotemporal dementia (FTD), the leading cause of dementia in middle age, comprises a heterogeneous group of neurodegenerative conditions characterized by diverse cognitive and behavioral manifestations (Olney et al., 2017). Familial forms of the disease are often caused by mutations of the progranulin, tau, or C9orf72 gene, but sporadic forms are more common (Olney et al., 2017). Resting CBF is reduced in the frontal, parietal and temporal cortex of presymptomatic patients, an effect independent of atrophy (Dopper et al., 2016). Pathological studies have described microvascular damage, microhemorrhages and white matter ischemic lesions, more pronounced in older individuals, suggesting that the vascular disease may be independent of FTD pathology and age related (Sudre et al., 2017; Thal et al., 2015). BBB alterations have also been reported (Janelidze et al., 2017). On the other hand, astrogliosis and astrocytic damage, prominent pathological features of FTD, correlate regionally with reductions in CBF (Martin et al., 2001), implicating astrocyte loss in the flow reduction and, possibly, in the mechanisms of the disease. Studies in mouse models reflecting selected aspects of FTD have confirmed some of these neurovascular alterations. In agreement with pathological studies showing microhemorrhages, progranulin-deficient mice exhibit increases BBB permeability and hemorrhages after stroke, but resting CBF or cerebrovascular reactivity were not reduced (Jackman et al., 2013). The BBB dysfunction was attributed to alterations in the structure of endothelial tight junctions, which appeared shorter and simplified (Jackman et al., 2013). BBB alterations have also been reported in mice overexpressing tau mutations found in FTD, e.g., P301L (Blair et al., 2015). However, in these mice the cerebrovascular reactivity to pCO₂ was reported to be increased (Wells et al., 2015), ruling out global suppression of vasomotor function.

Amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis (ALS) is characterized by dysfunction and degeneration of upper and lower motor neurons resulting in progressive weakness and paralysis (van Es MD et al., 2017). Up to 50% of cases have cognitive dysfunction of the type observed in FTD, and up to 15% of ALS cases also develop FTD (van Es MD et al., 2017). Alterations in the BBB and blood-spinal cord barrier were initially suggested to play a role in the disease based on increased plasma protein in the CSF and on microvascular alterations in spinal cord microvessels in autopsy specimens, e.g., reduced tight junction proteins, endothelial and pericyte damage, capillary rarefaction, plasma protein extravasation (Evans et al., 2013; Winkler et al., 2013). Studies in a mouse model of familial ALS caused by mutation of the free radical scavenging enzyme SOD1 have demonstrated uncoupling between spinal cord blood flow and glucose metabolism, as well as increased permeability of spinal cord capillaries and microhemorrhages, which precede motor neuron degeneration (Miyazaki et al., 2012; Zhong et al., 2008). Prevention of the barrier alterations with an activated protein C analog delayed the development of disease (Winkler et al., 2014). Although frank microhemorrhages were not observed by high resolution MRI (Verstraete et al., 2010), BBB dysfunction and red blood cell extravasation may be present in a subset of patients (Winkler et al., 2013). A separate line of investigations also suggests the involvement of neurovascular factors in motor neuron degeneration. In mice, deficiency of VEGF leads to motor neuron dysfunction and degeneration (Poesen et al., 2008), and administration of VEGF rescues motor neuron dysfunction in SOD1 mutant mice (Storkebaum et al., 2005). Initial genetic studies suggested a link of ALS with several VEGF haplotypes resulting in low VEGF production, which were not later confirmed except for a minor gender-specific effect (Lambrechts et al., 2009). The protection by VEGF is likely to result from its neurotrophic effects, although a microvascular contribution cannot be ruled out (Quaegebeur et al., 2011). However, vasoprotective effects are difficult to reconcile with the observation that increased vascular permeability may be damaging in ALS, since VEGF increases vascular permeability (Quaegebeur et al., 2011). Decreases in cortical perfusion, including non-motor areas, are well described early in the disease and correlate with disease progression (Murphy et al., 2012; Rule et al., 2010). In ALS patients with frontal cognitive dysfunction, the increase in CBF evoked by activation tasks in frontal association areas is reduced, an effect that is more widespread in patients with more severe cognitive deficits (Abrahams et al., 1996). These findings point to a more global alteration in neurovascular regulation, which could play a role in the cognitive manifestation of the disease.

Parkinson disease

Idiopatic Parkinson disease (PD) is a neurodegenerative disorder characterized by movement and cognitive deficits in which the pathology starts in the brainstem, particularly the nigrostriatal pathway, and spreads to the cortex (Przedborski, 2017). Postmortem studies of advanced cases have shown capillary damage and fragmentation in frontal cortex and brainstem nuclei (Guan et al., 2013), as well as increased endothelial cell nuclei in the substantia nigra (Faucheux et al., 1999). Increases in VEGF in the CSF have been reported (Janelidze et al., 2015), together with evidence of increased BBB permeability in the late stages of the disease (Kortekaas et al., 2005; Pisani et al., 2012). On the other hand, CBF reductions have been observed early in the disease course in neocortical regions without overt pathology (Borghammer et al., 2010; Fernández-Seara et al., 2012). In PD with cognitive impairment CBF reductions in parietal regions correlated with the cognitive dysfunction (Al-Bachari et al., 2014), suggesting a link between cognitive deficits and CBF reductions. However, cerebrovascular reactivity seems to be preserved since the hemodynamic response to visual stimulation or to increases in arterial pCO₂ (hypercapnia) are not attenuated (Al-Bachari et al., 2014; Rosengarten et al., 2010). An interesting hypothesis is that the CBF reductions is due alterations of the cholinergic, serotoninergic and noradrenergic neurovascular innervation of the neocortex (Al-Bachari et al., 2014; Fernández-Seara et al., 2012). The experimental observation that these subcortical pathways may influence resting CBF and not neurovascular coupling (see section on neurovascular coupling), would be consistent with this hypothesis.

Dementia with Lewy Bodies

CBF is reduced in frontal, insular, temporal and occipital cortex in dementia with Levy bodies (DLB), a condition characterized by fluctuations in cognition, hallucinations and parkinsonism (Galasko, 2017). The changes in CBF occur in the prodromal stage of the disease (Roquet et al., 2016), and may be due to a reduction in energy metabolism (Ishii et al., 2015). While vascular alterations, such as reduced VEGF, small vessel disease, and atherosclerosis, have been described in the occipital cortex (Ghebremedhin et al., 2010; S. Miners et al., 2014), there might be an inverse relationship between Lewy body pathology and vascular pathology (Ghebremedhin et al., 2010). Therefore, Levy body pathology may stave off vascular pathology and vice-versa.

Where do we go from here?

The concept of NVU has come a long way since 2001. Central to brain health, neurovascular interactions have emerged as essential contributors to brain development and vital for maintaining the homeostasis of the brain internal milieu. Neurovascular coupling has evolved from a unidimensional process driven by the direct effect of neuronal mediators on local blood vessels, to a multi-dimensional one in which a multitude of mediators released from multiple cell types engages unique signaling pathways and effector systems on different cerebrovascular segments in a coordinated fashion (Table 1). The resulting carefully orchestrated vascular response involves the entire cerebrovascular tree and leads to spatially and temporally focused increases in CBF. The evidence implicates neurons and astrocytes in the generation, modulation and transmission of the signal to local microvessels, capillary endothelial cells in the retrograde propagation of vasomotor signals, and contractile mural cells and SMC of progressively larger vascular segments as the main effector of the CBF increase (Table 1). Therefore, neural activity initiates the process, but the implementation of the hemodynamic response requires the coordinated interaction of other NVU cells across the whole cerebrovascular network. Significant progress has also been made in the pathobiology of the NVU. Increasing evidence implicates NVU dysfunction in major neurodegenerative diseases, in which the role of vascular factors was traditionally not considered. This new knowledge has raised a number of questions pertaining to the inner workings of the NVU and to its role in brain diseases.

Knowledge gaps in neurovascular coupling

The contribution of feed-forward and feedback mechanisms to the flow increase remains unclear. In particular, the region-specific factors determining the recruitment of one mechanism versus the other and the cells and mediators driving the attendant hemodynamic response have not been identified. The cellular bases of neurovascular interactions at the different levels of the cerebrovascular tree have not been established. Initial studies have highlighted the role of endothelial K_IR channel induced hyperpolarization at the capillary level and of neuronal NO in arterioles, but how these responses are generated, transmitted and appropriately timed requires further study. Furthermore, whether or not astrocytes contribute to the K⁺ surge at the capillary level and their potential contribution to the release of vasoactive factors, metabolic adaptation and retrograde vasodilatation at the arteriolar level has not been fully elucidated.

The role of capillaries in neurovascular coupling remains controversial. Are capillaries neural activity sensors or also effectors of the initial hemodynamic response? Are the small changes in capillary diameter observed in some studies physiologically relevant? Answering these questions would require a reexamination of what a capillary is, and a better characterization of the cells associated with them and their hemodynamic impact. Using vessel size and isolated cell markers as the sole identification tools for mural cells has generated considerable confusion. A different approach is needed based on: (a) establishing the ontogeny of mural cells, (b) defining the transcription and growth factors driving their development and localization, and (c) identifying their molecular signature. This effort would require a comprehensive reevaluation of the brain’s vascular development focusing specifically on mural cells, e.g., (Jung et al., 2017).

Relatively little is known about the upstream propagation of vasoactive signals from capillaries to pial arterioles. Are gap junctions and myoendothelial junctions the sole mode of inter-endothelial and endothelial-SMC transmission, or are diffusible factors also involved? Do mural cells, e.g., pericytes, play a role? The mechanisms focusing the hemodynamic response by selectively engaging arteriolar branches supplying the activated region, e.g., (Ngai et al., 1988), and the possible role of glial limitans astrocytes require further study. In systemic arteries perivascular sympathetic (noradrenergic) nerves control propagated responses (Segal, 2015), but it is unclear whether they play a similar role in cerebral arteries. In this regard, it is of interest that the noradrenergic innervation of the neocortex may play a role in focusing the vascular response (Bekar et al., 2012), a concept that requires further exploration.

Answering these questions will be facilitated by the use of new models and approaches to study the cerebral vasculature. For example, optical coherence tomography (Baran and R. K. Wang, 2016), 3-photon microscopy (Ouzounov et al., 2017), and ultrafast ultrasound localization microscopy (Errico et al., 2015), may offer an unprecedented look at the cerebral microcirculatory network at a depth, spatial extent and/or resolution previously unattainable. Whenever feasible, coupling these imaging tools with genetically-encoded activity sensors and cell specific reporters, as well as chemo- and opto-genetic approaches to activate specific neural pathways may provide novel insights into the spatiotemporal relationships linking neural activity and segmental vascular responses. Using conscious animals will eliminate anesthesia as a major confounder and will allow investigators to probe neurovascular coupling beyond the realm of olfactory, somatosensory or visual stimuli and towards complex behaviors in multiple brain regions (Gao et al., 2017). The introduction of more realistic multicellular in vitro models of the NVU, such as the 3-D microfluidics organ-on-chip technology for the BBB (Adriani et al., 2017) and cerebrovascular organoids derived from human induced pluripotent stem cells (IPS) (Appelt-Menzel et al., 2017), provides the opportunity to explore the molecular interactions among NVU cells beyond what can be achieved with conventional co-culture systems. Large-scale single-cell RNA sequencing applied to the cerebral vasculature (He et al., 2016; Tasic et al., 2016) will provide a glimpse into the molecular signature of specific cerebrovascular cells, and provide insight into the molecular bases of the functional differences of vascular and perivascular cells in the different segments of the cerebrovascular tree.

NVU dysfunction and neurodegeneration

Our understanding of the role of neurovascular dysfunction in neurodegenerative diseases is still rudimentary. Although there is overwhelming evidence that neurodegeneration is associated with structural and functional alterations of the NVU, often early in the disease course (Table 2), several questions remain to be addressed. Is the NVU dysfunction an epiphenomenon of the degenerative process or a pathogenic contributor? In most cases, the reductions in CBF are not severe enough to induce ischemic injury. However, it cannot be excluded that reduced cerebral perfusion, in concert with neurovascular uncoupling, may have deleterious effects on the brain in the long run. BBB dysfunction, neurotrophic failure and impaired clearance associated with NVU dysfunction may also contribute (Figure 11), but their pathogenic impact remains to the established. Which brings us to a related question. Does the neurovascular dysfunction promote the neurodegenerative process? In AD, experimental and clinical studies indicate that vascular dysfunction is an early manifestation of the disease and may promote AD pathology. Is this the case also in other neurodegenerative diseases? Considering that oxidative stress and inflammation, major causes of neurovascular dysfunction, have been implicated in virtually all neurodegenerative conditions (Ransohoff, 2016), these factors could be driving both the vascular and neurodegenerative pathologies independently, amplifying their respective pathogenic impact. In diseases like AD, in which the coexistence of neurodegeneration with cerebrovascular lesions is well documented, what is the role of the cerebrovascular pathology in the clinical expression of the disease? Vascular lesions could lower the threshold for cognitive deficits by increasing the overall lesion burden on the brain. Alternatively, the two processes could be synergistic, amplifying each other, as suggested by evidence that vascular dysfunction promotes AD pathology (see above). Although ischemic injury is not inconsequential to the clinical expression of AD (Kapasi et al., 2017), there is no definitive evidence of a synergistic interaction, e.g., (Vemuri et al., 2015). However, the reduction in dementia incidence attributed to better cardiovascular health (Pase et al., 2017) suggests a contribution of vascular pathology to neurodegeneration, an attractive hypothesis that requires further testing in randomized clinical trials.

The factors by which neurodegenerative pathology leads to neurovascular dysfunction remain to be defined. Are selected NVU cells and specific cerebrovascular segments particularly vulnerable, and, if so, why? Are there common signaling pathways and mediators responsible for the dysfunction in different diseases? What is the contribution of neuronal, glial and vascular cell metabolism to the vascular dysfunction? Difficulties in performing mechanistic studies in human patients, the long interval between onset of pathology and clinical diagnosis, lack of reliable biomarkers, and lack of animal models recapitulating in full disease phenotypes have been major impediments to progress. However, developments in both the clinical and experimental arenas may help overcome some of these barriers. Large-scale “omic” studies, advanced imaging with markers of disease, new molecular biomarkers, and IPS cells from patients, may provide mechanistic clues to be explored further in experimental models. In turn, new animal models incorporating neurodegenerative pathology with vascular co-morbidities and new cell-specific approaches to probe the NVU promise a dramatic expansion of knowledge that can then be verified in clinical studies. Therefore, there is a strong rationale for bolstering efforts to clarify the role of the NVU dysfunction in neurodegenerative diseases (Corriveau et al., 2016). New knowledge in this field may open new avenues in the prevention, early diagnosis and treatment of some of the most devastating illnesses affecting humankind.