Brain AVMs account for approximately 2% of all hemorrhagic strokes but cause almost half of all hemorrhagic strokes in children and young adults (Mahmoud et al., 2010; Nielsen et al., 2016; Solomon and Connolly, 2017). A hallmark of brain AVMs is the direct delivery of blood from artery to vein via an AV shunt, which forms in place of healthy capillary AV connections (Nielsen et al., 2016; Adhicary et al., 2022). In humans, brain AVMs can present as a tangle of enlarged, tortuous vessels—clinically termed a nidus (Figures 1A, B)—that may place contact pressure on surrounding brain tissue. The shunt effect driven by the AVM can cause hypoperfusion of surrounding brain tissue (steal effect). Additionally, because of the altered brain perfusion and increased pressure in both in the nidus and in AVM veins, brain AVMs have a risk of rupture that is around 1% per year. On average, the risk of spontaneous rupture and hemorrhage is low; however, there are factors that can increase this risk, such as older patient age, deep AVM location, and deep draining venous system (Stapf et al., 2006; Kim et al., 2014). Additionally, risk of hemorrhage is significantly increased in AVMs with previous hemorrhagic presentation (Gross and Du, 2013). By contrast, AVM size was not associated with increased rupture risk (Kim et al., 2014). Patients presenting with more than one risk factor have increased likelihood of rupture; patients with all three risk factors can have up to 34% chance of incurring a brain hemorrhage (Stapf et al., 2006).

Because of these alterations of the cerebral vasculature, brain AVMs can lead to seizures, headaches, stroke, hemorrhages, and neurological deficits (Nielsen et al., 2016; Solomon and Connolly, 2017; Scherschinski et al., 2022). Common methods to treat brain AVMs include embolization, radiation therapy, surgical resection of the unhealthy vasculature, or some combination thereof (Figure 2). However, these methods are not applicable to all brain AVM patients, as treatment options depend on AVM size, location, and on patient access to an AVM medical care team (Scherschinski et al., 2022). Given the heterogeneity of brain AVM presentation and angioarchitectural pathologies, there is controversy and longstanding debate over the best standard course of treatment for patients. For instance, combined treatment with embolization and radiosurgery can have differing results and success on patients with different AVM pathologies (Sackey et al., 2017). Repeated radiosurgery on AVMs can successfully resolve AVM pathologies but can leave the patient at higher risk for post-surgery complications (Kano et al., 2012). Amongst attempts to customize treatment plans for AVM patients, some clinical studies reported an increased rate of rupture in AVMs, thereby questioning the decision to pursue invasive treatments over medical management of symptoms (Luther et al., 2022). These findings have underscored the need for understanding patient and AVM heterogeneity and for developing and testing new treatments, such as targeted molecular or pharmacological therapies.

Research shows that brain AVMs affect vessel permeability and thus disrupt blood brain barrier (BBB) function (Hirunpattarasilp et al., 2019; Lendahl et al., 2019; Adhicary et al., 2023). Because brain capillaries (also known as microvessels) participate with other cell types as part of the neurovascular unit (NVU), it follows that brain AVMs lead to perturbed communication with other vascular cells (e.g., mural cells) and brain cells (e.g., astrocytes, neurons) (Winkler et al., 2022; Winkler et al., 2018; Chapman et al., 2022; Selhorst et al., 2022). Because NVU function is critical for brain homeostasis, including BBB integrity and neurovascular coupling, it is important to understand how NVU cells are affected during brain AVMs.

According to current statistics, less than 5% of brain AVMs have a known, heritable genetic lesion, while 95% of brain AVMs are currently considered to be sporadic—of those sporadic brain AVMs, a majority, but not all, appear to be associated with somatic mutations (Nikolaev et al., 2018; Adhicary et al., 2022). Genetic and RNA sequencing data from human brain AVMs have shown diversity in AVM phenotypes (Nikolaev et al., 2018; Huo et al., 2019; Hauer et al., 2020; Winkler et al., 2022). To understand the underlying diverse mechanisms of brain AVM formation and progression, many vertebrate models have been developed. Some models induce known genetic mutations identified in human patients, and other models target genes not yet known to cause brain AVM in humans. Pre-clinical mouse models are commonly used animal models that reliably reproduce features of brain AVM—AV shunting, nidus formation, hemorrhage, and abnormal blood flow—that are similar to those seen in humans (Figure 1C). Given these phenotypic similarities, research using mouse models has undoubtedly advanced our comprehensive understanding of brain AVMs and their genetic and cellular diversity.

Though CNS AVMs are more common in the brain, they can also occur in the spinal cord. In human patients, spinal cord AVMs account for approximately 25% of all vascular malformations identified in the spinal cord (Varshneya et al., 2019). Spinal cord vascular malformations are heterogenous, presenting with different anatomical pathologies and are classified into four subtypes, based on those vascular pathologies (Takai, 2017). Currently, spinal cord vascular malformations are classified as Type 1 (Spinal Dural Arteriovenous Fistula), Type 2 (Intramedullary Arteriovenous Malformation), Type 3 (Extradural-Intradural Arteriovenous Malformations), or Type 4 (Intradural Perimedullary Arteriovenous Fistula), depending on their location, symptoms, risk factors, and properties (Ferch et al., 2001; Patchana et al., 2020). Though classified as spinal vascular malformations, it is important to note that Type 1 and Type 4 are not true AVMs, but rather are vessel fistulas (Patchana et al., 2020). Type 2 AVMs are the most common type of spinal cord AVM, and they typically present as a nidus in the spinal cord (Ferch et al., 2001; Krings, 2010). Spinal cord AVMs may lead to hemorrhage and/or venous congestion (Krings, 2010) that can result in permanent disabilities (Hong et al., 2019). While spinal cord AVMs have been reported in mouse models (Milton et al., 2012), far less mechanistic insight has been gained, compared to brain AVMs. However the data from the few mechanistic studies that have been performed for spinal cord AVMs have shown that molecular expression of spinal cord AVMs is similar to retinal and brain AVMs (Vieira et al., 2020).

In animal models, AVMs are frequently observed in the retina. Retinal AVMs are generally reported as an enlarged AV connection together with other vascular defects. The retinas from animal models are widely used in AVM studies, due to their vascular anatomical simplicity (compared to brain tissue) and reproducibility. In mice, the retinal vasculature forms rapidly in the first weeks after birth (Mahmoud et al., 2010) resulting in a highly organized and hierarchical vascular network (Stahl et al., 2010; Crist et al., 2018). The simple and stereotyped artery-capillary bed-vein organization allows the presence of abnormal vasculature to be readily observed and imaged (Stahl et al., 2010). Further, because the retinal vasculature develops postnatally, this model can provide insight into AVMs that form during the early postnatal period (Stahl et al., 2010). Indeed, retinal vascular studies have been effectively translated to human and animal CNS AVMs.

HHT is an inherited disorder in which vascular abnormalities, including AVMs, occur in multiple parts of the body (Kim et al., 2018; Snodgrass et al., 2021). In HHT patients, AVMs can occur in multiple organs, including lung and brain (Thalgott et al., 2015; Kim et al., 2018; Snodgrass et al., 2021). Genetic studies revealed that most commonly, germline mutations in endoglin (ENG) or activin A receptor like Type 1 (ACVRL1) cause HHT1 or HHT2, respectively (Johnson et al., 1996; Mahmoud et al., 2010; Choi et al., 2014; Tual-Chalot et al., 2014; Thalgott et al., 2015; Baeyens et al., 2016; Ola et al., 2018). HHT animal models, which disrupt Endoglin and Alk1 (encoded by ACVRL1) signaling pathways, have been developed to study brain AVM onset and progression. These models have proven very useful for uncovering mechanisms of brain AVM pathogenesis and relating those findings to other brain AVM models.

Pericytes are mural cells that are found in most microvascular beds in the vascular system; however, the density and coverage of pericytes on microvessels varies among different tissue types (Armulik et al., 2011). In the CNS, the pericyte:endothelial cell ratio is estimated to be between 1:1 and 1:4, making the CNS microvasculature one of the most pericyte-enriched systems (Armulik et al., 2011; Smyth et al., 2018). Pericytes generally have a small cell body with multiple long, thin cytoplasmic processes that enwrap the underlying microvessel (Hill et al., 2015). One pericyte can extend its processes to many endothelial cells, permitting intercellular communication and influences with those endothelial cells (Gerhardt and Betsholtz, 2003).

Pericytes are critical cells within the CNS NVU. Here, at the level of the microvasculature, cells of the NVU collectively contribute to BBB formation and function. The BBB is inherently restrictive and selective, when determining what solutes are exchanged between the microvasculature and the CNS parenchyma. Of blood barriers in mammalian physiology (e.g., blood-cerebrospinal fluid, blood-testis, blood-retina barriers), the BBB is the most restrictive. Cell-cell junctions between neighboring endothelial cells provide barrier selectivity. However, endothelial cells are not the only NVU cells that contribute to BBB stringency. Specialized features of other NVU cells, like pericytes, play critical roles in development, regulation, and maintenance of tight junctions (Luissint et al., 2012), and thus promote BBB integrity. Pericytes and endothelial cells have a close cellular relationship, sharing a basement membrane and communicating with one another through multiple different signaling pathways (Adhicary et al., 2022). Pericytes communicate with endothelial cells through paracrine signaling, by diffusion of small molecules or, in some locations, via direct contact to allow for intercellular communication along a capillary segment (Adhicary et al., 2022). In fact, pericyte and endothelial cell communication is critical for proper formation and maintenance of the BBB. For example, pericytes can influence gene expression in endothelial cells that regulate vessel permeability (Freitas-Andrade et al., 2020).

Though the roles of pericytes can be generalized, it is important to note that pericytes are heterogenous in nature. CNS pericytes not only differ in their molecular and cellular profiles, when compared to other vascular beds (Su et al., 2021), but CNS pericytes also differ from one another and may be parsed into pericyte subtypes, based on morphological, molecular, and functional profiles (Brown et al., 2019). Within the CNS, different vessel segments have different types of pericytes with varying functions (Grant et al., 2019; Uemura et al., 2020). For example, pericytes that exist near high-flow vessel segments (such as arterioles) are broad cells that use their surface area to enwrap the vessel; which can have α-smooth muscle actin (α-SMA) fibers with greater contractile properties, perhaps to regulate blood flow and blood routing (Brown et al., 2019; Erdener et al., 2022; Hartmann et al., 2022). By contrast, thin-strand, mid-capillary pericytes extend longer, strand-like processes along the length of a vessel, spanning and contacting several endothelial cells. These pericytes can exhibit thin α-SMA fibers, perhaps to permit communication between cells and to promote blood brain barrier integrity, at the microvascular level (Erdener et al., 2022; Hartmann et al., 2022).

Contractile function of pericytes is a recent field of study, and we are beginning to learn how the role of contractile pericytes can influence vascular disease outcomes; for example, contraction of brain pericytes after stroke has been associated with capillary flow stalling and capillary dysfunction (Staehr et al., 2023).

Pericytes have also been reported to have stem-like cell properties that allow them to retain their potency and differentiate into other cell types, depending on the physiological situation (Özen et al., 2014; Nakagomi et al., 2015; Brown et al., 2019; Farahani et al., 2019). Though the mechanisms that trigger pericyte transdifferentiation are not clear, a few CNS AVM models suggest that pathological changes in pericytes—molecular and morphological changes—trigger pericytes to acquire SMC characteristics, such as contractile properties (Diéguez-Hurtado et al., 2019; Selhorst et al., 2022) (Table 1).

Pericytes can be affected either directly or indirectly during CNS AVMs. Genetic mutations in pericytes can lead to formation of AVMs, or mutations in other vascular cells (such as endothelial cells) can have indirect effects on pericytes, which may exacerbate AVM pathology (Table 1). One focus of this review is to describe direct and indirect changes to pericytes—including molecular, cellular, and functional changes—in the context of CNS AVMs. These changes are discussed in the text and are compiled in the Tables. In Table 1, we compiled a list of studies reporting changes to pericytes in mouse CNS AVM tissue. As our understanding of pericyte heterogeneity advances, so too will our understanding of how these cells are affected during CNS AVMs.

Within the CNS, perivascular SMCs enclose the endothelial tube of arteries, arterioles, some post-capillary venules, and veins, but are absent in capillaries (Uemura et al., 2020). Arteriolar SMCs are contractile and can regulate blood flow and pressure through vasoconstriction and vasodilation (Hill et al., 2015; Aldea et al., 2019). Contractility of arterial SMCs also maintains healthy vascular tone, in response to blood pressure changes and shear stress, and maintains arterial diameter within a normal physiological range (Aldea et al., 2019). In general, arteriolar SMCs are situated perpendicular to blood flow and vessel length and parallel to adjacent SMCs. SMC morphology differs, depending on the type of vessel on which they lie. Compared to arteriolar SMCs, venular SMCs have a nonuniform, stellate shape (Figure 3). Venous SMCs play important roles in maintaining venous blood volume depending on cardiac output by vasodilation and vasoconstriction (Downey and Heusch, 2001; Sykora et al., 2016). In Table 2, we have assembled a list of studies reporting changes to SMCs in mouse CNS AVM tissue.

Perivascular fibroblasts have recently come into focus in the brain vasculature as a cell type located in the arteries, between the vessel wall and astrocytic endfeet and more loosely attached to the vessels, as compared to pericytes or SMCs (Vanlandewijck et al., 2018). Perivascular fibroblast morphology is distinct from other perivascular cells, as they have a flat soma and non-overlapping processes (Vanlandewijck et al., 2018; Dorrier et al., 2022). Perivascular fibroblasts and fibromyocytes have recently been implicated in human CNS AVM samples (Winkler et al., 2022). While it was suggested that SMCs may differentiate into perivascular fibromyocytes, evidence for such cellular transdifferentiation has not yet been reported (Winkler et al., 2022).

The Platelet Derived Growth Factor (PDGF) family includes a series of multifunctional factors acting fundamentally in stromal cells. PDGF ligands signal through the cell-surface tyrosine kinase receptor PDGFR. The most relevant in vivo interaction of PDGFs with their receptors in the vascular system is that of PDGF-BB and PDGFRβ (Andrae et al., 2008). PDGF-BB is secreted by endothelial cells and signals through PDGFRβ in pericytes to regulate key functions in these cells, including recruitment, proliferation, and survival (Armulik et al., 2011). Consequently, mice lacking PDGF-BB or PDGFRβ exhibit a dramatic absence of pericytes. This paucity of pericytes results in increased capillary diameter, endothelial hyperplasia (Hellström et al., 2001), and microaneurysms (Lindahl et al., 1997), thus highlighting the importance of pericyte-endothelial cell interactions in the regulation of vascular diameter.

Interestingly, absence of pericytes and pericyte coverage has been reported in human samples of brain AVMs (Winkler et al., 2018) and in models of the disease (Chen et al., 2013; Tual-Chalot et al., 2014; Kofler et al., 2015; Baeyens et al., 2016; Crist et al., 2018; Ola et al., 2018; Nadeem et al., 2020; Orlich et al., 2022). With this evidence, it is tempting to speculate that loss of function mutations in the PDGF-BB/PDGFRβ axis might be associated with the development of AVMs. As a matter of fact, pathogenic variants of PDGFRB have been described associated with fusiform cerebral aneurysms (Karasozen et al., 2019; Chenbhanich et al., 2021; Parada et al., 2022) and in a small subset of patients with brain AVMs (Gao et al., 2022). Surprisingly, the pathogenic variants identified as associated with brain aneurysms result in gain of function of PDGFRβ (Wenger et al., 2020). Whether the already identified pathogenic variants in brain AVMs also lead to gain of function of the PDGFRβ, and how prevalent they are, remains to be established.

Notch signaling is a ubiquitous signaling pathway regulating critical functions in health and disease. Notch receptors (1–4) in the surface of the cell interact with the ligand (Delta-like 1, 3, 4 or Jagged 1, 3) in the neighboring cell. This interaction of Notch with its ligand creates a conformational change in the Notch receptor, exposing it for cleavage by ADAM metalloproteases and the γ-secretase complex and causing the release of the Notch intracellular domain (NICD). The NICD translocates to the nucleus where it binds to Recombination signal binding protein for immunoglobulin kappa J region (Rbpj) and a series of other co-activator factors to promote the transcription of its target genes (Siebel and Lendahl, 2017).

In the vasculature, endothelial cells express mostly Notch 1 and 4, while mural cells express Notch 1, 2, and 3. Of the ligands, Delta-like1 (Dll1), Dll4, and Jagged1 (Jag1) are the most prevalent in the endothelium, with Jag1, Dll4, and to some degree Dll1 and Jag2, also being found on mural cells (Hofmann and Iruela-Arispe, 2007; Baeten and Lilly, 2015; He et al., 2018). Consequently, Notch signaling in blood vessels has a wide range of functions mediated by ligand-receptor crosstalk between endothelial and mural cells (Gridley, 2010; Del Gaudio et al., 2022; Hasan and Fischer, 2023).

Notch signaling in the vasculature has been widely associated with brain AVMs in human patients and mouse models of the disease (Drapé et al., 2022). The first insights for a role of Notch signaling in AVM formation come from studies in embryos with mutations in different molecules of the Notch signaling pathway. Injection of India ink in mutants haploinsufficient for the Notch ligand Delta-like 4 (Dll4) or with deficiency of endothelial Rbpj revealed enlarged AV shunts. A few years later, it was reported in postnatal mice that Notch gain of function in endothelial cells gave rise to AVM formation, and together with further evidence from mutant embryos (Krebs et al., 2010), it became apparent that both gain and loss of function of endothelial Notch signaling can lead to AVM formation in mouse models. However, these early studies did not document whether AVMs were arising in the brain. Additional work using mouse models to activate (Murphy et al., 2008; Murphy et al., 2012; Murphy et al., 2014; Nielsen et al., 2023) or suppress Notch signaling (Nielsen et al., 2014; Selhorst et al., 2022; Adhicary et al., 2023) in endothelial cells was essential to pinpoint the role of Notch in brain AVMs.

Because of the strong requirement for Notch signaling in endothelial cells to prevent the development of brain AVMs, changes in the mural cell compartment were frequently overseen or not documented in the different mutants described above. In another study of Rbpj deletion from endothelial cells, numerous pericyte abnormalities were seen in brain and retina AVMs, including abnormal morphology and increased pericyte area with unchanged pericyte coverage of vessels (Selhorst et al., 2022). In this AVM model, pericytes expressed reduced levels of Pdgfrb, Cdh2, and Cd146 (Selhorst et al., 2022), suggesting that, in addition to preventing AVMs, endothelial Rbpj controls the expression of factors promoting pericyte recruitment and association with endothelial cells. Similar, yet somewhat contradictory, results were obtained by Li et al., (2011), who reported that loss of endothelial Rbpj using a different Cre-driver also resulted in reduced Cdh2 expression. However, while this study documented loss of pericytes, it did not document the formation of AVMs.

The function of Notch in CNS AVM formation was also evaluated using mice with germline deletion of one allele of Notch1, together with deletion of one or both alleles of Notch3 (Kofler et al., 2015). Abnormal, hyperdense vasculature was observed in Notch1 ^+/− and Notch3^−/− mice; however, AVMs were only observed in the double mutant mice Notch1 ^+/− ; Notch3^−/− mice. It was not clear from these studies whether Notch1 and Notch3 function in the mural cell compartment, in the endothelium, or in some combination thereof, was responsible for preventing AVM development. Subsequent studies, which impaired Notch signaling specifically in mural cells, demonstrated a role for Notch signaling in pericytes to prevent the development of retinal AVMs (Nadeem et al., 2020). Mechanistically, the Notch pathway prevents AVM onset by regulating the levels of Matrix Metalloprotease 14 (Kofler et al., 2015) and PDGFRβ (Kofler et al., 2015; Nadeem et al., 2020) in pericytes. Intriguingly, the work of Diéguez-Hurtado et al., (2019), using mouse models with mural cell deficiency of Rbpj, describes a Notch independent role for Rbpj in pericytes where vascular AV shunts are detected. However, these lesions lack the characteristic fast-flow associated with AVMs, and the overall lesions in this mouse model are described to resemble cerebral cavernous malformations (CCMs).

In addition to findings from animal models, dysregulated NOTCH signaling in mural cells has also been observed in brain AVM samples from human patients (ZhuGe et al., 2009; ZhuGe et al., 2013; Hill-Felberg et al., 2015). Upregulation of NOTCH4 and NOTCH1 was detected in both endothelial cells and SMCs in human brain AVM samples (ZhuGe et al., 2013; ZhuGe et al., 2009). Similarly enhanced levels of NOTCH4 were also reported by Hill-Felberg et al., (2015); however, these authors highlighted that upregulation of NOTCH3, expressed by endothelial cells and mural cells, was substantially more robust than that of NOTCH4 and that levels of NOTCH1 were decreased in brain AVMs samples.

It is intriguing that loss of Notch signaling in mural cells results in AVM formation in animal models, yet Notch signaling is mostly found to be upregulated in the mural cell compartment of human brain AVM samples. Further studies are needed to carefully and thoroughly delineate how Notch in pericytes and SMCs impacts CNS AVM formation and/or progression.

Bone Morphogenetic Proteins (BMPs) are soluble factors which belong to the Transforming Growth Factor-β (TGFβ) superfamily (Wrana, 2013). Several BMP members play important roles in the vasculature (Kulikauskas et al., 2022). BMP9 and BMP10, in particular, form a distinct subgroup within the TGFβ superfamily. These factors are released into the circulation from the liver and the heart, respectively, and signal through the activin receptor-like kinase 1 (Alk1)/BMPRII complex (David et al., 2007). This signal can be regulated by the co-receptor Endoglin, which interacts with the complex to promote phosphorylation of Smad1/5/8, resulting in their release, association with Smad4, and further translocation to the nucleus to regulate the expression of their target genes (Ross and Hill, 2008).

Mutations in the components of this pathway are associated with HHT and the development of AVMs. As such, mutations in the genes encoding for ENDOGLIN (ENG) and ALK1 (ACVRL1) result in the pathogenesis of HHT in 90% of the patients (McAllister et al., 1994; Johnson et al., 1996). Less common mutations in SMAD4 lead to a particular type of HHT, known as juvenile polyposis (JT)-HHT (Gallione et al., 2004).

The familial nature of HHT and identification of causal genes has facilitated the generation of several mouse models recapitulating the disease (Arthur and Roman, 2022). Interestingly, endothelial deletion of Alk1 or Smad4 leads to the formation of AVMs in the retina (Tual-Chalot et al., 2014; Baeyens et al., 2016; Crist et al., 2018; Ola et al., 2018) or the brain (Chen et al., 2013), which display decreased pericyte coverage, indicating a likely contribution of this cell type toward preventing the onset or progression of the lesions. Supporting this idea, treatment of Eng-deficient mice with Thalidomide ameliorated retinal AVM severity, in part, by increasing mural cell coverage (Lebrin et al., 2010).

Several studies have reported crosstalk between the Notch signaling pathway and the BMP/ALK/Smad pathway. During angiogenesis, ALK1 and Notch signaling pathways synergize to promote the expression of Notch target genes and repression of endothelial sprouting (Larrivée et al., 2012; Moya et al., 2012; Ricard et al., 2012; Kerr et al., 2015; Rostama et al., 2015). In mouse models of AVMs, in which BMP/ALK1/Smad signaling is impaired, Notch signaling was also downregulated (Yao et al., 2011; 2013; Tual-Chalot et al., 2014; Hwan Kim et al., 2020). Interestingly, Alk1 deficient HHT zebrafish models were not influenced by changes in the Notch signaling pathway, indicating that Notch signaling may not play a critical role in HHT associated AVMs (Rochon et al., 2015).

Most of the evidence gathered from mouse models of HHT has led to the general consensus that vessel abnormalities arise from loss of gene function in endothelial cells (Arthur and Roman, 2022). However, a role for the BMP/ALK1 axis in mural cell function has been established (Wang et al., 2021). More importantly, deletion of Eng/Alk1 in mural cells, using the SM22α-Cre driver to target mural cells, leads to the development of AVMs in the brain of mice (Milton et al., 2012; Choi et al., 2014; Han et al., 2022). However, studies have suggested that the SM22α-Cre driver targets a small population of endothelial cells, which might be responsible for the observed phenotypes. Moreover, alternative mouse models where Eng/Alk1 deletion was induced using other mural cell drivers (Ng2-CreER or Myh11-CreER^T2) did not result in AVMs (Chen et al., 2014; Garrido-Martin et al., 2014). Additional evidence is needed to establish a clear role for the BMP/ALK/SMAD axis in mural cells during the onset or progression of CNS AVMs.

The majority of CNS AVMs are sporadic and not linked to an inherited gene mutation. Recently, activating mutations in KRAS were uncovered in a high percentage of sporadic brain AVMs (Nikolaev et al., 2018). Thus far, the pathogenic variants of KRAS leading to brain AVMs seem to be restricted to endothelial cells and are not detected in the mural cell compartment (Nikolaev et al., 2018). The absence of pathogenic variants in mural cells is consistent with the data obtained using mouse models, in which expression of the pathogenic variant KrasG12D in endothelial cells is sufficient to recapitulate the phenotype observed in human lesions (Fish et al., 2020).

Further insight on the consequences of KRAS pathogenic variants in endothelial cells, and their interaction with pericytes, comes for co-culture experiments where Human Umbilical Endothelial Cells (HUVECs) expressing the pathogenic variant KRASG12V were cultured together with Human Brain Vascular Pericytes (HBVPs). KRASG12V expressing HUVECs formed abnormal vascular tubes with reduced pericyte coverage and reduced pericyte derived basement membrane, compared to control HUVECs (Sun et al., 2022). Among the most common factors that are produced by endothelial cells, yet control pericyte functions, PDGF-D and TGFβ2 expression was reduced in KRASG12V expressing HUVECs. PDGF-DD, similar to PDGF-BB, can drive pericyte migration and proliferation (Kemp et al., 2020). Future experiments will determine whether levels of pericyte coverage/numbers are also diminished in brain AVMs arising from the KRASG12V pathogenic variant in samples from human patients, whether levels of PDGF-DD are altered, and whether pericytes play an active role in the pathogenesis of these lesions.

As we described here, CNS AVMs are often characterized by loss or gain of mural cells or mural cell coverage (in many cases, data are specific for pericytes) associated with abnormal vessels. Where do pericytes go, or where do pericytes come from? Increased apoptosis accounts for reduced pericyte number and coverage following pericyte deletion of Rbpj (Nadeem et al., 2020). Transdifferentiation of pericytes, which thus lose their pericyte-specific molecular identity, may also account for pericyte loss. In support of this possibility, recent studies suggest reciprocal transdifferentiation between pericytes and mesenchymal stem cells (Tian et al., 2017; Yamazaki and Mukouyama, 2018). Further, evidence for transition of endothelial cells into mesenchymal cells (EndoMT) has been shown in human AVM tissue, suggesting that endothelial cells may be a course for mural cells via a two-step transformation (Shoemaker et al., 2020). Consistent with this reasoning, the previously mentioned fibromyocytes—which represent a potentially new mural cell cohort—show low expression of genes encoding contractile proteins, thus giving these cells an SMC-like identity (Winkler et al., 2022). Conversely, fibromyocytes do not express myocardin, a master regulator SMC molecular identity (Wang et al., 2003), thus distinguishing fibromyocytes from SMCs. Cell lineage tracing experiments will help researchers understand whether fibromyocytes indeed derive from a SMC lineage.

Evidence from human genomics and mouse genetics studies primarily shows that gene mutations (either germline or somatic gene mutations) in endothelial cells are the root cause of CNS AVM pathogenesis. However, data also suggest that other cells, including mural cells, perivascular astrocytes, and inflammatory cells (monocytes), are associated with AVM pathogenesis (Raabe et al., 2012; Zhang et al., 2016; Nadeem et al., 2020; Thomas et al., 2021; Orlich et al., 2022). For example, HHT-related AVMs are predicted to develop via a “two-hit” pathogenetic insult. Following this “two-hit” hypothesis, underlying genetic lesions are thought to contribute to the first pathological hit, while a later second hit to vessel stability places the vessel at risk for AVM formation. Perturbations to mural cells, either via an angiogenic insult or a direct cellular insult, may contribute to the second hit.

AVMs are characterized, in part, by abnormally high blood flow through AV shunts. The altered hemodynamics are predicted to affect endothelial cells directly, in response to increased blood flow and shear stress (Carlson et al., 2005; Murphy et al., 2014; 2012; 2008; Zhang et al., 2020). In turn, mural cell coverage of microvessels may increase, perhaps in response to a mechanical cue (vessel wall stretch) that promotes mural cell differentiation and maturation (Van Gieson et al., 2003). Indirect effects on mural cells, such as impaired recruitment to abnormal AVM blood vessels, have also been reported in an endothelial Alk1-deficient (Baeyens et al., 2016) mouse model of HHT. In two retinal HHT models [endothelial deletion of Eng (Mahmoud et al., 2010) or endothelial deletion of Smad4 (Ola et al., 2018)], increased blood flow through AVM reprograms venous endothelial cells to acquire an arterialized identity and mural cells to upregulate α-SMA expression, resulting in impaired SMC organization along retinal vessels. Following mural cell deletion of Srf in mice, altered blood flow affects pericyte migration during vessel patterning, impairing angiogenic vessel integrity and tone (Orlich et al., 2022). In human patients, brain AVMs can be characterized as having high or low blood flow rates. Expression studies showed that SMCs from low flow rate brain AVMs had increased expression of WNT signaling mediators FZD10 and MYOC, compared to high flow rate SMCs (Huo et al., 2019), suggesting that activated WNT signaling may be involved in AVM formation. Collectively, these findings indicate that mural cells are indeed affected by the significant hemodynamic changes that accompany AVM pathologies.