When, where and which PIK3CA mutations are pathogenic in congenital disorders

PIK3CA encodes the class I PI3Kα isoform and is frequently mutated in cancer. Activating mutations in PIK3CA also cause a range of congenital disorders featuring asymmetric tissue overgrowth, known as the PIK3CA-related overgrowth spectrum (PROS), with frequent vascular involvement. In PROS, PIK3CA mutations arise postzygotically, during embryonic development, leading to a mosaic body pattern distribution resulting in a variety of phenotypic features. A clear skewed pattern of overgrowth favoring some mesoderm-derived and ectoderm-derived tissues is observed but not understood. Here, we summarize our current knowledge of the determinants of PIK3CA-related pathogenesis in PROS, including intrinsic factors such as cell lineage susceptibility and PIK3CA variant bias, and extrinsic factors, which refers to environmental modifiers. We also include a section on PIK3CA-related vascular malformations given that the vasculature is frequently affected in PROS. Increasing our biological understanding of PIK3CA mutations in PROS will contribute toward unraveling the onset and progression of these conditions and ultimately impact on their treatment. Given that PIK3CA mutations are similar in PROS and cancer, deeper insights into one will also inform about the other. In this Review, the authors provide an overview of the pathogenic effects of somatic activating PIK3CA mutations in congenital disorders and discuss how the interplay between genetics, cell identity and the environment explains the onset, progression and severity of these disorders with a special focus on the vasculature.

P IK3CA encodes p110α, one of the four class I phosphatidylinositol 3-kinase (PI3K) catalytic subunits. p110α is an obligate heterodimer with a p85-type regulatory subunit, with no evidence of the existence of p85-free p110α 1 . For simplicity, from here on we will use PI3K when referring to the p110 isoforms. PI3Kα is ubiquitously expressed and is activated by tyrosine kinases. PIK3CA is the most frequently mutated oncogene across all human cancers with a high prevalence in breast and endometrial cancers 2 . PIK3CA mutations are acquired in a somatic fashion and are largely present in heterozygosity; nevertheless, there is evidence of double PIK3CA mutations in cis, which further increase its PI3K activity 3 .
Our Review stems from the remarkable discovery of oncogenic mutations in PIK3CA being causative of sporadic mosaic congenital disorders characterized by tissue overgrowth, with the vascular compartment being most frequently affected. These conditions have become widely known as PROS and they can range from isolated to complex and syndromic phenotypes where several tissues are affected (Fig. 1). PROS disorders are considered monogenic diseases, albeit emerging evidence indicates that co-occurrence of several genetic events, at least in the vasculature, is more frequent than previously anticipated [4][5][6] . This Review focuses on the pathogenic effects of somatic activating PIK3CA mutations acquired at different developmental stages. We discuss how the interplay between genetics, cell identity and the environment explains the onset, progression and severity of these disorders. Also, we provide an overview of the impact of distinct PIK3CA variants on these congenital conditions. Finally, we include a dedicated section on vascular malformations, given that the vascular compartment appears most affected in PROS. For congenital disorders caused by other PI3K signaling components, we refer the reader to Box 1.

Class I PI3Ks
PI3Ks are lipid kinases that catalyze the phosphorylation of the 3-hydroxyl group of the inositol ring of different phosphatidylinositol (PtdIns) lipid substrates present at the cellular membranes. In vertebrates, PI3Ks are divided into three classes (I, II and III) on the basis of structure, substrate preference, distribution, mechanism of activation and function 1,7 .
Class I PI3Ks are heterodimers composed of a catalytic and a regulatory subunit. The p110 subunit (here referred to as PI3K) confers the lipid kinase activity, while the regulatory subunit modulates the activity, stability and subcellular localization of the complex. Class I PI3Ks are subdivided into class IA and class IB depending on their ability to bind to different regulatory subunits 1,7 . The class IA catalytic subunits PI3Kα, PI3Kβ and PI3Kδ (encoded by PIK3CA, PIK3CB and PIK3CD, respectively) interact with one of five p85-type regulatory subunits: p85α (or its splice variants p55α and p50α), p85β and p55γ (encoded by PIK3R1, PIK3R2 and PIK3R3 respectively). p85 stabilizes the catalytic subunit but it inhibits its PI3K kinase activity in the basal state through its Src homology 2 (SH2) domains. Following stimulation, p85 allows PI3K activation by promoting recruitment to pTyr residues in receptor tyrosine kinases (RTKs) and adaptor molecules 8 . Both p85-mediated inhibition and recruitment to pTyr residues occur via the same SH2 domains in p85 (refs. 1,8 ). Class IB is solely composed of the PI3Kγ catalytic subunit (encoded by PIK3CG), which may interact with one of two regulatory proteins, p101 or p84/p87 (encoded by PIK3R5 or PIK3R6, respectively). The catalytic subunits of class I PI3Ks show specific expression patterns, with PI3Kα and PI3Kβ being ubiquitously expressed and PI3Kδ and PI3Kγ being enriched in some cell lineages such as immune cells, neurons and cardiomyocytes 1,7 .
Class I PI3Ks are activated by extracellular signals at the plasma membrane where they generate phospholipids that trigger a cascade of signaling events (Box 2). PI3Kα and PI3Kδ are recruited to the plasma membrane via binding of the SH2 domains of p85 to pTyr proteins. By contrast, the PI3Kγ heterodimer is activated by the Gβγ subunits released by activated G-protein-coupled When, where and which PIK3CA mutations are pathogenic in congenital disorders Ana Angulo-Urarte 1 ✉ and Mariona Graupera 1,2 ✉ PIK3CA encodes the class I PI3Kα isoform and is frequently mutated in cancer. Activating mutations in PIK3CA also cause a range of congenital disorders featuring asymmetric tissue overgrowth, known as the PIK3CA-related overgrowth spectrum (PROS), with frequent vascular involvement. In PROS, PIK3CA mutations arise postzygotically, during embryonic development, leading to a mosaic body pattern distribution resulting in a variety of phenotypic features. A clear skewed pattern of overgrowth favoring some mesoderm-derived and ectoderm-derived tissues is observed but not understood. Here, we summarize our current knowledge of the determinants of PIK3CA-related pathogenesis in PROS, including intrinsic factors such as cell lineage susceptibility and PIK3CA variant bias, and extrinsic factors, which refers to environmental modifiers. We also include a section on PIK3CA-related vascular malformations given that the vasculature is frequently affected in PROS. Increasing our biological understanding of PIK3CA mutations in PROS will contribute toward unraveling the onset and progression of these conditions and ultimately impact on their treatment. Given that PIK3CA mutations are similar in PROS and cancer, deeper insights into one will also inform about the other.
receptors (GPCRs) 1,7 . PI3Kβ is unique in that multiple active membrane receptors, including both RTKs and GPCRs, may potentially recruit and activate it [9][10][11] . All class I PI3K catalytic isoforms contain a RAS-binding domain (RBD), which allows them to interact with membrane-bound small GTPases and provide an extra activation input. Via their RBDs, PI3Kα, PI3Kδ and PI3Kγ interact with RAS, whereas PI3Kβ interacts with RAC1 and CDC42.

oncogenic PIK3CA mutations beyond cancer
In 2012, activating PIK3CA mutations were linked to mosaic, congenital and progressive overgrowth disorders for which the name PROS was coined [12][13][14][15] . PROS features an anatomically variable admixture of overgrown tissues, with the vasculature and adipose tissue being most severely affected macroscopically 16 . Figure 1 includes the list of all described disorders now grouped under the umbrella of PROS. Somatic activating PIK3CA mutations were later discovered as a cause of congenital sporadic venous malformations (VMs) and lymphatic malformations (LMs) [17][18][19][20] . Since then, the list of different subtypes of new PIK3CA-related disorders has continued growing. Strictly speaking, all these conditions involve tissue overgrowth (beyond other phenotypes), thereby suggesting that all should be grouped under the umbrella of PROS. Nevertheless, we favor the subclassification proposed by Mirzaa and colleagues, which distinguishes between isolated or syndromic PROS 21 . The former includes any clinical manifestations that occur as a focal lesion affecting only one tissue or body part, whereas syndromic PROS includes conditions in which tissue overgrowth is not focal, affects several tissues and presents other features (Fig. 1). The terms isolated and syndromic PROS will be used across this Review. Of note, Martinez-Glez et al. recently described PIK3CA mutations in individuals presenting segmental undergrowth of musculoskeletal tissues together with vascular malformations, with or without associated overgrowth 22 . There are some other individuals who are clinically diagnosed with some of the so-called PROS conditions (that is, Klippel-Trenaunay syndrome (KTS), diffuse capillary malformation with overgrowth (DCMO) and megalencephalycapillary malformation (MCAP) syndrome) but are not associated with mutations in PIK3CA 23,24 . These observations indicate that the understanding of PIK3CA-related congenital disorders is in its infancy and that the current classification may need to be revisited in the future.
Genetic patterns and severity of PROS. Overgrowth in PROS is characteristically present at birth and may progress during childhood and adulthood 16 . Activating PIK3CA mutations in PROS arise postzygotically and stochastically during embryonic development, leading to a mosaic distribution where only a subset of cells carries the mutation resulting in a variety of phenotypic features. Of note, an overgrown lesion may be composed of multiple cell types, which, potentially, may not all carry the PIK3CA mutation. The PIK3CA mutations that cause PROS are the very same ones that cause cancer, including the so-called cancer hotspots (p.Glu542Lys, p.Glu545Lys and p.His1047Arg). However, mutations in gene locations other than the common hotspots are more frequent in PROS than in cancer 23 . Given that PIK3CA is a gene widely associated with cancer, an intriguing aspect is whether individuals with PROS have an increased risk of developing cancer. However, current data seem to indicate that there is no tumor proneness in this context (Box 3).
Germ line versus mosaicism. Within PROS, most activating PIK3CA mutations have been detected in a mosaic fashion, with a very low allelic frequency in the affected tissues and absent in blood cells. PIK3CA activating mutations in the human zygote likely cause early embryonic death 25 . Indeed, expression of the Pik3ca p.His1047Arg variant in the germ line leads to embryonic lethality in mice 26,27 . This behavior is not unique to PIK3CA mutations, because many oncogenes have a dominant lethal activity that can only survive via mosaicism 28,29 . Germline PIK3CA mutations have been reported in 13 cases of PROS with macrocephaly. Ten of these carried missense mutations of uncertain significance, and none were identified in the cancer hotspot sites 12,[30][31][32] . There was a recent case of a child presenting a mild PROS phenotype and carrying a PIK3CA p.Gly364Arg germline mutation. This variant is annotated as a functional activating mutation in cancer 2 and was detected at a Defining clinical severity of individuals with PROS. Severity varies greatly between individuals with PROS, ranging from milder symptoms to painful and life-threatening severe forms 12,14,20,33,34 . Without exceeding the PI3Kα activity threshold that is compatible with life, it is expected that postzygotic activating PIK3CA mutations that arise at early developmental stages affect a large number of cell lineages, thus leading to a more widespread, pleiotropic and severe condition. On the other hand, if PIK3CA mutations were acquired later in development or after birth, this would result in lineage-specific pathogenesis, such as that found in isolated vascular malformations. These interpretations have led to the assumption that the latter is a less clinically severe form of PROS. The implementation of nextgeneration sequencing into clinical practice for PROS diagnosis has allowed us to study and follow large cohorts of individuals. Indeed, these genetic data have provided substantial evidence that there is not always a clear correlation between the type of cell lineages carrying the mutation, the VAF of the mutation in the affected tissue, and the severity of the clinical outcome 23 . In fact, some individuals develop an isolated, but very severe lesion and other individuals have widespread overgrowth but lower severity. These data suggest that severity primarily relies on the specific anatomic location and extension of the overgrown tissue rather than the degree of mosaicism. They also indicate that the severity and number of tissues carrying a PIK3CA mutation are not synonymous in PROS; in addition, it poses the notion that it is inaccurate to assume that the earlier a mutation appears, the more severe the pathogenic outcome would be.

Determinants of clinical phenotypes in PROS.
The variability in clinical phenotypes, symptoms and severity of individuals with PROS has posed difficulties in understanding what defines these outcomes. We propose that the intrinsic (cell-autonomous) and extrinsic factors that mutated clones are exposed to are key determinants of PROS clinical manifestations and severity; this includes the cell lineage that acquired the mutation (for example, mesoderm versus endoderm precursor, progenitor cell versus differentiated cell), the degree and mechanism of PI3Kα activation (PIK3CA variant) and the spatiotemporal environmental modifiers of PI3Kα signaling (availability of growth factors, paracrine activation of wild-type cells surrounding mutant cells, cell-cell and cell-extracellular matrix (ECM) interactions and mechanical signals, among others). We propose that the final phenotypic outcome of PROS would be the consequence of a unique combination of all these parameters (Fig. 2).

Box 1 | PIK3Copathies beyond PIK3CA
Mutations in several components of the PI3K pathway, other than PIK3CA, have been identified in congenital disorders characterized by aberrant activation of PI3K signaling and tissue overgrowth 125 . These mutations include loss of PTEN 126 , TSC1 and TSC2 (ref. 127 ), and germline or somatic activating mutations in AKT1 (ref. 128 ), AKT2 (ref. 129 ), AKT3 (ref. 12,130 ), PIK3R1 (ref. 131 ), PIK3R2 (ref. 12 ) and the mammalian target of rapamycin (MTOR) gene 45,132,133 . Heterozygous loss of PTEN and TSC1/TSC2 and activating mutations in AKT3, PIK3R2 and MTOR genes might occur in the germ line or somatically. The germline (heterozygous) loss of PTEN is known as PTEN hamartoma tumor syndrome (PHTS) and includes Bannayan-Riley-Ruvalcaba syndrome, Cowden syndrome, PTEN-related Proteus syndrome, and PTEN-related Proteus-like syndrome 126 . Although individuals with PHTS are prone to developing benign and malignant tumors, evidence has emerged that 50% of them also develop vascular malformations.
Autosomal dominant loss of TSC1/TSC2 leads to tuberous sclerosis complex, a neurocutaneous disorder frequently associated with abnormalities in the brain 127 . Somatic activating mutations in AKT1 lead to the so-called Proteus syndrome (PS) 128 and in PIK3R1 they have been associated with vascular malformations and overgrowth similar to those in PROS/PS 12,130,131 . Mutations in AKT3 (highly expressed in brain and heart) and in PIK3R2 have been associated with brain overgrowth. Only very few cases with AKT2 mutations have been found and they were associated with severe fasting hypoglycemia and asymmetrical growth 129 . Although there are overlapping characteristics between these genetic conditions, the specific expression and regulation of each of the proteins define the ultimate clinical outcome. The correct genetic diagnosis is important to find suitable treatments, as well as to better understand gene-specific functions during development. It is important to bear in mind that overactivation of the PI3K signaling pathway might not only be promoted by the PI3Kα isoform.
However, the exact cell type or types that carry the mutation in each case is not always clear. Most genetic testing has been performed on biopsy samples from affected tissues, in which many different cell types are found, including non-mutated cells. In addition, the VAF within these biopsy samples ranges from 0.5% to 50% indicating, for instance, that in cases where a 1% VAF is detected, only 2 of 100 cells carry the mutation (VAF being a measure of diploid zygosity). However, there is an intrinsic variability based on sample handling; often individuals with PROS need to have biopsy samples taken several times before the presence of a PIK3CA mutation is detected 35 . Cell-type-specific isolation and in vitro culture of patientderived cells have clarified that PIK3CA mutations are present in keratinocytes 36 , blood endothelial cells (BECs), lymphatic endothelial cells (LECs) 35,37,38 , fibroblasts 13,39-41 , adipose-derived stem cells and adipocytes 42,43 . Most tissues carrying a PIK3CA mutation are mesodermal derivates (vasculature, adipose, muscle and bone) and/or ectodermal derivates (nervous tissue, epidermis and connective tissues of the head) [44][45][46] . Within the nervous tissue, it is not clear which specific cell lineages carry PIK3CA mutations as biopsy samples do not discriminate between neurons, macroglia and microglia (neurons and macroglia are of neuroectodermal origin and microglia are derived from the mesoderm line) 43,44,47,48 . Also, it is incorrect to consider that all PROS-related neuropathies involve overgrowth of neuroectodermal derivates. For example, lipomatosis of nerve is a subtype of isolated PROS where individuals have nerve bundle enlargement primarily caused by overgrown adipose and fibrous tissue 46 . Endodermal-derived tissues (for example, epithelial lining of the gastrointestinal and respiratory tract and parenchyma of tonsils, liver, thymus, thyroid, parathyroid and pancreas) are not usually

Box 3 | tumor proneness in individuals with Pros
Given that the PIK3CA mutations that cause cancer are the very same as those found in PROS, an intriguing aspect is whether individuals with PROS exhibit high risk for developing cancer 16 . To our knowledge, PROS has only been associated with malignant renal disease in children, although tumor risk in that context is between 1% and 2% 139,140 . These data suggest that there is no clear evidence that pediatric patients with PROS have an intrinsic predisposition to develop cancer. In addition, no data on adults with PROS and cancer proneness are available. While the field requires a proper meta-analysis to determine the true risk of tumor appearance, there are important reflections to consider. First, low activating mutations are more common in PROS than in cancer: should individuals with common and non-common cancer hotspots be equally considered for tumor risk assessment? Second, why do PIK3CA mutations induce cell transformation in some linages and not in others (that is, endothelial cells (ECs) versus epithelial cells)? These mechanisms are quite lineage specific, thereby calling for a biological understanding of which unique factors and traits trigger PIK3CA-related transformation. We believe that unraveling which molecular mechanisms avoid transformation may offer an exciting therapeutic window to improve cancer therapy. Third, it is important to acknowledge that individuals with PROS are often managed by a fairly diverse team of physicians. While this clinical practice is not wrong, it has prevented the keeping of a thorough retrospective clinical record, thus the information currently available regarding these individuals is poor. Finally, pediatric and adult patients with PROS may have to be considered independently for tumor risk. Indeed, individuals with PHTS exhibit high risk of developing a tumor at the age of 35 years and older 141 . Together, an essential question that needs to be addressed is why the timing (development versus adulthood) and the cell lineage of PIK3CA mutation acquisition are so critical in determining tumorigenesis. We anticipate that current worldwide efforts to understand PROS will help us solve these conundrums and will also help to define whether tumor surveillance is warranted for these individuals.

Box 2 | Canonical class I PI3K signaling
Activated class I PI3Ks phosphorylate phosphatidylinositol 4,5-bisphosphate (PtdIns(4,5)P 2 ) at the plasma membrane and generate the second messenger PtdIns(3,4,5)P 3 , also known as PIP 3 . A transient rise in PIP 3 levels engages a signaling cascade that is required for regulation of a broad range of cellular functions, including growth, proliferation, metabolism, migration and survival 7 . How and when PIP 3 favors one or another cellular function is not well understood, but it quite likely involves several elements: (1) the activation of PI3K by different extracellular inputs; (2) the intensity and duration of PI3K activation; (3) the specific localization and amount of PIP 3 produced; (4) the activity, location and phosphate selectivity of the phosphatases that hydrolyze phosphoinositides; all ultimately leading to the activation of distinct downstream effectors 134 .
PIP 3 serves as a ligand and functional regulator of a group of proteins that contain a pleckstrin homology (PH) domain, such as phosphoinositide-dependent kinase 1 (PDK1) and serinethreonine kinase AKT (also known as PKB) 135 . Other PI3K effectors with PH domains are tyrosine kinases (for example, BTK in B lymphocytes), several guanine nucleotide exchange factors and GTPase-activating proteins that regulate small GTPases of the RAC, RAS, RHO and ARF families (for example, GRP1 and ARAP3) and protein adaptors (for example, GAB1, GAB2, TAPP1 or DAPP) 1 . Their recruitment and activation are isoform selective and cell-type-dependent compared with a more universal activation of AKT. PTEN and Src homology 2 domain containing inositol polyphosphate 5-phosphatase 1 and 2 (SHIP1 and SHIP2) counterbalance the transient increase in PIP 3 . PTEN reverts PIP 3 back to PI(4,5)P 2 , while SHIP dephosphorylates PIP 3 into PI(3,4)P 2 , which is then further dephosphorylated by INPP4B phosphatase 136 .
AKT is the most widely studied effector of PI3K and comprises three isoforms (AKT1, AKT2 and AKT3) that have different patterns of expression and localization 135 . AKT is recruited to the plasma membrane via its PH domain. There, AKT is phosphorylated at Thr308 by PDK1, which is also recruited to the membrane via its PH domain. Nevertheless, full AKT activation requires additional phosphorylation at Ser473 by mTOR complex 2 (mTORC2). Activated AKT can exert its function in the cytoplasm and nucleus, where it phosphorylates and consequently activates or inhibits different downstream substrates. More than 100 nonredundant substrates of AKT have been identified 135 . Among them, we highlight tuberous sclerosis complex 2 (TSC2) and forkhead box protein O1 (FOXO1) for their importance in cancer biology. Activated AKT phosphorylates and, in turn, inhibits both TSC2 and FOXO1. Following phosphorylation, TSC2 loses its ability to inhibit mTOR complex 1 (mTORC1). Of relevance, activation of mTORC1 occurs at multiple levels, thus it is incorrect to assume that PI3K signaling encompasses full activation of mTORC1 (ref. 137 ). FOXO1 is phosphorylated by AKT at three serine/threonine residues, which results in its nuclear exclusion and, in turn, the inactivation of its transcriptional activity 138 .
phenotypically affected in PROS. An enigmatic aspect is whether PIK3CA mutations are present but silent in non-affected tissues, or are they just not present. One possibility is that mesoderm-derived and ectoderm-derived tissues are more sensitive to PI3K overactivation, while in endoderm-derived tissues, PIK3CA mutations are not sufficient to cause pathogenesis. Another possibility is that mutation acquisition favors differentiation into specific lineages, as shown in breast cancer 49 . In fact, homozygous PIK3CA mutations induce self-sustained stemness and resistance to spontaneous differentiation in human induced pluripotent stem cells 50 . These data suggest that PIK3CA mutations persist better in less differentiated cell states. Nevertheless, we cannot rule out that PIK3CA mutant clones undergo negative selection in specific cell lineages during development. Of note, PIK3CA mutations have been found in healthy adult endoderm-derived tissues such as the esophagus 51 . Although PIK3CA mutant clones outcompete their wild-type neighbors in that context, they do not lead to abnormal tissue growth 51 . This outcome would fit with PIK3CA mutations being largely nonpathogenic in endoderm-derived tissues.
The bias of PIK3CA variants in PROS. Missense activating mutations in PIK3CA span almost the entire gene in cancer and PROS. In cancer, more than 80% of somatic mutations occur in three hotspots located in the helical (p.Glu542Lys and p.Glu545Lys) and kinase (p.His1047Arg) domains 2 . While the mutational profile of PIK3CA in PROS is similar to that of cancer, the occurrence of mutations outside these three hotspots is much higher 23 (Fig. 3).
Mutations with lesser gain-of-function activity are probably rare in cancer because of their reduced oncogenic potential. Instead, it seems that mild and weak activating PIK3CA mutations are sufficient to generate a pathogenic response when acquired at embryonic stages. Indeed, p.Gly914Arg and p.Glu726Lys, which are likely weak activating PIK3CA mutations, are also hotspots in PROS 3,23 . Intriguingly, sporadic cases of isolated PROS with two mutations have been identified, where a hotspot mutation is combined with a non-hotspot mutation 44,52 . However, it is not clear whether these mutations are present in the same clone in cis or trans, or in different clones. Of note, it is important to bear in mind that cancer hotspot mutations have been preferentially mapped for genetic testing in PROS, which has quite likely underestimated the occurrence of mutations beyond the aforementioned hotspots. In line with this, there is also a bias in the clinical visibility of severe cases, which tend to be overrepresented in genetic testing 53 .
The emergence of numerous genetic studies in the context of PROS is allowing us to infer phenotypes from genotypes. For example, the majority of individuals who have MCAP with a reported genetic diagnosis carry a non-hotspot mutation in PIK3CA, with p.Gly914Arg and p.Glu726Lys being the most common variants 12,23,33,41,[54][55][56][57] . MCAP is a subtype of PROS disorder in which the primary affected tissues derive from two different developmental layers, ectoderm (neurons and macroglia) and mesoderm (vasculature and microglia), thereby suggesting that weakly activating mutations in PIK3CA are more compatible with their existence before the divergence of the germline layers. Although rare, some individuals  programs, causing different tumor types and clinical outcomes 49 .
Modeling Pik3ca-related mouse brain overgrowth has shown that distinct variants induce different cellular effects (hyperplasia versus hypertrophia) 66 . Together, these data suggest that PIK3CA variants quite likely exhibit qualitative differences by means of variant-specific molecular signals in both cancer and PROS 65 .
Extrinsic factors in PROS pathogenesis. Tissue growth chiefly takes place during embryogenesis and early postnatal periods, when most cells in the organism divide and grow extensively. It is during these growth phases that activating mutations in PIK3CA favor PROS onset. In line with this, tissues with high plasticity that are in constant adaptation to the microenvironmental needs, like vasculature and adipose tissues, are greatly affected in PIK3CArelated congenital disorders. Recent data have shown that onset and growth of some Pik3ca-related vascular malformations rely on the synergy between Pik3ca mutations and growth factors 37,67,68 . This cooperative pathogenic effect may explain why in adulthood, when growth factor signals are residual, these lesions do not form de novo or that existing ones progress very slowly. This behavior also fits the observation that many lesions regrow after incomplete surgical removal, when the body reacts locally, boosting the production of growth factors to promote wound closure, and explains why some asymptomatic lesions suffer growth spurts during injury, adolescence and pregnancy (hormonal changes). Thus, individuals may benefit from therapies that, at specific timing windows (for example, after a resection), inhibit microenvironment-derived paracrine-specific signals 67 . Fibro-adipose vascular anomaly (FAVA) is an example of a PROS condition in which individuals are asymptomatic at birth with lesions developing in the extremities during late childhood and adolescence. Some individuals with FAVA have reported lesions appearing after an accidental event causing physical injury (E. Baselga, personal communication). Hence, it is tempting to speculate that tissue injury results in hypoxia and acute production of growth factors which would favor the progression of vascular lesions. Cells are in constant exposure to biomechanical cues (shear, tensile and compressive stresses) 69 , being at their foremost during developmental stages 70 . Indeed, mechanotransduction (the cellular response induced by biomechanical cues) is believed to contribute to cell fate decisions 71 . Mechanotransduction events are particularly relevant for those cells which coexist in cell-cell or cell-ECM contact. For example, ECs that establish adherent junctions with one another and are in constant interaction with both their luminal and abluminal extracellular space are extremely dependent on mechanobiology signaling 72 . Intriguingly, the anatomic location of a mutant clone has been recently identified as a critical factor for tumorigenesis 73 . Hence, it is tempting to speculate that tissue architecture and mechanical forces contribute to defining such anatomic-related pathogenesis. Much evidence suggests that oncogenic signaling synergizes with mechanotransduction to promote pathogenesis 74 .
Other reports have identified that PIK3CA-related pathogenesis is supported by non-cell-autonomous mechanisms. For example, Martin-Corral et al. have shown that the presence of Pik3ca mutant clones in lymphatic vessels results in the accumulation of immune cells, including macrophages, which then become a major source of vascular endothelial growth factor C (VEGF-C) that can further promote pathological lymphangiogenesis 67 . Other such studies have provided evidence that PIK3CA mutant clones in tumors, via paracrine communication, interfere with wild-type or ERBB2 mutant clones in close proximity and, thus, catalyze their aberrant behavior 75,76 . We anticipate that understanding how mutant clones hijack these other clones may open up new therapeutic opportunities.

PIK3CA mutations in vascular malformations
Vascular malformations occur in isolated and syndromic PROS. Even within syndromic PROS, which tends to feature a variable admixture of overgrown tissues, the vascular compartment/tissue is that most commonly affected. This observation highlights that vascular malformations are a hallmark of PROS (Fig. 1), which is coherent with PI3Kα being a master regulator of EC biology (Box 4). Vascular malformations are abnormal vessels that grow aberrantly during embryonic development and are manifested at birth or throughout the life of affected individuals. They grow slowly and do not regress spontaneously over time. Depending on the type(s) and localization of the affected vessels, individuals' symptoms can range from mild to severe, even life-threatening. These vascular lesions often cause pain, swelling or bleeding, together with cosmetic deformities that can interfere with the normal function of the affected areas. Occasionally, vascular malformations are not clinically manifested until their growth is triggered by events such as adolescence, pregnancy or injury 77 .
Subtypes of PIK3CA-related vascular malformations. Vascular malformations are divided into low-flow (venous, lymphatic and capillary) and fast-flow (arteriovenous) lesions. They can also be classified as simple, when only a specific vascular bed is affected such as capillary, venous, lymphatic or arteriovenous anomalies, or combined, when a lesion presents two or more vascular malformations or mixed vascular beds, characteristic of capillary-venous malformations (CVMs), lymphatic-venous malformations (LVMs) or capillary-venous-lymphatic malformations (CVLMs), among others. Intriguingly, PIK3CA-related vascular malformations are

Box 4 | PI3Kα in the endothelium
PI3Kα has emerged as a master regulator of vascular morphogenesis in blood and lymphatic vessels 142 . Genetically engineered mouse models have shown that both inactivation and overactivation of PI3Kα activity in the germ line or specifically in BECs cause profound defects in the blood vasculature, ultimately resulting in embryonic lethality 18,26,27,143 . This phenotype is explained by PI3Kα being the main producer of PIP 3 during RTK stimulation in ECs. In line with this, inactivation or deletion of other class I PI3K isoforms does not interfere with vascular nor embryonic development 143 . Lymphatic endothelial-specific depletion of PI3Kα results in perinatal lethality owing to selective alterations in mesenteric and intestinal lymphatic vessels 144 . Stanczuk et al. also showed that LECs rely differently on the VEGFR3-PI3Kα signaling pathway depending on their origin and tissue location. An intriguing aspect of PI3Kα is that it is activated via distinct mechanisms in BECs and LECs. BECs utilize the regulatory subunits to recruit PI3Kα to the plasma membrane following RTK stimulation, whereas LECs also require the RBD of PI3Kα to be intact. Therefore, mice carrying mutations in Pik3ca that block PI3Kα binding to RAS have selective defects in lymphatic vessel development 145 . PI3Kα regulates vessel growth via both AKT dependent and independent mechanisms 146,147 , including (1) control of cell cycle progression via the inactivation of FOXO1 (ref. 146  largely found in low-flow lesions, including isolated VMs, capillary malformations (CMs), LMs or in combination. PIK3CA mutations have not been reported in isolated arteriovenous malformations (AVMs) and it is not clear why this the case. One possibility is that PIK3CA mutations behave as silent mutations in arteries; in fact, arterial ECs have molecular refractoriness to other vascularrelated mutations 78 . However, targeted sequencing of the PIK3CA gene in more than 100 surgically resected human brain AVMs did not identify any positive cases 4,5 . Another possibility is that arterial mutant clones are negatively selected during vascular development or homeostasis. Of note, some CLOVES patients with a positive genetic test (carrying an activating PIK3CA mutation) have been associated with the presence of an AVM. However, it has not been confirmed whether these AVMs are composed of PIK3CA mutant cells or if a second hit (genetic or environmental) is necessary for the development of the AVM 14,79,80 .
Lymphatic malformations are a prototypical example of PIK3CA dominance. LMs are debilitating vascular anomalies classified as cystic (microcystic and macrocystic) LMs (also known as common) and complex lymphatic anomalies. While the former appears in isolation, complex lymphatic anomalies show a diffuse and multifocal pattern and may cause defects in the central collecting lymphatic channels, such as generalized lymphatic anomalies (GLAs), Gorham-Stout disease (GSD), kaposiform lymphangiomatosis and central conducting lymphatic anomalies. Activating mutations in PIK3CA have been detected in the majority of cystic LMs and GLAs 20,81 . The cancer hotspots are overrepresented in LMs 35 . While this could be explained by the notion that variants with lesser gain-of-function activity are not sufficient to induce pathogenesis in the lymphatic endothelium, it is also possible that this vascular compartment is more tolerant than other lineages to high PI3K signaling. Of note, mutations in the helical domain (p.Glu542Lys/p. Glu545Lys) are more common than in p.His1047Arg in LMs 35 , although the significance of this remains to be determined.
The generation of Pik3ca-related LM mouse models has provided key insights into the molecular and cellular factors that determine the onset and the subtype of LMs 67,81,82 . For example, the expression of the Pik3ca p.His1047Arg mutation in VEGFR3positive cells during early embryonic development recapitulates traits of macrocystic LMs. By contrast, if the same mutation is expressed in VEGFR3-positive cells during late embryonic or early postnatal stages, the mice develop microcystic LMs 67 . Another study showed that the expression of P110 (a dominant active Pik3ca transgene with 20-fold higher kinase activity than p.His1047Arg) in VEGFR3-positive cells in adult mice causes multifocal LMs 82 . Similarly, the expression of the Pik3ca p.His1047Arg mutation in PROX1-positive lymphatic cells after weaning resulted in GLAs 81 . These data suggest that the type of cell/precursor, time of activation and the mouse modeling genetic approach used to activate PI3Kα signaling are key factors in the onset of different subtypes of LMs. Martin-Corral et al. showed that VEGF-C-VEGFR3-dependent activation of mutated PI3Kα promotes the growth of microcystic LMs in mice. These data explain why combined inhibition of VEGF-C signaling and mTOR using a soluble VEGF-C trap and rapamycin, respectively, compared with their treatments alone, leads to the regression of microcystic LMs 67 . Alpelisib, a PI3Kα selective inhibitor, also ameliorates LM symptoms in mouse models and in individuals with cystic LMs who previously did not respond to rapamycin 82 . Currently, it is not clear why rapamycin and alpelisib induced different responses in LMs; however, it is tempting to speculate that the latter provides an overall better response owing to direct inhibition of the mutant protein. Also, given that rapamycin only targets a branch of the PI3K pathway, it is possible that PIK3CA-related pathogenesis occurs via mTOR-dependent and mTOR-independent mechanisms. Collectively, this emphasizes the importance of generating faithful preclinical models for each LM subtype for so-called personalized medicine. PIK3CA and TEK. VMs are bluish lesions caused by aberrant EC proliferation that show enlargement, tortuosity, reduced mural coverage and impaired functionality. Common VMs 83 are the most frequent (90% of all VMs) and are caused by activating mutations in TEK (60%) 84,85 or PIK3CA (20-25%) [17][18][19] , with both mutations being largely mutually exclusive. While PIK3CA mutations only occur in a somatic fashion, both somatic and germline mutations in TEK can cause VMs. PIK3CA and TEK mutations lead to increased PI3Kα signaling, although TEK mutations activate the pathway to a lower extent 17,37,86 . This difference might explain why some TEK mutations are compatible with their survival in the germline. There is a clear tissue-genotype association between TEK and PIK3CA, with TEK-related VMs being mostly found on the skin surface, whereas PIK3CA-related VMs are preferentially located in intramuscular areas 17,18 . An important aspect to consider is that TEK-related lesions tend to be purer VMs than PIK3CA-related lesions, which often also express some lymphatic markers. Data on the co-occurrence of PIK3CA and TEK mutations in the same lesion are emerging 18,86 , yet it is too early to say whether second events play a wider role in the severity of these diseases and whether multi-genetic events cooccur in the same cells.

PI3K overactivation in venous malformations is a matter of
Mouse models of Pik3ca-related common VMs have also flourished. For example, Castillo et al. reproduced the etiology of VMs by widespread mosaic induction of Pik3ca p.His1047Arg under the endogenous promoter in the mouse lateral plate mesoderm during embryonic development 19 . At present, it remains enigmatic as to why this mouse model solely develops vascular malformations, predominantly VMs, while avoiding major alterations in other tissues. Based on these data, it is tempting to speculate that the type of PROS also relates to the precursor subtype wherein the mutation occurs. Another study has shown that ubiquitous, but mosaic, expression of Pik3ca p.His1047Arg in adult mice also leads to a rapid and exclusive development of VMs 18 . Collectively, these mouse models have shown that Pik3ca-related VMs form via proliferation and are deprived of mural cells 18,19 . Xenograft models using human ECs derived from individuals with VMs have also emerged 86 . While these models have become a relevant tool for preclinical testing, they have limitations for the understanding of the onset and biology of these diseases.

Multigenic events in cerebral cavernous malformations pathogenesis.
Cerebral cavernous malformations (CCMs) are vascular malformations formed by closely clustered and enlarged capillary channels specifically located in the brain and spinal cord. There are two types of CCM disease, familiar (20% of CCMs) and sporadic (80% of CCMs). Until recently, it was believed that CCMs were monogenic diseases, largely caused by inherited or somatic loss-offunction mutations in one of three genes (KRIT1 (also known as CCM1), CCM2 or PDCD10 (also known as CCM3)) that encode the heterotrimeric CCM protein complex 87,88 . In 2021, somatic gain-offunction mutations in MAP3K3 (encoding MEKK3) were also found in sporadic orphan CCMs 89-91 , with MAP3K3 and CCM genes being mutually exclusive. Several studies have now shown that CCMs with prominent and rapid growth associated with strokes and seizures carry an additional somatic genetic hit in PIK3CA, chiefly in a cancer hotspot 5,89,90 . PIK3CA mutations may co-occur with MAP3K3, KRIT1, CCM2 and PDCD10 mutations.
Modeling CCMs in mice using endothelial-specific CreERT2 mouse models has confirmed that the synergy between loss of Krit1 and expression of Pik3ca occurs via interaction within or between ECs 5 . In line with this, isolation of ECs from human CCMs has confirmed that these cells carry PIK3CA mutations 89 . However, another such study in mice has proposed that Pik3ca mutations in CCMs occur in pericytes, a subtype of mural cells that adhere to and support capillary endothelial function 4,92 . These data remain controversial as the Cre mouse line used to activate Pik3ca p.His1047Arg expression is neither inducible nor pericyte specific. In addition, no proof was provided that the ECs in the mouse model did not carry Pik3ca mutations 93 . In support of possible mural cell involvement in CCM disease, another report showed that specific deletion of Pdcd10 in mural cells, including both pericytes and smooth muscle cells, resulted in CCM. While proof that human brain pericytes carry PIK3CA mutations would be required to validate these findings, it is possible that PIK3CA mutations in ECs and mural cells account for different subtypes of CCMs. Of note, pericytes do not rely on PI3Kα, but PI3Kβ to activate PI3K signaling, which would be coherent with the noninvolvement of PIK3CA mutations in pericytes in CCM progression 94 .
Intriguingly, sporadic CCMs are frequently found in close proximity to developmental venous anomalies (DVAs) 95,96 . DVAs are the most common vascular malformations in the brain (present in about 10% of the adult population) and largely develop before the age of 20 years. Emerging data suggest that sporadic CCMs may, at least partly, derive from DVAs. While comparing the mutational status of a DVA and its paired CCM (present in the same individual), the DVA and CCM were found to both carry the same somatic activating PIK3CA mutation, whereas the CCM lesions also harbored mutations in MAP3K3. Based on this finding, it was proposed that individuals who have a PIK3CA-related DVA are predisposed to develop sporadic CCMs in proximity to the DVA 91 . Collectively, these studies have catapulted PIK3CA as a critical genetic hit for CCM disease. However, at present, it is unknown whether PIK3CA mutations are required for the formation of CCMs or whether they serve as endothelial clonal amplifiers.
Capillary malformations are the least severe vascular malformations within PROS. CMs are anomalies consisting of dilated capillaries near the surface of the skin that normally present a macular, pink to red stain. PIK3CA-related CMs have been largely described as part of some PROS disorders, such as MCAP and DCMO, where they are mainly caused by activating non-hotspot mutations in PIK3CA 33,97 . Also, a recent case was reported with an acquired CM (after birth) associated with the p.Val344Met PIK3CA variant 98 . Overall, these lesions tend to be largely cosmetic, thereby indicating that they are less severe than LMs or VMs. Mutations in GNAQ and GNA11 (encoding Gα q and Gα 11 , respectively) are overrepresented in cutaneous CMs. Intriguingly, co-occurrence of PIK3CA and GNAQ/GNA11 has also been reported in combined vascular malformations 6 .

treatment options for Pros in the era of repurposing drugs
With the discovery of PIK3CA mutations to be causative of PROS, new treatment possibilities have emerged (Box 1). Below, we summarize the most promising attempts and we refer the reader to Table 1 for specific details on completed and ongoing clinical trials for PROS.
Sirolimus was the first targeted therapy to treat PROS. Sirolimus (also known as rapamycin) is an allosteric mTOR inhibitor approved by both the Food and Drug Administration (FDA) and the European Medicines Agency. The very first clinical trial using sirolimus was planned in patients with vascular anomalies, even before the implementation of genetic diagnosis as a precondition for trial inclusion. Together with other follow-up clinical trials, sirolimus was shown to reduce the overall volume of vascular malformations, with LMs exhibiting the highest susceptibility to the drug [99][100][101][102] . However, several adverse events were also identified in these studies 101 . The PROMISE study, which tested a lower dose of sirolimus, arose as the first clinical trial specifically for syndromic PROS patients 103 . In that context, sirolimus showed a modest reduction in the volume of the overgrown tissues, which was also accompanied by some adverse events. Although these results were not as good as initial expectations, it is fair to acknowledge that many individuals have benefited from this inhibitor. These beneficial responses explain why nowadays many individuals with PROS are treated with sirolimus as standard care. Currently, there are several ongoing clinical trials assessing the potential benefits of sirolimus for PROS, including the assessment of topical administration on superficial lesions.

Miransertib is an AKT inhibitor for the treatment of PROS.
Miransertib (currently known as MK-7075), an allosteric highly selective AKT inhibitor, was first used on a compassionate basis for Proteus syndrome (PS) and PROS patients 104,105 . While partial therapeutic efficacy and no major toxicities were observed, it seemed to be most effective in severe PROS. Preclinical studies in isolated vascular malformations have also shown promising data, even when using half the dose used in oncology 37 . Ongoing clinical trials are testing the safety and effectiveness in PS and PROS patients with a confirmed genetic diagnosis.

PI3K inhibitors for PROS.
A clinical trial using taselisib, a selective pan class I PI3K inhibitor, was initially set up for PROS patients 106 . Taselisib showed clinical improvement with reduced pain, chronic bleeding resolution and functional improvement. However, the presentation of severe drug-related adverse events in some individuals led to the early termination of the trial. Adverse events were thought to be caused by the impact of PI3Kδ/γ inhibition on the immune system, thereby suggesting that selective inhibition of the PI3Kα isoform would be a better choice in that context. Since then, expectations have been placed on alpelisib, which was first tested in mouse models of CLOVES-like syndrome 56 and LMs 82 . While these results showed higher efficacy than sirolimus, treatment withdrawal led to the recurrence of the lesions, indicating that chronic treatment may have to be considered 56 .
An unregistered case series of PROS patients with confirmed PIK3CA mutations treated with alpelisib on a compassionate basis reported evidence of efficacy with minor adverse effects 56 ; however, neither safety nor efficacy endpoints were prespecified in that study 56 . Later, other studies also confirmed promising results using alpelisib on a compassionate basis [107][108][109] . Nevertheless, the lack of evidence of benefit-risk assessments on a long-term basis pushed a retrospective and non-interventional study of 57 individuals treated with this inhibitor (EPIK-P1). The reported analysis of the first endpoint revealed that 37.5% of alpelisib-treated individuals exhibited a ≥20% reduction in lesion volume and 38.6% of individuals had hyperglycemia, aphthous ulcer and stomatitis 110 . EPIK-P3 is now continuing on from EPIK-P1 to evaluate long-term safety and efficiency. Alpelisib has also been tested in individuals with LMs who showed a general improvement with decreased LM volume and reduced tiredness and pain, while reporting moderate adverse events such as aphthous ulcers and diarrhea 82 . The promising data on alpelisib for compassionate use have finally catalyzed an ongoing prospective, multicenter, randomized, double-blind and placebo-controlled clinical trial (EPIK-P2) to demonstrate the efficacy, tolerability and safety of alpelisib. In April 2022, a gigantic step was achieved with the approval of alpelisib by the FDA for individuals ≥ 2 years of age with severe PROS manifestations who require systemic therapy.

Conclusions and perspectives
With the advent of next-generation sequencing, PIK3CA has emerged as a frequently mutated oncogene in congenital disorders with asymmetric overgrowth. This finding has accelerated the generation of mouse models of these diseases to understand the  mechanisms leading to their onset and progression. Lessons learnt suggest that the pathogenic score of activating PIK3CA mutations in congenital disorders is a combination of the developmental time (when) that these mutations appear, together with intrinsic factors (cell autonomous) such as the mechanisms and degree of PI3Kα activation (which), the cell lineage specificity of the mutated cell (where) and extrinsic factors (non-cell autonomous). Unraveling the specific contribution of each of these elements will be required to define improved treatments for these individuals. In line with this, mapping the entire PIK3CA gene should be considered when genotyping individuals with a suspicion of PROS. It is clear that some lineages are more sensitive to PI3K overactivation, yet the basis of such context-dependent pathogenesis remains to be discovered. We propose that single-cell RNA-sequencing approaches in combination with genetic mouse models, which will allow us to express Pik3ca mutations in specific subpopulations, will be critical to identify cell lineage histories of mutant clones and to understand how PI3Kα activation subverts cellular processes in a context-dependent manner. Indeed, emerging single-cell RNAsequencing data in mouse LMs have identified a new immunoregulatory subtype of dermal LECs as a driver of LM pathology 68 . Second or even triple genetic hits in vascular malformations are emerging in aggressive clinical cases, overruling the original idea that congenital vascular malformations are mostly monogenic disorders that behave differently from cancer. These data highlight the urge to look for multiple genetic hits in other PIK3CA-related disorders that may explain the severity or tissue-specific pathogenesis.
While we are still learning from all the pharmacotherapy studies, it is now clear that drugs targeting the PI3K signaling pathway have become essential tools to treat PROS. It is important to consider that individuals with PROS have high heterogeneity in their clinical manifestations, which probably explains the heterotypic responses to the different treatments. Therefore, once the safety of these treatments is well established, physicians will have to evaluate the riskbenefit balance of each patient according to their specific needs and responses. We believe that it is of vital importance to consider non-cell-autonomous mechanisms to refine current treatments. Response to co-inhibition of specific upstream or downstream signaling pathways may also be different among the PIK3CA variants. Targeted studies of molecular programs underlying distinct types of PIK3CA-related overgrowth would be needed to inform precision medicine strategies.