Introduction

The interest in hypoxia-inducible factors (HIFs) and their role in biology and the pathophysiology of various diseases has grown significantly over the past three decades, particularly following the discovery of VHL gene variants in Von Hippel–Lindau (VHL) disease, which is a hereditary tumor predisposition syndrome characterized by neuroendocrine tumors, including paraganglioma/pheochromocytoma (PPGL)1,2,3. Building on this foundation, we identified somatic variants in HIF2A (EPAS1) in multiple individuals with a syndrome of a triad of PPGL, polycythemia, and somatostatinoma, also known as Pacak–Zhuang syndrome (PZS)4. Since their initial discovery in VHL disease5, HIFs have long been recognized as critical mediators of homeostasis and biological processes, including angiogenesis, proliferation, and migration2. In the presence of oxygen, HIF-α proteins are hydroxylated by prolyl-hydroxylase domain-containing proteins (PHD), then ubiquitinated by VHL complex and degraded by the proteosome2. In response to hypoxia, HIF-α proteins are stabilized and dimerize with HIF-ß, and the complex travels to the nucleus, binding to hypoxia response elements (HRE) across the genome, altering gene expression. The discovery of HIF2A pathogenic variants not only explained the syndromic manifestation but also established the first direct causative link between hypoxia signaling and cancer biology4,6. Since the initial findings, we have continued to investigate this syndrome, expanding our understanding of the role of HIF-2α in clinically observed neurodevelopmental and vascular anomalies. Moreover, the investigation of HIF-2α in PPGL pathophysiology catapulted efforts to molecularly stratify these tumors into three main clusters: Cluster 1—pseudohypoxia cluster— related to HIF-2α and Krebs cycle defect, Cluster 2 related to increased kinase activity, and Cluster 3 related to the Wnt signaling pathway.

Genetic variant character and disease phenotype

Recently, Ferens et al. proposed a genotype-phenotype classification of PZS based on the structural biochemistry of HIF-2α variants and the correlated phenotypes7. Using wild-type and mutant synthetic proteins and microscale thermophoresis to measure protein-protein interactions, Ferens et al. demonstrated that HIF-2α variants affect the affinity of HIF-2α for binding to prolyl hydroxylase-2 (PHD2)7. The authors suggested that major alterations in the oxygen-dependent degradation (ODD) domain of HIF-2α would result in very low to no binding affinity to PHD2, resulting in more stable HIF-2α, and thus manifesting as the full disease triad of PZS7. Alternatively, less disruptive HIF-2α variants with moderate effect on the affinity of HIF-2α for PHD2 would manifest as only polycythemia (elevated red blood cell count). This is the first clinically relevant classification schema explaining the variability in the clinical phenotypes of patients. This represents an important milestone in the field, relating to our continuously evolving understanding of disease pathogenesis with disease phenotype, as did the molecular stratification of PPGLs into 3 clusters8,9. Our phenotyping studies of PZS have suggested that molecular stratification alone may not be the only consideration to explain disease extent, as the timing of acquiring these variants in development, e.g., early vs late somatic mosaicism (as defined below), may be as important as the class of variant in determining the disease phenotype, as evidenced by the varying allele frequencies throughout multiple tissues in our patients10. Whether there are other contributing factors, including acute or prolonged oxygen oscillations in utero, geographical location (especially areas of high altitude), or some specific environmental factors, including endocrine-disrupting chemicals, should be strongly considered and warrant further investigation.

Genetic variant spatiotemporal distribution and disease phenotype

The precise location and character of genetic variants are generally believed to be a critical determinant of disease phenotype in germline neoplastic syndromes11,12. For example, in VHL disease, which shares a molecular pathogenic pathway of increased hypoxia signaling with PZS, the location and type of pathogenic variants within the VHL gene contribute critically to phenotypic presentation13. Germline missense and truncating pathogenic variants with later somatic loss of heterozygosity (LOH) lead to a predisposition of hemangioblastomas, PPGLs, and renal cell carcinoma (RCC), i.e., VHL disease13. However, the biallelic c.598 C > T pathogenic variant in the same gene (3′ to the classic VHL-associated pathogenic variants) results in a different phenotype—polycythemia and congenital vascular anomalies in a syndrome known as Chuvash polycythemia14. This underscores the impact of the type and structural biochemistry of the VHL genetic variant on the resultant disease phenotype.

The discovery of PZS, which has a subtly different disease phenotype to VHL disease, provided the first direct evidence that aberrant hypoxia signaling mediated by HIF-2α is sufficient to cause tumor development. While VHL disease is caused by a germline variant and, therefore, may potentially affect every cell throughout the body, PZS is caused by somatic mosaicism with variable allele frequency, or degree of mosaicism, throughout the soma10. This insight led to our understanding that the absence of certain phenotypic features in patients with neoplastic syndromes may result from the absence, or variable frequency, of the variant in a given tissue, rather than purely the type of genetic variant, which determines the strength of association between HIF-2α and PHD2. Taken together, the insights from the study by Ferens et al.7 and our previous work suggest that to understand the phenotypic development in neoplastic syndromes, the structural biochemistry of a genetic variant and its distribution in the soma would need to be combined.

Illustrative cases

Two individuals from our cohort illustrate this point. First, an individual with PZS (Patient 1) presented with a mild phenotype—no somatostatinoma, polycythemia, and no evidence of PGL recurrence after resection of multiple tumors within the last six years. We would expect the genetic variant (EPAS1P531S) in this individual to result in a more severe phenotype based on the predicted loss of interaction between PHD2 and HIF-2α7. However, the phenotype was mild, which we believe is explained by the lower degree of mosaicism in multiple tissues (Table 1). In contrast, another individual in our cohort (Patient 2) presented with a more severe clinical course with recurrence of PGLs after four surgeries, frequent catecholaminergic crises, and a metastatic somatostatinoma (Table 1). This individual, notably, had a high variant allele frequency (25%) of EPAS1A530T, which we would expect to produce a milder phenotype according to the study by Ferens et al. These cases highlight the importance of considering the degree of mosaicism, i.e., the spatiotemporal distribution of acquired genetic variants, in disease classification. Clinical phenotypes fall on a spectrum from high to low mosaicism, following different natural histories and, therefore, may require more individualized clinical care, as reflected in our schema (Fig. 1).

Table 1 Pacak–Zhuang patient characteristics
Full size table
Fig. 1: Proposed molecular genetic schema to explain disease presentation in Pacak–Zhuang syndrome (PZS).
figure 1

Incorporation of multiple levels of molecular and genetic information enables the understanding of the phenotypic spectrum of PZS. a Diagram of possible disease outcomes resulting from differences in the spatiotemporal distribution of genetic variants in EPAS1, which encodes hypoxia-inducible factor 2α (HIF-2α). We previously found that somatic mosaicism in HIF-2α results in PZS with varying phenotypes, while others previously found that germline variants in EPAS1 may result in familial erythrocytosis. b Three-dimensional structure of the PHD2 protein (light blue) bound to the oxygen degradation domain (ODD) of HIF-2α (tan) (PDB ID: 7UJV). This domain contains mutational hotspots in HIF-2α that alter its binding affinity to PHD2, forming the basis of ‘class variants’ described by Ferens et al.1. Class 1 variants exhibit the weakest binding affinity, while class 2 variants bind more strongly than class 1 but still less than wild-type HIF-2α. c A graphical representation of genetic variants in EPAS1 and their classification according to Ferens et al. is shown: class 1 variants in red, class 2 in blue, and residues such as Y532 and D539 that may result in either class 1 or 2 in purple. d We propose an expansion of the Ferens et al. framework to include the spatiotemporal distribution of EPAS1 variants, based on its allele frequency across multiple tissues. Variants with higher allele frequencies are associated with more severe phenotypes, reflecting earlier acquisition, ranging from germline to later somatic events. We further suggest that this distributional component can be integrated with the structural biochemistry classification defined by Ferens et al. Created in BioRender. Alkaissi, H. (2025) https://BioRender.com/f30algg.

Full size image

Several neoplastic syndromes have a variable presentation and natural history, even in individuals harboring the same variant, as well as within individuals in families with hereditary cancer syndromes15. For example, germline pathogenic variants in neurofibromin 1 (NF1), lead to neurofibromatosis, a tumor predisposition syndrome characterized by 100% penetrance of systemic disease by age 5, while somatic and/or mosaic pathogenic variants in NF1 result in isolated neurofibromas with later onset16. Further, even within the same disease, e.g., germline neurofibromatosis (both de novo and hereditary), the disease phenotype may present differently based on the spatiotemporal distribution of the expression of the altered gene16, which can be mediated by the tissue specificity of the gene expression and parent-of-origin imprinting, and, therefore, may affect some tissues more than others16,17. Our schema may explain other diseases with tissue-restricted phenotypes. While our proposed schema considers the type and spatiotemporal distribution of the pathogenic variant, which likely explains most disease pathogenesis, factors that modify the expression of an acquired variant may also alter disease presentation. For example, a recent study including a remarkable international collaboration between European and South American centers reported significant phenotypic differences, such as age of onset or laterality, in patients with neuroendocrine tumors due to the same pathogenic variants18. Bilateral pheochromocytomas were more common in patients from Europe, as compared to South American patients, and were more likely to present the disease at an earlier age.

Concluding remarks

In summary, incorporating the degree of somatic mosaicism and the spatiotemporal distribution of pathogenic variants would be as important as identifying the pathogenic variant and its effect on protein functions and interactions. This would apply to PZS and, more broadly, to known diseases, including tumor predisposition syndromes and sporadic disease. With this new classification, patients with localized (late mosaicism) pathogenic variants would have a milder and more localized clinical manifestation even if the variant is molecularly classified as more deleterious, and vice versa.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.