Inherited genetics of adult diffuse glioma and polygenic risk scores—a review

Abstract

Knowledge about inherited and acquired genetics of adult diffuse glioma has expanded significantly over the past decade. Genomewide association studies (GWAS) stratified by histologic subtype identified six germline variants that were associated specifically with glioblastoma (GBM) and 12 that were associated with lower grade glioma. A GWAS performed using the 2016 WHO criteria, stratifying patients by IDH mutation and 1p/19q codeletion (as well as TERT promoter mutation), discovered that many of the known variants are associated with specific WHO glioma subtypes. In addition, the GWAS stratified by molecular group identified two additional novel regions: variants in D2HGDH that were associated with tumors that had an IDH mutation and a variant near FAM20C that was associated with tumors that had both IDH mutation and 1p/19q codeletion. The results of these germline associations have been used to calculate polygenic risk scores, from which to estimate relative and absolute risk of overall glioma and risk of specific glioma subtypes. We will review the concept of polygenic risk models and their potential clinical utility, as well as discuss the published adult diffuse glioma polygenic risk models. To date, these prior genetic studies have been done on European populations. Using the published glioma polygenic risk model, we show that the genetic associations published to date do not generalize across genetic ancestries, demonstrating that genetic studies need to be done on more diverse populations.

Background

Understanding of the genetics of adult diffuse glioma has significantly expanded in the past decade, both with respect to inherited and acquired genetics (inherited genetics refers to the DNA that humans are born with whereas acquired genetics refers to DNA changes that occur during a human’s lifespan). Genes with acquired alterations in the tumor (acquired genetics) also commonly have germline alterations in or near the gene that increase risk of glioma development (inherited genetics), implying that glioma development is governed by an interaction between inherited and acquired genetics. In 2016, Rice et al. reviewed the current state of glioma inherited genetics.¹ The clinical diagnosis of adult diffuse glioma has changed significantly since 2016, with the most significant change being the integration of classical histological features with acquired genetic alterations. As a result of the integrated diagnosis of glioma, the understanding of germline variants and their association with specific histologic and molecular subtypes of glioma has expanded. Thus, the current review provides an update regarding the interaction between inherited and acquired glioma genetics. Additionally, the clinical utility of inherited genetics has been investigated by developing polygenic risk models that attempt to identify patients at extreme risk of cancer. Thus, we also discuss how information across all known glioma inherited variants can be used to develop a glioma polygenic risk model that estimates relative and absolute risk of glioma. With respect to potential clinical utility of these polygenic risk models, it is important to note that most of the glioma inherited variants identified to date have been identified by studying cohorts of European ancestry. Thus, we demonstrate that glioma risk scores estimated from these European-based polygenic risk models do not generalize to all non-European ancestries, and discuss the importance of expanding genetic studies to diverse populations.

Clinical Relevance of Acquired Molecular Alterations

Prior to 2016, adult diffuse glioma was classified and graded based solely on tumor histological features as astrocytoma, oligodendroglioma, oligoastrocytoma, or glioblastoma (GBM) and tumor grade II, III, or IV. In 2016, the World Health Organization (WHO) updated the classification, and for the first time, used an integrated diagnosis that incorporated tumor histological features as well as acquired tumor molecular markers.² The most recent 2021 WHO classification expands the use of molecular markers and classifies adult-type diffuse gliomas into three entities: (i) astrocytoma, IDH-mutant, (ii) oligodendroglioma, IDH-mutant, and 1p/19q-codeleted, and (iii) glioblastoma, IDH-wildtype.³ Tumor grade is then assigned within each of the three entities, where tumor grade is determined using both histological features as well as tumor molecular markers that are known to be associated with prognosis.^3,4 For example, even if a tumor appears histologically to be a lower grade tumor, detection of the following acquired alterations will result in the tumor being classified as grade 4: CDKN2A/B homozygous deletion (for IDH-mutant gliomas) and TERT promoter mutation, EGFR amplification, or gain of chromosome 7 and loss of chromosome 10 (for IDH-wildtype astrocytoma). In summary, the 2016 WHO criteria for adult diffuse glioma was a paradigm shift in that it integrated molecular markers with classical histological features. However, the three main adult diffuse glioma entities did not drastically change between the 2016 and 2021 WHO criteria; the changes primarily entail the nomenclature and grading scheme.

Clinical Relevance of Inherited Genetics

Genomewide association studies (GWAS) are aimed at identifying germline variants that are associated with risk of a disease. GWAS are case-control studies that have historically been performed using array platforms that contain hundreds of thousands of variants and imputation is performed to estimate variants that are not on the array platform. The analysis approach is typically a logistic regression model. Each candidate germline variant is analyzed separately, where the dependent variable is a dichotomous variable denoting the disease status (case versus nondiseased control) and the independent variable is the number of risk alleles (e.g., minor allele) for the corresponding variant.⁵

A summary of all GWAS germline adult diffuse glioma risk variants discovered to date are shown in Figure 1. GWAS were initially performed treating adult diffuse glioma as a single disease and identified nine independent germline risk variants in or near eight genes; these variants were subsequently evaluated to determine if they were associated with specific histologic or molecular subtypes of glioma.^6–10 In 2015 and 2017, GWAS stratified by histologic subtype identified 18 additional novel variants: six that were associated with GBM risk and 12 that were associated with non-GBM risk (i.e., risk of developing an IDH-mutant glioma).^11,12 Importantly, these variants did not reach genome-wide significance (typically defined as P-value <5 x 10^–8, which corrects for multiple testing) when the GWAS was performed in all glioma; they only reached genome-wide significance when the analysis was performed within more homogeneous glioma subgroups. Utilizing the 2016 WHO criteria², a GWAS stratifying patients by IDH mutation and 1p/19q codeletion (as well as TERT promoter mutation) discovered that many of the known variants are associated with specific clinically relevant glioma molecular groups.¹³ In addition, two novel regions were identified: SNPs in D2HGDH were associated with tumors that had an IDH mutation, and a SNP near FAM20C that was associated with tumors having both IDH mutation and 1p/19q codeletion.¹³

The germline associations observed to date can be grouped into four general categories (Table 1). Note that while variants are often denoted by gene names, this does not imply that the variant resides within a protein-coding region.¹ Gene names are typically chosen based on the proximity of a variant to a gene or the hypothesized relationship between the variant and a nearby gene.

• Germline variants in or near genes associated with development of all gliomas (TP53).

• Germline variants in or near genes associated with the development of IDH-mutant gliomas (MDM4, AKT3, IDH1, D2HGDH, LRIG1, CCDC26, OBFC1, VTI1A, MAML2, ZBTB16, PHLDA1, AKAP6, ETFA, LMF1).

• Germline variants in or near genes associated with development of IDH-wildtype gliomas (RAVER2, EGFR, CDKN2A/B, FAM181B, MPG, HEATR3, RTEL1, SLC16A8).

• Germline variants in or near genes that are associated with development of IDH-mutant 1p/19q-codeleted gliomas, but not IDH-mutant noncodeleted glioma (EGFR and FAM20C), or vice versa (PHLDB1 and TERT). In fact, the EGFR and TERT variants were historically associated with IDH-wildtype tumors; however, there is evidence to suggest that they are also associated with specific IDH-mutated tumors.¹⁴

These germline association results are important for several reasons, including as a starting point for understanding pathogenesis. For example, some of the known germline risk variants are associated with age at diagnosis. The number of risk alleles for variants in TERT and RTEL1—both telomerase-related genes—are associated with older age at diagnosis across all glioma grades and histologies.^15,16 Conversely, the number of risk alleles for variants in CCDC26 and PHLDB1 are associated with younger age at diagnosis, particularly for oligodendroglial tumors.^15,16 It appears that the TP53 germline variant is involved in the development of all gliomas; it may interact with some acquired alterations to develop IDH-mutant gliomas, and other acquired alterations to develop IDH-wildtype glioma. Perhaps most intriguing is that some germline risk variants reside in or near genes that also frequently have acquired mutations in the tumor. The nature of the relationship between inherited and acquired alterations is not clearly known; however, there are likely several possibilities. The germline variant may nullify the need for a subsequent acquired mutation (e.g., CDKN2A/B) to allow glioma development. Conversely, a germline variant may facilitate the acquisition of a pathogenic acquired mutation in the same region. For example, the germline variant near IDH1 may predispose to the subsequent acquisition of a pathogenic IDH1 acquired mutation. Alternatively, germline variants may interact in some common pathway (e.g., IDH1, ETFA, and D2HGDH are involved in the glycolysis and oxidative phosphorylation pathways).

Germline Polygenic Risk Scores for Estimating Glioma Risk

Some germline variants (eg, CCDC26, TP53) have large enough effect sizes to possibly be used for early detection in high-risk populations and to predict glioma subtype (ie, tumor aggressiveness) prior to surgery. While the CCDC26 and TP53 variants have the largest effect sizes, for any given individual, all the glioma risk variants have a combined effect on risk. To account for this combined effect, all germline risk variants can be incorporated into a multivariable model to estimate relative and absolute risk of developing glioma. The multivariable model is often referred to as a polygenic risk model and is denoted by an additive model where each known germline variant is weighted by the corresponding logarithm transformed odds ratio (OR).¹⁷ The resulting value is called a polygenic risk score (PRS); a larger PRS denotes higher cancer risk. The clinical utility of a PRS is to identify individuals with significantly increased risk (eg, to identify patients in the top 5^th percentile of cancer risk). Seven diseases were identified where a PRS could have clinical utility, including breast and prostate cancer, coronary artery disease, obesity, type 1 and type 2 diabetes, and Alzheimer’s disease.¹⁸ Specifically, PRS were reported to be predictive of response to statin therapy for coronary artery disease,^19–21 to define high risk populations for whom to use serum prostate-specific antigen (PSA) to screen for prostate cancer^22,23 and to determine screening recommendations for breast cancer.^24–26

With respect to glioma, a polygenic risk model was developed that incorporated twenty-five glioma germline risk variants to calculate a person’s polygenic risk score, from which to estimate relative and absolute risk of overall glioma.¹⁴ With respect to polygenic risk scores, relative risk is typically calculated relative to individuals who have a median risk score or relative to the lowest risk score. This process is shown in Figure 2, where the data denote results from the published glioma risk model¹⁴:

• Step 1: The polygenic risk score is calculated on a large cohort of noncancer control individuals.

• Step 2: The distribution of polygenic risk scores from individuals in the noncancer control cohort are used to define percentile categories of polygenic risk.

• Step 3: The polygenic risk score is calculated on a large cohort of adult diffuse glioma patients. Subsequently, glioma patients are assigned to the risk score percentile categories that were defined on the noncancer controls in Step 2.

• Step 4: The relative risk of glioma cases versus noncancer controls for each percentile category is estimated (via logistic regression analysis), relative to a reference percentile group (e.g., the middle percentile category).

• Step 5: Since the absolute risk of developing an adult diffuse glioma is low, the lifetime absolute risk can be estimated by multiplying the lifetime risk in the general population by the relative risk for each percentile category. The absolute risk in the general population can be obtained by resources such as SEER or CBTRUS.

For glioma, the published polygenic risk model observed that patients with a risk score in the highest 5^th percentile of risk for the IDH-mutant 1p/19q codeleted subtype or for the IDH-mutant 1p/19q intact subtype had more than a 14-fold increased risk of developing a glioma with an IDH mutation in comparison to individuals with a median risk score.¹⁴ This increase in relative risk is more than two times larger than any published risk score to date for any cancer even though the other cancer models used more than three times as many germline risk variants.¹⁴

While the prevalence and absolute risk of adult diffuse glioma is too low to screen the general population using a polygenic risk model,¹ there are scenarios where an estimate of glioma risk could be helpful in clinical decision making. For example, there are instances where a brain lesion is seen in an eloquent location on a brain MRI, without features prompting immediate intervention. In these cases, there may be clinical equipoise as to whether to proceed with a biopsy or resection, or to take a watchful waiting approach; an estimate of glioma risk may influence whether and when to intervene or determine the nature and timing of follow-up. To further enhance risk estimates for clinical decision making, results from a polygenic risk score could be combined with other imaging and laboratory methods such as MRI-based machine learning,^27–30 magnetic resonance spectroscopy,³¹ circulating cell free DNA,^32–34 and tumor microvesicle³⁵ technologies. For example, MRI-based machine learning models have been evaluated for their ability to differentially diagnose GBM, solitary brain metastases, and CNS lymphoma.^{27–30,36–38} Preliminary results demonstrated that DNA methylation profiling of cell free DNA may be a diagnostic tool for glioma.³² While tumefactive multiple sclerosis (MS) and other tumefactive inflammatory brain lesions have not been included in these studies to date, distinguishing tumefactive inflammatory brain lesions from neoplasms is also critical.³⁹ Thus, noninvasive methods that provide more accurate discrimination of these brain lesions can allow for tailored therapy, optimal tumor control, and a reduced risk of iatrogenic morbidity and mortality.

To illustrate how the glioma polygenic risk model might discriminate glioma from other types of brain lesions, we applied the published glioma polygenic risk model to GWAS data from 924 Multiple Sclerosis (MS) patients that were in the MS and Related Disorders cohort of the Brigham & Women’s Hospital Multiple Sclerosis Genetic Collection.⁴⁰ The glioma risk score distribution in the MS patients were more similar to the glioma risk score distribution in healthy controls than in glioma patients (see Step 6 in Figure 2); 21% of the MS patients were in the upper 25^th percentile of the risk score versus 52% of IDH-wildtype glioma patients. While these preliminary results are interesting, it is necessary to further evaluate both the sensitivity and specificity of the glioma polygenic risk model on a larger cohort of candidate disease conditions that often present with brain MRI findings potentially suggestive of a broad list of conditions that include inflammatory diseases such as autoimmunity or MS, and neoplasms such as glioma, lymphoma, and brain metastases. And as discussed above, MRI-based machine learning and other imaging modalities could be integrated with polygenic risk estimates to refine differential diagnosis in the future.

Germline Polygenic Risk Scores for Determining Glioma Subtype

As a second illustration, germline risk variants were also utilized to develop a glioma polygenic subtype model for application to suspected glioma patients to estimate the probability having a brain tumor with an IDH mutation.¹⁴ The polygenic subtype model had a validation concordance index (c-index) of 0.85; the “validation” prefix denotes that the c-index was computed in an independent validation cohort. The c-index is analogous to the area under the curve (AUC) and is a measure of model discrimination, i.e., how well the model discriminates IDH-mutant tumors from IDH-wildtype tumors. A c-index of 0.5 implies that the prediction is like flipping a coin, whereas a value of one implies that the model predicts perfectly. MRI-based machine learning models have also been evaluated for their ability to predict acquired alterations in brain tumors, including IDH mutation, 1p/19q codeletion, and MGMT promoter methylation.^41–45 Identifying noninvasive techniques to predict glioma molecular subtype prior to surgery could impact surgical intervention, giving greater importance of gross total resection in some instances versus others. It may also allow more informed pre-operative discussion of surgical risks with a patient and selection of patients for more aggressive resections with appropriate use of advanced intraoperative monitoring and awake mapping procedures to maximize resection from eloquent regions when needed⁴⁶.

Importance of Ancestry in Genetic Studies

Most GWAS studies to date—regardless of disease/cancer type—have been performed on European populations, likely due to logistical or historical reasons.⁴⁷ Thus, while there may be clinical utility of polygenic risk models, it is important to note that genetic models derived from European populations may not generalize to all non-European populations.⁴⁸ To explore this hypothesis, we applied the published glioma polygenic risk model that was developed using data from European populations to independent cohorts of European and non-European ancestries; for the current application, age and sex were removed from the polygenic risk model. Glioma risk scores were calculated on individuals with glioma in The Cancer Genome Atlas (TCGA) of European (n = 764), African (n = 50), Ad Mixed American (n = 24), East Asian (n = 8), and South Asian (n = 7) ancestry. These five genetic ancestries are defined as the five “super populations” by the HapMap project and 1000 Genomes project (https://www.internationalgenome.org/). Affymetrix 6.0 data were analyzed and genetic ancestry was estimated using the ADMIXTURE program.⁴⁹ While the sample size is small for some of the groups, the shift in distributions across genetic ancestries in Figures 3a, 4a, and 5a demonstrate that the glioma risk score that was developed on Europeans does not generalize across ancestries. For example, the average IDH-mutant noncodeleted risk score for Africans and Ad Mixed Americans with IDH-mutant noncodeleted glioma is smaller than for Europeans with IDH-mutant noncodeleted glioma; the pink and green curves in Figure 3a are shifted to the left, in comparison to the black curves.

Glioma risk scores were also calculated on individuals in TCGA of European ancestry with primary tumor locations that are known to have the potential to metastasize to the brain: breast invasive carcinoma (BRCA), colon adenocarcinoma (COAD), kidney renal clear cell carcinoma (KIRC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), and rectum adenocarcinoma (READ) (Figures 3b, 4b and 5b). The risk score distribution for individuals of European ancestry with these other tumors was similar to the distribution observed in the Mayo (n = 443) and UCSF (n = 231) noncancer controls of European ancestry, suggesting that the PRS may help to distinguish between primary glioma and metastasis. Glioma risk scores were also calculated on non-European individuals in TCGA with these other primary (nonbrain) tumor locations. Figures 3c, 4c and 5c further demonstrate that the polygenic risk score distribution differs across ancestries. For example, the IDH-mutant noncodeleted risk score distribution for each of the non-European ancestries shown in Figure 3c are shifted to the left of the corresponding distributions in Europeans shown in Figure 3b. Similar shifts in risk score distribution can be seen for the IDH-mutant codeleted risk score (comparing Figures 4c and 4b), and to a lesser degree for the IDH-wildtype risk score (comparing Figures 5c and 5b). Figures 3c, 4c and 5c also contain a cohort of 236 African noncancer controls that were collected as part of a chronic lymphocytic leukemia study and similarly genotyped on the Affymetrix 6.0 platform. The results for this large cohort of Africans demonstrates that the observed shifts in the distribution of risk scores for Africans is not an artifact of the small sample sizes available in TCGA. Interestingly, the difference in inherited genetics between Africans and Europeans may partially explain the lower incidence of glioma in Africans compared to Europeans.⁵⁰ These observations further suggest that the scientific community and funding mechanisms need to expand genetic studies to include diverse populations to better understand glioma etiology.^47,48

Discussion

The three primary entities of adult diffuse glioma in the 2021 WHO criteria are based on two acquired tumor markers, IDH mutation and 1p/19q codeletion, and are: (i) astrocytoma, IDH-mutant, (ii) oligodendroglioma, IDH-mutant, and 1p/19q-codeleted, and (iii) glioblastoma, IDH-wildtype. These three entities have different survival outcomes, age at diagnosis, as well as different inherited genetics.^51,52 With respect to inherited genetics, the germline associations observed to date can be grouped into categories that reflect the three primary glioma entities: (i) germline variants associated with development of all gliomas, (ii) germline variants associated with the development of IDH-mutant gliomas, (iii) germline variants associated with development of IDH-wildtype gliomas, and (iv) germline variants associated with development of IDH-mutant 1p/19q-codeleted gliomas, but not IDH-mutant noncodeleted glioma, or vice versa. The germline variants, and their associations with specific glioma entities, are important for understanding the mechanism of how gliomas develop. There are likely important mechanistic relationships between the germline variants and the various tumor acquired molecular alterations, and this interplay between germline and acquired alterations will have important clinical and biologic significance. Recent advances in mouse modeling, human stem cell biology (e.g., iPS cells), and sequencing technology (e.g., single cell RNA sequencing, conformational chromosomal capture) will facilitate mechanistic experiments examining how germline and acquired alterations interact to enhance glioma development.

There is also mounting evidence across multiple diseases/cancers that germline variants can be used to estimate polygenic risk scores and identify patients at high risk of a particular disease/cancer, and that these risk scores may have clinical utility. In adult diffuse glioma, polygenic risk scores could have the potential to help with differential diagnosis of indeterminate brain lesions as well as to determine glioma molecular subtype (IDH mutation status) prior to surgery. Prospective studies are necessary to understand the utility of these polygenic risk scores applied to the intended clinical populations versus the research populations that they were developed on.⁵³ This includes verifying that the risk scores are able to sufficiently discriminate patients based on risk, and that the discrimination estimates account for absolute risk.⁵⁴ Additionally, it will be important to train physicians on how to communicate risk scores and to study the psychosocial impact on patients and families.⁵³ While there are few environmental risk factors for adult diffuse glioma, additional clinical data might improve the polygenic risk models, provided that the data can be captured consistently and accurately (e.g., history of seizures).⁵⁵ With respect to genotyping technology, large GWAS arrays, sequencing technology, or custom clinical assays can be designed and utilized. Polygenic risk scores can also be calculated using direct-to-consumer genotyping data.^56,57 While such tools already exist to obtain polygenic risk scores from direct-to-consumer tests,⁵⁷ it is important to consider data quality and the effect of quality on the accuracy of the polygenic risk models.

While there may be clinical utility of polygenic risk models, it is important to acknowledge that most GWAS studies to date have been performed on European populations, and the corresponding results do not always generalize to non-European populations. Thus, to not further exasperate health disparities, GWAS studies need to be performed on diverse populations. This will allow for a more thorough understanding of the development of glioma across populations (e.g., explain the lower incidence of glioma in Africans compared to Europeans), as well as the development of polygenic risk models that can be applied across all populations.

Going forward, there is evidence to suggest that circulating biomarkers—both inherited and acquired molecular biomarkers—can be detected and potentially used to help with diagnosis of brain cancer and determining molecular subtype.^14,32 Additionally, MRI-based models may have potential for the differential diagnosis of indeterminate brain lesions and to predict acquired genetic alterations.^{27–30,36–38,41–45} However, to fully evaluate and compare these approaches, it is important that manuscripts detail the experimental design, summarize patient clinical characteristics, and describe the internal validation approach in enough detail to understand how subjects were assigned to training, validation, and test sets. With respect to reporting the results, it is important to provide standardized performance metrics so that an evaluation of the method as a diagnostic test can be determined,⁵⁸ and to describe the variables/features that make up the models. Additionally, more investigation is required to validate these approaches in the clinical environment in which they will be applied, and to determine how best to implement them into clinical practice. Subsequently, clinical staff will need training with respect to interpretation of the results and discussing results with patients. Because similar research is being performed across diseases/cancers, much will be learned over the coming years. And, it is expected that in the near future, polygenic risk scores and MRI-based models will soon be incorporated into the medical record for diagnostic, predictive, and prognostic purposes.

Acknowledgments

The authors acknowledge study participants, clinicians and research staff at participating medical centers (including Annette Molinaro, PhD, Terri Rice, Helen Hansen, Lucie McCoy, Paige Bracci, PhD, Joseph Wiemels, PhD, John Wiencke, PhD, Alissa Caron, Terry Burns, MD, and Caterina Giannini, MD), Katherine Cornelius, Mayo Clinic Comprehensive Cancer Center Biospecimens and Processing and Genotyping Shared Resources, Gliogene Consortium, UCSF Helen Diller Family Comprehensive Cancer Center Genome Analysis Core, UCSF Cancer Registry, and the UCSF Neurosurgery Tissue Bank. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. The collection of cancer incidence data used in this study was supported by the California Department of Public Health pursuant to California Health and Safety Code Section 103885; Centers for Disease Control and Prevention’s (CDC) National Program of Cancer Registries, under cooperative agreement 5NU58DP006344; the National Cancer Institute’s Surveillance, Epidemiology and End Results Program under contract HHSN261201800032I awarded to the UCSF, contract HHSN261201800015I awarded to the University of Southern California, and contract HHSN261201800009I awarded to the Public Health Institute, Cancer Registry of Greater California. The ideas and opinions expressed herein are those of the author(s) and endorsement by the State of California Department of Public Health, the National Cancer Institute, and the Centers for Disease Control and Prevention or their Contractors and Subcontractors is not intended nor should be inferred.

Funding

This publication was supported by the National Center for Research Resources, National Center for Advancing Translational Sciences, and National Institutes of Health, UL1 RR024131 R01NS113803, R01CA230712), Bernie and Edith Waterman Foundation, and Ting Tsung and Wei Fong Chao Family Foundation. MW and the UCSF Adult Glioma Study were supported by the National Institutes of Health (R01CA052689, P50CA097257, R01CA126831, R01CA139020, R25CA112355, and R01CA207360), loglio Collective, Stanley D. Lewis and Virginia S. Lewis Endowed Chair in Brain Tumor Research, National Brain Tumor Foundation, and by donations from families and friends of John Berardi, Helen Glaser, Elvera Olsen, Raymond E. Cooper, and William Martinusen.

Conflict of interest statement. There are no conflicts of interest to disclose.

References

Author notes