Most Fanconi anemia heterozygotes are not at increased cancer risk: a genome-first DiscovEHR cohort population study

Abstract

Purpose:

Fanconi anemia (FA) is a bone marrow failure and cancer predisposition syndrome caused primarily by biallelic pathogenic variants in one of 22 genes involved in DNA interstrand crosslink repair. An enduring question concerns cancer risk of those with a single pathogenic FA gene variant. To investigate all FA genes, this study utilized the DiscovEHR cohort of 170,503 individuals with exome sequencing and electronic health data.

Methods:

5,822 subjects with a single pathogenic variant in an FA gene were identified. Two control groups were used in primary analysis deriving cancer risk signals. Secondary exploratory analysis was conducted using the United Kingdom Biobank and The Cancer Genome Atlas.

Results:

Signals for elevated cancer risk were found in all five known cancer predisposition genes. Among the remaining 15 genes associated with autosomal recessive inheritance cancer risk signals were found for four cancers across three genes in the primary cohort but were not validated in secondary cohorts.

Conclusion:

This is the first and largest FA heterozygote study to use genomic ascertainment and validates well-established cancer predispositions in five genes while finding insufficient evidence of predisposition in 15 others. Our findings inform clinical surveillance given how common pathogenic FA variants are in the population.

Introduction

Fanconi anemia (FA) is an inherited bone marrow failure (IBMFS) and cancer predisposition syndrome caused primarily by biallelic pathogenic variants in one of >22 genes involved in DNA interstrand crosslink repair¹. Most FA genes have an autosomal recessive (AR) inheritance pattern. FANCB is X-linked recessive (XLR), and FANCR (RAD51) is autosomal dominant (AD). Diagnosis of FA is confirmed with a chromosome breakage test of peripheral blood T-cells or skin fibroblasts cultured with DNA-crosslinking agents². Patients with FA are at particularly high risk of developing acute myeloid leukemia (AML), myelodysplastic syndrome (MDS), and head, neck and anogenital squamous cell carcinomas³.

Of the FA genes, five (FANCD1 [BRCA2], FANCJ [BRIP1], FANCN [PALB2], FANCO [RAD51C], and FANCS [BRCA1]) are known cancer predisposition genes when inherited in a monoallelic autosomal dominant manner⁴. Autosomal recessive inheritance of these genes is associated with rarer forms of FA, and there are differences in genotype-phenotype outcome associations across these FA genes^5,6.

An enduring question, especially in families of those with FA, concerns the cancer predisposition of those with a single pathogenic FA gene variant. Studies in 1980 and 1983 failed to identify increased risk of cancer, though these studies were performed before the cloning of the first FA gene FANCC in 1992 and were not informed by confirmed heterozygote status^7,8. After 1992, several larger studies also did not show an elevated risk of cancer, though one study did show increased rates of breast cancer in FANCC heterozygote grandmothers (6/33 relatives: SIR, 2.4; CI, 1.1–5.2)^9,10. More recent studies have proposed FANCI as a potential ovarian cancer predisposing gene, FANCC having a potential role in pediatric Ewing sarcoma, and FANCM protein-truncating variants as risk factors in breast cancer^11,12,13. In 2022, the National Cancer Institute (NCI) studied relatives of patients with FA within the 151 families enrolled in its IBMFS cohort. An increased risk of cancer was not found in these individuals, but the analysis was underpowered for each specific FA genotype except FANCA and FANCC⁴.

Large population-level exome/genome sequencing projects with linked health records are a promising adequately powered tool to study cancer risk using a genome-first approach, where subjects are ascertained based on genotype, not phenotype or family history. The DiscovEHR Cohort within the MyCode Community Health Initiative at Geisinger is one such cohort, consisting of 170,503 individuals with exome sequencing data and linked electronic health records¹⁴. Understanding cancer risk is important for the counseling and surveillance of those with a single pathogenic FA gene variant, who are more frequently identified with the ever-increasing use of genetic testing.

Materials and Methods

The study analyzed 170,503 individuals enrolled in the DiscovEHR (MyCode Community Health Initiative at Geisinger) cohort. Participants provided consent and biospecimen collection as previously described^14,15. Genomic analyses were performed as part of the DiscovEHR collaboration between Geisinger and the Regeneron Genetics Center. The Geisinger Health System institutional review board reviewed this work and determined it to be non-human subject research.

Pathogenic/likely pathogenic (P/LP) variants in 22 Fanconi anemia (FA) genes were identified from exome sequencing. Copy-number variants were identified using exome sequencing data followed by validation with microarray genotyping. Variants were annotated using ANNOVAR¹⁶, snpEFF¹⁷, and ClinVar (http://www.ncbi.nlm.nih.gov/clinvar/, September 2022). Pathogenicity was determined based on a modified ClinVar classification¹⁸ (Supplemental Figure 1). A total of 5,822 subjects with a single P/LP variant were identified and designated the heterozygote group (Supplemental File- spreadsheet of variants). A subset of 270 subjects were identified as having variants known to cause FA based on previous literature⁵ or within the NCI IBMFS cohort⁶ (NCT00027274). 125 patients were excluded from this study for separate analyses due to having multiple P/LP variants across genes or a single pathogenic variant in the AD or XLR (for males) FA genes.

ICD10 C-codes were curated to Phecodes¹⁹ and ICD-O-3 Site Recodes (https://seer.cancer.gov/siterecode/icdo3_dwhoheme/) for analyses using two separate control groups. The study employed this strategy to increase the evidence-basis and reduce potential control group bias. The first control group included all DiscovEHR subjects excluding the heterozygote group and those with a variant of uncertain significance in the gene of interest with a deleterious in-silico score (see Supplemental Methods). The second set of controls were derived from the NCI Surveillance, Epidemiology, and End Results (SEER) database (https://seer.cancer.gov/). We first determined adjusted odds ratios (OR) comparing the mapped Phecodes observed in the heterozygotes compared to the DiscovEHR internal controls (n≈160,000, varied per gene, see Supplemental Methods). ORs were adjusted for age, sex, smoking status, and Body Mass Index (BMI). Second, we used matched controls from the SEER database to determine observed/expected ratios (O/E). O/Es were adjusted for age, sex, and birth cohort.

Cancer phenotypes were analyzed across individual genes and multiple groupings (Supplemental Figure 2). When both arms of the analysis were complete, Phecodes and SEER analyses were compared, and findings were pursued if results showed statistically significant ratios in both cancer risk analyses. Findings were only considered if the number of subjects with the specific cancer type was greater than two.

Given statistical power of 80% and the study sample size, we could detect an odds ratio (OR) of 1.1 in heterozygotes across all 15 AR genes and 1.2 across the five known cancer predisposition genes. For individual genes, the minimally detectable OR was 2 or less (lowest 1.26) for all genes except FANCE (3.2), FANCV (3.9), and FANCW (2.2) (https://sample-size.net/proportions-effect-size/).

We also conducted validation analyses using data from the 469,787 individuals in the UK Biobank (UKBB, https://www.ukbiobank.ac.uk/)²⁰ and 2,847 germline samples from The Cancer Genome Atlas (TCGA, https://www.cancer.gov/ccg/research/genome-sequencing/tcga)²¹ to investigate cancer risk across four cancer types (endometrial, esophageal, lung, melanoma) and three FA genes (FANCI, FANCP, FANCT). Variants were identified in exome sequencing and curated as described above and with GEMSCAN (Supplemental Methods). UKBB phenotypes were derived from ICD-9 and ICD-10 codes.

Results

5,822 individuals (40% male) with a single pathogenic FA variant had a mean age of 56.9 comprising 331,272 person years (Supplemental Table 1). These 5,822 individuals represent a total heterozygote prevalence of 3.4%, compared to previous studies which estimated a prevalence of 0.5%²² based on patient cohorts and 2.6%²³ based on gnomAD (https://gnomad.broadinstitute.org/). The distribution of variants (Supplemental Figure 3) also deviated substantially from that of FA patients, where FANCA, FANCC, and FANCG typically compose most patients.

Across these 5,822 heterozygotes, we identified 1946 total cancers (0.33 cancers per subject) via electronic health record (EHR). In 152,108 controls, we identified 39,176 total cancers (0.26 cancers per subject). When stratified by location in the FA/BRCA DNA repair pathway (Supplemental Table 2), most cancers were in heterozygotes of variants in genes downstream in the pathway (1,046 cancers) followed by upstream (792) and ID complex (108). 807 cancers were identified across the five known cancer predisposition genes (1,647 subjects) and 1,139 cancers were identified across the other 15 AR genes (4,175 subjects).

In the primary cohort, signals for elevated cancer risk, defined by statistically significant excess in both control arms, were found in all five known cancer predisposition genes (O/E=2.15, OR=2.00, p=2.07e-32). Additionally, positive signals were found in 3 AR genes (Table 1). FANCI had a signal for endometrial cancer (O/E=3.89, OR=2.45, p=0.0151), FANCP (SLX4) for melanoma (O/E=2.94, OR=2.74, p=0.016) and lung cancer (O/E=3.13, OR=3.79, p=8.4e-5), and FANCT (UBE2T) for esophageal cancer (O/E=5.89, OR=5.28, p=1.18e-3). When a Bonferroni correction for multiple testing (adjusted for number of C-codes) is applied, only FANCP for melanoma and FANCT for esophageal cancer remain significant with a p value of 0.05.

Table 1.

Combined SEER and Phecode analyses for FA genes across different cancer types in DiscovEHR, adjusted for age, sex, smoking status, body mass index, and birth cohort.

Gene (Cancer Type)	O/E (SEER)	95% CI (SEER)	OR (Phecode)	95% CI (Phecode)	p Value (Phecode)
CPG (Pancancer)	2.15 (607/282.93)	1.98–2.32	2	1.78–2.24	2.07×10 −32
AR (Pancancer)	1.12 (870/776.48)	1.05–1.2	1.07	0.984–1.15	0.118
FANCS/BRCA1 (Pancancer)	3.55 (187/52.75)	3.06–4.09	3.54	2.80–4.47	4.3×10 −26
FANCD1/BRCA2 (Pancancer)	2.15 (263/122.53)	1.89–2.42	2.13	1.79–2.55	0.000487
FANCN/PALB2 (Pancancer)	2.02 (61/30.24)	1.54–2.59	2.27	1.62–3.17	0.0000016
FANCJ/BRIP1 (Ovarian Cancer)	3.80 (15/3.94)	2.13–6.27	3.03	1.33–6.89	0.00789
FANCO/RAD51C (Breast Cancer)	2.08 (9/4.32)	0.95–3.95	2.62	1.31–5.23	0.006
FANCO/RAD51C (Ovarian Cancer)	6.82 (3/0.44)	1.37–19.93	3.71	1.16–11.9	0.027
FANCM (Breast Cancer)	1.08 (27/25.11)	0.71–1.56	1.19	0.81–1.73	0.374
FANCI (Endometrial Cancer)	3.89 (7/1.8)	1.53–7.87	2.45	1.19–5.07	0.0151
FANCP (Melanoma)	2.94 (6/2.04)	1.07–6.40	2.74	1.2–6.24	0.0219
FANCP (Lung Cancer)	3.13 (10/3.2)	1.50–5.75	3.79	1.95–7.37	0.000084
FANCT (Esophageal Cancer)	5.89 (4/0.68)	1.58–15.07	5.29	1.93–14.4	0.00118

Significant values are bolded. Pancancer: all cancers in ICD-10 classifications; O/E: observed number of cancers/expected number of cancers; OR: Odds ratio; SEER: Surveillance, Epidemiology, and End Results program. CPG: cancer predisposition genes (FANCD1/BRCA2, FANCJ/BRIP1, FANCN/PALB2, FANCO/RAD51C, FANCS/BRCA1). AR: genes with autosomal recessive inheritance (FANCA, FANCC, FANCD2, FANCE, FANCF, FANCG/XRCC9, FANCI, FANCP/SLX4, FANCQ/ERCC4, FANCU/XRCC2, FANCT/UBE2T, FANCV/MAD2L2, FANCW/RFWD3)

The most common cancers seen in patients with FA (AML, MDS, squamous cell carcinomas) did not demonstrate significant increased predisposition amongst the 15 AR genes. The previously postulated increased risk of chronic lymphocytic leukemia and increased risk of female breast cancer in FANCC heterozygotes were not observed⁴. Pancancer associations were observed only in variants in downstream genes (O/E=1.72, OR=1.58, p=5.22e-21) [upstream (O/E=1.14, OR=1.08, p=0.10), ID complex (O/E=1.18, OR=0.979, p=0.074] (Supplemental Figure 2). Within the 270 subjects with variants observed in known FA patients, no significant association was identified across all cancers (p=0.102).

Within two secondary cohorts exploring FANCI, FANCP, and FANCT, no significant associations were identified between heterozygotes and cancer in the UKBB (Table 2). Additionally, germline samples of all four cancer types in TCGA found no pathogenic variants in FANCI and FANCT. In contrast, there were two patients with pathogenic variants in FANCP. Upon further review of clinical history, these two patients were determined as unlikely to have germline-influenced lung cancers due to significant smoking history and late onset of cancer.

Table 2.

Secondary analysis in UK BioBank of FANCI, FANCP, and FANCT for cancers with dual positive signals in DiscovEHR.

Gene (Cancer Type)	OR	95% CI	p Value
FANCI (Endometrial Cancer)	1.29	0.180–9.19	0.801
FANCP (Melanoma)	0.894	0.400–2.00	0.785
FANCP (Lung Cancer)	0.306	0.0760–1.23	0.0948
FANCT (Esophageal Cancer)	1.98	0.819–4.80	0.129

No significant values identified. Adjusted for age, sex, smoking status, BMI, and birth cohort.

Discussion

With the increasing use of genetic testing and identification of single pathogenic FA gene variants in patients, a robust understanding of cancer risk in these individuals is necessary for surveillance recommendations in clinical practice.

Building upon our previous research on FA relatives⁴, this study represents the largest investigation of FA heterozygotes and the only study to utilize a genome-first methodology to date. We utilized large population databases with sufficient power to study FA genes that previous studies were underpowered to analyze and include genes that are less represented in the FA patient population. To ensure the reliability for population-wide findings, we employed a multi-layered approach including a dual control group analysis and multiple secondary cohorts.

Additionally, our genome-first approach allows for evaluation of penetrance in a population unselected for family history of FA which is crucial to understanding the true effect of heterozygosity. Whereas previous estimates of heterozygote prevalence based on patient populations were as low as 0.5%, this study found a near seven-fold higher prevalence, even while only restricting to the ClinVar P/LP variants. These may reflect single variants being passed down that produce embryonically lethal phenotypes if inherited in biallelic fashion, though the true mechanisms are still unknown and require clinical corroboration. Additionally, since variants were genomically ascertained and thus identified in people without a family or personal history of FA or cancer (people who typically do not undergo genetic testing), these variants may be silent in the population and circulating more widely than previously recognized. If we consider, the coding length of each of the FA genes studied we do not see a correlation between the length of the coding region and the prevalence of heterozygotes in that gene or group of genes (Supplemental Table 3). For example, BRCA2 is large, 11,386 base pairs, with 669 heterozygotes identified, while UBE2T is relatively small, 939 base pairs, but with a mid-range number of heterozygotes, 377 within DiscovEHR. It is possible, that some of the FA genes are more tolerant to variation and thus retained in the population while other are not, and thus filter out over evolutionary time. The genome-first method of identifying variants has demonstrated frequencies differing from previous estimates in several cases²⁴, and elucidates variants that would otherwise be hidden in populations using traditional approaches. Interestingly, we also noted significant differences in the variant distribution across all the genes (Supplemental Figure 3). While FANCA, FANCC and FANCG are the most common within patients with FA, the most common heterozygotes identified were FANCM, with 13% of heterozygotes having a FANCM variant, and the second most common being BRCA2 (11%). These are relatively rare genotypes amongst patients with FA. The reason for these differences is not clear, but it is possible that some biallelic BRCA2 cases are missed as it is known to have a unique and severe FA phenotype. In contrast, it is possible that not all biallelic FANCM individuals have FA, it may depend on specific variants or modifiers, however, further research is needed.

Overall, increased cancer risk was not identified in most individuals. We did, however, confirm previously well-established cancer predispositions conferred by heterozygous autosomal dominant inheritance of pathogenic variants in BRCA1, BRCA2, PALB2, BRIP1, and RAD51C. These individuals should undergo personalized cancer screening per National Comprehensive Cancer Network guidelines (https://www.nccn.org/guidelines/nccn-guidelines). Further investigation into pathogenic variants in CPGs that may not necessarily confer cancer risk could provide clues to molecular mechanisms associated with inherited cancer susceptibility. For the remaining autosomal recessively inherited FA genes, although an initial study of the primary DiscovEHR cohort demonstrated limited positive cancer predisposition signals for FANCI, FANCP, and FANCT, further exploration studies did not confirm the initial findings. Thus, there was insufficient evidence of an increased risk of cancer in individuals with a single pathogenic variant.

The potential increased risk of breast cancer in FANCM heterozygotes has been studied over the last ten years, and while some studies²⁵ show and increased risk, others do not²⁶. Peterlongo and colleagues summarized these studies¹³, and commented that certain protein truncating variants may confer more risk than other based on degree of truncation of the protein, particularly for estrogen receptor negative and triple negative cases. They also comment that there may be genetic modifiers which alter the expressivity within certain populations. Here, we did not observe an increased risk of breast amongst FANCM heterozygotes (Table 1). It is possible that the DiscovEHR population is significantly different from those in prior studies such that the needed modifiers are not present, or that the variants represented are different from prior studies, however, all variants included here are ClinVar P/LP variants and were protein truncating. In general, genome-first studies have shown lower risk estimates than those observed in clinical cohorts and highlight the ascertainment biases present²⁴.

Despite the scale of this study, there are some limitations. While this study offers substantial power for most FA genes, FANCE, FANCV, and FANCW were underpowered to detect an OR under 2. Additionally, while the use of three cohorts increases robustness, there remains a lack of diversity in the cohort population. The DiscovEHR and UKBB cohorts are both predominately white populations and further study into additional cohorts are needed.

Overall, our findings are adequately powered, use multiple cohorts and are clinically relevant and useful and should be reassuring in the common genetic counseling situation of detection of a single P/LP FA allele from panel testing.

Data Availability Statement

All data described in this study are provided within the article and supplementary materials.