Paternal age effect in autosomal dominant or X-linked de novo variants identified by genome-wide sequencing

Summary

An increased frequency of sporadic autosomal dominant disorders has been observed among children born to older fathers. This paternal age effect is thought to reflect an accumulation of new mutations in the male germ line as DNA replication and cell division continue to occur as men age. Genome-wide sequencing is useful for identifying disease-causing genetic variants in patients with suspected genetic diseases and for determining inheritance or de novo mutation of the variants when done in patient-parent trios. We analyzed paternal ages in 593 families who received trio or quad exome or genome sequencing for suspected genetic disease. The mean age of fathers of children with de novo disease-causing variants (35.09 years) was significantly greater than that of children with inherited disease-causing variants (33.78 years, p = 0.04). The mean age of mothers of children with de novo disease-causing variants (31.86 years) was not significantly greater than that of children with inherited disease-causing variants (30.80 years, p = 0.09). Interestingly, when the de novo disease-causing variants were broken down into subgroups by variant type, both mean paternal age and mean maternal age of children with de novo indel variants (paternal = 36.33 years, maternal = 33.34 years) were significantly higher than in children identified to have de novo single-nucleotide variants (paternal = 34.35 years, p = 0.03; maternal = 31.15 years, p = 0.004). This observation, which may have implications for how indels arise, requires further study.

The mean maternal age was significantly greater for children with de novo disease-causing indels than for children with de novo disease-causing single-nucleotide variants in 593 families who received genome-wide sequencing for suspected genetic disease. A similar, but somewhat weaker, effect was seen for paternal age.

Main text

An increased incidence of sporadic genetic disorders that are thought to represent new mutations has been observed in children born to fathers of advanced age.^1,2,3,4 In contrast to the situation in females, mitotic division of pre-meiotic germ cell precursors continues throughout adulthood in males, with the male germ line estimated to replicate 150 times by age 20 years, 380 times by age 30 years, 610 times by age 40 years, and 840 times by age 50 years.^3,5 Because most mutations arise during DNA replication or cell division, the number of mutations accumulating in the male germ line is expected to increase with age.^3,5 Thus, the paternal age effect is attributed to an accumulation of DNA replication errors, largely thought to be as a result of single-base substitutions.^1,5 In contrast, mutations involving a larger number of nucleotides, such as large deletions or rearrangements, are not thought to exhibit a strong paternal age effect as they are less likely to be due to DNA replication errors.^1,4,5,6

Genome-wide (exome or genome) sequencing (GWS) has revolutionized the way that genetic disease is diagnosed. Specifically, GWS has greatly improved the ability of clinicians to recognize disease-causing genetic alterations in patients with suspected genetic diseases. Genotype-phenotype correlation is vital to confirm a suspected molecular diagnosis, and thus clinician assessment of the variants detected by GWS in light of the patient’s entire phenotype and clinical history is necessary to establish the diagnosis of a genetic disease.^1,7

Here, we studied 593 families of children with suspected genetic disease who underwent trio GWS and compared the ages of parents whose children were found to have de novo disease-causing variants with the ages of parents whose children were identified to have inherited variants or no diagnosis.¹

CAUSES and IMAGINe were studies of the clinical utility of trio GWS in children with suspected genetic disease approved by the University of British Columbia-BC Children’s and Women’s Hospital Research Ethics Board (CAUSES protocol H15-00092, IMAGINe protocol H16-02126). Consent was obtained from parents and assent from their affected children when possible.

The methods used for Illumina sequencing, bioinformatic analysis, alignment to hg19, and variant interpretation have been described in detail elsewhere.¹ GWS data were analyzed by a variant analyst who generated a short list containing the most promising candidate variants for the affected child or children in each family. The list of selected variants was reviewed by a multidisciplinary research team that included genome analysts, molecular laboratory geneticists, clinical geneticists, genetic counselors, and the referring specialist physician. The team assigned a diagnostic category (“definitely causal,” “probably causal,” “uncertain,” or “uninformative”) to each selected variant in each affected individual based on the patient’s full clinical picture (medical history, disease course over time, family history, physical examination findings, specialist consultations, imaging studies, and other laboratory test results) as well as all of the information and annotations available on the variant(s). Individuals who had variants that were deemed “definitely causal” or “probably causal” for the phenotype were considered to have been diagnosed with a genetic disease and were included in the parental age analyses. Individuals in whom no variant was found that appeared to be causal for the clinical features were classified as having “uninformative” or “negative” GWS.

Patients were assigned to the de novo group if they had a genetic disease diagnosed in association with a definitely or probably causal variant that was not present in either parent. All de novo X-linked mutations in these cohorts that occurred in females were dominant mutations, and both parental ages were considered. For de novo X-linked mutations that occurred in males, dominant or recessive, only maternal age was considered in the analyses. Paternity was confirmed in every case on the basis of the trio’s genomic findings at other loci. Patients were assigned to the inherited group if the disease-causing variant(s) found in the patient was also found in one or both parents. Patients who were diagnosed with two different genetic diseases (i.e., dual diagnoses) were counted twice: separately for each disease-causing variant, whether de novo or inherited. Affected identical twins were counted as one instance of a de novo or inherited disease-causing variant. Affected siblings found to both have inherited the same disease-causing variants from one affected parent (X-linked or autosomal dominant) or both heterozygous carrier parents (autosomal recessive) were counted as independent instances of inherited variants, with parental ages calculated for each child separately to reflect the chance of each child inheriting the reference allele, different variant, or a de novo variant in each conception.

Variants were classified as single-nucleotide variants (SNVs) if they represented a base pair change at a single nucleotide position. Variants were considered to be indels if they were insertions, deletions, or duplications of two or more but fewer than 50 base pairs. Changes were classified as structural variants (SVs) if they were deletions, duplications, or balanced rearrangements greater than 50 bp in length.

The age of the biological parents of each affected child was determined from the child’s date of birth and each parent’s reported birth date.

Normality of the parental ages was established through Normal QQ Plots and Shapiro-Wilk tests using R 4.2.2 software. One-tailed Student’s t tests were performed with R 4.2.2 software to examine the hypothesis that the mean paternal age for patients with disease-causing de novo variants is significantly greater than the mean paternal age of patients with inherited or no disease-causing variants. Differences in the mean ages of mothers and fathers were assessed by one-tailed Student’s t tests using R 4.2.2 software to test if the fathers were, on average, older than mothers, as expected. Differences in the mean paternal ages of the groups with disease-causing SNVs and indels or SNVs and SVs were calculated by two-tailed Student’s t tests using R 4.2.2 software, as there was no prior hypothesis for which group mean would be greater. There were only five individuals with de novo SVs; this sample size was considered to be too small for meaningful statistical analyses, so the group was omitted from comparisons with other variant types. The R function glm() in the stats package (version 4.2.2) was used to compute the logistic regression models by specifying the option family = binomial to fit logistic regression. Bayesian information criteria were then calculated using the BIC function in the stats package in R (version 4.2.2) for the maternal age model and the paternal age model to compare the strength of the association in the two models.

CAUSES included 415 families who underwent trio (usually affected child and both parents) or quad (affected child and affected sibling as well as both parents) exome sequencing and 85 families who underwent trio or quad genome sequencing. IMAGINe included 100 families who underwent trio or quad genome sequencing (Table 1). Seven families from CAUSES were excluded because the father’s birth date was not available. All de novo variants identified were associated with autosomal dominant or X-linked conditions.

Table 1.

Overview of the CAUSES and IMAGINe family members included in study

	n	Group	Total
Total families	–	–	593
Total variants included in study	–	–	342a
Total patients (probands and siblings) included in study	–	–	633
Patients with disease-causing variants	–	–	333a
Patients with de novo disease-causing variants	–	228	–
Patients with de novo disease-causing single-nucleotide variants	153	–	–
Patients with de novo disease-causing indels	70	–	–
Patients with de novo disease-causing structural variants	5	–	–
Patients with inherited disease-causing variants	–	105	–
Patients with no diagnosis	–	–	309

^aNine patients each had two variants that were included in the study. Each of these patients was counted twice in the analysis (i.e., as if each variant occurred in a different patient).

Before statistical analyses were performed, the normality of the distribution of both maternal and paternal ages for the entire dataset was established via quantile-quantile plots and Shapiro-Wilk tests (maternal age p = 0.35, paternal age p = 0.04) to ensure the validity of utilizing Student’s t tests (Figure S1). The mean age of fathers of patients with a genetic disease resulting from a de novo variant (35.09 years, SD = 6.08) was significantly greater than that of patients with inherited disease-causing variants (33.78, SD = 5.94), a difference of 1.31 years (95% CI, 0.06 to ∞; p = 0.04) (Table 2). Since all of the de novo disease-causing variants were autosomal or X-linked dominant, we also investigated this result in comparison with the paternal age of patients with dominant inherited variants and found that the mean paternal age for patients with de novo variants, although higher, was not significantly different than the paternal age for patients with inherited dominant variants (difference of means = 0.82 years, 95% CI, −0.88 to ∞; p = 0.21) (Table S1). The mean age of fathers of patients with de novo variants was also significantly greater (0.93 years, 95% CI, 0.06 to ∞; p = 0.04) than the mean paternal age of patients with no identified disease-causing variant (Table 2).

Table 2.

Comparison of the mean difference in paternal ages of patients identified to have de novo, inherited, or no disease-causing variants

Father’s age of patients with de novo disease-causing variants	Father’s age of patients with inherited disease-causing variants (n = 87, mean = 33.78 years, SD = 5.94)	Father’s age of patients with no diagnosis (n = 309, mean = 34.16 years, SD = 6.05)
All de novo variantsa (n = 225, mean = 35.09 years, SD = 6.08)	1.31 years (95% CI, 0.06 to ∞; p = 0.04)	0.93 years (95% CI, 0.06 to ∞; p = 0.04)
De novo SNVs (n = 150, mean = 34.35, SD = 5.51)	0.57 years (95% CI, −0.72 to ∞; p = 0.23)	0.19 years (95% CI, −0.74 to ∞; p = 0.36)
De novo indels (n = 70, mean = 36.33, SD = 6.70)	2.55 years (95% CI, 0.85 to ∞; p = 0.007)	2.17 years (95% CI, 0.72 to ∞; p = 0.007)

The difference between the indicated groups is shown, along with 95% CI of the difference and the p value obtained from the t test comparing the groups.

^aFive patients with de novo SVs are included in this total but were not statistically analyzed.

Interestingly, the mean paternal age for individuals with de novo disease-causing indels was significantly greater than the mean paternal age of patients with de novo disease-causing SNVs (1.98 years, 95% CI, 0.15 to 3.79; p = 0.03). The mean paternal age for patients with de novo disease-causing indels was also significantly greater than that of individuals with either an inherited disease-causing variant (2.55 years, 95% CI, 0.85 to ∞; p = 0.007) (Table 2), an inherited disease-causing indel (3.14 years, 95% CI, 0.90 to ∞; p = 0.01) (Table S1), or no diagnosis (2.17 years, 95% CI, 0.72 to ∞; p = 0.007) (Table 2).

Fathers in high-income countries are, on average, 2–4 years older than their female partners when their children are born.⁸ To assess the validity and robustness of our method, we compared the paternal and maternal ages of individuals with disease-causing variants in several groups. Fathers were about 3 years older than mothers in all comparisons, and this difference was statistically significant regardless of inheritance mechanism or de novo variant type (Table S2).

To lend further validity to our paternal age effect investigation, maternal ages were compared across all subgroups (Table 3). The mean maternal age of patients across all de novo variants ($\overline{\mathrm{x}}$= 31.86, SD = 5.16) was not significantly greater than that of patients with inherited variants (1.06 years, 95% CI, −0.16 to 2.29; p = 0.09) or no diagnosis (0.60 years, 95% CI, −0.29 to 1.50; p = 0.18). However, the mean maternal age of patients identified to have de novo indels ($\overline{\mathrm{x}}$= 33.34, SD = 5.34) was significantly greater than the mean maternal age of the inherited (2.54 years, 95% CI, 0.90 to 4.17; p = 0.003), no diagnosis (2.08 years, 95% CI, 0.67 to 3.49; p = 0.004), and de novo SNV subgroups (2.19 years, 95% CI, 0.69 to 3.68; p = 0.005) (Table 3).

Table 3.

Comparison of the mean difference in maternal ages of individuals identified to have de novo, inherited, or no disease-causing variants

Mother’s age of patients with de novo disease-causing variants	Mother’s age of patients with inherited disease-causing variants (n = 105, mean = 30.80, SD = 5.31)	Mother’s age of patients with no diagnosis (n = 309, mean = 31.26, SD = 5.27)
All de novo variantsa (n = 227, mean = 31.86, SD = 5.16)	1.06 years (95% CI, −0.16 to 2.29; p = 0.09)	0.60 years (95% CI, −0.29 to 1.50; p = 0.18)
De novo SNVs (n = 153, mean = 31.15, SD = 4.88)	0.35 years (95% CI, −0.93 to 1.64; p = 0.59)	−0.11 years (95% CI, −1.08 to 0.87; p = 0.83)
De novo indels (n = 69, mean = 33.34, SD = 5.34)	2.54 years (95% CI, 0.90 to 4.17; p = 0.003)	2.08 years (95% CI, 0.67 to 3.49; p = 0.004)

The difference between the indicated groups is shown, along with 95% CI of the difference and the p value obtained from the t test comparing the groups.

^aFive patients with de novo SVs are included in this total but were not statistically analyzed.

To explore this result further, the correlation of maternal and paternal ages was calculated for both the de novo SNV and indel variant subgroups (Figure S2). As expected, the ages of the mothers and fathers were strongly correlated for both children with disease-causing SNVs (R = 0.63) and children with disease-causing indels (R = 0.74). To determine if maternal or paternal age was more strongly associated with the child having a de novo disease-causing indel rather than another kind of de novo disease-causing variant, two separate logistic regression models were constructed with child diagnosis of a de novo indel variant as the dependent variable and either maternal or paternal age as the independent variable. The maternal age model has a slightly lower Bayesian information criterion (277.20) (Table S3) than the paternal age model (280.78) (Table S4), providing some evidence that it describes the association with de novo indels better.⁹

The average age of fathers is greater than expected among children clinically recognized to have sporadic occurrence of several different highly penetrant autosomal dominant disorders.^10,11 Additionally, an increase in the incidence of sporadic achondroplasia, a highly penetrant autosomal dominant condition, has been observed among children born to older fathers.^1,10,11 Myhre syndrome, an autosomal dominant condition caused by de novo SMAD4 mutations in the paternal germline, also exhibits a strong paternal age effect.¹² Similar paternal age effects have also been found in individuals with autism spectrum disorder or schizophrenia,^13,14 conditions that have been associated with increased genome-wide frequencies of de novo mutations.^15,16,17

Genome sequencing has enabled the recognition of de novo disease-causing variants in patients with a wide range of genetic conditions, even when the associated phenotype is variable or incompletely penetrant. GWS can be used to determine whether a clinically unaffected parent of a child with a disease-causing genetic variant also carries that variant (incomplete penetrance) or, alternatively, whether the child’s variant did, in fact, arise de novo. This approach has been used to demonstrate a paternal age effect among individuals with molecularly defined de novo variants across the genome as a whole without consideration of whether the variant produces a phenotypic effect.^18,19,20 In our study, we used diagnostic trio exome or genome sequencing to test for a paternal age effect in a wide variety of dominant disease-causing variants identified in children with suspected genetic conditions. We were unable to distinguish mosaic germline mutations in a parent from de novo somatic mutations very early in development of the embryo. Consequently, mosaic mutations in a parent’s germline would have been counted more than once if they were inherited in siblings, introducing potential biases in statistical analyses.

We assessed the parental ages of children with or without molecularly confirmed de novo variants that caused a genetic disease. Our finding that the average age of the fathers was significantly greater than the average age of the mothers across the subgroups analyzed is consistent with the older age of fathers than mothers at childbirth observed in all human cultures studied and provides a proof of principle for our method.⁸

To assess the paternal age effect in children with molecularly characterized disease-causing genetic variants, we compared the mean paternal age of individuals with de novo variants identified by the CAUSES or IMAGINe multidisciplinary research teams to the mean paternal age of patients with monogenic disease-causing genetic variants inherited from one or both parents. The de novo variant group had a significantly higher mean paternal age than the inherited variant group, a finding consistent with a paternal age effect for new dominant variants. The mean paternal age in the de novo variant group was also higher than that of patients in whom genetic testing was uninformative.

De novo mutations that cause high-penetrance autosomal dominant disorders in a child may be of several different types (e.g., SNVs, indels, or SVs). Analyzing the paternal age effect by mutation class offers the possibility of gaining insight into the underlying mutational mechanism because SNVs and indels usually arise during DNA replication—SNVs by substitution and indels from DNA polymerase slippage, while SVs arise through double-strand breakage and repair errors, often related to local genomic architecture.^21,22 Our sample of children with de novo disease-causing variants was sufficiently large to permit demonstrating that the mean paternal age of patients with de novo disease-causing indels was significantly greater than that of patients with de novo disease-causing SNVs (Table 1), suggesting that DNA polymerase slippage may be an especially important mechanism of mutation underlying the paternal age effect. Our study only included five patients with de novo SVs, so we could not perform meaningful comparisons to this subgroup. Paternal age was only weakly associated with the occurrence of de novo indels in the genome sequencing study reported by Jónsson et al.,¹⁹ but indels comprised only 6.8% of high-quality de novo mutations observed without reference to phenotype in this study. The mutation spectrum of protein-coding variants differs from that of variants that occur in non-coding genomic regions, especially those with repetitive sequences.²⁰

The paternal age effect is thought to result from the accumulation of mutations in the male germ line as it undergoes repeated mitotic renewal throughout life.^3,5 If this is true, all of the excess mutations in the children of older fathers should be found on a chromosome the child inherited from their father, i.e., the paternal haplotype. Two large genome sequencing studies in which parental origins of de novo mutations could be determined both found that most of these mutations appeared to have occurred in the paternal germ line and that this effect increased with paternal age.^19,20,23,24 As only whole exome rather than whole genome sequencing was performed in most of these cases, we were unable to detect potential disease-causing variants outside of protein-coding regions.

The diagnostic exome and genome sequencing in our study was done with Illumina short-read technology, and we were unable to determine whether the new variants we observed occurred on the maternal or paternal haplotype. Long-read sequencing has been used to phase parental haplotypes and could be used in future studies to determine the parental origin of de novo variants in the children of fathers or mothers of various ages.^25,26 Although additional studies are needed to establish the parent of origin of the mutations we observed, our current findings support the hypothesis that the paternal age effect can be attributed to DNA replication errors or reduced repair of such errors^27,28 in the male germ line as it continues to replicate throughout adulthood.

We also found that advanced maternal age was associated with de novo indels but not SNVs and that the occurrence of de novo disease-causing indels appears to be more strongly associated with maternal than paternal age. These observations require replication in future studies, but if the maternal age effect we observed for de novo disease-causing indels is real, it must arise by an unusual mechanism. Most indels are thought to result from polymerase slippage during DNA replication,²¹ but DNA replication in oocyte precursors only occurs during the embryonic and early fetal periods. It therefore seems unlikely that maternal age could influence the occurrence of indels by a replication-based mechanism.

Repair of DNA double-strand breaks by homologous recombination is essential to the successful completion of meiosis I, but this process is unlikely to produce indels because the repair is precise at the site of breakage. In contrast, repair of double-strand breaks by non-homologous end joining (NHEJ) is often imprecise, with the deletion or addition of one or a few nucleotides at the site of repair producing an indel. NHEJ occurs in oocytes and becomes an increasingly important mechanism of repairing double-strand breaks in the later part of meiosis I and in meiosis II,²⁹ which do not occur until just before ovulation or shortly after fertilization, respectively. Inefficient repair of non-replicative DNA damage with increasing age has been suggested as a source of de novo mutations in both spermatogenesis and oogenesis,^26,27,30,31 and, along with the accumulation of double-strand breaks in primary oocytes during their prolonged diplotene arrest, might account for the maternal age effect observed for indels in our study.

A maternal age effect was not seen for de novo indels that arose in the mother’s genome in the Jónsson et al. study,¹⁹ but this difference could reflect that fact only 6.8% of the de novo mutations found without consideration of phenotypic effect by Jónsson et al. were indels. In contrast, 31% of de novo mutations in our study were indels, and all were disease causing. In other words, our study group may have been “selected” for damaging indels. Long-read genome sequencing studies that permit determining the parent of origin of de novo mutations^32,33,34,35 should provide more insight into the mechanisms underlying the parental age effects we observed with de novo disease-causing indels.

Data and code availability

The datasets supporting the current study have not been deposited in a public repository as they contain unpublished study data but may be available from the corresponding author on request.

Acknowledgments

CAUSES Study investigators include Shelin Adam, Nick Dragojlovic, Christèle du Souich, Alison M. Elliott, Anna Lehman, Larry Lynd, Jill Mwenifumbo, Tanya Nelson, Clara van Karnebeek, and Jan M. Friedman. IMAGINe Study investigators are Patricia Birch, Madeline Couse, Colleen Guimond, Anna Lehman, Jill Mwenifumbo, Clara van Karnebeek, and Jan M. Friedman. We are truly grateful to the families who participated in the IMAGINe and CAUSES studies. The CAUSES project was funded by Mining for Miracles (BC Children’s Hospital Foundation) and Genome British Columbia, with support from the Provincial Health Services Authority and British Columbia Women's Hospital. The IMAGINe Study was funded by the Canadian Institutes of Health Research through the CHILD-BRIGHT Network, with support of the British Columbia Children’s Hospital Foundation and the Michael Smith Foundation for Health Research.

Author contributions

E.B. and J.M.F. contributed to the conception and study design. All authors contributed to data acquisition. E.B. and J.M.F. drafted the manuscript and all authors contributed to critical review and editing of the manuscript.

Declaration of interests

The authors declare no competing interests.