De novo mutations (DNM) are an important source of rare variants and are increasingly being linked to the development of many diseases. Recently, the paternal age effect has been the focus of a number of studies that attempt to explain the observation that increasing paternal age increases the risk for a number of diseases. Using disease-free familial quartets we show that there is a strong positive correlation between paternal age and germline DNM in healthy subjects. We also observed that germline CNVs do not follow the same trend, suggesting a different mechanism. Finally, we observed that DNM were not evenly distributed across the genome, which adds support to the existence of DNM hotspots.
Citation: Girard SL, Bourassa CV, Lemieux Perreault L-P, Legault M-A, Barhdadi A, Ambalavanan A, et al. (2016) Paternal Age Explains a Major Portion of De Novo Germline Mutation Rate Variability in Healthy Individuals. PLoS ONE 11(10): e0164212. https://doi.org/10.1371/journal.pone.0164212
Editor: Klaus Brusgaard, Odense University Hospital, DENMARK
Received: May 9, 2016; Accepted: September 21, 2016; Published: October 10, 2016
Copyright: © 2016 Girard et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Data are openly available on ENA using the Accession number PRJEB8308.
Funding: Guy A. Rouleau is grateful for the support received through his positions as Canada Research Chair in Genetics of the Nervous System and Jeanne-et-J.-Louis-Levesque Chair for the Genetics of Brain Diseases. Simon L. Girard is grateful for the financial support received from Fonds de Recherche Québec – Santé. This work was also supported by Calcul Québec / Calcul Canada. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Untill recently, little was known about the global prevalence of de novo mutations (DNM), and the factors that influence their rate of occurence. Some studies suggested that different factors such as pollution, tobacco smoke or magnetic field could impact the global mutation rate [1–3], but it was not possible to assess these on a genome-wide scale, or to look at different classes of DNM. New sequencing technologies now allow the interrogation of the full genome by sequencing. Using these technologies, it was recently demonstrated that paternal age effect (PAE) modulates the DNM rate in patients with psychiatric disorders[4, 5]. Here, we replicate these finding in ten disease free twin quartets. We show that rate of germline de novo single nucleotide variants (SNV) and indels but not CNV are associated with parental age. We also confirm the existence of DNM hotspots, suggesting new mechanisms for the occurrence of new mutations.
All patients gave informed consent in written form to participate in the Quebec Study of Newborn Twins. Ethic boards from the Centre de Recherche du CHUM, from the Université Laval and from the Montreal Neurological Institute approved this study. Amongst the cohort of the QSNT, we selected five families with young father at conception (range from 20–25) and five families with older father at conception (range from 40–47). The cohort did not allow us to select families with similar maternal age at conception for the two groups (range of 19–25 for the young father group and range of 30–36 for the older father group) (Table 1).
DNA from the 40 individuals was extracted directly from blood provided by the QSNT. Libraries were constructed using standard Illumina protocols. Sequencing was done on an Illumina HiSeq 2500 at Illumina sequencing facilities. Paired-end mode was used and the median fragment length was found to be ~300 bp for each individuals. Sequencing was performed until every sample reached an average coverage of 30x. For DNM rate calculation, only genome position with sufficient coverage (>20x) in both twins were kept for analysis (Table 1).
Log R Ratio (LRR) and B Allele Frequencies (BAF) were extracted by GenomeStudio (version 2011.1) from Illumina HumanOmni2.5 BeadChips using default parameters. CNVs were identified using QuantiSNP (version 2.3) with local GC correction (hg19) and default parameters. For de novo mutations, putative CNV regions were created using the twins’ calls of any length but with a logarithm of the Bayes factor higher than 10. Only regions with higher than 25% reciprocal overlap found in both the twins were kept for further analysis. From this set, regions showing any sign of presence in one of the two parents (raw calls) or with an overlap with any regions found in the Database of Genomic Variants (DGV, version 10, November 2010) were discarded.
SNV and indels calling
We received sequences from Illumina in the form of alignment (bam) files and variant (vcf) files. In order to have consistent metrics we applied our own bioinformatics pipeline and used the same genome reference (GRCh37), so we can compare with previous runs. We extracted the reads and created pools of 20M reads using Picard SAMtoFastq in order to optimise the use of our computing farm. Each of these pools were aligned using a modified version of the Burrows-Wheeler Aligner (bwa version 0.6.2-r126-tpx) with threading enabled). The options were 'bwa aln -t 12 -q 5' and 'bwa sampe -t 12'. Next we recombined the pools in a single alignment (bam) file using Picard MergeSamFiles. The realignment method took roughly 36 hours walltime to complete.
We used the realigned bam file in GATK 1.6 to process the data with indel realignments, read duplicates removal and quality score recalibration. We made use of the scatter-gather mechanism in GATK to accelerate the process; walltime for these operations was around 40 hours. Once all 40 samples were processed, we generated QC metrics using GATK DepthOfCoverage and custom scripts (https://bitbucket.org/mugqic/mugqic_pipelines).
We used GATK UnifiedGenotyper to call SNVs and Indels in a run containing all samples, and splitting by chromosome to accelerate the process on the cluster. Finally, the variants were filtered and ordered using GATK Variant Recalibration, which assigns a validity score to each variant using a training algorithm. For this purpose we strictly followed the procedure from the Broad Institute (https://www.broadinstitute.org/gatk/guide/best-practices.php).
Once the variant calls were produced, these were compared to the SNP-chip calls; all samples had a variant concordance higher than 99.5%.
WGS identification of SV
In order to achieve high-confidence structural variants calls for our samples, two different algorithms were used. The first one, CNVer v0.8.1, uses information from the paired-end mappings and from the depth of coverage in a given region to call CNVs (insertions and deletions). Prior to the mapping, the FastqMcf v1.1.2 (https://code.google.com/p/ea-utils/wiki/FastqMcf) utility was used to detect and remove fragments from sequencing adapters and primers and to remove poor quality bases at the end of reads. The Bowtie v0.12.9 aligner was then used with the “-v -2 –a–m 600—best—strata” options, as specified by the CNVer authors. To reduce the number of false positives, CNVs called by CNVer were merged when the distance to their neighbor was fewer than 5 kilobases.
The second algorithm, BreakDancer, uses paired-end mappings signatures to identify different types of structural variants, including insertions, deletions, inversions and translocations. The reads were processed as described in the SNV and indels calling section before using BreakDancerMax (cpp implementation, version 1.1.2) to call structural variants. Once again, CNVs were merged with a 5 kilobase proximity threshold.
The germline DNM detection and exclusion from this dataset required a maximum overlap between every twin and his parents of 1% and a minimum overlap to the other twin of 90%.
Sanger sequencing validation
Primers were designed using in-house scripts and the Batch Primer 3 web tool (http://probes.pw.usda.gov/batchprimer3/) using enough flanking sequence to have a good coverage of selected variants. PCR was performed using the AmpliTaq Gold DNA Polymerase (Applied Biosystems) according to the manufacturer’s instructions. To visualize DNA fragments, a small fraction of the PCR product was loaded on a 1.5% agarose gel containing ethidium bromide. PCR products were sequenced at the Genome Quebec Innovation Centre using a 3730XL DNAnalyzer (Applied Biosystems), and SeqMan from the DNAstar suite was used for mutation detection analysis.
In order to evaluate the effect of paternal age on the rate of DNM in healthy subjects, we selected ten families consisting of two monozygotic twins and their parents. All individuals were recruited by the Quebec Study of Newborn Twins (QSNT) project. These families have been prospectively followed since birth and no severe health conditions were reported for the twins, now all around 15 years old. These families were selected according to extremes in paternal age at the time of conception. A group of 5 younger-aged fathers (mean = 23.37±SD 2.06) and 5 older-aged fathers (mean = 43.33±SD2.75) (Table 1) were sequenced. Paternal age at conception was significantly different between the two groups (t-test, p-value < 1.00∙10−5). Despite our efforts to select comparably aged mothers, there was also a small but statistically significant difference in maternal age at conception (t-test, p-value < 0.0065). Thus, we will be able to directly assess the effect of parental age effect on DNM variability and we will need to establish parental origin of mutations to directly implicate the paternal or maternal age.
Whole Genome Sequencing (WGS) was performed on all 40 family members from the 10 families. Mutation detection was performed using an in-house bioinformatics pipeline to optimize the detection of variants present in the twins against the variant profiles of the parents. All samples were genotyped using Illumina HumanOmni2.5 BeadChips and the SNP calls were used to calibrate and validate variant detection with the WGS bioinformatics pipeline. de novo CNVs were identified using data from the Illumina SNP-chips as well as from the WGS data using two different algorithms[9, 11]. Consistent DNMs in a twin pair were categorized as germline (i.e. inherited from gamete cells of one of the parents), while DNM found in a single twin were categorized as post-zygotic (i.e. mutations happening post-conception).
Overall, 889 germline mutations and 106 post-zygotic mutations were detected using these criteria in the 10 twin pairs (S1 Table). In order to test the validity of our in silico detection pipeline, we tested a subset of the DNM identified using Sanger Sequencing. The proportion of SVs and indels validated was based on the total proportion observed in the full dataset. A total of 91 germline variants as well as 62 post-zygotic variants were tested using Sanger sequencing in the corresponding quartet. Almost 88% of the germline SNV variants were validated while the indels validation rate was slightly better at 93.5%. Based on these results, we calculated a false positive rate for each quartet and we used these individual rates to correct the total number of DNM. For family quartets for which no false positive was identified, we corrected with the overall false positive rate of 10.9%. For the post-zygotic variants, only 9 of 62 variants were confirmed, resulting in a false positive rate higher than 85%. Thus, given that we would predict that we have detected only 15 post-zygotic DNM, it is impossible to arrive at any conclusions regarding post-zygotic DNM rates, except to say that they are in limited number when looking at DNA extracted from blood.
Assuming a single mutation event for all SNV and indels germline mutations and collapsing all twin pairs to a single individual, we calculated the germline de novo mutation rate to be 1.37∙10−8, which is higher than the previously reported rate of 1.1∙10−8.. However, when stratified by paternal age, the estimated DNM rate in the groups of younger and older fathers respectively were 1.01∙10−8 and 1.74∙10−8 (Rao-Scott chi-square p = 1.38∙10−12)(Table 1). We then looked at the CNV DNM rates. CNVs were identified using three different methods, one using the SNP genotyping arrays and two using algorithms that predict CNVs from the, one focussing on the depth of coverage information (CNVer) and another that uses read pairs mapped with unexpected distance or orientation (BreakDancer). For the three algorithms, we kept only CNVs that had an overlap >90% in the two twins and no overlap at all with the parents. As getting a consensus for CNV is not trivial and was not needed here, we tested for parental age effect on each separate dataset without building a consensus dataset. The results were all concordant and no association with parental age was seen for any method (Table 2)
We next sought to find if there was a specific association between parental sex and the variability in DNM rate. We found that there was a strong correlation between the germline DNM and paternal age (Poisson regression R2 = 0.78, p = 0.00066) as well as with maternal age (Poisson regression R2 = 0.65, p = 0.004), though the effect size is smaller for the later. Then, we tested for parental effect looking separately at single nucleotide variants (SNV) and small insertions and deletions (indels) (Fig 1). We found a strong correlation between paternal age and single nucleotide germline DNM variants (Spearman correlation R = 0.7695, p = 0.0092). This correlation was also observed in the mother, although not as strong (Spearman correlation R = 0.745, p = 0.013) suggesting that parental age accounts for an important portion of the variability in the DNM rate. A positive correlation was also found for indel germline variants for both paternal (Spearman correlation R = 0.7631, p = 0.01) and maternal age (Spearman correlation R = 0.702, p = 0.0237). The weaker association with indels could in part result from the lower power due to the comparatively less frequent occurrence. This observation is nonetheless in concordance with other studies[4, 5]. Interestingly, germline de novo CNVs show no association with parental age (Fig 2) (Student T test > 0.05 for all five comparisons). Although further validation is required, this would suggest that the mechanisms explaining the increasing rate of SNV DNM with increasing parental age do not apply to CNVs, which may have implications in our understanding of this phenomenon.
This figure shows the correlation between different SNV and indel germline DNM (left: germline SNV, right: germline indels) with parental age (Top: paternal age, bottom: maternal age). The X-axis represents the parental age at conception. The Y-axis represents the number of DNM mutations identified through Whole Genome Sequencing.
We used three different algorithms to detect CNV in our dataset. QuantiSNP was used for genotyping assays while CNVer and BreakDancer were used for WGS. Although the number varies according to which algorithm was used, no difference between young parental age group and old parental group can be detected.
In order to separate out the paternal and maternal age effects, we used individual WGS reads to assign parental origin of the germline DNM. Working only with germline SNVs, as they were more frequent than indels, we assigned parental origin according to co-transmission of DNM with polymorphisms on single sequence reads. Out of 765 DNM, we were able to trace parental origin for 57 DNM. Of those, 43 DNM were of paternal origin and 14 were of maternal origin; providing an estimated paternal:maternal ratio of ~ 3:1 in our dataset. When limiting the comparison to the younger paternal age group, we see 16 paternal germline DNM and 5 maternal germline DNM (ratio of ~ 3:1) whereas the older paternal group has 27 paternal germline DNM and 9 maternal germline DNM (ratio of ~ 3:1). These data suggest that DNM rates increase both with maternal and paternal age, though the paternal age explains a greater number of DNMs than maternal age.
Many groups have shown an association of de novo mutations (DNM) to neurodevelopmental disorders, notably intellectual disability (ID), autism spectrum disorder (ASD)[15–17] and schizophrenia[18–20]. At the time these studies were published, it was yet unclear whether the DNM rate was fixed across our species or if it was subject to variation. Interestingly, it was previously shown that paternal age at conception is positively correlated with risk of schizophrenia[21, 22].
In this study, we show that parental age, more specifically paternal age, explains a major portion of the variation in the germline DNM rate. Contrary to other studies that have reported similar findings, we have studied disease free families, and we have isolated the germline DNM rate from post-zygotic mutations with respect to parental age[4, 5]. In our study, fathers aged 40 or greater had a rate of germline DNM that was almost twice as high as that of fathers aged twenty years younger. We also observe a smaller, yet significant increase in the mutation rate associated with maternal age. Although we were not able to fully account for age-dependent sex-effect because of assortative mating present in our families, the origin of mutation suggests that older maternal age also leads to an increase in DNM. This conclusion is also supported by other studies that also report a maternal age effect on DNM rate[23–25]. The rates that we observed in this study are comparable to the rates reported previously[4, 5]. These findings do not appear to apply to CNVs. We have observed that the CNV rate does not vary with parental age. This is not a surprising finding as it is known that CNV and SNV have different mechanism or origin although the technical variation observed specifically for CNV in this study does not permit to fully exclude the link between CNV and parental age. It is thought that most CNVs in the human genome arise from non-allelic homologous recombination while a vast majority of the SNVs occur during DNA replication. The implications of these findings are important, as an increased DNM rate could impact a number of Mendelian and complex genetic disorders. A study design that includes the DNA of two parents and their children has proved a valuable approach for the investigation of DNM. The ability to assess DNM rates offers a valuable means to further characterize the impact of rare variants on diseases and traits and should be encouraged in the design of future studies where parents are available.
We thank the families involved in our study. Guy A. Rouleau is grateful for the support received through his positions as Canada Research Chair in Genetics of the Nervous System and Jeanne-et-J.-Louis-Levesque Chair for the Genetics of Brain Diseases. Simon L. Girard is grateful for the financial support received from Fonds de Recherche Québec–Santé. We are grateful to Calcul Québec / Calcul Canada for the time and resources.
- Conceptualization: SLG GAR.
- Data curation: SLG.
- Formal analysis: SLG AB.
- Funding acquisition: PAD GAR.
- Investigation: SLG MAL.
- Methodology: SLG PAD MB (thirteenth author) MPD GAR.
- Project administration: SLG.
- Resources: MB (seventh author) FV GD RET MB (thirteenth author).
- Software: SLG LPLP MAL AA.
- Supervision: SLG GAR.
- Validation: CVB AN.
- Visualization: SLG.
- Writing – original draft: SLG.
- Writing – review & editing: SLG CVB LPLP MAL AB AA MB (seventh author) FV AN GD RET PAD MB (thirteenth author) MPD GAR.
- 1. Antipenko Ye N, Kogut NN. The experience of mutation rate quantitative evaluation in connection with environmental pollution (based on studies of congenital anomalies in human populations). Mutation research. 1993;289(2):145–55. pmid:7690882.
- 2. Ikehata M, Koana T, Suzuki Y, Shimizu H, Nakagawa M. Mutagenicity and co-mutagenicity of static magnetic fields detected by bacterial mutation assay. Mutation research. 1999;427(2):147–56. pmid:10393268.
- 3. Pleasance ED, Stephens PJ, O'Meara S, McBride DJ, Meynert A, Jones D, et al. A small-cell lung cancer genome with complex signatures of tobacco exposure. Nature. 2010;463(7278):184–90. pmid:20016488; PubMed Central PMCID: PMC2880489.
- 4. Michaelson JJ, Shi Y, Gujral M, Zheng H, Malhotra D, Jin X, et al. Whole-genome sequencing in autism identifies hot spots for de novo germline mutation. Cell. 2012;151(7):1431–42. pmid:23260136.
- 5. Kong A, Frigge ML, Masson G, Besenbacher S, Sulem P, Magnusson G, et al. Rate of de novo mutations and the importance of father's age to disease risk. Nature. 2012;488(7412):471–5. pmid:22914163.
- 6. Colella S, Yau C, Taylor JM, Mirza G, Butler H, Clouston P, et al. QuantiSNP: an Objective Bayes Hidden-Markov Model to detect and accurately map copy number variation using SNP genotyping data. Nucleic acids research. 2007;35(6):2013–25. pmid:17341461; PubMed Central PMCID: PMC1874617.
- 7. Li H, Durbin R. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics. 2009;25:1754–60. Epub 2009/05/20. btp324 [pii] pmid:19451168.
- 8. McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 2010;20(9):1297–303. pmid:20644199; PubMed Central PMCID: PMCPMC2928508.
- 9. Medvedev P, Fiume M, Dzamba M, Smith T, Brudno M. Detecting copy number variation with mated short reads. Genome Res. 2010;20(11):1613–22. pmid:20805290; PubMed Central PMCID: PMC2963824.
- 10. Langmead B, Trapnell C, Pop M, Salzberg SL. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 2009;10(3):R25. pmid:19261174; PubMed Central PMCID: PMC2690996.
- 11. Chen K, Wallis JW, McLellan MD, Larson DE, Kalicki JM, Pohl CS, et al. BreakDancer: an algorithm for high-resolution mapping of genomic structural variation. Nat Methods. 2009;6(9):677–81. pmid:19668202; PubMed Central PMCID: PMC3661775.
- 12. Boivin M, Brendgen M, Dionne G, Dubois L, Perusse D, Robaey P, et al. The Quebec Newborn Twin Study Into Adolescence: 15 Years Later. Twin research and human genetics: the official journal of the International Society for Twin Studies. 2012:1–6. pmid:23200437.
- 13. Genomes Project C, Abecasis GR, Altshuler D, Auton A, Brooks LD, Durbin RM, et al. A map of human genome variation from population-scale sequencing. Nature. 2010;467(7319):1061–73. pmid:20981092; PubMed Central PMCID: PMC3042601.
- 14. de Ligt J, Willemsen MH, van Bon BW, Kleefstra T, Yntema HG, Kroes T, et al. Diagnostic exome sequencing in persons with severe intellectual disability. The New England journal of medicine. 2012;367(20):1921–9. Epub 2012/10/05. pmid:23033978.
- 15. Neale BM, Kou Y, Liu L, Ma'ayan A, Samocha KE, Sabo A, et al. Patterns and rates of exonic de novo mutations in autism spectrum disorders. Nature. 2012;485(7397):242–5. pmid:22495311.
- 16. O'Roak BJ, Vives L, Girirajan S, Karakoc E, Krumm N, Coe BP, et al. Sporadic autism exomes reveal a highly interconnected protein network of de novo mutations. Nature. 2012;485(7397):246–50. pmid:22495309; PubMed Central PMCID: PMC3350576.
- 17. Sanders SJ, Murtha MT, Gupta AR, Murdoch JD, Raubeson MJ, Willsey AJ, et al. De novo mutations revealed by whole-exome sequencing are strongly associated with autism. Nature. 2012;485(7397):237–41. pmid:22495306.
- 18. Girard SL, Gauthier J, Noreau A, Xiong L, Zhou S, Jouan L, et al. Increased exonic de novo mutation rate in individuals with schizophrenia. Nature genetics. 2011. pmid:21743468.
- 19. Xu B, Ionita-Laza I, Roos JL, Boone B, Woodrick S, Sun Y, et al. De novo gene mutations highlight patterns of genetic and neural complexity in schizophrenia. Nat Genet. 2012;44(12):1365–9. pmid:23042115.
- 20. Xu B, Roos JL, Dexheimer P, Boone B, Plummer B, Levy S, et al. Exome sequencing supports a de novo mutational paradigm for schizophrenia. Nature Genetics. 2011;advance on. pmid:21822266
- 21. Croen LA, Najjar DV, Fireman B, Grether JK. Maternal and paternal age and risk of autism spectrum disorders. Archives of pediatrics & adolescent medicine. 2007;161(4):334–40. pmid:17404129.
- 22. Malaspina D. Paternal factors and schizophrenia risk: de novo mutations and imprinting. Schizophrenia bulletin. 2001;27(3):379–93. pmid:11596842; PubMed Central PMCID: PMC2993009.
- 23. Wong WS, Solomon BD, Bodian DL, Kothiyal P, Eley G, Huddleston KC, et al. New observations on maternal age effect on germline de novo mutations. Nat Commun. 2016;7:10486. pmid:26781218; PubMed Central PMCID: PMCPMC4735694.
- 24. McRae JF, Clayton S, Fitzgerald TW, Kaplanis J, Prigmore E, Rajan D, et al. Prevalence, phenotype and architecture of developmental disorders caused by de novo mutation. bioRxiv. 2016.
- 25. Goldmann JM, Wong WS, Pinelli M, Farrah T, Bodian D, Stittrich AB, et al. Parent-of-origin-specific signatures of de novo mutations. Nat Genet. 2016;48(8):935–9. pmid:27322544.
- 26. Conrad DF, Pinto D, Redon R, Feuk L, Gokcumen O, Zhang Y, et al. Origins and functional impact of copy number variation in the human genome. Nature. 2010;464(7289):704–12. Epub 2009/10/09. pmid:19812545; PubMed Central PMCID: PMC3330748.