Increased rate ofde novosingle nucleotide mutation in house mice born through assisted reproduction

Approximately 2.6% of live births in the United States are conceived using assisted reproductive technologies (ARTs). While some ART procedures, including in vitro fertilization (IVF) and intracytoplasmic sperm injection, are known to alter the epigenetic landscape of early embryonic development, their impact on DNA sequence stability is unclear. Here, we leverage the strengths of the laboratory mouse model system to investigate whether a standard ART regimen—ovarian hyperstimulation, gamete isolation, IVF, embryo culture, and embryo transfer—affects genome stability. Age-matched cohorts of ART-derived and naturally conceived C57BL/6J inbred mice were reared in a controlled setting and whole genome sequenced to ~50x coverage. Using a rigorous pipeline for de novo single nucleotide variant (dnSNV) discovery, we observe a ~30% increase in the dnSNV rate in ART-compared to naturally-conceived mice. Analysis of the dnSNV mutation spectrum identified signature contributions related to germline DNA repair activity, affirming expectations and evidencing the quality of our dnSNV calls. We observed no enrichment of dnSNVs in specific genomic contexts, suggesting that the observed rate increase in ART-derived mice is a general genome-wide phenomenon. Similarly, we show that the developmental timing of dnSNVs is similar in ART- and natural-born cohorts. Together, our findings show that ART is moderately mutagenic in house mice and motivate future work to define the precise procedure(s) associated with this increased mutational vulnerability. While we caution that our findings cannot be immediately translated to humans, they nonetheless emphasize a pressing need for investigations on the potential mutagenicity of ART in our species.

Assisted Reproductive Technology, in vitro fertilization, mutation rate, house mouse, mutation spectrum Classification: Biological Sciences, Genetics SIGNIFICANCE STATEMENT This study investigates whether assisted reproductive technologies (ARTs) increase the risk of inherited genetic mutations in offspring. Using a well-controlled mouse model system, we compared the de novo mutation burden in genomes of mice conceived through ART to a naturally conceived cohort. We find a ~30% increase in new DNA mutations in ART-conceived mice, suggesting that ART procedures have a genome destabilizing effect. This increase in mutation rate appears to be uniform across the genome, rather than attributable to specific genomic contexts. While we caution against the direct translation of our findings to humans, our work nonetheless highlights the need for further research into the genetic safety of ART in people.

Worldwide, approximately 1,000,000 babies are born by assisted reproductive technologies (ART) each year (1), accounting for over 2.5% of live-births in the United States (2) and many other Western countries (3, 4). This number has been steadily increasing over the last decade (5) and is expected to continue to increase in the face of rising infertility rates (6) and increases in the age at reproduction (7). The human health consequences of an increased reliance on ARTs are potentially substantial. Prior work has shown that various ART procedures are associated with altered methylation patterns in both model organisms (8–11) and humans (12–17), with downstream consequences for the expression of key genes required for totipotency and early development (18–22). These molecular changes are associated with increased rates of imprinting related disorders (23, 24), and may contribute to higher rates of adverse pregnancy, perinatal, neonatal, and long-term health outcomes in ART-derived offspring (25–28). While it is well-established that various ARTs can perturb the dynamics of epigenetic reprogramming in the pre-implantation embryo, it is not clear whether ARTs influence the integrity and stable transmission of the underlying DNA sequence itself. Prior investigations exploring this possibility have reached contradictory findings. Some studies have concluded that ARTs are associated with an increased burden of structural mutations and aneuploidy (29–31), while others have reported no significant increase in large-scale structural rearrangements in ART-compared to naturally-conceived offspring (32). Similarly, whereas a prior study in mice found no evidence for an increased point mutation rate in animals born via in vitro fertilization (IVF) (33), a recent retrospective comparative analysis in humans uncovered a significantly higher de novo point mutation rate in IVF-born children (34). However, the underlying etiology of sub- or in-fertility may predispose patients seeking ART to elevated rates of germline genome instability (35), such that any observed increase in the mutation burden associated with ARTs is potentially driven by ascertainment bias. Despite conflicting earlier findings, there are multiple compelling lines of evidence to support the hypothesis that ARTs may be intrinsically mutagenic. First, local epigenetic features, including methylation at CpG dinucleotides and H3K9me3 marks, directly shape local mutation rates (36, 37). Thus, the genome-wide epigenetic dysregulation observed in ART-derived embryos could secondarily alter the genomic distribution and rate of new mutations compared to natural conceptuses. Second, the first few early embryonic cell divisions are especially error-prone and uniquely vulnerable to the accumulation of large-scale structural mutations and aneuploidies (38, 39). Exposure to stressors unique to the in vitro environment may exacerbate genomic instability at this critical time point in early development. For example, ART-derived embryos are often cultured at oxygen levels that do not precisely replicate those encountered in the maternal reproductive tract (40). Importantly, exposure to either hyperoxic or hypoxic culture conditions can elicit DNA damage and promote genomic instability (41, 42). Additionally, compounds such as glucocorticoids present in culture media could also perturb the early epigenetic landscape of the developing embryo, leading to widespread regulatory and cellular metabolic changes with downstream implications for genome integrity (43). Here, we harness the efficiency of assisted reproductive technologies in the laboratory mouse to rigorously test whether a standard regimen of ARTs consisting of (1) ovarian hyperstimulation via administration of exogenous hormones, (2) gamete isolation and culture, (3) in vitro fertilization (IVF), (4) embryo culture, and (5) embryo transplantation into a receptive uterus influences the stable transmission of the mammalian genome. We use whole genome sequencing to identify de novo mutations in cohorts of age-matched ART-derived and natural-born mice. Our reliance on a single inbred strain mouse model allows us to rigorously control for potential genetic background effects and environmental exposures on mutation rates, eliminating these confounding factors from our analysis and overcoming a major limitation of retrospective clinical studies. We find a significant increase in the rate of de novo single nucleotide variants (dnSNVs) in ART-born mice, suggesting that this standard ART protocol adversely impacts genome integrity.

De novo single nucleotide variant discovery in naturally- and ART-conceived mice We aimed to test whether a standard ART protocol affects the overall incidence of dnSNVs accrued in gametes and/or during early embryonic development in a mouse model. Specifically, we focus on the sequential procedures of ovarian hyperstimulation by exogenous hormone injection, oocyte collection, sperm isolation, gamete culture, in vitro fertilization, embryo culture to the 2-cell stage (~1 day post fertilization), and embryo transfer to a pseudo-pregnant host dam. This series of ART procedures is routinely employed in the course of mouse husbandry for strain rederivation and colony expansion, and parallels protocols used in human fertility clinics with some exceptions. Notably, human embryos are typically cultured to the blastocyst stage (~5 days post fertilization), and in most cases, undergo a freeze-thaw cycle prior to transfer. We generated natural- and ART-derived cohorts of age-matched C57BL/6J mice derived from a common G0 founder pair (Figure 1a). Two G1 females were mated to a single male sibling, yielding two litters of naturally conceived mice. To obtain a matched ART-derived cohort, oocytes from two hormonally super-ovulated G1 females were harvested and in vitro fertilized with sperm from a second G1 male littermate. Sixteen 2-cell embryos derived from each of the two females were then transferred to the uteri of two pseudo-pregnant C57BL/6J recipient dams and reared to term. A total of 16 naturally born and 12 ART-conceived G2 progeny were whole genome sequenced to ~50–72x coverage using two independent sequencing libraries (Supplemental Table 1; Supplementary Figure 1). The six G1 parents and two G0 pedigree founders were additionally sequenced to enable discrimination between dnSNVs and segregating variants within our pedigree (>75x coverage; 2 libraries per sample; Supplemental Table 1; Supplemental Figure 2). By bottlenecking the breeders used in this experiment through a single G0 founder pair and using a common sire for each treatment group, we minimize the number of background segregating mutations present in our G2 cohorts, limiting this potential source of false positive dnSNVs. Simulations based on empirical coverage estimates indicate that our study design has 80% power to detect a ~30% increase in the dnSNV rate between the ART- and naturally-conceived cohorts, assuming a baseline mutation rate equivalent to that previously reported for C57BL/6J mice (46) (Supplemental Figure 3). Sequencing reads from each sample were aligned to the GRCm39 mouse reference genome, followed by SNV calling, dnSNV identification, and application of a post hoc filtering pipeline informed by tailored simulations (44) (see Methods). Our strategic use of the reference strain, C57BL/6J, effectively eliminates reference genome biases to maximize the accuracy of SNV calling and dnSNV discovery. On average, we identified 17 high-confidence autosomal dnSNVs in samples from the natural mating cohort (range = 11–25), corresponding to an estimated per base mutation rate of 3.88 × 10−9 per generation (per sample range: 2.47×10−9 to 5.61 × 10−9). This mutation rate estimate matches previously reported mutation rates for house mice (45–48). In contrast, offspring generated via ART exhibit a significantly elevated dnSNV burden, with an average of 22 autosomal dnSNVs per individual and an average dnSNV mutation rate of 4.95 × 10−9 (range: 13–29; range of mutation rate per sample: 2.91 × 10−9 to 6.50 × 10−9; one-tailed Wilcoxon rank-sum test, P-value = 0.038; Figure 1B). Our finding of a significant increase in dnSNVs in ART-derived samples suggests that the tested ART protocol has adverse impacts on genome stability in gametes, zygotes, or early-stage embryos. However, systematic differences in sequencing data quality could lead to differences in the number of false positive (or false negative) dnSNV calls between cohorts. Assuredly, sequencing data quality is universally high across our G2 samples (Supplementary Figure 1), dnSNV calls are supported by similar variant quality metrics in both cohorts (Supplementary Figure 4), and our findings are recapitulated using sequencing data from individual replicate libraries from each sample (see Methods; Supplementary Tables 3 and 4; Supplementary Figure 5). We conclude that the observed differences in mutation burden between ART- and naturally-conceived offspring are not attributable to cohort-specific technical artifacts. No difference in mutation spectra between naturally conceived and ART-born mice Exposure to specific environmental mutagens can impact the spectrum of mutations that accumulate in somatic tissues (49, 50), leading us to wonder whether the altered hormonal milieu and in vitro environment encountered during the generation of ART-born mice modifies the prevalence of specific types of mutations compared to those recovered in naturally-conceived animals. Both breeding cohorts exhibit qualitatively similar transition and transversion fractions (Wilcoxon rank sum, P-value > 0.05; Figure 1C), with estimates closely approximating those from prior studies of germline mutations in house mice (46, 51). Furthermore, the two cohorts show no difference in the relative frequency of individual mutation types (One-sided Wilcoxon rank sum test, P > 0.05) or the overall mutation spectrum (G-test, P-value > 0.05; Figure 1D). Further partitioning of dnSNVs based on their flanking nucleotide contexts reveals a significant increase in the C[C>A]A trinucleotide mutation fraction in ART-compared to naturally born samples (modified Chi-square P-value=0.019, Figure 2c). However, we note that resulting trinucleotide mutation count matrix is sparse, and we find no difference in the proportions of dnSNVs ascribable to defined mutational signatures between our two cohorts (Figure 2b). Overall, these findings reveal broad similarities in the mutation spectrum between ART- and natural-born mice, with the caveat that our analysis is likely underpowered to find differences. Evaluating the genomic landscape of dnSNVs in Natural- and ART-born mice We next sought to understand whether the distribution of new mutations differs between our two cohorts with respect to various genomic features, including GC content, functional annotations, and repetitive elements. dnSNVs ascertained in both cohorts are uniformly distributed across the genome (Poisson test applied to mutation counts in 10Mb bins, P > 0.3; Supplementary Figure 6A) and arise in regions of similar GC-content (Wilcoxon rank sum, P = 0.892, Supplementary Figure 6B). Likewise, we detect no significant enrichment or depletion of ART-associated dnSNVs in CpG islands (Fisher’s exact test, P = 0.6862; Supplementary Figure 6C). The majority of dnSNVs occur in intronic and distal intergenic regions, with no cohort differences in the distribution of dnSNVs across different genomic annotations (Wilcoxon rank sum, P > 0.05; Figure 3). Similarly, there is no difference in the predicted variant effects of new mutations between natural- and ART-born mice (Figure 3A). The number of dnSNVs in ART-derived mice is modestly increased within LINEs, but the difference does not reach statistical significance (Chi-square Test, Bonferroni-corrected P > 0.13; Figure 3D). Despite an overall increase in dnSNV rate in ART-born mice, the genomic landscape of dnSNVs is largely indistinguishable between breeding cohorts. Contrasting the epigenomic and genome regulatory context of dnSNVs Early development is characterized by the coordinated erasure, deposition, and redistribution of numerous DNA and histone modifications (52). Prior work has established that this epigenetic reprogramming is perturbed in vitro relative to in vivo (16). Given that diverse chromatin modifications are associated with intragenomic mutation rate variation (36), we next set out to explore whether dnSNVs arise in distinct epigenetic contexts in our two cohorts. We intersected the positions of dnSNVs ascertained in both the ART- and natural-born samples with genome-wide maps of several epigenetic marks in mouse embryonic stem cells (mESCs) (53) and early mouse embryos (54). There is no cohort-level difference in the proportion of dnSNVs that arise in regions associated with H3K27ac, H3K36me3, H3K4me1, H3K4me3, H3K9ac, H3K9me3 histone modifications in either mESCs or mouse embryos (Chi-square P > 0.4; Supplementary Figures 7 and 8). In somatic cells, transcription-coupled repair processes lead to reduced dnSNV rates in highly expressed genes relative to more lowly expressed or transcriptionally silenced genes (55). Although differences in gene expression have been documented between ART- and naturally born progeny (17), we find no cohort-level difference in the transcriptional activity of genes harboring dnSNVs in C57BL/6J-derived mESCs (Wilcoxon Signed Rank Test, P > 0.2; Supplementary Figure 9). A majority of dnSNVs are presumed to arise due to errors in DNA replication, with later-replicating regions of the genome exhibiting elevated mutation rates compared to early-replicating regions (56). Using published replication timing estimates derived from mESCs, we again find no evidence for differences in overall replication timing at dnSNV sites in our ART- and naturally-born cohorts (Mann Whitney U-Test, P > 0.05; Supplementary Figure 10) (57, 58). No difference in the developmental timing of dnSNVs between ART- and naturally-conceived mice The excess dnSNVs recovered in our ART-born mice may have arisen during the (i) formation of gametes in vivo, (ii) in vitro gamete manipulation and culture, (iii) the transition from the zygote to 2-cell stage, or (iv) at later stages of embryonic development. We profiled the allele depth ratio (i.e., the proportion of sequencing reads supporting the alternative allele) across all dnSNVs to gain insight into the developmental timing of mutations in our two cohorts. dnSNVs that arose in parental gametes will be constitutively present in the genome of offspring and manifest as an allele depth ratio close to ~0.5, whereas mutations that arise during early development will have lower allele depth ratios. We observed no significant difference in the allele depth ratios of dnSNVs between animals conceived via ART and those conceived naturally (Wilcoxon rank sum test, P = 0.795; Supplementary Figure 11), indicating no detectable difference in the temporal origin of dnSNVs between cohorts. No detectable difference in de novo structural variation rates between ART- and natural-born mice A recent report identified a significant increase in the rate of de novo structural variants (SV) in cattle born via ART compared to naturally conceived animals (31). We utilized an ensemble SV calling approach to identify de novo deletions and duplications >50 bp in our G2 samples (see Methods). Eight and seven high-confidence germline de novo SVs were identified in the natural (4 deletions and 4 duplications) and ART-born samples (4 deletions and 3 duplications), respectively, an insignificant difference (one-tailed Wilcoxon rank sum test; P = 0.931; Supplemental Table 5). The rate of de novo SV is predicted to be at least one order of magnitude lower than the single nucleotide mutation rate (59), a consideration that renders our study insufficiently powered to find small or modest differences in SV mutation rate.

Prior work has shown that ARTs are associated with epigenetic and transcriptomic changes in early embryos, but the potential impact of these procedures on DNA sequence integrity is not well-understood. Here, we took advantage of a well-controlled mouse model system to directly estimate the burden of dnSNVs in C57BL/6J inbred mouse cohorts conceived through a standardized ART protocol or via natural mating. We document a statistically significant increase in the overall dnSNV rate in ART-derived mice compared to their age- and genetically-matched naturally-born counterparts. Our dnSNV discovery pipeline surpasses field standards for rigor, relying on two independent sequencing libraries for each sample and invoking a simulation-informed protocol for dnSNV discovery in inbred mouse genomes (44). The biological mechanisms underlying the elevated rate of dnSNVs in ART-derived mice remain unclear. Known mutational processes often leave distinct genomic signatures or show regional enrichment (50, 60), yet we found no such signals in our data. While the modest number of dnSNVs identified may obscure subtle differences in the mutational landscape, we find no differences in the transition/transversion ratio, the mutation spectrum, or the genomic distribution of dnSNVs between cohorts. Similarly, we find no differential enrichment for dnSNVs across various epigenetic contexts, or with respect to GC content, local transcriptional activity, or replication timing. Future studies will be necessary to determine whether the elevated mutation burden observed in ART-derived mice arises from a specific step within the ART pipeline or reflects the cumulative effects of multiple procedures. One plausible contributor is ovarian stimulation using exogenous follicle-stimulating hormone (FSH) and human chorionic gonadotropin (HCG). These hormones induce the resumption of meiosis in oocytes, a process known to be highly error-prone (61). Hormone-induced ovulation could alter the fidelity of meiotic double-strand break repair or impair chromosome segregation, thereby contributing to the increased rate of de novo mutation. Although speculative, this possibility has some support. Ovulation induction has been associated with elevated risks of miscarriage and congenital anomalies in humans (62, 63), and a recent retrospective analysis suggested a link between this intervention and increased maternally transmitted mutations (34). Nonetheless, we cannot exclude the potential influence of other ART-related factors, such as mechanical stress during embryo manipulation or the physicochemical properties of the in vitro culture environment. Our use of male and female parents from a single inbred strain limits the ability to assign the parental origin of the dnSNVs reported here, an advantage that could help localize the ultimate source for the excess dnSNV burden associated with ART. Overall, the ~30% mutation rate increase observed in ART-derived offspring is unlikely to have substantial implications for the mutation load in laboratory populations maintained via recurrent cycles of IVF-based rederivation (64) or husbandry programs that utilize ARTs to rederive strains on regular intervals to curtail genetic drift (65). Assuming that ~0.5% of new single nucleotide mutations are strongly deleterious (66), a baseline mouse mutation rate of 𝜇 = 0.5 × 10−9/bp/gen (45, 46, 48), and genome size of 2.7Gb, we expect ~0.067 new deleterious mutations per mouse per generation under a traditional breeding program (0.5 × 10−9 × 2.7𝐺𝑏 × 0.005 = 0.067). Using ART, the number of expected new deleterious mutations is increased to ~0.088 per genome per generation. Thus, for every ~50 ART-derived mice, one additional highly deleterious dnSNV is expected to be recovered compared to the natural mating baseline. The magnitude of this impact is roughly equivalent to an increase in mouse paternal age of ~30 weeks (46). While our findings reveal a moderate mutagenic impact of ARTs in mice, we emphasize that our conclusions cannot be readily extrapolated beyond the mouse model system profiled here. Notably, mice and humans differ in their reproductive physiology, dynamics of epigenetic reprogramming, and timeline of early embryonic development, considerations that may influence both the rate and spectrum of dnSNVs and the sensitivity of mutation accumulation under ART. Further, ART protocols employed in human fertility clinics often include prolonged embryo culture to the blastocyst stage and an embryo freeze-thaw cycle, steps that present clear departures from the use of fresh 2-cell stage transfers in our mouse protocol. Whether or how these protocol differences impact potential mutation rate differences associated with ART is unknown. While we caution that our work has no immediate implications for human clinical practice, our findings nonetheless strongly motivate further investigation to assess the potential mutagenic impact of ART in humans. This need is particularly urgent in view of forecasted trends of increasing reliance on ART due to sociodemographic shifts and increasing democratization of access to ARTs (5).

Animal husbandry and establishment of breeding cohorts A single C57BL/6J breeding pair was obtained from The Jackson Laboratory and housed in a low barrier room in accordance with an animal care protocol approved by The Jackson Laboratory’s Animal Care and Use Committee (Protocol #17021). Two G1 females from the initial litter born to this G0 C57BL/6J founder breeding pair were naturally mated to a single male G1 littermate. One of these mated G1 females produced a litter of 9 live-born pups and the second gave birth to a litter of 7 mice (Figure 1a). All 16 naturally born G2 pups were reared to 4 weeks of age and euthanized by exposure to CO2 prior to terminal tissue collection. Two additional females and one male from the same G1 litter were transferred to The Jackson Laboratory’s Reproductive Services Facility at 4 weeks of age. Females were injected with 5 IU PMSG followed 48 hours later by a 5 IU HCG trigger to induce ovulation. Oocyte clutches were then harvested from the ampullae of each super-ovulated female and incubated in 150uL Cook RVF media supplemented with an additional 50uL of reduced glutathione (GSH) media at 37°C under mixed gas (5% CO2, 5% O2, and 90% N2) for 30–60 minutes. Concurrently, sperm were isolated from the caudal epididymides of the donor male and incubated in TYH media supplemented with 0.75mM Methyl-β-cyclodextrin at 37°C under mixed gas for 40–60 minutes. Following incubation, egg clutches and 10uL of sperm were transferred to a 1mL drop of Cook RVF media covered with mineral oil. Fertilization was allowed to occur over a ~2-6-hour period at 37°C under mixed gas. Zygotes were then washed by transfer through two sequential droplets of Cook RVF Media and incubated overnight in a final droplet of Cook RVF media at 37°C under mixed gas. Approximately 16 two-cell embryos from each donor female were subsequently transferred to the uteri of two pseudo-pregnant C57BL/6J dams and reared to term. A total of 29 live-born ART-derived G1 pups were euthanized by CO2 at approximately 4 weeks of age for terminal tissue harvests. DNA extraction, library preparation, and sequencing The G1 founder breeding pair, 6 G1 parents, 16 naturally born G2 mice, and 12 ART-derived G2 mice were selected for whole-genome sequencing (n = 36 samples; Figure 1a). Genomic DNA isolation, library preparation and sequencing were performed in duplicate for all samples, including an initial round of data collection performed in 2020 and a second batch completed in 2023. Genomic DNA was isolated from snap frozen mouse tails using the NucleoMag Tissue Kit (Machery-Nagel) according to the manufacturer’s protocol. DNA concentration and quality were assessed using the Nanodrop 8000 spectrophotometer (Thermo Scientific), the Qubit Flex dsDNA BR Assay (Thermo Scientific), and the Genomic DNA ScreenTape Analysis Assay (Agilent Technologies) and judged to be sufficient for library preparation for all samples. Paired-end 150bp whole genome libraries were constructed using the KAPA HyperPrep Kit (Roche Sequencing and Life Science) according to the manufacturer’s protocols, targeting an insert size of 400 base pairs. Briefly, the protocol entails shearing the DNA using the E220 Focused-ultrasonicator (Covaris), size selection targeting 400 bp, ligation of Illumina specific barcoded adapters and 9bp UMI adaptors, and 1 cycle of PCR amplification. Library quality was assessed using the D5000 ScreenTape (Agilent Technologies) and concentration was determined by a Qubit dsDNA HS Assay (ThermoFisher). The initial set of paired-end 150bp Illumina libraries were prepared and sequenced to ~20–30x coverage on an Illumina NovaSeq6000 using a combination of S2 and S4 flow cells. A second set of paired-end 150bp libraries were pooled and sequenced to ~30x coverage (~90 Gb/sample) on an Illumina NovaSeq 6000 using the S4 Reagent Kit v1.5. For clarity, we refer to sequencing data from these two libraries as “Seq1” and “Seq2”. All sequencing data are available on the NCBI Sequence Read Archive under PRJNA1282662. Read processing and mapping Read quality assessment and adaptor trimming were performed on each sample library using fastp (v. 0.23.4) (67). Processed reads were then mapped to GRCm39 mouse reference genome using default parameters in bwa mem (v. 0.7.17-r1188), indexed, and processed for duplicate read discovery using samtools (v. 1.21). Duplicate reads were identified by executing the following piped command series: samtools collate -O -u $BAM_FILE | \ samtools fixmate -m -u - - | \ samtools sort -u - | \ samtools markdup --use-read-groups -f $STATS_FILE \ -S -d 2500 --mode s --include-fails - $MARKED_BAM_FILE Mapping metrics were computed for each sequenced library by invoking the flagstat and idxstats command in samtools (Supplementary Table 1). The two independent libraries generated from each sample were then merged using samtools merge, and mapping metrics re-computed on the merged file (Supplementary Table 1). Single nucleotide variant calling Single sample variant calling was performed using DeepVariant (v. 1.6.1), invoking the pre-trained WGS model (68). A joint callset including all 36 samples was derived using GLnexus (v. 1.2.7). In parallel, we used Mpileup (via the ‘bcftools call’ command in bcftools v. 1.9–1) with default parameters to produce a second joint callset. These variant calling steps were executed on the individual Seq1 and Seq2 bam files from each sample, as well as the merged bam file integrating data from both Seq1 and Seq2. Calls from the merged data were used as the primary source for dnSNV discovery (see below), with the Seq1 and Seq2 call sets offering supportive confirmation of dnSNVs in two independent libraries. De novo mutation discovery dnSNV discovery was performed using the joint SNV call sets generated from merged BAM files containing reads from both sequencing libraries (Seq1 and Seq2). The joint VCF file was filtered using bcftools (v1.16) to retain only autosomal, biallelic single nucleotide variants that were present in a heterozygous state in a single G2 individual and absent from all other individuals in the pedigree. Following our simulation-based recommendations for dnSNV discovery in mice, we retained only those calls supported by both Mpileup and DeepVariant and applied regional filters to eliminate dnSNVs residing in genomic regions prone to false positive calls (44). Specifically, putative dnSNVs were excluded if they (1) overlapped with any of the following annotations defined by the repeatMasker track accessed from the UCSC Genome Browser (69): low complexity regions, rRNA, satellite DNA, scRNA, simple repeats, snRNA, srpRNA, and tRNA; (2) were flanked on one or both sides by an A/T homopolymer run of at least 5 bp; (3) were located within 35 bp of another SNV; or (4) overlapped a manually curated set of highly copy number variable genes in the mouse genome (Supplemental Table 6). We next assured that putative G2 dnSNVs surviving these strict filters were independently supported by calls in both Seq1 and Seq2. The final set of dnSNVs is provided in Supplemental Table 2. We employed an identical procedure for dnSNV discovery in the individual Seq1 and Seq2 data sets, excluding the requirement that dnSNVs are confirmed by both sequence batches (Supplemental Tables 3 and 4). De novo mutation rate estimation and analysis of mutation spectrum The per base de novo mutation rate (μ) was estimated for each G2 sample using the following formula: where M is the number of dnSNVs per G2 sample, and G is effective haploid genome size in base pairs. G was calculated as the total length of the 19 mouse autosomes (2.44 Gb) the minus masked regions described above, resulting in an effective genome size of 2.23 Gb. dnSNV spectra were obtained for each G2 sample using the TsTv-summary output flag in VCFtools version 0.1.16 (70). dnSNVs were further annotated by their trinucleotide context (i.e., the focal dnSNV and its immediate 3’ and 5’ flanking nucleotides) using the R libraries SomaticSignatures (version 2.38.2, Gehring et al. 2015) and VariantAnnotation (version 1.48.1, Obenchain et al. 2014). The proportion of mutations in a given sample that fall into each mutational category was used as input for a principal component analysis to evaluate variation in the dnSNV spectrum between ART- and natural-born mice. We used a modified Chi-square test with P-values corrected for non-independence to test for differences in the mutation fraction between the two cohorts. Finally, we aggregated the trinucleotide spectra across our two breeding cohorts to assess relationships with defined COSMIC single base mutation signatures (v 3.4) (50) computed for mm10 using SigProfilerAssignment in Python (version 0.2.3, (71)). We excluded mutation signatures not relevant to our unexposed cohort (e.g., Arisocholic Acid, Artifact, Colibactin, UV damage, Tobacco, Treatment Signatures, and immunosuppressants). To evaluate the statistical significance of signature contributions between the two groups, we randomly shuffled the cohort labels of the dnSNVs in their trinucleotide contexts and re-computed mutation signatures. We then compared the observed cohort difference in the proportion of each trinucleotide mutation class to the distribution of 1000 randomly simulated differences. A 1-sided P-value was calculated as the probability of observing a difference as large as or larger than the observed difference. All statistical analyses were performed using R (4.2.2) and RStudio (4.2.2). Genomic annotation and epigenomic enrichment of dnSNVs dnSNVs were annotated using SnpEff (v. 5.0, (72)). Bedtools intersect (v2.28.0) was used to determine the numbers of dnSNVs overlapping various classes of repeat elements annotated in the mm39 references using the RepeatMasker track extracted from UCSC Table Browser (69). Similarly, we interrogated SNP locations over CpG Islands by intersecting dnSNVs with the ‘CpG Island’ track from the UCSC Genome Browser. To contrast the GC content of dnSNVs in the natural and ART-derived cohorts, we calculated the GC content of a 201bp window centered on each dnSNV and compared the distributions of flanking GC content. dnSNVs were intersected with CTCF binding sites and various histone modifications assayed by ChIP-seq in C57BL/6J mouse ESCs under the mouse ENCODE project (73) and early mouse embryos (54). dnSNV coordinates were first lifted over to mm10 reference genome coordinates to ensure compatibility with ChIP-seq peaks associated with ENCODE datasets. Differential enrichment of dnSNVs between cohorts was assessed by Chi-Square tests. Similarly, dnSNVs were intersected with quantitative estimates of transcript abundance in C57BL/6J mouse ESCs (73), allowing a window of 2.5kb upstream and downstream of the gene start and end coordinates, respectively. Cohort differences in the mean expression level of genes neighboring dnSNVs and the proportion of dnSNVs neighboring active versus inactive genes were assessed by a Wilcoxon Rank Sum test and a Chi-square test, respectively. To evaluate potential cohort differences in the replication timing of genomic regions where dnSNVs arise, dnSNVs were intersected with published Repli-seq replication timing estimates on mESCs (57, 58). A Wilcoxon rank sum test was used to assess significance. Structural variant calling and de novo structural variant discovery To identify potential de novo structural mutations, we performed SV discovery on each cohort using DELLY (v.0.8.7) (74) and Manta 1.6.0 (75). We used default settings in DELLY to perform per sample germline SV calling against the GRCm39 reference and subsequently merged calls across all samples in our pedigree. In parallel, we used Manta to jointly call germline SVs in each of the 28 parent-offspring trios embedded in our pedigree (Figure 1A), followed by merging of these per trio SV call sets. (Running a joint sample analysis with Manta on larger sample cohorts caused run time challenges and proved to be infeasible with our compute resources). We then intersected the two final SV call sets from Manta and DELLY using the collapse command in Truvari (v4.0.0) (76) with the following parameters specified: -pctsize 0.75 –pctovl 0.5 –pctseq 0.7-s 20 -S 10000000 -k common --chain. We retained only the calls that were supported by both callers and that were unique to a single G2 sample. We focus on deletions and duplications, to the exclusion of complex and copy number natural SVs, owing to the inherent limitations of short-read data. These candidate de novo structural variants were then visually inspected for read depth signatures consistent with duplications and deletion calls using Samplot (version 1.1.6; (77)). Only calls visually supported by expected read depth patterns were retained. This manual filter resulted in the exclusion of 462 deletions and 108 duplications. De novo SVs were annotated for predicted functional effects using the Ensembl variation effect predictor (VEP; tool version 2.0; (78)). Supplementary Material

We thank members of the Dumont Lab, Baker Lab, and Mary Ann Handel at The Jackson Laboratory for critical feedback on this project. We are indebted to the technical expertise of the scientific staff in The Jackson Laboratory’s Reproductive Sciences and Genome Technologies Scientific Service for carrying out ART procedures and whole genome sequencing, respectively. We also thank the Research IT Staff at The Jackson Laboratory for their oversight and maintenance of the high-performance computing resources that made this work possible. This work was supported by start-up funds from The Jackson Laboratory and a MIRA from The National Institute of General Medical Sciences to BLD (R35 GM133415). Footnotes Competing Interest Statement: The authors have no competing interests to disclose.

- 1.ESHRE ART Fact Sheet. (2025). Available at: https://www.eshre.eu/Press-Room/Resources/Fact-sheets [Accessed 20 June 2025]. [Google Scholar] - 2.CDC, State-Specific ART Surveillance. Assisted Reproductive Technology (ART) (2024). Available at: https://www.cdc.gov/art/php/surveillance-state-specific/index.html [Accessed 20 June 2025]. [Google Scholar] - 3.Smeenk J., et al. ART in Europe, 2019: results generated from European registries by ESHRE. Human Reproduction 38, 2321–2338 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar] - 4.Katagiri Y., et al. Assisted reproductive technology in Japan: A summary report for 2021 by the Ethics Committee of the Japan Society of Obstetrics and Gynecology. Reproductive Medicine and Biology 23, e12552 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar] - 5.CDC, National ART Summary. Assisted Reproductive Technology (ART) (2024). Available at: https://www.cdc.gov/art/php/national-summary/index.html [Accessed 20 June 2025]. [Google Scholar] - 6.Nugent C. N., Chandra A., Infertility and Impaired Fecundity in Women and Men in the United States, 2015–2019. National Health Statistics Report 202, 1–19 (2024). [PubMed] [Google Scholar] - 7.Osterman M. J., Hamilton B. E., Martin J. A., Driscoll A. K., Valenzuela C. P., Births: Final Data for 2023. National Vital Statistics Reports 74 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar] - 8.de Waal E., et al. The cumulative effect of assisted reproduction procedures on placental development and epigenetic perturbations in a mouse model. Hum Mol Genet 24, 6975–6985 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar] - 9.Koustas G., Sjoblom C., Minute changes to the culture environment of mouse pre-implantation embryos affect the health of the conceptus. Asian Pacific Journal of Reproduction 5, 287–294 (2016). [Google Scholar] - 10.Yu B., Smith T. H., Battle S. L., Ferrell S., Hawkins R. D., Superovulation alters global DNA methylation in early mouse embryo development. Epigenetics 14, 780–790 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar] - 11.Ooga M., et al. Aberrant histone methylation in mouse early preimplantation embryos derived from round spermatid injection. Biochemical and Biophysical Research Communications 680, 119–126 (2023). [DOI] [PubMed] [Google Scholar] - 12.Barberet J., et al. Do assisted reproductive technologies and in vitro embryo culture influence the epigenetic control of imprinted genes and transposable elements in children? Human Reproduction 36, 479–492 (2021). [DOI] [PubMed] [Google Scholar] - 13.Choux C., et al. The epigenetic control of transposable elements and imprinted genes in newborns is affected by the mode of conception: ART versus spontaneous conception without underlying infertility. Human Reproduction 33, 331–340 (2018). [DOI] [PubMed] [Google Scholar] - 14.Ghosh J., Coutifaris C., Sapienza C., Mainigi M., Global DNA methylation levels are altered by modifiable clinical manipulations in assisted reproductive technologies. Clinical Epigenetics 9, 14 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar] - 15.Mani S., Ghosh J., Coutifaris C., Sapienza C., Mainigi M., Epigenetic changes and assisted reproductive technologies. Epigenetics 15, 12 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar] - 16.Håberg S. E., et al. DNA methylation in newborns conceived by assisted reproductive technology. Nat Commun 13, 1896 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 17.Song S., et al. DNA methylation differences between in vitro- and in vivo-conceived children are associated with ART procedures rather than infertility. Clin Epigenetics 7, 41 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar] - 18.Rivera R. M., et al. Manipulations of mouse embryos prior to implantation result in aberrant expression of imprinted genes on day 9.5 of development. Hum Mol Genet 17, 1–14 (2008). [DOI] [PubMed] [Google Scholar] - 19.Giritharan G., et al. Effect of in vitro fertilization on gene expression and development of mouse preimplantation embryos. Reproduction 134, 63–72 (2007). [DOI] [PubMed] [Google Scholar] - 20.Hayashi S., Mouse Preimplantation Embryos Developed from Oocytes Injected with Round Spermatids or Spermatozoa Have Similar but Distinct Patterns of Early Messenger RNA Expression. Biology of Reproduction 69, 1170–1176 (2003). [DOI] [PubMed] [Google Scholar] - 21.Kohda T., Effects of embryonic manipulation and epigenetics. J Hum Genet 58, 416–420 (2013). [DOI] [PubMed] [Google Scholar] - 22.Tan K., et al. Dynamic integrated analysis of DNA methylation and gene expression profiles in in vivo and in vitro fertilized mouse post-implantation extraembryonic and placental tissues. Mol Hum Reprod 22, 485–498 (2016). [DOI] [PubMed] [Google Scholar] - 23.Cortessis V. K., et al. Comprehensive meta-analysis reveals association between multiple imprinting disorders and conception by assisted reproductive technology. Journal of Assisted Reproduction and Genetics 35, 943 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar] - 24.Hattori H., et al. Association of four imprinting disorders and ART. Clinical Epigenetics 11, 21 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar] - 25.McDonald S. D., et al. Preterm birth and low birth weight among in vitro fertilization singletons: a systematic review and meta-analyses. Eur J Obstet Gynecol Reprod Biol 146, 138–148 (2009). [DOI] [PubMed] [Google Scholar] - 26.Palomba S., Homburg R., Santagni S., La Sala G. B., Orvieto R., Risk of adverse pregnancy and perinatal outcomes after high technology infertility treatment: a comprehensive systematic review. Reprod Biol Endocrinol 14, 76 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar] - 27.Luke B., et al. The risk of birth defects with conception by ART. Hum Reprod 36, 116–129 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 28.Hart R. J., Wijs L. A., The longer-term effects of IVF on offspring from childhood to adolescence. Frontiers in Reproductive Health 4, 1045762 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 29.Bean C. J., Fertilization in vitro increases non-disjunction during early cleavage divisions in a mouse model system. Human Reproduction 17, 2362–2367 (2002). [DOI] [PubMed] [Google Scholar] - 30.Bonduelle M., Prenatal testing in ICSI pregnancies: incidence of chromosomal anomalies in 1586 karyotypes and relation to sperm parameters. Human Reproduction 17, 2600–2614 (2002). [DOI] [PubMed] [Google Scholar] - 31.Lee Y.-L., et al. The rate of de novo structural variation is increased in in vitro–produced offspring and preferentially affects the paternal genome. Genome Res. 33, 1455–1464 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar] - 32.Zamani Esteki M., et al. In vitro fertilization does not increase the incidence of de novo copy number alterations in fetal and placental lineages. Nat Med 25, 1699–1705 (2019). [DOI] [PubMed] [Google Scholar] - 33.Caperton L., et al. Assisted reproductive technologies do not alter mutation frequency or spectrum. Proc. Natl. Acad. Sci. U.S.A. 104, 5085–5090 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar] - 34.Wang C., et al. Association of assisted reproductive technology, germline de novo mutations and congenital heart defects in a prospective birth cohort study. Cell Res 31, 919–928 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 35.Ruth K. S., et al. Genetic insights into biological mechanisms governing human ovarian ageing. Nature 596, 393–397 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 36.Makova K. D., Hardison R. C., The effects of chromatin organization on variation in mutation rates in the genome. Nat Rev Genet 16, 213–223 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar] - 37.Chen C., Qi H., Shen Y., Pickrell J., Przeworski M., Contrasting Determinants of Mutation Rates in Germline and Soma. Genetics 207, 255–267 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar] - 38.Vanneste E., et al. Chromosome instability is common in human cleavage-stage embryos. Nat Med 15, 577–583 (2009). [DOI] [PubMed] [Google Scholar] - 39.Currie C. E., et al. The first mitotic division of human embryos is highly error prone. Nat Commun 13, 6755 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 40.Konstantogianni O., et al. Culture of Human Embryos at High and Low Oxygen Levels. J Clin Med 13, 2222 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar] - 41.Gille J. J., van Berkel C. G., Joenje H., Mutagenicity of metabolic oxygen radicals in mammalian cell cultures. Carcinogenesis 15, 2695–2699 (1994). [DOI] [PubMed] [Google Scholar] - 42.Bristow R. G., Hill R. P., Hypoxia DNA repair and genetic instability. Nat Rev Cancer 8, 180–192 (2008). [DOI] [PubMed] [Google Scholar] - 43.Zhao C., et al. Single-cell multi-omics of human preimplantation embryos shows susceptibility to glucocorticoids. Genome Res. 32, 1627–1641 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 44.Garretson A., Blanco-Berdugo L., Roberts A., Dumont B. L., Benchmarking Genomic Variant Calling Tools in Inbred Mouse Strains: Recommendations and Considerations. [Preprint] (2025). Available at: https://www.biorxiv.org/content/10.1101/2025.05.28.656711v1 [Accessed 20 June 2025]. [Google Scholar] - 45.Uchimura A., et al. Germline mutation rates and the long-term phenotypic effects of mutation accumulation in wild-type laboratory mice and mutator mice. Genome Res 25, 1125–1134 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar] - 46.Lindsay S. J., Rahbari R., Kaplanis J., Keane T., Hurles M. E., Similarities and differences in patterns of germline mutation between mice and humans. Nat Commun 10, 4053 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar] - 47.Bergeron L. A., et al. Evolution of the germline mutation rate across vertebrates. Nature 615, 285–291 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar] - 48.López-Cortegano E., et al. The rate and spectrum of new mutations in mice inferred by long-read sequencing. Genome Res. 35, 43–54 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar] - 49.Pleasance E. D., et al. A comprehensive catalogue of somatic mutations from a human cancer genome. Nature 463, 191–196 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar] - 50.Alexandrov L. B., et al. The repertoire of mutational signatures in human cancer. Nature 578, 94–101 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar] - 51.López-Cortegano E., et al. Variation in the Spectrum of New Mutations among Inbred Strains of Mice. Mol Biol Evol 41, msae163 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar] - 52.Reik W., Dean W., Walter J., Epigenetic reprogramming in mammalian development. Science 293, 1089–1093 (2001). [DOI] [PubMed] [Google Scholar] - 53.Yue F., et al. A comparative encyclopedia of DNA elements in the mouse genome. Nature 515, 355–364 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar] - 54.Liu X., et al. Distinct features of H3K4me3 and H3K27me3 chromatin domains in pre-implantation embryos. Nature 537, 558–562 (2016). [DOI] [PubMed] [Google Scholar] - 55.Moore L., et al. The mutational landscape of human somatic and germline cells. Nature 597, 381–386 (2021). [DOI] [PubMed] [Google Scholar] - 56.Stamatoyannopoulos J. A., et al. Human mutation rate associated with DNA replication timing. Nat Genet 41, 393–395 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar] - 57.Pratto F., et al. Meiotic recombination mirrors patterns of germline replication in mice and humans. Cell 184, 4251–4267 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 58.Dey S. S., Kester L., Spanjaard B., Bienko M., van Oudenaarden A., Integrated genome and transcriptome sequencing of the same cell. Nat Biotechnol 33, 285–289 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar] - 59.Belyeu J. R., et al. De novo structural mutation rates and gamete-of-origin biases revealed through genome sequencing of 2,396 families. Am J Hum Genet 108, 597–607 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 60.Garcia-Salinas O. I., et al. The impact of ancestral, genetic, and environmental influences on germline de novo mutation rates and spectra. Nat Commun 16, 4527 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar] - 61.Hassold T., Hunt P., To err (meiotically) is human: the genesis of human aneuploidy. Nat Rev Genet 2, 280–291 (2001). [DOI] [PubMed] [Google Scholar] - 62.Hansen M., Kurinczuk J. J., Milne E., de Klerk N., Bower C., Assisted reproductive technology and birth defects: a systematic review and meta-analysis. Hum Reprod Update 19, 330–353 (2013). [DOI] [PubMed] [Google Scholar] - 63.Davies M. J., et al. Reproductive Technologies and the Risk of Birth Defects. New England Journal of Medicine 366, 1803–1813 (2012). [DOI] [PubMed] [Google Scholar] - 64.Yamauchi Y., et al. Loss of mouse Y chromosome gene Zfy1 and Zfy2 leads to spermatogenesis impairment, sperm defects, and infertility. Biol Reprod 106, 1312–1326 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 65.Taft R. A., Davisson M., Wiles M. V., Know thy mouse. Trends Genet 22, 649–653 (2006). [DOI] [PubMed] [Google Scholar] - 66.Dukler N., Mughal M. R., Ramani R., Huang Y.-F., Siepel A., Extreme purifying selection against point mutations in the human genome. Nat Commun 13, 4312 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 67.Chen S., Ultrafast one-pass FASTQ data preprocessing, quality control, and deduplication using fastp. iMeta 2, e107 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar] - 68.Yun T., et al. Accurate, scalable cohort variant calls using DeepVariant and GLnexus. Bioinformatics 36, 5582–5589 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 69.Kent W. J., et al. The Human Genome Browser at UCSC. Genome Res. 12, 996–1006 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar] - 70.Danecek P., et al. The variant call format and VCFtools. Bioinformatics 27, 2156–2158 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar] - 71.Díaz-Gay M., et al. Assigning mutational signatures to individual samples and individual somatic mutations with SigProfilerAssignment. Bioinformatics 39, btad756 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar] - 72.Cingolani P., et al. A program for annotating and predicting the effects of single nucleotide polymorphisms, SnpEff. Fly (Austin) 6, 80–92 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar] - 73.Stamatoyannopoulos J. A., et al. An encyclopedia of mouse DNA elements (Mouse ENCODE). Genome Biology 13, 418 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar] - 74.Rausch T., et al. DELLY: structural variant discovery by integrated paired-end and split-read analysis. Bioinformatics 28, i333–i339 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar] - 75.Chen X., et al. Manta: rapid detection of structural variants and indels for germline and cancer sequencing applications. Bioinformatics 32, 1220–1222 (2016). [DOI] [PubMed] [Google Scholar] - 76.English A. C., Menon V. K., Gibbs R. A., Metcalf G. A., Sedlazeck F. J., Truvari: refined structural variant comparison preserves allelic diversity. Genome Biology 23, 271 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar] - 77.Belyeu J. R., et al. Samplot: a platform for structural variant visual validation and automated filtering. Genome Biology 22, 161 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar] - 78.McLaren W., et al. The Ensembl Variant Effect Predictor. Genome Biology 17, 122 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article.