In utero exposure to zidovudine-containing antiretroviral therapy and clonal hematopoiesis in HIV-exposed uninfected newborns

Abstract

Objective:

Zidovudine (ZDV) has been extensively used in pregnant women to prevent vertical transmission of human immunodeficiency virus (HIV), but few studies have evaluated potential mutagenic effects of ZDV during fetal development.

Design:

Our study investigated clonal hematopoiesis in HIV-exposed uninfected (HEU) newborns, 94 of whom were ZDV-exposed and 91 antiretroviral therapy (ART)-unexposed and matched for potential confounding factors.

Methods:

Utilizing high depth sequencing and genotyping arrays, we comprehensively examined blood samples collected during the first week after birth for potential clonal hematopoiesis (CH) associated with fetal ZDV exposure, including clonal single nucleotide variants (SNVs), small insertions and deletions (indels), and large structural copy number or copy neutral alterations.

Results:

We observed no statistically significant difference in the number of SNVs and indels per person in ZDV exposed children (adjusted ratio (95% confidence interval) for expected number of mutations=0.79 [0.50, 1.22], P=0.3), and no difference in the number of large structural alterations. Mutations in common CH driver genes were not found in the study population. Mutational signature analyses on SNVs detected no novel signatures unique to the ZDV-exposed children and the mutational profiles were similar between the two groups.

Conclusions:

Our results suggest that CH at levels detectable in our study is not strongly influenced by in utero ZDV exposure; however, additional follow-up studies are needed to further evaluate the safety and potential long-term impacts of in utero ZDV exposure in HEU children as well as better investigate genomic aberrations occurring late in pregnancy.

INTRODUCTION

Zidovudine (ZDV) was the first approved antiretroviral medication used for preventing the transmission of human immunodeficiency virus (HIV) from mother to child [1]. The availability of ZDV, especially as part of combination antiretroviral therapy (ART), has dramatically reduced vertical transmission of HIV [2]. However, the long-term safety of in utero exposure to antiretroviral medications in HIV-exposed uninfected (HEU) children remains a concern.

ZDV is a nucleoside reverse transcriptase inhibitor (NRTI) which competitively inhibits HIV reverse transcriptase and is incorporated into viral DNA, resulting in DNA chain termination [3]. NRTIs can also become incorporated into human nuclear and mitochondrial DNA [4]. Previous studies have shown potential impact of in utero exposure to ZDV and other NRTIs. For instance, Chenadec et al. observed transiently lower hemoglobin which disappeared within 3 months and reduction in neutrophil, lymphocyte, and platelet counts which persisted from birth to at least 18 months in the French Perinatal Study [5]. The European Collaborative Study reported that reduced neutrophil count could be observed in kids exposed to antiretroviral therapy even after 8 years [6]. Finally, Pacheco et al. demonstrated that changes in platelet and lymphocyte counts in children exposed to antiretroviral therapy persisted 2 years after birth [7]. In vitro studies also suggest ZDV could have genotoxic effects [8,9]. Furthermore, evidence from animal studies suggests ZDV could have transplacental effects and cause genetic abnormalities and genotoxicity [10–12]. Finally, aneuploidy rate in cord blood cells as well as expression of genes related to DNA repair have been shown to be altered in children with exposure to ZDV [13,14].

These hematologic and genotoxic effects have raised concerns regarding the long-term risk of cancer in HEU children exposed in utero to ZDV. The International Agency for Research on Cancer has classified ZDV as a probable carcinogen for humans (class 2A agent) [15]. As the population of surviving HEU children expands and ages, knowledge of potential genotoxic and carcinogenic effects of early-life exposure to ZDV is essential for weighing potential risks and benefits of in utero antiretroviral regimens that include ZDV or other NRTIs.

Clonal hematopoiesis (CH) refers to clonal expansion of a subset of leukocytes carrying somatic mutations [16]. These mutations range in size from a single nucleotide variant (SNV) to large somatic copy number alterations (SCNAs) at the chromosomal scale. The detection of CH relies on the size of event: SNVs require next-generation sequencing whereas SCNAs are better detected by virtual karyotyping methods using single nucleotide polymorphism (SNP) genotyping arrays. Besides being a characteristic of myelodysplastic syndromes as well as leukemia [17,18], CH has been found to be associated with cardiovascular disease [19], Alzheimer disease [20], and cancer [21–23]. Previous studies on CH have also found associations with chemotherapy [24], cigarette smoking [16,25] and air pollution [26] suggesting that CH can serve as a biomarker reflecting environmental exposure to a carcinogen or mutagen. However, no study has characterized potential adverse effects of ZDV exposure on fetal CH using currently available genomic technologies.

In this study, we investigate CH as a marker of genomic damage detected in peripheral blood mononuclear cell samples obtained from ZDV-exposed and, for comparison, ART-unexposed neonates during the first week after birth. We utilize high-depth DNA sequencing and SNP genotype array-based methods to comprehensively identify clonal SNVs, small insertions and deletions (indels), and large acquired SCNAs.

METHODS

Study participants

We started with 200 HEU children born to women living with HIV from two studies. One hundred newborns who were exposed to ZDV in utero for the longest duration (most of them in combination with other antiretrovirals) and born between 1995 and 2017 were selected from the Surveillance Monitoring for ART Toxicities (SMARTT) study of the Pediatric HIV/AIDS Cohort Study (PHACS) network. SMARTT enrolled children with detailed information available on in utero antiretroviral medication exposure from 22 US sites since 1995 [27,28]. The comparison group was comprised of ART-unexposed children who were frequency-matched to the exposure group by age, gender, and ethnicity mostly from the Women and Infants Transmission Study (WITS) (one ART-unexposed child from SMARTT). WITS enrolled newborns born during 1990–1994 and their mothers living with HIV from 6 US sites [29]. The current study shared the same ZDV-exposed and most ART-unexposed children with our prior study, and details of participant selection have been described previously [30]. We retrieved information from PHACS and WITS on gestational age, birthweight, maternal demographic and health information including HIV RNA levels and CD4 cell count before delivery, and antiretroviral use during pregnancy. After DNA extraction, 6 children from WITS and 4 from SMARTT were excluded due to insufficient DNA yield. Further examination of genomic data revealed that 1 child from WITS and 2 childern from SMARTT had evidence for contamination and were removed. An additional 2 children from WITS were excluded due to gender discordance, leaving 94 ZDV-exposed and 91 ART-unexposed HEU children included in the current analysis. Institutional Review Board approval was obtained at each participating center and all participating mothers and/or caregivers provided written informed consent.

Laboratory methods

Exome and untranslated region (UTR) sequencing

Whole exome sequencing (including neighboring UTRs) for the HEU newborns was performed at the Cancer Genomics Research Laboratory of the National Cancer Institute, as previously described [31]. In brief, 1.1 μg of genomic DNA was extracted from peripheral blood mononuclear cell samples utilizing magnetic-bead based DNA extraction and purification chemistry on the QIAsymphony SP according to the manufacturer’s validated protocols (QIAGEN, Germany). Exome+UTR capture was performed using SeqCap EZ Human Exome Library v3.0, (Roche NimbleGen, Inc., Madison, WI, USA). The captured DNA was coded with unique molecular index then underwent paired-end sequencing on an Illumina HiSeq following Illumina-provided protocols for 2×100-bp paired-end sequencing. Exome sequencing was conducted to high depth to achieve a targeted minimum threshold of 80% of coding sequence bases with coverage of over 250-fold, based on the University of California Santa Cruz (UCSC) hg19 “known gene” transcripts database (http://genome.ucsc.edu/)[32]. The exome data analyzed are now archived in a controlled-access dbGaP repository under the following accession number (phs002061).

Variant alignment, calling, and annotation

Details of variant alignment and calling methods have been published previously [31]. Briefly, sequencing reads were trimmed using the Trimmomatic tool [33] that reported the longest high-quality stretch of each read. Reads were then aligned to the reference genome (hg19) using the Novoalign software v3.00.05 (http://www.novocraft.com). To refine alignments, the following reads were removed from further analysis: 1) duplicate reads (using the MarkDuplicates module of the Picard software (v1.126) (http://picard.sourceforge.net/)), 2) sequence pairs not in complementary directions when mapped to the reference genome, and 3) sequence reads not reflecting the expected fragment length (300 ± 100 bp). A local realignment around known and novel sites of insertion and deletion mutations was performed using the RealignerTargetCreator and IndelRealigner modules from the Genome Analysis Toolkit (GATK v3.1) [34]. BAM file level recalibration was also performed using BaseRecalibrator module from GATK. Variants were identified by HaplotypeCaller and UnifiedGenotyper modules from GATK as well as FreeBayes [35]. All called variants were annotated by snpEff with a focus on predicted medium and high impact mutations in subset analyses [36].

Identification of somatic SNVs and small indels from sequence variant calls

Variant calls were further filtered by the following criteria: 1) must be within 5 bp of designed exome probe capture region, 2) must reside within coding regions of the genome, and 3) must be called by HaplotypeCaller, UnifiedGenotyper, and FreeBayes. We defined SNVs and small indels as variants which 1) had two or more reads for the alternative allele to remove potential sequencing noise or artifacts, 2) had a minor allele frequency ≤ 0.1% in Exome Aggregation Consortium (ExAC) [37], NHLBI GO Exome Sequencing Project (ESP) [38], 1000 Genomes Project phase 3 [39] and gnomAD [40] databases which suggests the called variants were not a common germline variant, 3) had a minor allele frequency < 5% in the current sequenced study population to exclude potential sequencing artifacts, and 4) were not in highly repetitive regions (which often result in poorly mapped reads and false variant calls). The highly repetitive regions include a) loci which failed Hardy-Weinberg equilibrium in the 1000 Genomes Project [39], b) regions with unusually high read depth reported previously [41], c) low complexity regions identified by mdust [42], and d) highly duplicated regions [43]. Finally, we created a theoretical distribution of variant allele frequencies (VAFs) by symmetrically mirroring the observed distribution of variants between VAF of 0.50 and 0.98 for VAFs < 0.5. A variant with VAF < the 0.25% quantile of this theoretical distribution was classified as a detectable somatic SNV or indel. VAF is defined as the percentage of reads containing a specific alternative allele divided by the overall coverage for that locus.

Characterization of somatic SNV mutational signatures

Somatic SNV calls from each ZDV exposure group were aggregated and used as input for SigProfilerMatrixGenerator [44] to identify potential de novo and known single base substitution (SBS) mutational signatures (COSMIC Mutational Signatures v3) [45]. Results from both the de novo and decomposed mutational signatures analyses were visualized using tidysig [46] and compared between ZDV-exposed and ART-unexposed children. We analyzed all 96 trinucleotide mutational contexts as well as the 6 single nucleotide SNV somatic substitutions by principal component analysis to explore potential clustering of samples by ZDV exposure group. Binomial tests of each mutational context and Chi-squared tests for overall association between ZDV exposure and mutational context were employed to identify potential differences in mutational signature. Mutational signatures were further confirmed with SigneR [47].

SNP array genotyping and somatic copy number alteration detection

Among the 185 individuals, 184 had enough DNA for genotyping, and 5 samples failed QC leaving 179 genotyped. Genome‐wide genotyping was independently performed using Illumina Global Screening Arrays (version 1). We estimated log₂ R ratio (LRR) and B allele frequency (BAF) values from raw array intensity data to detect somatic copy number change or copy neutral alterations sized > 100kb as described previously [48]. The LRR and BAF values were re-calculated after quantile normalization [49], then analyzed utilizing the MoChA [50] framework to systematically detect deviations in LRR and BAF values which indicate the presence of somatic copy number or copy neutral alterations. All resulting somatic calls were manually inspected to ensure high true positive rates.

Statistical analysis

Fisher’s exact tests using the mid-p method [51] and Wilcoxon rank sum tests were performed to compare characteristics of ZDV-exposed and ART-unexposed children. Significance level for multiple comparison was determined by Bonferroni correction. We employed negative binomial regression to evaluate associations between ZDV exposure and number of detectable clonal SNVs, in univariable models and multivariable models adjusting for infant’s sex, race/ethnicity, gestational age, and maternal use of alcohol, marijuana, and tobacco. As only 2 SCNA events were identified, we did not conduct statistical comparisons. All statistical analyses and visualizations were performed using R v4.0.2 [52].

RESULTS

Characteristics of the 185 participants are described in Table 1. No statistically significant differences were observed in infants’ sex, race/ethnicity, and mothers’ use of tobacco, alcohol, marijuana, or illicit drugs during pregnancy between the two groups. ZDV-exposed children were younger than ART-unexposed children (median 38.3 versus 39 weeks, P=0.021) and were slightly lighter in birth weight (median 3 versus 3.13 kg, P=0.045). Mothers of ZDV-exposed children were older at delivery (median 28.6 versus 26 years, P<0.001) and had lower CD4+ T-cell count (median 502 versus 577.5 cells/ul, P=0.015) and HIV viral load (median 48 versus 9684 copies/ml, P<0.001).

Table 1.

Study participant characteristics of the newborn children in ZDV-exposed and ART-unexposed groups used for genomic characterization in our study of CH.

Variable	ZDV-exposed, N=94	ART-unexposed, N=91	P
Birth weight (kg), median (range)	3 (1.74, 4.38)	3.13 (1.91, 4.60)	0.045
Gestational age (weeks), median (range)	38.3 (32.6, 42.1)	39.0 (3.20, 43.0)	0.021
Female, n (%)	49 (52.1)	46 (50.6)	0.83
Group, n (%)			<0.001
Public WITS	0 (0)	90 (98.9)
SMARTT w/o WITS	93 (98.9)	1 (11)
SMARTT WITS	1 (11)	0 (0)
Year of birth, n (%)			<0.001
1990–1994	0 (0)	88 (96.7)
1995–2017	94 (100)	3 (3.3)
Mothers age at delivery (years), median (range)	28.58 (16.92, 44.97)	26 (15, 40)	<0.001
Participant’s race/ethnicity, n (%)			0.36
White	6 (6.4)	13 (14.3)
African-American	31 (33.0)	40 (44.0)
Hispanic	55 (58.5)	32 (35.2)
Other	2 (2.1)	6 (6.6)
Days from sample draw to birth, median (range)	1 (0, 7)	1 (0, 7)	0.072
Last CD4+ cell count during pregnancy (cells/μl), median (range)	502 (34, 1198)	577.5 (74, 2330)	0.015
Last HIV RNA level during pregnancy (copies/ml), median (range)	48 (0, 64361)	9684 (0, 672005)	<0.001
<400	82 (88.2)	10 (11.0)	<0.001
≥400	11 (11.8)	81 (89.0)
Mother ever used tobacco, n (%)	26 (27.7)	32 (35.2)	0.28
Mother ever used alcohol, n (%)	25 (26.6)	32 (35.2)	0.21
Mothers ever used marijuana/hashish, n (%)	13 (13.8)	8 (8.8)	0.29
NNRTIs use during pregnancy, n (%)	18 (19.2)	0 (0)	<0.001
PIs use during pregnancy, n (%)	70 (74.5)	0 (0)	<0.001

Footnote:

^*:NNRTI: Non-Nucleoside Reverse Transcriptase Inhibitor; PI: Protease Inhibitor.

Clonal somatic single nucleotide variants and small indels

We identified a total of 80 clonal somatic SNVs and 2 indels in ZDV-exposed and 93 somatic SNVs and 9 indels in ART-unexposed children. Mean mutation burden per person was 0.01 (range 0 to 0.11) in ZDV-exposed and 0.02 (range 0 to 0.16) mutations/Mb in ART-unexposed infants. The range of VAF for detectable clonal SNVs and indels ranged from 2.5% to 30.3% (Figure S1). The overall mean number of detected somatic SNVs or indels per participant was 0.94 with the mean count among ZDV-exposed children of 0.84 and among ART-unexposed children of 1.03 (Figure 1, P=0.3). In univariable negative binomial regression analyses, we did not observe significant associations between the number of clonal SNVs and indels with ZDV exposure (P=0.36) or with sex, race/ethnicity, substance use, or mother’s last measured viral load or CD4+ T-cell count (Table S1). In a multivariable model adjusting for infant’s sex, race/ethnicity, gestational age, and maternal use of alcohol, marijuana, and tobacco, the somatic SNV or indel count appeared lower in ZDV-exposed than ART-unexposed children but the difference was non-significant (adjusted ratio for mutation counts = 0.79 [95% CI 0.50 to 1.22], P=0.3, Table 2). We also investigated somatic SNVs and indels with potential high or medium impact predicted by snpEff (Tables S2). Among the top 50 mutated genes with detected mutations in our sample set, 34 genes had at least one medium or high impact somatic SNVs or indels detected; 29 and 11 of these top genes had medium or high impact mutations in ZDV-exposed and ART-unexposed infants, respectively.

Figure 1.

Number of somatic SNVs and indels combined in peripheral blood mononuclear cell derived DNA. Dotted lines indicate median for each group.

Table 2.

Multivariable associations of the number of whole-exome detected clonal SNVs and indels with zidovudine exposure and other descriptive characteristics in both ZDV-exposed and ART-unexposed newborn children.

Characteristic	Adjusted ratio for expected mutation counts [95%CI]	P
Zidovudine exposure	0.79 [0.50, 1.22]	0.30
Gestational age (week)	0.99 [0.88, 1.12]	0.87
Infant’s Male sex	0.78 [0.50, 1.23]	0.29
Infant Non-African American race	1.08 [0.68, 1.72]	0.74
Maternal Tobacco Use	0.96 [0.58, 1.56]	0.85
Maternal Marijuana Use	1.36 [0.68, 2.74]	0.38
Maternal Alcohol Use	0.85 [0.51, 1.42]	0.51

We curated a list of 37 CH driver genes (Table S3) commonly observed to be somatically mutated in studies of clonal hematopoiesis [53–55]. We did not detect the presence of any clonal SNV or indel in these genes.

To compare potential mutational processes and characteristic signatures between ZDV- exposed and ART-unexposed children, we analyzed SBS mutational signatures (Figure 2). Results were highly similar with no visible overall differences in mutational context by fetal ZDV exposure status. We likewise did not observe differences based on principal component analysis (Figure S2). In addition, we did not identify enrichment based on 96 SBS groups (Table S4) or mutational contexts (Table S5, Figure 2) adjusted by Bonferroni corrected thresholds. Finally, we further performed mutational signature analyses to identify the contribution of known signatures to the somatic SNVs detected in our study. We identified SBS signatures 1 and 5 (two clock-like mutational signatures [56]) as the predominant signatures in both groups. We also found SBS signature 15 which is one of the mutation signatures associated with defects in DNA mismatch repair (MSI). However, we did not observe any mutational signatures exclusively in the ZDV-exposed individuals and found no evidence suggesting a novel signature associated with ZDV exposure.

Figure 2.

Mutation count by base context for somatic SNVs detected by deep coverage whole exome sequencing. Top panel: 94 ZDV-exposed infants. Lower panel: 91 ART-unexposed infants. Footnote: *: Base mutations are displayed in the reference>alternate format (e.g., C>A represents a C to A mutation). The X axis shows base context in which the base pair before and after the mutation are displayed. The Y axis represents the total number of mutations detected from each base context.

Detected structural somatic copy number and copy neutral alterations

We detected 2 clonal SCNAs in the peripheral blood mononuclear cell-derived DNA in ART-unexposed and none in ZDV-exposed children. The first child had a partial duplication of chromosome 6 (Figure 3a) that impacted a small fraction of sampled leukocytes (approximately 5%). And the second child was a female with a total loss of one copy of the X chromosome that impacted approximately 62% of sampled leukocytes (Figure 3b).

Figure 3.

Log intensity ratio (LRR) and B-allele frequency (BAF) of detected SCNA by SNP arrays. (a) Detected mosaic SCNA on chromosome 6 from one participant. (b) Detected SCNA on chromosome X from another participant. Footnote: *: Black dots denote LRR of each SNP with scale on the left (Y1 axis), and red dots denote BAF of each SNP with scale on the right (Y2 axis). Three red bands represent BAF of SNPs with homozygous major alleles (bottom), heterozygous alleles (middle), and homozygous minor alleles (top). (a) Vertical grey lines and the horizontal blue line indicate the region with detected SCNA event (from 9,661,265 to 39,016,096). The two panels to the right are the probability density of BAF (red) and LRR (black) in normal (norm) and SCNA (Seg1) regions. The two black lines are the modes of two potentially overlapping BAF distributions, and the green shaded area shows the 95% confidence interval for the predicted mode. Based on the mode of BAF, about 5% of the cells carry this SCNA. (b) Both LRR and BAF suggested the event covered the whole chromosome. The decreased intensity in LRR suggested copy number loss, and the modes of BAF indicated that 62% of sampled cells carried this event.

Discussion

We employed both deep whole-exome sequencing and genome-wide genotyping arrays to investigate CH as a potential result of in utero ZDV exposure in 94 newborn HEU children and 91 matched ART-unexposed newborns. With high depth, broad coverage, and integration of unique molecular index, we were capable of detecting clonal mutations of low frequency across the whole exome. We did not observe statistically significant differences in the number of somatic SNV and indels among children who were exposed in utero to ZDV. Further analyses found that neither the mutational spectrum nor the underlying mutational signatures differed between the ZDV-exposed and ART-unexposed children. Our genome-wide scan for SCNAs identified two individuals with SCNAs in the ART-unexposed group. Cumulatively, these results suggest in utero exposure to ZDV has limited impact on the types and frequencies of clonal somatic mutations in peripheral leukocytes detectable by our approaches.

The nonsignificant ratio reduction in somatic SNV and indels in ZDV-exposed children compared to ART-unexposed controls (mean count of 0.84 vs 1.03, P=0.3) is intriguing, but we did not find evidence from mutational context or mutation signature analyses for any ZDV-specific biologic mechanisms driving this slight reduction among ZDV-exposed children. We also did not observe any instance of CH within commonly observed CH driver genes. Interestingly, the genes observed to be somatically mutated in our investigation often contained a large proportion of medium or high impact mutations, suggesting these mutations may have deleterious impacts on leukocyte function.

We detected 2 ART-unexposed children with mosaic aneuploidies but found no evidence for aneuploidies among the ZDV-exposed children. These findings differ from those reported by André-Schmutz et al. [14] and Vivanti et al. [13] in which an increase in aneuploidy was observed in cord blood at birth in children exposed in utero to ART containing ZDV. The observed differences could be attributable to different laboratory methods: André-Schmutz et al. and Vivanti et al. used cytogenic approaches to assess cell karyotypes for aneuploidy, whereas our analysis of genotyping array intensity data is equivalent to creating a “virtual karyotype” [57]. Moreover, virtual karyotyping allows for an assessment of all circulating peripheral leukocytes, while the cytogenic karyotyping analyses by André-Schmutz et al. and Vivanti et al focused exclusively on CD3⁺ T cells. Although virtual karyotyping is not sensitive enough to capture all genomic aberrations down to single cell level, the MoChA algorithm is sensitive to detect clonal events carried by clones which disproportionately expand relative to non-mutated cells to a detectable cellular fraction of 1% when the event spans beyond 100Mb. For events as small as 1Mb, the minimum detectable clonal fraction is around 10%, and there was no significant difference in the detection of gain or loss except for potential false positive from constitutional gain events which is removed in the pre-processing step of MoChA [50]. The survival and proliferation of such clones provide preliminary indication that these genomic aberrations might confer some functional advantage. The absence of detectable mosaic aneuploidy among ZDV-exposed children in our investigation cannot entirely rule out the presence of elevated aneuploidy in children exposed in utero to ZDV, but suggests that even if elevated rates of aneuploidy exist, they have not resulted in disproportionate growth of aberrant clones at the time point we sampled for DNA collection.

Our study used both deep whole exome sequencing and SNP genotyping arrays to detect CH in blood-derived DNA collected from newborns during the first week after birth. The use of high-depth sequencing allowed for us to detect CH clones with SNV or indel VAFs ranging from 2.5% to 30.3%. Likewise, intensity data from genotyping arrays allowed us to generate virtual karyotypes for detecting mutated cell fractions for SCNAs ranging from approximately 5% to 95%. While our genomic approaches were able to thoroughly characterize CH in ZDV-exposed and ART-unexposed HEU children reaching the frequencies described above, our methods were unable to detect SNVs or indels above the VAF threshold which is indistinguishable from germline mutations or low-frequency clones below our limits of detection (e.g., mutations in single cells). All of these somatic mutations, while not studied by CH investigations, could have associations with in utero ZDV exposure with potential relevance for future disease risk (i.e., elevated hematologic cancer risk), but we have limited ability to examine this hypothesis in the current study. Furthermore, our study was a snapshot at birth which might not be the best time point to evaluate the impact of mosaic events ocurred in late pregnancy. Besides, selective forces during infancy, adolescence and adulthood may promote clonal expansion of low-frequency clones currently below our window of detection to become detectable CH later in life. Additionally, our study only investigated somatic mutations in blood-derived DNA. Mutations could be present in other tissues and have relevance for increased disease risk such as cancer or cardiovascular diseases in relation to in utero ZDV exposure. Future studies should utilize single cell sequencing approaches to determine excess frequency of mutational load and evaluate specific components and combinations of ART in greater detail. Finally, with an overall clonal mutation rate of 0.94/person during gestation and approximately balanced design of 91 and 94 individuals, the power of our study to detect a 2-fold increase in mutation rate was excellent (94%, with two-sided significance level of 0.05). However, the power to detect smaller differences was lower. For instance, if the actual increase in mutation rate is 1.6 fold, our study’s power was only 65%.

The current study focused on the comparison between ZDV-exposed and ART-unexposed individuals. A particular challenge of investigating potential genotoxic impacts of in utero ART exposure is the varied content of drugs that comprise the treatment. Different combinations of antiretroviral drugs might have varied impact on the development of somatic mutations and clonal selection of mutated cells needed for the detection of CH. While modern ART regimens often replace ZDV with newer generation of NRTIs, they still share similar mechanisms of action as ZDV. Therefore, our study on ZDV could have implications for the safety of more recent NRTIs included in the ART regimens. Consistent with our findings which suggest that ZDV exposure did not resulted in detectable abnormal clonality in blood cells, the French Perinatal Study also failed to detect changes in risk of blood cancers from ages of 0–15 [58].

In conclusion, we did not find statistically significant evidence for the contribution of in utero ZDV exposure to CH in HEU children. Future larger long-term follow-up studies are warranted to better assess genotoxicity and to ensure the long-term safety of in utero NRTI exposure.