Genomic Imbalances in the Placenta Contribute to Poor Fetal Growth

Background: Fetal growth restriction (FGR) is associated with increased risks for complications before, during, and after birth, in addition to risk of disease through to adulthood. Although placental insuciency, failure to supply the fetus with adequate nutrients, underlies most cases of FGR, its causes are diverse and not fully understood. One of the few diagnosable causes of placental insuciency in ongoing pregnancies is the presence of large chromosomal imbalances such as trisomy conned to the placenta; however, the impact of smaller copy number variants (CNVs) has not yet been adequately addressed. In this study, we comprehensively evaluate the contribution of both placental aneuploidy and CNVs to fetal growth. Methods: We used molecular-cytogenetic approaches to identify aneuploidy in placentas from N=101 infants born small-for-gestational age (SGA), typically used as a surrogate for FGR, and from N=173 non-SGA controls from uncomplicated pregnancies. We conrmed aneuploidies and assessed mosaicism by microsatellite genotyping. We then proled CNVs using high-resolution microarrays in a subset of N=53 SGA and N=61 control euploid placentas, and compared the load, impact, gene enrichment and clinical relevance of CNVs between groups. Candidate CNVs were conrmed using quantitative PCR. Results: Aneuploidy was over 10-fold more frequent in SGA-associated placentas compared to controls (11.9% vs. 1.1%; p=0.0002, OR=11.4), was conned to the placenta, and typically involved autosomes, whereas only sex chromosome abnormalities were observed in controls. We found no signicant difference in CNV load or number of placental-expressed or imprinted genes in CNVs between SGA and controls, however, a rare and likely clinically-relevant germline CNV was identied in 5.7% of SGA cases. These CNVs involved candidate genes INHBB, HSD11B2, CTCF, and CSMD3. Conclusions: We conclude that placental genomic control groups by Student’s t-test for maternal age and birth weight, Mann-Whitney U-test for gestational age, and Fisher’s exact test for all categorical variables. SGA, small-for-gestational age; PE, preeclampsia; N/A, not available. in R. Based on a previous report of a large effect size (d > 0.95) in the difference in CNV load in control vs. SGA placentas (19), we assumed a slightly lower but still large effect size (d) of 0.8. Based on the minimum sample size in each group per cohort (N = 24) and using an α = 0.05, our study had > 80% power to detect signicant differences in each cohort individually. Analyses were performed in R version 3.5.1 (38), and plots were generated using the ggplot2, ggbio, and ggpubr packages.

and function, thus the etiologies of FGR and placental insu ciency are complex and intertwined. A major known cause of placental insu ciency in a viable pregnancy is con ned placental mosaicism (CPM), where some or most cells in the placenta are aneuploid, while the fetus has a predominantly normal diploid chromosome complement. CPM identi ed prenatally is associated with increased risk for FGR and other pregnancy complications depending on the levels of abnormal cells and the chromosome(s) involved (10,11). Screening placentas postnatally has also con rmed a contribution of CPM to FGR (12)(13)(14)(15). We previously identi ed trisomy CPM in 4/43 FGR pregnancies, but none in 85 controls nor 18 cases associated with preeclampsia (PE) without FGR (16). Despite the evidence that large genomic imbalances in the placenta are associated with FGR, few studies have investigated the role of smaller genetic imbalances (< 5-10 Mb), copy number variants (CNVs). To date, studies investigating CNVs associated with FGR have either not studied placental tissue (17,18) or had small sample sizes and found con icting results (19,20).
In this study, we aimed to thoroughly evaluate the contribution of placental genomic imbalances to poor fetal growth. To this end, we assessed i) the incidence of large aneuploidies (> 15 Mb) in 274 placentas from control and SGA pregnancies, and ii) the load, impact, and clinical relevance of placental CNVs (< 15 Mb) to SGA in a subset of 114 euploid placentas. This is the largest study to date of its kind; it enhances our understanding of the underlying causes of placental dysfunction and poor fetal growth, and further establishes the importance of assessment of CPM in the clinic.

Research ethics approval
Ethics approval for use of human research subjects in this study was obtained from the University of British Columbia/Children's and Women's Health Centre of British Columbia Research Ethics board (H17-01545) and from the Hospital for Sick Children (1000038847) and Mount Sinai Hospital (05-0038-E) Research Ethics boards. Informed written consent was obtained from all study participants.

Sample collection and cohort characteristics Vancouver Cohort
Placental samples for the Vancouver cohort were ascertained and processed as described 19 and include cases used in previous studies (16,(21)(22)(23)(24). Clinical information, including newborn sex and birth weight, gestational age at delivery, maternal age, and ethnicity were collected. Placental and maternal samples were processed and DNA was extracted as previously described (16).
This cohort (N = 207) included 136 controls from uncomplicated pregnancies (no SGA, hypertension/PE, or known abnormal maternal serum screen results) and 71 cases of SGA (Table 1). Exclusion criteria were a prenatally-diagnosed chromosome abnormality or congenital anomaly in the fetus. SGA was de ned as birth weight < 10th percentile, adjusted for sex and gestational age at birth based on Canadian growth charts (25). The majority, 55/71 (77%) of SGA cases met criteria for FGR, de ned as birth weight < 3rd percentile, or < 10th percentile with additional ndings suggestive of placental insu ciency, including i) persistent uterine artery notching at 22-25 weeks, ii) absent or reversed end diastolic velocity on umbilical artery Doppler, and/or iii) oligohydramnios (amniotic uid index < 50 mm). One FGR case had a birth weight > 10th percentile but was diagnosed as FGR from prenatal measurements and severe oligohydramnios. Preeclampsia (PE) was de ned according to Canadian criteria(26) as previously described (24). Following aneuploidy assessment, euploid placentas from a subset of N = 24 control and N = 29 SGA cases, 90% of which ful lled criteria for FGR, were selected for further CNV pro ling (Table 1). These were randomly selected after excluding cases and controls associated with a twin pregnancy (N = 23), or known maternal smoking during pregnancy (N = 3). Figure 1 summarizes the study design and number of cases per cohort used at each analysis step.

Toronto cohort
The Toronto cohort was ascertained and processed as part of a distinct study, and ndings from the two cohorts were then subsequently compared. Placental samples were obtained as previously described (27). Clinical information including newborn sex, birth weight, and gestational age were collected for all cases. The original cohort included N = 99 samples, however following microarray quality ltering, N = 67 remained, including placentas from 37 control and 30 SGA pregnancies (Table 1, Fig. 1). De nitions for control and SGA followed the same criteria as the Vancouver cohort, described above. Exclusion criteria were a prenatally-diagnosed chromosome abnormality or congenital anomaly in the fetus, CMV or toxoplasmosis infection, or clinical amnionitis. Additionally, cases or controls were excluded if mothers were diagnosed with: i) preconceptional severe hypertension; ii) clinically signi cant thrombophilia; iii) advanced renal, heart or liver failure; iv) type I diabetes mellitus or gestational diabetes requiring treatment with insulin; or v) anemia and autoimmune disorders requiring therapy during pregnancy.

Aneuploidy screening and CPM follow-up
Aneuploidy was detected using several methods in this study. In the Vancouver cohort, samples were assessed by comparative genomic hybridization (CGH), which can detect aneuploidies greater than 15 Mb, or by multiplexed ligation-dependent probe ampli cation (MLPA) of subtelomeric probes (SALSA MLPA Subtelomeres Mix, MRC-Holland, NL), designed to detect aneuploidies that extend to the ends of the chromosome (Fig. 1). A subset of these samples (N = 85 control and N = 43 SGA), all screened by CGH, have been previously published (16); the current study is a retrospective re-assessment of aneuploidy in those cases, with additional samples collected. For more recent cases, MLPA was used to screen for aneuploidy because it is a reliable and cost-effective method to identify whole chromosome aneuploidies (monosomy and trisomy), as well as terminal duplications and deletions. In the Toronto cohort, aneuploidy was detected using CNV pro ling by microarray (see below). All cases with an aneuploidy detected by any method was further assessed by microsatellite polymorphism genotyping of probes on the involved chromosome (Additional File 1: Supplementary Methods). Aneuploidies identi ed by MLPA were also con rmed using CNV pro ling by microarray to determine the extent of the alteration, particularly in cases where results suggested abnormalities restricted to one chromosome arm (see below, Supplementary Methods).

Microarray processing and CNV detection
Placental DNA was assessed on the In nium Omni2.5-8 BeadChip array (Illumina, USA) for the Vancouver cohort, and on the Affymetrix CytoScan HD array (ThermoFisher Scienti c, USA) for the Toronto cohort ( Fig. 1) at The Centre for Applied Genomics, Toronto, Canada (28,29). In the Vancouver cohort, an additional DNA sample from a different location in each placenta was also run on the array to assess the possibility of detecting mosaicism of CNVs by high-density microarray (Supplementary Methods).
Following sample quality checks unique to each array type, all 54 Vancouver cases and 67/99 Toronto cases were available for analysis ( Fig. 1). CNVs were detected using in-house pipelines (28, 29) applying 3-4 CNV-calling algorithms speci c to each array platform (Supplementary Methods). Following CNV quality checks, high-con dence CNVs called by at least two algorithms with a minimum 50% reciprocal overlap, ≥ 5 probes, and ≥ 10 kb were kept for analysis. CNV boundaries were compared to the Database of Genomic Variants and in-house databases of CNVs in controls, and rare CNVs were de ned as those present in < 0.1% of controls and at least 50% unique. Given discordance in CNV calls between technical replicates of placental DNA (Supplementary Methods, Supplementary Fig. 1), mosaicism of CNVs was not investigated and the DNA sample with the higher microarray quality scores from each placenta was selected for CNV analysis for the Vancouver cohort. Ancestry was assessed using SNP genotypes by Placental-enhanced and imprinted genes A list of 356 genes with elevated expression in the placenta was downloaded from the Human Protein Atlas (32), including 78 with placental-speci c elevated expression. A database of imprinted regions was curated from the OTAGO Imprinted Genes (33) and GeneImprint (34) databases, and reported placental imprinted differentially methylated regions (DMRs) (35,36) (Table S2). Outer genomic boundaries were used to generate a consensus region for those genes associated with a placental imprinted DMR.
Functional pathway enrichment Enrichment of 2,191 GO and KEGG (37) pathways in genes with coding sequences impacted by rare CNVs in SGA was assessed using a generalized linear model with universal gene count correction in the cnvGSA R package. Sex and cohort (array) were included as covariates, and thresholds of 100-1,500 genes were used to limit pathways assessed. A false-discovery rate (FDR) of < 0.1 was used to de ne signi cantly enriched (coe cient > 0) or de cient (coe cient < 0) pathways in SGA CNVs.

Statistical analyses
Continuous variables were compared using the Student's t-test or Mann-Whitney U test depending on whether the data was normally-distributed by the Shapiro-Wilk normality test. Categorical variables were compared by Fisher's exact test. Bonferroni correction for multiple testing was used where applicable. Statistical power for comparing CNV load was assessed using the pwr package in R. Based on a previous report of a large effect size (d > 0.95) in the difference in CNV load in control vs. SGA placentas (19), we assumed a slightly lower but still large effect size (d) of 0.8. Based on the minimum sample size in each group per cohort (N = 24) and using an α = 0.05, our study had > 80% power to detect signi cant differences in each cohort individually. Analyses were performed in R version 3.5.1 (38), and plots were generated using the ggplot2, ggbio, and ggpubr packages.

Results
Poor fetal growth is associated with placental aneuploidy Aneuploidy screening was performed in 207 placentas from the Vancouver cohort and 67 placentas passing microarray quality checks from the Toronto cohort. Amongst 173 control placentas, no cases of CPM or autosomal aneuploidy were detected. Two (1.1%) controls had constitutional abnormalities involving the sex chromosomes ( Table 2), one of which only impacted Yqter. In contrast, amongst 101 SGA cases, 12 (11.9%) had a whole or partial autosomal trisomy present in the placenta (Table 2) (p = 0.00017, OR = 11.4; Fisher's exact test). These were found both in cases of isolated SGA and SGA in association with maternal PE. Of the cases with successful follow-up (10/12), all abnormalities in SGA placentas were determined to be CPM based on microsatellite genotyping ( Table 2). Four of these cases were previously published 19 , however 8 are new and con rm that CPM is a signi cant cause of SGA. Of the 9 cases with available maternal DNA, uniparental disomy (UPD) was excluded in the diploid cell population from all but one previously-published case with CPM for trisomy 2 and probable upd(2)mat (16). The incidence of aneuploidy did not differ between cohorts (2/136 vs. 0/37 controls and 7/71 vs. 5/30 SGA in the Vancouver and Toronto cohorts, respectively). Overall, our cohorts had high maternal ages (Table 1), and among the SGA cases, maternal age tended to be higher in pregnancies associated with CPM than those without a placental aneuploidy (Supplementary

Load of CNVs does not differ between SGA and control placentas
To explore the role of placental CNVs in in utero growth, 114 euploid placentas from control and SGA newborns were assessed using high-density microarrays (Fig. 1). We found one SGA case (PM324) with mosaicism for 8 large 2-4 Mb duplications in the placenta (Supplementary Fig. 3). As the combined level of aneuploidy exceeded 27 Mb, it was an outlier that was excluded it from further comparisons, so as to not bias results; we instead considered it as an additional case of placental segmental aneuploidy. Due to signi cant differences in load of CNVs between the different array platforms (Supplementary Table 4), we performed case-control comparisons within each cohort independently. We found no difference in total number and cumulative extent (bp) of CNVs per placenta, except for a greater cumulative bp of rare CNVs in SGA placentas in the Vancouver cohort (p = 0.03, Mann-Whitney U test) ( Table 3). When comparing these measures by gains and losses separately, there were also no signi cant differences (Table 3). As larger CNVs are more likely to be impactful, we compared CNV size across all placentas in each group. In the Vancouver cohort, CNVs were larger in SGA placentas (p = 0.002, Mann-Whitney U test; Supplementary Fig. 4). When considering CNV gains and losses separately, only the losses were signi cantly larger (p = 0.010, Mann-Whitney U test; Supplementary Fig. 4). When separated by sex, the larger CNV sizes in SGA were signi cant only amongst females ( Supplementary Fig. 5). There were no signi cant differences between groups in the Toronto cohort.
To further assess whether SGA placentas had a greater CNV load, we compared the number of gains or losses per placenta at size bins ranging from < 15 kb to > 3 Mb in all CNVs or only in rare CNVs between groups. There were no consistent differences between SGA and control placentas. SGA placentas in the Vancouver cohort had fewer small losses (< 15 kb, p = 0.002; Mann-Whitney U-test), and those in the Toronto cohort had more large losses (500 kb-1 Mb, p = 0.001; Mann-Whitney U-test). Both of these ndings withstood multiple test corrections at a Bonferroni-corrected p-value threshold of p = 0.005, but were not observed in rare CNVs (Fig. 2).

Candidate CNVs identi ed in SGA placentas
We next focused on rare CNVs ≥ 200 kb as these are most likely to contribute to the SGA phenotype. There were 34 large rare CNVs present in SGA placentas and 53 in controls. CNVs with potential roles in placental function and/or fetal growth were identi ed 5.7% (3/53) of SGA placentas but not in controls (0/61). The 3 candidate CNVs were categorized as variants of uncertain signi cance (VUS)-likely pathogenic and impact the functionally relevant genes IHNBB, HSD11B2, CTCF, and CSMD3 (Table 4). These were con rmed by qPCR to be present in both placenta and cord blood, thus were not con ned to the placenta. To investigate potential impact of CNVs, we compared the number of genes involved in CNVs per case. We found no differences in the Vancouver cohort, however there was a trend for a greater number of genes affected by losses in SGA placentas in the Toronto cohort (p = 0.049, Mann-Whitney U-test; Fig. 3). There were no signi cant differences when focusing on rare CNVs.
We did not nd an enrichment of genes with enhanced placental expression in SGA CNVs, however there were more losses of placental-enhanced genes in controls in the Toronto cohort (p = 0.02, Fisher's exact test; Supplementary Table 5) that was not reproduced in the Vancouver cohort. Gains impacting ERVV-1 and ERVV-2, and CNVs impacting several PSG family genes, a region known to be copy number variable in the human population(39), were common in both cases and controls.
We did not nd any signi cant enrichment of imprinted regions in placental CNVs from SGA cases (Supplementary Table 6). Several common CNVs impacting imprinted regions were recurrent, including placental imprinted DMRs for SPRN and CYP2E1 (Supplementary Table 7). CNVs deemed as rare were also recurrent, including gains impacting KCNK9 and the DMR near PRMT2 (Supplementary  Table 8).

Discussion
In this study, we investigated the contribution of genomic imbalances in the placenta to poor fetal growth.
In our otherwise low-risk population, we found that CPM involving trisomy or large segmental aneuploidy was present in 11.9% of SGA cases, or 12.7% when including the case with duplications totaling > 27 Mb.
While the incidence of placental aneuploidy associated with SGA in this study is comparable to past reports (13,15,16), it is expected to be population-dependent. The frequency of trisomy, and thus CPM, increases with advanced maternal age, which is also a risk factor for SGA. Indeed, we found that maternal age tended to be higher in SGA pregnancies with CPM (mean: 36.7 y) than those that were chromosomally-balanced (mean: 35.0 y). Conversely, CPM should contribute to fewer cases of SGA in populations with high rates of other risk factors for SGA, such as maternal smoking or poor nutrition (43,44). A higher CPM incidence is also expected using a stricter de nition of FGR rather than SGA, e.g. fetal weight < 3rd percentile or by using biomarkers like placental growth factor (PlGF) in maternal serum that are predictive of placental-mediated FGR (45). Although we could not measure maternal PlGF levels, our SGA group was likely enriched for cases of pathological growth restriction as a large proportion of cases were < 3rd percentile (68% Toronto cohort, 48% Vancouver cohort) and the majority of cases in the Vancouver cohort met criteria for FGR.
Overall, we could not con rm previous reports nding decreased (19) or increased (20) load of CNVs in SGA placentas compared to controls. Small sample size may explain these discrepancies, as both past studies had < 10 cases per group. With greater sample size and low incidence of other risk factors in our population, we were well poised to detect genetic contributors to SGA. Although we identi ed trends that suggest that some SGA placentas have an increased load of large CNVs, our ndings did not support that placental CNVs commonly contribute to SGA. We also did not nd signi cant differences in number of total or placental-expressed genes or imprinted regions in CNVs, which also suggests that either these are not major drivers of poor fetal growth in our cohort or their effects are subtler than we had power to detect.
Nonetheless, a candidate VUS-likely pathogenic germline CNV was identi ed in 5.7% of SGA placentas in this study. This incidence is similar to past studies of prenatal samples, which identi ed pathogenic CNVs in 3-7% of cases of isolated FGR with normal karyotypes (17,18,46). Case 7665 has a duplication of INHBB, which encodes a subunit for the activin and inhibin proteins that play important roles in trophoblast growth and invasion (47,48), and altered mRNA or protein levels of these compounds are associated with miscarriage, severe PE, and FGR (49). Case 6234 has a deletion encompassing HSD11B2 and part of CTCF. HSD11B2 is highly expressed in placental trophoblast cells, and encodes 11β-HSD2, which regulates fetal exposure to maternal glucocorticoids (50). Reduced placental gene expression or protein levels has been associated with FGR (51-54), and patients with rare mutations in HSD11B2 have signi cantly lower birth weight (55). CTCF is a highly-conserved transcription factor, and rare loss-offunction variants or deletions of the gene are associated with low birth weight, postnatal growth retardation, microcephaly and intellectual disability (56). Case 10506 had a 3 Mb deletion encompassing CSMD3, which is reported to be intolerant to loss-of-function variants (upper bound o/e = 0.3 in gnomAD (57)), and Csmd3 knockout mice display lower body length and body fat (58).

Strengths and limitations
This is the rst study to our knowledge to characterize both aneuploidy and copy number variants in the placenta in association with poor fetal growth. It also contributes the largest sample evaluated for the association between placental CNVs and SGA to date, incorporating rigorous data processing using wellestablished pipelines, and several thorough lines of investigation. Due to the retrospective nature of this study, differences exist in clinical characteristics and methodologies between the cohorts and are a limitation of the study. Certain exclusion criteria used in the Toronto cohort were not available in the Vancouver cohort (e.g. infection during pregnancy), therefore we could not exclude such cases.
Additionally, the aneuploidy screening methods used were not fully equivalent, as MLPA cannot detect large interstitial duplications or deletions. Despite this, the Vancouver and Toronto cohorts had similar clinical characteristics (Table 1, Supplementary Table 1) and the methods to screen for aneuploidy all accurately identify whole chromosome or chromosome arm abnormalities, therefore we combined the cohorts to improve our power to establish the contribution of placental aneuploidy to SGA. We were unable to combine the two cohorts to study CNV load associated with SGA due to the signi cant differences between the high-density microarrays used for CNV detection. However even when assessed separately, each cohort had adequate power to identify differences at the large effect sizes described in previous reports (19,20), and testing the two cohorts independently gave us the opportunity to assess the reproducibility of our ndings.

Research and clinical implications
An appreciation for the impact of placental aneuploidy on SGA/FGR is relevant for both research and clinical applications. For studies investigating the etiology of idiopathic SGA/FGR, excluding cases explained by CPM may increase the power of association studies. When identi ed prenatally, CPM may signify that the pregnancy is at increased risk for complications depending on the extent of the abnormality and the chromosome(s) involved. For example, CPM of trisomy 8 has low risk of complications (59), while that of trisomy 16 is associated with a high risk for FGR and PE (10,(60)(61)(62).
Additionally, there is an increased risk of UPD in the diploid cell population which can be associated with imprinting disorders; for example, upd (7)mat and upd (20)mat are associated with FGR and several longterm health complications (63,64). Reassuringly, follow-up studies of cases of CPM without UPD suggest that most growth-restricted infants tend to have catch-up growth, normal neurodevelopment, and no global developmental delay (41,(65)(66)(67). Identifying cases that were growth-restricted due to CPM can inform further long-term outcome studies, particularly in relation to speci c trisomies, to improve our understanding of the developmental trajectories and risks for complications in affected infants, and address the clinical utility of screening for CPM and UPD in cases of FGR.
Our ndings also provide evidence that CNVs impacting genes relevant to growth or placental function may contribute to idiopathic SGA. In contrast to ndings of aneuploidy CPM, the CNVs identi ed in our study were germline alterations and may therefore have clinical implications beyond birth. Future studies pro ling CNVs associated with SGA or FGR may add to ours and improve the annotation of CNVs found in cases of obstetric complications, for which information is largely absent in population databases.
Given the widespread use of non-invasive methods to detect placental DNA in maternal blood and the development of methods to identify CNVs from these samples (68)(69)(70), the feasibility of identifying pathogenic CNVs prenatally is increasing. This will have relevant implications for both predicting pregnancies at risk of FGR and its associated complications and for post-natal counselling if CNVs are not con ned to the placenta. Additional research on the incidence and impact of CNVs on obstetric outcomes is thus needed to assess the potential clinical utility of this information.

Conclusions
Overall, we nd consistent evidence that trisomy and segmental aneuploidy con ned to the placenta are an important cause of poor fetal growth, and that rare germline CNVs overlapping genes of functional interest may also underlie a subset of idiopathic SGA cases.

Availability of data and materials
The data that support the ndings of this study are available from the corresponding author upon reasonable request.

Competing interests
The authors report no con ict of interest. Author's contributions GFDG contributed to the study design, prepared samples for microarray, performed microsatellite genotyping and data analysis, interpreted results, and wrote the manuscript draft. YY and EAB contributed to data analysis. JW performed initial data processing. SC prepared samples for microarray analysis and collected clinical information. RW provided samples and clinical information. PVD and HB contributed to subject recruitment. ERS contributed to study design, and interpretation of results. RKCY and WPR conceived and supported the study and contributed to the data analysis and interpretation of results. All authors read and provided critical revisions of the manuscript and approve the nal version.  Total number of genes impacted by placental CNVs from control and SGA pregnancies. The cumulative total of unique RefSeq genes impacted by CNVs for each case in the Toronto and Vancouver cohorts are shown, separated by all CNVs or exclusively rare CNVs, and by gains and losses. Toronto cohort SGA placentas had slightly more genes affected by losses than controls. A similar trend was found in Vancouver cohort, but the difference was not statistically signi cant. p-values calculated by Mann-Whitney U-test.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. AdditionalFile3SuppFigure15.pdf AdditionalFile2SuppTables18.xlsx AdditionalFile1SuppMethods.pdf