METHODS

2.1 Study design

The VIRTUAL Placenta cohort was embedded in the ongoing prospective Rotterdam Periconception Cohort 21, 22. Between January 2017 and March 2018, women who were at least 18 years old, carried a singleton pregnancy <10 weeks gestational age (GA), and gave written informed consent were recruited from an academic hospital. Both naturally conceived pregnancies and pregnancies achieved via in vitro fertilization (IVF) with or without intracytoplasmic sperm injection (ICSI) were eligible for inclusion. Pregnancies achieved via oocyte donation and miscarriages were excluded from analyses. At enrolment, participants filled out a questionnaire on general characteristics, medical and obstetrical history and lifestyle behaviours, and a Food Frequency Questionnaire (FFQ).
For all participants, two or more study visits were scheduled in the first trimester at 7, 9 and 11 weeks GA during which 3D PD transvaginal ultrasound scans of the whole gestational sac including the placenta and utero-placental vasculature were obtained using the GE Voluson E8 (GE, Zipf, Austria). Standardized ultrasound settings were previously described (quality: max; pulse repetition frequency (PRF): 0.6; wall motion filter (WMF): low1; compound resolution imaging (CRI): off; power Doppler (PD) gain: -8.0) 19. Ultrasound examinations were performed according to international guidelines on safe use of Doppler ultrasound in the first trimester of pregnancy (ALARA-principle)23.
At the first study visit, height and weight were measured according to protocol to calculate the body-mass index (BMI). Pregnancy outcomes were collected through a questionnaire filled out by the participant within 1 month after giving birth and complemented with medical delivery records.

2.2 Pregnancy dating

For naturally conceived pregnancies in regular cycles (25-35 days), GA was calculated from the first day of last menstrual period (LMP). In case of unknown LMP or irregular cycle, GA was calculated from Crown-Rump-Length (CRL). If the two methods varied >6 days, the CRL-based GA was assumed the true GA. For fresh IVF/ICSI pregnancies, GA was calculated from oocyte pick-up day +14 days. In case of cryopreserved embryo transfer, GA was calculated from transfer date +19 days.

2.3 Periconceptional maternal dietary intake

We used a standardized semi-quantitative food frequency questionnaire (FFQ) validated for women in the reproductive age 24. The FFQ consists of 191 food and beverage items and collects detailed information about dietary intake, the frequency of consumption, portion size and method of preparation over the previous four weeks. Energy and nutritional intake of each food item was determined with the Dutch food composition table by Wageningen University.
First, we extracted total daily energy intake (kcal/day) from the FFQ. Using the Goldberg cut-off, designed for an average population as described by Black 25, participants reporting an unrealistically low value of energy-intake were excluded from analysis.
Next, we calculated the percentage energy intake (PEI) of each food item. Then, using the NOVA classification, each food item in the FFQ was categorized as ‘unprocessed or minimally processed food’, ‘processed culinary ingredient’, ‘processed food’ or ‘ultra-processed food’9. The classification of all items was performed by three researchers independently. In case of discrepancies, items were discussed with a nutritional epidemiologist until consensus was reached. Hereafter, we calculated the percentage of energy intake from ultra-processed food consumption (PEI-UPF, %) for each participant.
To assess the intake of macronutrients, we used the FFQ to calculate the total daily intake of carbohydrates, proteins and fats (g/day). In addition, we calculated the total daily intake of macronutrient compounds, for which we distinguished between mono-/disaccharides and polysaccharides, animal proteins and plant-based proteins, and saturated fatty acids and unsaturated fatty acids (g/day).
To identify distinct dietary patterns, we first reduced all 191 food items into 25 food groups based on similarities in origin and nutrient content, which we adapted from the European Prospective Investigation into Cancer and Nutrition (EPIC) project 26, see Table S1. All food groups were entered in a principal component analysis (PCA) to identify dietary patterns (principal components) based on the degree of reciprocal correlation between specific food groups. We extracted dietary patterns with eigenvalues >1.0 and used a scree plot to only select dietary patterns that explain a large proportion of the variance in the food groups and exclude the residual components27. We provided a nutritional summary per dietary pattern. The PCA automatically calculated a factor loading for each food group, showing the extent to which that specific food group is correlated with each dietary pattern. Finally, participants received a factor score representing their adherence to each dietary pattern.

2.4 Imaging markers of first-trimester utero-placental vascular development

Image quality was scored on a four-point scale ranging between zero (optimal) and three (unusable) based on the presence of artefacts, the ability to distinguish between myometrium and trophoblastic tissue, and completeness of the placenta. Images with a quality score of three were excluded from the analyses.
The placental volume (PV) was measured using VOCAL software according to the previously published validation study 28. In short, the placental outline and gestational sac contours were repeatedly traced in rotational steps of 15 degrees to calculate total pregnancy volume and gestational sac volume respectively. The gestational sac volume was subtracted from the total pregnancy volume to calculate PV (cm3) 28.
The utero-placental vascular volume (uPVV) was measured using a virtual reality (VR) desktop system with the V-Scope volume rendering application. First, the threshold for 8-bit Doppler magnitude data was set at a value of 100 and PD artefacts were removed with a virtual eraser. Then, VR segmentation was used to erase the Doppler signal in the embryo, the umbilical cord and the uterine tissue surrounding the placenta (Figure S1A-B). The V-Scope application automatically calculated the volume of all remaining PD voxels to measure the uPVV (cm3), a volumetric vascular characteristic, as published previously 19 (Figure S1C).
The utero-placental vascular skeleton (uPVS) was generated by applying a skeletonization algorithm to the uPVV segmentations20. The skeletonization algorithm repeatedly peels off the outermost layer of voxels from the uPVV, reducing the diameter of the PD signal at each point in the vascular network until one central voxel remains, thereby creating a network-like structure representing the vascular morphology (Figure S1D) (18). Following the construction of the network, the skeletonization algorithm classifies each 26-connected voxel based on the number of neighbouring voxels as endpoint (n) (1 neighbour), bifurcation point (n) (3 neighbours), crossing point (n) (4 neighbours) or as normal vessel point (n) (2 neighbours). Voxels with >4 neighbours are considered an anomaly and excluded from analyses. Further, the algorithm measures total network length and average vascular thickness (mm) (Figure S1E). The 6 uPVS characteristics represent absolute morphologic development of the first-trimester utero-placental vasculature. Also, we calculated ratios of the uPVS end-, bifurcation- and crossing points to the uPVV (n/cm3) to identify 3 imaging markers to represent the density of vascular branching in the utero-placental vascular volume. Women who had no PV, uPVV or uPVS measurement available were excluded from analysis.

2.5 Statistical analysis

Baseline characteristics were presented as mean with standard deviation. If needed, non-volumetric parameters were transformed using a square root transformation to approximate a normal distribution. For volumetric parameters and ratios a cubic root and natural log transformation were used, respectively.
We used linear mixed models to estimate the association between maternal intake of PEI-UPF, total energy, macronutrients and their compounds and dietary patterns, and imaging markers of utero-placental vascular development, assessed with PV, uPVV and uPVS morphologic and density characteristics. We constructed three different models to explore the potential effects of confounding: model 1 (adjusted for gestational age only); model 2 (model 1 additionally adjusted for maternal age, BMI, parity, conception mode, foetal sex and periconceptional alcohol consumption, smoking and folic acid supplement use); and model 3 (model 2 additionally adjusted for total energy intake). Possible confounders were selected based on literature and discussion amongst authors using a directed acyclic graph.
All analyses were performed using SPSS (version 25.0; SPSS Inc., Chicago, IL, USA) and R (version 4.2.2, R Core Team, Vienna, Austria, 2022). P-values <0.05 were considered statistically significant.