Advanced magnetic resonance imaging detects altered placental development in pregnancies affected by congenital heart disease

Congenital heart disease (CHD) is the most common congenital malformation and is associated with adverse neurodevelopmental outcomes. The placenta is crucial for healthy fetal development and placental development is altered in pregnancy when the fetus has CHD. This study utilized advanced combined diffusion-relaxation MRI and a data-driven analysis technique to test the hypothesis that placental microstructure and perfusion are altered in CHD-affected pregnancies. 48 participants (36 controls, 12 CHD) underwent 67 MRI scans (50 control, 17 CHD). Significant differences in the weighting of two independent placental and uterine-wall tissue components were identified between the CHD and control groups (both pFDR<0.001), with changes most evident after 30 weeks gestation. A Significant trend over gestation in weighting for a third independent tissue component was also observed in the CHD cohort (R = 0.50, pFDR=0.04), but not in controls. These findings add to existing evidence that placental development is altered in CHD. The results may reflect alterations in placental perfusion or the changes in fetal-placental flow, villous structure and maturation that occur in CHD. Further research is needed to validate and better understand these findings and to understand the relationship between placental development, CHD, and its neurodevelopmental implications.


Introduction
The placenta delivers oxygen and nutrients to the developing fetus, removes carbon dioxide and waste products, and performs a host of endocrine and immune functions.Placental development and function is key to healthy fetal development and impaired development can be linked to future health outcomes 1 , extending beyond the fetal and early neonatal period, into adulthood 2 and can even affect future generations 3 .
Congenital heart disease (CHD) is the most common congenital malformation, affecting ~ 1% of livebirths 4 , and is associated with impaired brain development and adverse neurodevelopmental outcomes, including motor, language and cognition, 5 which persist into adulthood [6][7][8] .Previous studies suggest that placental development is altered in pregnancies where the fetus has CHD: placental weight and volume can differ, depending on the type of CHD 9,10 and gross morphological changes, such as the insertion site of the umbilical cord, have been recorded 11,12 .Histological studies have identi ed differences 13,14 , including an increased incidence of thrombosis and infarction 15 and vascular malperfusion lesions 16,17 .In certain CHD subtypes, placental gene expression and nutrient transfer is abnormal 18 and MR imaging studies, although limited in number, have revealed differences in placental function 19,20 .
The fetal heart and placenta are both embryological fetal vascular organs, with shared expressed gene pathways 21,22 .It is plausible that placental vasculature may also be disrupted in fetal CHD, although the mechanisms behind exactly how remain unclear 23 .Recently, focus has shifted to the "heart-brainplacenta axis" [24][25][26][27][28] to improve our understanding of the impaired brain and placental development seen in CHD.For example, there was a trend towards more severe brain injury in neonates with CHD when placental pathology was present 13 and individuals with both CHD and placental abnormalities have signi cantly lower cognitive and motor performance scores in early childhood 29 , although understanding the exact relationship between these factors is complex.
Identifying differences in structure and/or function of the CHD placenta is now a key area for research in the fetal CHD population 23,30 , with the hypothesis that altered placentation can result in decreased cerebral oxygen delivery and thus may be associated with impaired early brain development and subsequent adverse neurodevelopmental outcomes in CHD.
Placental histopathological examination is undoubtedly helpful for identifying gross morphological and microscopic changes in the CHD placenta 31 .However, placental histopathology has been likened to performing an autopsy, where important information only becomes available after delivery, and by which time placental structure and functional properties have changed substantially [32][33][34] .There is therefore a need for techniques to quantify placental health, structure and function in-utero.
Ultrasound, the current 'gold-standard' screening tool during pregnancy, is not suitable for assessing placental function or microstructure.The complex cascade of events, from in ow of highly oxygenated maternal blood through the spiral arteries, to exchange across the syncytiotrophoblast calls for comprehensive in-utero investigations exploring factors such as perfusion and tissue oxygenation.
Placental MRI is a safe, non-invasive technique allowing such in-utero analysis, with both T2*-relaxation and diffusion imaging techniques used to assess placental function and microstructure throughout gestation [35][36][37][38][39] .T2*-relaxation exploits the blood oxygen level dependent effect linking shorter T2* values to, amongst other factors, a higher concentration of deoxygenated hemoglobin, and is often interpreted as a proxy for placental oxygenation or function 40,41 .Since placental T2* is sensitive to the balance between oxygenated maternal blood delivery and fetal oxygen demand, it is particularly well-suited to identify altered placental oxygenation or blood ow patterns, as might be seen in CHD 20,39,[41][42][43][44][45] .
Diffusion MRI is sensitive to the speed and directionality of motion of water molecules and thereby provides information relating to tissue microstructure.Importantly, the combination of an acquisition comprising multiple MR images and a mathematical model unlocks sensitivity to structures much smaller than the voxel size 46 .
Emerging combined diffusion-relaxation MRI techniques enable the acquisition of multi-modal diffusion and relaxation data in a single, e cient MR scan, instead of the conventional sequential scans.
Measuring diffusion and relaxation simultaneously allows disentanglement of distinct tissue microenvironments that cannot be distinguished with T2*-relaxation or diffusion MRI alone 47,48 .
Combined diffusion-relaxation MRI has been demonstrated in the placenta using the ZEBRA 49 technique, and shows promise for identifying placental dysfunction 48 .Additionally, an unsupervised, data-driven analysis technique known as InSpect has been developed, enabling simultaneous assessment of placental oxygenation, microstructure and microcirculation from T2*-diffusion data 48,50 .Combining such advanced MRI acquisition and analysis techniques enables in-vivo investigation of multiple microstructural and perfusion environments, such as those found in the placenta.
This study utilized advanced combined diffusion-relaxation MRI and an unsupervised, data-driven analysis technique (InSpect) to test the hypothesis placental microstructure and perfusion are altered inutero in pregnancies affected by CHD.

Results
Participant demographics 48 individual participants (36 controls, 12 CHD) satis ed the inclusion criteria, resulting in imaging data from 67 scans in total (50 control, 17 CHD).Maternal demographics for all participants are in Table 1.There was a signi cant difference in maternal age at scan between the control and CHD samples, (median age 35.0 years (33.2-37.1)vs 32.3 years (27.9-34.9)respectively, p = 0.0079).There was no difference in scan GA or BMI between groups.Diagnoses included in the CHD cohort were: Coarctation of the aorta (CoA) = 4; Tetralogy of Fallot (ToF) = 1; Transposition of the Great Arteries (TGA) = 3; Hypoplastic Left Heart Syndrome (HLHS) = 3; Truncus Arteriosus (TA) = 1.

Interpreting microenvironments
The derived T2*-ADC control spectra for each of the seven InSpect components, after running 'full InSpect' on data from 36 control participants, are shown in Fig. 2. Plots showing the mean MRI signal weighting for each ROI as a proportion of the total signal for each of the seven components across gestation, for all control participants, are also shown.
Figure 3 shows composite spatial maps for both a control and a CHD participant, acquired at comparable GAs, highlighting differences in voxelwise weightings for all seven components (rows) and all placental slices (columns).Component one has two spectral peaks, both with relatively low T2* (< 0.06s) and ADC (< 0.001mm 2 s − 1 ) values, representing poorly oxygenated tissues with lower diffusivity.The spatial maps for this component show the highest signal in the periphery of placental lobules.Component three contains some spectral peaks with higher T2* (> 0.07s) and ADC (> 0.1mm 2 s − 1 ) values and, particularly at later gestations, is conspicuously absent from within placental lobules.It has a high signal towards the edge of the placenta and adjacent uterine wall and could be interpreted as representing connective tissue structures such as placental septa, as well as blood in vasculature within the uterine wall.Component seven has multiple spectral peaks, all with relatively high T2* values (> 0.09s), re ecting well oxygenated tissues.The spatial maps show it is con ned within the placental lobules, with 'hot-spots' at the center of each lobule, and may therefore represent blood owing into the lobules via uterine spiral arteries, before being slowed abruptly as it travels through the villous tree architecture at the fetal-maternal exchange surface.
An overview of the spectral peaks, mean signal-weighting contribution, how this weighting changes over gestation and the spatial distribution for each component, used for interpretation of the underlying tissue environments, are in Supplementary Table 1.Selected mid-placental slices for all participants for each component are in supplementary Figs.1-7.

Component weighting plots
All component weighting plots are shown in Fig. 4.
After accounting for GA at scan and maternal age, there was a signi cant difference in mean ROI weightings between control and CHD groups for component three and component four (both p FDR <0.001). For

Discussion
This is the rst study utilizing combined diffusion-relaxation MRI to explore placental structure and function in-utero in CHD-affected pregnancies.We used a data-driven approach simultaneously sensitive to oxygenation, microstructure and microcirculation 50 to show that independently derived placental and adjacent uterine wall tissue environments change signi cantly during key periods of fetal development, between 20 and 40 weeks gestation, in both normal pregnancies and those where the fetus has CHD.
For multiple components, different trends over gestation for control and CHD data are observed (Fig. 4).The weighting of component three increases noticeably after 30 weeks in CHD cases, in contrast to the control sample.For component four, the increase in weighting after 30 weeks in control participants is not re ected in the CHD data.Based on their MR properties and spatial distribution (see Supplementary Table 1), component three could represent poorly perfused structures such as placental septa, as well as blood within vasculature in the uterine wall and component four may represent blood returning from the fetus and draining into maternal veins.Importantly, however, this result is independent of any interpretation of the speci c placental microenvironments these components represent.A signi cant trend over gestation in weighting for an additional independently-derived component was also observed in the CHD cohort, but not in the control cohort.
These ndings add to existing evidence that placental development is altered in CHD, and complements research using combined diffusion-relaxation MRI to identify placental compartments with distinct T2*-ADC combinations 51 and abnormal placentation associated with pregnancy-related conditions PE or FGR 48,52 .The results reported here have the potential to help with understanding of the interlinked pathways between placental and cardiac development 22 .
One hypothesis is that changes observed here may re ect alterations in placental perfusion seen in CHD 43,53 .However, given the small difference in mean placental T2* between groups, and the difference in trajectories between T2* and the weighting of components three and four over gestation, it is unlikely that reduced perfusion alone is driving these differences.Changes in fetal-placental ow 54 and villous structure 18 that occur in CHD may also be contributing.As pregnancy advances, speci c microstructural changes occur within the placenta, particularly in the third trimester 55 , including terminal villi development 56 and fetal villous angiogenesis 57 .This is an important adaptation that ensures e cient oxygen and nutrient exchange in the later stages of gestation, to meet fetal demands, but as terminal villous development is directly in uenced by placental oxygen levels in normal pregnancy 58 , this process may be altered in CHD.Altered villous maturation, consistent with an 'immature' placental microvasculature, could also be preventing maximal oxygenation of fetal blood in CHD 23 .It is interesting to note that the GA after which the differences in the weighting of components three and four between groups becomes most apparent − 30 weeks -is consistent with the GA at which volumetric brain development also deviates from normal in fetuses with CHD 59 .
The approach we have used involves no a-priori understanding of different placental compartments or microstructural environments, but identi es them based on shared T2*-ADC characteristics and an understanding of placental structure (Fig. 5).The fetal circulation is intra-capillary and has a relatively low oxygen saturation at the exchange surface, whereas the maternal circulation enters the placenta as a highly-saturated blood pool, but is extravascular in the human placenta 47 .Despite this complex, heterogeneous structure, previous studies have used mean whole-placental MRI biomarkers 60,61 , making interpretation of the results challenging.Identifying unique placental tissue compartments that might be altered in CHD ts the inherent complexity of the placenta and thus helps provide a focus for future research studies.The associated spatial maps aid in localisation of these compartments, helping differentiate tissue environments as they change throughout the placenta, i.e. from basal to decidual plate.
Additionally, the complexity of placental physiology bene ts from a comprehensive assessment approach, such as this combined use of diffusion-relaxation and a bespoke analysis tool like InSpect.Whilst a small but signi cant decrease in whole placental T2* was identi ed in the CHD cohort, there was no signi cant difference in whole placental ADC values between groups.This further highlights the enhanced sensitivity of InSpect to detect changes in placental function and microstructure beyond the use of T2* or ADC independently.
However, it is important to emphasize that using a data-driven approach means components don't have to neatly de ne placental compartments such as "fetal" and "maternal", or "intracellular" and "extracellular", but can also re ect combined tissue environments.Placental anatomy means tissue environments with different T2*-ADC properties can sit in close proximity, i.e. at the villous exchange surface, where pooled maternal blood lies close to fast owing fetal blood in small diameter vessels, or at the umbilical cord insertion site, where similar sized vessels containing blood with very different T2* properties intertwine.
It is also worth highlighting that InSpect provides the proportion or weighting of the MR signal in each voxel that each component represents.This results in component weightings that are intrinsically linked, and may explain why there appears to be such a 'reciprocal' change in components 3 and 4, since a reduction in the weighting for one component necessitates an increase in another.This makes it challenging to interpret whether both components are affected, or just one, and this might re ect underlying changes in placental microstructure or perfusion.However, this result is independent of any interpretation of the speci c placental microenvironments and suggests a clear difference in at least one compartment.Furthermore, any changes in component weighting, i.e. those seen over gestation, or between groups, suggests that the tissue environment(s) represented by that component change as a proportion of total placental volume over gestation, and not necessarily that there is a change in absolute volume.
As expected, we show that placental volume increases with advancing gestation.Consistent with previous work, we also identify a decrease in both T2* and ADC values with advancing gestations 39,40,62 , that appear consistent in healthy participants undergoing two scans 63 .
This study is limited by relatively small numbers at early gestations.The CHD cohort is also diagnostically heterogenous, and different diagnoses may impact placental development or function in different ways 13,15 .However, all CHD diagnoses were critical or serious, which is where the greatest alterations in placental development in CHD might be expected 12,64 .Different types of CHD may affect fetal-placental ows in different ways 43,65,66 , or involve different genes linking placental and vascular development 18 , so future work with larger cohorts is needed to explore how certain CHD diagnoses or physiologies might be associated with impaired placental development.
Using a data-driven technique such as InSpect involves speculation as to the underlying tissue environments or microstructures represented.There is currently no 'ground truth'.We also did not collect ultrasound information, such as that relating to uterine artery resistance or dopplers, which may be helpful for establishing the presence of uteroplacental dysfunction 43 .Future work should involve attempts to validate and better understand these ndings, either through histopathological examination, by invasive sampling 67 or in comparison with complementary ultrasound techniques 68 .
Future work including data from other cohorts where placental dysfunction is better understood/characterised, for comparison to both CHD and control placentas, would also be bene cial.For example, others have previously identi ed associations between fetal CHD and maternal hypertensive disorders of pregnancy 69,70 , including the risk of pre-eclampsia 16 , hinting at a potential common etiology, which could be explored using this approach in future.
Furthermore, we did not collect data related to maternal or fetal haematinics.However, it is plausible that levels of fetal or maternal haemoglobin in uence placental T2* values.Given that maternal haemoglobin levels change over gestation 71 , and that fetal haematinics are affected by both impaired placentation and CHD 72,73 , future work should attempt to capture and include this information.
The trends over GA of several component weightings, seen in both CHD and control placentas, could re ect normal changes in the microstructure the placenta as pregnancy advances 31,32,74 , with the corresponding differences in tissue microstructure and perfusion that occur 75 , and in future could serve as imaging biomarkers of both normal and abnormal placental development.Given enough data from typically developing placentas, InSpect could also be used to generate a quantitative 'placental abnormality' score, taking into account the deviation from normal of each component weighting for a given GA.This would enable both quanti able analysis of impaired placentation, and help identify where within the placenta this occurs.

Conclusions
We report using combined diffusion-relaxation MRI and a data-driven approach to detect altered placental tissue environments in pregnancies affected by fetal CHD, with changes most evident after 30 weeks gestation.We speculate that these changes are driven by impaired perfusion and microstructure in the CHD placenta, although future work is needed to de nitively link these imaging ndings to potential alterations in the underlying placental structure and function.

Ethics and Recruitment
Data were acquired as part of The Congenital Heart Disease Imaging Programme (CHIP) at St. Thomas' Hospital in London.All methods were carried out in accordance with relevant guidelines and regulations and all experimental protocols were approved by a named institutional and/or licensing committee [NHS REC 21/WA/0075].Control participants experiencing a low-risk pregnancy, with the absence of pregnancy-induced hypertension (PIH), preeclampsia (PE), fetal growth restriction (FGR), or gestational diabetes (GD) at the time of enrolment, were recruited after their antenatal booking or screening appointments.Participants with a fetus with severe or critical CHD, as de ned previously 76 , con rmed on fetal echocardiography, were recruited from the fetal cardiology clinic.Participants with PIH, PE, FGR, GD, or where the fetus had con rmed genetic abnormalities were also excluded from the CHD cohort.All participants were invited to have up to two fetal MRI scans.
Data were subsequently excluded if the pregnancy resulted in a delivery before 37 weeks gestational age (GA), if PIH, PE, FGR or GD were newly diagnosed between scan and delivery, if any genetic abnormalities were detected on antenatal testing, or if any signi cant incidental fetal or placental ndings were reported on imaging.Data sets with insu cient quality, including cropping of the placenta, extensive geometric distortion artifacts, or visible contractions during the scan were also excluded.

Image acquisition and reconstruction
Informed, written consent was obtained from all subjects prior to imaging.Images were acquired on a Philips Achieva 3T scanner using a 32-channel surface coil.All imaging was performed in supine position with frequent verbal interaction, continuous heart rate and oxygen saturation monitoring, and bloodpressure measurements at ten minute intervals.
Following a pilot scan and B0 and B1 calibration scans, anatomical imaging using T2-weighted turbospin-echo sequences, as well as a multi-echo gradient-echo sequence, a combined T2*-diffusion scan (ZEBRA 49 ) was performed, with parameters de ned in Table 3. Acquisition time for this sequence was 8m30s.
The acquired data were then anonymised and reconstructed using in-house tools, including denoising and motion-correction, as previously described 49 .The reproducibility of this T2*-diffusion sequence has been demonstrated in MR phantom, adult brain, and placental studies 49,63 .with the lowest echo time, by an experienced clinician, who was blinded to the maternal demographics and fetal diagnosis.The T2*-ADC model described in Eq. 1 was then t voxelwise using a modi ed version of the diffusion microstructure imaging in python toolbox 77 , as described in 63 , enabling calculation of the mean T2* and ADC values for the whole ROI.

Data-driven analysis with InSpect:
We ran InSpect on the rst scans from all 36 control participants using the InSpect toolbox (https://github.com/PaddySlator/inspect)(Fig. 6).Seven InSpect components were xed as this number has previously been shown to best explain the placental and adjacent uterine-wall T2*-diffusion signal 50 .This full InSpect is a process akin to independent component analysis, albeit under the assumption that the data is generated by the underlying dynamics of Eq. 1. Full InSpect hence identi ed seven components in the data, with each component having a corresponding T2*-ADC spectra (e.g. Figure 6).
The spectral peaks in these T2*-ADC spectra represent different tissue microenvironments within the ROI.InSpect has no a-priori information about the tissue or organ being imaged.For placental imaging, this full InSpect analysis is analogous to a 'reverse-recipe': it takes T2*-ADC data as input, and infers information about the unique tissue microenvironments (or components) that are required to 'make' each placenta as the output.The relative weighting of each of these components is calculated voxelwise during this full InSpect process, allowing maps quantifying the spatial distribution and relative amount of each component present in every voxel to be created (e.g. Figure 6).These components and their corresponding T2*-ADC spectral peaks in data from control participants were assumed to be representative of typically developing placentas.We then quanti ed how the relative fractions of these components change over GA.
where S 0 is the signal at the proton density (b=0), T E is the echo time, T E is the shortest echo time acquired, is the effective transverse-relaxation time, b is the b-value and ADC is the apparent diffusion coe cient.
These spectra were then used to infer voxelwise spatial maps for all CHD scans and additional 'repeat' control scans, in a process we term 'reduced InSpect'.Reduced InSpect xes the values of the T2*-ADC spectra associated with each component, only calculating the corresponding maps, ensuring the components are identical for all data being analyzed with the additional bene t of computational e ciency.Continuing the previous analogy, this corresponds to using 'reduced InSpect' to quantify the proportions of each of these prede ned components in each individual placenta.The results of these analyses were used to determine how much each component differs from normal for each CHD dataset.

Interpreting microenvironments and plotting component weightings
Next, the overall contribution, or weighting, of each component of the InSpect analysis was plotted against GA for each dataset.The MR tissue properties of each component were then assessed, taking into account their T2*-ADC spectral peaks, with relatively higher T2* values representing more welloxygenated tissue and higher ADC values (above free-water = 0.3mm 2 s − 1 ) representing perfusing or fastowing blood, usually interpreted as within vasculature.Combining this information with the MR signal weighting contributed by each component, how the weightings change over gestation, and how the components are spatially distributed, enabled speculations about the tissue microenvironments encoded by each component to be made.

Statistical analyses
A Shapiro-Wilk test was used to test normality.An ANCOVA was used to compare placental volume, T2*, ADC and InSpect derived component-weightings between groups, after accounting for gestational age at scan and maternal demographics.Pearson's correlation coe cient was calculated to determine the direction and strength of trends between continuous variables across the GA ranges studied.For analyses of component weightings, Benjamini and Hochberg false discovery rate (FDR) was applied to correct for multiple comparisons (reported as P FDR ).P FDR -values < 0.05 were considered signi cant.All statistical analyses were performed using statsmodels v0.13.2 78 and Jupyter Notebook, python3.

Declarations Figures
Figure6outlines the data-processing work ow.First, a region of interest (ROI) containing the whole placenta and adjacent uterine wall section was manually segmented on the anonymised rst b = 0 image

Plots
showing mean placental and adjacent uterine wall T2* (top left), ADC (top right) and volume (bottom left) over gestation for all scans.Dotted lines join participants who had two scans.

Figure 3 Whole
Figure 3
-* Values given are mean (± standard deviation) for parametric data, or median (25th centile-75th centile) for non-parametric data P-value from ANCOVA between control and CHD cohort, after accounting for gestational age at scan and maternal age.Results in bold are considered signi cant.