Ultrasound detection of altered placental vascular morphology based on hemodynamic pulse wave reflection

Here, we describe a novel approach to noninvasively detect microvascular changes in the fetoplacental circulation using ultrasound. The technique is based on detecting reflection pulse pressure waves that travel along the umbilical artery. Using a proof-of-principle study, we demonstrate the feasibility of the technique in two strains of experimental mice.

Abstract

Abnormally pulsatile umbilical artery (UA) Doppler ultrasound velocity waveforms are a hallmark of severe or early onset placental-mediated intrauterine growth restriction (IUGR), whereas milder late onset IUGR pregnancies typically have normal UA pulsatility. The diagnostic utility of these waveforms to detect placental pathology is thus limited and hampered by factors outside of the placental circulation, including fetal cardiac output. In view of these limitations, we hypothesized that these Doppler waveforms could be more clearly understood as a reflection phenomenon and that a reflected pulse pressure wave is present in the UA that originates from the placenta and propagates backward along the UA. To investigate this, we developed a new ultrasound approach to isolate that portion of the UA Doppler waveform that arises from a pulse pressure wave propagating backward along the UA. Ultrasound measurements of UA lumen diameter and flow waveforms were used to decompose the observed flow waveform into its forward and reflected components. Evaluation of CD1 and C57BL/6 mice at embryonic day (E)15.5 and E17.5 demonstrated that the reflected waveforms diverged between the strains at E17.5, mirroring known changes in the fractal geometry of fetoplacental arteries at these ages. These experiments demonstrate the feasibility of noninvasively measuring wave reflections that originate from the fetoplacental circulation. The observed reflections were consistent with theoretical predictions based on the area ratio of parent to daughters at bifurcations in fetoplacental arteries suggesting that this approach could be used in the diagnosis of fetoplacental vascular pathology that is prevalent in human IUGR. Given that the proposed measurements represent a subset of those currently used in human fetal surveillance, the adaptation of this technology could extend the diagnostic utility of Doppler ultrasound in the detection of placental vascular pathologies that cause IUGR.

NEW & NOTEWORTHY Here, we describe a novel approach to noninvasively detect microvascular changes in the fetoplacental circulation using ultrasound. The technique is based on detecting reflection pulse pressure waves that travel along the umbilical artery. Using a proof-of-principle study, we demonstrate the feasibility of the technique in two strains of experimental mice.

abnormalities of the fetoplacental blood vessels are the most common pathological finding in placentas from pregnancies complicated by intrauterine growth restriction (IUGR) (14, 41). These abnormalities are part of the diagnostic category termed maternal vascular malperfusion and comprise infarction of villous tissues, arterial wall thickening, and reduced elaboration of the fetoplacental vascular trees including the distal villi and decreased vessel diameters, villous volumes, arterial number and branching (17, 27, 28, 31, 38). These progressive morphological defects act to restrict O₂ and nutrient transfer and thus have profound consequences for fetal growth, possibly resulting in stillbirth or significant adverse health outcomes postnatally (11, 41). As such, accurate evaluation of the integrity of the fetoplacental circulation is a critical component in the diagnosis and management of IUGR.

One of the most important methods for identifying fetuses at risk for IUGR is umbilical artery (UA) Doppler ultrasound (14, 51), where downstream placental vascular defects are associated with absent or reversed UA end-diastolic blood velocity waveforms (4, 27). These types of abnormal waveforms are typically seen in early onset IUGR, whereas the more common late onset IUGR pregnancies typically have normal UA Doppler (14, 26, 33) and are identified instead by Doppler assessment of the middle cerebral arteries (44). Clinically, the UA velocity waveforms are most commonly evaluated in terms of the UA pulsatility index (1, 5), defined as the ratio of peak-to-peak velocity to the mean velocity over a cardiac cycle; this is widely viewed as a surrogate for downstream fetoplacental vascular function and thus oxygenation of the fetus (1). This conclusion is broadly valid because pulsation in the UA is correlated with defects in the fetoplacental vessels (7, 36) and related to the downstream placental vascular resistance (1). However, a major limitation of measuring pulsatility is that it varies substantially between placental and fetal ends of the umbilical cord (52). This may be one explanation as to why the UA pulsatility index may not be sensitive to mild late onset placental vascular disease, since abnormalities in the UA velocity waveforms only begin to appear when placental dysfunction is severe (4, 13). Indeed, in pregnancies with a high risk for IUGR, current methods to evaluate placental function (based on UA velocity pattern assessment) have a poor sensitivity for detecting IUGR and fetal distress (55% sensitivity) (40). For instance, in one retrospective cohort study (12) in pregnancies where the estimated fetal weight was less than the 10th percentile for gestational age and placental abnormalities were evident in 79% of the cases, it was found that the sensitivity for Doppler indexes (the UA pulsatility index and systolic-to-diastolic ratio) in detecting placental defects was 42%.

An explanation for the poor sensitivity of UA Doppler to placental defects is that the pulsatility in the UA is not only a function of the downstream placental characteristics but also upstream characteristics starting from the contractions of the fetal heart and continuing into the umbilical cord itself. Here, we propose to overcome the current diagnostic limitation in detecting fetoplacental vascular disease by using wave mechanics theory to isolate the placenta-associated signal from observed UA Doppler waveforms. Under this framework, the observed UA Doppler flow waveform is the summation of two counterrunning waves: a forward-traveling wave (generated by the fetal heart), which propagates along the UA, and a backward-traveling wave, which is reflected back from the fetoplacental arterial tree and propagates counter to the direction of net forward flow. This reflected waveform is the combination of many individual reflected waves that arise in the downstream fetoplacental vasculature whenever there is a mismatch in the vascular impedance of the fetoplacental vessels (i.e., if a vessel bifurcates, becomes stiffer, etc.). As such, the composite UA reflected flow wave is dependent on the properties of the downstream placental network and thus has the potential to identify abnormalities in the placental vasculature.

Wave reflection methodology has been shown by others to be useful for assessing the mechanical properties of arteries in both humans (8, 23, 25, 37, 39, 53) and mice (20–22, 34, 58). The idea that reflection measurements could be a useful diagnostic marker is supported by both empirical and theoretical arguments. Measurements of arterial cross-sectional areas proximal and distal to bifurcations in major conduit arteries of the human have been found to closely match theoretical predictions for minimizing wave reflections (45). Also, the geometry of a bifurcating arterial tree structure is proposed to trap propagating waves, as a bifurcation geometry that is efficient for transmitting the wave in a forward direction is also inefficient for transmission in the opposite direction (46). Both these observations support the notion that elevated wave reflection could be associated with inefficient or pathological configurations of the downstream vascular bed. The motivation for the present work was to learn whether wave reflections are present in the UA and to determine whether this phenomenon could plausibly explain known associations between UA pulsation and placental pathology in human pregnancy.

The aim of this study was to develop a noninvasive methodology to measure UA wave reflection and determine if the reflected flow wave is sensitive to variations in placental morphology. This was assessed through a proof-of-principle experiment in which we developed a novel, ultrasound based technique to measure and compare the patterns of the UA reflected wave in two strains of mice, CD1 and C57BL/6. Mice have been widely used as a model for human placental development due to the similarities in their placental structure and genetics (9). Furthermore, previous microcomputed tomography work has shown that while the fetoplacental vascular patterning between CD1 and C57BL/6 mice is similar at embryonic day (E)15.5, the vascular patterning diverges between the two strains at E17.5 (47). Thus, we predicted that the composite UA reflected wave would differ between the two strains only in late gestation. Finally, we compared the sensitivity of the reflected wave metrics and UA pulsatility index with late-gestation interstrain fetoplacental morphological differences.

MATERIALS AND METHODS

Pregnant CD1 mice at E15.5 (23 fetuses from 4 litters) and E17.5 (30 fetuses from 7 litters) as well as pregnant C57BL/6 mice at E15.5 (22 fetuses from 4 litters) and E17.5 (26 fetuses from 6 litters) were obtained from Charles Rivers Laboratories (St. Constant, QC, Canada) and mated in house. The morning that a vaginal plug was detected was designated as E0.5. All animal experiments were approved by the Animal Care Committee of the Toronto Centre for Phenogenomics.

Animal preparation.

Anaesthesia of the pregnant mouse was induced at 5% isoflurane and maintained under 2.5% isoflurane in 21% oxygen (medical air). Isoflurane at 2.5% was chosen compared with the 1.5% isoflurane concentration often used for noninvasive mouse ultrasound imaging (61) to limit the effects of maternal respiration on UA motion and so that maternal respiration rate was reduced and longer recordings of the UA wall signal could be analyzed between breaths.

To obtain UA ultrasound recordings of CD1 and C57BL/6 fetuses at E15.5 and E17.5 time points, the pregnant dam was placed in a supine position and hair from the abdominal region was removed using Nair (Church & Dwight, Erwing, NJ) (6). Warmed ultrasound gel was then placed on the abdomen, and the body temperature of the pregnant mouse was maintained at 35–37°C using a temperature-regulated platform (61). Heart rate and respiration of the pregnant mouse were monitored throughout the entire recording session. At a constant level of 2.5% isoflurane, there was no difference in maternal heart rate between strain or gestational age (512 ± 10 beats/min). There was an effect of gestational age on the maternal respiratory rate (69 ± 3 vs. 48 ± 3 breaths/min at E15.5 and E17.5, respectively) as determined by two-way ANOVA (F value = 14.13, P < 0.01), perhaps explained by a difference in the dams’ plane of anesthesia at each gestational age.

Ultrasound imaging.

Recordings were obtained from the fetal end of the UA using a high-frequency (40 MHz) linear array transducer (Vevo 2100, VisualSonics). UA M-mode and Doppler signals were recorded at approximately the same location during a 5-s window. The UA luminal diameter change was measured using M-mode ultrasound (frame rate: 3,000 frames/s) with the ultrasound beam perpendicular to the vessel. The velocity spectrum was measured using pulsed wave (PW) Doppler (pulse repetition frequency: 4−15 kHz) and by choosing a sample volume large enough to cover the entire UA luminal region (the angle of insonation between the ultrasound beam and direction of flow was <60°). The Doppler wall filter was set between 40 and 375 Hz to account for low-frequency Doppler signals due to vessel wall motion.

Data quality control.

To account for noise and maintain physiological consistency of the M-mode waveforms, acceptance criteria were applied before and after image analysis. Data sets were excluded if during ultrasound imaging it was observed that there were high levels of maternal gasping/fetal movements or if the raw UA wall signal was clearly nonphysiological (i.e., the signal pattern was completely noncyclical) and had excessive high-frequency components.

If the M-mode wall signal met this preprocessing criteria, the following postprocessing criteria were applied to the extracted, filtered, and partitioned area waveforms:

1. Individual area waveforms that appeared to be contaminated with maternal gasps or appeared nonphysiological were removed before the alignment and averaging process.

2. Data sets were excluded if >35% of the total aligned waveforms were outside bands representing ±20% of the average area change. Visually, at this threshold, the area waveforms separated into two distinct clusters representing data sets with low and high variation between the individual area waveforms.

3. Data sets were excluded if there were not enough individual area traces to average (<4 area traces).

Image analysis.

All image analysis was performed in Python (version 2.7.3, https://www.python.org). For each M-mode acquisition, the two walls of the UA were automatically outlined based on the inner wall location where the pixel intensity crossed 95% of the maximum (Fig. 1A). The tracked luminal outline was smoothed using a low-pass, second-order Butterworth filter (from scipy.signal library in the SciPy package). The cutoff frequency of the filter was set to five times the fundamental frequency of the fetal heart. This cutoff (through visual inspection) preserved the shape of the diameter signal while removing jitter due to noise. The resulting smoothed waveform was partitioned into individual fetal cardiac cycles based on the onset of systole.

Fig. 1.

M-mode and Doppler image analysis. A: tracking of the top and bottom umbilical artery (UA) vessel walls outlined in red. The starts of systole are indicated by the yellow crosshairs. B: the centroid of the UA Doppler velocity spectrum outlined in red. The yellow arrows indicate the starts of systole. C: from the perpendicular distance between the UA top and bottom walls, the cross-sectional area across time was calculated for each fetal cardiac cycle. The area waveforms were aligned to have a uniform length (in gray) and were subsequently averaged to obtain a single area waveform (in red). D: the extracted velocity for each fetal cardiac cycle was aligned (in gray) and averaged (in red). E: scaled area (A_m) and flow (Q_m) waveforms after baseline subtraction. F: calculation of the forward (Q_f) and reflected (Q_r) waveforms.

A specific challenge to applying this method in the fetus is that the electrocardiogram is not available to determine the start of systole. To overcome this, the sharp change in the UA speckle pattern between diastole and systole was used to define the beginning of consecutive cardiac cycles on M-mode images (i.e., with increasing fetal red blood cell velocity, the decorrelation time for scattering decreases and the globular speckle pattern becomes striated) (Fig. 1A).

Diameter estimates as a function of time were converted to cross-sectional area and assumed a circular cross-section for subsequent calculations. The individual UA area waveforms from each fetal cardiac cycle were temporally aligned to have a uniform length and then averaged into a single area waveform (Fig. 1C).

Finally, the centroid between the manually specified upper envelope and lower bound of the Doppler velocity spectrum was extracted to determine the mean velocity waveform (Fig. 1B). After application of a low-pass Butterworth filter with the same properties as the filter used for M-mode recordings, the smoothed velocity waveform was partitioned into individual fetal cardiac cycles, temporally aligned, and averaged (Fig. 1D). The start of the fetal cardiac cycle was determined as the point where the blood velocity first began to rise at the end of diastole (Fig. 1B).

Wave decomposition.

From the average velocity and area waveforms, the average flow waveform was calculated as flow = velocity × area. Next, the average flow waveform was plotted against the average area waveform (QA loop) and a line was fitted (using total least squares) to 20–80% of the maximum systolic flow region in the QA loop. This initial systolic portion was assumed to be free of reflections (34). The slope of this line was calculated to obtain the pulse wave velocity (PWV). Together, the measured area and flow waves (Fig. 1E), along with their PWV, were used to decompose the observed flow waveform into its forward and reflected components (Fig. 1F) using the following equations (30, 32, 56):

$${\text{Q}}_{\text{f}}\left(t\right)=\frac{1}{2}\left\{{\text{Q}}_{\text{m}}\left(t\right)+{\text{Q}}_{\text{m}}\left(0\right)+\text{PWV}\left[A_{\text{m}}\left(t\right)-A_{\text{m}}\left(0\right)\right]\right\}={\text{Q}}_{\text{m}}\left(t\right)-{\text{Q}}_{\text{r}}\left(t\right)$$

$${\text{Q}}_{\text{r}}\left(t\right)=\frac{1}{2}\left\{{\text{Q}}_{\text{m}}\left(t\right)-{\text{Q}}_{\text{m}}\left(0\right)-\text{PWV}\left[A_{\text{m}}\left(t\right)-A_{\text{m}}\left(0\right)\right]\right\}$$

where Q_f and Q_r are the decomposed forward and reflected flow waveforms, t is time, and Q_m and A_m are the measured flow and area waveforms, respectively (as illustrated in Fig. 2).

Fig. 2.

Illustration of UA wave decomposition with exaggerated waveforms. Measured at a particular point along the UA, the observed flow and cross-sectional area waveforms represent the superposition of the forward- and backward-traveling wave illustrated at top. As the reflected flow waveform is inverted with respect to the reflected area waveform (57), the two measured waves differ as shown and provide the basis for uniquely decomposing the two waves into their forward (in orange) and reflected (in green) components. The three metrics used to summarize the decomposed waves are shown on the right: specifically, reflection coefficient, time delay, and dispersion.

The reflected placental wave was summarized in terms of the reflection coefficient (ratio of the absolute peak-to-peak variation between backward and forward waves), time delay (the time difference between the peak of the backward and forward waves), and dispersion (the difference between the full-width at half-maximum of the backward and forward wave) (Fig. 2). Furthermore, the UA pulsatility index was calculated from the extracted Doppler velocity spectrum as previously described (18). Finally, receiver operating characteristic (ROC) curves were computed to evaluate the wave reflection metrics and UA pulsatility index for differentiating between the two strains at E17.5 (C57BL/6 mice were used as the ground truth).

Statistical analysis.

The following statistical tests were performed using R statistical software (version 3.1.1, http://www.r-project.org/):

1. Two-way ANOVA was used to evaluate the interaction of strain and gestational age on wave reflection metrics and UA pulsatility index. The interaction was further analyzed by a two-tailed independent-samples t-test.

2. A two-tailed independent-samples t-test was used to determine if the wave reflection metrics differed between E15.5 and E17.5 time points within each strain.

3. A Pearson's correlation test was used to assess the extent to which factors independent of placental vascular geometry, such as the fetal heart rate and UA mean flow rate, could explain the wave reflection parameters and UA pulsatility index.

Results and barplots are reported as measn ± SE. Statistical significance was defined as P < 0.05.

RESULTS

Data quality.

Nine of the M-mode data sets had errors during data acquisition (i.e., significant image artifacts or the sample volume was not sufficiently large to see the entire lumen wall) and were excluded from analysis. Fourteen data sets did not meet the preprocessing acceptance criteria, and an additional 35 data sets were excluded following the postprocessing quality control. Therefore, after application of the pre- and postprocessing steps to the M-mode recordings, 46% of the UA scans met the quality control criteria and were included in the wave decomposition analysis. A second rater reanalyzed the M-mode data, and there was a 99% match between the two raters for exclusion of data based on these acceptance criteria. The physiological and hemodynamic parameters for the data sets that were included in the analysis are shown in Table 1.

Table 1.

Physiological and hemodynamic parameters

Strain	Gestational Age	Fetal Heart Rate, beats/min	Mean UA Velocity, mm/s	Peak UA Velocity, mm/s	Pulse Wave Velocity, m/s
C57BL/6	E15.5	280 ± 13	32 ± 3	70 ± 7	2.8 ± 0.6
C57BL/6	E17.5	232 ± 10*	29 ± 4	67 ± 8	2.5 ± 0.4
CD1	E15.5	285 ± 29	28 ± 2	60 ± 4	2.6 ± 0.5
CD1	E17.5	320 ± 10	30 ± 3	62 ± 5	3.1 ± 0.3

Data are presented as means ± SE. UA, umbilical artery.

^*P < 0.01 compared with CD1 mice at embryonic day (E)17.5 (as determined by two-way ANOVA followed by a Tukey post hoc test).

Reflection coefficient.

The average reflection coefficient showed a significant strain and gestational age interaction as determined by two-way ANOVA (F value = 4.32, P < 0.05). Post hoc analysis using a t-test showed that while the reflection coefficient was not different between the two strains at E15.5, this metric was 33% higher in late-gestation C57BL/6 mice compared with CD1 mice (P < 0.05; Fig. 3A). With respect to intrastrain effects, the average reflection coefficient was unchanged across gestation for CD1 mice. In comparison, this parameter increased by 26% in C57BL/6 mice from E15.5 to E17.5 time points (P < 0.05; Fig. 3A). Finally, for all strains and gestational ages, the within-group Pearson’s correlation between reflection coefficient and fetal heart rate was weak (r = −0.30 to 0.34; P = 0.34−0.62) and ranged from weak to moderate for reflection coefficient and mean flow rate (r = −0.53 to −0.21, P = 0.18−0.97).

Fig. 3.

Wave reflection parameters and UA pulsatility index. *P < 0.05 denotes a significant strain and gestational age interaction, as determined by two-way ANOVA. In addition, significant main effects of strain or gestational age are noted, where present, as p_strain and p_time. n = 10 C57BL/6 fetuses at embryonic day (E)15.5 and E17.5; n = 8 CD1 fetuses at E15.5 and 14 CD1 fetuses at E17.5.

UA pulsatility index.

Two-way ANOVA of the pulsatility index showed a significant strain and gestational age interaction (F value = 5.38, P < 0.05). A post hoc t-test confirmed that the pulsatility index was not different between strains at E15.5, while it was 21% higher in C57BL/6 mice compared with CD1 mice at E17.5 (P < 0.05; Fig. 3B). Between E15.5 and E17.5, the UA pulsatility index decreased by 32% in CD1 mice (P < 0.01), and this metric was unchanged in C57BL/6 mice (Fig. 3B). Finally, the within-group Pearson’s correlation between the pulsatility index and fetal heart rate was weak to moderate (r = −0.55 to 0.23, P = 0.15−0.43), whereas the pulsatility index and mean flow rate correlation varied between weak to moderate and showed a trend toward significance (r = −0.59 to −0.39, P = 0.04−0.26).

Time delay and dispersion.

Time delay showed a main effect of strain as determined by two-way ANOVA (P < 0.01), whereas the interaction between strain and gestational age was nonsignificant. Dispersion did not show a main effect of strain, gestational age, or an interaction but was different from zero (P < 0.01). Finally, within each strain, both the time delay and dispersion parameters did not differ between E15.5 and E17.5 (Fig. 3, C and D).

Reflection coefficient and UA pulsatility index sensitivity.

The reflection coefficient metric was more sensitive than the UA pulsatility index to differences in the fetoplacental networks of CD1 and C57BL/6 mice at E17.5. The area under the curve for the ROC curve was 0.94 for the reflection coefficient metric and 0.83 for the UA pulsatility index metric (Fig. 4).

Fig. 4.

Reflection coefficient and UA pulsatility index receiver operating characteristic (ROC) curves for detecting interstrain fetoplacental vascular differences at E17.5. The area under the curve was greater for the reflection coefficient metric compared with the pulsatility index (0.94 vs. 0.83).

DISCUSSION

In the present study, we describe a novel, noninvasive method to measure and characterize UA wave reflections from the fetoplacental vascular tree in mice. Importantly, our approach was able to discriminate, in late gestation, between two mouse strains with known differences in fetoplacental vascular morphology. While catheter-based methods for detecting wave reflections in conduit arteries have been well studied (30, 57), the invasiveness of these methods has presented a significant challenge to measure reflections that originate from the fetoplacental vasculature. Anticipating that the noninvasive method presented here could have important diagnostic applications for detecting placental pathology in human pregnancy, we sought to identify those features of the reflected wave that were most affected by the morphology of the fetoplacental circulation and to infer what aspects of fetoplacental vascular morphology give rise to wave reflections.

Among the parameters that we selected to characterize the observed reflections, the reflection coefficient and time delay varied between study groups, whereas relative dispersion was unchanged. In previous studies using invasive methods to measure wave reflection in other vascular beds (24, 43), the reflection coefficient metric has been used to detect arterial disease and infer vascular properties, such as vessel stiffness. For instance, in pulmonary arterial hypertension patients, where pathological changes are often observed in the pulmonary vasculature, the reflection coefficient is more than twice that of healthy human subjects (29). In C57BL/6 mouse lungs exposed to chronic hypoxia, structural remodeling of the vasculature (i.e., increases in proximal pulmonary arterial stiffness) is thought to result in greater pulse wave reflection (55). An elevated reflection coefficient may therefore be useful as a marker of pathology in the fetoplacental circulation. As the morphology of the fetoplacental vasculature for the strains and gestational ages used here has previously been described in detail using X-ray microcomputed tomography and stereological techniques (47), we examined the relationship between the reflection coefficient and vascular morphology.

Wave reflection and capillary bed growth.

While there is no growth in the fetoplacental arterial network of CD1 mice between E15.5 and E17.5 (i.e., arterial diameters, lengths, and branching pattern are largely unchanged), the capillary volume increases substantially over this time period (47). We observed that the reflection coefficient did not change during this period for CD1 mice, suggesting that this parameter is insensitive to changes at the level of capillaries. Arterial pulsations are highly attenuated by the time they reach a capillary bed (1) such that arteriole-capillary junctions, while large in number, should make a negligible contribution to the observed reflections. This observation is consistent with previous work on the microvasculature of the rat spinotrapezius muscle, where variations in the dimensions and network topology of the capillary bed had little effect on the frequency dependence of the input impedance spectrum at the arteriolar level (16). Indeed, it has been suggested that, due to cardiac pulse attenuation, the frequency dependence of the input impedance is independent of physical properties of vessels distal to the arterioles (1).

Wave reflection and arterial geometry.

In contrast to CD1 mice, both fetoplacental arterial and capillary compartments in C57BL/6 mice grow between E15.5 and E17.5 (47). Given the argument just presented and previous observations that the number of fetoplacental arteries does not increase during this period, the observed 26% increase in the reflection coefficient is most likely caused by changes in the dimensions of fetoplacental arteries.

Wave reflection and fractal geometry.

The fetoplacental arterial tree has a stochastic but particularly regular structure that has been well described in terms of fractal geometry (59). In particular, the relationship between the size of parent and daughter branches is preserved across the approximately eight generations (48) of vessels that connect the UA to the capillary bed. Anticipating that these geometric relationships could explain the observed wave reflections, we used available data (47) to compute the bifurcation area ratio, the ratio of the combined cross-sectional area of daughter vessels to that of the parent vessel, for each study group. This area ratio determines how rapidly diameter drops off with successive generations of vessels and also determines the proportions of a wave that are reflected and transmitted at each bifurcation. Using the diameter scaling coefficient reported in Ref. 47 and assuming symmetric bifurcations, the average bifurcation area ratio in C57BL6/J mice increased from 1.26 at E15.5 to 1.35 at E17.5. In contrast, the CD1 area ratio was unchanged at 1.26 between E15.5 and E17.5.

The bifurcation area ratio has a number of interesting properties. An area ratio of 1.26, as observed in CD1 and C57BL/6 mice at E15.5, corresponds to a diameter scaling coefficient of 3 and was predicted by Murray in 1926 (42) to minimize energy dissipation due to viscous friction in a hierarchical branching network with fixed total blood volume. In contrast, under the assumption that only the radius is allowed to vary at a bifurcation, wave reflections are completely eliminated when the area ratio is 1.0 (i.e., total area is preserved at each bifurcation) (50). A theoretical curve (50) relating the area ratio of the reflection generated by a single bifurcation is shown in Fig. 5 with the area ratios reported at E15.5 and E17.5 marked for reference. Keeping in mind that the observed wave reflection represents the aggregate of reflections originating from all bifurcations, the measured reflection coefficients of 0.35 and 0.44 for C57BL/6 mice at E15.5 and E17.5 seem plausible based on the prediction of 0.115 and 0.149 as the proportion of the wave reflected from each bifurcation. It is interesting to note that this increase in area ratio from 1.26 to 1.35, while effective at reducing vascular resistance, is neither optimal under the conditions set out by Murray nor for minimizing energy losses associated with wave reflection.

Fig. 5.

Reflection coefficient for a single bifurcation (independent of wave velocity and the density of blood) as a function of the area ratio. At E17.5, the area ratios for CD1 and C57BL/6 mice were 1.26 (white circle) and 1.35 (gray circle), respectively.

Wave reflection and other physiological parameters.

The observation that the reflection coefficient was not correlated to heart rate or mean flow rate supports the notion that wave reflection is determined by the geometry of the fetoplacental arteries and not the physiological state of the fetus at the time of the measurement. This is in contrast to the pulsatility index, which showed a trend to being related to mean flow rate. Previous studies in humans have also noted that pulsatility is negatively correlated with fetal heart rate, estimating that heart rate variation accounts for 17% of pulsatility index variation (60).

UA pulsatility index.

The UA pulsatility index decreased across gestation in CD1 mice. Electrical models of the Doppler waveform pulsatility suggest that if the arterial pressure pulsatility (which drives flow through the placental vasculature) is constant, then the UA pulsatility index is dependent on the ratio of the placental to UA resistances (54). Assuming that the aortic pressure pulsatility and UA properties do not vary across gestation in CD1 mice, then the observed decline in the UA pulsatility index is likely due to decreases in the placental resistance across gestation. Indeed, it has been reported that the capillary volume increases substantially in CD1 mice from E15.5 to E17.5 time points (47). These vascular changes decrease placental resistance and likely account for the decline in the UA pulsatility index. Unlike CD1 mice, the increase in capillary volume is known to be modest in C57BL/6 mice (47). The blunted fetoplacental capillarization may explain why the pulsatility index does not decrease across gestation in C57BL/6 mice and is elevated in late gestation compared with CD1 mice.

Diagnostic comparison of the reflection coefficient and UA pulsatility index.

Since the morphologies of the fetoplacental vasculature of CD1 and C57BL/6 mice are known to differ at E17.5, we compared the diagnostic performance of the reflection coefficient and UA pulsatility index for discrimination between these two strains. ROC analysis (19) demonstrated that the reflection coefficient metric was better able to discriminate between strains compared with the UA pulsatility index. A previous study (1) has shown that the UA pulsatility index is related to the total umbilicoplacental vascular resistance under the conditions where the pressure pulsatility (dependent on cardiac function) and the properties of the UA are invariant. It is possible that interstrain variation in these parameters may have lowered the diagnostic performance of the UA pulsatility index. The waveforms obtained from the wave reflection methods are also dependent on factors such as pressure pulsatility. However, the wave reflection metrics, specifically reflection coefficient and change in dispersion, have been constructed to minimize the sensitivity to confounding physiological variables associated with cardiac function. This could prove to be an important advantage for detecting placental defects in high-risk pregnancies where congenital heart defects are present.

Detection of IUGR.

IUGR is identified as an underlying cause in ~40% of stillbirths of normally formed fetuses in the third trimester (10). A variety of defects in the fetoplacental arterial and capillary compartments have both been associated with IUGR. For instance, vessel wall thickening, degeneration of the small- to medium-sized arterial vessels (50−149 μm in diameter), and reduced stem, intermediate, and terminal villi volumes have been previously found in IUGR placentas (15, 28, 35). However, the sensitivity of UA Doppler (the current clinical gold standard) to detect this group of placental defects is low, especially after 34 wk of gestation, when pulsatility in the UAs is preserved. Indeed, in one study (49), the abnormal stereological parameters of intermediate and terminal villi in the IUGR group showed no correlation with various UA Doppler indexes (such as the pulsatility index, resistivity index, etc.). Furthermore, as a change in UA Doppler indices represents a late pathological process, it has been shown that resistance measures of intraplacental arteries perform better in detecting early placental disease compared with the pulsatility index of the UA (3).

Since our results suggest that the reflected waveform can detect subtle differences in fetoplacental arterial morphology and is more sensitive to detecting interstrain fetoplacental vascular differences compared with the UA pulsatility index, this technique may offer an advantage in detecting early and milder forms of the placental pathologies underlying IUGR. Moreover, since the measurements required for computing pulsatility index are a subset of those needed for wave reflection, the two methods may prove complementary. Previous work in fetal sheep has demonstrated that increasing placental vascular resistance by microsphere embolization of capillaries produced measurable changes in the pulsatility index (2), whereas the theoretical arguments presented above suggest that the effect of capillary morphology on wave reflection should be small except in extreme pathological cases.

Study limitations.

Despite the successful observation of UA wave reflections, a large proportion of data sets (~54%) were discarded that did not meet the pre- and postprocessing quality control criteria for the UA M-mode recordings. Applying such stringent criteria was necessary as even small amounts of maternal respiration and fetal/umbilical cord movements had a large effect on the lumen diameter signal. For instance, contamination of the M-mode signal with maternal gasping often resulted in nonphysiological area waveforms, and these signals had to be excluded in the postprocessing steps. Furthermore, as the measured UA area change was quite small (on the order of 2–9%), even minor sources of noise, such as out-of-plane UA motion during M-mode scanning or fetal motion, degraded the data quality. It is important to note that the measurement technique depends on detecting subtle shifts in the grayscale intensity of the vessel wall as many of the displacements are smaller than the point spread function of the M-mode ultrasound pulses.

A second limitation of this study is that the viscoelastic properties of the UA are not considered. Previous work in sheep has demonstrated that pulse waves are attenuated as they propagate along the cord (1). As the point of measurement was not controlled from fetus to fetus, this effect could represent an additional source of variability in the present study.

A final limitation is that the site of the earliest reflection source could not be determined from the reflection data. While in principle using the PWV and time delay should give the distance to the first reflection sites, uncertainty about the PWV in vessels beyond the umbilical cord prevents this interpretation. PWV is dependent on the ratio of the wall thickness to lumen diameter as well as elastin and collagen content (34). Assuming that PWV was constant throughout the placental tree (distance to reflection site = ½ × PWV × time delay) led to implausible estimates for the distance to the reflection site of 7–11 cm. It is important to note, however, that time delay was defined here based on the time of the peak amplitude. While this is a convenient metric to compute, it does not provide an estimate of the timing of the earliest reflection, which may precede the arrival of the wave peak. The observation of nonzero dispersion confirmed our expectation that multiple reflection sites contribute to the observed waveform. It is nonetheless surprising that a preponderance of reflections arrive at a later time than a simple estimate based on distance traveled would predict and suggests that proximal reflection sites such as the insertion point of the umbilical artery into the placenta make at most a small contribution to the measured reflection.

Conclusions.

We have developed a novel, noninvasive, and clinically motivated method to measure and characterize UA wave reflections in mice. Using high-frequency ultrasound measurements in the UA and semiautomated image analysis, we isolated the forward wave originating from the heart and the reflected wave originating from the placenta. Based on our finding that this method is able to sensitively discriminate between mouse strains with differing fetoplacental vascular morphology, we believe that wave reflection measurement could open a new avenue for detecting placental vascular pathology in human pregnancy.

GRANTS

Funding for this work was provided by Eunice Kennedy Shriver National Institute of Child Health and Human Development of Health Grant U01-HD-087177-01 and by Canadian Institutes of Health Research Grant MOP130403.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

A.R. and Y.-Q.Z. performed experiments; A.R., L.S.C., and Y.Y. analyzed data; A.R., L.S.C., C.M., and J.G.S. interpreted results of experiments; A.R. prepared figures; A.R. and J.G.S. drafted manuscript; A.R., Y.Y., L.S.C., C.M., and J.G.S. edited and revised manuscript; A.R., Y.-Q.Z., Y.Y., J.D., L.S.C., J.C.K., C.M., and J.G.S. approved final version of manuscript; J.D., L.S.C., J.C.K., C.M., and J.G.S. conceived and designed research.