Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2024 Jan 1.
Published in final edited form as: Ultrasound Med Biol. 2022 Oct 22;49(1):356–367. doi: 10.1016/j.ultrasmedbio.2022.09.008

Longitudinal In Utero Analysis of Engrailed-1 Knockout Mouse Embryonic Phenotypes Using High-Frequency Ultrasound

Orlando Aristizábal 1, Ziming Qiu 2, Estefania Gallego 1, Matias Aristizábal 1, Jonathan Mamou 3, Yao Wang 2, Jeffrey A Ketterling 3,†, Daniel H Turnbull 1,2,†
PMCID: PMC9712241  NIHMSID: NIHMS1837773  PMID: 36283941

Abstract

Large scale international efforts to generate and analyze loss-of-function mutations in each of the approximately 20,000 protein-encoding gene mutations are ongoing using the “knockout” mouse as a model organism. Because one third of gene knockouts are expected to result in embryonic lethality, it is important to develop noninvasive in utero imaging methods to detect and monitor mutant phenotypes in mouse embryos. We demonstrate the utility of three-dimensional (3D) high-frequency (40-MHz) ultrasound (HFU) for longitudinal in utero imaging of mouse embryos between embryonic days (E)11.5 to E14.5, which represent critical stages of brain and organ development. Engrailed-1 knockout (En1-ko) mouse embryos and their normal control littermates were imaged with HFU in 3D, enabling visualization of morphological phenotypes in the developing brains, limbs and heads of the En1-ko embryos. Recently developed Deep Learning approaches were used to automatically segment the embryonic brain ventricles and bodies from the 3D-HFU images, allowing quantitative volumetric analyses of the En1-ko brain phenotypes. Taken together, these results show great promise for the application of longitudinal 3D-HFU to analyze knockout mouse embryos, in utero.

Keywords: Longitudinal 3D ultrasound, In utero imaging, mouse embryo, mutant phenotype analysis

INTRODUCTION

In the post-genomic era, after completing the DNA sequencing in the Human Genome Project, a major challenge has been analyzing the functions of the ~20,000 protein-encoding genes. In this effort, the mouse has been a critical model organism (Collins et al. 2007), which resulted in the formation of the International Mouse Phenotyping Consortium (IMPC) whose stated goal over the past decade has been to undertake broad-based phenotyping of the 20,000 mouse genes (Brown & Moore 2012). Early in this effort, it was recognized that loss-of-function or “knockout” mutations of approximately one third of the mouse genes lead to embryonic or peri-natal lethality, which has motivated the development of three-dimensional (3D) imaging approaches to assess the morphological phenotypes of knockout mouse embryos (Dickinson et al. 2016). The resulting phenotyping pipelines have relied heavily on ex vivo 3D imaging of mouse embryos that are harvested and prepared (fixed and stained) for each imaging modality at defined embryonic stages. Commonly used imaging methods include optical projection tomography (OPT) (Sharpe et al. 2002; Wong et al. 2015), micro-computed tomography (micro-CT) (Wong et al. 2012; Wong et al. 2014), and high-resolution episcopic microscopy (HREM) (Weninger et al. 2014).

Missing in these efforts has been the incorporation of in utero imaging approaches – noninvasive, in vivo imaging of embryos within the maternal uterus – in the phenotyping pipelines. The current 3D imaging methods are all ex vivo and static, providing morphological data at a single time-point in each fixed embryo sample. We have demonstrated that high-frequency (40 MHz) ultrasound (HFU) can provide high resolution (<40-μm axial, <100-μm lateral) in utero images of live mouse embryos, without the need for fixation or staining, over a wide range of stages, enabling high-throughput longitudinal analyses of mouse brain development (Turnbull et al. 1995; Aristizábal et al. 2006; Mamou et al. 2009; Aristizábal et al. 2013). In utero HFU can also be used to analyze the cardiovascular system in mouse embryos (Phoon & Turnbull 2016; Ketterling et al. 2017), and likely other organ systems in the future.

In the current study, we aimed to test the feasibility of employing longitudinal 3D HFU for in utero qualitative and quantitative analyses of Engrailed-1 (En1) mutant mouse embryos (Wurst et al. 1994). To this end, we employed a custom 40-MHz annular-array HFU scanner to acquire longitudinal 3D HFU images from embryonic day (E)11.5-14.5 En1 homozygous mutants (En1−/−, denoted En1-ko), and their wildtype (En1+/+ or WT) and heterozygous mutant (En1+/−) control littermates. Our results demonstrate the ability to detect and monitor over time the En1-ko mid-hindbrain and limb phenotypes (Wurst et al. 1994; Loomis et al. 1998) in individual mouse embryos. Moreover, recently developed Deep Learning approaches were utilized (Qiu et al. 2021), enabling automated segmentation of the bodies (including heads) and brain ventricles (BVs) for quantitative volumetric analyses of the HFU data. Taken together, these results demonstrate high potential for the use of in utero HFU to analyze knockout mouse embryos, as well as the potential to incorporate longitudinal 3D HFU in future embryonic phenotyping pipelines.

MATERIALS AND METHODS

Animals

All mice used in this study were maintained under protocols approved by the Institutional Animal Care and Use Committee at New York University School of Medicine. Mice carrying a homozygous mutation in the En1 gene were generated by breeding together En1 heterozygous mice (provided by Dr. Alexandra Joyner at Memorial Sloan Kettering Cancer Center, New York NY), maintained on a mixed genetic background including the (outbred) Swiss-Webster strain (Taconic Biosciences, Germantown NY). Progeny of this cross were genotyped at embryonic stages, using PCR of embryo tissue DNA to identify homozygotes (En1−/−, denoted En1-ko), heterozygotes (En1+/−), and wild-type (En1+/+, WT) mice. The stage of each embryo was specified as the embryonic day (E), where E0.5 was defined as noon of the day on which a vaginal plug was detected after overnight breeding. This is a standard biological method of defining the developmental stage of mouse embryos.

HFU Acquisition System

The 3D HFU data were acquired with a custom 5-element HFU (40-MHz) annular-array (AA) system similar to that previously described for in utero imaging of mid-gestation mouse embryos (Aristizábal et al. 2013) (Figure 1). AA transducers generate an extended focal zone, compared to single-element focused transducers (Ketterling et al. 2005), and have the advantage over linear arrays of a radially symmetric ultrasound beam (Filioux et al. 2011). In this study, the AA transducer was fabricated with a 9-µm poly(vinylidenedifluoride-cotrifluoroethylene) membrane (PVDF-TrFE; Piezotech, Arkema-CRRA., Pierre-Benite, France) bonded to a copper clad flex circuit etched with an AA pattern (Figure 2a). Microvias routed the elements to connection pads on the reverse side of the flex circuit (Figure 2a), allowing the containment of all trace lines within the transducer assembly. Previous designs had trace lines on the same side of flex as the annuli which were wrapped around the side of transducer shell and exposed to the environment (Ketterling et al 2005). A spherical curvature was formed with a press fit (Ketterling et al. 2005) and made permanent by backfilling with epoxy (Figure 2b). The total active aperture of the array was 6 mm, and the geometric focus was 12 mm. The 5 array elements had equal area and were separated by 0.1 mm. After press-fit assembly, the transducer was mounted into a metal cap and series inductors were placed just behind the elements terminating in a circular connector (Micro 360, Omnetics, Minneapolis, MN) (Figure 2c). This module was sealed in a copper tube and mated to a matching circular connector with cabling (Figure 2d). The transducer was mounted to a high-speed motion stage (LAS35, SMAC, Carlsbad, CA) which permitted real-time imaging (Figure 2d). The system was controlled with custom LabVIEW software (National Instruments, Austin, TX). Imaging was performed in contact mode using a plastic assembly with a small water bath sealed at the bottom with an acoustic window. This configuration eliminated the need for an external water bath, as was used previously (Aristizábal et al. 2013).

Figure 1: Schematic of the annular array system for in utero HFU.

Figure 1:

Open in a new tab

The 5-element Annular Array (AA) transducer was coupled to the shaved abdominal skin of the pregnant mouse using an integrated water bath (W) with an acoustic window and a thin layer of gel (G) to visualize and acquire HFU image data from the mouse embryos and nearby anatomical structures, including the bladder (B), uterus (U) and vagina (V). The position of the AA transducer was controlled using XYZ stages while transmitting and receiving from each element independently with a 5-channel pulser/receiver (5 Ch P/R). A respiratory pillow (P) was used to detect the respiratory waveform, enabling real-time (RT) 2D imaging and respiratory-gated 3D imaging. Image data could be displayed in real time after GPU beamforming.

Figure 2: Fabrication of the 5-element annular array transducer.

Figure 2:

Open in a new tab

a) The arrays were fabricated by bonding a PVDF-TrFE film to a copper-clad polyimide film etched with an array pattern on one side and contact pads on the opposite. b) A press-fit technique was used to form a spherical curvature and an epoxy plug was cast to maintain the geometry. The array was mounted in a metal shell (A). (c) Series inductors (L) were placed near the array elements and then terminated at a circular connector. The array was placed in a copper tube (B) and the tube was filled with epoxy. d) The array was attached to high-speed motion stage that had an integrated water bath module (W) with an acoustic window. Imaging was performed in contact mode using a thin layer of gel to provide acoustic coupling.

Data were acquired in a high-speed, single-pass mode using a custom 5-channel pulser/receiver (Daxsonics, Halifax, Nova Scotia, Canada) and PCIe digitizer (ATS9440, Alazartech, Pointe-Claire, Quebec, Canada). At fixed lateral intervals, a spatial trigger initiated a rapid sequence of 5 staggered triggers, one to each array channel, to permit collection of all transmit-to-receive element pair echo data. When acquiring 3D data, an automated cross-axis motion stage was used to move fixed intervals between adjacent planes. Data were streamed in real-time to a graphics-processing unit (GPU) (GTX-970, Nvidia, Santa Clara, CA, USA) for beamforming and the result displayed and saved to file. The improvements to the annular-array system permitted real-time frame rates higher than 10 frames per second and reduced the time to acquire a typical 3D volumetric data set from 2 minutes to 20 seconds.

HFU Acquisition Protocol

For in utero volumetric HFU acquisition, a pregnant mouse was anesthetized using a digital anesthesia machine (Somnosuite, Kent Scientific, Torrington CT) and maintained by delivering 1.5–2.0% isoflurane in air at a flow rate of 50 mL/min. It should be noted that this flow rate is approximately 10x lower than conventional anesthesia vaporizers, which significantly reduced the exposure levels to both the animals and the personnel. Once anesthetized, the mouse was secured to a custom platform and the respiration was monitored using a respiratory pillow interfaced to a Biopac MP100A (Biopac System Inc., Goleta, CA, USA). The respiration rate was between 60–75 breaths/min for all the mice included in this study. The respiration pillow signal was also converted to a digital signal for respiratory gating. The internal temperature of the pregnant mouse was monitored using a rectal thermometer, and a heat lamp was used to keep the maternal body temperature between 35–37°C. The abdomen of the mouse was wet shaved and ultrasound gel was used to couple the transducer to the abdomen. Using the real time B-mode imaging mode, the bladder and adjacent embryos were located and imaged. For a particular embryo, the volume of interest was set such that it covered the whole conceptus (embryo and surrounding uterus). Each volume was saved as a series of 2D slices and immediately loaded into Amira (V2019.1, Thermo Fisher Scientific, Waltham MA USA). Using the multi planar viewer, the embryo was sagittally oriented and mid sagittal sections were inspected to assess whether the embryo exhibited the expected En1-ko brain phenotypes. This procedure took about 2 minutes per embryo, and was repeated for one or two embryos on either side of the bladder for an average acquisition time of 15 minutes per pregnant mouse. The sagittally reconstructed volumes also revealed if the data was free of motion artifacts and, if present, a new dataset was acquired. All motion artifacts were recognized when the planes perpendicular to the reconstructed plane where inspected. Also, embryonic stability was assessed while imaging an individual embryo in the real time mode before acquiring 3D data. Within the 2-minute acquisition of the 3D volumetric data from each embryo, it was very rare that motion artifacts required re-acquisition of a dataset.

After the final day of in utero imaging (between 0–24h after), volumetric data sets were acquired from the whole litter using a semi-invasive method (Aristizábal et al. 2006). Briefly, an intact conceptus was exposed through midline laparotomy and into a petri dish full of saline solution. The transducer was then lowered into the solution and the 2D real-time imaging mode was used to center the conceptus in the 3D field of view. The 3D dataset was acquired and the same procedure repeated until all the embryos on the left and right sides of the uterine horn were imaged. After imaging, tissue from each embryo was collected and analyzed to correlate the genotype with the phenotype in each embryo.

Image Analyses

In addition to the changes in image acquisition hardware and software, an image preprocessing step was included in the pipeline to improve image quality for 3D analyses. Specifically, the raw image data were passed through the following filters in Amira: 1) a 3D Gaussian filter (kernel size:3x3x3, sigma:0.4x0.4x0.4), 2) a 3D contrast limited adaptive histogram equalization (CLAHE) filter (contrast limit: 2), and 3) a 3D brightness and contrast filter (brightness: 0, contrast:1.2). These filters were implemented and adjusted empirically to enhance the contrast between the fluid-filled BVs and the surrounding embryonic and extra-embryonic tissues (Supplemental Figure 1).

Automatic segmentation of the embryo body and BV was achieved using a novel Deep Learning-based pipeline which consisted of two stages, described in detail previously (Qui et al. 2021). Briefly, in the first stage a fully convolutional network (FCN) was trained to output low-resolution segmentations of the body and the BV. In the second stage, these coarse segmentation maps were used to crop a region of interest around the target object and then input as context information to the refinement FCNs which further refined the body and BV segmentations at the high resolution. This pipeline was trained with manually segmented body and BV labels obtained from 3D annular array HFU acquired from 139 E10.5-14.5 mouse embryos imaged. Accurate segmentation was achieved on a separate testing set containing 46 HFU images with manual labels. It should be noted that although the pipeline was originally trained on mouse embryos staged between E10.5-14.5 (Qui et al. 2021), in the current study the same pipeline was applied to embryos staged between E11.5-14.5. This validated model was applied to the current in utero dataset of 32 embryos, as well as the full set of semi-invasive data from the 101 embryos in the 9 pregnant mice of this study. The automatic segmentation results were inspected manually by an expert (OA) and found to be satisfactory.

Quantitative Analyses

For each embryo, the body and BV labels were imported to Amira to calculate the volumes and surface areas. Because the labels were in 8-bit format, they were first converted into native Amira labels by connecting to the CastField module with the Output Datatype set to LabelField. The labels were then connected to the SurfaceGen module with smoothing set to none and then the SurfaceArea module was attached to the generated surface yielding a volume and a surface. In order to efficiently process all of the body and BV segmentations, a script was written in Python and run within Amira to automatically calculate the volumes and surface areas and write them into a comma separated file that was used for statistical analyses.

Statistical Analyses

All data were expressed as mean ± standard deviation. Comparisons between two groups were analyzed with the two-tailed Student’s t test, for parametric distributions with unequal variances. Statistical significance level was set at p<0.05, and was also corrected from p<0.05 for multiple comparisons using the Bonferroni correction.

RESULTS

HFU was used to Acquire 3D In Utero Data

Noninvasive HFU was successfully used to acquire 3D in utero images of En1-ko (En1−/−) embryos and their WT and En1+/− littermates in 9 pregnant mice spanning E11.5 to E14.5 (Table 1). Because En1+/− mice have no known phenotypes (Wurst et al. 1994), we grouped WT and En1+/− embryos together as controls (CTLs; N=23) for the En1-ko embryos (N=9). 3D HFU images were acquired using longitudinal (i.e., imaged at two or more timepoints) imaging (Table 1). In all, 26 in utero datasets were acquired in the E11.5-14.5 En1-ko embryos, and 67 datasets from the CTL littermate embryos in the 9 pregnant mice. Table 1 shows details of the embryos and the timepoints imaged for all of the in utero longitudinal datasets.

Table 1: Summary of HFU data acquired from En1-ko and CTL mouse embryos.

The number of En1-ko and CTL embryos imaged longitudinally (non-invasively) in utero between E11.5-14.5 is summarized in (A). Details of the timelines for each embryo imaged longitudinally are shown in (B). Arrows indicate that each embryo was imaged at each stage between the beginning to the end of each arrow. For example, in the section on En1-ko embryos, the first arrow shows that 3 En1-ko embryos were imaged at each stage between E11.5 to E14.5 (i.e., at E11.5, E12.5, E13.5 and E14.5). Finally, the number of embryos imaged semi-invasively is summarized in (C).

graphic file with name nihms-1837773-t0008.jpg
Open in a new tab

In addition to the noninvasive in utero HFU imaging, each of the 9 pregnant mice used for in utero studies were imaged in a semi-invasive mode. In this case, every embryo (N=101) in each pregnant mouse (N=9) was externalized (within the uterus) for a final acquisition of high-quality 3D HFU data immediately before euthanasia and harvesting of the embryos for genotype analysis (Table 2). These semi-invasive externalized HFU imaging sessions provided another 26 datasets from E12.5-14.5 En1-ko embryos, and 75 datasets from E12.5-14.5 CTL embryos (Tables 1 & 2).

Table 2: Summary of the numbers of embryos and their genotypes in the 9 pregnant mice used in these studies.

The start- and end-points for longitudinal imaging are shown for each pregnant mouse. Note that Endpoint refers to the day that all the embryos were imaged semi-invasively and harvested for genotyping. Only a subset of the En1-ko and CTL from each pregnant mouse were imaged longitudinally, as shown in Table 1.

Mouse Stages imaged: Start-Endpoint En1-ko CTL
1 E11.5-12.5 3 10
2 E11.5-13.5 3 6
3 E11.5-13.5 5 9
4 E11.5-14.5 4 8
5 E11.5-14.5 2 3
6 E11.5-14.5 1 12
7 E12.5-14.5 2 9
8 E12.5-14.5 2 10
9 E13.5-14.5 4 8
Open in a new tab

HFU Reveals Brain Phenotypes in En1-ko Embryos

As previously reported (Wurst et al. 1994), the most obvious brain phenotype in En1-ko embryos was the partial deletion of mid-hindbrain, including most or all of the early forming cerebellum (Figure 3). Semi-invasive HFU images of whole litters of embryos from En1+/− x En1+ crosses were followed by genotype analysis of dissected embryos to determine the accuracy of HFU in detecting the brain phenotype in En1-ko embryos (N=26), compared to the CTL embryos (N=75) (Figure 4, Table 2). Interestingly, a small residual cerebellum, while missing in most En1-ko embryos, was present and detected by HFU in a few knockout embryos (2/8 at E13.5, 4/15 at E14.5; Figure 4). This variability in the En1-ko brains is likely due to the mixed background strain used in the current study, which has previously been shown to affect the En1-ko brain phenotype (Kuemerle et al. 2007). Careful examination of the embryonic cerebellum in 3D HFU images was sufficient to discriminate between En1-ko and CTL embryos in all cases. In addition to the brain phenotypes, in utero HFU images of En1-ko embryos showed the associated abnormal cranial/head shape described in earlier studies (Figure 4; Wurst et al. 1994; Deckelbaum et al. 2012).

Figure 3: HFU detection of brain and head phenotypes in En1-ko embryos.

Figure 3:

Open in a new tab

Mid-sagittal (a, b) and transverse (c, d; plane position shown by yellow dashed lines on panels a, b) sections of E12.5 control (CTL) and En1-ko embryos show the developing cerebellum in the CTL embryos (white arrows on panels a, c) above the choroid plexus (cp). The cerebellum is missing in the En1-ko embryo (red arrows on panels b, d). 3D renderings of the heads and segmented BVs (purple) (e, f) show the prominent bulge at the back of the head in the CTL embryo (white arrow on panel e), which is significantly reduced in the En1-ko embryo (red arrow on panel f). Visualizing the segmented ventricles (g, h) alone, the midbrain aqueduct (which connects the 3rd and 4th ventricles) in the CTL embryo (white arrow on panel g) was clearly missing in the En1-ko embryo (red arrow on panel h). Scale bar = 1 mm.

Figure 4: Semi-invasive HFU images of an En1+/− x En1+/− litter of embryos.

Figure 4:

Open in a new tab

Semi-invasive HFU of the exteriorized uterus was used to acquire images of each litter of embryos, which could be unambiguously correlated to the genotypes after harvesting the embryos. In this example, E14.5 mid-sagittal sections are shown for the 4 En1-ko embryos (right column) and 8 CTL littermate embryos (left two columns). In this litter, 3 of the 4 En1-ko embryos had the obvious missing cerebellum above the choroid plexus (cp). One En1-ko embryo (bottom left panel) had some cerebellar tissue remaining, which was detected by HFU (red arrow) but clearly distinguishable from the normal cerebellum in a CTL embryo (white arrow). Scale bar = 1 mm.

HFU Enables Longitudinal Imaging of En1 Mutants and Control Littermates

Importantly, HFU enabled in vivo longitudinal imaging of individual En1-ko embryos (N=9), and their CTL littermates (N=23) over the time interval between E11.5 and E14.5 (Table 1; Figure 5). Longitudinal imaging of individual embryos was achieved using external anatomical landmarks, most notably the maternal bladder. The mid-hindbrain deletion, described above, was detected at each developmental stage in all 9 of the En1-ko embryos imaged longitudinally, including one case of the residual cerebellum detected at E13.5, after which the embryos were imaged semi-invasively and harvested to correlate genotype with phenotype (Table 2). Our results indicated a 100% reproducibility in detecting the En1-ko mid-hindbrain phenotype (including the one E13.5 embryo with residual cerebellum) from the longitudinal in utero HFU images (compared to semi-invasive images), with 0% false-positive detection in the CTL embryos.

Figure 5: Longitudinal in utero HFU of control and En1-ko embryos.

Figure 5:

Open in a new tab

Example longitudinal HFU images acquired from an En1-ko embryo (bottom row) and its CTL littermate (top row) between E11.5 and E14.5. Scale bar = 1 mm.

HFU Reveals Limb Phenotypes in En1-ko Embryos

Loss of En1 gene function also results in limb phenotypes, including ectopic digits and fusion of the normal digits (Loomis et al. 1998). In utero longitudinal HFU images revealed abnormal digit formation in all the forelimbs of the E14.5 (N=6) and E13.5 (N=7) En1-ko embryos imaged, but did not have sufficient resolution to detect abnormalities at E12.5 (N=8) or E11.5 (N=5) (Figure 6). We further confirmed that the forelimb phenotypes were present and could be detected with HFU in all 15/15 of the E14.5 En1-ko embryos imaged semi-invasively and in 7/8 of the E13.5 En1-ko embryos imaged semi-invasively, while 0/3 of the E12.5 En1-ko embryos imaged semi-invasively showed abnormal forelimbs. As with the brain phenotype, there was 100% reproducibility of the forelimb phenotype detection on the longitudinal HFU images of E13.5-14.5 En1-ko embryos compared to the semi-invasive images, with 0% false-positive detection in the E13.5-14.5 CTL embryos. Taken together with the brain and head phenotypes described above, these results show that in utero HFU provides a highly reliable and reproducible in vivo approach for identifying a variety of stage-dependent morphological phenotypes in En1-ko mouse embryos.

Figure 6: HFU detection of limb phenotypes in En1-ko embryos.

Figure 6:

Open in a new tab

3D renderings of E12.5 (a, b) and E14.5 (c, d) CTL and En1-ko mouse embryos show expanded views of the 3D rendered limbs (top insert) and a 2D HFU section (bottom insert) in each panel. At E12.5, the developing digits were clearly visualized in both the CTL (a) and En1-ko (b) embryos, with no obvious phenotype in the mutants. By E14.5, compared to the CTL embryos (c), the fusion of digits (d) was detected in all of the En1-ko embryos (red arrows). Scale bars: Bars in main 3D images = 1 mm; Bars in insert images = 0.5 mm.

Automated Segmentation of 3D HFU Images Enables Quantitative Analyses

Recently developed image analysis approaches, based on a Deep Learning framework, have enabled simultaneous, automated segmentation of the BVs and body from 3D HFU images (Qiu et al. 2021). These approaches were applied to analyze stage-dependent volumetric changes in the BVs and bodies of the En1-ko embryos and their control littermates (Figure 7). Despite obvious differences in the morphologies of the BVs of the En1-ko mice compared to controls, there were no significant differences (with or without Bonferroni correction) in BV volumes at any embryonic stage (E11.5-14.5), nor in volumes or surface areas of the BV and bodies, comparing En1-ko and control embryos (Figure 7). There were also no differences at any stage between En1-ko and control BV volumes or surface areas after normalization to their respective body volumes or surface areas (Figure 7).

Figure 7: Quantitative analyses of BV and body volumes and surface areas.

Figure 7:

Open in a new tab

Quantitative analyses of the volumes (a-c) and surface areas (d-f) of the BVs and bodies showed no significant differences, with or without Bonferroni correction, between the E11.5 to E14.5 En1-ko embryos and their CTL littermates. There were also no significant differences between En1-ko and CTL embryos in the BV/body volume and surface area ratios at any stage between E11.5 to E14.5.

DISCUSSION

In this study, we have demonstrated the utility of 3D HFU imaging for noninvasive, in utero detection and longitudinal analysis of mutant phenotypes in En1-ko mouse embryos. We focused primarily on the En1-ko brain phenotypes, providing in utero visualization and both qualitative and quantitative analyses between the embryonic stages E11.5 to E14.5. The embryonic En1-ko phenotype is subtle, with the missing tissue representing a very small fraction of the total brain tissue. 3D HFU images enabled detection and characterization of these brain defects, visualized in both 2D sections and in 3D renderings. In addition to the brain defects, longitudinal HFU images also enabled visualization and monitoring of En1-ko phenotypes in the limb buds and head shape. All of the longitudinal HFU phenotype results were validated with semi-invasive HFU and genotype analyses at the end of each longitudinal imaging study. Taken together, these results show great promise for applying 3D HFU for in utero longitudinal analyses of a variety of mutant embryo phenotypes over a wide range of embryonic stages relevant to brain, limb and other organ development.

In this study, we utilized a custom 40-MHz AA system for 3D HFU imaging. AA transducers provide superior focusing properties compared to linear arrays (Filioux et al. 2011), with the disadvantage of requiring mechanical systems for scanning. Given the widespread availability of commercial linear array HFU scanners, it will be interesting in future to characterize the differences of annular vs. linear arrays for in utero phenotyping in mouse embryos. Beyond some reduction in time to acquire volumetric data, one advantage of linear arrays is their potential to extend phenotype analysis to include cardiovascular features through traditional Doppler or ultra-fast acquisition using plane wave imaging (Huang et al. 2017; Ketterling et al. 2017).

Compared to our previous HFU measurements (Aristizábal et al. 2013), the BV volumes and surface areas of CTL embryos were approximately 20–25% larger in the current study. The relative trend in BV volume and surface area between E11.5-14.5 was similar in the two studies. Another study using 40-MHz HFU (Chau et al. 2015) reported very similar BV volumes to the current study at E11.5-E12.5, with a 25% reduction at E13.5. It is likely that the volumetric differences found in these studies are due to the different background mouse strains used in each report. Finally, the current volumetric HFU results are also in good agreement with our previous measurements made using in utero MRI, including the lack of any significant reduction of BV volume at E13.5 (Parasoglou et al. 2013). Notably, our results do not show any significant differences in BV volume or surface area between En1-ko and CTL embryos between E11.5-14.5.

In utero longitudinal imaging poses significant challenges for mutant mouse phenotype analyses. Specifically, it is often difficult to identify individual embryos unambiguously over several days of longitudinal imaging because the embryos grow significantly and their positions relative to each other can shift over time. For longitudinal HFU, we employed a previously described approach in which the maternal bladder is used as the primary anatomical landmark for identifying individual mouse embryos in utero (Ji & Phoon, 2005). By imaging along the uterine horn, we could confidently identify 1–2 embryos on each side of the bladder (2–4 embryos per pregnant mouse) in each imaging session. With experience, it is likely that skilled sonographers could confidently identify larger numbers of individual mouse embryos in each litter, which can include 12–14 embryos in total (Table 2). In the current study, we identified 9 individual En1-ko embryos and 23 individual control embryos during longitudinal HFU imaging sessions in 9 pregnant mice, for an average of 3.5 embryos per mouse (Table 1).

Repeated exposure to anesthesia has the potential to affect fetal development in mice. In this study, the isoflurane levels employed were below those expected to cause obvious morphological changes in developing mouse embryos (Thaete et al. 2013). Nevertheless, we sought to minimize the anesthesia exposure to the pregnant mice and embryos, utilizing a protocol that required an average of 15 minutes of anesthesia per imaging session. It should also be noted that the anesthesia conditions for both the En1-ko embryos and their CTL littermates were the same during each image acquisition.

Despite the success of ex vivo 3D imaging methods such as micro-CT and OPT for mouse embryo phenotyping (Dickinson et al. 2016), there is a clear need for non-lethal in vivo methods for longitudinal, in utero analyses. Previous reports have shown some promise for in utero imaging of mouse embryos using magnetic resonance imaging (MRI) (Deans et al. 2008; Nieman et al. 2009; Berrios-Otero et al. 2012; Parasoglou et al. 2013; Wu et al. 2015; Zhang et al. 2018) and optical coherence tomography (OCT) (Larina et al. 2012). In comparison to HFU, OCT has finer resolution (<10-μm) but more limited penetration in tissues (~1-mm vs. 10-mm), and in utero imaging at mid-gestation has been confined to the surgically exposed uterus, similar to our semi-invasive HFU method. This has limited the applications of OCT for in utero longitudinal imaging and has led to most studies being performed on dissected embryos in culture (Lopez & Larina 2021). MRI has no penetration limitations, but 1–2 hours per embryo are typically required to obtain sufficient signal-to-noise and resolution for phenotype analysis, which drastically limits throughput compared to HFU. One advantage of MRI over HFU is its ability to acquire low-resolution, whole-abdomen 3D images to assess the positions of the entire litter of mouse embryos prior to focusing on single embryos to acquire higher resolution data in each session, including diffusion MRI studies of the developing brain (Wu et al. 2015; Zhang et al. 2018). In the future, a similar approach could be used with ultrasound by first using a low-frequency, wide-field of view ultrasound probe prior to the HFU acquisition.

Recently, we demonstrated a Deep Learning-based segmentation approach, applied to in utero HFU images of mouse embryos, which enabled automated segmentation of the embryonic bodies and BVs (Qiu et al. 2021). In the current study, this automated segmentation method was used to analyze volumetric changes in the body and BVs of En1-ko embryos and their control littermates. It was interesting that despite obvious morphological differences between the En1-ko and control mouse embryos, there were no differences in the quantitative measurements of volume and surface area at any stage between E11.5 to E14.5. In the previous report, we also developed a Deep Learning-based classifier to identify En1-ko mutant embryos based on the 3D HFU data (Qiu et al. 2021). In that study, the classifier was found to make its decisions mainly based on the data in the mid-hindbrain region, even though the classifier was given the images of the entire embryonic body. This finding strongly suggests that future shape-based analyses should be useful for characterizing the En1-ko mice and other knockout embryos with brain defects. Independent of such future analyses, the Deep Learning methods have already been shown to be highly effective for automated detection of embryonic brain phenotypes, including the En1-ko mid-hindbrain deletion. At this point, no quantitative shape or volume analyses have been performed on the En1-ko limb phenotypes.

These results provide strong evidence of the utility of longitudinal in utero HFU for high-throughput detection and analyses of mutant phenotypes in mouse embryos. Although we used a custom annular-array imaging system, the methods we developed should work well with 3D data sets obtained from traditional linear array HFU systems. Because of the radial symmetry of the annular array acoustic field, our approach can achieve superior out-of-plane (i.e., elevation) resolution compared to a linear array (Filoux et al. 2011; Filoux et al. 2012). Thus, our annular array approach represents a “gold standard” to test future linear array results against.

Supplementary Material

1. Supplemental Figure 1: Image processing pipeline for visualization of HFU images.

Example sagittal and coronal planes of an E12.5 embryo. The raw HFU data were sequentially processed with a Gaussian Filter, followed by a Contrast Limited Adaptive Histogram Equalization (CLAHE) Filter. Finally, brightness and contrast (B&C) were adjusted for optimal visualization of embryonic anatomical structures.

ACKNOWLEDGEMENTS

This research was funded by NIH grant R01 EB022950. We thank Drs. Alexandra Joyner (Memorial Sloan Kettering Cancer Center) and Cynthia Loomis (NYU Grossman School of Medicine) for discussions and advice on the En1-ko embryo phenotypes, and Dr. Alexandra Joyner for providing the En1+/− mice used for breeding. We also thank Alfred Yu and Billy Yiu (U. Waterloo) for providing the core GPU code used for the real-time beamforming.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare.

DATA AVAILABILITY STATEMENT

All data included in this paper will be made available upon request.

REFERENCES

  1. Aristizábal O, Ketterling JA, Turnbull DH. 40-MHz annular array imaging of mouse embryos. Ultrasound Med Biol 2006; 32: 1631–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Aristizábal O, Mamou J, Ketterling JA, Turnbull DH. High-throughput, high-frequency 3D ultrasound for in utero analysis of embryonic mouse brain development. Ultrasound Med Biol 2013; 39: 2321–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Berrios-Otero CA, Nieman BJ, Parasoglou P, Turnbull DH. In utero phenotyping of mouse embryonic vasculature with MRI. Magn Reson Med 2012; 67: 251–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Brown SD, Moore MW. The International Mouse Phenotyping Consortium: past and future perspectives on mouse phenotyping. Mamm Genome 2012; 23: 632–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Chau KF, Springel MW, Broadbelt KG, Park H-Y, Topal S, Lun MP, Mullan H, Maynard T, Steen H, LaMantia AS, Lehtinen MK. Progressive differentiation and instructive capacities of amniotic fluid and cerebrospinal fluid proteomes following neural tube closure. Dev Cell 2015; 35: 789–802. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Collins FS, Rossant J, Wurst W, International Mouse Knockout Consortium. A mouse for all reasons. Cell 2007; 128: 9–13. [DOI] [PubMed] [Google Scholar]
  7. Deans AE, Wadghiri YZ, Berrios-Otero CA, Turnbull DH. Mn enhancement and respiratory gating for in utero MRI of the embryonic mouse central nervous system. Magn Reson Med 2008; 59: 1320–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Deckelbaum RA, Holmes G, Zhao Z, Tong C, Basilico C, Loomis CA. Regulation of cranial morphogenesis and cell fate at the neural crest-mesoderm boundary by engrailed 1. Development 2012; 139: 1346–58. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Dickinson ME, Flenniken AM, Ji X, et al. High-throughput discovery of novel developmental phenotypes. Nature 2016; 537: 508–514. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Filoux E, Mamou J, Aristizábal O, Ketterling JA. Characterization of the spatial resolution of different high-frequency imaging systems using a novel anechoic sphere phantom. IEEE Trans Ultrason Ferroelect Freq Contr 2011; 58: 994–1005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Filoux E, Mamou J, Moran CM, Pye SD, Ketterling JA. Characterization of the effective performance of a high-frequency annular-array based imaging system using anechoic-pipe phantoms. IEEE Trans Ultrason Ferroelect Freq Contr 2012; 59: 2825–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Huang CC, Chen PY, Peng PH, Lee PY. 40 MHz high-frequency ultrafast ultrasound imaging. Med Phys 2017; 44: 2185–2195. [DOI] [PubMed] [Google Scholar]
  13. Ji RP, Phoon CK. Noninvasive localization of nuclear factor of activated T cells c1-/- mouse embryos by ultrasound biomicroscopy-Doppler allows genotype-phenotype correlation. J Am Soc Echocardiogr 2005;18: 1415–21. [DOI] [PubMed] [Google Scholar]
  14. Ketterling JA, Aristizábal O, Turnbull DH, Lizzi FL. Design and fabrication of a 40-MHz annular array transducer. IEEE Trans Ultrason Ferroelect Freq Contr 2005; 52: 672–681. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Ketterling JA, Aristizábal O, Yiu BYS, Turnbull DH, Phoon CKL, Yu ACH, Silverman RH. High-speed, high-frequency ultrasound, in utero vector-flow imaging of mouse embryos. Sci Rep 2017; 7: 16558. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kuemerle B, Gulden F, Cherosky N, Williams E, Herrup K. The mouse Engrailed genes: a window into autism. Behav Brain Res 2007; 176: 121–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Larina IV, Syed SH, Sudheendran N, Overbeek PA, Dickinson ME, Larin KV. Optical coherence tomography for live phenotypic analysis of embryonic ocular structures in mouse models. J Biomed Opt 2012;17: 081410. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Loomis CA, Kimmel RA, Tong CX, Michaud J, Joyner AL. Analysis of the genetic pathway leading to formation of ectopic apical ectodermal ridges in mouse Engrailed-1 mutant limbs. Development 1998;125: 1137–48. [DOI] [PubMed] [Google Scholar]
  19. Lopez AL 3rd, Larina IV. Dynamic Imaging of Mouse Embryos and Cardiac Development in Static Culture. Methods Mol Biol 2021; 2206:129–141. [DOI] [PubMed] [Google Scholar]
  20. Mamou J, Aristizábal O, Silverman RH, Ketterling JA, Turnbull DH. High-frequency chirp ultrasound imaging with an annular array for ophthalmologic and small-animal imaging. Ultrasound Med Biol 2009; 35: 1198–208. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Nieman BJ, Szulc KU, Turnbull DH. Three-dimensional in vivo MRI with self-gating and image coregistration in the mouse. Magn Reson Med 2009; 61: 1148–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Parasoglou P, Berrios-Otero CA, Nieman BJ, Turnbull DH. High-resolution MRI of early-stage mouse embryos. NMR Biomed 2013; 26: 224–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Phoon CK, Turnbull DH. Cardiovascular imaging in mice. Curr Protoc Mouse Biol 2016; 6: 15–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Qiu Z, Xu T, Langerman J, Das W, Wang C, Nair N, Aristizábal O, Mamou J, Turnbull DH, Ketterling JA, Wang Y. A deep learning approach for segmentation and classification of 3D high frequency ultrasound images of mouse embryos. IEEE Trans Ultrason Ferroelectr Freq Control 2021; 68: 2460–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Sharpe J, Ahlgren U, Perry P, Hill B, Ross A, Hecksher-Sørensen J, Baldock R, Davidson D. Optical projection tomography as a tool for 3D microscopy and gene expression studies. Science 2002; 296: 541–5. [DOI] [PubMed] [Google Scholar]
  26. Thaete LG, Levin SI, Dudley AT. Impact of anaesthetics and analgesics on fetal growth in the mouse. Lab Anim 2013; 47: 175–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Turnbull DH, Bloomfield TS, Baldwin HS, Foster FS, Joyner AL. Ultrasound backscatter microscope analysis of early mouse embryonic brain development. Proc Natl Acad Sci USA 1995; 92: 2239–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Weninger WJ, Geyer SH, Martineau A, Galli A, Adams DJ, Wilson R, Mohun TJ. Phenotyping structural abnormalities in mouse embryos using high-resolution episcopic microscopy. Dis Model Mech 2014; 7: 1143–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Wong MD, Dorr AE, Walls JR, Lerch JP, Henkelman RM. A novel 3D mouse embryo atlas based on micro-CT. Development 2012; 139: 3248–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Wong MD, Maezawa Y, Lerch JP, Henkelman RM. Automated pipeline for anatomical phenotyping of mouse embryos using micro-CT. Development 2014; 141: 2533–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Wong MD, van Eede MC, Spring S, Jevtic S, Boughner JC, Lerch JP, Henkelman RM. 4D atlas of the mouse embryo for precise morphological staging. Development 2015; 142: 3583–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Wu D, Lei J, Rosenzweig JM, Burd I, Zhang J. In utero localized diffusion MRI of the embryonic mouse brain microstructure and injury. J Magn Reson Imaging 2015; 42: 717–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Wurst W, Auerbach AB, Joyner AL. Multiple developmental defects in Engrailed-1 mutant mice: an early mid-hindbrain deletion and patterning defects in forelimbs and sternum. Development 1994; 120: 2065–75. [DOI] [PubMed] [Google Scholar]
  34. Zhang J, Wu D, Turnbull DH. In utero MRI of mouse embryos. Methods Mol Biol 2018; 1718: 285–296. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1. Supplemental Figure 1: Image processing pipeline for visualization of HFU images.

Example sagittal and coronal planes of an E12.5 embryo. The raw HFU data were sequentially processed with a Gaussian Filter, followed by a Contrast Limited Adaptive Histogram Equalization (CLAHE) Filter. Finally, brightness and contrast (B&C) were adjusted for optimal visualization of embryonic anatomical structures.

NIHMS1837773-supplement-1.jpg (1MB, jpg)

Data Availability Statement

All data included in this paper will be made available upon request.

RESOURCES