2.1.

Sample Preparation

Fresh embryos ($n=24$) of CD1xFVB mice were obtained from the Transgenic, Knockout, and Tumor Model Center (TKTC) at Stanford. The mice were subjected to hormone treatment with pregnant mare serum gonadotropin and human chorionic gonadotropin 72 and 24 h before mating, respectively. One half day after mating, the embryos were extracted from the mice and transported to our lab imaging facility for vitrification and imaging. All vitrification was performed by a trained clinical embryologist to emulate the protocol typically followed during assisted reproduction. Vitrification and thawing were performed with the Sage vitrification and warming kits (Cooper Surgical), according to the included protocols. Embryos were transported and imaged in Sage modified human tubal fluid with 10% Sage protein supplement. During imaging, a heated stage kept the media temperature around 37°C.

2.2.

Workflow and Experiment Design

On two separate occasions, fresh embryos were received from the TKTC $\sim 1\mathrm{h}$ after collection. Embryos on both days were randomly assigned to one of two groups. The embryos in the first group ($n=14$) were imaged using FF-OCT, vitrified and thawed, and subsequently imaged with FF-OCT once more. The embryos in the second group ($n=10$) were stained, imaged with laser scanning microscopy (LSM), imaged with FF-OCT, vitrified and thawed, imaged with FF-OCT, and then imaged with LSM. Care was taken to minimize rotation of the embryos between FF-OCT and LSM imaging sessions to ensure high fidelity and correlation of the images.

Combining the two groups, a total of 24 embryos were imaged and used in this study. Due to various challenges with embryos lysing or data corruption, the final dataset includes pre- and postvitrification FF-OCT volumes from 8 embryos in group 1, pre- and postvitrification FF-OCT volumes from 9 embryos in group 2, and pre- and postvitrification LSM datasets from 10 embryos in group 2. Table 1 provides a detailed summary of how many embryos / datasets were acquired and maintained during the process.

Table 1

Summary of all data collected. Each table cell contains the number of embryos imaged at that step followed by the number of usable images obtained from that step. Row 3 (vitrified and thawed) contains the number of embryos vitrified followed by the number of embryos that survived the thaw and could be imaged further.

2.3.

Fluorescence Imaging

The embryos used for fluorescence imaging were incubated in PBS containing 100 nM MitoTracker Green (M7514, LifeTechnologies) according to directions from the manufacturer. Embryos were imaged using an inverted LSM780 laser scanning confocal microscope (Zeiss). The laser wavelength of 488 nm was chosen to coincide with the excitation wavelength of the selected dye (490 nm). The laser was operated at 3% of the maximum intensity to avoid damaging the embryo and photobleaching. Using a $25\times$ LCI Plan NEO water-immersion objective (0.8 NA, Zeiss), 3-D image stacks were obtained.

2.4.

Full-Field Optical Coherence Tomography Imaging

All embryos were imaged using the FF-OCT system shown in Fig. 1. The system is based on a Linnik interferometer and incorporates a high-power white LED (Indus Star A007-GW765-R5, LEDdynamics Inc.) focused at the back focal plane of air-immersion microscope objectives ($20\times$ Plan APO LWD, 0.42 NA, Mitutoyo). The reference arm contains a piezoelectric actuator (PZT) (Thorlabs, AE0203D08F) onto which we mounted a fragment of an absorptive neutral density filter having $\sim 4\%$ reflectivity. An empty glass bottom Petri dish was also inserted into the reference arm for dispersion compensation. Images were acquired using a PCO Edge sCMOS camera over a small region of interest (860 by 800 pixels), using an integration time of 10 ms per frame. The incident light from the LED is a 6500 K color temperature, which was measured to be centered about 580 nm. It was made to pass through a 1500-grit diffuser (Thorlabs DG10-1500-MD) and a long-pass filter (Thorlabs FEL0550) that attenuates wavelengths below 550 nm. Thus, the total intensity illuminating the embryos was measured to be $11\mathrm{mW}/{\mathrm{cm}}^2$ at the peak wavelength of 580 nm. The imaging sensitivity of the system was 60 dB but was increased by averaging. We performed accumulation of 40 to 160 images to produce a single en-face image. The number of images used was increased linearly with depth to compensate the reduced signal-to-noise ratio due to scattering and attenuation. This yielded a system sensitivity of $\sim 76\mathrm{dB}$ at all depths. The resolution of the system was determined to be 1.5 and $1.0\mu \mathrm{m}$ in the axial and lateral directions, respectively. Acquisition of a 60-stack image took, on average, 8 min because of the low light intensity used; however, the time can be decreased significantly by reducing the integration time and increasing the illumination intensity. A custom LabView program was written to control the position of the actuators, conduct phase stepping with the PZT, provide a trigger for the camera, and save the data in 16-bit format. The data were acquired by using a phase stepping algorithm similar to that described in Ref. 23.

Fig. 1

Full-field optical coherence tomography (FF-OCT) system schematic. The system features an inverted microscope in the sample path to provide minimal disturbance to embryos during imaging.

2.5.

Image Processing

The image processing steps used to isolate the cell cytoplasm for further analysis are shown in Fig. 2. Embryos underwent an initial preprocessing step to remove artifacts and ensure consistent brightness for all images in the 3-D stack. Wiener filtering and 3-D Gaussian blurring were then performed to improve the signal from the embryo while suppressing noise (step 1). Edge detection and dilation were conducted on a two-dimensional projection image of the 3-D stack, and the cell cytoplasm was distinguished from the zona pellucida and background by repeated thresholding (step 2). The cell outline was identified by the application of circle Hough transform to the remaining pixels (step 3). The outlines of the cell and pronuclei in each slice were identified via repetition of steps 2 and 3 on each slice in the 3-D stack of images (step 4). Finally, a 3-D binary mask was created to identify the embryo cytoplasm (step 5).

Fig. 2

Overview of the image processing procedure performed to identify pixels inside the cell cytoplasm.

3.1.

Full-Field Optical Coherence Tomography Reveals Visible Changes in Embryos after Vitrification

We imaged one-cell mouse embryos before and after vitrification using both FF-OCT and fluorescence microscopy; as described previously, embryos in group 2 were stained to enable visualization of mitochondria. Consistent with prior literature,^12,24 we found qualitative changes in the distribution of the intensity signal from mitochondria before and after vitrification in the fluorescence images, which also appeared to correlate with the distribution of high-intensity structures we observed in the FF-OCT images (Fig. 3). In both imaging modalities, the intensity signal appears fairly evenly distributed in fresh embryos before vitrification. After vitrification, mitochondria seem to aggregate in clusters close to the center of the embryo; similarly, there is a clustering of bright structures seen in the FF-OCT images after vitrification. The two imaging modalities appear quite similar, which implies that a significant component of the FF-OCT signal may come from light scattering and reflection caused by mitochondria.

Fig. 3

Vitrification changes the distribution of mitochondria in embryos, as seen in images of fluorescently stained mitochondria. Similar changes are also visible with FF-OCT. Two example embryos from our dataset are shown. For each embryo, all four images are from a single slice of the three-dimensional volume and are taken from the same embryo at the z-location with the maximum radius. $\text{Scale bar}=20\mu \mathrm{m}$.

3.2.

Changes in Embryo Cytoplasm Can Be Quantified with Our Automated Algorithm

In order to more precisely quantify the changes we saw in embryos due to vitrification, we developed an image processing algorithm to extract parameters from the FF-OCT images of fresh ($n=22$) and vitrified ($n=18$) embryos. We thus devised a mechanism to score embryos based on our general observations of the clustering of cellular contents. Extraction of the parameters to quantify such CA in the embryo is described below.

First, the intensity profile within the embryo cytoplasm was calculated going radially outward for all slices bordering the pronuclei in three dimensions. To maximize the signal associated with aggregation near the pronuclei and cell membranes, only slices contained within $5\mu \mathrm{m}$ of the centroid of the pronuclei were considered. For each slice, the cytoplasm of the embryo was split into 20 concentric disks of equal radius, and the average intensity of all cytoplasmic pixels in each disk was calculated [Fig. 4(a)]. Pixels inside the pronuclei and outside the cell boundary were excluded from the average. The average intensity from each disk was plotted as a function of disk radius, yielding a radial intensity profile for each embryo.

Fig. 4

Method to extract a cytoplasmic aggregation (CA) score. (a) Schematic of the radial segmentation of embryos for extracting the radial intensity profile. (b) Qualitative comparison of radial intensity profile shapes of embryos with high and low aggregation, as scored manually. Images of typical embryos with aggregated cytoplasm and a homogeneous cytoplasm are shown alongside the plot for reference.

The radial intensity profiles for all slices within $5\mu \mathrm{m}$ of the pronuclei were averaged to generate a final radial intensity profile for each embryo. Figure 4(b) shows radial intensity profiles of embryos having differing cytoplasmic distributions, with representative images of embryos having cytoplasmic distributions that were homogeneous or aggregated (classified manually) shown alongside for reference. From these distributions, it is evident that the embryos with manually classified homogeneous distributions have intensity profiles that are generally flat or increase slightly toward the outside of the cell. The embryos with manually classified aggregated distributions show intensity profiles that are high in the inner 50% of the cell and then appear to drop off toward the outside.

We then calculated various features based on the radial intensity profiles that were linear combinations of the elements of the radial intensity profile vector (i.e., sums and differences of various points along the intensity profile) and calculated the accuracy with which these features could reproduce our manually classified cytoplasm distributions. We ultimately found that a single feature was sufficient to automatically classify the embryos and that our automatic classification showed good agreement with our manual classification. We call this feature the CA score, and it is calculated as follows:

The CA scores of embryos with manually classified aggregated and homogeneous cytoplasm distributions exhibited only a small degree of overlap. Figure 5(a) shows a smoothed kernel density plot of the CA scores. In Fig. 5(b), we show a receiver operating characteristic (ROC) curve that evaluates the performance of the CA score as a classifier. The area under the ROC curve is 0.94, which is indicative of a highly accurate classifier. It should be noted that the ground truth classification (i.e., manual classification as homogeneous or aggregated) is itself subject to human error. In particular, upon visual inspection, some embryos appeared to be slightly aggregated or in-between the two patterns. Because our automatic classifier lacks the bias and subjectivity of a human, its accuracy may be understated.

Fig. 5

(a) The CA score extracted from the radial intensity profile can accurately reproduce manual classification of embryo cytoplasm distributions. (b) The area under the ROC curve is 0.94 when using the CA score to distinguish embryos by clumping.

3.3.

Vitrification Causes an Increase in Embryo Cytoplasmic Aggregation

We next sought to determine whether vitrification induced significant changes in the distribution of the FF-OCT signal within embryo cytoplasm distribution. Using the CA scores we had previously calculated for our fresh ($n=22$) and vitrified ($n=18$) embryos, we found that vitrification caused significant changes in embryo cytoplasm morphology ($p\ll 0.001$) (Fig. 6). After vitrification, 94% of the embryos for which we had both pre- and postvitrification images exhibited an increase in the CA score. On average, the change in the CA score after vitrification was 0.084 with a standard deviation of 0.057.

Fig. 6

Cytoplasm aggregation quantified from FF-OCT images increases postvitrification. The CA score between fresh ($n=22$) and vitrified ($n=18$) embryos shows a statistically significant difference ($p<1\mathrm{e}-6$).

We also calculated a CA score for the fluorescence images of mitochondria to confirm that vitrification causes aggregation of the intensity signal for this imaging modality. We applied a similar processing method to the fluorescence images to extract radial intensity profiles and then calculated the CA score of embryos before ($n=10$) and after ($n=10$) vitrification. It is important to note that in this case, the CA score directly correlates with the distribution of mitochondria. As shown in Fig. 7(a), the radial intensity profiles of the fluorescence images appear slightly different from those of the FF-OCT images: the radial intensity profiles of the fluorescence images are dim near the outer edge of the cell because they lack the bright signal from the cell membrane that is seen in many FF-OCT images.

Fig. 7

The radial intensity profile extracted from fluorescent images of mitochondria quantifies the increased aggregation of mitochondria in the center of the cell postvitrification. (a) While these images do not possess a bright outline at the edges seen in FF-OCT, the radial profile shows a steeper decline away from the center in vitrified embryos compared to previtrification. (b) The mitochondria show significantly more aggregation after vitrification based on the CA score ($p=0.006$).

Otherwise, the radial intensity profiles of the fluorescence images are similar to those of the FF-OCT images in that the profiles of vitrified embryos experience a drop-off in signal closer to the center of the cell compared to fresh embryos. When quantifying the differences in the radial intensity profiles with the CA score, it can be seen that vitrified embryos again have a higher CA score than fresh embryos. Therefore, the CA score accurately captured a significant ($p=0.005$) increase in aggregation of mitochondria in vitrified embryos [Fig. 7(b)].

3.4.

Full-Field Optical Coherence Tomography May Be Used as a Label-Free Measure of Mitochondria Distribution

Because we observed that the fluorescence and FF-OCT images appeared qualitatively very similar, we sought to determine the extent to which the FF-OCT signal likely originated from the mitochondria and, therefore, whether FF-OCT could indirectly provide insight into the effects of vitrification on embryo mitochondria. We compared the CA scores of embryos that were imaged with both fluorescence and FF-OCT ($n=10$) and found a visible relationship between the CA scores from the two imaging modalities (Fig. 8). Our results suggest that aggregation in FF-OCT statistically correlates with mitochondrial aggregation viewed with fluorescence with high significance ($p=0.002$, $r\text{-}\text{squared}=0.42$). Hence, we conclude that, aside from the bright signal from the cell membrane, a significant component of the FF-OCT signal must originate from the distribution of mitochondria.

Fig. 8

Embryos that show aggregation under FF-OCT also show aggregation of stained mitochondria ($n=8$ for previt, $n=10$ for postvit). The correlation between the two imaging modalities ($r\text{-}\text{square}=0.42$, $p=0.002$) suggests that the FF-OCT signal partly originates from mitochondria.