OCT-SFS methods and analysis metrics

Imaging of the 3D cultures was performed using a custom built, spectral-domain optical coherence tomography (SD-OCT) system as shown in Figure 1A, which has been described in detail previously (30). Briefly, the system consists of three parts: the light source, a Michelson interferometer and a custom spectrometer. The light source consisted of a Ti:Sapphire laser with a central wavelength of 800 nm and bandwidth of 120 nm. Light from the source was linearly polarized (horizontal to the surface of the optical table; H polarized) and directed into the Michelson interferometer. The H polarized light was split into reference and sample arms, where the sample arm light (~6 mW) was focused by a 30 mm focal length lens upon the sample. The co-polarized backscattered light (H) interfered with the reference at the beam splitter. The interfered light (HH) was then directed to the custom spectrometer where spectral interferograms were collected into the first 2,048 pixels of a 4,096-pixel Dalsa Piranha line scan CCD camera, operated at an A-line rate of 2 kHz for this study. The resolution of the OCT system was ~10 µm × 3.0 µm (in aqueous medium) in x × z, and the signal-to-noise ratio (SNR) was ~108 dB. B-mode (cross-sectional) image frames were collected into 1,000×1,024 pixels over 1.5 mm × 1.5 mm (in aqueous medium) in x × z, respectively. 300 frames per time series were collected at a frame rate of 0.876±0.004 Hz in order to capture cellular dynamics (12,13), and then divided into three groups of n=100 consecutive images for independent analysis with cross-validation.

Overview of OCT-SFS hardware, data collection, and analysis pipeline. (A) Spectral domain OCT system; (B) a representative time-stack of 100 successive B-mode images for analysis; (C) metric for motility amplitude; (D) power spectral analysis of OCT fluctuations. FC, fiber coupler; BS, beam splitter; PBS, polarizing beam splitter; QWP, quarter wave plate; ROI, region of interest; OCT-SFS, optical coherence tomography speckle fluctuation spectroscopy.

All image analysis was performed as described in our previous studies (9,12,13). Briefly, depth-resolved OCT images were computed from raw spectral images after reference subtraction and digital dispersion compensation (31), where the intensity at each image pixel, I(x, z), was computed from the absolute value of the complex analytic signal obtained from Fourier transform of the spectral domain OCT data. A time series of OCT images of MEC organoids, as shown in Figure 1B, were used to generate movies for the SFS analysis. Each organoid, containing hundreds to thousands of cells, was identified and segmented as one region of interest (ROI) by custom semi-automated MATLAB scripts; two representative ROIs are color-coded in yellow in Figure 1B. The temporal fluctuation of speckle intensity at each pixel, I(x, z, t), was extracted from the image stack, as shown in Figure 1C, which was attributed to the motions of the intracellular particles that backscatter OCT light. These speckle fluctuations were characterized by two parameters: the so-called “motility amplitude”, M, that characterizes the modulation amplitude of the speckle fluctuations in time, and the power-law exponent of the decay of the power spectral density in frequency, α. M is an autocorrelation-based modified standard deviation that is normalized by average pixel intensity (12), and provides two complementary benefits for data analysis: the autocorrelation at each image pixel obtained at a given sampling time Δt, Γ(x, z, Δt), naturally omits shot noise that decorrelates instantaneously, while normalization by average pixel intensity eliminates the depth-dependent SNR roll-off, making M intensity- and depth-invariant. M at each pixel was then spatially averaged over each ROI to represent the M of an organoid. To characterize the frequency (f) dependence of the motility signals, the power spectral density S(x, z, f) was computed by a discrete Fourier transform of the time signal I(x, z, t) at each pixel. The spectra were then spatially averaged over each ROI, then fitted to an inverse power-law model with the power exponent α (12), as illustrated in Figure 1D. A goodness-of-fit test was performed for model fitting of each ROI spectrum, where only data exhibiting R²>0.98 compared to the model were included in the subsequent analysis. Figure 2A shows a cross-sectional OCT image of a representative data set. The motility metrics derived from this data set are visualized in Figure 2B,C,D, with more details of the visualization method described in (12). Together, the two metrics α and M characterized the intracellular motility of an organoid (one ROI), with ~12–24 values of α and M each computed over n=3 cultures at each culture condition (inhibitor type, inhibitor concentration, and cell line). To assess the time-evolution of intracellular motility, multiple comparison t-tests with Bonferroni correction were then performed to compare α and M for each culture condition against the pre-exposure α and M for that same condition. P values were calculated to indicate the statistically significant differences, where the critical P value was set to 0.01 for all comparisons.

Visualization of the motility metrics derived from a representative data set. (A) A representative cross-sectional OCT image of 6 live MEC organoids in 3D mono-culture of MCF7 cells (seed density 30,000/cm³) at 14 days culture time after seeding; (B) visualization of motility metrics on (A), where two of the organoids (yellow arrows) are enlarged in (C) and (D). The background gray-scale image indicates M at each pixel, and the yellow contour lines indicate the ROIs determined from the semi-automated segmentation. α is overlaid in each ROI as spherical glyphs with different colors and sizes. The color of the glyph indicates the value of α, while the size of the glyph represents R² from the power-law fitting. OCT, optical coherence tomography; MEC, mammary epithelial cell; ROI, region of interest.

As previously discovered (12,13), values of α and M are cell type-dependent, which is attributed to differences in morphology, metabolism, invasiveness, and ECM interactions. Therefore, in order to define a typical dynamic range for each metric, cell line-specific baselines were established based on the α and M values of live and fixed cells measured previously (13). A linear mapping was used from the raw α and M values to normalized values where 0% and 100% represent the fixed and live values, respectively. Note that raw α exhibits an increase under cell fixation, indicating that the speckle fluctuation spectra more rapidly decay in frequency (suppressing high frequency motions), while the linear scaling releases the users from constantly recalling this fact when assessing longitudinal data. It should be noted that the fixed values represent “death by fixation”, which is mechanistically different from inhibitor-induced cell death evaluated in this paper. In addition, the live values may be subject to biologic variations over time such as the phase of cell growth, which may cause untreated cells to exhibit initial values that are above or below 100%. However, this does not affect the observations and conclusions from longitudinal data, as effects of inhibitors are only reflected by the relative changes between the pre-exposure and post-exposure data.