Quantitative and longitudinal monitoring of cancer cell invasion in a three-dimensional in vitro model of oral cancer using optical coherence tomography

Abstract

We previously developed a three-dimensional (3D) organotypic culture model of oral squamous cell carcinoma (OSCC) by incorporating cancer-associated fibroblasts to replicate oral tissue architecture and the tumor microenvironment (TME). This model provides a relevant platform for investigating cancer cell invasion. Optical coherence tomography (OCT), a noninvasive, high-resolution imaging technique, enables both real-time and longitudinal observations. This study assessed the applicability and feasibility of combining OCT with deep learning for the quantitative, longitudinal monitoring of cancer cell invasion in our 3D model. OCT effectively captured cross-sectional images and identified three regions-original cancer cell region, invasive cancer cell region, and stromal layer-based on scattering intensity and optical density, demonstrating nondestructive visualization of tissue microarchitecture. Sequential OCT imaging facilitated 3D image reconstruction and repeated monitoring. Planimetric and volumetric analyses of 3D OCT images revealed internal structural alterations and enabled comparative evaluation of invasion behaviors across OSCC cell types, TME conditions, and culture durations. Moreover, the invasiveness parameter obtained from 3D OCT images strongly correlated with histomorphometric-based data, confirming its reliability. These findings support the use of OCT imaging as a promising tool for noninvasive, quantitative assessment of invasiveness in organotypic cancer models.

Keywords: 3D organotypic model; Cancer cell invasion; Cancer-associated fibroblasts; Deep learning; Optical coherence tomography; Oral squamous cell carcinoma; Tumor microenvironment.

Fig. 1

Research scheme of the study. Schematic of the research plan, methodology, and overall study design.

Fig. 2

Representative OCT images (X–Y slices) at three levels from top to bottom within 3D oral cancer models under eight conditions (HSC-2 and HSC-3 cells; repopulated with NOFs and CAFs; 14 and 21 days in culture). (A) X–Y slice of the original cancer cell region. Scale bars 100 μm. (B) X–Y slice of the invasive cancer cell region. Scale bar: 100 μm. (C) X–Y slice of the stromal layer. Scale bar: 100 μm. A color bar shown by gray-scale indicates the signal intensity (0 to 65535) of this OCT images.

Fig. 3

Representative panels of OCT cross-sectional (X–Z) images (left), deep learning–processed images (middle), and histological images (right) of 3D oral cancer models under 10 conditions (HSC-2 and HSC-3 cells; acellular stromal layer or repopulated with NOFs and CAFs; and 14 and 21 days in culture). (A) Representative panel of 3D oral cancer models in which HSC-2 or HSC-3 cells were seeded on top an acellular stromal layer cultured for 21 days. White scale bar: 100 μm; black scale bar: 50 μm. (B) Representative panel of 3D HSC-2 oral cancer models in which NOFs or CAFs repopulated the stromal layer cultured for 14 days. White scale bar: 100 μm; black scale bar: 50 μm. (C) Representative panel of 3D HSC-3 oral cancer models in which NOFs or CAFs repopulated the stromal layer cultured for 14 days. White scale bar: 100 μm; black scale bar: 50 μm. (D) Representative panel of 3D HSC-2 oral cancer models in which NOFs or CAFs repopulated the stromal layer cultured for 21 days. White scale bar: 100 μm; black scale bar: 50 μm. (E) Representative panel of 3D HSC-3 oral cancer models in which NOFs or CAFs repopulated the stromal layer cultured for 21 days. White scale bar: 100 μm; black scale bar: 50 μm. Cohesive invasion patterns and well-defined invasion fronts were observed in the HSC-2 CAF models, whereas the invasive cancer cell region in HSC-3 CAF models was more prominent and larger than in the HSC-2 CAF models.

Fig. 4

OCT cross-sectional image (X-Z slice) of the HSC-3 CAF model at 21 days in culture after deep learning processing. The right panel shows a higher-magnification view of the boxed area outlined with red solid line in the left panel. This image demonstrates the resolution of the OCT system used in this study. Brackets 1, 2, and 3 indicate the original cancer cell region, invasive cancer cell region, and stromal layer, respectively.

Fig. 5

Depth of invasion in 3D oral cancer models under different conditions. (A) Quantification of the “Depth of invasion (µm)” comparing four independent conditions (HSC-2 vs. HSC-3 and CAFs vs. NOFs). Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01. (N = 5). (B) “Depth of invasion (µm)” comparing models cultured for 14 vs. 21 days. Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01. (N = 5).

Fig. 6

Representative surface images of the hypothetical invasive front reconstructed from OCT data (A and B) and planimetric measurements of surface area (C, D) in 3D oral cancer models. (A) Surface images of the hypothetical invasive front in HSC-2 cells under four different conditions. (B) Surface images of the hypothetical invasive front in HSC-3 cells under four different conditions. (C) Quantitative comparison of surface area across four conditions (HSC-2 vs. HSC-3 and CAFs vs. NOFs). Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01. (N = 3–6). (D) Longitudinal comparison of surface area in the same models over time (14 vs. 21 days). Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01. (N = 3–6).

Fig. 7

Representative 3D OCT reconstructions of the invasive cancer cell region (A and B) and volumetric measurements in 3D oral cancer models (C and D). (A) Three-dimensional reconstructions of invasive cancer cell regions in the HSC-2 model under four different conditions. (B) Three-dimensional reconstructions of invasive cancer cell regions in the HSC-3 model under four different conditions. (C) Quantitative comparison of total invasive cancer cell volume across four conditions (HSC-2 vs. HSC-3 and CAFs vs. NOFs). Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01. (N = 3–6). (D) Longitudinal comparison of total invasive cancer cell volume in the same model over time (14 vs. 21 days). Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01. (N = 3–6).

Fig. 8

Comparison of mass invasion index from OCT-based volumetric analysis with invasion index from histomorphometric analysis in four 3D cancer models cultured for 21 days. Data are presented as means ± S.D. ns: not significant, *p < 0.05, **p < 0.01 (N = 3–6).