Abnormal morphology biases hematocrit distribution in tumor vasculature and contributes to heterogeneity in tissue oxygenation

Abstract

Oxygen heterogeneity in solid tumors is recognized as a limiting factor for therapeutic efficacy. This heterogeneity arises from the abnormal vascular structure of the tumor, but the precise mechanisms linking abnormal structure and compromised oxygen transport are only partially understood. In this paper, we investigate the role that red blood cell (RBC) transport plays in establishing oxygen heterogeneity in tumor tissue. We focus on heterogeneity driven by network effects, which are challenging to observe experimentally due to the reduced fields of view typically considered. Motivated by our findings of abnormal vascular patterns linked to deviations from current RBC transport theory, we calculated average vessel lengths [Formula: see text] and diameters [Formula: see text] from tumor allografts of three cancer cell lines and observed a substantial reduction in the ratio [Formula: see text] compared to physiological conditions. Mathematical modeling reveals that small values of the ratio λ (i.e., [Formula: see text]) can bias hematocrit distribution in tumor vascular networks and drive heterogeneous oxygenation of tumor tissue. Finally, we show an increase in the value of λ in tumor vascular networks following treatment with the antiangiogenic cancer agent DC101. Based on our findings, we propose λ as an effective way of monitoring the efficacy of antiangiogenic agents and as a proxy measure of perfusion and oxygenation in tumor tissue undergoing antiangiogenic treatment.

Keywords: anti-angiogenic agents; hematocrit dynamics; mathematical modelling; oxygen heterogeneity; tumor vasculature.

Fig. 1.

(A) Multiphoton image (0.83 $\mathrm{\mu }$m $\times$ 0.83 $\mathrm{\mu }$m $\times$ 3 $\mathrm{\mu }$m resolution) of a MC38 tumor vessel network obtained via an abdominal imaging window in mouse. Red, endothelial cells; green, tumor cells. White box shows region of interest. (Scale bar, 100 $\mathrm{\mu }$m.) (B) Region of interest showing endothelial cell-staining channel alone. Abnormal vascular morphology: reduced interbifurcation distance (between arrowheads) and complex topology (three- to two-branch merger, star). Arrows indicate direction of flow. (Scale bar, 50 $\mathrm{\mu }$m.) (C) RBC flow simulation in realistic network with vessel identities in computational domain reconstructed from region of interest. (D) RBC flow-simulation analysis. RBC enrichment/depletion after branching point. Solid line represents proportional partitioning based on flow rates. Existing PS theory predicts enrichment in higher-flow branch. RBC simulations (Sim) show inversion of effect in reduced interbifurcation case and strong attention in complex topology case. PS theory refers to ref. .

Fig. 2.

(A) Maximum intensity projection of multiphoton image stack of a tumor vessel network obtained via an abdominal imaging window in mouse from the MC38 dataset. Red, perfusion; cyan, endothelial cells; green, GFP tumor cells. (Scale bars: 1 mm.) (B) The stack is subsequently segmented and skeletonized, and distributions of vessel diameters and lengths are calculated. Red, vascular network segmentation; dark, skeletonization. Mouse MC38-5 in Table 1. (C–E) Scatter plots of vessel lengths vs. diameters for different cell lines studied: MC38 (mouse 3 in Table 1) (C), B16F10 (mouse 1 in Table 1) (D), and LLC (mouse 1 in Table 1) (E).

Fig. 3.

(A–C) Discharge hematocrit at different geometry branches: extended double-t geometry (A), double-t geometry (B), and cross geometry (C). (D) Example simulated RBCs in the double-t geometry: The vessel network is rendered semitransparent in gray, and the RBC membranes are rendered in red suspended in transparent blood plasma. (E) Schematic describing the impact of CFL dynamics on hematocrit split. (F and G) CFL width in opposite sides of channel 1: double-t geometry (F) and extended double-t geometry (G).

Fig. 4.

(A and B) For $\lambda =4.0$, the model with memory effects yields more pronounced oxygen heterogeneity than previous theory (11) predicts (i.e. more dispersed oxygen distribution) in A (model without memory effects) and B (model with memory effects) (spatial scales are in micrometers; vessels are shown in black for reference). (C) Violin plots show oxygen distributions for varying $\lambda$ and the two HS models under consideration. Heterogeneity increases with $\lambda$ for the model without memory effects, as expected, but the model with memory effects predicts increased heterogeneity for very low $\lambda$. The horizontal lines in oxygen distributions in C represent 25th, 50th, and 75th percentiles.

Fig. 5.

Vascular phenotypes in MC38 tumors over time following DC101 treatment compared with control (n = 5). DC101 $\lambda$ ratio raw data are given in SI Appendix, Table S3. (A) $\lambda$ ratio. (B) Bifurcation density. (C) Fraction of perfused vessels. Plots show mean (solid markers) and SE (error bars). t test for statistical significance, *$P<0.05$; **$P<0.01$.