2.1.

Design and Fabrication of Microchannel Network to Simulate Tumor Vasculature

The normal and tumor vasculature was adopted from Carmeliet and Jain⁵ and Less et al.⁶ with modification, as shown in Fig. 1. The main arteriole branches into a small arteriolar network to supply the capillaries. Then the capillaries drained into the small venules to incorporate into the main venules. The tumor vasculature incorporates trifurcations, true loops, and self-loops, which are specialized features of tumor vasculature. The diameter of vessel in our vasculature decreases as the branch increases. The mean values of main vessel widths of tumor vasculature we designed were 0.4 mm and the width of smallest branch was $\sim 0.02\mathrm{mm}$, which were designed according to Less et al.⁵

PDMS phantoms were fabricated using a SYLGARD^® 184 silicone elastomer kit (Dow Corning, Midland, Michigan). The material was used because its refractive index ($\sim 1.4$) approximates that of soft mammalian tissue (1.33 to 1.50)³³ and it is stable, elastic, and of adequate strength after curing.³⁰ The master mold was produced using SU-8 negative epoxy (MicroChem Corp., Newton, Massachusetts), and silicon pellet. The fabrication process followed the standard procedure of soft lithography.³⁴

Figure 2 illustrates the design of a sandwiched PDMS phantom that simulates oxygen metabolism and hypoxia in tumor vasculature. The phantom consists of an upper piece and a lower piece separated by a dialysis membrane. Fresh chicken blood is circulated through the microchannel in the upper piece to simulate tissue vasculature. Sodium sulfite solution is circulated through the microchannel in the lower piece to simulate oxygen consumption. The microchannels are 0.6-mm wide, 0.1-mm deep, and 20-mm long. The dialysis membrane enables selective transport of sodium sulfite to the upper microchannel and blockage of hemoglobin to the lower microchannel. Such a sandwiched phantom design will facilitate an effective control of blood oxygenation by adjusting the flow and the concentration of the sodium sulfite solution.

Fig. 2

Design of a sandwiched polydimethylsiloxane (PDMS) phantom with microfluidic assembly to simulate blood flow dynamics and metabolism in vascular network. The phantom consists of an upper piece and a lower piece separated by a dialysis membrane.

To prepare the above sandwiched PDMS phantom, the upper and the lower pieces were first fabricated by soft lithography. The two pieces were then sandwiched by a dialysis membrane and assembled by a modified plasma bonding process.³⁵ Materials for the upper piece and the lower piece were prepared by mixing PDMS and the curing agent at ratios of $10:1$ and $20:1$, respectively. Because the upper and the lower PDMS pieces had different material concentrations, reactive molecules remain at the interface between these two pieces according to Unger et al.³⁶ The upper and the lower pieces were placed in an oven (DHG-9070A, Shanghai Sanfa Scientific Instrument Co., Ltd., China) at a constant temperature of 65°C for curing. Right before completion of curing, a dialysis membrane was soaked, unfolded, and placed between two PDMS pieces, with the membrane carefully positioned to fully cover the microchannels. The assembly was then treated by a plasma bonding machine (WH-1000Z, Wenhao Chip Technology Co., Ltd., China), followed by further curing at 80°C for 12 h. The results of an additional experiment demonstrated that the permeability of the dialysis membrane was not degraded when it was placed under 80°C for 12 h. The fabrication and assembling process for the sandwiched phantom is illustrated in Fig. 3. The PDMS phantoms that simulate normal and tumor vascular networks, as shown in Figs. 1(a) and 1(b), were produced by a soft lithographic process, as described earlier.³⁵ First, the master mold of the vascular phantom was fabricated by soft lithography. Second, PDMS was cast onto the master mold and cured at 65°C for 1 h. Third, the PDMS replica of the master containing vascular network channels was peeled away from the silicon wafer. Finally, the PDMS replica and a flat slab of PDMS were treated by a plasma bonding machine, followed by further bonding at 80°C for 12 h as was done for the sandwiched PDMS phantom.

Fig. 3

Microfabrication processes for microchannel network with simulated metabolism.

2.2.

Numerical Simulation of Blood Flow Velocity in Vascular Network

The blood flow patterns of microvascular phantoms at different conditions were simulated using the commercial computational fluid dynamics (CFD) package FLUENT 6.3.26 (Ansys, Canonsburg, Pennsylvania). The simulation was based on the approximations of constant, uncompressible, and laminar flow and a rigid microfluidic device. The input flow rate was $10\mathrm{ml}/\mathrm{h}$, the blood density was $1060\mathrm{kg}/{\mathrm{m}}^3$, and the dynamic viscosity was $0.05\mathrm{Ns}/{\mathrm{m}}^2$.

2.3.

Experimental System

The simulated blood flow patterns were evaluated by our dual-mode imaging system in comparison with the actual experimental results. The dual-mode imaging system integrated LASCA and multispectral imaging for noninvasive assessment of blood perfusion and oxygenation.³⁷ The scientific principle and the engineering design of the dual-mode imaging system have been discussed in detail by Ren et al.³⁷ Figure 4 shows the dual-mode imaging system setup. The imaging component of the system was a 12-bit CCD camera (Microvision Digital Imaging Technology Co. Ltd., China) with a resolution of $1392\times 1040\text{pixels}$. An acousto-optic tunable filter (AOTF), tunable light source (500 to 900 nm, 10-nm bandwidth at 633 nm, Brimrose), and a laser device ($\lambda =785\mathrm{nm}$, 150 mW, Changchun New Industries, China) were connected to a light ring via a bifurcated optical fiber. A syringe pump (Pump 33, Harvard Apparatus) was used to drive blood and sodium sulfite solution and to control their flow rate in the phantom. Such a dual-mode imaging system enabled noninvasive and simultaneous imaging of tissue oxygenation and perfusion.

Fig. 4

Experimental setup for noninvasive imaging vascular perfusion and oxygenation.

2.4.

Multimodal Imaging of Perfusion and Oxygenation

2.4.2.

Oxygenation imaging of the phantom that simulates tumor hypoxia

A sandwiched phantom, as shown in Fig. 2, was fabricated to simulate blood oxygen metabolism, vascular permeability, and tumor hypoxia. Figure 5 illustrates a schematic drawing of an experiment setup. Fresh chicken blood was circulated through the microchannel in the upper piece of the sandwiched phantom to simulate the blood oxygen supply. Sodium sulfite solution ($25\mathrm{mg}/\mathrm{ml}$) was circulated through the microchannel in the lower piece of sandwiched phantom to simulate cellular oxygen consumption. A dialysis membrane was sandwiched between two microchannels to simulate vascular permeability. Tumor vascular metabolism was simulated by varying blood flow rate ($Q_{\mathrm{b}}$) and sodium sulfite solution flow rate ($Q_{\mathrm{s}}$), which might be equivalent to change the rates of oxygen supply and consumption. Dynamic control of oxygenation was demonstrated following three steps. First, sodium sulfite was continuously added and mixed with fresh blood until the oxygenation level approximated zero, as measured by a Moor oxygenation monitor (VMS-OXY, Moor Instruments Inc., Devon, United Kingdom). Then fresh chicken blood and oxygen-free blood were used to calibrate the oxygenation measurements in the microchannel. Second, the inlet blood flow rate $Q_{\mathrm{b}}$ was controlled at 10 levels from 1 to $10\mathrm{ml}/\mathrm{h}$, and the inlet sodium sulfite flow rate $Q_{\mathrm{s}}$ was controlled at two levels of 1 and $3\mathrm{ml}/\mathrm{h}$, respectively. Third, the inlet sodium sulfite flow blood flow rate, $Q_{\mathrm{s}}$, was controlled at 10 levels from 1 to $10\mathrm{ml}/\mathrm{h}$, and the inlet blood flow rate, $Q_{\mathrm{b}}$, was controlled at two levels of 1 and $3\mathrm{ml}/\mathrm{h}$, respectively. The designated flow rates were maintained for 2 minutes until the ${\mathrm{StO}}_2$ of the circulated blood was stabilized. Then the multispectral images were acquired at each ${\mathrm{StO}}_2$ plateau, the ${\mathrm{StO}}_2$ map was reconstructed by a wide gap second derivative reflectometry algorithm,³⁸ and was calibrated to full scale.

Fig. 5

Schematic drawing of an experimental setup for simulating blood oxygen metabolism and tumor hypoxia.

To obtain multispectral data set, a sample to be measured and a 99% reflected diffuser (National Institute of Standards and Technology, Gaithersburg, Maryland) were placed within the field of view of the CCD camera at an equal height. Then an image of the dark environment was captured to record the dark current noise of the CCD chip. An AOTF tunable light source was controlled to illuminate the phantom with the specific wavelength beam (544, 552, 568, 576, 592, and 600 nm) successively, while the autoexposure adjusting function was run to set up the optimal exposure time of each wavelength in case of improper intensity of the sample. After that, the monochromatic images of every wavelength were taken and stored at a corresponding exposure time for further processing.

The imaging algorithm includes four consecutive steps of calculating absorption–scattering ratio, deriving wide-gap second derivative, obtaining analytical expression of the second derivative ratio, and curve fitting for ${\mathrm{StO}}_2$ calculation. More details about the algorithm were introduced by Huang et al.³⁹

2.5.

Simulation of Blood Perfusion in Tumor Vasculature

PDMS phantoms designed as in Fig. 1(b) were used to simulate the blood perfusion of the tumor vasculature. The three-dimensional size of PDMS phantoms were $30\mathrm{mm}\times 30\mathrm{mm}\times 4\mathrm{mm}$. The microchannels of tumor vasculature were 0.1 mm deep. The maximum width of microchannels was 0.4 mm, and the minimum width was 0.02 mm. The blood perfusion was evaluated by LASCA at different flow rates and compared with that of a normal tissue vascular network.

3.1.

Multimodal Imaging of Blood Perfusion and Oxygenation

3.1.1.

Perfusion imaging of the phantom that simulates tissue vascular network

Figure 6(a) shows the photographic image of the microchannel network that simulated the regularly patterned and normally functioned tissue vascular network. The microchannel network was circulated with chicken blood to simulate tissue perfusion. Figure 6(b) shows the numerical simulation of the blood velocity distribution for the vascular network, as designed in Fig. 6(a). Figure 6(c) shows the blood perfusion map of the actual microchannel network measured by laser speckle contrast imaging. LASCA enabled real-time and dynamic imaging of relative blood flow within tissue vasculature. Since the technique is based on pixel operation of the laser speckle pattern, it has relatively low spatial resolution in comparison with other imaging methods.^40,41 In Figs. 6(b) and 6(c), the data are normalized by the inlet flow velocity. According to Figs. 6(b) and 6(c), numerical simulation and the actual experiment yield similar blood flow patterns at the same inlet flow rate of $10\mathrm{ml}/\mathrm{h}$. Furthermore, such a vascular topology that simulates normal tissue yields uniform blood flow distributions at the individual network levels, as denoted by numbers 1 to 6 in Fig. 6(b). Six regions of interest (ROIs) as marked in Figs. 6(b) and 6(c) were selected in order to explore the correlation between the simulated and the experimental flow patterns of the designed vascular network. The perfusion levels within the individual ROIs as marked in Fig. 6(c) were calculated by the LASCA algorithms and normalized by the inlet flow rate of $10\mathrm{ml}/\mathrm{h}$ for 10 consecutive perfusion maps. These 10 perfusion maps were then averaged and the normalized perfusion values were compared with the simulation results for each ROI. Figure 6(d) plots actual the blood velocity distribution normalized by the inlet velocity, which is linearly correlated with the simulation. Linear correlation is observed between the simulated and the measured blood velocities ($R=0.985$), indicating the technical feasibility of using microchannel networks to simulate the designated microvascular networks and the resultant blood flow patterns. Correlation between the simulated and the measured blood velocities follows a regression equation of $y=0.986x+0.022$.

Fig. 6

Microfabricated vascular network is able to achieve the designated flow pattern in coincidence with that of the numerical simulation. (a) Photographic image of the microfabricated vascular network. The scale bar is 5 mm. (b) Blood velocity distribution in the vessel network simulated by a CFD software package. (c) Actual blood perfusion map of the vascular simulating phantom imaged by laser speckle technique. (d) Actual blood velocity distribution normalized by the inlet velocity is linearly correlated with the simulation. $X$ axis: simulated blood velocities at different levels in the vascular network. $Y$ axis: actual blood velocities at the corresponding levels in the vascular network. The data are normalized by the inlet flow velocity.

3.1.2.

Oxygen imaging of the phantom that simulates tumor hypoxia

For the sandwiched phantom as shown in Fig. 2, microchannels in the upper layer and the lower layer were circulated with the chicken blood and the sodium sulfite solution, respectively. The blood ${\mathrm{StO}}_2$ in the upper layer was imaged by our multispectral imaging system as the flow rates of the blood and the sodium sulfite solution were changed at different levels. The blood ${\mathrm{StO}}_2$ levels of fresh blood and oxygen-free blood we measured with our systems were defined as 1 and 0, respectively. The blood ${\mathrm{StO}}_2$ level of the simulated blood vessel was normalized by the blood${\mathrm{StO}}_2$ level of fresh blood and oxygen-free blood. Figures 7(a) and 7(b) plot the averaged ${\mathrm{StO}}_2$ level of the simulated blood vessel in response to the flow rate changes for the chicken blood and the sodium sulfite solution. According to Fig. 7(a), for a fixed sodium sulfite flow rate $Q_{\mathrm{s}}$, the blood ${\mathrm{StO}}_2$ level increases as the blood flow rate $Q_{\mathrm{b}}$ increases and tend to reach the blood ${\mathrm{StO}}_2$ level of fresh blood. Meanwhile, Fig. 7(b) indicates that for a fixed blood flow rate $Q_{\mathrm{b}}$, the blood ${\mathrm{StO}}_2$ level decreases as the sodium sulfite flow rate $Q_{\mathrm{s}}$ increases and approaches close to the blood ${\mathrm{StO}}_2$ level of oxygen-free blood. Considering that the oxygen consumption is by sodium sulfite through the dialysis membrane, this experimental result demonstrates the technical potential of simulating oxygen metabolism and vascular permeability by controlling $Q_{\mathrm{b}}$ and $Q_{\mathrm{s}}$ in the sandwiched phantom. A further increase of $Q_{\mathrm{s}}$ and reduction of $Q_{\mathrm{b}}$ (for example, $Q_{\mathrm{s}}=10\mathrm{ml}/\mathrm{h}$ and $Q_{\mathrm{b}}=1\mathrm{ml}/\mathrm{h}$) could simulate the hypoxic microenvironment and oxygen dynamics in solid tumor.

Fig. 7

Microfabricated vessel network is able to achieve the designated oxygenation levels by controlling the flow rates for blood and for sodium sulfite solution. (a) $X$ axis: the blood flow rates in the upper microchannel and $Y$ axis: actual blood oxygen saturation (${\mathrm{StO}}_2$) level of simulated blood vessel, normalized by the blood ${\mathrm{StO}}_2$ level of fresh and oxygen-free blood. (b) $X$ axis: the blood flow rates in the lower microchannel and $Y$ axis: actual blood ${\mathrm{StO}}_2$ level of simulated blood vessel, normalized by the blood ${\mathrm{StO}}_2$ level of fresh and oxygen-free blood.

However, increasing the flow rate of the sodium sulfite solution may build up the internal pressure significantly and cause leakage of the dialysis membrane. The limiting flow rate of the existing experimental setup is $30\mathrm{ml}/\mathrm{h}$. By improving the design of the sandwiched vascular phantom and optimizing the manufacturing process, it is better to simulate the extremely hypoxic tissue condition and facilitate the validation of various therapeutic strategies that overcome hypoxia-induced chemoresistance in cancer treatment.

Despite the successful demonstration of the oxygenation control using the sandwiched tissue-simulating phantom, further improvement is necessary in order to realize the proposed tool, a clinically useful platform for understanding the mechanism of cancer metabolism, validating the performance of spectral imaging devices, and optimizing the techniques for anticancer drug delivery. First, the stimulated tumor hypoxia was based on a single microchannel instead of a microchannel network, owing to the difficulty to seal a large area of microchannels without leakage. It is necessary to develop a better packaging process to simulate the vascular permeability in a large tumor area. Second, the existing method of controlling the upper layer oxygenation by a lower layer flow of sodium sulfite is not the most efficient way to simulate tissue oxygen metabolism. It is technically possible to incorporate oxygen-consumption ingredients in the phantom or integrate blood vessels and oxygen-consumption vessels in one piece in order to improve the oxygenation control efficiency. Third, the blood ${\mathrm{StO}}_2$ level of the simulated blood vessel is not sufficiently accurate, which is the result of the experimental system error and oxygen calibration error. It is more accurate and efficient to measure the blood oxygenation of a simulated blood vessel with the mature commercial instrument. Finally, the current work of hypoxia simulation is still very preliminary. Quantitative analysis and numerical simulation are necessary in order to optimize the oxygen transport between the two layers for the best simulation outcome.

3.2.

Simulated Blood Perfusion in Tumor Vasculature

Figure 8(a) shows the microfabricated microchannel network that simulated the tortuous tumor vasculature and irregular branching. Such an irregular vascular topography leads to nonuniform and poor regional blood velocity, as simulated by the CFD software in Fig. 8(b). The actual blood perfusion pattern of the microchannel network was acquired by laser speckle contrast imaging and is shown in Fig. 8(c). An inlet flow rate of $10\mathrm{ml}/\mathrm{h}$ was used in both numerical simulation and experimental validation. According to the figures, the blood perfusion distribution acquired by the experiment is in agreement with that of numerical simulation. Comparison between the perfusion patterns for normal and tumor vessels at the same inlet flow rate [i.e., Fig. 6(c) versus Fig. 8(c)] indicates that the structural anomaly of tumor vasculature leads to less uniform blood perfusion, consistent with the previous observations by Dudar and Jain.⁴² Furthermore, the nonuniform perfusion pattern in Fig. 8(c) is able to simulate different heterogeneous regions as observed in a solid tumor.⁴³ For example, ROI#3 has an extremely large blood perfusion that simulates the “well vascularized region” where tumor cells have a sufficient oxygen and nutrition supply for rapid proliferation.⁴³ ROI#2 has a relatively small blood perfusion that simulates the “seminecrotic region” where the slow rates of oxygen and nutrition supply lead to hypoxic cells.⁴³ ROI#1 has either no blood vessel or close to zero blood perfusion, simulating the “necrotic region” that leads to tissue death.⁴³

Fig. 8

Simulated and actual blood velocity distribution in the tumor vessel network. (a) Photographic image of the tumor vascular network. The scale bar is 5 mm. (b) Simulated blood velocity distribution in the tumor vessel network. (c) Actual blood perfusion map of the tumor vascular simulating phantom imaged by laser speckle technique.

Figures 9(a) and 9(b) compare the perfusion maps of the normal and the tumor vascular networks at the same inlet flow rate of $10\mathrm{ml}/\mathrm{h}$. Sixteen ROIs at the lowest branch level, as marked in Fig. 9(a), are selected to calculate the averaged perfusion values of the normal vascular network. Similarly, 16 ROIs at the lowest branch level, as marked in Fig. 9(b), are selected to calculate the averaged perfusion values of the tumor vascular network. The ROIs are selected based on the designated distances to the inlets and are, therefore, reproducible for different experiments. For each ROI, the perfusion value is calculated by averaging five perfusion maps. Statistical analysis shows that the normal vascular network has a normalized perfusion of ($0.359\pm 0.111$) while the normalized perfusion for the tumor vascular network is ($0.100\pm 0.063$). A significant difference exists between the normal and tumor vascular perfusion levels, with $P<0.05$. Twelve ROIs, as marked with circles in the inlet and outlet regions of tumor vasculature, respectively, are selected for the calculation of the average perfusion values, as shown in Fig. 9(b). The criteria of selecting ROIs is the distance to the inlet and outlet. The perfusions of each the ROIs in five perfusion maps are calculated by LASCA algorithms and the average values are obtained. Twelve average perfusions of inlet and outlet ROIs in tumor vascular network are averaged and normalized. The normalized perfusion for the inlet region of tumor vascular network is ($0.253\pm 0.145$), whereas the outlet region has a normalized perfusion of ($0.237\pm 0.111$). This illustrates that perfusion is similar between the inlet region and the outlet region.

Fig. 9

Flow comparison between normal and tumor vasculature. (a) Actual blood perfusion map of the normal vascular simulating phantom imaged by laser speckle technique. Sixteen ROIs at the lowest branch level are selected to calculate the averaged perfusion values of a normal vascular network. (b) Actual blood perfusion map of the tumor vascular simulating phantom imaged by laser speckle technique. Seven ROIs are selected from the tumor vascular network for the calculation of the average perfusion values. All the data are normalized by the inlet blood perfusion.

Similar analysis will also enable us to validate and optimize the drug delivery strategies considering that the structural anomalies of tumor vasculature significantly affect the delivery efficacy of many anticancer therapies.

3.3.

Simulated Hypoxia in Tumor Vasculature

Fresh chicken blood was mixed with sodium sulfite solution ($25\mathrm{mg}/\mathrm{ml}$) at volume ratios of $1:0$, $10:1$, and $5:1$, respectively. The mixture solution was injected to the simulated tumor vascular network by a syringe pump. At each oxygenation plateau, multispectral images were acquired. The ${\mathrm{StO}}_2$ levels of the simulated tumor vascular network were calculated by our imaging algorithm.^23,39

The level of hypoxia in tumor vasculature was simulated by adjusting the volumetric ratio between blood and sodium sulfite solution. Figure 10 shows that the ${\mathrm{StO}}_2$ of the tumor vascular network decreases as the ratio of blood and sodium sulfite solution decreases. In this experiment, the oxygenation control was achieved by mixing blood with sodium hydrosulfite solution at different ratios rather than circulating sodium hydrosulfite and blood in different microchannels, as discussed earlier. One major obstacle is the difficulty of sealing the dialysis membrane between two layers of the tumor vascular network without leaking. It is our future work to integrate a dialysis membrane with tumor vascular network in order to produce microchannel networks that simulate the structural and metabolic anomalies of tumor tissue.

Fig. 10

Oxygenation maps of PDMS tumor vascular phantom with the different proportions of blood and sodium sulfite solution. (a) Blood, (b) blood mixed with sodium sulfite solution at a ratio of $10:1$ by volume, and (c) blood mixed with sodium sulfite solution at a ratio of $5:1$ by volume.