A Clinical‐Scale Microfluidic Respiratory Assist Device with 3D Branching Vascular Networks

Abstract Recent global events such as COVID‐19 pandemic amid rising rates of chronic lung diseases highlight the need for safer, simpler, and more available treatments for respiratory failure, with increasing interest in extracorporeal membrane oxygenation (ECMO). A key factor limiting use of this technology is the complexity of the blood circuit, resulting in clotting and bleeding and necessitating treatment in specialized care centers. Microfluidic oxygenators represent a promising potential solution, but have not reached the scale or performance required for comparison with conventional hollow fiber membrane oxygenators (HFMOs). Here the development and demonstration of the first microfluidic respiratory assist device at a clinical scale is reported, demonstrating efficient oxygen transfer at blood flow rates of 750 mL min⁻1, the highest ever reported for a microfluidic device. The central innovation of this technology is a fully 3D branching network of blood channels mimicking key features of the physiological microcirculation by avoiding anomalous blood flows that lead to thrombus formation and blood damage in conventional oxygenators. Low, stable blood pressure drop, low hemolysis, and consistent oxygen transfer, in 24‐hour pilot large animal experiments are demonstrated – a key step toward translation of this technology to the clinic for treatment of a range of lung diseases.

The model sets the concentration of the oxygen on the PDMS surfaces in the oxygen channels by the gas pressurein the oxygen channels (760 mmHg + back pressure) multiplied by the solubility of the membrane (S mem ). The partition coefficient between the membrane and blood (K mb ) is calculated by dividing the solubility of the blood by the solubility of the membrane. This value is used to calculate the oxygen transfer between the membrane and the blood (the discontinuity in solubility at that interface equates to a discontinuity in the concentration because the partial pressure must be equal).
In total, there are four parameters that the model uses to calculate oxygen transfer; the diffusivity and solubility of the membrane and the blood. Values of diffusivity (D mem , D eff ) and solubility (S mem and S eff ) of the membrane and blood were previously determined empirically by fitting the modeling results to data from prior experiments that varied the channel height, length, oxygen pressure and flow rate. There is a 3 μm cell-free plasma layer that is included in the model; this layer is so thin compared to the rest of the channel that its properties can change by multiple orders of magnitude without affecting the results. The diffusivity of this layer was set to literature values for plasma in this study (see Table S1).
The flow was modeled as a laminar flow entering the channel fully developed. There is a 3 μm cell-free plasma layer that has Newtonian properties and viscosity set to 0.0035 Pa*s and density of 1025 kg/m 3 .
[76] The rest of the blood channel is modeled as a Carreau fluid with properties defined as shown in Table S1.
The width of the channel is fixed at 500 μm (250 um for half-width in the simulation due to symmetry plane) and the distance between ribs is similarly fixed at 1000 μm due to machining constraints. The height and length of the channel were varied to determine the optimum dimensions for oxygen transfer (given the membrane thickness, oxygen pressure, and desired blood flow). Various blood flows from 5 to 150 mL/min were run to determine the dependency of oxygen transfer on blood flow rate.

Expression
Value Description  The simulations were run with a structured mesh in the blood layer, a free tetrahedral mesh in the membrane, and boundary layers were added in the blood and plasma layers ( Figure S1). A total of 61,697 elements were used to accurately model both blood flow and oxygen transfer. Figure S1: Image of the mesh distribution in the COMSOL oxygen transfer model.
All results from the simulation were used to derive an optimum channel height and length given design constraints (discussed in methods section of main text). The number of channels per bank was fixed at 184 due to the channel width and a physical limit to the maximum overall width of the device, so the number of banks was calculated based on the required channel length. The 4and 5-bank designs had similar overall device length (Table S2); the 4-bank was chosen to maintain a bifurcating manifold leading into the banks.
The shear rates were calculated for the 4-bank design and determined that the channel height needed to be slightly taller to reduce the shear rate in the channels. The channel height was increased from 147 µm to 160 µm, and the channel length recalculated to be 9.1 cm to accommodate the increased height.

Flowlo distribution
The flow distribution and pressure were checked with CFD. One bank of the device was modeled independently to check that the flow rate in each channel was uniform, and the manifold that distributed flow to the banks was modeled separately with a single resistive channel between the inlet and outlet manifolds for computational efficiency. Both models show uniform flow ( Figures  S2 and S3) and pressure drop (Figures S4 and S5) across the channels and device.

Updated oxygen layer with additional membrane support features
In the original BLOx design, there was significant overlap of the oxygen channels with the wide blood distribution channels ( Figure S7a), which lead to a high degree of channel deformation due to the deflection of the membranes in these regions at elevated oxygen pressures. An updated oxygen channel layer and the addition of extra membrane support structures (Figure S7b/c/d) had the effect of minimizing the membrane deflection to an acceptable level at 400mmHg, resulting in a greatly reduced pressure drop across the device (original layout = 367 mmHg, updated layout = 110 mmHg) at the nominal flow rate of 100 mL/min. Figure S7: Updated oxygen distribution layer to minimize overlap with blood distribution channels: a) original channel overlay with oxygen transfer channels overlapping with blood finger channels, b) updated channel overlay with no oxygen transfer channels over the blood finger channels and addition support features in the oxygen trunk channel where it overlaps with the blood finger channel, c) isometric view of additional oxygen channel support features, d) image of blood flowing through device with updated oxygen channel layer. Figure S8: Improvement in shear stress distribution in inlet transition region of the BLOx device compared to our previous design [75] using the iterative geometry refinement technique described in the main text.