C. Experimental setup

The experimental system used in this study refers to the one applied by Lv et al. (202025. Koullapis, P. G., Stylianou, F. S., Sznitman, J., Olsson, B., and Kassinos, S. C, “Towards whole-lung simulations of aerosol deposition: A model of the deep lung,” J. Aerosol Sci. 144, 105541 (2020). https://doi.org/10.1016/j.jaerosci.2020.105541), which consists of two main parts: a respiratory control system and a micro-PIV visualization system. As shown in Fig. 1(d), the experimental equipment includes a microscope, a laser, a precision syringe pump, PIV software and hardware, a synchronizer, etc. Combining the schematic shown in Fig. 1(c), the fluid flows in position 1, flows through the previous generation of alveoli and then splits evenly into two bundles of fluid at position 2, enters the next generation of alveolar tubes, and after repeating a similar motion as the previous generation, exits from positions 3 and 4, respectively. Positions 5, 6, and 7 are connected to the same pressure chamber, and this pressure chamber is connected to the air feeder on the syringe pump via a catheter. While the gas in the chamber is being pumped, the pressure in the chamber changes, thus controlling the expansion and contraction of all alveoli simultaneously.The micro-PIV system consists of a CCD camera (specifications of the 630 091 PowerView 4MP-HS camera, TSI), a double pulsed Nd-YAG laser (Vlite-135, Beamtech), a microscope (IX73, Olympus) with a 40× objective, a synchronizer, and a computer. The camera has a resolution of 2048 × 2048 pixels, with a pixel size of 7.4 × 7.4 μm2. The spatial resolution of PIV is defined by the interrogation window (i.e., 64 × 64 pixels), which corresponds to 11.84 μm × 11.84 μm2 (Lv et al., 202025. Koullapis, P. G., Stylianou, F. S., Sznitman, J., Olsson, B., and Kassinos, S. C, “Towards whole-lung simulations of aerosol deposition: A model of the deep lung,” J. Aerosol Sci. 144, 105541 (2020). https://doi.org/10.1016/j.jaerosci.2020.105541; Meinhart et al., 199926. Lv, H., Dong, J., Qiu, Y., Yang, Y., and Zhu, Y., “Microflow in a rhythmically expanding alveolar chip with dynamic similarity,” Lab Chip 20, 2394–2402 (2020). https://doi.org/10.1039/C9LC01273G; and Santiago et al., 199829. Oakes, J. M., Hofemeier, P., Vignon-Clementel, I. E., and Sznitman, J., “Aerosols in healthy and emphysematous in silico pulmonary acinar rat models,” J. Biomech. 49, 2213–2220 (2016). https://doi.org/10.1016/j.jbiomech.2015.11.026). It should be noted that the value of time interval δt in the timing settings of the PIV software needs to be adjusted according to the velocity of the fluorescent fluid movement. Considering that there are significant differences in the magnitude of the flow velocity in the alveolar ducts of different generations, the corresponding values of δt are different, which in any case need to satisfy the criteria reported by Boillot and Prasad (Boillot3. Boillot, A. and Prasad, A. K., “Optimization procedure for pulse separation in cross-correlation PIV,” Exp. Fluids 21, 87–93 (1996). https://doi.org/10.1007/BF00193911 and Prasad, 19963. Boillot, A. and Prasad, A. K., “Optimization procedure for pulse separation in cross-correlation PIV,” Exp. Fluids 21, 87–93 (1996). https://doi.org/10.1007/BF00193911). Based on the summary of previous studies, the flow with typical characteristics at T/4 and 3T/4 moments was focused on in this study, and in these two moments, we can capture the saddle point phenomenon and study the flow details in it. The camera takes pictures of the two peak moments of respiratory flow, T/4 and 3T/4, and obtains a set of two frames for each capture. The flow field was calculated from each pair of images captured by the intercorrelation algorithm of the software package Insight 4G, and the measurements were repeated for each alveolus for about 70 cycles. The overall average velocity field was obtained by combining dozens of data sets, and, finally, the velocity distribution contours and streamlines were processed using Tecplot software. It is worthy to note that although both geometric and dynamic similarities were matched in the present experimental setting, it is still a two-dimensional model with a uniform thickness, not a three-dimensional one as the real human alveoli. This limitation may make us miss some details of complex and new 3D alveolar flow patterns.