Enhancing large particle recovery in high-throughput functional cell sorting through ΔBOP optimization Authors: Naohisa Sakamoto, Eikichi Shibata, Mitsuo Yoshimura, Haruto Namura, Yudai Yanashita, Shin Masuhara, Yusuke Uehara, Kenji Yamane, Hiroto Kasai, Motohiro Furuki, Yoshiharu Hayashi Published: 2025 ABSTRACT AbstractFlow cytometry is a critical technology for single-cell analysis; however, sorting large particles (> 30 μm) remains challenging owing to low recovery rates. This study investigated the relationship between particle size and the Break-off point (BOP) and assessed the impact of large particles on droplet formation through a combination of simulations and experiments. The results revealed that interference between the particle-induced BOP and vibration-driven BOP leads to droplet instability—identified as the key factor limiting recovery. We introduce ΔBOP as the difference between these two BOPs and propose an optimization strategy to stabilize droplet formation. Implementing this ΔBOP-based approach increased recovery from 72.2% (794/1100) to 92.4% (1016/1100) for 35-µm particles and from 23.0% (115/500) to 75.6% (378/500) for 50-µm particles. This strategy demonstrates broad applicability for high-throughput single-cell functional analyses involving various particle types, including hydrogel particles, double emulsions, and agarose beads, and opens new possibilities for applications such as secretion profiling and intercellular communication research.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-32698-0.Keywords: ΔBOP (Delta break-off point), Flow cytometry, Large-particle sorting, Functional cell analysis, Recovery rateSubject terms: Biological techniques, Biotechnology, Cell biology, Engineering Introduction Single-cell analysis has emerged as a transformative approach in biological research, offering critical insights into cellular heterogeneity and function. In recent years, the integration of next-generation sequencing (NGS) with droplet microfluidics has expanded the scope of single-cell analysis, enabling the profiling of comprehensive single-cell RNA (scRNA) and spatially resolved single-cell analysis². These advancements have allowed for detailed exploration of both spatial and temporal dimensions of cellular behavior, thereby broadening the horizons of cellular research. As the field continues to evolve, the demand for functional studies focused on intercellular communication, particularly those investigating cell secretion, is growing to highlight the increasing importance of single-cell analysis in understanding cellular mechanisms and interactions. Cellular functional analysis, encompassing cell secretion profiling and intercellular communication studies, plays a pivotal role not only in basic research but also in uncovering disease mechanisms, developing therapeutic strategies, and designing antibody-based drugs1–3. Cytokines, secreted proteins that mediate immune responses and other physiological processes, are key targets of investigation. Understanding cytokine dynamics is essential for elucidating disease mechanisms. For instance, the therapeutic efficacy of CAR-T cell therapy is closely linked to cytokine regulation, which is critical for accurately evaluating clinical outcomes. Similarly, technologies that enable screening of antibodies secreted by single B cells are vital for cancer therapy and for rapid development of neutralizing antibodies and drugs during pandemics4. Despite significant progress, current methodologies for cellular functional analysis face several critical challenges. The primary limitation lies in the difficulty of performing single-cell analysis while maintaining cell viability throughout the isolation, sorting, and analysis processes5. Although fixed-cell analysis is often more straightforward, analyzing live cells offers substantial advantages, including studies conducted under near-physiological conditions and for subsequent cell culture. ELISpot is a notable method for single-cell secretion analysis; however, the method does not allow for cell recovery in a viable state and permits the detection of only a limited number of markers. More advanced platforms, such as the single-cell barcoding chip, enable multiplexed analysis6, but similar to ELISpot, such methods cannot preserve cell viability7. ELISA permits live-cell analysis but suffers from limited sensitivity, hindering its use at the single-cell level. Flow cytometry can be used to measure cytokines and other factors, yet it typically involves intracellular staining of fixed cells, precluding live-cell analysis. A second major challenge is the insufficient throughput of current analytical methods relative to experimental demands. For example, B-cell screening typically involves analyzing 1–100 million candidate cells, necessitating rapid, high-throughput techniques8. The Optofluidic Nanopen assay supports viable cell recovery and real-time monitoring during cultivation, representing an advanced analytical platform. However, a low throughput poses major limitations, with only 20,000 chambers per chip9–11. Droplet microfluidics enables the capture and analysis of single cells within droplets12 but operates at only 200–600 Hz, compared to the maximum of 70,000 Hz achievable with conventional flow cytometry9–11. Previous studies made significant strides in addressing the challenges associated with live-cell functional analysis13. By leveraging the high-throughput sorting capabilities of conventional flow cytometry, researchers have developed methods for isolating viable cells. Although flow cytometry is a powerful technique, it is traditionally limited to detecting fluorescent molecules directly bound to the cell surface and cannot detect secreted substances that have diffused away from the cells. To overcome this limitation, de Rutte et al.14developed a hydrogel-based particle system that allows the simultaneous detection of cells and their secretions. This innovation has facilitated high-throughput single-cell secretion analysis and enabled detailed studies of intercellular communication using conventional flow cytometry. A persistent challenge associated with this approach is the size limitation of sortable events in flow cytometry. Most commercial flow cytometers feature orifice systems limited to 130 μm, constraining the maximum sortable particle diameter to approximately 30 μm15. Furthermore, de Rutte et al.14 optimized the sort delay in flow cytometry, achieving recovery rates of 40–60% for large particles. Comparatively, sorting standard cells such as lymphocytes typically yields above 90%. Furthermore, manually adjusting the sort delay is time-consuming, posing a practical challenge16. Although microfluidics-based approaches can achieve recovery rates of approximately 90%, they remain constrained by the throughput limitations9–11. Additional challenges include channel clogging, limited dynamic range, and particle aggregation14. The approach proposed by de Rutte et al.13 effectively addresses the challenges of live-cell analysis and high throughput. However, further improvements in recovery rates are required to enhance overall system efficiency. The approach is applicable not only to hydrogels but also to double emulsions, agarose gel beads, and other encapsulation systems, offering significant potential to advance cellular functional analysis17,18. Therefore, investigating and resolving the causes of reduced recovery rates observed during the sorting of large particles are essential. While manual adjustment of sort delay can improve recovery, it falls short of 90–95% recovery rates typically achieved with lymphocytes. Thus, identifying and addressing the fundamental mechanisms underlying these limitations is critical for optimizing performance. In this study, we combine computational simulations and experimental analyses to identify and mitigate the primary mechanism limiting large-particle recovery. Using computational fluid dynamics (CFD) and bench experiments, we identify a phenomenon we term particle-induced Break-off point (pBOP) as the main contributor to instability and low recovery rates. We introduce a novel optimization parameter, ΔBOP—the difference between the vibration-driven BOP (vBOP) and the pBOP—and demonstrate that increasing this offset stabilizes droplet formation. This stabilization significantly improves recovery efficiency. Our approach provides a robust and scalable solution for enhancing high-throughput flow cytometry in live-cell applications, with broad implications for functional single-cell analysis. Results Recovery rate and sort delay analysis Figure 1a illustrates the sorting mechanism used in this study, modeled with SH800/MA900 cell sorters (Sony). Sorting was based on the relative positions of target and non-target events. A piezoelectric actuator generated droplets by maintaining a stable Break-off point (BOP), and when a target was detected by the laser, the sort delay—defined as the number of droplets between detection and the target-containing droplet—was calculated. This droplet was then electrically charged and deflected via sorting plates. A droplet-monitoring camera continuously tracked the BOP throughout sorting. Principles of flow cytometry sorting and effects of particle size on sort delay. (a) Flow cytometry sorting relies on the precise spatial alignment of target cells, achieved through piezoelectric actuation to maintain a stable Break-off point (BOP). Although BOP timing can vary under different conditions, proper calibration of the sort delay ensures accurate isolation of target events. (b) For standard-sized particles (20 µm), an optimal sort delay of 18 results in 100% recovery. As particle size increases, the optimal delay shifts: 16 for 40 µm and 14 for 50-µm particles. Interestingly, 30 µm particles retain the same delay as 20 µm particles (18) but exhibit a significantly lower recovery rate (59%). Recovery rates for particles of different sizes were evaluated using a 3-drop charge mode (Fig. 1b). For 20-µm particles, 100% recovery was achieved at a sort delay of 18. However, recovery for 30-µm particles dropped to 59% at the same delay. For 40-µm and 50-µm particles, recovery rates peaked at 68% and 88%, respectively—but only when sort delay was manually adjusted to 16 and 14. Notably, the optimal sort delay for 30-µm particles remained at 18, yet recovery remained significantly lower, indicating an underlying mechanism beyond delay miscalibration. Particle-induced BOL analysis To elucidate the mechanisms underlying low recovery rates and variable sorting delays observed with particles exceeding 26 μm in diameter19, we performed computational fluid dynamics (CFD) simulations to analyze droplet Break-off behavior under different conditions (Fig. 2). In simulations involving particles, the BOL was defined as the distance from the chip outlet to the position where the first droplet containing a particle detaches from the jet. In contrast, in particle-free conditions—especially without applied vibration—the BOL could not be uniquely defined due to Rayleigh–Plateau instability, which introduces substantial uncertainty in droplet formation. It is important to note that several studies have established relationships between the BOL and the Rayleigh–Plateau instability, and have offered analytical formulations to predict the breakup length theoretically20–24.However, these theoretical relationships predominantly apply to the linear regime. Both classical and recent studies21–24,33 demonstrate that as jet velocity increases and nonlinear interactions intensify, the BOL becomes increasingly sensitive to initial conditions and ambient noise, rendering it inherently stochastic24. Consequently, the BOL values in the present study, obtained under particle-free conditions, are represented with error bars (n = 10) to reflect variability, as shown in Fig. 2b. Detailed BOL values are listed in Extended data table 1. Particle Size systematically shifts BOL: 20-µm particles extend BOL by minimal jet disturbance, while 50-µm particles reduce BOL by 15–20% through early jet fragmentation. (a) Flow simulation demonstrates size-dependent jet disruption patterns with and without vibration. (b) BOL distribution boxplots under particle-free conditions with and without vibration, showing median values and interquartile ranges. (c) Experimental validation shows that BOP shift correlates with particle size (20, 35, and 50 μm); with 35-µm particles matching standard sorting BOP. * A true OFF state could not be achieved due to jet instability; thus, a minimal actuation state was used to approximate OFF conditions. (d) Microscopic validation of sorted 35-µm Nanovials. (e) Mechanistic model: particle size determines BOP timing even without piezoelectric actuation; representative sorted Nanovial images included. Two key observations emerged from Fig. 2a and b. First, in the absence of particles, BOL values exhibited substantial variation, with deviations reaching up to 3 mm. Additionally, BOL decreased as particle diameter increased. Second, the application of periodic vibration, as described in Eq. (2), reduced BOL variation under particle-free conditions and stabilized BOL positions for 35 and 50 μm particles. To further investigate these effects, we compared the flow fields across various scenarios. Representative results from all simulation conditions—including particle sizes of 20, 35, and 50-µm without vibration, and particle-free flows with and without vibration—are shown in Fig. 2a. The movie files of each case are listed in Extended data table 2, and the associated data are available online. We examined droplet formation under two conditions: first, vibration-free cases, where the flow field is affected only by particle size, are examined. In the 20-µm case, the particle had minimal impact on the surrounding flow, which closely resembled the particle-free flow field. In contrast, the 35-µm particle induced stronger perturbations in the jet, causing it to break into larger fragments and form more rounded droplets. This disturbance led to an earlier BOL compared to the 20-µm case. For 50-µm particles, breakup occurred even earlier than in the 35-µm case, with the jet breaking downstream of the particle and producing a spherical droplet that fully encapsulated the particle. These results suggest that while both surface tension-induced and particle-induced waves interact in the 35-µm case, the particle-induced effect dominates in the 50-µm case. All of these behaviors were observed under non-vibrating conditions, where predicting BOL precisely proved difficult, emphasizing the need for external control mechanisms. Next, simulation results under vibrating conditions for different particle sizes are also shown in Fig. 2a. Vibration clearly defined the wave structure around the particle and promoted earlier and more consistent jet break-off. The resulting droplets were closer to spherical in shape, indicating more stable droplet formation. These simulations demonstrate that both vibration and particle size jointly influence jet behavior, providing practical guidance for optimizing droplet formation through vibration control. To validate the simulations, we conducted experimental observations using particles of various sizes under minimal actuation conditions, maintaining sufficient flow stability for imaging (Fig. 2c). The experimental results aligned well with the simulations: BOL systematically shifted with increasing particle size, with larger particles producing earlier jet break-off. Notably, for 35-µm particles, the experimentally observed BOL closely matched that seen during standard piezoelectric sorting conditions. Microscopic images of sorted 35-µm Nanovials are shown in Fig. 2d. A schematic summary of the simulation (Fig. 2a) and experimental (Fig. 2c) findings is provided in Fig. 2e, illustrating that even without vibration, BOP timing shifts proportionally with particle size. Hypothesis and mechanism identification Based on these findings, we proposed a hypothesis linking particle-induced BOP (pBOP) to recovery loss (Fig. 3a). For small particles (≤ 20 μm), pBOP formed well below the standard vBOP used in sorting, minimizing interference and ensuring high recovery. However, for 35-µm particles, pBOP and vBOP overlapped, disrupting droplet formation via capillary wave interference. For 50-µm particles, pBOP occurred slightly above vBOP, potentially introducing delay misalignments and further reducing recovery. Particle-induced BOP Interference and stability analysis. (a) A schematic shows the interaction between vBOP and pBOP. For 20 μm particles, pBOP is longer than vBOP, resulting in a large ΔBOP, stable droplets, and high recovery rates. With 35 μm particles, pBOP closely overlaps with vBOP, leading to wave interference and droplet instability. For 50 μm particles, pBOP is shorter than BOP, causing misaligned sort timing and reduced recovery. (b) The effect of particle size on BOP stability is shown, where 20 μm beads exhibit stable BOPs, 35 μm Nanovials show moderate vertical fluctuations, and 50 μm Nanovials display significant instability. To test this, we used a droplet-monitoring camera with a 130-µm nozzle (the maximum commercial standard, limiting sortable diameter to ~ 26 μm¹⁷). As shown in Fig. 3b, BOP remained stable for no-particle and 20-µm conditions but became unstable for 35- and 50-µm Nanovials, exhibiting erratic vertical motion. Figures 4 and 5 show that high ΔBOP (the difference between vBOP and pBOP) correlated with high recovery. Specifically, 20-µm particles exhibited the largest ΔBOP and highest recovery, while 35- and 50-µm particles showed reduced ΔBOP and recovery. This supported our hypothesis that small ΔBOP destabilizes droplet formation, decreasing sorting efficiency. ΔBOP optimization improves 35-µm particle recovery from 72% to 92%: increasing ΔBOP > 0.4 reduces the distance from the laser interrogation point variance by 58% (CV: 0.04→0.02) and establishes a recovery threshold effect. (a) Strategy: vBOP shortening increases ΔBOP separation to minimize vBOP-pBOP interference. (b) Quantitative comparison: Optimized ΔBOP shifts mean BOP from 12.90 (SD=0.50) to 10.68 (SD = 0.21), improving recovery 72.2%→92.4%. (c–d) Multi-device validation (48 experiments, 4 devices) reveals ΔBOP threshold at 0.4, below which recovery remains <80% and above which recovery consistently exceeds 80% (frequently >90%). (e) Purity validation: Purity was evaluated at several ΔBOP values, showing that it remained above 85% regardless of the ΔBOP. ΔBOP optimization improves 50-µm particle recovery from 23% to 76%: Lowering vBOP to create positive ΔBOP reduces CV from 0.03 to 0.02, though the improvement was less pronounced than for 35-µm particles. (a) Strategy: vBOP lowering compensates for elevated pBOP timing in large particles, increasing ΔBOP magnitude. (b) Quantitative outcome: Optimized settings shift mean BOP from 10.71 (SD=0.30) to10.88 (SD=0.21) with recovery improving from 23.0% to 75.6%. (c) Multi-device recovery validation (n=13 experiments, 2 devices): Positive ΔBOP-recovery relationship confirms larger ΔBOP enhances stability, although efficiency gains are attenuated compared to 35-µm optimization. (d) Purity validation: Purity was evaluated at several ΔBOP values, showing that it remained above 85% regardless of the ΔBOP. ΔBOP optimization for 35-µm particles To resolve droplet instability for 35-µm particles, we implemented a ΔBOP optimization strategy (Fig. 4a). By shortening vBOP, we increased ΔBOP, thereby reducing interference between pBOP and vBOP. As shown in Fig. 4b, under standard conditions, the distance from the laser interrogation point was unstable (upper panel). With the optimized setting, that distance became consistent and sharply defined (lower panel). Quantitative analysis of 25 images revealed substantial improvement: the mean distance from the laser point shifted from 12.90 mm (SD = 0.50, CV = 0.04) to 10.68 mm (SD = 0.21, CV = 0.02), a statistically significant change (p < 0.001, paired t-test). Across 11 independent trials (n = 100 each), recovery improved from 72.2% to 92.4%. Correlation between ΔBOP and recovery rate To assess generalizability, we analyzed the relationship between ΔBOP and recovery across 48 experimental points from four devices using 30-µm beads and 35-µm Nanovials (Fig. 4c and d). When ΔBOP was < 0.4, recovery consistently fell below 80%. When ΔBOP exceeded 0.4, recovery surpassed 90% in most cases. A strong positive correlation was observed (r = 0.78, p < 0.001), confirming ΔBOP as a robust predictive parameter for recovery performance. Statistical analysis confirmed these relationship (30-µm: r = 0.97. p = 1.45 × 10− 11, 35-µm: r = 0.94, p = 1.26 × 10− 15), indicating that increasing ΔBOP systematically improved the stability of droplet formation and sorting recovery rate. To account for potential variation between devices, we initially fitted linear mixed-effects models with devices as a random intercept (Y ~ X + (1 | device)) (Table 1)). Table 1Results of the OLS fixed-effects model.ParameterEstimateSDp-value(two-sided t-test)95% CI30-µmIntercept55.07212.0643.68 × 10^−11[48.603 61.540]30-µmX (ΔBOP)63.86410.6896.48 × 10^−9[52.771, 74.958]35-µmIntercept55.6709.7447.14 × 10^−25[52.152, 59.187]35-µmX (ΔBOP)60.35722.4781.26 × 10^−15[52.771, 68.473] Results of the OLS fixed-effects model. The intraclass correlation coefficient (ICC) was calculated as the ratio of between-device variance to the total variance. For Fig. 4c, the ICC was 0.001068, and for Fig. 4d was 0.000159. In both cases, ICC ≤ 0.01 indicated negligible between-device variability; therefore, we adopt the OLS fixed-effects model for each dataset. In Fig. 4c, the association between ΔBOP and recovery rate was significant (β = 63.86, SD = 20.69, 95% CI [52.77, 74.96], p = 6.48 × 10^−9, n = 16; MA#1 = 10, SH#2 = 6). In Fig. 4d, the association was also significant (β = 60.36, SD = 22.48, 95% CI [52.24, 68.47], p = 1.26 × 10^−15, n = 32; MA#1 = 10, MA#2 = 6, SH#1 = 10, SH#2 = 6). These findings suggest a robust positive relationship between ΔBOP and recovery rate, independent of device. The source code for the random effects model statistical software can be downloaded from https://github.com/sony/FCM_Large_Particles. ΔBOP optimization for 50-µm Nanovials We extended our strategy to 50-µm Nanovials (Fig. 5a). Although the pBOP was already farther from vBOP than in the 35-µm case, it still caused instability due to insufficient ΔBOP. By further lowering vBOP, we increased the temporal offset and stabilized the BOP (Fig. 5b). Quantitative image analysis (n = 25) showed a mean of distance from the laser interrogation point change from 10.71 mm (SD = 0.30, CV = 0.03) to 10.88 mm (SD = 0.21, CV = 0.02) following sort delay adjustment, indicating significantly improved droplet formation stability. Across five 100-event trials, recovery increased from 23.0% to 75.6% (P = 0.00005, two-tailed test). Figure 5c shows pooled results from 13 experiments on two devices, confirming a consistent positive relationship between ΔBOP and recovery rate. Statistical analysis confirmed this relationship (50-µm: r = 0.76, p = 2.50 × 10− 3), indicating that increasing ΔBOP systematically improved the stability of droplet formation and sorting recovery rate. Validation using spheroid cells To validate biological applicability, we tested the ΔBOP strategy using HEP G2-derived spheroids—large, fragile cellular aggregates not compatible with conventional flow cytometers²¹. Spheroids were filtered to ~ 40 μm using a 40-µm strainer, with final selection based on FSC-W gating (Fig. 6a and b). Correlation Between ΔBOP and HepG2 Spheroid Recovery. (a) Spheroid Preparation and Sorting: HepG2 spheroids were cultured and filtered through a 40 μm cell strainer. Target spheroids (FSC-W 600–800; Calcein-AM area > 10) were sorted using 3-drop charge mode. (b) Post-Sorting Cell Imaging: Representative images of sorted spheroids. (c) Spheroid Size Distribution: Microscopic analysis of 84 sorted spheroids revealed diameters ranging from 14.38 μm to 44.27 μm. (d) ΔBOP vs. Recovery Rate: Recovery rates increased with larger ΔBOP values, reaching a peak of 82% at ΔBOP = 0.75, confirming that ΔBOP optimization enhances sort performance for biological particles. Microscopy of 30 sorted events confirmed spheroid identity and viability, with diameters ranging from 14.38 to 32.22 μm (mean = 28.36 μm, Fig. 6c). As ΔBOP increased, recovery improved, reaching 85% from 66%at a ΔBOP of 0.75 (Fig. 6d) (P = 0.016, two-tailed test), validating the method in a physiologically relevant context. Discussion Analysis of cellular function—including profiling secreted factors and investigating intercellular interactions—is fundamental to both basic research and translational medicine. Such analyses yield critical insights into disease mechanisms, guide therapeutic development, and facilitate antibody drug discovery. However, conventional methods often struggle to achieve single-cell resolution while maintaining cell viability and high throughput. Previous efforts to overcome these challenges employed flow cytometry combined with hydrogel particles, yet sorting large particles still suffers from low recovery rates requiring further improvement. In this study, we systematically explored the causes of reduced recovery rates in large particle sorting and developed an effective optimization strategy. Integrating computational simulations with experimental validation, we identified variations in particle-induced Break-off points (pBOP) as the primary factor driving recovery loss. We demonstrated that particles generate their own BOPs independently of piezoelectric actuation, with pBOP timing shifting according to particle size, density, and material properties. Crucially, interference between vibration-induced BOP (vBOP) and pBOP caused droplet instability, manifested as random vertical fluctuations in BOP timing, temporal misalignment during sorting, and decreased recovery. By optimizing the time lag between vBOP and pBOP (ΔBOP), we successfully minimized this interference and stabilized droplet formation. To improve recovery rates, ΔBOP must be increased with corresponding sort delay re-optimization. This becomes critical for ~ 50 μm particles, where pBOP dominates; using vBOP timing determined for 10 μm particles would result in mistimed charging. Therefore, it is important to expand the ΔBOP and re-optimize the sort delay simultaneously, particularly when the particle size is large (Extended data fig. 4). Furthermore, for samples with large particle size and a large variation in particle size distribution, the pBOP varies significantly, and this proposed method might not be sufficient. If the particle size is non-uniform, it is necessary to align the particle size distribution sufficiently in advance by filtering or similar pre-processing methods. There is no direct correlation between the abort rate and recovery. The abort rate is determined by the Clock (the number of droplets generated per second), the number of guard droplets, and the Events Per Second (EPS). An abort occurs when a particle enters the guard range; however, whether a particle falls within this range is determined probabilistically. Under all sample conditions (both standard and optimized), the Clock frequency was fixed at 12 kHz, the guard range at 3 drops, and the EPS at less than 300. Therefore, the abort rate remained constant. The theoretical relationship between EPS and sorting efficiency (100% - Abort Rate = Efficiency) is shown in Extended data Fig. 1. While the polarity of ΔBOP may influence BOP stability, experimental verification was not feasible in the present study. Reversing the polarity (making vBOP longer than pBOP for 35-µm particles) would shift BOP outside the optimal range required for maintaining high sort performance, as established in droplet formation theory20,26,27. Similarly, for 50-µm particles, the pBOP-induced breakoff already approached the boundary of the acceptable range, limiting the parameter space for achieving appropriate vBOP adjustment. Therefore, investigating polarity effects while preserving sort performance remains a subject for future investigation. The recovery improvements achieved through ΔBOP optimization significantly enhance the practicality of flow cytometry for cellular function analysis. This approach enables efficient sorting of viable cells encapsulated within microcarriers or double emulsions, facilitating functional assays such as antibody discovery from B cells and cytokine secretion profiling. Furthermore, the ability to culture sorted cells post-sorting expands the method’s applicability to longitudinal studies of cellular behavior and function. Combining the high throughput of conventional flow cytometry (up to 70,000 Hz) with our improved recovery rates addresses a critical need for applications such as B cell screening, which typically involves analyzing millions of candidate cells28. Previous approaches using microfluidics are limited to 200–600 Hz9–11, making comprehensive screening impractical. Microfluidic approaches continue to make rapid advances; while excellent performances have been reported in this regard29, they require novel devices and specialized microfluidic systems. In contrast, if automated large particle-sorting using ΔBOP optimization is implemented, the proposed approach can be applied to widely used commercial flow cytometers, offering superior advantages in terms of cost, operability, and reliability. Importantly, this method preserves the inherent high-throughput advantage of flow cytometry while achieving recovery rates comparable to microfluidic platforms. The optimization principles established in this study can be extended to particles with diverse physical properties, potentially broadening the range of carrier systems for effective sorting using flow cytometry. The versatility of this method enables researchers to select optimal carriers for specific applications without compromising recovery efficiency. While our study demonstrates significant improvements in sorting large particles up to 50 μm in diameter, several limitations should be acknowledged. First, testing has been limited to particles not exceeding 50 μm, and challenges remain for larger particles including channel clogging, reduced throughput, and increased abort rates. Second, although larger nozzles (> 130 μm) can potentially accommodate larger particles, these typically operate at lower frequencies, substantially reducing overall throughput30,31. Third, we have not yet validated compatibility with adherent-cell carriers, which may present unique fluid dynamics challenges. Alternative approaches, such as multi-drop sorting, may address some of these challenges but they introduce new complications, including increased abort rates or potential compromises in sample purity. This study addresses a critical bottleneck in flow cytometry sorting of large particles by elucidating the mechanism behind recovery loss and presenting a robust optimization strategy. By characterizing the interplay between pBOP and vBOP timing and establishing ΔBOP as a key parameter, we markedly improve recovery rates to levels comparable with standard lymphocyte sorting. Implementing automated large particle sorting with ΔBOP optimization stands to expand the utility of commercial flow cytometers for high-throughput applications such as antibody discovery from B cells. These advancements enhance the practical scope of flow cytometry for cellular function studies, balancing throughput and cell viability while offering significant cost and time efficiencies over conventional methods. Continued development of analytical devices incorporating these principles promises to accelerate progress across diverse biological fields—from immunology to cancer research and regenerative medicine—ultimately enabling new discoveries and therapeutic innovations. Materials and methods Numerical framework We performed CFD simulations to validate the proposed strategy. The seminal work by Lord Rayleigh paved the way for a study of the droplet formation process from a continuous jet32. The breakup of a liquid jet into droplets due to surface tension is referred to as the Rayleigh–Plateau instability. As controlling this instability is difficult, external forces such as periodic vibrations are generally applied to regulate droplet formation20. In addition to surface-tension-driven disturbances, the presence and size of particles within the jet can introduce additional types of perturbations. The simulations in this study sought to investigate different types of disturbances by comparatively analyzing scenarios with particles of various sizes individually flowing in the jet. Prior research has examined the effect of these perturbations through theoretical20,33, experimental34, and computational approaches22. In this study, we integrated experimental and numerical simulations to comprehensively investigate this phenomenon. Using Star-CCM + 2310 (Siemens PLM Software Inc.), we performed coupled simulations of droplet formation and solid particle motion. Water and air served as the working fluids for the liquid and gas phases, respectively, with physical properties summarized in Table 2. To capture the liquid–gas interface, we employed the volume of fluid (VOF) method, wherein the local volume fraction in each mesh cell is denoted by γ (γ = 1 for liquid phase and γ = 0 for gas phase). A single particle was introduced into the fluid domain, and its motion within the jet was tracked before and after passing the Break-off length (BOL). Particle dynamics were simulated using an overset mesh combined with a dynamic fluid–body interaction (DFBI) method. Prior to the main simulations, a mesh sensitivity analysis was conducted to ensure mesh independence; the mesh size was refined until BOL converged. The final mesh employed a minimum cell size of 3.5 μm, and the temporal resolution was set to 2.2 × 10⁻⁸ s. Table 2Physical properties of water and air at 25 °C.Physical propertiesWaterAirDensity [kg/m3]9971.18Viscosity coefficient [Pa・s]0.89 × 10− 31.85 × 10− 5Surface tension [N/m]0.072- Physical properties of water and air at 25 °C. Simulation setup The simulation domain, illustrated in Fig. 7, corresponds to the nozzle section of the microfluidic chip used in the numerical analysis in this study. To reduce computational cost, only the nozzle region (highlighted by the red circle in Fig. 7) was modeled under the assumption of axisymmetry. Although the actual chip has a rectangular cross-section, the simulation used a circular cross-section with an equivalent area and flow rate to ensure consistency with the experimental system. (In Method section) Geometry of the chip tip. Computational domain is highlighted in red. A parabolic flow profile was applied at the inlet, expressed as Here, D is the outlet diameter (1.4 × 10⁻⁴ m), Dc ​is the inlet diameter (1.86 × 10⁻⁴ m), r is the radial position within the channel, and U0 represents the mean flow velocity derived from experimental flow rates (5.936 m/s). A Dirichlet boundary condition (zero pressure) is applied at the outlet. Moreover, a cylinder representing the ambient gas is placed outside the tip outlet, with a diameter six times and a length 200 times that of the outlet diameter, respectively. Simulation scenarios To evaluate the influence of vibration-induced perturbations and particle size on droplet formation, we conducted multiple simulation scenarios. First, a baseline two-phase flow simulation without particles was run until a stable droplet formation pattern was established. A single particle was then introduced into the established flow field (as shown in Fig. 8), and the simulation was restarted. The simulation continued until droplet pinch-off occurred, at which point the breakup timing was analyzed. (In Method section) Initial particle position in a flow (the red region indicates liquid, the blue region indicates air, and the arrow points to the particle). The simulated particle properties—density (1.05 g/cm³) and diameter—matched those used in experiments, as deviations in these parameters produced results inconsistent with observed behavior. The following two primary conditions were studied: With jet vibration. With jet vibration. Without jet vibration. Without jet vibration. Under each condition, particles of diameters 20 μm, 35 μm and 50 μm were analyzed to evaluate size-dependent effects. Jet vibrations were modeled as sinusoidal velocity fluctuations, as described by Eqs. (2), where A is the amplitude (0.01) and f is the frequency (12 kHz). The resulting velocity perturbation u is calculated using Eq. (1). Equation (2) corresponds to the piezoelectric excitation used in the experiments. The vibration amplitude A was carefully adjusted in advance to match the experimental trends in BOL as closely as possible. Sample Preparation Fluoro-Max Green fluorescent beads (30, 40, and 50 μm; Thermo Scientific) were prepared by suspending separately weighed quantities in phosphate-buffered saline (PBS, pH 7.4) containing 0.1% [w/v] Pluronic F-68 to prevent aggregation. Suspensions were sonicated for 5 min to ensure uniform dispersion. Final bead concentrations were adjusted to: 2 × 10⁶ beads/mL for 30- and 40-µm beads. 2 × 10⁶ beads/mL for 30- and 40-µm beads. 1 × 10⁶ beads/mL for 50-µm beads. 1 × 10⁶ beads/mL for 50-µm beads. Microcarriers NV35 and NV50 (35- and 50-µm Nanovials; Partillion Bioscience) were prepared following the manufacturer’s protocol. They were washed three times using the provided buffer via centrifugation at 300 × g for 5 min. Conjugation with PE Streptavidin (SAv-PE, BioLegend) was performed at 5 µg/mL in conjugation buffer for 30 min at room temperature with gentle rotation. Following conjugation, excess SAv-PE was removed via three additional washes. Final concentrations were: 2 × 10⁶ particles/mL for NV35. 2 × 10⁶ particles/mL for NV35. 1 × 10⁶ particles/mL for NV50. 1 × 10⁶ particles/mL for NV50. Both were suspended in sorting buffer (PBS with 1% BSA and 0.1% Pluronic F-68). Both were suspended in sorting buffer (PBS with 1% BSA and 0.1% Pluronic F-68). Control experiments used standard-sized Fluoro-Max Green beads (5, 10, and 20 μm; Thermo Scientific), prepared similarly. Final concentrations were set at 5 × 10⁶ beads/mL. Calculation of ΔBOP used in the flow cytometer The time gap between vBOP and pBOP is defined as ΔBOP. ΔBOP can be computed as: In Eq. (3), ΔBOP represents the time difference between vBOP and pBOP, expressed in milliseconds. Equation (4) defines vBOP as the break-off time generated by the piezoelectric actuator, and pBOP as the time at which particle-induced break-off occurs, both measured in milliseconds. The variable Clock refers to the frequency of voltage applied to the piezoelectric element, expressed in hertz [Hz], representing the number of droplets generated per second. Delay corresponds to the number of voltage cycles from particle detection at the laser interrogation point to the onset of droplet charging. Phase is calculated by dividing a full Delay cycle into 360 segments, expressed in degrees [°]. Sort delay and Phase are determined by the auto calibration optimized for 10 μm beads. In Eq. (5), FlowRate is defined as the volume of liquid [µL min⁻¹] ejected from the sorting chip over one minute while the flow cytometer (FCM) is in the Ready state. The coefficients a and b are used to estimate pBOP as a function of FlowRate and are influenced by parameters such as temperature, sheath pressure, and particle size. These coefficients are empirically determined through the following procedure: FlowRate Measurement: FlowRate is measured by collecting the liquid expelled from the sorting chip over 1 min while the FCM is in the Ready state. The mass of the collected liquid is used to calculate the corresponding volumetric flow rate. FlowRate Measurement: FlowRate is measured by collecting the liquid expelled from the sorting chip over 1 min while the FCM is in the Ready state. The mass of the collected liquid is used to calculate the corresponding volumetric flow rate. pBOP Determination: pBOP is experimentally determined at three different FlowRate conditions by varying the sheath pressure. The pBOP is computed using Eq. (6). For each Delay setting, images are captured to observe the particle break-off behavior. The proportion of images showing particle break-off is calculated for each Delay. The BOP Delay is defined as the Delay at which the break-off occurrence reaches 50%. Dividing the pBOPDelay by the Clock frequency yields the pBOP for the corresponding FlowRate. The relationship between BOP and Delay is illustrated in Extended data fig. 2. pBOP Determination: pBOP is experimentally determined at three different FlowRate conditions by varying the sheath pressure. The pBOP is computed using Eq. (6). For each Delay setting, images are captured to observe the particle break-off behavior. The proportion of images showing particle break-off is calculated for each Delay. The BOP Delay is defined as the Delay at which the break-off occurrence reaches 50%. Dividing the pBOPDelay by the Clock frequency yields the pBOP for the corresponding FlowRate. The relationship between BOP and Delay is illustrated in Extended data fig. 2. Linear Regression Analysis: Flow rate and pBOP timing were plotted on the horizontal and vertical axes, respectively. A linear regression was applied to derive the coefficients a and b. The correlation between flow rate and pBOP is presented in Extended data fig. 3. Linear Regression Analysis: Flow rate and pBOP timing were plotted on the horizontal and vertical axes, respectively. A linear regression was applied to derive the coefficients a and b. The correlation between flow rate and pBOP is presented in Extended data fig. 3. Definition of evaluation metrics Recovery Rate: The sort recovery rate (RR) is the percentage of target particles successfully collected relative to the total number of sorting attempts. The calculation method for R is provided in Eq. (7). Purity: Purity (P) represents the percentage of target cells among all collected cells after sorting. It represents how selectively the sorting system isolates the intended population from contaminants or non-target cells. The calculation method for P is provided in Eq. (8). Efficiency: Efficiency (E) is the number of sorts attempted as a percentage of the targeted cells that passed through the lasers. The calculation method for E is provided in Eq. (9). ​: number of cells in the collection tube that satisfy the sort criteria. : total number of sort attempts made by the instrument. ​: number of contaminating cells in the collection tube that do not satisfy the criteria. : number of cells that should satisfy the sort gate and that pass through the laser(s). Experimental procedure Cell sorting experiments were performed using the Sony SH800S (version 2.1.7) and MA900 (version 3.3) systems, equipped with 130-µm sorting chips (Sony). Prior to each session, instruments were calibrated using Auto Setup Beads (Sony) following the manufacturer’s protocol, which included auto calibration of the sort delay value. These values were recorded for reference. To evaluate recovery rates, fluorescent beads of 30, 40, and 50 μm in diameter were sorted under varying conditions. The sort delay was systematically adjusted around the automatically determined value (± 5 units) to identify optimal settings for each particle size. For each delay value, 25 beads were sorted in single three-drop mode onto a glass slide. Each condition was repeated four times to ensure reproducibility. Sorted beads were counted using a fluorescence microscope (BX53, Olympus) equipped with a GFP filter set. Recovery rate was calculated as the percentage of beads successfully recovered relative to the number of sort attempts25. The mean recovery rate and standard deviation were computed from the four replicates per condition. Purity was assessed by tube sorting followed by flow cytometric reanalysis of the sorted fraction. A mixed sample was prepared containing 90% unstained and 10% PE-stained Nanovials. Two gates were established to separate unstained and PE-positive populations. Sorting was conducted by targeting the PE-positive (10%) gate, and the collected tube was subsequently remeasured by flow cytometry. Purity was calculated as the percentage of reanalyzed events that remained within the PE-positive gate. The specific adjustment procedures are described below: Common settings: Frequency under 12 kHz, temperature 25 °C (room temperature), sorting mode: Single 3-drop. For + ΔBOP (vBOP is shorter than pBOP): The sheath pressure was adjusted and set between 50% and 100% of the standard conditions, and the sort delay was determined by autocalibration. This procedure afforded a shorter vBOP than that under the standard condition, thereby expanding + ΔBOP. For –ΔBOP (vBOP is longer than pBOP): After auto calibration under standard conditions, the PZT drive amplitude was manually adjusted to < 50% of the standard value to minimize the influence of vBOP, ensuring that the number of droplets increased by a factor of at least two. This approach expanded − ΔBOP. Additionally, the sort delay was manually adjusted to match the timing of pBOP. These adjustments allowed precise control over vBOP. Experiments were repeated across five different devices to ensure statistically robust analysis. The relationship between ΔBOP and recovery rate was assessed using Pearson’s correlation coefficient, with statistical significance set at p < 0.05. Droplet image acquisition and analysis High-speed imaging of droplet formation was conducted using the MA900 sorter (version 3.3, Sony) equipped with a synchronized high-speed camera system. The camera recorded at 10,000 frames per second (fps) with an exposure time of 1 µs, enabling clear visualization of droplet dynamics. All experiments were performed after auto calibration with Auto Setup Beads to ensure consistent baseline parameters. Various particle types were introduced into the sorter, and for each, at least 100 consecutive frames were captured to document droplet formation behavior. Image analysis was conducted using custom-developed software that automatically detected the BOP timing in each frame based on pixel intensity gradients. The timing of pBOP was determined by visual inspection of high-speed images for each experimental condition. The pBOP was defined as the point where the continuity of the droplet stream is interrupted. This criterion reflects the requirement for continuous liquid connection during charging, as droplet charging must occur before detachment. For each particle type, 25 randomly selected frames were analyzed to calculate the mean BOP timing, standard deviation (SD), and coefficient of variation (CV). A red line was drawn on the captured images to indicate the distance from the laser interrogation point to where the droplet stream becomes discontinuous. To assess the influence of ΔBOP, sheath pressure, the piezoelectric drive amplitude and frequency were adjusted, altering vBOP while maintaining stable droplet generation. Droplet images were then acquired and analyzed using the same procedure. Comparisons between standard and optimized ΔBOP settings were evaluated using paired t-tests, and the relationship between BOP stability (quantified by SD and CV) and recovery rate was analyzed via linear regression. All statistical tests used a significance threshold of p < 0.05. Cell preparation HepG2 cells were cultured to form spheroids by seeding 1 × 10⁶ cells into a 10-cm dish and incubating for 72 h. On the day of spheroid collection, all reagents (medium, PBS, and trypsin/EDTA) were pre-warmed in a 37 °C water bath. The culture medium was gently aspirated to avoid disturbing the cells. 5–10 mL of PBS was used to rinse the dish, followed by the addition of 1 mL trypsin/EDTA for 4 min to detach the cells. To stop the enzymatic reaction, 5 mL of medium was added, and the suspension was pipetted to achieve a uniform single-cell solution. The cell suspension was transferred to a 15-mL conical tube and centrifuged at 200 × g for 5 min at room temperature. The supernatant was discarded to remove residual trypsin, and the cells were resuspended in 5–10 mL of fresh medium. Cell viability was assessed using an automated cell counter. Spheroid formation and sorting For 3D culture, 1.5 mL of medium was added to each well of a 6-well Elplasia® plate (Corning), ensuring removal of air from the microcavities by pipetting. The plate was then centrifuged at 500 × g for 5 min. Cells were seeded into the wells at a density of 1.5–3 × 10⁴ cells/well, and the medium volume was adjusted to at least 3 mL per well. Cultures were incubated for 3 days to promote spheroid formation. On the day of sorting, Calcein AM was prepared in 6 mL of medium, and 1 mL was added to each well. The plate was incubated for 15–20 min in the dark at room temperature to stain viable cells. Formed spheroids were then collected using a wide-bore pipette tip and transferred to a 5-mL tube. Before sorting, the spheroid suspension was filtered using pluriStrainer® 40 μm (pluriSelect) to ensure the spheroid size was smaller than 40 μm. Flow cytometry experiments were performed in compliance with MIFlowCyt guidelines, and all required criteria are documented in Extended data table 3. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Acknowledgements The authors thank Claudia Krause, Masahiro Matsumoto, and Michael Zordan for their critical review of the manuscript and valuable discussions. Abbreviations Break-off point Break-off length Vibration-driven BOP Particle-induced BOP Time lag between vBOP and pBOP Single-cell RNA Next-generation sequencing Formalin-fixed paraffin-embedded Author contributions NS, ES, YY and HK designed and conducted the primary experiments. MF and YH conceptualized the overall framework of the study. MY carried out the computational simulations, and HN conducted the spheroid-related experiments. SM, KY and YU drafted the manuscript with input from all authors. YH finalized the writing of the manuscript. Data availability The datasets generated or analyzed during the current study are available from the corresponding author upon reasonable request. Declarations Competing interests All authors are employees of Sony corporation. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement The datasets generated or analyzed during the current study are available from the corresponding author upon reasonable request.