High-Reynolds Microfluidic Sorting of Large Yeast Populations

Microfluidic sorting offers a unique ability to isolate large numbers of cells for bulk proteomic or metabolomics studies but is currently limited by low throughput and persistent clogging at low flow rates. Recently we uncovered the physical principles governing the inertial focusing of particles in high-Reynolds numbers. Here, we superimpose high Reynolds inertial focusing on Dean vortices, to rapidly isolate large quantities of young and adult yeast from mixed populations at a rate of 107 cells/min/channel. Using a new algorithm to rapidly quantify budding scars in isolated yeast populations and system-wide proteomic analysis, we demonstrate that protein quality control and expression of established yeast aging markers such as CalM, RPL5, and SAM1 may change after the very first replication events, rather than later in the aging process as previously thought. Our technique enables the large-scale isolation of microorganisms based on minute differences in size (±1.5 μm), a feat unmatched by other technologies.


Results
Single cell size characterization of yeast. Budding yeast divide asymmetrically, producing a smaller daughter cell and a bud scar on the mother (Fig. 1A). Division is asymmetric, resulting in the accumulation of aging factors by the mother cell, while the daughter is rendered pristine 2 . To rapidly characterize yeast populations, we created an algorithm using CellProfiler 16 that uses images of calcofluor white stained yeast (Fig. 1B) to identify single cells and quantify their diameter and scar count (Fig. 1C,D). Following experimental validation of our automated algorithm, we analyzed over 5,000 yeast cells demonstrating a linear increase in yeast diameter starting from 2.3 ± 1 μm for a virgin bud (0 scars) and plateauing around 12.2 ± 2 μm at 11 scars (Fig. 1E). Yeast diameter growth rate was 0.8 ± 0.1 μm/scar, showing significant overlap between age groups. To verify our membrane and scar detection, we analyzed populations of yeast in the exponential growth phase (Fig. 1F), of which 51.5 ± 1.5% of the yeast are newborn. These results confirm and reiterate previous measurements 17 showing that replicative age distribution among the exponential phase population correlates well with yeast generation time, for yeast with 10 scars or less 18 . Analysis of yeast cells grown in high-density culture for several days showed that 32 ± 8% of the overall population was newborn, 67 ± 8% of the overall population had 0 to 2 scars, and 33 ± 8% had 3 or more scars (Fig. 1F). Less than 0.5% of the population had 11 scars or more. The different age distribution in the high-density culture suggests that the generation time homogeneity among adult yeast (<10 scars) is not maintained due to age dependent stress response. Interestingly, our analysis suggests that HSP104 inclusion increased gradually from the first steps of replicative aging (Fig. 1G). CellProfiler code is available upon request. Microfluidic size-dependent sorting in high flow rates. Recently we showed that opposing shear-induced forces push large particles toward the concave side of curved channels in high flow rates ( Fig. 2A). This force increases as a function of particle diameter to the 4 th power and predominantly works on larger particles 10 . Interestingly, a second lift force appears in curved channels due to the Dean vortices induced rotation (Fig. 2B). We note that particles will be pulled to the vortex center, due to higher particle velocity relative to the fluid in the inner side of the spinning particle 19 . Thus, a stable trap dominates at the center of the channel (Fig. 2B, arrow). These forces increase as a function of particle diameter to the 2 nd power 20,21 , and thus predominantly work on smaller particles.
To demonstrate our separation principle, we introduced fluorescent polystyrene beads to microfluidic devices containing repeating curves of 0.9 mm radius of curvature (methods). Bead diameter ranged from 4.1 to 15.5 µm while flow rates ranged from 1500 to 1800 μl/min, corresponding to Reynolds of 215 to 257 (Fig. 2C). We show that small particles are predominantly focused to a single streamline at the center of the channel due to Dean vortices induced lift force, while large particles are forced to the concave edge of the channel due to opposing shear-induced forces. Interestingly, increasing fluid velocity changes the threshold between focusing mechanisms, rejecting larger particles from the Dean vortices toward the center (Fig. 2C).
The 2 nd power difference in the size dependence of both mechanisms allows us to separate not only particles with different sizes, but also closely similar particles with overlapping distributions (Fig. 2D-H). Particles were isolated at 1.5 mL/min flow rate (Re = 215). High purity of small 4.1 μm particles was obtained at convex side of the channel but at low 12 ± 1% yield (Fig. 2D,E). The small 4.1 μm particles were primarily focused by Dean vortices to the channel center with a yield of 55 ± 5%. Purity of 4.1 μm ranged from 79 ± 1% compared to similarly sized 5.56 μm particles, with overlapping distribution (Fig. 2D,F) to 97 ± 1% when separated from larger 9.95 μm particles (Fig. 2E). The large 5.56 and 9.95 μm particles were primarily pushed toward the concave side of the channel by opposing shear-induced forces 10 . Purity of larger particles ranged from 76 ± 9% to 90 ± 5% for 9.95 μm and 5.56 μm particles, respectively, with similar yields of 60 ± 9%. To demonstrate that similar mechanisms may hold for yeast isolation, we captured a streakline image of calcofluor-stained yeast population. As most yeast in a heterogeneous mixture are young, the streakline coincides with 4.1 μm beads (Fig. 2G). Thus, our device should be able to separate adult yeast (3-10 scars) with a diameter of 5.4 to 11 μm, from young cells (0-2 scars) that are 2.3 to 4.7 μm in diameter with high fidelity (Fig. 2D,E). We note that a flow rate of 1.5 mL/min allows us to isolate 10 7 yeast/min from standard suspension for bulk metabolomic and proteomic analysis with minimal clogging due to the high pressure applied.
Microfluidic sorting of yeast. Yeast cells suffer minimal damage when exposed to shear stress of up to 1292 Pa 22 and thus can be separated in high Reynolds flows without damage. Our microfluidic device contains 5 exit ports (Fig. 2H) for simultaneous, parallel collection. Yeast cells were introduced into the device at 1.5 mL/ min, experiencing a shear stress of 691 Pa, significantly below their damage threshold. Samples were taken from the central port that should predominantly contain young yeast (0-2 scars) and from the concave port that should be enriched with adult yeast (3-10 scars). FACS measurement of the sorted yeast populations shows 34% increase in forward scattering, a measurement of particle size (Fig. 3A). This result confirms the small but distinct size difference between the sorted populations. Calcofluor white staining of the sorted populations shows 3-fold higher intensity in concave port cells (Fig. 3B). Since the budding scars are the source of the bright fluorescence ( Fig. 1A,B), the difference in scar number between the populations results in a large difference in calcofluor fluorescence. Single cell analysis showed that young cells (0-2 scars) were enriched from 77 ± 5% in the mixture to 89 ± 7% in the central port ( Fig. 3C), with an average scar number of 1.3 ± 1.6, while adult cells (3-10 scars) were enriched from 23 ± 2% in the mixture to 69 ± 3% in the concave port ( Fig. 3C), with an average scar number of 3.4 ± 1.7, suggesting the concave port isolate was primarily adult yeast that underwent at least one replication cycle.
Age dependent growth rate of the sorted yeast populations. To verify that the subpopulations of yeast cells sorted by microfluidics correlate with familiar age-related properties of yeast 23 , we used single cell analysis to compare vacuolar acidity, whose absence is an early aging marker 24 in our sorted and unsorted cell populations. We show 16.1 ± 2% lower acidity in adult compared to young yeast sorted using our microfluidic device (Fig. 3D), compared to 15.8 ± 3.4% in the unsorted population (p = 0.452, n = 3), showing that shear stress had no effect on traditional age-related properties of yeast. Similarly, while growth rates in the exponential stage were similar for both populations (Fig. 3E), the lag times were different (Fig. 3E,F). The adult yeast subpopulation taken from the concave port showed a lag time of 5.9 hours, which was 8% higher than the original mixture (p = 0.007). However, the young subpopulation showed a lag time of 3.9 hours, which was 27% lower than the original mixture (p = 0.001). This analysis demonstrated that the size-sorted subpopulations have different, age-related, functional properties similar to traditional assays, leading us to run a comparative analysis of protein degradation in the sorted populations.
Protein degradation in sorted yeast populations. Proteasome degradation was shown to progressively decrease with aging in both mammals and yeast 25 , and suggested as a mechanism of age-dependent neurodegenerative diseases 26 . Attempts to quantify age-dependent degradation rate of misfolded proteins have thus far been limited to single-cell analysis of hundreds of cells resulting in high variance, and significant differences identified only for old yeast with 6-7 scars (see Fig. 1G). Taking advantage of the high yield of yeast populations sorted by our device, we quantified the age-dependent degradation rate in a high throughput assay. We expressed Von Hippel-Lindau protein (VHL), a spontaneously misfolded protein in yeast, fused to GFP as a tag protein, under the control of GAL inducible promoter. This allowed us to shut down expression of the VHL by providing glucose instead of galactose in the media, and following the degradation rate of the previously accumulated VHL. We   14 . Stable equilibrium points are indicated by black arrows. Numerically derived velocity profile is superimposed. (B) Illustration of centrifugal forces acting on a dense 4 μm particle exposed to two Dean vortices. Stable equilibrium point indicated by black arrows. (C) Long-exposure images of fluorescent beads focused in curved microfluidic channels with 0.9 mm radius of curvature. Inertia focusing predominates in large particles, pushing beads to the concave side of the channel, while Dean vortices traps small particles in the center of the channel. Particle size threshold for inertial focusing increases with fluid velocity. (D,E) Cytometry-based particle analysis of microfluidic outflow compared to unsorted particle mixture. Particles were isolated at 1.5 mL/min flow rate (Re = 215). High purity of small 4.1 μm particles was obtained at convex side (outlet 1) but at low 12 ± 1% yield. Small 4.1 μm particles were primarily focused by Dean vortices to the channel center (outlet 2) with a yield of 55 ± 5%. Purity of 4.1 μm ranged from 79 ± 1% compared to similarly sized 5.56 μm particles, with overlapping distribution (D) to 97 ± 1% when separated from larger 9.95 μm particles (E). Large  normalized the GFP-VHL intensities to a stable NLS-tRFP that is expressed at identical levels under the control of a GAL promoter. Then we isolated 5 samples of 100,000 to 200,000 cells of each subpopulation and ran a neck-to-neck comparison of degradation using FACS (methods).
Interestingly, we identified significant differences in misfolded protein degradation between the adult (3.4 ± 1.7 scars) and young (1.3 ± 1.6 scars) yeast populations (Fig. 3G). While GFP-VHL degradation rate was not significantly different between young yeast and the original mixture (Fig. 3H), it was 4.5-fold lower in the adult yeast subpopulation (p = 0.009). We note that 87% of the adult yeast population has 5 scars or less, and that the average scar number in this population was 3.4, suggesting that misfolded protein degradation rates are affected already at early stages of replicative aging in yeast. To verify that the lower GFP measured is a result of the VHL degradation and not a result of GFP degradation, we performed western-blot of VHL using anti-GFP antibodies and quantified the degraded percentage of VHL using image processing. Western-blot analysis shows 29 ± 3% difference in VHL degradation between adult and young cells (p = 0.031, n = 3) confirming that the decrease in GFP measured is a result of VHL degradation (Fig. 3I).

Mass spectrometry of sorted yeast populations.
Our microfluidic approach permits the continuous isolation of large quantities of yeast cells for proteomic analysis, allowing us to identify differences in the yeast proteome at the first steps of replicative aging. Yeast cells were separated into young (1.3 ± 1.6 scars), adult (3.4 ± 1.7 scars) and the original mixture, lysed, trypsin-digested and subjected to system-wide proteomic analysis by LCMS. Samples of each group from two independent experiments were analyzed in triplicates. We identified 932 proteins present in all six sample groups (Table S1). According to the Gene Ontology (GO)-based annotation of cellular localization, the majority of our identified proteins are cytoplasmic (389), and associated with mitochondria (255), roughly half of them (112) are localized to the mitochondrial matrix, others to the organelle membranes (132), vacuole (8), stress granules (9), Golgi vesicles (13), and nucleus (21) (Table S2) in agreement with general distribution of yeast proteins 27 .
Label-free quantification (LFQ) showed that the majority of the proteins did not change between the populations, with mean standard deviation less than two-fold (Fig. S1), demonstrating the robustness of our analysis. ANOVA identified 64 proteins that were differentially expressed between all three groups (p < 0.05). As expected, the proteomic profile of the original mixed population was more similar to the young than the adult yeast cells (Fig. S1, Table S3). To elucidate differences between populations we used a stringent criterion (FDR = 0.05, p < 0.05) to identify 95 proteins enriched at least 2-fold in the young and 9 proteins enriched in the adult cells (Fig. 4A,B). Enrichment analysis showed that young cells differentially expressed proteins involved in protein homeostasis, including chaperones and co-chaperones. Other enriched proteins included regulators of mRNA metabolism and translation (Fig. 4A,B, Table S4). Western blot analysis validated our findings, showing SPB1 abundance did not change, while EGD1 and KAP95 were enriched (p = 0.04, n = 3) in the adult and young populations, respectively (Fig. 4C).
Analysis of protein-protein interactions showed clustering of proteins differentially expressed in young cells (Fig. 4D, blue). One such cluster represents mRNA binding and metabolism associated proteins, where DHH1, STO1 and SSD1 are highly interacting hub proteins, with 18, 14 and 9 connections respectively among the enriched proteins in young cells. Both DHH1 and SSD1 are associated with longevity in yeast 28,29 . Similarly, chaperones CCT3 and SBA1 formed a network with upregulated proteins in young cells, suggesting upregulation of protein folding, and its downregulation after the first few budding events in adult cells (3.4 ± 1.7 scars). CCT3 is a subunit of TRiC which is crucial for VHL recognition and degradation 30 , and its downregulation may be a cause of slow VHL degradation in adult cells (Fig. 3G-I).
Annotation and connectivity analysis of upregulated proteins in the adult cells pointed to a small subset of translation regulatory network where translation regulator SBP1 is a highly connected hub protein. Interestingly, Calmodulin, RPL5 and SAM1 showed significant upregulation in adult cells (Fig. 4A,D, red, Table S4). Calmodulin and RPL5 play a significant role in apoptosis and senescence 31,32 and the deletion of SAM1 extends life cycle in yeast 33 . Thus, our microfluidic separation allowed us to detect a distinct proteomic profile of two extremely close populations of young and adult cells, originating from the same culture, and separated by 2-3 replication events.

Discussion
The mechanism that regulates senescence is among the most ancient enigmas. While most cellular and organismal systems are thought to decline late in the aging process 34 , recent efforts attempt to analyze the mechanism of aging by identifying discrete aging determinants and the stage in which they appear. Saccharomyces cerevisiae, the baker's yeast, allows us to dissect the basic mechanisms of aging at high temporal resolution. There are two aspects of aging in yeast: chronological and replicative aging 35 . Chronological aging corresponds to the life span of a non-dividing cell, while replicative aging corresponds to the number of cell divisions that an individual yeast were primarily pushed toward the concave side of the channel (outlet 4) by opposing shear-induced forces. Purity of larger particles ranged from 76 ± 9% to 90 ± 5% for 9.95 μm and 5.56 μm particles, respectively, with similar yields of 60 ± 9%. (F) Long exposure image of 4.1 μm (green) and 5.56 μm (red) particle streaklines. The green particles are focused to the center of the channel, and red particles are primarily pushed toward the concave edge of the channel. (G) Long exposure image of a heterogeneous yeast sample (blue). The streakline's highest intensity location (white histogram) coincides with the small bead streakline (4.1 μm, green arrow) rather than the large bead streakline (5.56 μm, red arrows). cell goes through. Each cell division is asymmetric, producing a daughter cell that is smaller than the mother cell, with the latter accumulating a bud scar (Fig. 1A). Budding yeast divide asymmetrically with respect to aging as well, resulting in the accumulation of aging factors, such as damaged proteins and unfit organelles by old mother cells (Fig. 1G) 1 .
Microfluidic cell isolation using inertial focusing is an appealing process due to its low cost and continuous operation of the disposable polydimethylsiloxane (PDMS) chip 10,14 . However, current approaches predominantly sort particles in low flow rates, increasing the risk of clogging and resulting cross-contamination. In addition, the low resolution of current devices led to primarily coarse separations of budded yeast from mother-daughter pairs, where significant size difference makes the separation trivial 12 . In contrast, our approach relies on our recently described 14 opposing shear-induced forces that push large particles towards the concave side of curved channels ( Fig. 2A,C). Superimposed on this mechanism are the effects of Dean vortices confining small particles to the center of the channel due to rotation induced lift forces (Fig. 2B,C). As inertial forces scale as a 4 , and Dean vortices induced forces scale as a 2 , our method permits sensitive separation that theoretically scales as a 2 , where a is the particle diameter. High Reynolds that limit clogging and permits the processing of up to 10 7 cells/min offers a unique tool to separate microbiota for metabolomic and proteomic analysis. We applied this separation mechanism for the challenging sorting of 4.1 μm beads and 5.56 μm beads, reaching purity of 90.6% ± 8% from a heterogeneous mixture. The separation resolution is much higher than previous works that demonstrated separation of cells at least 5 μm in diameter larger between the two sorted populations 10,36 .
Using microfluidic separation we isolated predominantly young (1.3 ± 1.6 scars) and adult (3.4 ± 1.7 scars) yeast from the center and concave ports, respectively, without affecting viability or function (Fig. 2F,G). Surprisingly, even that small difference in replicative age led to significant changes. Young yeast showed a significantly 27% shorter lag time to exponential growth (Fig. 3F), while adult yeast were not significantly different from the original mixture. VHL is a human tumor suppressor that misfolds in yeast 37 , undergoing proteasome degradation 38 . VHL degradation rates were 4.5-fold lower in adult cells possibly due to downregulation of CCT3 which is crucial for VHL recognition 30 . These results suggest that protein quality control is already affected at early stages of replicative aging, explaining the accumulation of HSP 38,39 during early stages of replicative aging (Fig. 1G).
Our high-Reynolds approach allowed us to isolate sufficient cell quantities for mass spectrometry, showing that young (1.3 ± 1.6 scars) and adult (3.4 ± 1.7 scars) cells, originating from the same culture but separated by 2-3 replication events, have distinct proteomic profiles. We show that changes in classical aging and youthfulness markers such as CalM1, SAM1, RPL5, SSD1, and DHH1 may actually occur earlier than previously reported. Our results support earlier proteomic analysis of age-dependent changes in Caenorhabditis elegans, reporting a decline in protein homeostasis in adult nematodes.
In summary, our microfluidic platform offers the ability to rapidly isolate large quantities of microorganisms for proteomic and metabolomic analysis, offering insights into population composition and dynamics. While our current findings could have implications for understanding cellular aging in eukaryotes, the technology can be readily applied to study pathogens, different yeast populations and the diverse human microbiome.

Methods
Yeast strains, growth conditions, and materials. Yeast growth, media preparation, and manipulations were performed as previously described 40 . To increase the adult yeast population, cell were grown in high-density cultures in YPD medium for 3 days, and the medium was switched to fresh medium every 24 hours, without diluting the culture. Yeast strain used in this study is BY4741 except for the VHL degradation assay. For the VHL degradation assay, plasmid pESC-URA GAL1p GFP-VHL/GAL10p NLS-TFP (DK5 plasmid) 40 was used. The VHL gene was fused to GFP and was expressed under the control of a galactose-regulated bi-directional promoter (Gal1p); NLS-TFP was used as a nuclear marker and expressed under Gal10p.

Microscopy. Yeast cells were grown in liquid medium to mid-Log phase and seeded on concanavalin A
(Sigma-Aldrich) coated four-well microscope plates (IBIDI). To label the cell walls of living cells, cells were stained for 30 min at room temperature with rhodamine-concanavalin A (Vector Laboratories) diluted 1:100 from 5 mg/ml stock solutions, and with Calcofluor White (Sigma-Aldrich) diluted 1:10 from 1 mg/ml stock before experiments to stain the bud-scars. The nuclei were tagged with Htb2-GFP from endogenous GFP-tagged yeast library. Confocal 3D images were acquired using a dual point-scanning Nikon A1R-si microscope equipped with a PInano Piezo stage (MCL), using a 60x PlanApo VC oil objective NA 1.40. Calculations of the fluorescence intensity and image processing were performed using NIS-Elements software.
Single Cell Analysis Algorithm. Single cell analysis of yeast was performed with CellProfiler software 16 .
For yeast membranes identification, circular shapes were enhanced in the raw image. Circles ranging from 1.5 to 15 µm (15-150 pixels) diameter were identified using Otsu-Adaptive method and two-class thresholding. Membrane geometrical properties were measured. Threshold correction factor of 2 ± 0.5 was applied to avoid background noise identification as cells. Clumped objects were distinguished by shape. Budding scars and HSP104 foci were identified in a raw image masked by the formerly identified yeast membranes, using Robust-Background-Per-Object method. Threshold correction factor of 1.5 ± 0.5 was applied to determine the identified objects. In all objects identifications, exact threshold for an image series was calibrated using side-by-side examination of 5-10 images. Scars and foci were related to cells using Relate-Objects function. CellProfiler code is available upon request. HSP104 aggregation assay. Yeast cells expressing HSP104 endogenously tagged with GFP 41 were grown as described, without any heat shock, labeled with calcofluor white and imaged. Pictures were analyzed for HSP104-GFP inclusions and scar number using CellProfiler program.
Protein degradation assay. Cells were grown as described above. The expression of GFP-VHL variants was induced by switching to galactose-containing medium for 3 days. The degradation was calculated as a decrease in the green fluorescence intensity after cells were transferred to glucose medium. The fluorescence intensity of single cells was calculated using FACS. All data points were normalized to GAL induced nuclear localization sequence tag red fluorescence protein (NLS-tRFP) fluorescence decay expressed on the same plasmid.

Analysis of levels of GFP-fusion proteins and VHL-GFP degradation. Strains, containing
GFP-fusion proteins corresponding to hits from the proteomic data were obtained from the GFP-fusion collection. They were then grown and separated in standard conditions. Lysates were obtained using alkaline lysis 41,42 After SDS-PAGE and transfer to nitrocellulose membrane, the membranes were stained with Ponceau S to obtain data on the amounts of protein in the samples. This was essential, since the amount of cells for analysis was low and protein loading was difficult to equalize by other means. After western blotting with anti-GFP antibodies (ab290 Abcam), the resulting Ponceau S and chemiluminscence images were subjected to densitometry using ImageJ software. Correction for protein loading was performed by dividing the ratio observed in the blotting images, by the ratio calculated for the Ponceau S images. The latter was calculated as the densitometric ratio between the relevant samples, multiplied by 2 for each 1.2 fold difference in Ponceau S ratios. The correction coefficient was calculated from a calibration curve obtained from staining serial dilutions of Ponceau S. The calculated correction factor correlates well with previous results 43 .
Assay for VHL degradation was performed using the W303 strain with four integrated copies of VHL-GFP 44 . VHL-GFP was under the control of a CUP1 promoter and its production was induced by overnight incubation in SD-medium containing 100. VHL-GFP was under the control of a CUP1 promoter and its production was induced by overnight incubation in SD-medium containing 100 µM CuSO4. The chase procedure was done according to Buchanan et al. 45 . Quinacrine staining was performed according to Hughes and Gottschling 24 .

Fluorescence-activated cell sorting analysis. Flow cytometry was carried out on Becton Dickinson
FACSAria III. Yeast strains were grown on SC-URA selection medium for 2 days, diluted twice a day to log phase, and on the third day, the fluorescence intensity of the GFP tag fused to VHL and NLS-tRFP was analyzed by fluorescence-activated cell sorting with a 488 and 561 laser using BD FACSDiva (BD Biosciences) software.
Statistics for image processing, VHL degradation and lag time. The experiments were repeated at least 3 times. Multiple correlation coefficients were calculated for a single variable in the analyzed group, with subsequent regression analysis to determine p-values. Standard comparisons were performed using t-Test.
Mass spectrometry. Mass spectrometry of yeast cells was carried out as previously described 46 .
Sample preparation for MS analysis. Pellet of 10 7 yeast cells was re-suspended in 0.1 M NaOH and incubated 5 min at room temperature. After washing out NaOH, the cells were lysed in buffer containing 100 mM Tris, 2% SDS, and 10 mM DTT for 60 min in room temperature. Lysed samples were prepared using the FASP protocol 47 . Briefly, lysate was transferred to microcon centrifugal filters, Amicon-ultra-0.5 tubes, MW cutoff of 10 kDa (Merk Millipore, Germany) for reduction, alkylation, washing and trypsin digestion. The protein lysate was diluted and washed by denaturation buffer containing 8 M urea in 0.1 M Tris-HCL set at pH 8.5. Proteins were reduced with 10 mM of DTT for 1 hour and then alkylated with 50 mM iodoacetamide supplemented to the denaturation buffer for 1 hour in dark. To remove denaturants and alkylation reagents the proteins were washed five times with the denaturation buffer. Then the protein were digested with 0.4 μg Trypsin (sequencing grade modified trypsin V5111, Promega, Madison, WI) in digestion buffer (10% ACN, 1.3 M Urea, 25 mM Tris pH 8) overnight at 37 degrees. To remove excess of Urea, the peptides were washed with 10% ACN, 25 mM Tris pH 8. The tryptic peptides were eluted from the Amicon tubes by centrifugation at 12,000 g for 10 min, evaporated till dryness, and resuspended in 0.1% formic acid (FA). The peptide mixture was desalted using in-house packed C18 StageTips 48 . The StageTips were conditioned at first with 100% Methanol and then with 80% acetonitrile (ACN). The equilibration was performed with 2 × 0.1 ml 0.1% FA before application of the samples at a total amount of 15 µg total protein per stage tip. The bound peptides were washed with 2 × 0.1 ml 0.1% FA and eluted with 120 µl 80% ACN plus 0.1% FA. The eluted fractions were dried under vacuum and resuspended in 0.1% FA to a final concentration of around 0.4 µg/µl. Nano-LC-MS/MS Analysis. The peptides were injected into a Nano Trap Column, 100 μm i.d. × 2 cm, packed with Acclaim PepMap100 C18, 5 μm, 100 Å (Thermo Scientific) for 8 min at flow 5 ul/min, and then separated on a C18 reverse-phase column coupled to the Nano electrospray, EASY-spray (PepMap, 75 mm × 50 cm, Thermo Scientific) at flow 300 nl/min using an Dionex Nano-HPLC system (Thermo Scientific) coupled online to Orbitrap Mass spectrometer, Q Extactive Plus (Thermo Scientific). To separate the peptides, the column was applied with a linear gradient with a flow rate of 300 nl/min at 35 °C: from 1 to 25% in 82 min, from 25 to 35% in 45 min, from 35 to 55% in 43 min, from 55 to 90% in 5 min, and held at 90% for an additional 30 min, and then equilibrated at 1% for 20 min (solvent A is 0.1% formic acid, and solvent B is 80% acetonitrile, 0.1% formic acid). The Q Exactive was operated in a data-dependent mode. The survey scan range was set to 200 to 2000 m/z, with a resolution of 70,000 at m/z. Up to the 12 most abundant isotope patterns with a charge of ≥2 and less than 7 were subjected to higher-energy collisional dissociation with a normalized collision energy of 28, an isolation window of 1.5 m/z, and a resolution of 17,500 at m/z. To limit repeated sequencing, dynamic exclusion of sequenced peptides was set to 60 s. Thresholds for ion injection time and ion target value were set to 70 ms and 3 × 10 6 for the survey scans and to 70 ms and 10 5 for the MS/MS scans. Only ions with "peptide preferable" profile were analyzed for MS/MS. Data was acquired using Xcalibur software (Thermo Scientific).
Data Analysis and Statistics. For protein identification and quantification, we used MaxQuant software 49 , version 1.5.3.30. We used Andreomeda search incorporated into MaxQuant to search MS/MS spectra against the UniProtKB database of Saccharomyces cerevisiae proteome, (Uniprot release, Aug 2015). The identification allowed two missed cleavages. Enzyme specificity was set to trypsin allowing cleavage N-terminal to proline and up to two miscleavages. Peptides had to have a minimum length of seven amino acids to be considered for identification. Carbamidomethylation was set as a fixed modification, and methionine oxidation was set as variable modifications. A false discovery rate (FDR) of 0.05 was applied at the peptide and protein levels. Initial precursor mass deviation of up to 4.5 ppm and fragment mass deviation up to 20 ppm were allowed. Only proteins identified by more than 2 peptides were considered. To quantify changes in protein expression we used the label-free quantification (LFQ) using the MaxQuant default parameters 49 . For statistical and bioinformatic analysis, as well as for visualization, we used the Perseus software 50 . (http://www.biochem.mpg.de/5111810/perseus). For annotation, the DAVID webserver 51 was used. STRING server (http://string-db.org/) 52 was used to define interacting proteins, which were visualized by using Cytoscape software 53 .
Device fabrication. Serpentine shape molds were fabricated by photolithography of SU8 on silicon wafers. Channels were replica molded in PDMS and bonded to glass using oxygen plasma as previously described 10,14 . Channels and glass were cleaned using M3 low-residue adhesion tape. This cleaning made acid processing redundant and enabled device durability to velocities up to 12 m/s. High aspect ratio channels were designed to achieve distinct horizontal focusing locations. Channels are composed of four half-circle arcs each of them is 33 μm high, 200 μm wide, and the radius of curvature is 900 μm. Prior to the experiment the channel was coated with Pluronic F-68 (Sigma-Aldrich) for 1 hour at room temperature to prevent non-specific adhesion of fluorescent particles.
Yeast sorting. Yeast cells were grown as described, pelleted by centrifugation and re-suspended in water.
Cells density diluted to OD of 0.2 and injected to the microfluidic device at 1500 μl/min. The high fluid velocity enables binning directly from a wide cone shaped drainage section (Fig. 2D,E) avoiding the risks of particle sedimentation and division sub-channels blocking.