A high-throughput microfluidic diploid yeast long-term culturing (DYLC) chip capable of bud reorientation and concerted daughter dissection for replicative lifespan determination

The high-throughput microfluidic DYLC chip was designed to capture single budding yeast cells, reliably retain mother cells, and effectively remove daughter cells so as to perform long-term culturing and microscopic monitoring for later-on image analysis to determine yeast RLS. Figure 1 shows the micrographs and schematics of the microfluidic chip, which simply consists of an 8-µm-high PDMS microchannel bonded on a glass substrate. The microchannel features an inlet for cell suspension and medium infusion, 53 cylindrical posts, 1100 "leaky bowl"-shaped traps formatted in an array, and an outlet for waste collection (Fig. 1A). The posts, with the diameter of 30 μm, were designed to support the ceiling of the wide microfluidic channel from collapse during chip bonding. In order to prevent the whole channel clogging among the dense traps, the trapping array was patterned into 5 subarrays with a spacing of 100 μm in between (Fig. 1B). Each subarray contains 220 traps (10 rows × 22 columns) with optimized distances of 30 and 34 μm between adjacent columns and rows, respectively. Traps in each column are aligned to the middle of spacings between two neighboring traps in adjacent columns. Such setting of the trapping array ensures both efficient trapping and enough growing space of yeast cells. Each trap (outer profile: 15 μm × 7 μm) comprises two pillars facing each other, together forming a "bowl"-shaped wide opening (width: 8 μm) upstream and a "leaky" orifice (width: 3 μm) downstream. The length of the wide opening was optimized to 5 μm to ensure the yeast trapping in single-cell resolution, and the length of the orifice was set to 2 μm to minimize the structural squeezing or mechanical stress on newborn buds during their growth. Once yeast suspension was infused into the microchannel, cells were dragged to be docked inside the empty "bowls" by hydrodynamic forces (Fig. 1C). Since the height (8 μm) of each trap is larger than the diameter of budding yeast cells (4 ~ 6 μm), immobilized single cells with small buds have sufficient space vertically to rotate in the "bowls" under hydrodynamic forces, and buds could be ultimately orientated and clamped in the orifice downstream when growing bigger (Fig. 1D). After the completion of cytokinesis, the matured buds (i.e., daughter cells) could be automatically detached by hydrodynamic shearing forces whereas mother cells were stably retained in their traps. Therefore, the immobilized yeast cells during their entire lifespans could undergo consistent hydraulic microenvironments, such as the hydrodynamic forces in cell growth, mother retention and daughter removal. In particular, for accurate RLS determination with the diploid budding yeast, which features random budding spots on mother cells, the delicate "leaky bowl"-shaped traps with the optimal dimensions could thus provide the merits of free mother rotation in the traps, bud reorientation to the narrow orifices and concerted daughter dissection.

To better understand the cell-trapping process and therefore optimize the geometric arrangement of cell traps, the fluid field distribution in the array was calculated in CFD simulation under three different settings of the distance between adjacent rows (d_r) and the misalignment in rows (d_m). Contour plots in Fig. 2A show the flow velocity distributions across the 5 × 5 cell-trap array in the xy-plane. Compared to the flow velocity distribution in setting 2, the region with high value of flow velocity in setting 1 is almost interconnected so that cells are prone to bypass traps and flow along this region towards downstream. In setting 2, dissected daughters from upstream tend to be captured by traps downstream, while in setting 3 they are inclined to make a detour around traps downstream due to larger d_r. Moreover, for practical considerations, slightly larger d_r in setting 3 could reduce the occurrence of channel clogging caused by abnormally large cells when approaching apoptosis. Hence, the distance between adjacent rows and the misalignment in rows of the cell-trap array were suggested to be 34 μm and 17 μm, respectively.

With the suggested geometric setting of the array, particle trajectory tracing was then simulated to investigate the process of cell trapping before and after the traps upstream were docked with cells (Fig. 2B). Due to the low flow resistance through the orifices of the "leaky bowl"-shaped traps, suspended cells could be initially dragged towards the empty traps under hydrodynamic forces. Once the wide openings were occupied, drastic increase of flow resistance in traps rendered subsequent cells bypassing these traps and flowing towards the empty traps downstream (Additional file 2: Video S1). Such dynamic process of single-cell trapping was observed in experiments (Additional file 3: Video S2). A budded cell, rotating with the medium stream, was immobilized in an empty trap, and then cell suspension from upstream bypassed this occupied trap successfully. Therefore, the simulation results demonstrated that cells in the loading process could rapidly fill the array by evading the occupied traps and giving preference for empty ones.

The efficiency and reliability of single-cell immobilization plays a vital role in sample collection for high-throughput imaging and data analysis. With the optimal design (d_r = 34 μm and d_m = 17 μm) of the cell-trap array, the cell-trapping efficiency was investigated in experiment. After cell loading, an initial trapping rate could reach 70.5% on average. The trapping rate could be raised by increasing the concentration of cell suspension and the duration of cell loading. However, a higher trapping efficiency in the beginning of the experiments was not demanded since empty traps could be filled with daughter cells removed from traps upstream in the next several hours during cell culturing (Fig. 3A). Within the first 4 h of yeast culturing, the cell-trapping efficiency could be raised from 70.5 to 92.3% (Fig. 3B), thereby providing a high-throughput platform for yeast immobilization. With the ability of self-filling among the staggered traps, low concentration of cell suspension was preferred for single-cell immobilization.

Next, we investigated the capability of single-cell maintenance in traps that differed in the length of the "bowl"-shaped wide opening (i.e., 4 μm, 5 and 6 μm). In the experiment with 4-µm-long "bowl" of traps, large mother cells, when approaching apoptosis, could be dragged away by unrotated buds that were located upstream (Fig. 3C). Such loss of mother cells could dramatically cut down the number of samples for RLS determination. In the experiment with 6-µm-long "bowls", which provide sufficient space to dock two small yeast cells, traps occupied with single yeast cells could capture extra cells coming from the upstream (Fig. 3D). The extra cell trapping could affect the imaging quality and identification of original mother cells in long-term cell culturing. Figure 3E shows the percentages of occupied traps that were occurred with mother loss and extra cell capture during 0 to 4 h and 24 to 28 h after cell loading. We can see that the mother-missing rate gradually decreases with the increase of "bowl" length, while the extra-trapping rate rises up. Therefore, the length of wide-opening was optimized to 5 μm for the sake of stability in single-cell retention.

A haploid budding yeast always forms new buds adjacent to the scars of previous daughters [19, 20]. Thus, haploid yeast that initially sprouted downstream in the microfluidic traps could undergo successive removal of daughters in following generations [14, 15]. However, diploid budding yeasts, which forms new buds either adjacent or opposite to the previous bud site, would have high risks of being dragged away when buds are towards the upstream of traps. Hence, the microfluidic DYLC chip was designed to ensure concerted daughter removal of diploid budding yeast cells within consecutive generations by taking the advantages of bud reorientation into the unified orifices downstream.

Firstly, bud reorientation of immobilized budding yeast under continuous fluidic perfusion was theoretically analyzed by numerical simulation of hydrodynamic forces on a single cell docked in a trap (Additional file: Fig. S1A). When the bud was set to 0 (i.e., along -x direction) or 45 degree, the hydrodynamic force on bud points obliquely upwards, showing the motional tendency of the bud. After the bud rotated over 90 degree (i.e., along + z direction), the force points obliquely downwards, indicating the bud to be orientated into the orifice. At 180 degree (i.e., along + x direction), the z-component of the hydrodynamic force on the bud is negative, meaning that the reoriented bud could be stably maintained in the orifice. Therefore, the simulation result shows the feasibility of bud rotation from upstream to downstream in the orifice under hydrodynamic forces. Moreover, with the increase of bud diameter from 2 to 4 μm, the x-component of hydrodynamic forces gradually rises up and dominates during the bud growth (Figure S1B), suggesting that larger bud could be retained in the orifice more stably till the consequent daughter dissection had occurred after the completion of cytokinesis. Also, larger bud possesses higher hydrodynamic force for dragging the bud downstream to accomplish daughter detachment. 1

Secondly, occurrences of bud rotation in experiment were verified by time-lapse images, especially with the mother cells that had two buds during a certain time period (Fig. 4A). In the case of Cell 1, the small bud (2nd bud) sprouted next to the unseparated daughter (1st daughter), and the 2nd bud remained in the narrow orifice downstream after the separation of 1st daughter. The 3rd bud successfully oriented into the orifice following the same path as the 2nd one did. The cases of Cell 2 and Cell 3 demonstrate the reorientation of new buds that sprouted on opposite and random sites of previous daughters, respectively. We could clearly observe that newborn buds (2nd buds) appeared before the removal of daughters (1st daughters). In the next images, the buds were oriented into the orifice downstream and meanwhile the mature daughters disappeared. Afterwards, the buds remained downstream throughout their division periods, followed by their dissection and the reorientation of new progenies (3rd buds) into the orifice. Furthermore, Videos S3 and S4 show the dynamic rotation and reorientation of immobilized budding yeast cell. Initially, the mother cell with a tiny bud was freely rotating in the "bowl"-shaped trap. With the bud growth, the cell stopped rotating with its bud clamped in the narrow orifice and swayed under the continuous medium perfusion. As the bud grew larger, the bud became motionless until it was dissected after the completion of cytokinesis (Additional file 4: Video S3). Right after the removal of the mature progeny, the following bud was immediately rotated into to the orifice (Additional file 5: Video S4).

Thirdly, the rotation rates of cells immobilized in two trap arrays that differed in height (7 μm vs. 8 μm) were compared in terms of the initial six generations of buds. The percentage of successful reorientation towards the downstream in traps with the height of 7 and 8 μm were 64.67% and 81.12%, respectively (Fig. 4B). The result indicated that 8-µm-high traps provided sufficient space for cell rotation and consequently concerted daughter dissection. Video S5 recorded the whole lifespan of an immobilized budding yeast cell. We could observe the successful orientation of buds towards the orifice downstream followed by concerted daughter dissection in the orifice for the 25 generations. However, bud rotation occurred less as the mother cell aged. This phenomenon could be attributed to following reasons: One is that the size increase of the senescent mother severely restricted the mobility of buds in traps; the other one is that the delayed detachment of daughter cells rendered the new buds growing too large to be rotated.

Therefore, the above experimental results demonstrated that the "leaky bowl"-shaped yeast traps enable efficient cell rotation by hydrodynamic forces, and cells could subsequently undergo the concerted processes of bud reorientation and daughter dissection in successive division cycles until apoptosis.

To validate the capability and reliability of the microfluidic DYLC chip in high-throughput aging study of diploid budding yeast cells, long-term monitoring of single immobilized cells for over 60 h was performed. Figure 5A shows one representative cell that continuously gave birth to new buds until its death. The corresponding data analysis of this cell was conducted to enumerate the budding and dissection events throughout its lifespan by using the sequential two-digit coding. RLS of this cell, 23 generations, was calculated from the recorded digital matrix and its converted oscillogram (Additional file 1: Fig. S2). According to the data analysis of 786 cells trapped in the initial 10 h of 3 independent experiments, we obtained the reduction in cell viability as a function of replicative ages (Fig. 5B), resulting in an average RLS of 24.29 ± 3.65 generations. Then, additional 211 cells captured after the first 10-hour running of experiments were included in RLS analysis. We redistributed all 997 cells by grouping them according to the time point of the mother-cell trapping at a 2-hour interval (Fig. 5C). The fitting curve shows statistically shortened lifespans through the whole data set. The mean RLS (24.87 generations) was highest at the first 2 h of the experiments, while the minimum (only 14.25 generations) appeared on cells captured between 44 and 46 h. Moreover, the mean RLS of cells captured after 20-hour running was 17.21, a large decrease of 29.15% compared to that in the first 10 h. This phenomenon implies that a portion of senescent mother cells after 20-hour culturing may produce daughters that could not inherit full lifespan [25]. Due to the asymmetric division between mother and daughter yeast, the shortened lifespan in daughters from old mothers has been debated at the protein level. The function of asymmetric distribution of proteins, which accumulate steadily as mother cells aged, usually keep the protein level in daughter cells lower than that in their mothers. However, as was proposed and verified that daughters born from old mother cells have a protein level comparable to middle aged mothers, the lifespan could be limited when the inherited proteins reach a proper level [26]. The downward trend of the linear fitting curve in Fig. 5C also agrees well with the reported results that lifespan reduction in daughters may accumulate progressively as the age of their mothers increased.

The microfluidic DYLC chip enables large scale screening of cell samples to investigate the age-related changes in BTI. Since the time duration for the first G1 phase of haploid yeast fluctuated over a tenfold range, the mean budding time interval (BTI) of the first generation of haploid yeast could show great extension than subsequent generations [27]. To investigate the dynamic BTI of diploid budding yeast cells at birth, 33 mother cells in a small size and initially showing no obvious bud were selected and analyzed. The budding time interval (BTI) was obtained from the digital matrix (Additional file 1: Fig. S2). As illustrated in Fig. 6A, mean BTI of the first generation of diploid cells exhibits an extension of 9%, compared to that of the next five generations, similar to the haploid yeast. To investigate the BTI variation upon cell apoptosis, BTI of the 33 mother cells was then calculated in death-centric perspective by aligning all data to the last generation in their lifespan. The result in Fig. 6B demonstrates that BTI of diploid yeast cells is almost steady during most of the lifespan, whereas it rises up dramatically in the last several generations.

It is commonly recognized that the stochasticity and robustness are inherent properties of living cells [28]. However, experimental results derived from conventional approaches were usually based on assumptions that cells performed uniform growth and division and the group behaviors could hide heterogeneities among individual cells. The cellular de-synchronization in diploid yeast aging was verified through 400 samples from 3 independent experiments, where cells were relatively small in size and with no obvious bud initially. For better visualization, BTIs in every 5 consecutive generations were periodically marked with 5 different colors in a kymograph. As shown in Fig. 6C, the BTI distribution remains synchronized within the first 8 to 10 generations, whereas it gradually differentiated and elongated as cells progressively approached to the end of their lives. The BTI divergence may demonstrate the heterogeneity among individual cells even under the same microenvironment. In addition, the result of diploid yeast cells may exhibit better stability than haploids, which started to de-synchronize after about 5 divisions [14]. Therefore, the consistence of the mean BTI and BTI distribution until the last few generations indicates that diploid yeast cells have a better resistance to extrinsic fluctuations and intrinsic noises, such as environmental stress, protein accumulation and DNA damage during their lifespans [29].

The microfluidic DYLC chip was fabricated using a standard PDMS-based soft-lithography process. First, an SU-8 (SU-8 3010, MicroChem Co., USA) master was patterned directly onto a 4-inch silicon wafer using photolithography. The master was then silanized with trichloro (1 H,1 H,2 H,2 H-perfluorooctyl) silane (Sigma-Aldrich Co., USA) in vapor phase to prevent PDMS from sticking to SU-8 surface in replication. Afterwards, a mixture of PDMS oligomer and cross-linking polymer with a weight ratio of 10:1 was degassed and poured onto the SU-8 master, and was cured by baking on a hotplate (PR 5 SR, Harry Gestigkeit GmbH, Germany) at 90 °C for 2 h. Then, the PDMS replicas with channels and micro structures were peeled off from the master, and cut into pieces, which were then punched with 1-mm-diameter holes as inlets and outlets for fluid perfusion. To finalize the chip fabrication, by activating both PDMS and glass surfaces with an oxygen plasma cleaner (PDC-002-HP, Harrick Plasma, USA), each PDMS sheet was irreversibly bonded onto a glass slide that was previously diced from a 500-µm-thick 4-inch Borofloat 33 glass wafer.

Diploid budding yeast (Saccharomyces cerevisiae, strain BY4743 was employed in this work. Dextrose complete (SD) medium was used for cell culturing by standard methods as described in the literature [19]. Solid SD plates (pH 6.5) are composed of 0.17% w/v yeast nitrogen base (YNB) without amino acids, 2% w/v glucose, a complement of amino acids and nucleotides and 15% w/v agar. Liquid SD medium (pH 6.0) has the same compositions as solid SD medium except agar, and the concentration of each composition in liquid SD is the same as that in solid SD medium. Additionally, 0.05% w/v Pluronic^® F127 is added in liquid SD to prevent hydrophobic compositions from sticking together or to PDMS traps. All chemicals were purchased from Sigma-Aldrich Co., China.

For cell culturing, yeast cells were first grown on solid SD plates. To prepare cell suspension, yeast was scraped up with a sterile inoculating loop and suspended in liquid SD medium. Suspended yeast cells were then incubated in a flask at 30 ℃ with shaking at 300 rpm for 12 h until cells growing in stationary phase. Afterwards, 100 µL cell suspension was inoculated into another flask containing 10 mL fresh liquid SD medium and incubated for 6 h to resume exponential growth. Before loading cells into the DYLC chip, cell suspension was diluted to approximately 1 × 10⁶ cells per mL.

To optimize the cell-trapping performance in the array, computational fluid dynamics (CFD) simulation in COMSOL Multiphysics software (COMSOL Inc., Burlington, MA, USA) was performed under different geometric conditions, including the distance between adjacent rows (d_r: 30 μm, 30 and 34 μm, respectively) and the misalignment in rows (d_m: 10 μm, 15 and 17 μm, respectively). To simplify the simulation, the geometric model was established in a 5 × 5 cell-trap array (Additional file 1: Fig. S3A, Table S2). Geometric structures and dimensions of the model were derived from the microfluidic DYLC chip (cf. Fig. 1D). Settings of density of the fluid subdomain, dynamic viscosity, and boundary conditions in numerical simulation are listed in Table S3. Specially, to amplify the subtle differences between different array settings, the laminar flow speed at the entrance was set larger than the equivalent value from the medium perfusion conditions used in practice. Contour plots referring to the flow velocity distribution were used to investigate the fluid field in the 5 × 5 array with different settings of d_r and d_m. Based on the obtained fluid field in the middle cross-section of the array (Additional file 1: Fig. S3B), time-dependent particle trajectory tracing was calculated to study the dynamic process of cell trapping (Additional file 1: Table S4). In addition, for better illustration of cell trajectories before and after cell capture in traps, two columns of cell traps upstream were set empty and filled with single cells, respectively.

To investigate the bud rotation and concerted daughter-cell dissection, hydrodynamic forces over a budding yeast cell immobilized in a single trap were calculated [36]. The budding yeast was modeled as a conjugation of an ellipsoidal mother (5 μm × 4.4 μm) and a spheroidal bud (2 μm in diameter), contacting with each other at the upstream end of the major axis of the ellipsoid (Additional file 1: Fig. S4). To imitate the bud rotation in the trap, the mother cell together with its bud, were set to rotate from -x to + x direction along the xz-plane with a 45-degree interval, meanwhile the mother was kept contacting with the "bowl" bottom. To imitate bud growth after the bud was relocated in the narrow orifice, the diameter of the sphere was set 2 μm, 3 and 4 μm, respectively. Hydrodynamic forces over the bud of each setting of orientation and diameter were numerically calculated. Arrows with quantified lengths were labeled on the cell to visualize the force values in both 3D and planar projection views.

To calculate yeast RLS and budding time interval from time-lapse images, cell events, including budding and daughter dissection, were analyzed by comparing the changes of bud number between two adjacent images. In experiment, sometimes cell events did not occur in order, or could not be recorded within the 10-minute time interval of imaging, thereby resulting in three representative cases (Fig. 7A). (i) Former bud was observed from its appearance to removal, after which came the latter bud orderly with an observable time gap. At least one image could show that the mother cell held no bud after the removal of the first bud and before the appearance of the second bud. (ii) Latter bud sprouted before the removal of former bud. Images in this case could show that the mother cell held two buds in a certain time period. (iii) The removal of former bud and the appearance of latter bud occurred orderly without an observable time gap. In this case, a big bud could be observed in an image, but in the next image it disappeared and meanwhile a small bud was observed.

To efficiently analyze cell events of yeast budding and daughter dissection of each immobilized mother cell, sequential two-digit coding, which corresponded to each frame of time-lapse images, was devised to mark the change of bud number ('no bud', 'one bud', 'two buds' and 'three or more buds') (Fig. 7B). Each code, comprising 'Δn' and 'ΔT', was defined as follows: 'Δn' denoted the change of bud number between two adjacent images; '0', '1' and '-1' of 'Δn' indicated no change in bud number, production of a new bud and removal of an old bud, respectively. 'ΔT' marked whether there was an observable time gap in images showing the mother holding no bud for clearly separating two adjacent generations. For instance, 'ΔT' = '1' indicated that the removal of former bud and the appearance of a new bud occurred within a 10-minute interval (cf. Case (iii) in Fig. 7A). Then, the sequential codes of the entire time-lapse images of a recorded cell were obtained to enumerate all budding and dissection events of a recorded cell and transformed into a digital matrix (Fig. 7C). Consequently, the yeast RLS was determined by summating the total number of '1' in 'Δn' and 'ΔT'. Additionally, the budding time interval, which was defined as the time duration between two adjacent successive budding events, could be calculated by the following steps: (1) subtracting the former image number, of which 'Δn(i)' or 'ΔT(i)' were marked as '1', from the latter one; (2) multiplying the difference from last step by 10 min. Then, the budding time intervals throughout the whole lifespan of a recorded cell could be obtained as a function of the generation of buds. Finally, an oscillogram developed from the digital matrix could intuitively illustrate the calculation processes of RLS and budding time intervals (Fig. 7C).

In data statistics, standard error (SE), which was calculated by dividing the square root of independent runs into the standard deviation, was used to evaluate the precision between the sample mean and the population mean unless stated separately.

Additional file 1: Figure S1. CFD simulation of hydrodynamic forces exerted on an immobilized budding yeast cell. (A) Hydrodynamic forces on the bud quantified by arrow length during hydrodynamic rotation of budding yeast from upstream to downstream in a 45-degree interval on the xz plane. (B) Hydrodynamic forces on a growing bud (diameter: 2 μm, 3 μm and 4 μm) towards the downstream to evaluate the cell retention and daughter detachment in the orifice. Figure S2 Calculation of RLS and BTI by using the digital matrix and oscillogram recorded from the representative cell in Figure 5(A). Figure S3 Wireframes of the model geometry containing channel walls (black), a 5 × 5 array of cell traps (blue) and immobilized yeast cells (brown) and culturing medium filled in the residual space of the model. (A) 3D view of the simplified geometric model established as follows: A 5 × 5 cell-trap array with different settings of geometric features (dr and dm) was placed in the middle of a microchannel, which featured 400 μm × 8 μm outer profile in xz-dimension and variable length dy in y-axis for fitting 5 columns of traps. The distance between adjacent columns was set to 30 μm. (B) 2D cross-section view of the array for particle trajectory tracing. dr, dm and dy in modeling were set to 34 μm, 17 μm and 158 μm, respectively. Figure S4. 3D wireframe of the model geometry in simulation of bud rotation in a single trap, containing channel walls (black), a trap (blue) and a budding yeast cell (brown) and culturing medium filled in the residual space of the model. Geometric conditions were set as follows: The trap was placed in the center with 30 μm × 34 μm × 8 μm outer profile in xyz-dimension; the budding yeast cell was set as a conjugation of an ellipsoid (5 μm × 4.4 μm) and a sphere (2 μm in diameter), contacting with each other at the upstream end of the major axis of the ellipsoid initially. To imitate the bud rotation in the trap, the mother cell together with its bud, were set to rotate from -x to +x direction along the xz-plane with a 45-degree interval, meanwhile the mother was kept contacting with the "bowl" bottom; to imitate bud growth after the bud was relocated in the narrow orifice, the diameter of the sphere was set 2 μm, 3 μm and 4 μm, respectively. Table S1. A comparison of reported RLS data of budding yeast using conventional and microfluidic methods. Table S2. Geometric settings of the cell-trap array in CFD simulation. Table S3. General settings and parameters in CFD simulation. Table S4. Settings and parameters of particle trajectory tracing in cell-trapping simulation

Additional file 2: Video S1 Video of CFD simulation showing the dynamic process of cell trapping in the array. Cells in suspension bypass the occupied traps and are captured in empty traps downstream.

Additional file 3: Video S2 A budded yeast cell was rotating when flowing towards an empty trap, and then immobilized at the trap. Afterwards, cells coming upstream were bypassing the occupied trap. In order to record the cell movement, the flow rate was slowed down to 0.5 μL/min. Scale bar is 10 μm.

Additional file 4: Video S3 Bud rotating, swaying and fixedness of immobilized budding yeast cells with their buds at different sizes. Scale bar is 5 μm.

Additional file 5: Video S4 Momentary daughter dissection and immediate bud reorientation towards downstream. Scale bar is 5 μm.

Additional file 6: Video S5 Time-lapse images of the whole lifespan of an immobilized budding yeast cell from newborn to death (RLS: 25 generations). Scale bar is 5 μm.