If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
DispenCell is an innovative impedance-based method for single-cell isolation.
DispenCell automation increases throughput, robustness and user-friendliness.
The impedance signal provides a reliable proof of monoclonality
DispenCell is compatible with a workflow employing standard plates.
High clonal outgrowth is maintained with DispenCell compared to limiting dilution.
Single-cell isolation is a truly transformative tool for the understanding of biological systems. It allows single-cell molecular analyses and considers the heterogeneity of cell populations, which is of particular relevance for the diagnosis and treatment of evolving diseases and for personalized medicine. Single-cell isolation is also a key process in cell line development, where it is used to obtain stable and high producing clonally-derived cell lines, thus contributing to the efficiency, safety and reproducible quality of the drug produced. High producing clonally-derived cell lines are however rare events and their identification is a time-consuming process that requires the screening of thousands of clones. Therefore, there is an unmet need for a device that would allow the fast and efficient isolation of single cells, while preserving their integrity and providing an insurance of their clonality.
We proposed earlier an impedance based pipetting technology for isolation of single cells (Bonzon et al., 2020), with initial validations for state-of-the-art stem cell in-vitro and in-vivo assays (Muller et al., 2020). Here, we present the transition from this pioneering technology developed in an academic setting into an automated instrument, called DispenCell-S1, allowing for traceable isolation of single cells.
We developed and validated models predicting the performances for 96-well plates single-cell isolation. This resulted in a time of dispense down to 3 min and a plate filling rate up to 96%. Finally, we obtained an impedance signal reliability for proof of single particle isolation of 99% with beads and ranging from 93 to 95% with CHO cells.
Single-cell isolation is of paramount importance for many bioengineering related applications. A first example of its usage is the production of monoclonal antibody therapies, which were spotlighted by the COVID-19 pandemic [
]. To generate these therapeutic proteins, the cell is engineered to produce the recombinant protein by itself. Such a methodology can be applied to monoclonal antibodies but more generally to many other types of therapeutic proteins, such as epoetin to treat anemia, interferon β−1a used in the treatment of multiple sclerosis or etanercept as a remedy to arthritis [
. Because immortalized cell lines are used to meet production needs, they tend to be subject to genetic and phenotypic drift. The transgene expression is also not homogeneous over cells. Therefore, single-cell isolation is performed to minimize heterogeneity and isolate the rare high producing clones [
]. Whether single-cell isolation is applied to cell line development, omics or personalized medicine, the required specifications for the dispensing approach remain the same: performance, sterility, preservation of the cell viability and functionality, simplicity, accessible cost, timing, and flexibility in terms of integration into existing workflows. Moreover, tangible proof of clonality ought to be provided to the user.
Originally done via several rounds of limiting dilution, single-cell seeding was often tedious, lengthy and did not provide a reported proof of clonality. A first solution proposed was the Biomek i7 Workstation (Beckman Coulter, USA) [
], where each well is automatically imaged right after seeding, in order to exclude wells that do not contain exactly a single cell. However, with this protocol, the repartition of cells in a well plate still relies on Poisson statistics. This makes it impossible to override the theoretical limit of 37% of wells with one cell inherent to Poisson distribution. Moreover, keeping a negligible number of wells with 2 cells or more, in order to not compromise the single-cell origin of colonies, comes with the cost of diminishing the number of wells with a single cell [
]. Fluorescent-Activated Cell Sorting (FACS) allowed better handling and sorting of cells, while increasing the filling rate of well plates, meaning the proportion of wells with a single cell. Although such an approach can be used for single-cell isolation - one good example being the MoFlo Astrios™ (Beckmann Coulter, USA) [
]. Besides, FACS comes with a large dead volume, making it prone to contamination and most of all to loss of potential rare cells of interest. Other automated single-cell isolation techniques have been developed. For instance, Cytena's Single Cell Printing (Cytena, Germany) [
]. All these techniques include one or several automated imaging steps, which necessitate high quality imagers, specific plates and well visible cells. Cell visualization will depend on cells characteristics, such as their fluorescence signal (when applicable), their shape, their position in the well plate of dispense, the time elapsed since dispense, while other parameters impacting image quality may be, among others, the color of the well plate, the magnification or the properties of the dispensing medium.
In this scope, a novel single-cell isolation technology was previously introduced by some of the authors [
, relying on the Coulter principle: the passage of a cell in solution between two electrodes induces an impedance change of the solution. Therefore, by recording the impedance, one can know when a cell passes and get a trace of this cell via the impedance signal. Practically, we use a 200 µL tip with a golden external electrode and a stainless-steel internal electrode. To increase sensitivity, the size of the bottom aperture is reduced by adding a 15 µm parylene membrane, drilled with a laser [
]. The tip is loaded with a solution of cells. Dispensing is then performed by immersing the tip in a well prefilled, typically with culture medium, and applying pressure in the tip, until a cell passes. This single-cell impedance-based seeding technology was first embedded in a manual pipette, as thoroughly described by Bonzon et al. and characterized with beads and cells [
]. Among others, we described the structure of the tip and the performance in terms of sensitivity to particle size, single particle dispensing capabilities, and confirmed the cell viability after dispensing. Muller et al. [
] then validated this first design iteration with stem cells for cloning application, through a holoclone assay, a transplantation assay and a tumorigenic assay. Nonetheless, in order to be broadly used in the industry, the set-up needed to be automated and the throughput increased, which led to the pilot series DispenCell-S1. Key features of the hand-held pipette maintained in this new version of the instrument were the capacity to handle a small number of cells, the use of a disposable tip that would not compromise sterility, and the compatibility with standard workflows while maintaining very high cell viability and functionality. Because DispenCell-S1 does not come with an embedded imaging platform, it was also of importance to verify that the impedance trace provides a reliable assessment of single-cell isolation.
In this article, we will describe the technical evolution of the DispenCell-S1 platform, from an academic innovative technology to an industrial product ready for commercial launch, and the in-depth characterization of the platform in its final state. (i) We will first describe the technical modifications of the design; namely how automated movements were made possible and how fluidic control was improved by increasing the viscosity of the dispensing solution. (ii) Using beads as a model, we will then validate DispenCell-S1 performances in terms of plate filling rate and total time of dispense. This will continue with a thorough assessment of the impedance signal reliability, in particular through an inventory of the eventual sources of error. (iii) Finally, the use of Chinese hamster ovary (CHO) cells will provide an accurate and comprehensive testing of the platform: plate filling rate, total time of dispense, outgrowth, impedance signal reliability and proof of clonality will be investigated.
2. Material and methods
2.1 Bead and cell dispensing preparation
Dispensing models and parameters were assessed with 15 µm polystyrene beads, either non-fluorescent (Ref #18328–5, Polysciences, USA) or fluorescent (Ref #2106 L, Phosphorex, USA). For each experiment, beads in solution in 1x DPBS (Ref #14190144, Gibco, Ireland) and 1x TWEEN (Ref #P2287, Sigma-Aldrich, USA) were resuspended in a methylcellulose-based solution (CloneMedia, Molecular Devices, USA), at 2 × 104 beads/mL, unless stated otherwise. Depending on experiments, dispensing was performed either in 96-well plates (Ref #655101, Greiner, Germany) or in 384-well plates (Ref #781906, Greiner, Germany) pre-filled with 1x DPBS.
DispenCell-S1 was then characterized with cells. Suspension Chinese hamster ovary (CHO) cells (ExpiCHO-S™ Cells, Ref #A29127, Thermo Fisher Scientific, USA) stably expressing GFP were maintained in Hyclone SFM4CHO serum-free medium (Cytiva, USA) supplemented with 1x HT Supplement (Gibco, Ireland) and 8 mM L‐glutamine (PAA, Austria). Cells were passaged every 3 to 4 days and grown at 37 °C, 5% CO2, in 6 mL in flasks or in 3 mL in bioreactor tubes (Ref #87050, TPP, UK) placed on an orbital shaking platform at 250 rpm. Viable cell concentration was assessed with 0.1% Erythrosin B (Ref #ACR40945, Thermo Fisher Scientific, USA) staining and enumerated with the CellDropBF (DeNovix, USA) cell counter. Experiments of single-cell isolation were performed 2 to 3 days after the last passage. For dispensing, cells in suspension were filtered with a 20 µm strainer (Ref #130–101–812, Miltenyi Biotec, Germany) and diluted to a concentration of 2 × 104 cells/mL in the methylcellulose-based solution.
DispenCell-S1 automatically loads the tip with the methylcellulose-based solution for a chosen duration. Wash cycles (W) are then performed in a 25 mL 1x DPBS containing reservoir. For each dispense, an impedance threshold and a timeout are set. The threshold is used to discriminate particles from noise while the timeout is the maximum time allowed for particle detection. Parameters for each dispense are described in detail hereafter.
2.2 Impedance signals interpretation
Impedance signals were verified manually, except when said otherwise. To be qualified as such, an impedance event has to be sharp, rather symmetric, around 30 to 50 ms long (depending on whether pressure is applied or not), and bigger than 50 Ω, to account for the noise. In a minority of wells, usually less than 2% per plate, the signal exhibited too much noise or an event was not clearly defined, in which case it was excluded.
2.3 Microscopy for optical validation
For microscopic observation, we used the inverted microscope Eclipse Ts2 (Nikon, Japan). For fluorescent observation, illumination was made at 470 nm, with a BP filter cube 470/40–500–534/55. When beads were dispensed, plates were covered with a plastic film (Ref #T329–1, Simport, Canada) and left at rest overnight to allow particle sedimentation before imaging. To reduce microscopy uncertainty, we used 384 well plates, which field can be visualized entirely using a 10X objective. We also implemented a centrifugation step for the cells to be all in the same focal plane. Two independent operators performed the microscopy observation directly using the eyepiece. The high fluorescence intensity of the cells allowed a clear identification.
2.4 DispenCell-S1 characterization with beads
Bead dispensing was performed in standard non-sterile conditions, with the parameters used for each experiment detailed in Table 1.
Definition of the Plate Filling Rate (PFR) model
Table 1Experimental conditions for DispenCell-S1 characterization with beads. Tcc represents the time between two detection events.
Undetected leaking events
Impedance signal reliability assessment
15 µm beads
2 × 104
2 × 104
2 × 104
2 × 104
Washing cycles (W)
Standard dispenses with different amounts of beads
• Immersed rate: immerse the tip in a well with no pressure and acquire impedance • Emerged rate: set the tip above a well for 2 h, then rinse it
Dispense in one well, then immerse the tip in 2 wells without pressure
Different quantities of 15 µm non fluorescent (NF) beads were resuspended in the methylcellulose-based solution, with loading of varying duration and 3 x wash cycles (W). Dispensing was performed as in Table 1. Impedance graphs were verified manually, and raw data extracted with a Python script to compute different timings.
Assessment of undetected leaking events
96-well plates pre-filled with 100 µL 1x DPBS were used. Loading of 15 µm beads to a height of 17 mm in the tip (corresponding to a loading of 60 s at room temperature) and 10 x W were performed. We first estimated the immersed leaking rate with NF beads. After washing, DispenCell-S1 was turned off and the tip manually immersed about 3 mm in a well. Impedance variations were acquired during 1.5 min for three occurrences. The emerged leaking rate was then estimated with fluorescent (F) beads. The tip was positioned above a well for 2 h, rinsed in 2 other wells, and beads in these wells were counted with fluorescent microscopy.
Assessment of transport events
384-well plates pre-filled with 80 µL 1x DPBS were used to dispense 15 µm F beads, as described in Table 1. A 60 s continuous dispense (thanks to a very high threshold, typically 10 kΩ, and a timeout of 60 s) was done in one well to compute the Tcc with an event detection threshold of 200 Ω. Dispense was then made in 96 wells of the plate as followed: one well with pressure ON, followed by two wells with pressure OFF. The threshold was set at 10 kΩ to control dispensing time via the timeout, which was chosen as circa twice the Tcc. Impedance signals and fluorescent observations were compared (N = 3).
Assessment of undetermined events and threshold placement
384-well plates pre-filled with 80 µL 1x DPBS were used to continuously dispense 15 µm F beads. 60 s loading and 5 x W were performed. A 300 s continuous dispense was done in one well of the plate. After overnight sedimentation, the well was imaged and beads counted using FiJi software. In parallel, impedance graphs were manually analyzed.
Assessment of the impedance signal reliability
15 µm F beads were dispensed in 384-well plates pre-filled with 60 µL 1x DPBS. 60 s loading and 5 x W were performed. A 60 s continuous dispense was done in one well to check that there were between 10 and 60 impedance events higher than 200 Ω. Dispensing was then performed in 96 wells of the plate as in Table 1. The plate was observed with a microscope eyepiece as described above, alongside a manual inspection of the impedance graphs.
2.5 DispenCell-S1 characterization with cells
Single-cell isolation with DispenCell-S1
50 µL of filtered cells at 2 × 105 cells/mL were added to 450 µL of methylcellulose-based solution pre-warmed at 37 °C, to obtain a final concentration of 2 × 104 cells/mL. 96-well plates (Ref #0030730.011, Eppendorf, Germany) or 384-well plates (Ref #3542, Corning, USA) appropriate for cell culture were filled with respectively 100 µL or 32 µL of medium supplemented with 1x ClonalCell-CHO ACF (Ref #03820, STEMCELL Technologies, Canada) and stored at 37 °C, 5% CO2. 1x DPBS for washing was also warmed at 37 °C. Loading of 60 s for a 96-well plate and 240 s for a 384-well plate and 4 x W were performed. A 60 s continuous dispense was done in one well to check for 10 to 60 impedance events higher than 200 Ω. Single-cell isolation was then performed (threshold 200 Ω, timeout 60 s). The 96-well plates were observed under the fluorescent microscope by one operator after the dispense while for 384-well plates, microscopic observation as described above was performed by two operators after a centrifugation step at 300 g, 5 min, carried out 10 min after the end of dispense. A minority of wells in which the fluorescence signal was unclear were excluded.
Pseudo limiting dilution with DispenCell-S1
Sample, plate preparation and analysis were performed as described above in 384-well plates. Pseudo limiting dilution was performed in 160 wells after single-cell isolation in 96 other wells of the same plate and 4 x W. A 60 s continuous dispense was done in one well to compute the Tcc with an event detection threshold of 200 Ω. Pseudo limiting dilution was then performed in 160 wells (threshold 10 kΩ, timeout 0.7 x Tcc).
Single-cell isolation with conventional limiting dilution
The initial cell suspension was diluted serially to reach a concentration of 5 cells/mL in medium containing 1x ClonalCell-CHO ACF supplement. A 96-well plate was filled with 100 µL of this cell suspension, then incubated for 4 h at 37 °C, 5% CO2 to allow cell sedimentation before fluorescence imaging.
After single-cell dispensing, plates were kept at 37 °C, 5% CO2 to assess clonal outgrowth on days 7 and 14, with a Breathe-Easy film seal (Ref #BEM-1, Diversified Biotech, USA) to permit gas exchange and limit evaporation in 384-well plates, while 100 µL of medium was added to 96 well-plates on day 7.
3.1 Automation of the instrumentation; from the handheld pipette to DispenCell-S1
, we showed the development of a sensor allowing to detect each cell leaving the pipette tip using the impedance signal. This technology enabled single-cell dispensing with a significantly improved efficiency compared to conventional limiting dilution (allowing at the most 37% wells with a single cell) [
. But a hand-held system (Fig. 1A) was not suited to industrial throughput requirements. Thus, a fully automated instrument embedding the pipetting system was developed with a XYZ robotic platform as shown in Fig. 1B–D, the DispenCell-S1, allowing the throughput to increase by one order of magnitude, from circa 1000 to 10,000 dispensing events per day (Fig. 1E).
In parallel, such dispensing automation made the instrument more user-friendly, less prone to user error, while increasing its efficiency by preventing targeting twice the same well for single-cell dispensing, and increasing the accuracy, with a constant immersion in wells. Other optimizations were compatibility with a biosafety cabinet and sterility requirements while maintaining a light weight.
3.2 Reliability of the fluidic control
DispenCell-S1 applies a flow rate through the sensing tip to perform dispensing and cancels it as soon as a single cell is detected. The effective flow rate through the aperture is given by the equilibrium of (i) the pressure applied by the pumping system, (ii) the hydrostatic pressure given by the liquid height and (iii) the capillary pressure given by the liquid interfaces. The applied pressure is the only one that can be precisely controlled using a sufficiently well engineered system. With the initial setup [
, the 3 pressures were all in the same order of magnitude, ranging from -1.5 to 1.5 mBar. This was leading to an effective flow rate varying all along the experiment and decreasing the efficiency of the system. Here, the viscosity of the dispensed media is increased from 1 cSt to 100 cSt and the applied pressure increased by a 100-fold. Conversely, the hydrostatic and capillary pressures remained constant. Thus, the flow rate is now independent of external and variable parameters.
3.3 Development of a plate filling rate model and experimental validation
We first quantified the achievable dispensing time and single particle isolation efficiency, i.e. the number of wells where the impedance signal effectively indicates one particle, called the plate filling rate (PFR). The PFR and time of dispense are assessed versus a parameter that the user can access. Many factors play an intertwined role in DispenCell-S1 performances, such as the efficiency of cell mixing in the viscous methylcellulose-based dispensing solution, the particle concentration, the tip effective aperture diameter or even the room temperature. We therefore decided to assess performances with regards to a parameter intrinsic to each experiment, and accessible by the user prior to performing single-cell isolation. This parameter, Tcc, describes the average time between two consecutive events while pressure is applied by DispenCell-S1. We built models linking dispensing time and PFR to this Tcc.
Fig. 2A illustrates the impedance signal obtained by DispenCell-S1 for a typical experiment of dispensing. Here, only one particle is detected, and the different time periods corresponding to the phases of dispensing are indicated and reported in Table 2. Tno control is introduced in addition to Tcc and represents the time window during which the pressure is applied but DispenCell-S1 has no control over the dispense of a particle. This can happen either because event detection is not yet activated (T3), or because a particle has been detected but pressure is still activated to avoid re-aspiration of this particle (T5). We therefore have the following formula:
We first built a model linking Tcc to dispensing time. We defined the dispense process time for a well Tdispense process, well. This time is composed of a constant component Twork, given by the machine incompressible times required for the dispensing process, and a varying component, linked to Tcc. By extension, Tdispense process, 96 wells is 96 times Tdispense process, well. Given Fig. 2A, we have the following formula:
Table 2Different timesteps of a typical dispense using DispenCell-S1.
Start acquisition and positioning above the well of interest (XY movement)
Entrance in the well (Z movement)
Pressure ON but not peak detection
Pressure ON, peak detection ON, lasts until a peak is detected
Pressure ON but a peak has been detected, timestep to avoid reaspiration
Pressure OFF but tip still in the well, timestep to avoid reaspiration
Exit the well (Z movement), acquisition for a well stops after this timestep
Computations between stop acquisition for a well and start acquisition for the next one
Concerning PFR, our model is based on Poisson distribution, which describes the probability of a number k of occurrences to happen, given the average number of occurrences for a fixed time λ:
In particular, the PFR is defined as the probability of zero occurrence to happen during the time when pressure is applied, but the dispense of a particle cannot be avoided, or detected, i.e. Tno control. Tcc being the time between two events when pressure is applied, PFR is given by the Poisson law of parameters k = 0 and . Tcc is calculated by dividing the time when pressure is applied by the number of events detected during that time window.
To experimentally assess the dispensing parameters and validate our models, dispensing experiments in 96-well plates were performed with 15 µm beads (one experiment consisting of 96 dispensed wells, i.e. a plate). Different concentrations of beads were arbitrarily tested, in order to scan a large range of Tcc. For each dispense in a well, DispenCell-S1 provides a similar impedance signal which manual analysis gives access to the PFR. Combined with the extraction of raw data, ie T1 to T7, we obtained Twork = 1241 ± 22 ms and Tno control = 226 ± 1 ms (Fig. 2B). This allowed computing Tcc and Tdispense process, 96 wells. Theoretical predictions and experimental results are plotted on Fig. 2C. With regards to Eq. (4), the dispensing time model predicts a linear increase of Tdispense process 96 wells with Tcc as shown in the following equation for a time given in minutes:
This would lead to a dispensing time ranging from 3 min to 11 min for a Tcc ranging from 1 to 6 s. A linear model was fitted to the experimental measurements of Tdispense process, 96 wells (red dashed line, R2 = 0.99 and with a linear regression of 1.61 Tcc + 2.06 min). The slope is close to the expected one, while the y-intercept is slightly different. This suggests a good similarity between experiments and the model but shows that Twork and Tno control could be estimated more accurately.
A PFR of 87% is obtained at a Tcc of 2 s to then reach a plateau around 98% for a Tcc greater than 13 s. A Tcc ranging from 1 to 6 s should lead to a PFR ranging from 76% to 96%. Experimental measures, although showing dispersion, tend to follow theoretical predictions and validate the model.
3.4 Assessment of the impedance signal and causes of reliability errors
The impedance signal has to be a reliable way to ensure single-cell isolation, i.e. we need to assess whether it provides a proof of what is effectively in the well. Here, the reliability error of the DispenCell-S1 impedance signal is defined as the probability of observing at least two particles in a well, for an impedance signal only indicating one event. Such phenomena can be the consequence of several events which were experimentally quantified, as listed in Table 3.
Assessment of undetected leaking events
Table 3Impedance signal, single-cell detection and error analysis. A reliability error is defined as an event that could lead to an impedance signal exhibiting one dispensing event, while two particles or more are observed in the well, i.e. effectively dispensed. Different classes of error and their rate for typical 96-well dispenses are presented. Error values were obtained experimentally.
As mentioned in Paragraph 3.2., the pressure on the aperture is the sum of three pressures, so that even when pneumatical pressure is not applied, particles can drip from the tip because of the residual hydrostatic and capillary pressures. This was defined as a leaking event.
We first assessed the immersed and emerged leaking event rates by estimating the average number of 15 µm beads leaking per min. We focused on a loading of 60 s, typical of 96-well plates. We recorded impedance variations when the tip was immersed in a well filled with 1x DPBS and no pressure applied. By counting recorded events, we obtained an immersed leaking rate of 1.4 beads/min.
The emerged leaking rate was estimated by letting a loaded tip at rest during 2 h above a well filled with 1x DPBS. We then immersed the tip manually for a few seconds in a first, then a second well, in order to rinse the tip, and counted beads in each well: no bead in the first well, two in the second. The emerged leaking rate was thus inferior to one bead per hour, i.e. 2.7 × 10−4 beads/min. Therefore, we only considered immersed leaking events for the evaluation of this error.
We subsequently defined Tcc_OFF as the time between two events when no pressure is applied. We used Poisson law to estimate the undetected leaking event rate, namely the probability of a leaking event to occur and not being detected by the user via the impedance signal. TimeOFF is defined as the amount of immersed time when pressure is not applied and the signal is not exploitable due to XYZ platform movements. It was estimated using the times extracted from Tcc experiments (Fig. 2B and Table 2) as followed, where Tunexploitable is the time where the impedance signal is not exploitable, estimated on typical impedance signals as 0.25 s for a well:
This led to the average number of occurrences for a fixed time , and the probability of having no leaking event .
We obtained a probability of undetected leaking event during a typical 96-well plate single particle isolation of 0.6 ± 0.2%.
Assessment of transport events
A source of error can be the undesired transport of a particle from one well to another, for instance by non-specific adsorption on the tip surface. Instead of having one particle in both wells, one could have zero particle in the first well and two in the second, while the impedance signal would indicate one event in each well. To assess this error rate, 15 µm F beads were dispensed with the pressure ON only one well out of three. We checked with the microscope whether beads were indeed in the wells where the impedance signal indicated they had been dispensed, or whether they had been transported to one of the two following wells. No undesired transport event was observed over three sets of 96-well experiments, performed with different tips. Thus, the non-specific transport of beads from one well to another can be neglected.
Assessment of undetermined events and threshold placement
Histograms of event impedance usually exhibit a major Gaussian distribution around the average impedance (1000 Ω with 15 µm particles) and another distribution with lower impedances, representing the noise (below 200 Ω). Depending on the experiments, a few events can lie in between these populations. These are defined as undetermined events, since there is no certitude that they represent events of particle dispense. They hamper impedance event threshold placement. To assess their contribution to error, 15 µm F beads were continuously dispensed in the same well and the impedance trace compared to fluorescent observation. Histograms of event size were plotted with a 50 Ω bin size for every experiment assessing the impedance signal reliability (Supp. Fig. 1). Calculations for the rate of undetermined events were made with a threshold at every possible position between the two impedance event populations mentioned. The end of the lowest population was considered as the first empty bin above 50 Ω, to account for noise, while the beginning of the biggest population was defined as the last empty bin before that population (red rectangles, Supp. Fig. 1). If there was no empty bin between populations, the lowest minimum between populations was considered. μk is the average value of the assessed parameter and σk the standard deviation per repeat k. We finally averaged these values and computed a combined standard deviation per experiment, where K is the number of repeats k and μ the average of μk.
We obtained a rate of 3.1 ± 3.2% for undetermined events.
Assessment of impedance signal reliability error
We then measured the impedance signal reliability error of the DispenCell-S1 single-cell isolation as followed:
This error was experimentally measured in triplicates with 96-well single particle isolation. We analyzed the impedance graphs manually, with the thresholds placed as described above, and compared them to fluorescent observation. The impedance signal reliability was 99.2 ± 0.7%. Interestingly, histograms were clean, with barely any noise or undetermined events, suggesting a very high quality and optimal impedance signal reliability.
3.5 Characterization of DispenCell-S1 performance with cells
We further characterized the performances of DispenCell-S1 for single-cell isolation with an industrial model of cells, CHO cells stably expressing GFP.
First, we tested the model linking Tcc to Tdispense process, 96 wells and PFR on cells. Five single CHO cell isolations were independently performed from samples at 2 × 104 cells/mL, as recommended by DispenCell-S1 guidelines. We obtained a Tcc of 2.5 ± 0.5 s, Tdispense process, 96 wells of 5.7 ± 0.8 min and a PFR of 87 ± 3%. These results fitted well to the model linking Tcc, Tdispense process, 96 wells and PFR (Fig. 2A) and confirmed the accuracy of our model for both beads and cells. The recommended concentration of 2 × 104 cells/mL allows the compromise between a low dispensing time and a high plate filling rate.
Second, we assessed the impedance signal reliability for cells. The impedance signal was compared to microscopic observation on a subset of the 96-well plates previously used to estimate Tcc. The impedance signal reliability was 94.6 ± 4.0%.
Because transparent 96-well plates are not optimal for accurate cell observation, dispensing was then performed in black 384-well plates with transparent bottom entirely visible with a 10X magnification. We added a centrifugation step after single-cell isolation and two independent visualizations were performed by different operators at day 0.
We first assessed the reliability of fluorescence microscopy method for the 384-well plates experiments. We performed a 160-well pseudo limiting dilution using DispenCell-S1, with the goal of obtaining around 100 empty wells. By counting the number of ghost wells, which are defined as wells observed as empty at D0 but exhibiting outgrowth at D7, one can estimate the error of microscopic observation. We assumed that all ghost wells only contained one cell missed with the microscope at D0, since it is less likely an observer will miss two events in a well. Dividing the percentage of ghost wells by the outgrowth of each plate, we obtained 0.4 ± 0.8% of error for microscopic observation, which confirms the relevance of microscopy to assess impedance reliability.
In parallel we performed a standard dispense using DispenCell (Fig. 3A,B). This experiment resulted in an impedance signal reliability of 93.4 ± 4.6%. We then tested whether we could set up a rule for choosing the impedance threshold. We delineated a common criterion that would most often lead to the best impedance signal reliability for the dispense of CHO cells: first empty bin above 50 Ω, to account for noise, or if there are no empty bin between the noise population and the cell population, the lowest bin at the minimum between populations. For 96-well plates, we obtained an impedance signal reliability of 95.1 ± 3.3%; for 384-well plates, of 93.3 ± 3.3%. We also observed that for a given experiment, the more possibilities to set the threshold, the lower reliability was (data not shown).
In addition, we designed a test based on best practices from the actual literature [
] to further assess the reliability of the system with regards to usual workflows. To that purpose, we took into account wells outgrowth at day 7, defined as the number of wells containing at least one cell at day 7. We compared the number of wells that were verified single cells at day 0 (one impedance events, one observed cell with the microscope) and exhibited outgrowth at day 7 with the number of wells predicted by DispenCell-S1 as being single cells at day 0 (one impedance event) and exhibiting outgrowth at day 7 (Fig. 3C). This led to a reliability with outgrowth at day 7 of 93.0 ± 3.8% for 96-well plates experiments and 92.3 ± 5.9% for 384-well plates experiments.
Finally, the impact of DispenCell-S1 on clonal outgrowth was assessed, benchmarking single-cell isolation by DispenCell-S1 versus single-cell isolation by limiting dilution, i.e. a gentle isolation method. On D0, plates were observed under the microscope to select wells that contained a single cell. We quantified the number of these wells containing more than 10 cells at D7 and showing more than 25% confluency at D14 (Fig. 3D). Clonal outgrowth was similar for both methods, with 53% of wells for DispenCell-S1 versus 61% for limiting dilution at D7, and 35% versus 42% at D14.
We characterized DispenCell-S1 for automated impedance-based single-cell isolation. We presented the instrument and designed models to predict the performances of a specific single-cell isolation. Using beads, we tested the experimental fit of these models and assessed the different sources of impedance signal reliability error. Finally, we isolated GFP expressing CHO cells to validate DispenCell-S1 performances in an industrially representative situation.
We first introduced Tcc to guide the user during his/her experiment. This parameter gives an insight into the expected dispensing performances through models we built. It also allows the user to set a sample quality criterion prior to perform an experiment with an emphasis on the sample mixing quality. Tcc is intrinsic to each experiment and governed by many factors, among which mixing of the particles in the methylcellulose-based dispensing solution is crucial. Because of the solution high viscosity, obtaining homogeneous cell dispersion is indeed a difficult task. Together with the tip aperture variation, this explains why experiments performed at the same concentration can exhibit variations of Tcc. Experimental values of Tdispense process 96 wells and PFR obtained with beads and CHO cells showed good similarity with the models we developed using Tcc. In particular, single-cell isolation performed using CHO cells indicated that a concentration of 2 × 104 cells/mL was a good compromise between a low dispense time and a high PFR with an average dispensing time of 5.7 min and PFR of 87%. This characterization also allowed us to set a quality control prior to dispensing samples which are often very precious. We recommend a 60 s continuous dispense prior to single-cell isolation to evaluate the Tcc. Isolation shall not be performed if the Tcc is lower than 1 s or higher than 6 s, respectively to avoid a too low PFR or a too long dispensing time. In the future, working on what we called incompressible times, both with hardware and software, would be promising to lower dispensing time while sustaining or even increasing a high PFR.
A crucial aspect of DispenCell-S1 validation was to show that the impedance signal could provide a reliable proof of single-cell isolation. We obtained impedance signal reliability of 99% with beads and ranging from 93 to 95% using CHO cells. For CHO cells, impedance signal reliability with outgrowth at day 7 was in the same range. We showed that reliability error mainly arises from undetermined events which complicate threshold placement. This could explain why reliability is higher with beads, whose population size dispersion is very sharp. Besides, although cell cycle timings strongly depend on the type of cell and conditions, cells can divide between their dispensing and their visualization, leading to a result misinterpretation. This points out the utility of providing proof of clonality at the time of dispense, either with an image or a cell-specific signal (like the impedance signal), and not afterwards. Finally, the impedance reliability difference between beads and CHO cells could be caused by transport events. Indeed, we showed with beads, which are in suspension with TWEEN, that transport events are neglectable, but cells may not be subject to the same inter-particle interactions.
This study therefore highlighted that the correspondence between an impedance event and a given single particle requires a cautious placement of the detection threshold to maximize the reliability. As shown with impedance signal reliability assessment using beads, some undetermined events can lie between the noise and particle populations. Thanks to the impedance signal reliability experiment, we defined a common criterion to set up a detection threshold, delimiting noise or debris and particles. Of course, such a criterion is particle size-dependent and has to be validated on a broader range of experimental conditions. Finally, although not investigated here, being able to set an upper limit threshold could be a great improvement, especially with some types of cells that might be strongly subject to aggregation.
Interestingly, we also noticed that the number of undetermined events seemed to correlate with the impedance signal reliability. This paves the way to setting a quality criterion once dispensing has been performed. One could even imagine computing an impedance event histogram prior dispensing to anticipate the experiment reliability. With regards to our experiments, this criterion should be based on the separation between impedance event population compared to noise and debris distribution.
Altogether, these results show that impedance-based single-cell isolation can be successfully performed in a single step workflow. The impedance signal acts as a report of what has been dispensed. Using existing automated plate imager could increase this single-cell isolation reliability even further [
] and other types of cells, with a size ranging from 7 to 20 µm (Supplementary Table 2). Detection of particles up to 30 µm should be feasible with the same tip, as the only limit is that the aperture might get clogged. The tip topology could also allow the detection of bigger particles, with similar electrical parameters, as the aperture would simply need to be widened by impacting only one step of the fabrication process (i.e. aperture ablation).
Most importantly, we verified that the instrument maintains cell integrity after 7 and 14 days. In fact, clone outgrowth was similar to limiting dilution, that is known to be the most gentle cloning method. DispenCell-S1 can also handle small sample volumes as its design comes with no dead volume. In addition to the fact that it does not compromise cell viability, this would make it particularly well-suited for rare samples, such as circulating tumor cells from cancer patient blood or fetal cells from maternal blood.
As for the next steps, it would be interesting to leverage impedance spectroscopy to investigate cell dielectric properties. Cell membrane capacitance and cytoplasmic conductivity can indeed be measured by impedance spectroscopy [
]. Another technical development in the long-term could be to implement cell sorting capability based on the impedance signal only or even relying on a double readout such as impedance characterization and fluorescence.
DispenCell-S1 provides an innovative and simple impedance-based method for isolation of intact single cells. Its incorporation into an automated platform allowed the increase of dispensing throughput and reliability. The instrument is compatible with a workflow employing standard plates and tips. Because it uses a disposable sensing tip, no cleaning step is required to ensure sterility. DispenCell-S1 is also easily transportable and fits into a biosafety cabinet, so that sterility is easy to maintain. All these attributes make the instrument adequate for cell culture and user-friendly. Finally, DispenCell-S1 design is well-suited for rare cell handling and the impedance spectroscopy technology it uses could be leveraged to investigate other cell characteristics.
CRediT authorship contribution statement
Héloïse Hannart: Methodology, Formal analysis, Writing – original draft. Audrey Berger: Visualization, Methodology, Formal analysis, Writing – original draft. Luc Aeberli: Methodology. David Forchelet: Methodology. Nicolas Uffer: Methodology. Georges Muller: Methodology. Yann Barrandon: Conceptualization. Philippe Renaud: Conceptualization. David Bonzon: Visualization, Methodology, Writing – original draft.
Declaration of Competing Interest
D.B., G.M., N.U., Y.B. and P.R. have financial interests in SEED Biosciences SA and intellectual property described herein.
This study was supported by SEED Biosciences SA and a grant from the Swiss Innovation Agency/Innosuisse (grant 37852.1 IP-LS). The authors would like to thank the Protein Production and Structure Core Facility of Ecole Polytechnique Fédérale de Lausanne for providing with the GFP-expressing polyclonal CHO cells.