Microparticle-enabled single cell multiparameter electronic immunophenotyping for selective electroporation

Abstract

Electroporation (EP) is one of the leading non-viral intracellular delivery methods used in various applications across research and cell therapy development and manufacturing. Currently widely used bulk EP methods, while they offer scalability, cost efficiency and simplicity, cannot be used for targeted or selective delivery to a defined subset of an input cell population. Here, we present a microparticle-enabled selectively permeabilizing impedance cytometer (ME-SPICy), a microfluidic single-cell EP platform that enables targeted EP of selected cell subpopulations based on their surface markers. Antibody conjugated microparticles (MPs) are used to label selected cell subpopulations within a larger heterogeneous sample. In this work, we demonstrate multiple multi-parameter impedance-based sensing schemes that enable real-time discrimination of non-labeled and labeled single cells within a mixed population as they flow through a 3D printed biconical micro-aperture. This allows for the system to analyze in real time if each single cell is a target cell, based on user set parameters, and selectively apply a low voltage (<16 V) for targeted single-cell EP. Simulations and experimental validation demonstrate that MP binding substantially alters cell impedance and phase signature, enabling accurate label-based discrimination. We demonstrated selective EP first using Jurkat cells by targeting either the labeled or non-labeled populations. Then we demonstrated targeted delivery with both dyes and EGFP mRNA in subpopulations of primary cells (peripheral blood mononuclear cells, activated T cells, and naïve T cells). ME-SPICy achieved high precision, with up to 98% purity and >5 fold enrichment in the electroporated cell population. This approach expands the capabilities of EP, offering a promising solution to decrease manufacturing complexity in both research and clinical cell engineering workflows.

Introduction

Engineered cell therapies have emerged as a promising treatment option for many cancers and diseases, with thousands of clinical trials either completed or in progress.¹ A critical step in engineering, or modifying, cells for these therapies is the intracellular delivery of gene editing cargo across the cell membrane. This delivery can be achieved by using chemical, biological, or physical methods.² Chemical methods, such as lipofection, while commonly used in research settings, can result in inconsistent delivery, especially with primary cells.^3–5 Viral vector-based methods are popular due to their high efficiency, often exceeding 90%, but can also have significant limitations.^6,7 These include high costs, lengthy production time for clinical grade vectors, limited cargo payloads, and lingering long-term safety concerns.^6,8

Physical methods to permeate the cell membrane, such as electroporation (EP), minimize many of the above challenges associated with viral methods.⁹ In conventional EP for intracellular delivery, a suspension of cells and cargo is placed between two electrodes, typically spaced a few millimeters apart. A high-voltage pulse (often >1000 V) is applied across these electrodes. This creates an electric field which increases the transmembrane potential (TMP) of the cells, inducing transient pores in the cell membrane and allowing the cargo to enter.^10,11 However, conventional EP methods can experience inconsistent delivery efficiency and reduced post-EP cell viability, especially for heterogeneous primary cells, due to electrolysis and Joule heating from the high voltages required.^11,12

Microfluidic EP systems have emerged as a subtype of electroporation that operate at significantly lower voltages due to their microscale architecture.¹³ These systems offer improved efficiency and viability by minimizing electrolysis.^4,12,14–17 Additionally, they allow for single-cell resolution and can be integrated with fluorescent imaging or electrical measurements for real-time monitoring or precise control of EP parameters.^18–23

Despite these advances in delivery precision, a critical hurdle remains selectivity. Increasingly, a key step in the cell engineering workflow is the need to engineer certain subpopulations of cells, which EP devices cannot deliver to on their own.²⁴ Currently, most biological and clinical cell targeting depends on external cell identification and isolation via optical detection or magnetic separation based on cell surface markers.²⁵ While effective, these methods can be limited by expensive imaging tools, processing costs, and time.^26,27

Microfluidic multifrequency impedance cytometry offers an alternative for targeting subpopulations. As an electrical measurement technique, it enables single-cell phenotyping based on cellular dielectric properties.^28–31 At low measurement frequencies (<1 MHz), the impedance signal correlates with cell size and volume, while at higher frequencies (>1 MHz), it is related to membrane and intracellular properties.²⁹ As such, surface bound microparticles or nanoparticles have shown promise in their use as electronic tags using the ratio of high to low frequency impedance magnitude, or opacity, for affinity-based multiplexed microfluidic sorting of cells.^30,32,33

Here, we present a microparticle enabled selectively permeabilizing impedance cytometer (ME-SPICy) to achieve selective EP of a cell subpopulation within a heterogeneous mixture. By combining single-cell microfluidic EP with multifrequency impedance cytometry we identify and selectively electroporate, in real time, subpopulations of cells from mixtures. This enables targeted EP which is not achievable with conventional EP methods. Cells are flowed through a 3D printed microchannel where we use high and low frequency signals to identify a cell and determine, in real-time, if the cell is labeled with an MP, and thus is a target cell for EP. Once a target cell is identified, a low voltage DC pulse is applied to electroporate the targeted cell. This enables selective EP within cell lines and primary cells, including naïve T cells, for both labeled and non-labeled populations, with propidium iodide and EGFP mRNA. This method expands the capabilities of EP while also being a cost-effective and efficient alternative to viral vectors for targeted intracellular delivery.

Results and discussion

Principle and microfluidic system

A 3D printed biconical vertical micro-aperture is coupled with fluidics and electronics for multifrequency impedance measurements³⁴ and real-time discrimination of non-labeled and MP-labeled cells for selective EP. Cells flow rapidly through the narrowing micro-aperture which concentrates the electric field at a constriction (Fig. 1a). As a cell flows through the first half of the constriction, sinusoidal (AC) sensing voltages at multiple measurement frequencies are simultaneously applied by a lock-in amplifier (LIA) to identify the cell via real-time multifrequency impedance cytometry. Thus, multi-frequency impedance magnitude and phase values for each single cell are obtained as it passes through the micro-aperture, Fig. 1b. These are analyzed, in real-time, on an embedded microcontroller, to discriminate MP-labeled cells which exhibit distinct impedance signatures compared to non-labeled cells based on their opacity (high to low frequency impedance ratio). For more details of this real-time analysis see the section on sensing and a full block diagram of the electronic system in Fig. S1. If a target cell for EP is identified, as determined in real-time by the microcontroller algorithm, the microcontroller triggers an external switch to redirect the signal source at the electrodes from the LIA to a DC voltage supply for EP of the target cell before it leaves the region of the concentrated electric field. Thus, the region of the micro-aperture where the cell is sensed is called the “identification zone” and the region where EP can take place is called the “EP zone”.

Fig. 1.

Schematic and concept of ME-SPICy: (a) overview of ME-SPICy concept, as individual cells flow through a narrowing micro-aperture, cells are identified by a multi-parameter impedance signature measured with a lock in amplifier and analysed in real-time on a microcontroller and if a target cell is identified, the microcontroller switches the voltage applied to the top electrode (b) simultaneous individual data of a single cell across 6 different frequencies for impedance and phase output by the LIA for a single Jurkat cell flowing through the system (c) simulated electric field in the micro-aperture with 15 V applied to the top electrode (d) expanded view of the device with the 3D printed micro-aperture sandwiched between two PCB electrodes which are connected to the cell reservoir and syringe pump. SEM of the top view of the 3D printed micro-aperture is included (e) assembled flow-cell set up.

Here, this system uses a micro-aperture, 3D printed in biocompatible non-conducting resin (see Methods), with the smallest dimension being 25 μm in diameter. The constriction geometry ensures that only one cell passes through the constriction at a time. Additionally, at the narrowest part of the constriction, the cell experiences a high electric field, as evidenced by COMSOL simulations (Fig. 1c) while using a much lower voltage (~15 V) compared to bulk EP (~100–1000 V).

This micro-aperture is sandwiched between printed circuit board (PCB) electrodes with plated-through holes (PTHs) (Fig. 1d and e). These PCBs are then assembled with a cell reservoir and flow connector, to form a flow-cell. As the cells flow through, the rise in aperture impedance due to the cell is measured via the PCB electrodes. Notably, this device does not require any complex multi-step microfabrication process as it is directly 3-D printed from a design file enabling flexibility of aperture shape. It enables the use of low voltages via electric field focusing, making it much simpler, reusable, and lower cost to manufacture compared to standard PDMS channel and microelectrode-based microfluidic EP devices. Typically, PDMS-based microfluidic EP devices need multiple microfabrication steps including microelectrode and mold fabrication, soft lithography and aligned bonding of PDMS on microelectrodes and yet result in devices which are often not reusable due to clogging or electrode damage via corrosion due to electrolysis.

Simulation of effects of MP on cell impedance measurements

To enable selective in-flow EP, it is critical to identify the MP-labeled subpopulation in real-time in our system. Cell subpopulation identification is often based on surface marker expression and is typically achieved using fluorescent labeling which is detected optically. However, previous work has shown that surface markers can also be detected electronically by labeling them with MPs.^30,35 To assess the feasibility of electronic surface marker detection using MPs in our system and to guide the selection of MP size, we performed a series of simulations in COMSOL Multiphysics.

We first developed a model which replicated the experimental baseline and single cell multifrequency impedance measurements in the system (Fig. S2). As shown in Fig. 2a, the maximum impedance change, or where the signal to noise ratio (SNR) is the highest, occurs at the location of the maximum electric field at the narrowest part of the microchannel. This indicates that peak impedance should be used for MP discrimination for maximum SNR, and thus subsequent simulations were focused on the MP-cell interaction at this location. A 6 μm cell was used due the size of lymphocytes, a cell of interest in cell engineering, being in the range of 6–8 μm.³⁶

Fig. 2.

Simulated effects of MPs on the impedance of cells (a) change in impedance of a single cell, 6 μm in diameter, with varying vertical position in the micro-aperture, where y = 0 μm is the narrowest part of the constriction. The electric field along the central axis of the microchannel vs. position is overlayed. (b) Maximum impedance change from baseline detected with different cell sizes at the narrowest part of the constriction. (c) The increase in maximum impedance change of a cell caused by one MP of different diameters directly on top of the cell surface. (d) The increase in maximum impedance change of a cell caused by varying quantities of 5 μm MPs on the cell surface. Orientations of these MPs can be found in Fig. S3. All impedance changes in a–d are at a frequency of 45 kHz. (e) Example orientations of the MPs on a cell as it flows through the microchannel which correlate to the positions in f and g. (f and g) Impedance and phase spectra of 5 μm MP labeled cells at the different orientations shown for a cell 6 μm in diameter.

Initially, we modeled non-labeled cells as conductive spheres of different sizes, each surrounded by a thin non-conducting shell representing the cell membrane. The results indicate that as the cell diameter increases, the maximum change in impedance increases at low frequencies (45 kHz) (Fig. 2b). This is consistent with the Coulter principle which relates the resistance change caused by the presence of a nonconductive sphere in conductive medium in a microchannel, or ΔR, with the sphere diameter, d, diameter of the microchannel, D, and the resistivity of the suspension media, ρ, assuming that D ≫ d:³⁷

$$\Delta R=\frac{4\rho {\text{d}}^3}{\pi D^4}$$ (1)

This cubic relationship shows that the volume of the sphere, or cell, drives changes in low frequency impedance. When a cell is labeled with MPs, they increase its effective volume, leading to a greater ΔR. For our system, assuming the total change in impedance is additive with volume, and disregarding that D changes slightly even in the narrowest part of the microaperture, the total impedance change when MPs label a cell in ME-SPICy can therefore be approximated as:

$$\begin{gathered} \Delta R_{\text{total}}=\sum _{\text{total}\text{MPs}}\left(\Delta R_{\text{MP}}\right)+\Delta R_{\text{cell}} \\ =\sum _{\text{total}\text{MPs}}\left(\frac{4\rho d_{\text{MP}}^{\text{3}}}{\pi D^4}\right)+\frac{4\rho d_{\text{cell}}^{\text{3}}}{\pi D^4} \end{gathered}$$ (2)

Thus, as the impedance change scales with MP volume, detectability can theoretically be improved by using either many small MPs or few large MPs. To evaluate this, we simulated varying diameters (1, 2.5, or 5 μm) and quantities (1–4 MPs) of MPs on cells ranging from 6–12 μm in diameter (Fig. 2c and d). To investigate MP size, a single MP was simulated directly attached to the top pole of the cell. The simulations also indicated that MPs located on the vertical poles of the cell have less of an effect on impedance compared to MPs bound at the perpendicular pole of the cell. This is likely due to the proximity to the highest electric field regions along the microchannel walls (Fig. 1b). Thus, to understand the minimum expected impedance change, MPs were simulated on the vertical pole, Fig. 2c. These results confirm that larger diameter MPs increased the impedance more than smaller diameter MPs. However, since MPs are smaller than cells, the overall impedance change remains cell dominated. This explains why, in Fig. 2c, the larger cells have a smaller increase in impedance change due to the MP constituting a smaller fraction of the total volume.

To isolate the effect of MP quantity at low frequencies, we simulated cells labeled with increasing numbers of the 5 μm MPs, which were identified as the most effective size in Fig. 2c. Fig. 2d demonstrates that impedance change increases with increasing MP quantity. MP positions used in these simulations can be found in Fig. S3. Interestingly, when four MPs are present (with two on the vertical poles and two on perpendicular poles), larger cells experience the greatest impedance change, contrary to the trend in Fig. 2c. This is likely due to edge effects of the electric field, because at the perpendicular poles the MPs and large cells and cells with many MPs interact more with the regions of highest electric field than small cells (Fig. S3).

Overall, however, these simulations revealed that increased presence of MPs on the cell surface, attributed to either larger MP diameters or numerous MPs binding, yielded an increase in low frequency impedance across all cell sizes. These findings guided the selection of 5 μm MPs for experiments. This diameter, based on simulations, provides a significant electronic signature while maintaining high likelihood of multiple MPs binding to a single cell in experiments.

Given the observed edge effects in Fig. 2d causing the approximation involved in eqn (2) to no longer be applicable, as D ≫ d is no longer true, we also characterized different MP locations in the micro-aperture across the full frequency spectrum in the range that our system is capable of, Fig. 2e–g. High frequency impedance has previously been used to electronically discriminate between MP-labeled and non-labeled cells, with less positional dependance.³⁰ Here, 6 μm cells were simulated due to the dramatic effects of MP volume as seen in Fig. 2c and d.

These simulations confirmed that MPs bound on the sides result in a greater change in impedance than those on the vertical poles across all frequencies but also that MP-labeling increases the change in impedance magnitude and phase across the spectrum, regardless of location of the MP. Impedance magnitude is constant at lower frequencies then falls at higher frequencies, while phase steadily increases until approximately 1 MHz. This spectrum shape matches the general impedance signature of cells shown in our previous work, which reflects the resistive and capacitive effects of the cell.^34,38 In impedance cytometry literature, a cell is often approximately modeled as a resistor and capacitor in series.³⁹ At low frequencies, the cell and MP thus act as a non-conductor impeding the flow of current which causes the volume dependent impedance change. However, at higher frequencies, ~1 MHz,⁴⁰ capacitive effects become significant, resulting in lower impedance and increase phase shift so that eqn (1) no longer directly applies (Fig. 2g). These capacitive effects can arise from the non-conducting cell membrane and MP acting as additional dielectric layers where charge can accumulate. This indicates that using the impedance value at not just a single frequency, but at multiple frequencies, allows for the probing not just of the size-based effects of the MP but also of their capacitive effects on the cell impedance to verify their presence on the cell surface. Since experimentally MP orientation may be difficult to control, such multi-frequency measurements can be critical in robustly distinguishing MP-labeled from unlabeled cells.

All together, these simulations show that the increase in low frequency impedance is primarily volume driven while the high frequency increase is also affected by the capacitive effects. Consequently, we hypothesized that the ratio of high to low frequency impedance, also known as opacity, can be used to more reliably distinguish between non-labeled and labeled cells in our system. This has also been used by other groups for similar discrimination.³⁰ Thus, these simulations provided the justification for using large MPs (5 μm) and establish the necessity of multifrequency impedance detection to discriminate between labeled and non-labeled cells for selective EP.

Verification and electronic analysis of MP-labeling of cells

To experimentally validate and quantify MP labeling on cells, Jurkat cells (~10 μm in diameter) were incubated with anti-CD45 conjugated 5 μm MPs and analyzed with imaging cytometry to quantify MP labeling (Fig. 3a). Labeled cells exhibited higher side scatter (SSC) than non-labeled cells, with increased SSC indicating a greater number of MPs bound (Fig. 3b). Based on the SSC and confirmed visually with the images from the imaging cytometer, cells were gated into groups based on the number of MPs bound. 85% of all cells had at least one MP bound, with ~50% of cells having three or more MPs (Fig. 3c). Due to the clear correlation between SSC and MP labeling, flow cytometry was here onwards used for MP labeling verification in further experiments. Flow cytometry data from this imaging cytometry experiment is included in Fig. S4.

Fig. 3.

Verification of Jurkat cells labeling with anti-CD45 conjugated MPs (a) anti-CD45 conjugated MPs bind to CD45 receptors on Jurkat cells. (b) Imaging cytometer verified SSC correlation to the number of MPs bound to cells (c) identification of the number of MPs cells are labeled with (d) bode plots for mean impedance and phase spectra of non-labeled (control) and MP-labeled cells. The bold line is the mean while standard deviations are the highlighted regions. Additionally, the individual single-cell frequency impedance and phase responses which are represented by the bode plots are given, with the top row representing the impedance values and the bottom row the phase, n = 2334 cells (e) LDA separation of populations with and without MPs (f) loadings plot of the LDA coefficients (g) opacity plot of non-labeled and labeled cells, using only 45 kHz and 1.8 MHz.

Next, populations that were MP-labeled (Jurkats + MPs) or had no MPs added (Jurkats only) were independently flowed through ME-SPICy for initial multifrequency impedance characterization. This allowed simultaneous measurement of impedance and phase magnitudes at six frequencies (from 45 kHz to 5 MHz) at the single cell level. This data was then analyzed to find the maximum change in the impedance and phase signal for individual cells at each frequency. Population level results (n > 800 cells) of the maximum changes are shown in the Bode plot Fig. 3d. These results showed significantly higher ( p < 0.05) impedance magnitude and phase across all frequencies for MP-labeled cells (Table S1). Single-cell multifrequency impedance magnitude and phase histogram-based visualization of this data is also shown in the later panels in Fig. 3d. These panels are included here to highlight the high-throughput single-cell nature of these measurements which can be hidden by the Bode plot. These findings were consistent with simulation predictions that MPs would cause an increase in impedance across all frequencies, driven by increased volume and increased capacitive effects. It should be noted that the population mean for the MP-labeled cells in Fig. 3d, is likely an underestimate for the maximum change in impedance due to the presence of the small fraction (~15%) of non-labeled cells in the MP-labeled population.

Linear discriminant analysis (LDA) was then applied to the dataset using the maximum impedance and phase values. Using the full feature set (12 parameters per cell – impedance magnitude and phases at 6 frequencies) achieved a separation accuracy of 0.85 (Fig. 3e). The LDA coefficients, shown in Fig. 3f, indicate the importance and direction of the features in separating the two populations. We see that frequencies between 750 kHz and 1.8 MHz drive discrimination between the populations, likely due to the capacitive effects of the MPs on the cell surface, which simulations showed as well. Due to better electronic signal-to-noise performance, the values at 1.8 MHz were used to plot opacity, or the ratio of high to low frequency impedance magnitude (Fig. 3g). This opacity plot shows that given just the low frequency value there is only a slight increase in impedance magnitude with the MP-labeled population (x-axis, volume driven). However, the opacity shows a clear increase for the MP-labeled population (y-axis). This also establishes that despite lack of control on MP orientation relative to the aperture, the opacity metric can be used to more reliably distinguish MP-labeled cells. Because of this clearer discrimination between populations, only opacity with these two frequencies was used in further experiments to demonstrate selective EP.

Real-time single-cell sensing for opacity calculation and selective EP

Having thus established opacity as an effective metric for MP-labeled single-cell identification in ME-SPICy, we now aimed to selectively electroporate single cells, based on MP-labeling, identified in real-time as they flow through the micro-aperture. This requires both knowing what opacity value would identify a MP-labeled cell, as well as measuring and comparing each individual cell’s opacity against that threshold, in real-time, to determine if EP should occur for that particular cell and then applying the EP pulse on that cell before it flows out of the high E-field EP zone. Thus, for selective EP, first there is a cell characterization stage to set the opacity threshold, as well as then the real-time sensing of each individual cell to implement the threshold which enables selective EP. This requires the use of a real-time sensing and actuation loop. Notably, again, to state this extremely clearly and simply, for the selective EP using ME-SPICy, sensing and EP are done simultaneously in real-time for each single cell as it flows through the micro-aperture and not in separate experiments. This real-time signal processing and actuation scheme is shown in Fig. 4 and described below.

Fig. 4.

Real-time single-cell feedback based sensing and selective EP: (a) the lock in amplifier sends out a multi-frequency signal to the micro-aperture, then demodulates the resulting signal and outputs them as separate signal voltages at individual frequencies. These signals are then high-pass filtered, digitized, and read by a microcontroller for real-time analysis and thresholding to determine if electroporation should occur, based on an opacity threshold. (b) Post-peak detection is triggered at the point when the derivative of the 45 kHz signal crosses 0 (black circle on the derivative signal). At that point the opacity is instantly calculated (red circles) and the decision is made to electroporate and switch to the DC voltage or to continue sensing. (c) Pre-peak detection is triggered at the point when the derivative of the 45 kHz signal passes 8 (black circle on the derivative signal). At that point both the high and low frequency signals are integrated for 200 μs (until the red circles), then the integrated values are used to calculate opacity and the decision is made to electroporate and switch to the DC voltage or to continue sensing.

We now use the LIA to output the measurement signal at the two selected frequencies (45 kHz and 1.8 MHz) to measure the impedance of each cell at two frequencies as it passes through the micro-aperture. The LIA output signals from both measurements are then high-pass filtered, digitized with an analog-to-digital converter, and analyzed in an embedded algorithm by an external microcontroller to calculate the single-cell opacity in real-time. This identifies if a cell is MP-labeled or not (Fig. 4a). For cell characterization, to set the threshold, the opacity of ~2000 cells are first recorded, and the opacity threshold is set at 95% of the non-labeled population to maximize specificity of cell selection. During selective EP, the microcontroller calculates the opacity of each cell as it flows through, comparing it to the set threshold to determine if EP should occur, and based on that, sends an output digital signal to an external switch which controls whether the EP voltage is applied or not. All of this processing, analysis, and switching occurs in less than 5 μs, allowing for real-time sensing and processing of each individual cell as it flows through the system.

If a decision to electroporate is made, the input voltage to the electrodes switches from the AC sensing voltage to a ~15 V DC voltage for a specified EP pulse time, after which it switches back to sensing. If the decision is made to not electroporate, there is no switching from the AC to DC input voltage and instead sensing continues. This entire process is outlined in Fig. 4b and c, where the switching between DC and sensing voltages can be seen. The observed dip in 45 kHz signal post DC voltage application can be largely attributed to the filtering that occurs to the 45 kHz signal, the signal returns to baseline before the next cell is sensed.

To calculate the opacity value, we employ two different real-time analysis methods of the multifrequency impedance signal, either calculating the opacity immediately after the maximum, or peak, of the low frequency impedance signal or immediately before this peak. In both analysis schemes, the microcontroller takes the LIA output voltage and analyzes it in real-time to calculate an averaged impedance signal for each frequency, using a smoothing algorithm to minimize the effects of noise. For the post-peak detection, the first derivative of the 45 kHz signal is taken to identify the peak of the signal, which is at the zero-crossing point of that derivative (Fig. 4b). The 45 kHz signal is used for this due to it having a better SNR than the 1.8 MHz signal. To prevent identification of smaller peaks from noise rather than cells, an additional voltage threshold on the 45 kHz signal itself is also applied so that only derivative crossings above this threshold can identify a peak for analysis. At the point that the peak has been identified, the instantaneous values of the low and high frequencies are analyzed to calculate the opacity to determine whether EP will occur for that cell.

For pre-peak detection, a threshold is set on the derivative of the 45 kHz signal itself, rather than waiting for a zero crossing. At that point both signals are integrated for 200 μs, allowing for better separation of MP-labeled and non-labeled cells compared to the signal taken at just one point pre-peak (Fig. S5). No additional threshold to prevent identification of noise peaks is needed in this case because the threshold on the derivative of the signal is greater than the maximum derivative of a noise peak, as seen in Fig. 4c. Additionally, it can be seen that the time the opacity calculation occurs is ~0.5 ms sooner in the pre-peak detection case compared to post-peak case. Although the details of this depend on specific flow rates used and the thresholds and delays set, this generally allows for much greater time for EP.

The timing of both schemes in the entire context of the time a cell flows through the system can be seen in Fig. 4a, with the time at which EP would occur represented by the dashed lines. Thus, the major advantage of post-peak detection is that the signal is at the point of having the maximum SNR. However, pre-peak detection allows for quicker cell identification to allow for more time for EP. Overall, this is a highly flexible system that can accomplish opacity-based discrimination between MP-labeled and non-labeled cells in multiple ways. As such, additional more complex real-time sensing schemes could also be implemented in the embedded algorithm to further maximize sensitivity and time in the electroporation zone simultaneously.

Selective EP of non-labeled and MP-labeled Jurkat cells

Next, a model selective EP experiment was designed with a mixture of MP-labeled and non-labeled Jurkat cells, to target a specific population for EP as shown in Fig. 5. Here, post-peak detection is utilized to determine cell opacity and trigger EP. As introduced above, the experimental workflow begins with a rapid calibration step to establish the electrical opacity threshold used for the experiment. This step is completed in less than 5 minutes, allowing the system to adapt to the unique impedance signatures of any given cell population, ensuring accurate real-time discrimination within ME-SPICy.

Fig. 5.

Selective electroporation of labeled and non-labeled Jurkat cells. (a) The multifrequency opacity threshold, red line, set for targeting either non-labeled or labeled cells, n = 2000 cells. (b) Scheme for selecting non-labeled cells as the target cell. (c) Scheme for selecting labeled cells as the target cell. (d) Flow cytometry plots of PI delivery in non-labeled and MP-labeled populations. (e) Non-labeled and labeled percents of the population of cells that were delivered to and alive. All data are from N = 2 technical repeats. Error bars shown are standard deviations. (f) Ratio of the number of non-labeled cells to the number of labeled cells within the population of cells that are live and electroporated with and without selection. N = 2 trials, where number of cells is between 2000–8000 cells each trial separately.

For this calibration step, both populations were first independently flowed through ME-SPICy, using the microcontroller to record the opacity calculated with post-peak detection on 1.8 MHz and 45 kHz signals (Fig. 5a). The numerical range of opacity values is different here than in Fig. 3g due to increased amplitudes used now for the 1.8 MHz and 45 kHz signals output by the LIA. The EP threshold was subsequently set at 95% of the opacity of the non-labeled population to maximize specificity. Labeled and non-labeled cell populations were then mixed for EP to ensure both populations would be present in the final mixture of cells used for demonstrating selective delivery. Note that unbound MPs were also present in this mixture. However, their low frequency impedance was significantly lower than that of the cells so they were electronically filtered out by the microcontroller algorithm to avoid unnecessary switching to the EP voltage which could cause spurious EP or electrolysis.

Selective EP was then tested by either selecting the non-labeled population (Fig. 5b) or the labeled population (Fig. 5c) as the target population for EP. Propidium iodide (PI) was delivered and delivery efficiency was defined as the percentage of live and electroporated (PI+) cells out of live cells only within the respective non-labeled or labeled populations. Flow cytometry was used to analyze viability, delivery efficiency, and to identify the labeled cells (Fig. 5d). For both schemes, it was found that >80% of the electroporated population was on target, with selecting non-labeled cells yielding ~95% purity of the target population in the electroporated population of cells (Fig. 5e). Here, we define purity to be the percent of on target cells that were delivered to (in this case, PI+) and are alive out of all the live cells that were delivered to, represented by the percent of the population of cells that is non-labeled or labeled. Additionally, comparing the initial ratio of target to non-target cells with the final ratio of electroporated cells revealed a >10-fold enrichment of the target population when selecting non-labeled cells and ~4-fold enrichment when targeting labeled cells (Fig. 5f). Cell viability remained high, >75% across the selection trials. Both cell viability and delivery efficiency are shown in Fig. S6. Intriguingly, delivery efficiency was found to be higher when selecting for non-labeled cells than when selecting for MP-labeled cells, likely causing the lower enrichment factor when selecting labeled cells. To improve this efficiency, the opacity threshold could be optimized for, however this could be at the cost of purity in the electroporated sample representing a typical yield vs. purity trade-off in any selection method. Note that each of the flow cytometry results shown here involve characterizing >1000 cells each in an unbiased manner (Fig. 5d). These are the results that are then used to derive the purity and enrichment results (Fig. 5e and f). This can be contrasted with selectively showing a few imaging fields of view in microscopy of cells with ~10–20 cells at most each, which can easily involve experimenter bias in selection of which fields of view to select for imaging.

Selective EP of primary cells

Having demonstrated selective EP with a cell line, ME-SPICy was next used to selectively electroporate primary cells and perform mRNA delivery, as shown in Fig. 6. Given the growing demand for lymphocyte-based cell therapies, such as NK and CAR-T cell therapies, we aimed to selectively EP a range of cells to show the versatility of this system, including non-labeled lymphocytes in primary human peripheral blood mononuclear cells (PBMC), MP-labeled and non-labeled activated T cells, and MP-labeled naïve T cells.

Fig. 6.

Selective electroporation of non-labeled lymphocytes from PBMC (a) PBMC are labeled with anti-CD14 conjugated MPs. Proportions of lymphocytes and monocytes in the labeled and non-labeled populations. (b) The multifrequency opacity threshold, red line, set for targeting non-labeled lymphocytes. (c) Non-labeled and labeled percents of the population of cells that were delivered to (PI+) and alive. n of cells is between 830 and 1800 for all trials. (d) Ratio of the number of non-labeled (target) cells to the number of labeled cells within the population of cells that are live and were electroporated with and without selection. (e) Flow cytometry plots of EGFP mRNA expression in both non-labeled and labeled subpopulations of activated T cells labeled with anti-CD45 MPs, control was incubated with mRNA but not electroporated (f) EFGP mRNA expression in activated T cells, selecting for either the non-labeled or the labeled population. n of cells is between 1000 and 3800 for all trials (g) ratio of the number of target to non-target cells (depending on the selection case) within the population of cells that are live and were electroporated without selection. (h) Normalized viability and mRNA expression for naïve T cell selection of MP-labeled cells, using pre-peak detection. n of cells is between 7000 and 55 000 for all trials. All data are from N = 2 technical repeats. Error bars shown are standard deviations.

Initially we aimed to target non-labeled lymphocytes within PBMCs, due to the additional advantage of monocytes and lymphocytes being different size as wells. PBMCs contain lymphocytes (NK, B, and T cells which are ~7 μm in size) and monocytes (which are ~14 μm in size). Of the batch of PBMC used, ~38% were monocytes, as determined by flow cytometry (Fig. S7). To optimize for purity of the non-labeled fraction, monocytes were labeled via anti-CD14 conjugated MPs, resulting in a mixed population of non-labeled lymphocytes and MP-labeled monocytes. Flow cytometry confirmed that when MPs were added to PBMC, ~80% of the monocytes are labeled, leaving the non-labeled target population to primarily consist of lymphocytes (Fig. 6a). Of the cells exposed to MPs, the labeled population, on the other hand, does consist of ~40% lymphocytes, accounting for 25% of the total lymphocyte population. While this MP-labeling could be optimized further, via changing the antibody concentration, the bead-to-cell ratio, or the kind of MP used, the 90% purity of lymphocytes involve the non-labeled population was considered sufficient to proceed with selective EP of non-labeled cells.

For selective EP of lymphocytes, the cells were first characterized to determine the threshold for selection, which was set to include 95% of non-labeled PBMC (Fig. 6b). Below this threshold, non-labeled cells, which are primarily lymphocytes, were selectively electroporated, while cells above the threshold were identified as MP-labeled monocytes and were not electroporated. As was observed previously, there is a clear shift in the opacity of MP-labeled primary cells (Fig. 6b), with a high opacity population emerging with the addition of MPs, similar to what was seen with MP-labeled Jurkat cells. The full frequency spectrum of the measured single cell impedances can be found in Fig. S8. PI was delivered to the non-labeled lymphocyte population. Similar to the Jurkat cells, a purity of ~95% was achieved in the electroporated cells, meaning that 95% of the cells that were electroporated were on target, Fig. 6d. This enabled the ratio of non-labeled cells to labeled cells to increase by factor of ~5 within the electroporated population (Fig. 6e). Viability and efficiency were similar to what was accomplished for Jurkat electroporation and can be found in Fig. S9. Again, note that each of these results are derived from flow cytometry results shown here involve characterizing >1000 cells each in an unbiased manner.

To further test ME-SPICy with other cells and cargos of clinical interest we selectively electroporated both activated and naïve primary T cells with EGFP mRNA. Furthermore, to display the flexibility of the system, both the pre- and post-peak detection schemes described above were utilized here. For these demonstrations mixtures of CD-45 MP-labeled and non-labeled populations were used. Electroporated cells were placed back in culture to allow time for GFP expression and analyzed 24 hours after selective EP was performed.

For activated T cells, post-peak detection was utilized for selection of either the non-labeled or labeled populations. The setting of opacity threshold can be seen in Fig. S10. In either case, ~40% of the target population was EGFP+ after 24 hours, as compared to <15% of the non-target population, demonstrating selection delivery of mRNA to either population (Fig. 6e and f). Similar to selection of lymphocytes within PBMCs, a maximum of ~5-fold increase in the ratio of target to non-target cells with non-labeled selection, Fig. 6g. The select non-labeled condition had much higher enrichment due to there being more non-labeled cells than labeled cells in this mixed population. The viability of the activated T cells post-EP is shown in Fig. S11.

With naïve or non-activated T cells, which are known to be difficult to transfect,⁴¹ pre-peak detection was utilized to highlight the other operating mode for selection as well. Only the select-labeled experiment was conducted here, due to the larger overlap between the non-labeled and MP-labeled cells (Fig. S5). Future further optimization of MP-labeling would alleviate this issue. Viability of the electroporated population remained high in both the target and non-target populations after 24 hours and selection accomplished ~20% EGFP expression in naïve T cells, Fig. 6h.

Across cell types and cargos, we did observe some drop in viability for MP-labeled cells when they are electroporated with the same parameters as non-labeled cells. This occurs regardless of selection (Fig. S6, S9 and S11). This could be due to steric effects from MPs on the cell, or effects on the localized electric field on the cell surface. Given the wide parameter space for the EP voltage, such as number of pulses, pulse length, and amplitude, there is considerable room for further future optimization to improve both delivery efficiency and viability, especially for naïve T cells, which are known to be difficult to transfect and can suffer from lower viability upon electroporation.⁴¹

A known limitation of many single-cell microfluidic techniques is their throughput. Although ME-SPICy can currently process up to 0.5–1 million cells per hour, this remains lower than other bulk or macro-fluidic EP approaches.^14,42 Throughput could be increased in the future in a number of different ways. It can be increased by decreasing the time spent in the identification zone and electroporation zones for each single cell. Decreasing identification time could be accomplished by using a faster flow rate, higher cell concentrations, or different identification algorithms, as evidenced here by the use of pre-peak rather than post-peak detection. Decreasing the time in the electroporation zone could significantly improve throughput as short EP pulses, on the scale of 1–100 μs, rather than the 1 ms used here, have been shown to be effective for gene delivery.^43,44 Finally, multiaperture designs can be used to further scale up throughput. Here, a significant advantage of the ME-SPICy approach is that it is inherently easier to scale due to the flexibility and reusability provided by the way it is designed and fabricated, as described earlier, which could allow for ease of parallelization with multiple microchannels to increase throughput. Impedance detection and selective EP electronics would also need scaling to multi-channel sensing and EP for this, which is achievable either using board-level custom electronics or via application-specific integrated circuits (ASIC).

With further optimization of MP labeling, ME-SPICy could also be used to target specific lymphocyte subpopulations such as NK cells, CD8⁺ cytotoxic T cells, or CD4⁺ helper T cells. Beyond lymphocytes, this strategy could also be adapted for stem cells or other clinically relevant rare populations. Although current selection relies on MP labeling, future integration of cleavable antibodies, shear-removable MPs, or magnetic separation could facilitate MP removal post-electroporation.⁴⁵ We also note that while EP coupled with downstream label-based sorting can accomplish similar selection goals, ME-SPICy simplifies such workflows and enables additional functionalities through real-time single-cell impedance measurement. For example, the platform could be adapted for selective killing, applying high-voltage irreversible EP to eliminate unwanted cells, such as cancerous B cells in CAR-T manufacturing or even circulating tumor cells (CTCs).⁴⁶ Furthermore, the impedance-based feedback control in the system could enable dynamic EP voltage modulation, adjusting pulse strength based on cell size or other properties, reaching a threshold that relates to permeability, thereby improving delivery precision.²⁰

Conclusions

We developed an integrated microfluidic EP platform which couples multi-frequency impedance cytometry, real-time analysis of single-cell impedances and transient EP triggering to achieve high-purity selective EP of labeled subpopulations. MP-labeling, coupled to single-cell opacity measurements, was found to be a robust method to convert surface marker binding to a distinct electronic signature, despite heterogeneity in MP counts and orientation on the cell surface. The overall method was demonstrated with both immortalized cell lines and human primary cells for small molecule and mRNA delivery. To the best of our knowledge this is the first demonstration of multi-parameter single-cell feedback-controlled EP as well as surfacemarker selective EP-mediated nucleic acid delivery. ME-SPICy thus adds a new dimension to EP, which at the commonly used bulk scale, lacks selection or targeting ability. It has thus the potential to eliminate the need for cell sorting for workflows that use EP, such as cell engineering, reducing costs from reagents and equipment, and simplifying the cell manufacturing process. Moreover, the performance of ME-SPICy is on par with conventional bulk EP in terms of delivery efficiency, with potential for further optimization to increase efficiency and throughput, including via parallelizing apertures and electronics, broadening its clinical applicability. This microfluidic platform also allows for flexibility and integration into existing manufacturing workflows, offering an alternative to traditional viral methods with additional benefits for integration. This positions ME-SPICy as a transformative tool for both research and clinical applications in cell engineering.

Experimental methods

Finite-element modeling and simulations

Finite element analysis was performed in COMSOL Multiphysics 5.6 using the AC/DC and electrical circuit modules. The electrical circuit module was used to couple circuit elements to match the baseline experimental impedance of the device. Cells were modeled as spheres of conductive media, σ = 1.1 S m⁻¹, using the contact impedance feature to model the cell membrane as a thin non conducting layer. Microparticles were modeled as solid non conducting spheres.

Device fabrication

ME-SPICy consists of a narrowing vertical micro-aperture which is 3D-printed using a Nanoscribe Photonic Professional GT2 and sandwiched, using epoxy, between 2 PCBs with gold-coated copper electrodes. Further details of it can be found in previous work.^34,38

Cell Culture

Jurkat cells (Clone E6-1, Fisher Scientific) were cultured in RPMI-1640 medium (ThermoFisher, A1049101) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. Activated T cells and naïve T cells were cultured in RPMI-1640 supplemented with IL-2 (200 units per mL, Sino Bio 11848-HNAE) and IL-7 (10 ng mL⁻¹, Sino Bio 11821-HNAE). Cells were passaged every 3–4 days and kept in a humidified incubator at 37 °C. Human PBMC, purchased from STEMCELL Technologies (70025), were aliquoted, refrozen, and kept in liquid nitrogen. Human T cells were isolated from buffy coat (research blood components), using an isolation kit (STEMCELL 15061), and frozen and kept in liquid nitrogen. For experiments, Jurkat cells (from culture), PBMC (after thawing for 10 minutes), activated T cells (thawed for 10 minutes, washed, activated (ThermoFisher 11161D), and placed in culture for a week), or naïve T cells (thawed and placed in culture for 24 hours), were washed with PBS for 5 minutes at 400 g. For all experiments, cells were then resuspended at 3 × 10⁶ cells per mL of delivery buffer (90% Opti-MEM (Thermofisher, 31985070), 10% OptiPrep (STEMCELL Technologies, 07820)). Before MP-labeling, PBMC were resuspended at 3 × 10⁶ cells per mL in PBS 1% BSA and then were stained with anti-CD3, anti-CD19, and anti-CD56 for 30 minutes at room temperature before being washed in PBS with 1% BSA for verification of cell subpopulations. Similarly, due to the cells going back into culture post-EP for mRNA experiments with the excess MPs present with the cells, non-MP-labeled activated or naïve T cells were stained with anti-CD45 PE for later verification of cell populations. For microparticle labeling itself, cells were exposed to MPs in PBS with 1% BSA and then washed and resuspended in Opti-MEM. For mRNA experiments cells were not resuspended with BSA. A Countess cell counter was used for cell counting and sizing.

Microparticle labeling

5.7 μm polystyrene streptavidin functionalized MPs were obtained from Spherotech (SVP-50). MPs were washed at 3000 g for 3 minutes with PBS 1% BSA and incubated for 30 minutes at room temperature on a shaker with anti-CD45 biotin (ThermoFisher, 13-0459-82) for labeling Jurkats or T cells, or with anti-CD14 for PBMC (ThermoFisher, 13-0149-82) for labeling monocytes. Across all MP-labeling, 2.5 μgof antibody was added for 1 × 10⁵ beads. The MPs were then washed twice before adding cells suspended in PBS with 1% BSA at a concentration of 3 × 10⁶ cells per mL, at a ratio of ~;40 MPs per cell. The cell and MP mixture was then incubated for 30 minutes at room temperature on a shaker. Initial verification of MP binding was achieved by using an Amnis MKII imaging cytometer and a Cytek Aurora Flow Cytometer.

Impedance measurements for machine learning

A lock-in-amplifier and transimpedance amplifier (LIA) (HF2LI, HF2TA, Zurich Instruments) were used to detect cells as they flow through the aperture, only interfacing ME-SPICy to the LIA with no additional circuit elements. Impedance and phase were recorded at 6 frequencies, ranging from 45 kHz to 5 MHz at 1 V each. Cells were pulled through the system at 1 μL per minute via a syringe pump, Chemyx F200X. A peak finding algorithm is then applied to this data to identify the cell peaks at the lowest frequency, filtering out noise and unbound MP. The time index at which this peak is identified is then used to extract the remaining change in impedance and phase measurements at other frequencies. Linear discriminant analysis (LDA) was then used to reduce the dimensionality of the data to identify the two most relevant frequencies that would be used for EP experiments. The EP circuit is external to the LIA but uses the LIA auxiliary outputs which relate to 1 selected frequency each. The training data for the LDA model was datasets of the peak impedance and phase values at the different frequencies for non-labeled cells and MP-labeled cells. The data was standardized and split so that 70% was used for training and 30% for testing.

Selective EP control mechanism

A 3 V 45 kHz and a 4.5 V 1.8 MHz sine wave were applied to the top PCB electrode by the LIA for impedance detection. The lock-in amplifier then internally demodulates and amplifies the signal 300 times and removes the DC offset. The signals from the individual frequencies are then output by the LIA, through a 1st order high pass filter ( f_c = 70 Hz), to an analog to digital converter, AD7366 (analog devices), which communicates to a Teensy 4.1 microcontroller at 220 000 samples per second. The microcontroller is constantly running an opacity calculating algorithm on the moving average of the signal to detect cells in the micro-aperture, filtering out noise and any MPs not attached to cells, and calculate if the ratio of high to low frequency impedance (opacity) falls above or below the opacity threshold. The microcontroller sends a signal to an external switch (TMUX8109) which switches the voltage applied to the top electrode from the detection voltage to a 15.5 V DC EP voltage for 1 ms before switching back to the detection voltage. If a cell is not within the target opacity range, there is no switching.

Selective EP

PBS with 1% BSA was flowed through the aperture prior to beginning experiments for 15 minutes. The impedance data of unbound cells and labeled cells in delivery buffer were individually recorded before beginning EP experiments to determine where the opacity range should be set for selective EP of the target cells. The bounds of this range were set to include 95% of the target population. 95% was selected to mimic traditional gating strategies. For each PI trial, cells were diluted to 1.5 × 10⁶ cells per mL with delivery buffer that contained 130 μg mL⁻¹ propidium iodide (ThermoFisher, P1304MP) for a final PI concentration of 65 μg mL⁻¹. For mRNA experiments, 25 μg of CleanCap EGFP mRNA was added to 50 μL of cells in delivery buffer (Trilink Bio Technologies L7601100). 50 μL of cells were then pipetted into the reservoir and flowed through the system at 0.5 μL per minute for 20 minutes. For PI delivery, post EP, cells were immediately removed from the system, stained with 1 : 50 volume ratio of 50 μM Calcein Blue AM for viability, and analyzed on an Aurora Cytek Flow Cytometer. For mRNA delivery, cells were washed immediately post EP and plated with supplemented RPMI for 24 hours. Immunofluorescent staining was used to confirm cell subpopulations, and all data was analyzed in FlowJo and visualized in GraphPad Prism. PI+ or EGFP+ classification was set above the fluorescence of 95% of cells that were incubated with the cargo but were not electroporated. Cell viability was calculated as the number of live cells normalized to nonelectroporated controls, accounting for pre-EP viability that was ~80%. Bulk electroporation was performed with a Bio-Rad Gene Pulser XCell with a 350 V, 1 ms, square pulse in a 2 mm cuvette, with the same cell and cargo concentrations used above. Similarly, cells were washed and plated for 24 hours.

Data availability

The data supporting this article have been included as part of the supplementary information (SI).

Supplementary information is available. See DOI: https://doi.org/10.1039/d6lc00180g.