Tobias
Gerling‡
ab,
Neus
Godino‡
a,
Felix
Pfisterer
a,
Nina
Hupf
a and
Michael
Kirschbaum
*a
aFraunhofer Institute for Cell Therapy and Immunology, Branch Bioanalytics and Bioprocesses IZI-BB, Am Muehlenberg 13, 14476 Potsdam, Germany. E-mail: michael.kirschbaum@izi-bb.fraunhofer.de
bInstitute of Biochemistry and Biology, University of Potsdam, Karl-Liebknecht-Str. 24/25, 14476 Potsdam, Germany
First published on 7th June 2023
Continuous flow cell sorting based on image analysis is a powerful concept that exploits spatially-resolved features in cells, such as subcellular protein localisation or cell and organelle morphology, to isolate highly specialised cell types that were previously inaccessible to biomedical research, biotechnology, and medicine. Recently, sorting protocols have been proposed that achieve impressive throughput by combining ultra-high flow rates with sophisticated imaging and data processing protocols. However, moderate image quality and high complex experimental setups still prevent the full potential of image-activated cell sorting from being a general-purpose tool. Here, we present a new low-complexity microfluidic approach based on high numerical aperture wide-field microscopy and precise dielectrophoretic cell handling. It provides high-quality images with unprecedented resolution in image-activated cell sorting (i.e., 216 nm). In addition, it also allows long image processing times of several hundred milliseconds for thorough image analysis, while ensuring reliable and low-loss cell processing. Using our approach, we sorted live T cells based on subcellular localisation of fluorescence signals and demonstrated that purities above 80% are possible while targeting maximum yields and sample volume throughputs in the range of μl min−1. We were able to recover 85% of the target cells analysed. Finally, we ensure and quantify the full vitality of the sorted cells cultivating the cells for a period of time and through colorimetric viability tests.
To overcome this limitation, various high-impact works have been published in the recent years, combining sorting with bright field and fluorescence 2D-imaging data and thus being able to make spatial cues in cells (e.g., subcellular localisation of fluorescently-labelled markers, protein co-localisation etc.) accessible for cell classification and sorting.12,22–28
In image-activated flow sorting, the correct alignment of the cells in the central focal plane is essential to achieve high image quality and high control over cell movement. For this purpose, the width of the sample stream is usually confined to a few μm, either by 3D sheath flows22,28 or small constriction channels.24 Nevertheless, to achieve a high sample volume throughput (i.e., at least in the range of μl min−1), high flow velocities of up to 1 m s−1 must be applied.
At such high flow velocities, imaging of faint fluorescence signals, data processing and sorting have to take place within microseconds, which is technically enormously demanding. As a result, the proposed systems are complex, require elaborate imaging-, flow control- and data management mechanisms and are still limited in the achievable image quality and depth of analysis.22,23,26,28
In cell biology, however, spatial information with nm precision can be crucial. This is the case, for example, when analysing subcellular structures such as mitochondria or cytoskeleton or the close interaction of proteins and their localisation within the cell.29,30 In this respect, fluorescence microscopy via direct imaging offers a good compromise between high image quality, high resolution and low system complexity. However, to use this widely used technique for blur-free imaging under flow, it is necessary to work with flow velocities several orders of magnitude lower than in the studies mentioned above, as the acquisition of fluorescence images with commercially available cameras offers only limited frame rates and requires relatively long integration times (i.e., in the sub-ms range).
In this paper, we present a low-complexity microfluidic approach that enables flow-through image-based cell analysis and precise sorting based on standard high numerical aperture wide field microscopy, achieving a level of image quality comparable to steady state conditions. To be able to work at the low flow velocities required by the image acquisition and still achieve both a reasonable sample throughput in the μl min−1 range and a precise control of the cell position in the channel, we focus the cells in the microchannel by the well-established concept of dielectrophoresis31–36 instead of confining the sample flow.
Besides focusing, we also employ dielectrophoretic forces to sort the cells. Taking advantage of the high precision of microelectrode fabrication, we are able to reduce the effective length of the sorting window (i.e., the minimum distance between two consecutive cells to be sorted independently) almost down to the size of a single cell. This allows us to achieve high purity and yield even for very dense cell samples, the latter being a prerequisite for processing an acceptable number of cells in a reasonable amount of time even at low flow velocity.
Thanks to the simple and robust concept of our sorting protocol, no external instruments other than microfluidic pumps and a multichannel signal generator are required to control the cell flow in our system. Furthermore, the frame rates and associated amount of data to be processed are comparably small, so that a standard Windows PC is sufficient to operate the whole system.
We demonstrated the ability of our low-complexity image-activated cell sorter to sort out cells from heterogeneous cell samples based on image processing (i.e., extraction of pure colour information or the spatial distribution of subcellular fluorescence signals). By maximising sorting yields at the expense of cell throughput, we achieved target cell purities between 80% and 95% at yields around 85%.
Besides the performances of the sorting, maintaining cell health and viability is a fundamental requirement for any cell sorter. However, the high shear stress to which the cells are subjected when moving through very small channels at high speed, as well as cell-stressing sheath buffer compositions are often a limiting factor to achieve desirable vitality levels. We do not use sheath buffers, so the cells remain in their preferred medium throughout the complete sorting process as dielectrophoretic cell handling in physiological medium is still possible when using negative dielectrophoresis.37,38 To demonstrate the high biocompatibility of our gentle and low-shear rate process, we show and quantify the vitality of the sorted cells through both colorimetric viability assays and continuation of the processed cells in cell culture.
![]() | ||
Fig. 1 Schematic overview of the microfluidic system and the dielectrophoretic sorting principle (not to scale). a, Top view of the microchannel. Top and bottom glass slides of the microchannel are equipped with congruent microelectrodes (black lines) which cells can initially pass freely unless an electric field is applied between them (1 MHz, 3 to 4.5 V, depending on the electrode pair). In this case, the electrode segment acts as a barrier for the cells (flowing at 580 μm s−1 or 40 μl h−1), deflecting them from their path through the channel. A cell sample containing target and non-target cells (depicted in red and green, respectively) is introduced into the microfluidic channel through the sample inlet. Buffer from a second inlet ensures that the opposing half of the microchannel and target cell outlet are not accidently contaminated with cell debris from the sample inlet. Electrode pairs E1 and E2 focus all cells into a pearl chain configuration. A CMOS camera continuously captures images of the region of interest (ROI), which are analysed by a standard PC for the presence and velocity of target cells. On this basis, an activation pattern of the sorting electrode array E3 is generated and transmitted to the electrodes via a signal generator. After the sorting step (details in b), target and non-target cells are further separated by electrode pair E4, forwarded to the respective outlets and recovered from the system. b, Time series of enlarged schematic of the ROI and sorting electrode array E3 (red frame) for channel top view. t1: Target and non-target cells arrive one after the other at the ROI. The system automatically detects target cells and tracks their position. t2: The position of the detected target cell is tracked continuously across multiple frames and its velocity is calculated. Non-target cells pass the eight individual electrodes of the sorting electrode array which remain inactive. t3: Based on the calculated target cell velocity, the targeted activation of individual sorting electrodes ensures that each target cell is deflected. t4: The subsequent electrodes are activated according to the flow velocity of the target cell to achieve complete separation of the cells. c, Depiction of different force effects on cells flowing in the microchannel with active electrodes (channel side view). In addition to the hydrodynamic force FH exerted by the fluid flow and driving the cells through the channel, the gravitational force FG and the buoyant force FB act with constant strength. The dielectrophoretic force FDEP acts towards the minimum of the gradient of the electric field generated by the electrodes, which is located in the vertical centre of the channel in front of the electrodes. If the field strength is sufficient, FH and horizontal component of FDEP are in equilibrium, preventing cells from passing through the electrodes. At the same time, an equilibrium is established between FB, FG and the vertical component of FDEP, whereby the cells arrange in a common vertical plane. For information on electric field distribution between active electrodes, please see ref. 37. |
Pressure-driven flow is used to hydrodynamically drive the cells through the channel while several platinum electrodes extending in different geometries along the channel deflect the cells across flow lines by dielectrophoretic forces. The electrodes are arranged in pairs on the top and bottom inner faces of the microchannel and were supplied with square-wave radio frequency electric signals between 3 V and 4.5 V peak-to-peak by custom-made signal generators. Cells brought by the flow into the immediate vicinity of the electrodes are polarised and experience a force directed to the minimum of the electric field gradient (i.e., dielectrophoresis). Due to the pairwise arrangement of the electrodes at top and bottom of the channel, the resulting force vector has both a horizontal and a vertical component directed towards the central plane of the microchannel39,40 (Fig. 1c). For details on dielectrophoresis and the function of the microfluidic chips see ref. 37, 38 and 41.
The different electrode pairs along the channel (see also ESI,† Fig. S1) serve different purposes. At the beginning of the channel, long transversing electrodes (E1 and E2 in Fig. 1a) focus cells into a streamline. The subsequent sorting array (E3 in Fig. 1a) consists of eight consecutive short electrodes that are operated sequentially (Fig. 1b). The length of the individual electrodes has a direct impact on the precision of sorting target cells, as it determines the effective length of the sorting window. Due to the short length of only 50 μm of the individual electrode segments, only a small cell spacing is required for efficient sorting of target cells, which enables the processing of high cell densities (see ESI†).
Finally, diverging electrodes (E4 in Fig. 1a) further separate target and non-target cells and guide them towards different channel outlets. Here, the target cells are forwarded to a rinsing flow, flushed out of the system and collected into standard cell culture equipment.
The analysis of the cells is based on a series of high-quality images that are acquired for each cell in the region of interest directly in front of the sorting array. The velocities of the cells are calculated after tracking them according to the central coordinates of the cells in several consecutive frames (Fig. 1b). In the case of a target cell, the computer switches sequentially individual sorting electrodes according to the expected position for each target cell.
![]() | ||
Fig. 2 Frequency distribution of cell velocity and vertical alignment of dielectrophoretically guided cells compared to free-flowing cells. While flowing through the microchannel, cells were either dielectrophoretically guided (i.e. electrodes active) or free flowing (i.e. electrodes inactive). Images of the cells were taken and their velocity determined as they appeared in the ROI. a: Frequency distribution of velocity of dielectrophoretically guided (blue) and free-flowing cells (orange). Free-flowing cells exhibited a much broader velocity distribution than dielectrophoretically guided cells, due to the random vertical distribution of the cells in the microchannel and the associated different flow regimes of the parabolic flow profile in which they were located. In contrast, the dielectrophoretically guided cells experienced a vertical force that directed them into a common vertical plane and flow regime, equalising their velocities. b and c: Exemplary images of free-flowing (b) and dielectrophoretically guided cells (c). The different vertical positions of free-flowing cells were also reflected by the highly variable image sharpness. In contrast, the dielectrophoretically-guided cells were perfectly aligned in focus. The numbers in white represent the measured velocity of each cell in μm s−1. Scale bar, 20 μm. A larger overview can be found in the ESI† in Fig. S3. |
![]() | ||
Fig. 3 Evaluation of the colour-based sorting performance of the system. a, The achieved target cell purity as a function of cell throughput for sortings at 60× magnification. Experimental results are depicted in blue while the orange line indicates ideal results based on Poisson's statistics (see ESI† for theoretical description). b, Representative images of a cell sample initially containing 10% of red cytosol-stained cells and 90% of green cytosol-stained cells before (left) and after sorting red cytosol-stained cells as target cells (right) at a throughput of 12![]() |
To investigate the cell recovery rate and the precision of the sorting process in more detail, we repeated the sorting of red and green cytosol-stained cells as described above but at a lower magnification (i.e., 4×), enabling the observation of the whole sorting array during operation. A corresponding sorting sequence is shown in ESI† Video S1.
Counting of target cells in the recorded videos of the “yield sorting” experiments at low magnification revealed that our system correctly identified and sorted ca. 96% of the total analysed red target cells into the target outlet. We also compared the actual amount of target cells recollected from the device with the number of target cells counted in the recorded videos before sorting at 60× and at 4× magnification. On average, 85% of all analysed target cells were successfully sorted and recollected from the system with a standard deviation of only 5% over all sorting experiments (see ESI† Table S1). Both purity and recovery rate of the sorted sample kept constant along the whole experimental time, as well as in between independent repetitions. Across all experiments, the system was stable and did not require any external intervention for the entire duration of the experiment, which in some cases lasted up to 4 hours.
![]() | ||
Fig. 4 Image analysis. a, Exemplary images of cytosol- (first row) and membrane-stained cells (second row) flowing in the microchannel at approximately 580 μm s−1. The micrographs were taken at 60× magnification and 0.5 ms exposure time. Scale bar, 10 μm. b and c, Basics for classification of cytosol- and membrane-stained cells, respectively. For each cell, sixteen radially equally distributed brightness profiles across the outer cell border were collected and each normalised for brightness and profile length, respectively. Cytosol-stained cells usually show brightness profiles with gradually increasing values towards the cell centre, as depicted in (b). However, most of the membrane-stained cells show a large number of brightness profiles with a maximum value in the outer region and decreasing values towards the cell centre, which is associated with a higher concentration of fluorescence in the membrane (c). Profiles that fall below the threshold of 0.9 after reaching the maximum value are defined as M-type profiles (marked in orange), while profiles that do not fall below this threshold are defined as C-type profiles (marked in blue). After testing various classification criteria, those cells were classified as membrane-stained that had at least three or more M-type profiles among the sixteen profiles (see ESI†). |
Once the classification parameters were defined, we conducted a series of sorts across two different days at different cell densities with a running time of approximately 30 minutes each. Cells were again imaged at 60× magnification and sorted according to their subcellular brightness distribution. The sorted cells were recollected from the system, analysed by fluorescence microscopy and target cell purity was assessed by human-based counting of cells in the images. For this morphology-based sorting we achieved target cell purities of up to 83% and enrichment factors between 4.1 and 11.9 (see Table 1).
Throughput | Analysed cells | Target cell concentration | Enrichment | ||
---|---|---|---|---|---|
(cells per h) | Total | Target | Initial | Final | Factor |
1206 | 1069 | 214 | 20.0% | 83.0% | 4.1 |
1491 | 1341 | 108 | 8.1% | 61.0% | 7.6 |
2789 | 1229 | 248 | 20.2% | 82.2% | 4.1 |
3960 | 3467 | 240 | 6.9% | 82.6% | 11.9 |
7692 | 6902 | 645 | 9.3% | 76.5% | 8.2 |
The results of this experiment are shown in Fig. 5. The normalised cell growth of the reference sample increased throughout the observed days. After a short lag phase, the cell number of the sorted sample followed this increase. The measured absorbance, normalised to cell number, was comparable for both samples, with a slight increase over the days.
Fig. 4a shows the level of image quality achieved by the system when cells flow at 580 μm s−1. The detailed morphology of the cells and the high resolution of the images (i.e., 9.3 px μm−1) without blurring is comparable to the image quality achieved under stationary conditions. The acquisition of these fluorescent images requires a relatively long (500 μs) exposure time, which only leads to non-blurred images when flow velocities are low.
The importance of high-quality images to identify morphological differences between groups of cells and to use them for cell sorting applications is evident in recent works where the high capabilities of artificial intelligence using convolutional neuronal network (CNN) approaches help to classify and even sort blood cells based on label-free bright field images.42–44 These results highlight the major importance of having high-quality images of the cells to detect unexpected morphological patterns that only an algorithm can identify after processing thousands of cells. We believe that high-quality microscopy (bright field in the combination with fluorescence) is fundamental for linking physiological properties of a cell to its morphology.
In this regard, a remarkable feature of our system is that the time available for image analysis can be adjusted according to computing time requirements. This is done by simply adjusting the relative position of the cell inspection area with respect to the sorting microelectrodes. Due to the fact that the movement of the cells in the microchannel along the electrodes is well-controlled and kept in a precise vertical plane, the cells can travel long distances (i.e., several mm) without the need for additional control of speed and trajectory, unlike other systems.22,23 The velocity of each cell is calculated individually, making possible to create an expected trajectory for each cell.
The controlled movement in combination with the low speed of the cells (i.e., a few hundred μm s−1) supports image analysis times in the order of hundreds of milliseconds. This flexibility in decision time opens up the possibility to combine our system with more advanced artificial intelligent computing.43,45 Apart of that, the long frame acquisition and processing times reduced the amount of data and allowed us to use a colour camera, including processes of Debayering and on-line dynamic range colour adjustment for each frame. This is advantageous, since large amounts of data require a more complex data management that would exceed the capabilities of a standard PC.
The low flow velocities in our system are achieved without compromising the sample volume flow rate (i.e., lower μl min−1 range), keeping the real amount of total processed sample per time acceptable. This property is described by the ratio of sample volume throughput to flow velocity (we call this factor “flow efficiency”) and is compared for different approaches to image-based cell sorting in Table S4 (see ESI†). The flow efficiency of our system is orders of magnitude larger than other approaches, as our approach does not require 3D hydrodynamic cell confinement or the use of a small channel cross-section to control the position of the cells along the microchannel.
The performance of a cell sorting process in terms of throughput, yield and purity also depends on the so-called sorting window (i.e., the minimum distance accepted by the sorter to distinguish between two different cells). Our proposed concept of handling the cells by stepwise-arranged microelectrodes allows the active sorting window to be extremely small, down to the size of the object to be sorted, by just reducing the length of the sorting electrodes. This permits working with relatively high particle densities, which leads to reasonable throughputs even at low flow rates. It is easy to see that the sorting window in our approach can be further reduced by a factor of 3–5, which would allow processing of 3–5 times more concentrated cell samples, increasing throughput by the same amount without compromising image quality or other sorting parameters. The size of the sorting window has also a direct impact on the maximum achievable ratio between yield and purity of the sort. The fact of having small sorting windows makes it possible to maximize sorting purity and yield at the same time. We achieved purities of sorted cell samples of up to 95%, while at the same time the yield of sorted target cells was kept at a maximum (i.e., “yield sort”). During the characterisation of the sort based on image-derived colour information, the system was in operation for several hours at times. The average cell recovery rate of 85% with only 5% standard deviation over the individual tests and the robust purities even over longer process durations demonstrates the stability and reliability of the DEP-based sorting principle. This enabled us to sort several 10000 cells at maximum yield. The results on the achieved purity are in agreement with the theoretical estimate based on Poisson statistics (ESI†) for “yield sort” operation mode. According to Poisson statistics, with the given length of the sorting window and a throughput of 12
000 cells per h we can expect a maximum purity of approximately 80% even when target cells accompanied by non-target cells are not actively discarded in favour of purity (Fig. 3a).
While the amount of processed sample volume per time is similar to other approaches, the effective throughput of analysed cells is significantly lower, which is an indirect consequence of the low flow velocities. It is true that the number of cells processed per time could be further increased by a factor of 3–5 without sacrificing other performance parameters (see above). However, we will not be able to achieve cell throughput comparable to FACS sorters without compromising purity, yield and image quality and it is in the simultaneous optimisation of the latter three parameters where the strength of our approach lies. Especially when working with small and valuable cell samples, such as those frequently encountered in organoid- or stem cell research, the study of disease models or small-tissue biopsies, high throughput is less important than the precise, reliable and gentle processing of small cell numbers with minimal cell loss and high-quality cell imaging and -classification.15
In the demonstration of a morphology-based sorting, the obtained purities are consistent with the ones obtained for the colour-based sorting (see Table 1), taking into account the inherent error associated with the more complex sorting algorithm (96% accuracy, see ESI†). Based on the different dye distribution between cytosol- and membrane-stained cells, we could clearly isolate membrane-stained cells from a mixed population. We identified the membrane of the cells and achieved an enrichment factor up to 11.9.
The simplicity of our approach is largely based on the use of a single physical effect for both cell focusing and sorting (i.e., DEP). This allows all electric elements to be controlled by one single external device (an AC signal generator connected to a computer), which makes the system very compact, easy to place on a standard microscope, and unlike complex large-scale instruments suitable for almost any type of laboratory. Moreover, the use of a standard wide-field fluorescence microscope not only allows high-resolution imaging with minimal effort, but also offers extremely high flexibility in terms of the imaging technique. Transmitted light-, phase contrast- or (multi-colour) fluorescence imaging of the most different excitation and emission wavelengths can be combined with our approach as easily as less common imaging techniques (e.g. polarisation-, dark field- or RAMAN microscopy, optical thickness analysis by quantitative phase imaging46etc.). Another important feature is the possibility to add more outlet ways. Our eight sorting electrodes separate target and non-target cells sufficiently to drive them fluidically and dielectrophoretically into two different outlets. For instance, adding another eight sorting electrodes would allow to separate three classes and adding 16 more sorting electrodes four classes, and so on. For the last case, the total amount of sorting electrodes would be 24, which would be feasible without major adaptation of the generators nor electric or fluidic interfaces (our current set-up allows to run up to 30 individually switchable electrodes or eight inputs and outputs in parallel).
Finally, we have shown and verified the vitality of our sorted cells after being processed inside the chip. Vitality is a fundamental parameter for further cultivation of the cells, but is not always specified in other cell sorter approaches. In contrast to conventional droplet-based cell sorting, we do not rely on sheath buffers but are able to work with complete (or even conditioned) cell culture media and low pressures (i.e., single-digit psi range) to image and manipulate the cells at low flow velocities. This helps to reduce shear forces and sorter-induced cellular stress (SICS).47 As a consequence, cells survive the sorting process without major damage, which makes the system also very interesting for sorting shear-sensitive cells or cells that do not tolerate prolonged incubation in PBS. The high vitality rates of the processed cells in comparison with standard cell cultures prove the biocompatibility of our approach and also prove the gentleness of electric field exposition in our set-up.
Electrodes were operated using a custom-made multichannel electric signal generator, which transmitted individually switchable square-wave signals for each electrode with a frequency of 1 MHz and amplitudes between 3 V and 4.5 V peak-to-peak. Data transmission was controlled via USB from the mentioned computer and a Python-based custom-written script, with the commands of the switching states of all electrodes being transmitted in less than 4 ms.
For each frame, the previously defined ROI was analysed for the presence of cells based on colour and/or brightness values. The size, position and time (from the camera clock) were calculated or assigned to each detected cell. In the next frames, the cells were tracked, i.e. the code updated the positions of previously detected cells, gave new cells a new identification number and stored the new information of each cell. Once a cell reached the end of the ROI, its velocity was calculated and a sorting decision was made (note that in colour-based sorting, an explicit sorting decision became obsolete due to the fact that only those cells were tracked that matched the target colour).
In the case of a positive sorting decision (i.e., for a target cell), the expected position of this cell in the next cycles was calculated from its velocity, and, on this basis, the triggers for activating or deactivating the respective sorting electrodes were sent to the generator. It was possible to control several cells simultaneously. As soon as a cell was anticipated at the end of the sorting electrode array, it was deleted from the process.
Parallel to the principal process of reading, analysis and generator control, we implemented another process to visualise the frames and save videos in real time with OpenCV-Python. The parallelisation was performed using multiprocessing module. Having the saving function in a parallel process allowed us to record long-duration videos.
For the case of 4× colour-based sorting, there was a post processing code to analyse the videos and quantify the efficiency of the sorting (i.e., count the number of red and green cells which were directed to the target and waste outlets). This code was written using python modules for image processing like: NumPy, OpenCV-Python, SciPy and Scikit-image.
In the case of colour-based sorting, the cells were automatically detected by the system and assigned to the red target or green non-target population based on their fluorescence signal. This assignment was visually checked and verified on a random basis. In the case of morphology-based sorting, the images were evaluated by visual inspection following assignment of the cells to one of the two classes.
5 mL of a cell sample with a concentration of 1 × 106 cells mL−1 was carefully resuspended and divided into two samples. One sample was passed through the system for four hours as described above with the rinsing buffer containing conditioned cell culture medium. Another sample was kept at room temperature as a reference during that time. To compensate for the dilution of the first sample due to the rinsing buffer, the reference sample was manually diluted with conditioned cell culture medium after two hours to equalise the final cell concentration of both samples to approximately 4 × 104 cells mL−1.
Subsequently, 1.1 mL of each sample was mixed with 110 μL of the cell proliferation assay solution (WST-8, Dojindo, Japan). Then 110 μL of each of this mixture was added to 8 wells of a 96-well plate and incubated at 37 °C and 5% CO2 for three hours. The absorbance at 450 nm for each well was determined using a photometer (EnVision 2105, Perkin Elmer, Germany) and the values from wells of the respective sample were averaged.
In parallel, four cell counting slides (Luna-FL, Logos Biosystems, South Korea) were loaded with each sample and 20 images of each sample were taken under a microscope (Olympus IX73, Olympus) at 10× magnification. The cell number in these images was then automatically evaluated using a custom-made Python code. The cell concentration of the sample was calculated based on the determined cell count and the known size of the image section.
The entire vitality test process was repeated three times in total over the course of several weeks and the results of all three runs were averaged.
A major advantage of the system is its flexibility with regard to the optical components and the image acquisition method employed, as it is compatible with virtually any microscopic imaging technology. An equally large degree of flexibility is provided with regard to the analysis algorithm used. Due to the low flow velocity and flexible positioning of the image acquisition relative to the sorting array, long iteration times and thus also complex analysis procedures can be implemented.
The proven biocompatibility of the system in combination with low shear forces and gentle cell handling makes the system seem particularly suitable for (shear-) sensitive cell samples.
In summary, the low complexity and high flexibility in technical design and operation, as well as the moderate cost of the optical and fluidic components used in our system, set the system apart from other approaches. This paves the way for the development of small and affordable benchtop instruments. In the long term, this could lead to the approach becoming an interesting alternative for smaller laboratories and working groups beyond core facilities.
Footnotes |
† Electronic supplementary information (ESI) available: Video of color-based cell sorting, video of morphology-based cell classification, theoretical estimation of the achievable purity based on Poisson statistics, recovery rate definition, development of criteria for image-based cell classification, microscopic images of the sorting channel and the focal alignment and velocities of dielectrophoretically-guided cells, comparison of different sorting methods. See DOI: https://doi.org/10.1039/d3lc00242j |
‡ These authors contributed equally to this work. |
This journal is © The Royal Society of Chemistry 2023 |