Capture and reagent exchange (CARE) wells for cell isolation, labeling, and characterization

Cell therapy is an emerging field that uses cells as living drugs to treat a broad array of acute and chronic diseases. Most cell therapies in clinical trials are made using standard bench methods, whose open processing require manufacturing in expensive GMP cleanrooms. As cell therapies progress, new methods are needed to enable scalable manufacturing while maintaining process integrity, reducing environmental exposure, and limiting critical cell and reagent use. Here, we introduce capture and reagent exchange (CARE) wells that allow critical processing steps to be integrated into a closed microfluidic device. The unique property of CARE wells is that they allow reagent exchange from an attached channel without cell loss from wells. We show through simulation and experiment that this feature is present in cylindrical wells whose depth is sufficient to generate multiple recirculating vortices and is independent of flow rate in the channel. We demonstrate that CARE wells can be used to perform cell separation, on-chip labeling, and characterization of monocytes as the first steps toward a closed microfluidic system for production of dendritic cell therapies. Immunomagnetic separation of CD14 + monocytes from peripheral blood mononuclear cells (PBMCs) into wells was performed with purity of 97 ± 2% and capture efficiency of 50 ± 17%. On-chip labeling, washing, and characterization were performed using two cell surface markers (CD14 and HLA-DR) on over 3000 cells captured in a 5193-well device. The combination of high purity separation and reagent exchange without cell loss with robust performance over wide range of input and operating conditions makes this technique a promising approach for scalable manufacturing and analysis of cell therapies.


Introduction
Cell therapy is an emerging paradigm of medical treatment that uses cells to treat a broad array of acute and chronic diseases (Ankrum and Karp 2010;Anguille et al. 2014;Holzinger et al. 2016). Most cell therapy products are made using typical bench cell culture techniques performed inside of a clean room. New methods are needed to enable scalable manufacturing of cell therapies while maintaining protocol integrity and meeting required safety and potency requirements. Closed systems that integrate the manufacturing process into a single instrument are desirable to reduce skilled technician time, reduce reagent use, and reduce or eliminate expensive GMP lab space (Iancu and Kandalaft 2020).
Additionally, due to the fragmentation of cell processing and analysis, cell product testing consumes large quantities of cells, increasing both cells harvested from the patient and total costs (Boudousquie et al. 2020). There is a clear need to develop tools for both integrated closed manufacturing of cell therapies and minimally destructive functional test platforms to enable broader access to these life-saving therapies.
Dendritic cells (DCs) are antigen-presenting cells that act as the sentinel of the immune system and can stimulate both naïve and recall responses (Steinman and Banchereau 2007) and have shown promise for use as cancer therapy (Anguille et al. 2014). The most common method for production of therapeutic DCs includes immunomagnetic isolation of CD14 + monocytes from the patient's blood, culture with cytokines to differentiate them to into immature dendritic cells, and a second culture with cytokines and a target antigen to produce mature, antigen-presenting dendritic cells (Dietz et al. 2006a, b, Laborde et al. 2014. After cell culture is completed, millions of cells are sacrificed for phenotypic 1 3 60 Page 2 of 10 characterization before they are injected back into the patient where they promote an immune response against the target antigen-expressing cells. Depending on the application, the antigen can be tumor lysate (Fadul et al. 2011), autologous tumor lysates (Parney et al. 2020), or peptides that are highly expressed on the target tumor (Bedrosian et al. 2003). These therapies have demonstrated good safety and efficacy in feasibility studies for lymphoma , glioblastoma (Parney et al. 2020), and ovarian cancer (Knutson et al. 2020). Efforts to develop closed systems for manufacturing DCs have primarily relied on daisy-chaining existing biomanufacturing instruments with closed fluidic connections (Erdmann et al. 2018;Uslu et al. 2019). Their use still requires skilled operators and specialized facilities. A closed, fully integrated system for producing dendritic cells would overcome these limitations and allow much broader deployment of DC-based cell therapies.
Immunomagnetic isolation is an established separation technique that uses magnetic beads conjugated to an antibody against a specific cell surface marker to isolate cells expressing that marker within a mixture of cells (Thiel et al. 1998). When placed in a magnetic field, labeled cells can be drawn away from other cells and collected. A variety of methods have been developed for using this technique in microfluidic devices, including continuous flow isolation (Inglis et al. 2004;Kim et al. 2013;Darabi and Guo 2016), immobilization in a microfluidic channel (Hoshino et al. 2011;Mohamadi et al. 2015), isolation into microwells (Huang et al. 2018;Armbrecht et al. 2019), and multistage isolations coupled with size-based sorting (Mishra et al. 2020), or in combination with deformability ). There is a large catalog of commercially available magnetic isolation kits based on surface markers, including stem cells, hematopoietic cells, and circulating tumor cells, and many of these have been used in clinical manufacturing of cell therapy products (Van Driessche et al. 2009;Priesner et al. 2016).
Microwells are a versatile tool for performing diverse biological workflows in microscale devices. A single microwell allows cells (Rettig and Folch 2005), organoids (Brandenberg et al. 2020), or other objects to be retained in a fixed location, while the composition of the fluid above the microwell is changed to accomplish a number of analytical tasks. Arrays of microwells allow populations of cells to be quantified, characterized, and monitored with single cell resolution (Wang et al. 2007). Complex biological workflows have been integrated into microfluidic devices using microwells, including digital PCR (Podbiel et al. 2020), single cell RNA-Seq (Yuan and Sims 2016), multiplex secretion profiling (Armbrecht et al. 2019), nutrient adaptation (Woronoff et al. 2020), immunostaining (Kobayashi et al. 2015), drug discovery (Jorgolli et al. 2019;Ai et al. 2020), organoid culture (Brandenberg et al. 2020), tissue microRNA analysis (Nagarajan et al. 2020), proliferation assays (Park et al. 2010), and lineage analysis (Luro et al. 2020). A feature of microwells that has been used by some groups is to increase the depth of the well to prevent flow from sweeping out cells within the well (Pilat et al. 2017;Avesar et al. 2018;Jorgolli et al. 2019;Luro et al. 2020). Avesar et al. showed with simulations that 200-μm-wide by 100-μm-tall by 400-μm-deep wells had recirculating flow at the well opening and no convective flux between the wells and the channel (Avesar et al. 2018). While this feature of deep microwells has been used by several groups, no one has identified the key parameters that allow microwells to retain cells when there is flow in an attached channel.
In this paper, we present a microfluidic device that allows many key functions of cell therapy manufacturing to be integrated into a single closed system using cell isolation and reagent exchange (CARE) wells. The device is designed to take advantage of the flow properties in deep microwells to allow of integration of immunomagnetic separation and subsequent reagent exchange operations without cell loss in scalable manner. We performed simulations and experiments to understand the well geometries necessary to get these capabilities and then demonstrated the functionality of this technique by performing separation, labeling, and characterization of monocytes from a sample of peripheral blood mononuclear cells (PBMCs) without cell loss during fluidic operations. The combination of high purity separation and reagent exchange without cell loss makes this technique a promising approach for minimally destructive analysis of cell cultures and scalable manufacturing of cell therapies.

Results
Capture and reagent exchange (CARE) wells are designed to take advantage of the flow properties of deep microwells to allow for integration of immunomagnetic cell capture and reagent exchange operations, such as immunofluorescent labeling and washing. A schematic of the device operation is shown in Fig. 1. The device is composed of a channel with wells patterned along the upper wall and a single inlet and outlet. Cells labeled with antibody-conjugated magnetic nanoparticles are imported into the channel and pulled into the wells by a magnet, while unlabeled cells remain in the main channel or sediment under the influence of gravity (Fig. 1A). The wells are designed such that flow from the channel does not enter the well, so only cells labeled with magnetic nanoparticles will be captured and unlabeled cells can be flushed from the device (Fig. 1B). After cell capture, the device is flipped so gravity maintains captured cells in the wells (Fig. 1C) and fluid in the main channel can be changed to allow reagent exchange, such as in situ labeling with fluorescent markers, washing, and enumeration and characterization (Fig. 1D, E).
To achieve integrated separation and reagent exchange, cells need to be shielded from flow in the channel after capture. Simulations were performed in COMSOL 5.4 (Burlington, MA, USA) to understand the key geometric features necessary to retain cells when there is flow in the channel. Figure 2A shows streamline plots for fluid flow in a geometry composed of a cylindrical well with variable diameter from 10 to 100 microns with 100 micron depth connected to a 200-micron-deep channel with average inlet velocity of 1 mm/s in the channel. The streamline plots show that flow from the channel does not enter the well for any well diameter. Rather, a series of recirculating vortices are generated starting at the entrance of the well with the number of vortices increasing as the well diameter decreases. The number of vortices as a function of well diameter for a fixed well depth of 100 microns is plotted in Fig. 2B (blue squares). Wells with diameters of 50 microns or less have multiple vortices, while those with diameters of 60 microns or more have a single vortex. Cells or objects driven by flow in the main channel will follow streamlines unless acted on by external forces (White 1974). Therefore, the only way for a cell to enter or exit the well is via non-fluidic forces (magnetic, gravitational, thermal, etc.). This ensures that immunomagnetic separation (shown in Fig. 1A) will capture a pure population of cells by surface marker expression.
We tested the effect of vortices in microwells experimentally with beads and found that multiple vortices are needed for it to retain objects during fluidic operations. A microwell device was made that had arrays of wells with variable diameters from 10 to 100 microns patterned inside a channel with a single inlet and outlet (layout shown in Figure  S1). 5.8 micron COMPEL magnetic beads (Bangs Laboratories, Fishers, IN, USA) suspended in buffer at 0.01% w/v were pulled into the wells with a magnet, the magnet was removed, and the total number of beads loaded in the array of wells at each diameter was counted. The device was then flushed with 1 mL of liquid at 100 µl/s with a syringe pump and total number of beads remaining in the array of wells at each diameter was counted. The percentage of beads initially loaded into wells that was retained after flushing at each diameter is shown in Fig. 2B (black circles). Wells with diameters between 60 and 100 microns (those shown to only have a single vortex) lost a percentage of beads during the flush, while those with diameters of 50 microns or less (those shown to have multiple vortices) retained all the beads during the flush. This result can be understood through the streamline plots. In a well with a single vortex, an object in the bottom of the well will be driven to the mouth of the well by flow in the channel where it may divert from the streamline due to thermal motion and be lost. In a well with multiple vortices, an object at the bottom of the well will be brought partially up the well and will remain protected from flow in the channel. We do not expect there to be a strong Removing the magnet and flipping the chip allows captured cells to be retained by gravity while fluidic operations are performed in the channel. D On-chip labeling is performed by pulling a solution containing fluorophore-conjugated antibodies into the channel and allowing them to diffuse into the wells. E Cells are washed by flushing with buffer until residual antibodies diffuse from wells dependence on cell/bead size and retention in the well so long as the object remains entirely below the first vortex.
Additional simulations established that the recirculating flow pattern in the wells is independent of flow rate. Simulations of 30-micron-diameter wells with 100 micron depth attached to a 200-micron-deep channel were performed with varying average inlet flow rate from 10 cm/s to 1 μm/s. Streamline patterns are plotted as well as the log of velocity magnitude normalized to the inlet velocity (Fig. 2C). The plots show that both the streamlines and the normalized velocity magnitude are nearly identical over five orders of magnitude difference in flow rate and the velocity in the bottom of the well is reduced by a factor greater than 10 8 from the channel. Cells or beads in the bottom of wells with this geometry will experience essentially no fluid motion (1 nm/s) even when fluid is moving quickly (10 cm/s) in the channel. This is ideal for maintaining a low-shear environment for cell culture and enabling a wide variety of flow rates to be used for various analytical operations, such as perfusion for culture or fast flushing for labeling (Young and Beebe 2010). This result contrasts with inertial vortex capture approaches where the flow pattern in the capture reservoir changes depending on the flow rate (Hur et al. 2011;Chung et al. 2013) and provides a robust performance over a wide range of operating conditions. Cell retention across a range of flow rates was further verified by exposing beads in wells to a sequence of flushes at varying flow rates and monitoring for bead loss. 207 magnetic beads with 4 micron diameter (Bangs Laboratories, Fishers, IN, USA) were loaded into an array of wells with 30 micron diameter and 100 micron depth. The device was then flushed with 1 mL of buffer at 100 μL/s, imaged, and beads were counted. This was repeated for 10 μL/s, 1.0 μL/s, and 0.1 μL/s in sequence. During this process, which exposed beads to flow rates varying over three orders of magnitude and greater than three hours of flushing, 207 beads were observed after every step and no beads were lost. The process and results are summarized in Table 1. Combined with the simulation results, this verifies that objects captured in We use these simulation results and experiments to define capture and reagent exchange (CARE) wells as those that contain two or more recirculating vortices as shown by streamline plots. Within the geometric space explored in this paper (cylindrical wells with 100 micron depth connected to a 200-micron-deep channel), wells with diameter less than 50 microns exhibited the CARE functionality, while those with larger diameters did not. Because the flow field was independent of flow rate, these criteria will remain consistent if the geometric parameters are scaled up or down by a constant value (Sedov 1982). These criteria may differ for different well shapes and ratios of well depth to channel height and a more in-depth exploration remains for future work.
Using CARE wells, we demonstrated immunomagnetic isolation of CD14 + monocytes from PBMCs with high purity over a wide range of input cell densities. Devices contained 5193 microwells with 30 micron diameter and 100 micron depth patterned over a 2 by 8 mm area. A flow channel with a single inlet and outlet over the well area had 200 micron depth. External connections were made with 23-gauge needle tips mated to EVA plastic tubing to vertical holes punched through the channel layer. One port was connected to a syringe for fluid operations, while the other was open for sample input/output. A schematic of the device layout and cross section is shown in Fig. 3A. PMBCs were isolated from donor blood using Histopaque density gradient separation and labeled with Miltenyi CD14 MicroBeads and fluorescent antibodies against CD14 (monocytes) and CD45 (hematopoietic cells) surface markers. All cells within the PBMC population are expected to express CD45. Cell density, viability, and CD14 abundance were measured in the input cell sample (Fig. 3B). CD14 abundance in the input PBMC samples varied by donor with a range of 13-36% with mean of 24 ± 9%. The input CD14% for each experiment is listed in Table S1. To load cells, 100 μL of cell sample was pulled into the chip at 10 μL/s, flow was stopped and two stacked 32 lb pull force neodymium magnets were placed on top of the device for 10 min to pull bead-labeled cells into wells (Fig. 3C). After 10 min, cells remaining in the channel were flushed with 1-2 ml buffer at 100 μL/s, the magnet was removed, and the chip was flipped to retain captured cells by gravity. This approach yields a uniform density of captured cells across the chip but does not capture any cells outside of the well area. After cell capture, a composite brightfield and fluorescent image was captured of the well area of each device (Fig. 3D) with an EVOS Cell Imaging system (Thermo Fisher, Waltham, MA, USA). For each device, cells were counted using a semi-automated program written in MATLAB to determine the total number of CD45 + cells and CD14 + /CD45 + cells. A detailed description of the cell counting program is available in the supplemental material.
Twelve immunomagnetic CD14 isolation experiments, each with a unique donor, were performed with a range of input cell densities from 6.2e5 to 1.6e7 cells/mL, and an average captured CD14 + cell purity of 97 ± 2% was observed with estimated capture efficiency of 50 ± 17%. Purity is defined as number of captured CD14 + cells divided by the total number of captured CD45 + cells. Estimated capture efficiency is defined as the number of captured CD14 + cells divided by the number of CD14 + cells calculated to be in the well area of the chip during separation (live cell density x CD14% in PBMC input x chip volume). Individual experiment results are tabulated in Supplementary Table S1 and purity and capture efficiency as a function of input cell density are plotted in Supplementary Fig. S2. A linear fit shows that purity does not depend strongly on input density, while there is a decrease in capture efficiency with increasing input cell density. Some strategies toward increasing capture efficiency include increasing the magnetic force on cells by increasing capture bead concentration or size or increasing magnetic field strength. Increasing the surface density of wells will increase the capture area the channel area (some cells may be pulled up to the surface of the channel but not into a well). Increasing the channel depth may also allow more cells to be loaded at lower input cell densities if cell clumping at higher densities is limiting capture efficiency.
On-chip labeling with fluorophore-conjugated antibodies, washing, and characterization without cell loss were performed on cells captured in CARE wells. A CD14 isolation with cells only labeled for CD45 was performed and cells were labeled on-chip with antibodies against CD14 and HLA-DR. Low or negative expression of HLA-DR has been associated with an immunosuppressive state that can hinder dendritic cell maturation (Vuk-Pavlovic et al. 2010;Young and Beebe 2010). Labeling was performed by pulling a solution containing fluorescent antibodies into the chip and incubating for 10 min. Washing was performed by flushing the chip with buffer for 10 s at 100 μL/s to remove the labeling solution in the channel and then 30 min at 1 μL/s to allow unbound labeling antibodies to diffuse from the wells. Figure 4A shows a cropped region of the device in each fluorescent channel (CD45/CD14/HLA-DR) before labeling, during labeling, and after washing. Cells that were initially unlabeled for CD14 and HLA are clearly labeled in the wells. A significant fluorescent background can be observed in the labeling step that is removed after the wash, allowing fluorescence intensity of individual cells to be characterized. Figure 4B shows scattergrams (CD45 vs CD14 and CD45 vs HLA-DR) of fluorescence intensity for 3123 CD45 + cells identified after immunomagnetic isolation before and after labeling, with clear separation between unlabeled and labeled populations observed. This capability allows non-disruptive, in situ analysis of surface marker expression of the captured cell population within the device. The timing of the washing step was informed by a numerical simulation of diffusion of antibodies from the wells during a 1 μL/s flush. Diffusion from the well was simulated in COMSOL 5.4 using the boundary conditions and geometry shown in Fig. 4C. A single 30-micron-diameter, 100-micron-deep well was attached to a 200-micron-deep channel. Average inlet velocity was set to be 2.5 mm/s, which is the equivalent velocity to 1 μL/s on the full device. The concentration was set to 1 mol/m 3 in the geometry with a concentration of 0 mol/ m 3 at the inlet to the channel. The diffusion coefficient was set to 10 μm 2 /s based on estimates for diffusivity of an antibody at room temperature (Nauman et al. 2007). A time-dependent solution was calculated with 10-s intervals for 30 min. The plot in Fig. 4C shows the average concentration in the well in blue and the concentration at the bottom surface of the well in pink. In 30 min, the Fig. 3 Immunomagnetic separation of monocytes in the CARE well device. A CARE well devices were made in PDMS and composed of a 100-micron-deep well layer with 5193 microwells with 30 micron diameter patterned over a 2 by 8 mm area and a 200-micron-deep flow layer with a channel over the wells connected inlet and outlet ports. External connections were made with hollow metal pins mated to plastic tubing with one port connected to a syringe pump for fluid operations and the other left open for sample input/output. B Input samples composed of PBMCs isolated from whole blood labeled with CD14 MicroBeads and fluorescent antibodies against CD14 and CD45. Scale bar is 50 microns. C Input cells are pulled into PDMS microfluidic device in wells-up configuration and magnet is placed on top of device for 10 min to capture cells labeled with antibody-conjugated magnetic nanoparticles. Cells remaining in channel are flushed with 2 mL buffer, magnet is removed, and chip is flipped for imaging. D Composite (bright field/CD14/CD45) microscopic image of 5193 microwell device after monocyte capture. From an input PBMC sample containing 19% CD14 + /CD45 + cells, 3123 cells were captured with 98% CD14 + /CD45 + purity. Full chip image scale bar is 250 microns. Inset scale bar is 50 microns average concentration in the well is 5e-4 mol/m 3 , while the concentration at the bottom of the well is 1e-3 mol/ m 3 . This corresponds to a 2000 times reduction in concentration in the well on average and a 1000 times reduction in concentration at the bottom of the well, effectively washing the labeled cells from unbound fluorescent antibody. After the first few minutes, both curves show good fit with exponential decay curve with the same decay constant of − 0.24 min −1 , so each 9 min and 36 s of additional flushing under these conditions will result in a further 10 times reduction in concentration. Combined with the experimental results, these simulations provide confidence that reagent exchange operations can be performed on cells captured in the wells.

Conclusion
There is a clear need to develop tools for integrated closed manufacturing of cell therapies and minimally destructive functional test platforms to enable broader access to these life-saving therapies. Here, we have presented a new microwell that allows cell capture and reagent exchange (CARE) to be integrated into a simple, valveless design. We showed through simulation and experiment that fluid motion in CARE wells is characterized by multiple recirculating vortices that shield cells or objects in the wells from flow in an attached channel and allows reagent exchange without cell loss. By exploiting immunomagnetic isolation, gravity, and CARE wells, we can separate cells from a mixture with high purity over a wide range of input densities and perform subsequent reagent exchange without cell loss. With the addition of simple fluorescent imaging and image processing software, the device can be used to count and phenotype cells. These devices could be patterned over a large area to accommodate millions of cells, manufactured in plastic using high volume manufacturing techniques, and operated as a closed system rather than requiring GMP space. In situ analysis of cell surface marker expression without loss could be performed to monitor process steps and ensure the safety and efficacy of the final cell product. The combination of high purity separation and reagent exchange without cell loss, with robust performance over wide range of input and operating conditions, gives this approach broad applicability in cell therapy manufacturing, including diagnostic processes, cell isolation, exposure to differentiating or transformative agents, phenotyping, culture, and expansion.

Microfluidic device fabrication
Microfluidic devices were designed using CAD software (AutoCAD 2020, AutoDesk, San Rafael, CA, USA) and were fabricated in PDMS using photolithography and soft lithography (Becker and Gartner 2000). The devices consist of a microwell layer with 5193 cylindrical microwells and a channel layer for fluid transport over the microwells. Microwells had 30 micron diameters and 100 micron depth and the channel layer is 200 micron thick. Replica molded PDMS layers were aligned and bonded to each other and a glass slide by oxygen plasma (YES-G1000, Yield Engineering Systems, Livermore, CA, USA). Fluidic connections were made by through holes punched in the channel layer.

Cell preparation
Cell samples were obtained from blood donated by volunteers at the Division of Transfusion Medicine, Mayo Clinic, Rochester, Minnesota, in accord with current regulations by the AABB and US FDA. Following an apheresis procedure, a leukoreduction system chamber from the apheresis apparatus (Trima Accel, Gambro BCT, Lakewood, CO, USA) was collected, and PBMCs were isolated using density gradient separation in Histopaque (MilliporeSigma, St. Louis, MO, USA) following a previously established protocol (Dietz et al. 2006a, b). Cell density and viability were enumerated using a Countess Cell Counter (Thermo Fisher, Waltham, MA, USA) and trypan blue exclusion. CD14 percentage in PBMC samples was determined by taking a composite fluorescent image of labeled input cells (one channel each for CD14 and CD45 markers) loaded in a Countess Cell Counting Chamber Slide with an EVOS Cell Imaging system (Thermo Fisher, Waltham, MA, USA) and quantifying the number of CD45 and CD14 cells. The CD14 percentage was calculated as the ratio of number CD14 cells detected to the number of CD45 cells detected. Cells were labeled with CD14 MicroBeads (Miltenyi, Bergisch Gladbach, Germany) and stained with antibodies (Biolegend, San Diego, CA) against CD45 and CD14 surface markers following the manufacturers' protocols before being centrifuged, washed, and resuspended to a desired cell density. In the on-chip labeling experiment, cells were only labeled with CD45 antibodies before on-chip capture.

Device operation
Devices were wet with a solution of 1% (w/v) Pluronic F-127 in PBS with fluid flow driven by a hand-operated syringe or syringe pump (KD Scientific, Holliston, MA, USA). Onchip magnetic isolation was performed by pulling labeled cell suspensions into the flow channel, stopping flow, and placing a magnet on top of an inverted chip. Two 32 lb. pull force neodymium magnets (K&J Magnetics, Pipersville, PA, USA) were stacked and placed on top of the chip for 10 min, then the flow channel was flushed with greater than 1 mL of buffer at approximately 100 uL/s. The magnets were removed, and the chip was flipped to a wells-down configuration for imaging and other fluidic operations.

Microscopy and image analysis
Imaging was performed on an EVOS Cell Imaging system. Stitched, composite brightfield, and fluorescent images were captured over the entire well region for each device. Individual channels (brightfield and fluorescent) were split for semi-automated analysis in MATLAB2019a (Mathworks, Natick, MA, USA). Individual wells were identified using the circular Hough transform and used to mask fluorescent images. After adjusting for brightness/contrast, cell quantification was performed on the masked fluorescent images and detected cells were assigned to a specific well. Any well that contained a fluorescent cell in any channel was then manually inspected to confirm the count. The resulting data included total number of cells detected for each fluorescent channel and the number of cells per well. More detail on this process is available in the supplementary material.

On-chip labeling
Alexa 488 anti-human HLA-DR (200 μg/mL), PE antihuman CD14 (25 μg/mL), and Alexa 647 anti-human CD45 (150 μg/mL) antibodies were used in the labeling experiment (Biolegend, San Diego, CA, USA). Antibodies were diluted 10:1 in buffer, pulled onto the chip, incubated for 10 min at room temperature, and subsequently flushed from the chip. Washing was performed by flushing the chip with buffer for 10 s at 100 μL/s and then 30 min at 1 μL/s. The initial flush is to remove the label from the main channel, while the secondary flush allows antibodies within the well time to diffuse out. Full chip images were captured in fluorescent channels corresponding to each antibody before staining, during labeling, and after washing. Exposure times were constant in CD45 channel (500 ms) across all steps, while exposure times were reduced during the labeling step in the CD14 (500 ms to 20 ms) and HLA-DR (300 ms to 100 ms) channels before being returned to initial values in the post-wash image.