Assay and Isolation of Single Proliferating CD4+ Lymphocytes Using an Automated Microraft Array Platform

Abstract

Objective:

While T lymphocytes have been employed as a cancer immunotherapy, the development of effective and specific T-cell-based therapeutics remains challenging. A key obstacle is the difficulty in identifying T cells reactive to cancer-associated antigens. The objective of this research was to develop a versatile platform for single cell analysis and isolation that can be applied in immunology research and clinical therapy development.

Methods:

An automated microscopy and cell sorting system was developed to track the proliferative behavior of single-cell human primary CD4+ lymphocytes in response to stimulation using allogeneic lymphoblastoid feeder cells.

Results:

The system identified single human T lymphocytes with a sensitivity of 98% and specificity of 99% and possessed a cell collection efficiency of 86%. Time-lapse imaging simultaneously tracked 4,534 alloreactive T cells on a single array; 19% of the arrayed cells formed colonies of ≥2 cells. From the array, 130 clonal colonies were isolated and 7 grew to colony sizes of >10,000 cells, consistent with the known proliferative capacity of T cells in vitro and their tendency to become exhausted with prolonged stimulation. The isolated colonies underwent ELISA assay to detect interferon-γ secretion and Sanger sequencing to determine T cell receptor β sequences with a 100% success rate.

Conclusion:

The platform is capable of both identification and isolation of proliferative T cells in an automated manner.

Significance:

This novel technology enables the identification of TCR sequences based on T cell proliferation which is expected to speed the development of future cancer immunotherapies.

I. Introduction

The adaptive immune system is characterized by an acquired immune response to foreign pathogens, including viruses and bacteria, and endogenous threats such as tumors. In response to pathogens and tumors, a cellular immune response is mounted through lymphocytes such as CD8+ T lymphocytes (cytotoxic T cells) and CD4+ T lymphocytes (helper T cells). CD8+ T cells are primarily responsible for removing threats by inducing cell death through cytotoxicity while CD4+ T cells are responsible for mediating the immune response through intracellular signaling and activation [1]. The adaptive immune system has been harnessed in the clinic to reduce and remove blood borne and solid tumors through increasing, regulating, or inhibiting lymphocytic activity [2][3][4]. Typical cell-based immunotherapies include the use of engineered CD8+ T cells (TCR-T) and chimeric antigen receptor cells (CAR-T) to eradicate cancerous cells by inducing apoptosis in target cells [5]. While TCR-T and CAR-T cell therapies are most represented in immunotherapy research and clinical trials, the importance of helper T cells in developing an effective therapy has recently gained recognition [6]. TCR-T and CAR-T often fail due to a lack of longevity or suppression of the immune response within the tumor microenvironment [5], [7], [8]. To overcome the downregulation or exhaustion of TCR-T cells, helper T cells can be included in cell therapies to activate and maintain CD8+ effector and memory cells [9]. CD4+ T cells provide essential cytokines such as interleukin (IL)-2 and interferon gamma (IFN-γ) for improving CD8+ T cell proliferation and viability. The co-transfer of helper T cell and cytotoxic T cells has been shown to improve the quality and lifetime of the immune response [10][11][12][13].

Helper T cells recognize antigens presented on human leukocyte antigen complex (HLA) class II molecules, primarily located on the surface of professional antigen-presenting cells (APCs). The T cells interact with the antigen-HLA complex with T cell receptors (TCRs) and become activated upon recognition of their cognate antigen. Antigens presented on HLA class II molecules are derived from breaking down foreign pathogens or endogenous dysfunctional cells such as cancer cells. Importantly, T cells can distinguish between self-antigens and foreign antigens due to negative selection during maturation in the thymus [14]. Activated helper T cells undergo proliferation and differentiation into memory and effector cells [15].

While antigen-specific CD8+ T cells can be identified and sorted through peptide-HLA tetramers and flow cytometry, the low avidity of CD4+ TCRs to HLA class II molecules has rendered such techniques ineffective in sorting antigen-specific CD4+ T cells [16], [17]. Alternative methods for identifying and isolating antigen-specific helper T cells include laborious techniques such as limiting dilution, however single-cell analysis and sorting of this lymphocyte subpopulation remains extremely challenging [18], [19]. While single-cell CD4+ T cell proliferation has been tracked using microdevices in the past, these systems were not capable of both assaying and sorting lymphocytes based on proliferation greatly limiting their utility [20], [21]. A technology capable of tracking the proliferation of individual CD4+ T cells in a high-throughput and parallel manner with sorting capabilities would provide an optimal system for identifying antigen-specific helper T cells based on functionality.

To meet the critical need of identifying activated CD4+ T cells based on proliferation, polydimethylsiloxane (PDMS) microwell arrays containing releasable elements termed microrafts, were used to track T cell proliferation in an automated manner and isolate individual T cells of interest. Instrumentation and software were developed to provide an automated platform for single-cell analysis of T lymphocytes. The platform’s capabilities were demonstrated by time-lapse imaging CD4+ T cells stimulated by allogeneic (derived from a separate donor than the T cells) Epstein-Barr virus-immortalized lymphoblastoid cell lines (EBV-LCLs). To demonstrate cell isolation and post-isolation analysis, colonies were isolated from the arrays, expanded, and then samples from clonal populations were genetically analyzed for their TCR α and β sequences and assayed for interferon-γ secretion using enzyme-linked immunosorbent assay (ELISA).

II. Experimental

A. Microraft fabrication

Microraft arrays were manufactured using methods previously described [22], [23]. The microraft array templates were fabricated using photolithography and a 1002-F-100 negative photoresist. The microraft arrays were then fabricated using a standard soft lithography process and PDMS (Sylgard 184 silicone elastomer kit: Dow Silicones Corp., Midland, MI). A single microraft array contains 19,600 microwells in a square pattern with 140 rows and columns, each microwell measuring 100 μm × 100 μm × 187 μm (L × W × H) with 50 μm spacing between wells. The arrays were dip-coated in a solution of 20% poly(styrene-co-acrylic-acid) in gamma butyrolactone (GBL) containing 1% γFe₂O₃ nanoparticles. The arrays were subsequently baked at 95° C to evaporate the GBL solvent and solidify the polystyrene. The dipcoating and baking process resulted in microwells each containing a hard, concave polystyrene element, termed a microraft which was doped with superparamagnetic nanoparticles.

B. Microscopy Setup

An MVX10 MacroView (Olympus, Center Valley, PA) fluorescence microscope was inverted and mounted to an optical table with a cutout in the center to accommodate the microscope. The microscope was outfitted with a 2x widefield objective with a 0.5 numerical aperture (NA). The microscope was equipped with an ORCA-Flash4.0 V.3 digital CMOS camera (Hamamatsu, Bridgewater, NJ). The system included a motorized H112A stage controlled by a ProScan III Controller, a PS3H122 Motorized Focus Drive, and a PJ2J100 joystick (Prior Scientific Inc., Rockland, MA). A Lambda 10–3 optical filter changer (Prior Scientific Inc., Rockland, MA) was used to electronically control an excitation filter wheel with a SmartShutter (LB10-NWIQ) and an emission filter wheel (LB10- NWE) (Sutter Instrument, Novato, CA). The filter wheels were outfitted with a sedat filter set (89000 – ET – Sedat Quad; Chroma Technology Corp, Bellows Falls, VT), providing a multiband dichroic filter, five excitation filters with central wavelengths (CWL) and bandwidths (FWHM) of 350 ± 50 nm, 402 ± 15 nm, 490 ± 20 nm, 555 ± 25 nm, 645 ± 30 nm, and four emission filters with CWLs and FWHMs of 455 ± 50 nm, 525 ± 36 nm, 605 ± 52 nm, 705 ± 72 nm. The sedat filter set allowed for fluorescence imaging in the blue, green, red and far red portions of the light spectrum. A Lumen 200 mercury lamphouse (Prior Scientific Inc., Rockland, MA) was used to provide excitation light. The system was controlled and operated using MATLAB (MathWorks, Natick, MA) and a Micro-Manager (Open Imaging, San Francisco, CA) core. The system was enclosed in a custom incubation chamber, providing temperature, humidity, and CO₂ regulation during experiments. The incubation chamber was maintained at 55% humidity, 37° C, and 5% CO₂ while imaging cells.

C. Microraft Sorting System

The release device was comprised of a single 19000 Series Can-Stack stepper motor (Haydon Kerk Motion Solutions Inc, Waterbury, CT), two steel dowels (60 mm length, 6 mm diameter: McMaster-Carr, Atlanta, GA), and a micro needle (10 mm length, 100 μm diameter base, 10μm tip) attached to a clear polycarbonate window. These components were attached to a custom aluminum chassis comprised of two concentric discs, aligned by the two dowels serving as guide rods. Together, these components actuated the micro needle in the z-axis to pierce microraft arrays from below, releasing target microrafts.

The 3-axis pick-and-place system consisted of two VLST45 ball screw linear actuators (Moog, Mountain View, CA) to control motion in the x-axis (250 mm travel) and y-axis (350 mm travel), and a 31mm 19000 Series Can-Stack stepper motor (Haydon Kerk Motion Solutions Inc, Waterbury, CT), providing motion in the z-axis, to actuate a magnetic wand and collect released microrafts. The magnetic wand was composed of a hollow polycarbonate tube (43.5 mm length, 4.7 mm diameter), with one closed off end and one open end with threads on the outer surface. A cylindrical magnet (12.7 mm length, 3.175 mm diameter,) was inserted into the wand body, allowing free motion in the z-axis. The wand attached to the stepper motor via a custom thread adapter. A secondary magnet (10 mm length, 5 mm diameter) was mounted above the magnetic wand to push the magnet in the wand to the wand tip. The secondary magnet was attached with a 3D-printed custom clip. The pick-and-place system was secured to the microscope stage via custom aluminum mounts. The release device and pick-and-place system were controlled by a custom MATLAB program (MathWorks, Natick, MA). To operate the stepper motors, MATLAB controlled an Arduino Uno (SparkFun Electronics, Boulder, CO) outfitted with an Adafruit Motorshield (Adafruit Industries, New York, NY).

D. Isolating Non-Adherent Cells

Isolation of non-adherent cells using microrafts requires cell encapsulation within the microraft prior to collection. Cell encapsulation was performed by pipetting 3% type B bovine gelatin in PBS phosphate buffer solution (PBS, pH 7.4), prepared as previously described [24], onto microraft arrays at 37°C and centrifuging to seed the liquid gelatin into the microwells. The arrays were then placed at 4°C for 5 min, resulting in a sol-gel transition and encapsulating cells on each microraft. Using an automated release process, microrafts containing gelatin-encapsulated cells were released and collected with a magnetic wand. Once collected, microrafts were deposited into a 96-well plate containing Roswell Park Memorial Institute (RPMI) 1640 cell media (Gibco, Life Technologies, Grand Island, NY).

E. Post-Isolation Viability of Jurkat Cells

Post-isolation viability using the microraft array platform was evaluated based on cell expansion after single cell isolation. Jurkat cells were seeded on an array with 3 mL of cell media and imaged for 1 h. Microrafts containing single cells were identified using brightfield images and 36 microrafts were chosen for isolation. The selected cells were encapsulated in 3% gelatin at 4°C on the microraft array, individual microrafts containing the target cells were released, and the microrafts containing the gelatin-encapsulated cells were placed into a 96-well plate containing 200 μL of cell media (1:1 fresh media to conditioned media) per well [24]. The well-plate was incubated at 37°C for 3 weeks and wells containing an isolated microraft were manually screened for the presence of a cell colony. The results were then compared to a limiting dilution control group. In the limiting dilution group, Jurkat cells were seeded at a density of 0.5 cells/well into a 96-well plate [25], [26] containing 200 μL of media (1:1 conditioned to fresh media) per well and allowed to incubate for 3 weeks. Colony growth was manually screened and analyzed based on an assumed Poisson’s distribution of single cells in the well plate. The experimental and control group were performed in triplicate.

F. TCR α/β Sequencing

Ribonucleic acid (RNA) was isolated from 10,000 clonally expanded T cells. First-strand complementary deoxyribonucleic acid (cDNA) synthesis and polymerase chain reaction (PCR) amplification of the TCR α or β chain was performed using the SMARTer Human TCR a/b Profiling Kit (Takara Bio USA, Inc., Mountain View, CA) according to manufacturer protocols. TCR sequences were obtained with a custom sequencing primer complementary to the Illumina TruSeq HT Adapter (5’-TTCCCTACACGACGCTCTTCC-3’) and Sanger sequencing (Eton Bioscience, Inc., San Diego, CA). Sequence alignment was performed using IMGT/V-QUEST [27].

G. Cell Culture and Staining

Jurkat cells were incubated with 2 mM Hoechst 33342 for 30 min in RPMI 1640 complete media. The cells were then rinsed and added to 4 mL of complete media accompanied by 32 ng of Sytox Green. CD4+ T cells were incubated with 3 mM CellTracker Red for 30 min in PBS then in RPMI 1640 complete media for 30 minutes. The cells were rinsed and added to media containing 30 nM Sytox Green, 100 U/mL IL-2, 20 U/mL IL-7, 20 U/mL IL-15, and 100 ng/mL anti-CD3 monoclonal antibody.

H. Image Acquisition, Processing, and Analysis

Brightfield and fluorescence images were acquired by raster scanning each microraft array at specified time intervals. The images overlapped by at least 300 μm (sum of the width of a single microraft and the gap between microrafts) to ensure each microraft was fully captured in the data set. Images were acquired at either 6.4x or 8x total magnification, resulting in pixel sizes of 1.03 μm per pixel or 0.82 μm per pixel respectively.

Images were processed and analyzed in MATLAB in an automated, parallel pipeline during the time-lapse scan. Brightfield images were acquired to identify locations of individual microrafts. The brightfield images underwent flat-field correction to correct for uneven illumination, thresholding via Otsu’s method [28] to segment microrafts, and morphological filtering to remove incomplete microrafts and debris. The processed brightfield images were used to create a binary mask of the microrafts to define regions in which to extract fluorescence data and a unique index was assigned to each microraft. After each brightfield image was taken, images were captured in selected fluorescence channels. The fluorescence images underwent Wiener filtering to remove noise, Top-hat filtering [29] for background subtraction, and thresholding (Otsu’s method) to segment individual cells. Cells were then counted by finding local maxima with a peak finding algorithm. Fluorescence data (pixel locations and intensities) and cell count were saved for each indexed microraft.

I. Statistics

All statistics were calculated in GraphPad Prism. A Wilcoxon rank sum test (a non-parametric test of two independent samples) was used to test for a statistically significant difference between two groups in multiple experiments; comparing microraft isolation with and without gelatin, comparing cell viability of Jurkat cells after sorting with the microraft array platform and limiting dilution, and comparing the experimental and control groups in the T cell proliferation assays. Additionally, proliferating single CD4+ T cells were clustered using K-means clustering, analyzed using linear regression, and tested for a statistically significant difference using a one-way ANOVA followed by a Tukey’s multiple comparison test.

III. Results and Discussion

A. Overview of Assay and Platform

Developing a platform capable of assessing the functionality of helper T cells based on proliferation is particularly challenging due to the characteristics of primary immune cells. To identify T cells with receptors directed at tumor-associated antigens (TAAs), the system must be capable of assaying a large number of cells, as TAA-specific T cells occur at a frequency of 1 in 100 to 1 in 10,000 cells after multiple rounds of stimulation [30], [31]. Furthermore, T cells become exhausted when serially stimulated, leading to a decrease in proliferative capacity and viability. Due to the risk of exhaustion, the system must be able to sort at least 100 cells while maintaining a physiological environment, as less than 3% of human primary T cells are expected to be competent to expand after isolation [26]. Last, activated T cells divide approximately every 10–24 h so that cell numbers need to be measured every 12 h over a minimum of 48 h to assess proliferative behavior [21], [32]. The length of the proliferation assay thus requires a platform capable of maintaining cell viability for at least two days.

To meet these demanding requirements, an automated microscope with an integrated cell isolation system was developed to image immune cells on microraft arrays, identify proliferative cells and then isolate these cells (Fig. 1). The platform was used to investigate the proliferative responses of individual CD4+ T cells when stimulated by allogeneic EBV-LCL feeder cells. T cells were assayed on microraft arrays with feeder cells for 48 h, proliferative cells were identified, and proliferative T cells were collected. The isolated T cells were then expanded into colonies and underwent ELISA analysis to detect IFN-γ secretion and Sanger sequencing to ascertain their TCR α and β sequences.

Figure 1:

Process overview for CD4+ T cell proliferation experiments. CD4+T cells (red) were seeded on microraft arrays along with feeder cells (blue) and imaged for 48 h. Magnetic microrafts containing proliferative cells were identified, gelatin was overlaid onto the array to entrap cells on the microraft surface, and finally cells were isolated into a 96-well plate. The cells were then clonally expanded and underwent either Sanger sequencing to determine their TCR α and β sequences or ELISA to detect IFN-γ production.

To assay a sufficient number of cells given the expectation of T-cell exhaustion, the system was designed to image 6 arrays, tracking approximately 40,000 single cells in total (assuming 1/3 of microrafts contain a single cell), with the capability of imaging each array within 10 min, permitting 6 arrays to be imaged in 1 h. The total number of single cells that could be tracked was improved by 24-fold with respect to previous generations of microraft array platforms [26]. Furthermore, imaging was greatly enhanced in comparison to previous systems, by inverting and rebuilding a widefield microscope used for animal imaging: a unique setup of such instrumentation. Inversion of the microscope enabled imaging and counting of cells through a flat, transparent PDMS film, to eliminate the multiple surfaces with changing refractive indices [33]. A high quantum efficiency camera was integrated into the system to enhance light gathering capabilities and resolution. Further the system was enclosed in an incubation chamber that maintained 5% CO₂ levels and 37° C, allowing longitudinal studies of cells (≥ 4 d) while maintaining cell viability. An image processing pipeline was established to segment and count individual cells within indexed microrafts. To isolate cells, a microraft release device and 3-axis pick-and-place device was developed to quickly and robustly isolate individual microrafts in under 1 min. Each subsystem is fully described in the ensuing paragraphs.

B. Software

To track temporal characteristics of cells based on fluorescence markers, an automated image processing and analysis pipeline was developed with the capabilities of locating and segmenting microrafts and individual cells on the microrafts. Microrafts were segmented and indexed to identify the regions in which cell fluorescence data was to be collected and to provide a real-world positional address for each raft. Microrafts were segmented and indexed, with 99.99 ± 0.02% (n = 5 arrays, 19,600 microrafts each) of rafts successfully identified, demonstrating excellent segmentation accuracy. Next, the cell segmentation and counting were evaluated using Jurkat cells, a T lymphocyte-derived cell line. Jurkat cells were stained with Hoechst 33342, seeded on the arrays, and imaged at 30 m intervals over 1 h (3 scans). Imaging each array in brightfield and a single fluorescence channel required a total of 5.7 ± 0.5 min (n = 6 arrays, 19,600 microrafts each) at 6.4x magnification (spatial resolution of 1.03 μm per pixel). Cells were counted and assigned to each indexed microraft during the image acquisition and analysis process. Automated image analysis identified cells with 97.9 ± 1.9% sensitivity and 98.9 ± 0.9% specificity (n = 100 rafts, 3 arrays) as compared to manual cell counting (Table I). Both the microraft and cell detection capabilities of the system were shown to be extremely effective, enabling high-quality monitoring of labeled cells during time-lapse imaging and demonstrating the ability to isolate target cells of interest based on their real-world coordinates.

Table I:

Microraft and cell segmentation metrics.

Microraft segmentation sensitivity (%)	99.99 ± 0.02
Cell segmentation sensitivity (%)	97.9 ± 1.9
Cell segmentation specificity (%)	98.9 ± 0.9
False positives per microraft, cell count	0.01 ± 0.01
False negatives per microraft, cell count	0.02 ± 0.02
Rafts with correctly counted cells (%)	97.7 ± 1.5

C. Instrumentation and System Design

To isolate cells of interest from the microraft arrays, a microraft-release device and 3-axis pick-and-place system were implemented (Fig. 2, 3). The release device (Fig. 3c) was mounted below the microscope stage and vertically actuated a microneedle, piercing the PDMS array to dislodge a single microraft from the PDMS mold. The 3-axis pick and place robot (Fig. 3b) then positioned a magnetic wand—a hollow polycarbonate tube containing a cylindrical magnet with mobility in the z direction—above the location of the released microraft, capturing the element and associated cell. The wand (Fig. 3f) was then moved to position it above a single well in a 96-well plate, located above a block magnet. The wand was lowered vertically into a well with culture medium on the multi-well plate. The block magnet opposed the wand magnet, forcing the wand magnet upwards into the polycarbonate tube, and thereby releasing the microraft into the desired well (Fig. 3d–e, Supporting Fig. 1, Appendix A). During the microraft isolation process, images were acquired before and after each release attempt to verify the release of the target microraft. The change in average contrast between the two images was used to provide a metric by which the program decided whether the microraft had been successfully released or additional needle punctures were necessary (Supporting Fig. 2, Appendix A). This entire microraft release process was fully automated.

Fig. 2:

(a) Individual microraft array containing 19,600 releasable microrafts in a 140 × 140 pattern with each microraft measuring 100 μm × 100 μm × 187 μm (length, width, height of microwell) and having a 50 μm inter-raft gap. (b) Merged fluorescence and brightfield image of Hoechst labeled CD4+ T cells (magenta) on a subregion of a microraft array showing 9 microrafts. (c) Layout of the platform, displaying 6 mounted microraft arrays, an inverted microscope, microraft collection system, motorized stage, and a 96-well plate for cell collection.

Fig. 3:

Instrumentation overview. (a) Overall system model viewed from the side. (b) Pick-and-place system to move the magnetic wand during cell-microraft collection. (c) Release device to actuate the needle vertically for microraft release. (d) Collection of microraft (not to scale) from microraft array by inserting the magnetic wand into the medium overlying the raft array. (e) Release of microraft into well of a 96-well plate; the block magnet opposes the wand magnet, forcing the wand magnet into the upward position and releasing the microraft. (f) Magnetic wand assembly for collecting released microrafts. (g) Front view of system with cell incubator system in place. (h) Inside view of platform with a single microraft array and 96-well plate in place for imaging and microraft isolation.

Isolation times are comprised of three main components; stage movement times, release times, and collection times. Stage movements to place the microneedle below the targeted microraft required 1.8 ± 0.8 s on average (n = 60 microrafts). Successful release of a microraft required 6 ± 2 needle actuations through the PDMS mold requiring 1.3 ± 0.01 s per needle actuation (n = 60 microrafts). The large number of needle actuations was required to force the microraft with its cell cargo up through the gelatin layer. After each microraft release, the pick-and-place system collected the released microraft and deposited it into a 96-well plate with an average time of 21.8 ± 0.02 s per single raft collection (n = 60 microrafts; Fig. 3, 4). The majority of the collection time was due to wait steps in which the magnetic wand was positioned over the released raft for 8 s to capture the microraft and then subsequently positioned in the well-plate for 3 s to permit microraft deposition. The wait steps provide time for the microraft to move through the medium from the array to the magnetic wand during collection and for the microraft to move from the wand into the well-plate during deposition. The total time from target identification to release into a well-plate required an average of 45.0 ± 6.6 s (n = 60 microrafts). Target microrafts were isolated from arrays without gelatin at an isolation efficiency of 95.3 ± 3.1% (n = 3 arrays, 50 microrafts per array) and from gelatin-coated arrays at an efficiency of 87.1 ± 3.6% (n = 3 arrays: 42, 48, and 48 microrafts per array); a statistical difference was seen between coated and non-coated arrays (p < 0.01, Wilcoxon rank sum test). Isolation efficiency was defined as the ratio of rafts successfully isolated into a well-plate relative to the total number of attempts. A failed microraft isolation was primarily due to a raft not being fully released from the PDMS mold after several needle actuations. During the release process, the needle may fail to fully dislodge the microraft from the mold, which prohibits the raft from moving to the magnetic wand during the collection process. The rafts were deposited into the intended well at an efficiency of 100 ± 0% (n = 3 arrays: 50 microrafts per array). The automated platform demonstrated a significant improvement in performance compared to manual microraft isolation methods (Supplemental Discussion, Appendix A). Overall, the platform consistently performed well at releasing, collecting, and sorting microrafts in under one min per target.

Fig. 4:

Microrafts were released from the array by actuating a microneedle from below the array, displacing a single raft which was then collected using a magnetic wand. To collect non-adherent cell types, the cells were encapsulated in gelatin on the array before isolation. (A) Microraft isolation with cell media (pink) and gelatin (purple) adhering the target cell to the microraft. (B) A magnetic wand was extended into media on the microraft array for microraft collection. (C) The magnetic wand was retracted from the array, with the microraft collected on the tip. (D) Images of an isolated microraft and single Jurkat cell, labeled with Hoechst 33342 and CellTracker Red, in the well of a 96-well plate.

The platform was developed primarily for imaging and isolating non-adherent cell types, and in particular immune cells, which requires cells to be encapsulated in a hydrogel during the microraft release and collection process so that the cell remains on the microraft [24]. To test the efficiency of cell isolation using gelatin (cells successfully isolated/attempted), Jurkat cells (E6–1) stained with Hoechst 33342 and CellTracker Red were seeded on microraft arrays. The arrays were then coated in 3% gelatin to encapsulate and entrap the cells within the curved surface of the microraft. Next, microrafts containing a single Jurkat cell were identified and isolated into a 96-well plate, with 98.4 ± 1.4% (n = 3 arrays: 42, 48, and 48 microrafts per array) of isolated microrafts containing their original cell contents (Fig. 4). The platform successfully isolated cells with an overall single cell isolation efficiency of 85.7 ± 4.6% (n = 3 arrays: 42, 48, and 48 microrafts per array). Single cell isolation efficiency was calculated by multiplying the raft isolation efficiency by the percentage of rafts containing their original cell contents after isolation. Importantly, the raft isolation was the limiting factor in the overall single cell isolation efficiency, as nearly all successfully isolated microrafts contained the target cells of interest. By improving the microraft release and collection process, which is hindered by the gelatin coating, the overall isolation efficiency could approach 98%. Combined, the isolation metrics demonstrate that the platform isolated non-adherent cell types using hydrogel encapsulation with high efficiency.

D. Post-Isolation Cell Viability

An optimal cell isolation system is capable of sorting cells while maintaining high levels of cell viability. To evaluate the impact of the cell seeding and isolation process on cells, Jurkat cells were seeded on arrays, encapsulated in gelatin, and isolated into well-plates (Fig. 4a–c). First, Jurkat cells were seeded on microraft arrays at a 1:1 ratio of cells to microrafts. Cell encapsulation was then performed by pipetting 3% bovine gelatin in PBS onto microraft arrays at 37°C and centrifuging the array to load the liquid gelatin into the microwells. The arrays were then placed at 4°C for 5 min, resulting in a solid-to-liquid transition and encapsulating cells within a gelatin plug above each microraft. During cell overlay with gelatin, 99.2 ± 1% (n = 10 arrays, 240–245 microrafts per array) of cells remained on their original microraft after encapsulation. Single cells were then isolated into the wells of a 96-well plate using automated sorting. Post-sort cell viability of Jurkat cells was evaluated based on the clonal expansion capacity after isolation and compared to clonal expansion using limiting dilution. On average, 43.5 ± 12.2% (n = 3 arrays; 34, 35, and 30 single cells each) of Jurkat cells isolated from the microraft arrays expanded into colonies and 67.5 ± 14.4% of single cells expanded into colonies using limiting dilution (n = 3 plates, 16 single cells per plate) over the course of three weeks (p = 0.012, Wilcoxon rank sum test). While the microraft-array based isolation decreased the viability of Jurkat cells as compared to limiting dilution, a substantial proportion of the isolated cells maintained viability suggesting that microrafts will be suitable for collection of fragile cells such as T lymphocytes.

E. CD4+ T Cell Proliferation

Technologies for exploring T cell activation on a single-cell level have relied primarily on single-time point measurements of biomarkers or affinity to protein-HLA (pHLA) tetramers, with very few technologies able to measure T cell functionality and then sort single cells based on this outcome [18]. In particular, there is little focus on the activation state and antigen-specificity of CD4+ T cells due to their low avidity to pHLA tetramers and it remains extremely challenging to identify and isolate individual, antigen-specific, helper T cells based on proliferation [34], [35]. To address this vacancy and to demonstrate the capabilities of the microraft array platform, the system was used to analyze the proliferative response of CD4+ T cells in response to non-specific stimulation through patient-donor HLA mismatch [36], [37]. The proliferative response of CD4+ T cells to allogeneic feeder cell stimulation was investigated using time-lapse imaging of cells on microraft arrays.

To examine reactivity to HLA-mismatch, CD4+ T cells were stimulated with allogeneic EBV-LCLs [38], expecting a broad proliferative response [39]. T cells that became activated after encountering allogeneic EBV-LCLs respond to multiple allo-antigens arising from the mismatch in HLA molecules between the donor CD4+ T cells and recipient EBV-LCLs. T cells were stained with CellTracker Red, mixed with allogeneic EBV-LCL feeder cells (at 5:1 ratio), and seeded on microraft arrays (Supporting Fig. 3, Appendix A) at T cell:microraft ratio of 1:1. The feeder cell to microraft ratio was expected to follow a Poisson’s distribution based on previous work [22]; with a cell seeding of 5:1 of feeder cells to microrafts, 96% of microrafts were expected to contain 2 or more feeder cells and only 0.7% of microrafts were expected to contain zero feeder cells. Thus, nearly every microraft contained multiple feeder cells for T cell interaction and activation—a high cell density can be seen in Fig. 5d. Control arrays were prepared by seeding T cells on microraft arrays without EBV-LCL cells. In total, three biological replicates were performed in which one experimental array and one control array were prepared from primary human cells derived from a single donor sample. The arrays were then imaged every 12 h for 48 h. Each array was imaged in brightfield and two fluorescence channels (CellTracker Red and Sytox Green), requiring 12.1 ± 0.8 min per array at 8x magnification (n = 6 arrays). Sytox Green was included in the cell media to provide information regarding cell death rates on the microraft arrays during the experiments. Microrafts containing a single T cell at the initial timepoint were identified through automated cell counting using the CellTracker Red stain. A total of 10,967 single cells were identified across the three experimental arrays (T cells with feeder cells) and 14,015 single cells across the three control arrays (T cells without feeder cells) and changes in cell count were tracked throughout the imaging process.

Fig. 5:

Proliferation of primary T lymphocytes on the microraft arrays. (a) Representative heatmap of cell count for 75 example microrafts, initially containing a single CD4+ T cell and stimulated with allogeneic feeder cells, in which proliferation occurred. (b) Total CD4+ cell count increase for cells stimulated with allogeneic feeder cells over 48 h with or without stimulation on microraft arrays from Replicate #1: n,experimental = 3438 microrafts, n,control = 4759 microrafts. A significant difference was seen between the experimental and control groups (p < 0.001). (c) The average cell count for microrafts (Replicate #1) originally containing a single cell and showing no proliferation, low proliferation, medium proliferation, and high proliferation with allogeneic stimulation (k-means clustering, n = 3438 microrafts). The slope of each cluster was significantly different (p < 0.05) when compared using linear regression and a one-way ANOVA followed by a Tukey’s multiple comparisons test. (d) Image of a 9 × 9 and 3 × 3 section of an array with 100 μm wide microrafts at 8x magnification, plated at a T cell:feeder cell ratio of 1:5. (e-g) Images of example microrafts from each cluster in panel (c). A T cell (red) and surrounding allogeneic feeder cells are shown over 48 h, demonstrating low (e), medium (f), and high (g) levels of proliferation.

In the presence of allogeneic EBV-LCLs, 26.5 ± 6.5% of the T cells divided at least once while 17.4 ± 2.8% of T cells divided at least once without on-array stimulation (Fig. 5). There was a statistically significant difference in the change in cell count of microrafts originally containing a single cell between the paired arrays (experimental and control) in each biological replicate (p < 0.001, Wilcoxon rank sum test). Generally, comparing paired control and experimental arrays from each biological replicate—rather than the average from all three replicates—provides the most useful information, as lymphocyte populations vary based on the primary samples acquired from donors. Due to the absence of on-array stimulation, the control groups in all three replicates proliferated to a lesser degree than the experimental arrays as expected (Supporting Fig. 4–5, Appendix A) [32].

Microrafts with proliferative cells were then clustered into non-proliferators, low proliferators, medium proliferators, and high proliferators using k-means clustering (Fig. 5c) for the first biological replicate to demonstrate the feasibility of grouping T cells based on proliferative capacity. Linear regression analysis was applied to the four clusters and a statistically significant difference was seen between the slopes (p < 0.05 between none and low; p < 0.01 between low and medium; p < 0.001 between medium and high) as tested by a one-way ANOVA followed by Tukey’s multiple comparisons test, indicating that cells can be grouped based on proliferation levels. Identifying groups of T cells with similar proliferative capacity could be used to further dissect the response of lymphocytes to antigenic-stimulation in future work.

F. Tracking CD4+ T Cell Proliferation

To assess the system performance and verify the accuracy of the proliferation results, the automated cell count from the experiments was compared to manual cell counts. Using CellTracker Red (T cells) and brightfield images, a custom GUI was developed to display spatially and temporally random microrafts to the user, accept a manual cell count, and then compare the value to the automated cell count for that microraft. T cell counting was achieved with a sensitivity of 92.1 ± 2.7% and a specificity of 85.2 ± 4.8% (n=3 arrays, 100 rafts per array) compared to manual counting. The decrease in sensitivity and specificity of counting T cells as compared to Jurkat cells is likely due to the wide range of normal primary T cell shapes and sizes as well as loss of the Cell Tracker Red dye with each cell division. On average, the automated cell counting program correctly counted the number of cells in 89.0 ± 1.7% of rafts, with 0.08 ± 0.02 false positives per raft and 0.06 ± 0.02 false negatives per raft (n = 3 arrays, 100 rafts per array). The platform’s accuracy in tracking cell count throughout long time-lapse imaging should in the future enable dissection of the T cell proliferative response.

G. Isolation of Alloreactive CD4+ T Cells

To fully harness the power of a T cell functional assay for immunotherapy research, single or clonal T cells must be isolated for additional characterization and TCR sequencing. The isolation capabilities of the system were demonstrated by imaging proliferative T cells on microraft arrays with allogeneic-feeder cell stimulation and then collecting proliferative T cells. After imaging for 48 h, the cells were incubated for an additional 48 h on the arrays to allow each identified clonal colony to increase in cell number prior to isolation. The automated system tracked the proliferation of 4,534 microrafts originally containing a single cell during the imaging process; a total of 18.99% of the T cells divided at least once. The platform provided a target list of proliferative T cells based on total increase in cell count. The list of target cells was provided to the viewer who then selected a final smaller list of target cells for isolation (180 microrafts). The arrays were then removed from the system, cells were encapsulated in 3% gelatin at 4° C, and the arrays were returned to the system for automated isolation. The microrafts were released, collected, and deposited into a 96-well plate containing allogeneic EBV-LCLs and donor peripheral blood mononuclear cells (PBMCs) to stimulate and promote further expansion of the isolated cells. A total of 130 single microrafts containing proliferative T cells were successfully isolated. After 3 weeks of stimulation, cells from 7 microrafts (5.4%) expanded into colonies—a similar expansion rate observed in previous primary T cell studies [26].

H. Post-Sort T Cell Characterization

Cytokine production continues to be an area of interest in immunology as cytokines play a key role in regulating the immune response in the tumor microenvironment and can be investigated through immunostaining assays [40], [41]. As such, samples from four of the expanded T cell colonies isolated from the microraft arrays were assessed by ELISA for IFN-γ secretion. From each colony, 100,000 cells were challenged overnight with allogeneic EBV-LCLs and then measured for IFN-γ secretion. Of the three sampled colonies, none showed specific IFN-γ release (Supporting Fig. 6, Appendix A). However, the non-specific IFN-γ secretion of the three colonies was not unexpected as the cells experienced multiple rounds of stimulation over the course of 3 to 5 weeks. Continual T cell stimulation over an extended time period can lead to T cell exhaustion and dysfunction, including a loss of cytokine secretion [42][43][44]. The assay demonstrated that while the cells can be immunoassayed after isolation and expansion, the results may not provide useful information.

In the majority of single cell T lymphocyte research, sequencing the TCR α and β genes is arguably the ultimate goal, as the TCR genes determine antigen-specificity and can provide the tools to redirect T cell activity to selected target peptides and subsequent development of T cell therapeutics. As a proof of concept, the TCR α and β genes of the isolated allogeneic-stimulated T cells were sequenced. RNA was isolated from approximately 10,000 cells from each of the expanded 7 colonies and used for TCR α and β sequencing. Complete sequences for the TCR α V, CDR3, and J regions were obtained for 6 of the 7 clones and complete TCR β V, CDR3, and J regions were obtained for 7 of the clones (Supporting Table I, Appendix A). The β chain of the TCR provides the unique label for the T cell as only a single β chain can be expressed by a cell due to allelic exclusion while multiple α chains can be expressed by the same cell [45]. Along with sequence information, this data also demonstrates the mono-clonality of the T cells isolated from each microraft. The microraft array platform produces a reliable system for identifying TCRs from antigen-responsive, proliferative T cells, providing a powerful tool for immunotherapy research.

IV. Conclusions

Single cell research has propelled advances in scientific and industry research, especially in the field of cancer immunotherapies. The response of individual T cells to cancer-associated antigens has been of particular importance, as it determines the efficacy of cell-based therapies. While methods exist to identify T cells specific to cancer-associated antigens through peptide-HLA tetramers, a robust method for assessing T cell functionality and isolating single cells or colonies has yet to be developed (Supporting Table II, Appendix A). Here we describe a technology to determine the functional activity of individual helper T cells and to isolate cells based on proliferative capacity. A majority of cancer research focuses on cytotoxic T cells; however, helper T cells play a vital role in the clinical efficacy of T cell therapies. The described automated platform can be used to identify antigen-specific helper or cytotoxic T cells, based on their activation and proliferative profile, and isolate the cells for TCR sequencing. The system is capable of robust time-lapse imaging, producing accurate temporal cellular metrics, and single/multi-cell isolation, while retaining high cell viability. The T cell proliferation assay provides a proof of principle for future immunotherapy research in which T cells could be stimulated by feeder cells presenting cancer-antigens, paving the way for identifying tumor-antigen specific CD4+ T cells. Very few cancer-antigen-specific TCR sequences for Class II HLA molecules have been identified to date and the automated microraft array platform has the potential to open the doors for TCR discovery.