Microraft array-based platform for sorting of viable microcolonies based on cell-lethal immunoassay of intracellular proteins in microcolony biopsies

Abstract

The majority of bioassays are cell-lethal and thus cannot be used for cell assay and selection prior to live-cell sorting. A quad microraft array-based platform was developed to perform semi-automated cell sampling, bioassay, and banking on ultra-small sample sizes. The system biopsies and collects colony fragments, quantifies intracellular protein levels via immunostaining, and then retrieves the living mother colonies based on the fragments’ immunoassay outcome. To accomplish this, a magnetic, microwell-based plate was developed to mate directly above the microraft array and capture colony fragments with a one-to-one spatial correspondence to their mother colonies. Using the Signal Transducer and Activator of Transcription 3 (STAT3) model pathway in basophilic leukemia cells, the system was used to sort cells based on the amount of intracellular STAT3 protein phosphorylation (pSTAT3). Colonies were detected on quad arrays using bright field microscopy with 96 ± 20% accuracy (true-positive rate), 49 ± 3% of the colonies were identified as originating from a single cell, and the majority (95 ± 3%) of biopsied clonal fragments were successfully collected into the microwell plate for immunostaining. After assay, biopsied fragments were matched back to their mother colonies and mother colonies with fragments possessing the greatest and least pSTAT3/STAT3 were resampled for expansion and downstream biological assays for pSTAT3/STAT3 and immune granule exocytosis. This approach has the potential to enable colony screening and sorting based on assays not compatible with cell viability, greatly expanding the cell selection criteria available to identify cells with unique phenotypes for subsequent biomedical research.

Graphical Abstract

A magnetic, microwell-based plate was established enabling automated collection of colony biopsy fragments into known locations for immunofluorescence assays.

1. Introduction

Intracellular proteins are involved in almost all cellular processes, including development, proliferation, differentiation, stress responses, and apoptosis.¹ Analysis of intracellular signaling networks and intracellular protein states in live cells is required to monitor changes in cellular physiology such as that occurring during infection or mutation accumulation. Protein labeling methods are extensively used to measure intracellular protein abundance, protein co-localization, and post-translational protein modification within cells.² In particular, fluorescence-based methods such as gene-engineered proteins or tagged antibodies have been widely utilized to identify intracellular proteins.^3,4 Fluorescent protein indicators or protein fusions are a powerful approach, but require genetic alteration of cells and may alter the function of the intracellular protein as a consequence of the addition of a large fluorophore to the target protein.^5,6 In contrast, the use of exogenous probes such as fluorophore-labelled recognition antibodies or ligands employs readily available reagents.⁷ However, most exogenous probes such as antibodies cannot cross the intact plasma membrane and, thus, require cell-lethal fixation with membrane permeabilization so that the probes might access the cell interior to label an intracellular protein.³ As a result, a major drawback to intracellular immunofluorescence assays is that they are typically end-point assays without straightforward options to relate single cell fluorescence measurements to downstream biological studies requiring viable cells.

Current strategies to utilize cell-lethal assays for phenotype selection typically employ large scale-up culture efforts (>10⁵ to 10⁶ cells) after cell cloning, for example by limiting dilution or cloning ring. The scaled up culture can then be subsampled and employed in an assay that destroys cell viability since the macro-culture remains available for re-sampling.⁸ These cloning methods are laborious and time-intensive, requiring weeks to months to grow up a sufficiently large number of cells from a single founder cell during which significant phenotypic drift might occur. Maintenance of the large-scale mother colonies during the time of the cell-lethal assay on the subsample also requires significant manpower and time. Another drawback is that in all of these methods cells are initially cultured at extreme dilution which can impose stress due to lack of autocrine and/or adhesion-initiated signaling from other cells.^9,10 Thus, there is a need for cell sorting methods that permit higher throughput sampling while keeping the cells at reasonable density.

Automated systems can address some of the limitations of these macro-culture systems by eliminating manual processing, improving speed, throughput, and scalability, and enhancing reproducibility. Many approaches exist to automate the individual processes of cell culture scale-up, culture splitting, and cell banking. These technologies combine multiple steps, for example, automated liquid-handling, cell culture, and reagent application, while also gathering cell count measurements from microscopy images.¹¹ For example, the CellCelector technology combines imaging software with colony harvesting using a glass capillary with subsequent cell dispensing into a multi-well plate.¹² Despite improvements in throughput made by automated systems, long culture times are still required to propagate cells in hundreds to thousands of individual macro-wells to increase cell numbers for cell splitting and downstream assay. The miniaturization of assay volumes afforded by microfluidics, microarrays, and other micropatterned devices has the potential to address these challenges while also enhancing throughput of cell cloning and sampling. Previously, we developed an automated microraft array-based system for colony biopsy and cell sorting utilizing automated microscopy, computer vision, and smart algorithms for cell identification and colony tracking.¹³ The foundation of the system, quad microraft arrays, were comprised of clusters, or ‘quads’, of 2 × 2 superparamagnetic polystyrene cell microcarriers, or ‘microrafts’, within microwells on a polydimethylsiloxane (PDMS) substrate.¹⁴ The quad microraft arrays supported hundreds of discrete micro-scale colonies in culture in a small media volume while simultaneously enabling the biopsy, release and collection of biopsied fragments from each colony. Miniaturization of colony culture sites and biopsy regions permits the sampling of colonies with fewer than 100 cells, whereas standard macro-culture methods (described above) typically involve efforts to generate >10⁵ to 10⁶ cells prior to sampling.⁸ By reducing the colony (or culture) size that can be subsampled, the cell culture duration and manual labor required to maintain the cells is also minimized. However, this system (and other technologies) lacked the capabilities to efficiently use the fragments for a destructive bioassay or to match assay results from the daughter fragments back to their microscopic mother colonies. Thus cell sorting using complex sort criteria such as that enabled by cell-destructive assays was not possible on this system. Other related methods including micropallet¹⁵ and microcup¹⁶ arrays face similar limitations in addition to being challenging to automate. Microengraving technologies permit the matching of an assay result back to the original cells, however, isolation of the cells based on the assay was very slow and challenging to automate.¹⁷ Although microarray technologies can increase cell-based assay throughput, there remains a significant need for a robust micro-assay platform that combines cell cloning and culture with colony sampling, collection, and assay in parallel.

A semi-automated pipeline was developed to support parallel microcolony growth and sub-sampling followed by biopsy of a small portion of each colony for use in a cell-lethal assay (i.e. immunoassay of intracellular proteins). The outcome of the assay was then used to identify the desired microscopic mother colonies for collection and propagation based of the ratio of pSTAT3/STAT3. Selected mother colonies could undergo a broader range of assays to understand how the ratio of pSTAT3/STAT3 might correlate with proliferation and immune granule release. A microwell collection plate was designed and optimized to interface with the quad microraft array platform to enable automated collection of biopsied colony fragments into known locations on the collection plate. Methodologies for immunostaining cells within the collection plate were developed, and automated software identified the immunostained cells. Finally, customized software matched the assayed daughter colony fragment back to its original mother colony permitting colony sorting based on the outcome of the cell-lethal assay of the biopsy fragment. Using this technology, we assessed analytical benchmarks of the semi-automated pipeline: colony detection accuracy, colony biopsy efficiency, and fragment collection success. As a demonstration of the technology, the phosphorylation of an intracellular protein, STAT3, in rat basophilic leukemia (RBL) micro-colonies was measured in biopsy fragments and the viable cell colonies sorted according to their differing levels of phosphorylated protein (pSTAT3). pSTAT3 in the biopsied sample was then correlated with the phenotype of the living mother colony. This work establishes a platform permitting live-cell colony screening and sorting based on immunofluorescence of intracellular proteins and should be readily expandable to the use of many other cell-destructive assays as a cell sorting parameter.

2. Experimental

2.1. Materials and reagents

Poly (dimethylsiloxane) (PDMS) Sylgard 184 silicone elastomer kit was purchased from Dow Corning (Midland, MI) and poly (acrylic acid) (PAA) (MW 30,000, 30% in H₂O) from Polysciences Inc. (Warrington, PA). Methanol, bovine serum albumin (BSA), and Triton X-100 were from Fisher Scientific (Hampton, NH). RBL-2H3 (CRL-2256) and H1299 (CRL-5803) cells were purchased from American Type Culture Collection (ATCC) (Manassas, VA). Penicillin/streptomycin (P/S), Minimal Essential Medium (MEM) Alpha, Roswell Park Memorial Institute (RPMI) medium and 0.05% trypsin-Ethylenediaminetetraacetic acid (EDTA) were obtained from Life Technologies (Waltham, MA). Fetal bovine serum (FBS) was from Atlanta Biologicals (Norcross, GA) and sodium azide (NaN₃) from VWR (Radnor, PA). Primary antibodies for STAT3 (ab119352) and STAT3 phosphorylated at Y705 (ab76315) were purchased from abcam (Cambridge, MA). Secondary antibody with Alexa Fluor 647 (715-605-150) was from Jackson ImmunoResearch Inc. (West Grove, PA) and with Alexa Fluor 555 (A-31572) from Thermo Fisher (Waltham, MA). Hoechst 33342 (62249) was from Thermo Fisher. Tween-20 was obtained from Sigma-Aldrich (St. Louis, MO). Phosphate buffered saline (PBS) and rat tail collagen type I were purchased from Corning (Corning, NY).

2.2. Microarray Fabrication

Quad microraft arrays (21.4 × 21.4 mm) comprising 10,000 clusters of 2 × 2 microrafts (‘quads’) were fabricated using previously reported methods.¹³ Each 79 × 79-μm microraft was embedded within a 60 μm-deep microwell. Microrafts within the same quad were separated laterally by 25 μm-wide walls of PDMS, whereas each quad was separated laterally by a 31-μm barrier of PDMS with a height 60 μm above the microrafts.

Microwell collection arrays (21.4 × 21.4 mm) comprising 2,500 microwells were designed to match one microwell to every four (2 × 2) quad sites on the microraft array. Each microwell was 183 × 183 μm wide and separated laterally by a 31 μm-wide section of PDMS with a wall height of 175 μm. The arrays were molded from photoresist micropillar array templates on glass which were fabricated using one-layer photolithography as previously described.¹⁸ PDMS microwells were molded from the template onto a #2 glass coverslip (25 × 25 mm wide, and 0.23 mm thick) by modification of a previously reported patterning method.¹⁹ To produce the mold, a thin layer of PAA was spin-coated onto a glass slide and the coverslip positioned in the center. The slide-coverslip was baked at 95 °C for at least 2 hrs. PDMS (10:1) was degassed onto the template and sandwiched together with the slide-coverslip. After curing at 95 °C for 45 min, the slide-coverslip was demolded from the template and placed in a 70 °C water bath overnight to dissolve the PAA, thus releasing the coverslip comprising adhered PDMS wells from the slide.

2.3. Hardware Design

A disk magnet (3 × ½”) was purchased from K&J Magnetics (Pipersville, PA) and inverter-power electroluminescent (EL) sheet from Electro Luminescence Inc. (Aromas, CA). Original designs for the microraft array cassette, collection base plate and snap-on fastener, and magnet/EL light sheet holder were designed using SolidWorks (2016, Dassault Systems, Waltham, MA) and Computer Numerical Control (CNC) g-code was generated using the SolidWorks CAM plug-in (Autodesk HSMXpress, San Rafael, CA). Designed components were milled from clear polycarbonate sheets acquired from McMaster-Carr (Douglasville, GA) using a MicroMill DSLS 3000 milling machine from MicroProto Systems (Chandler, AZ).

2.4. Automated Microscopy

Bright field and fluorescence microscopy scanning was performed using an IX81 microscope (Olympus Corporation, Tokyo, Japan) equipped with a MS-2000 motorized stage (ASI, Eugene, OR) and Flash 4.0 V2 camera (Hamamatsu, Shizuoka, Japan). The microscope was situated within a Plexiglass incubator to maintain a 37 °C temperature, 60% humidity and 5% CO₂ during all cell experiments. The objectives used were an Olympus UPLFLN 4× (NA 0.13) and 10× (NA 0.3). Fluorescence filter cubes utilized included: Chroma ET-DAPI 49000, Semrock TxRed-4040B, and Chroma ET-Cy5 49006. Microscopy was automated using a MATLAB (2018A, MathWorks, Natick, MA) graphical user interface (GUI) that controlled the automated components using Micro-manager’s Java API, as previously described.¹³

2.5. Quad Microraft Image Analysis

Customized MATLAB image-processing pipelines (Fig. S1) were used to detect cellular material on quad microraft arrays from the bright field microscopy images. Microscopy was performed before cell seeding to measure the background noise and then every 2 days after seeding to track cell locations. To enhance image contrast, images were acquired at 4× magnification (NA 0.13) at two different focal planes: i) focal plane of in-focus cells, ii) 58 μm below the in-focus plane (outside of the objectives’ 25 μm depth of focus). First, images across all time points were flat-field corrected, illumination corrected, and registered. Images were morphologically processed to segment the microrafts and quad sites, and a digital identifier in the form of row and column numbers was applied to each microraft and quad site across the array grid.¹³ Background subtracted images from the two focal planes were used for cell segmentation of micro-colonies via intensity thresholding at an empirically set threshold and morphological processing to exclude non-cellular objects based on area, major and minor axis length, eccentricity, solidity, and orientation. The segmented images were combined and cellular area quantified. Colonies suitable for biopsy were identified as having ≥2 microrafts within the quad colony site possessing ≥25% of the microraft surface area covered by cells. To aid in the identification of clonal quad sites, a MATLAB GUI was developed. For each clonal colony, the automated system was set to select one microraft to biopsy, chosen as the microraft with the least cellular outgrowth onto the adjacent PDMS borders.¹³ Proximity filtering²⁰ of the chosen microrafts was implemented by iteratively identifying and eliminating microrafts that possessed too many neighboring microrafts with cells (<1070 μm away) and repeating the process until the nearest neighbors of all remaining microrafts selected for tracking were ≥1070 μm distant from one another. The automated system was set to perform the first biopsy attempt at the side of the target microraft with the least cell coverage on the adjacent PDMS borders, and subsequent attempts at the centroid of the microraft.¹³

2.6. Cell Culture

Wild-type RBL-2H3 rat basophilic leukemia cells and H1299 human non-small cell lung carcinoma cells were used in this work, in addition to wild-type H1299 cells altered to stably express green fluorescent protein (GFP) as described in previous work.²¹ RBL-2H3 and H1299 cells were cultured on rat tail collagen type I (0.0433 μg/mL) coated substrates in MEM and RPMI medium, respectively, supplemented with 10% FBS and 1% P/S. For subculture, the cells were detached from the substrate using 0.05% trypsin-EDTA. When seeding RBL-2H3 cells at low cell numbers (≤ 1,000 cells), cells were cultured in 50% v/v conditioned medium for the first 2 days with the medium replaced every other day. Conditioned medium⁹ was prepared by collecting the supernatant from the bulk RBL-2H3 cultures and filtering the supernatant through a sterile 0.22 μm pore filter (Fisherbrand #09-720-004). Colony fragments biopsied from mother colonies were cultured an additional 2 days prior to assay. Cells were limited to 12 passages post-authentication.

2.7. Immunofluorescence Staining

Cells were fixed with 60% methanol in complete MEM for 15 min at −20 °C and washed with 3% BSA. After fixation, cells were then permeabilized with 0.5% Triton X-100 for 20 min and washed with 3% BSA. Permeabilized cells were blocked with 1% BSA in an immunofluorescence buffer (IFB, 0.05% Tween-20, 0.2% Triton-X 100, 0.1% BSA, and 0.05% NaN₃). After 60 min, cells were incubated with 1:200 STAT3 and STAT3-phosphoY705 in IFB overnight at 4 °C. The next day, cells were incubated in 1:500 Alexa Fluor 555 and 647 secondary antibodies in IFB for 45 min. The cells were washed with 1× PBS and incubated with 0.16 μM Hoechst 33342 for 15 min. Unless otherwise specified, solutions were prepared v/v in 1× PBS (pH 7.4) and incubation steps performed at 20–25 °C.

Biopsied and collected colony fragments were cultured for an additional 2 days prior to immunofluorescence staining. A magnet was also placed below the collected fragments throughout the assay to hold the samples in place.

2.8. Collection Array Imaging and Analysis

For image acquisition, regions of interest (ROIs) on the microwell collection array possessing microrafts with cells were automatically located from low-magnification (4×) bright field microscopy scans of the entire array. Microrafts were identified by intensity-thresholding bright field images followed by morphological filtering based on object area, major axes lengths, and minor axes lengths. To minimize the total number of ROIs, the ROIs were automatically combined whenever more than one microraft would be circumscribable by one ROI. Fluorescence microscopy images of identified ROIs were acquired at 10× magnification for Hoechst 33342, pSTAT3, and STAT3 stains. To correct for uneven illumination, ROIs were adjusted so that the microraft(s) were located within a central region (75% the size) of each acquired image or field of view (FOV) (100% FOV=1329 μm × 1329 μm). The selection of FOV positions was automated by selecting the FOV that encompassed the most colonies, assigning the encompassed colonies to the FOV, and iteratively repeating until all colonies were assigned an FOV.

Hoechst 33342 fluorescence was used to identify the correct focal plane for fluorescence imaging of the collection array. Focusing was performed using a published software autofocus routine that utilized a modified Laplacian focus measure.^22,23 Fluorescence intensities were measured as mean raw intensity values. Cells obscured by microrafts were excluded from the analysis to reduce the impact of microraft autofluorescence on intensity measurements. The location of the wells with cell colonies was used to match the cells back to their mother colony on the quad array by manually registering stitched images of the quad array before and after microraft release (Supplemental methods).

2.9. Statistics

Statistical analyses were performed using GraphPad Prism 8.2.0 software (San Diego, CA). D’Agostino-Pearson omnibus, Shapiro-Wilks, and Kolmogorov-Smirnov normality tests were performed on all data. A Mann-Whitney test (unpaired, two-tailed, nonparametric t-test) was used for two-variable comparisons. For multiple comparisons, a Kruskal-Wallis and Dunn’s multiple comparisons test (unpaired, two-tailed, nonparametric tests) were performed to compare the mean of each condition to the mean of every other condition (p<0.05 = *, p<0.01 = **, p<0.001 = ***). Unless otherwise noted, measurements are reported as the mean ± sample standard deviation.

3. Results and Discussion

3.1. Overview of Colony Culture, Biopsy, Collection, and Fragment Assay

A semi-automated pipeline was developed to enable micro-colony separations based on immunofluorescence assay of intracellular protein (Fig 1). The pipeline employed a novel microarray platform to biopsy colonies, collect colony fragments, assay the biopsy fragments, and match the assay result from the fixed daughter fragment to the living mother colony. Quad microraft arrays which are arrays of microwells containing 2 × 2 clusters (‘quads’) of superparamagnetic, releasable cell culture elements, served as the foundation of the platform (Fig. 2A).¹⁴ Cells were plated and cultured on quad arrays and the colonies tracked over time utilizing bright field microscopy. A microwell collection plate for collecting the biopsied colony fragments (~40 cells per biopsied fragment) was mated directly above colonies on the quad array with a one-to-one spatial correspondence (Fig. 1). The biopsy fragments were assayed by immunofluorescence staining after cell fixation and the mother colonies sorted based on the properties of the daughter micro-colony fragment. Sorted mother colonies were expanded for downstream assays to investigate additional phenotypic characteristics of the cells.

Fig. 1.

Overview of workflow for the quad microraft arrays with biopsy of colony fragments. (A) Single cells (red and green objects) were seeded onto a quad microraft array. Shown is a side view through the array. Each quad site contains four magnetic microrafts (only two are shown in brown on this side view). The light tan marks the PDMS substrate. (B) The cells were cultured into clonal micro-colonies (2 red colonies and 1 green colony are depicted). (C) A PDMS-on-glass microwell collection plate (light gray) was mated directly above the quad array, and colonies were biopsied by microneedle-based release of a single microraft within a quad site. The microrafts with their attached biopsied cells or colony fragment were magnetically collected into the overlying collection plate and cultured for 2 days. (D) After colony biopsy expansion, the collection plate is detached from the quad array. Immunofluorescence staining of select intracellular proteins (marked by yellow sunburst) is performed on cells in the microwell collection plate after fixation. (E) The mother colonies which when biopsied demonstrated the selected trait are released from the array. (F) The collected mother colonies are assayed or expanded for later use.

Fig. 2.

Colony growth across quad microraft arrays. (A) Image of a quad microraft array mounted in a culture cassette (left-side) and a close-up view of a quad microraft array (right-side), where the scale bars are 20 mm and 400 μm, respectively. (B) Example of clonal colony expansion across a quad site where a single cell resides on the site immediately after cell seeding. Scale bar is 100 μm. (C) Semi-logarithmic plot showing the number of quad colony sites located on an array (306 colonies located over 10,000 total quad sites) plotted against the number of microrafts with cells in each quad colony site. The threshold area for a microraft to be considered as occupied by cells was ≥25% surface area.

3.2. Cell Colony Tracking on Quad Microraft Sites

Image analysis scripts were developed to identify and track cells as they grew into clonal colonies on quad microraft arrays. To validate the methods, RBL-2H3 cells were seeded on a quad microraft array at low density and were tracked over time using bright field microscopy (Fig. 2B). Images acquired immediately after cell plating were used to identify the quad microraft sites that received only a single cell on the entire quad region. The colonies were imaged by bright field microscopy at two different focal planes on each of 5 days (Fig. S1).¹³ To measure colony area, background subtracted images from the two focal planes at day 5 were used to segment cells and then the two images were combined into a single segmented image. Compared to colony detection performed by hand, the two-focal plane method resulted in true positive, true negative, false positive, and false negative identification rates of 96 ± 20%, 90 ± 17%, 4 ± 20%, and 10 ± 17%, respectively (N=35 technical replicates, 185 total colonies). To assess the feasibility of identifying clonal colonies, cells were seeded onto the array and tracked over time. After 5 days of culture, 307 colonies were detected on the microraft array with 74 of these colonies covering ≥2 microrafts (Fig. 2C). A microraft was considered to have a cell colony if cells occupied ≥25% of the microraft surface area. Of the 74 colonies with cells covering ≥2 microrafts, 38 (51%) were clonal (i.e. originating from a single cell at time zero) and the remaining 36 (49%) originated from 2 cells (17, or 23 %), >2 cells (2, or 3%), or an indeterminate number of cells (11, or 15%). A minority of colonies starting with a single cell appeared to become cross-contaminated by cells not clonal to that colony, as a result of cell migration from other colony sites on the array (6, or 8%) (Fig. S2). On average, 49 ± 3% of colonies on quad microraft arrays were identified as clonal (N=3 experimental replicates, 627 total quad colony sites). In a separate experiment, cross-contamination of colonies resulting from cell migration was quantified by seeding co-mixed wild-type H1299 and GFP expressing H1299 cells on a quad microraft array and measuring the number of non-GFP expressing H1299 colonies contaminated with GFP-expressing cells after 5 days in culture. Over the 5 days, 6 ± 8 % of non-GFP expressing colonies were contaminated with GFP expressing cells (N = 10 technical replicates, 116 total quad colony sites). The image analysis pipeline enabled the exclusion of non-clonal colonies when >1 cell was present at time zero or when a colony was contaminated by cells not clonal to that colony. However, cross-contamination might still occur from cells breaking free from a colony and then settling and attaching onto another colony. More frequent imaging times or seeding cells at lower density would likely address this challenge. The power of this image analysis method was that cells could be identified and tracked on microraft arrays from bright field microscopy images, without the use of cell-modifying fluorescent probes.

3.3. Selection and Biopsy of Target Cell Colonies

A colony screening pipeline utilizing bright field microscopy images was developed to select colonies on a quad site for biopsy as well as the optimal microraft within that quad site to release. To enhance colony tracking post-release, only microrafts with colonies that were ≥1070 μm (or 10 microrafts) from another colony were biopsied. Of the 38 RBL-2H3 cell clonal colonies identified as spanning ≥2 quad microrafts, 30 of the colonies were well separated from adjacent colonies. For these 30 colonies, the quad microraft with the smallest number of cells bridging onto the adjacent PDMS borders was targeted for release (Fig. S1).¹³ The goal was to minimize the number of intercellular contacts that needed to be disrupted to successfully release a microraft with attached cells to biopsy the colony. Biopsies were performed using a computer-actuated microneedle with computerized tracking of the microraft (Supplemental methods). The computer actuated the needle and then tracked the location of the microraft. The needle was actuated repeatedly until the microraft was no longer detected in the FOV (i.e. released from the quad array). To biopsy a microraft with adhered cells, the microneedle pierced through the PDMS base of the microraft array (2 mm/s needle actuation speed¹³) applying a mechanical force to the bottom of the microraft. The microraft itself is a hard plastic (polystyrene) and is not pierced by the needle during microraft ejection. Thus, the only forces experienced by the cells are the fluid flow forces as the microraft is ejected. Prior work has demonstrated that multiple needle actuation attempts do not impact cell physiology.^{13, 24} Under automated biopsy conditions, 76 ± 13% of microrafts with cell colonies spanning across multiple adjacent microrafts were successfully biopsied in less than 12 attempts requiring only 5 ± 3 needle actuations (N=3 experimental replicates, 207 total microrafts). Off-target release of an adjacent microraft in a quad was 0.7 ± 0.6% (N=3 experimental replicates, 207 total microrafts). Microrafts not successfully biopsied with the automated system (within 12 attempts) could be released by manual needle actuations, resulting in 100 ± 0% of all microraft targets biopsied (N=3 experimental replicates, 207 microraft total). For the biopsied arrays, 100 ± 0% of the mother colonies (30 total colonies) remained attached to the quad microraft array. Using the developed pipeline, microrafts with adhered clonal colony fragments were successfully detected and biopsied from quad microraft arrays.

3.4. Collection of Released Biopsies into Microwell Plate

To collect released biopsy fragments efficiently and without manual intervention, a microwell collection plate was developed to mate with the quad microraft array and capture the released microrafts containing a colony fragment. The collection plate was comprised of three interlocking parts: i.) a base plate, ii.) a microwell collection array, and iii.) a snap-on fastener (Fig. 3A). The base plate served as the foundation or support system for the microwell collection array while the snap-on fastener held the microwell collection array in place on the base plate. During collection of the microrafts with attached cells, the assembled platform consisted of: i.) a microraft array mounted within a cassette, ii.) a collection plate, and iii.) a disk magnet with mounted electroluminescent (EL) light sheet (Fig. 3B and 3C). The disk magnet supplied the requisite magnetic force to attract the magnetic microrafts upward into the collection plate while the EL light sheet placed between the magnet and collection plate provided illumination to visualize microraft collection.

Fig. 3.

Platform for retrieval and assay of biopsied colony fragments. (A) Microwell collection plate components: i) base plate, ii) microwell collection array, iii) snap-on fastener, and iv) assembled microwell collection plate. Scale bar is 3 cm. (B) Exploded view schematic of the entire assembly used during colony biopsy and collection (left side). From top to bottom: disk magnet, EL light sheet, inverted collection plate housing the microwell collection array (blue), and quad microraft array (orange). Top right panels: ESEM micrographs of microwell collection array and a close-up of a sectioned microwell. Lower right panels: ESEM micrographs of quad microraft array and a close-up of sectioned quad sites. Scale bars are 200 μm. (C) Steps involved in assembly of the mated device: i) quad microraft array with cultured microcolonies, ii) microwell collection plate inverted and situated on top of the quad microraft array, iii) the magnet and EL light sheet mounted in their holder is secured on top of the collection plate and quad microraft array, and iv) microwell collection array detached from the mated components (panel iii.) and transferred into another chamber for assay. Scale bar is 3 cm.

Prior to colony biopsy, the assembled collection plate was mated to the quad microraft array by inverting the plate and positioning it into designated slots on top of the microraft array cassette. Holes drilled into the collection plate’s base plate enabled oxygen exchange between gas in the tissue-culture incubator and the culture medium overlying the colonies on the microraft array. Additionally, the snap-on fastener had a rounded cavity engraved into the top to permit the release of entrapped bubbles when mating the collection plate with the microraft array. Entrapped bubbles were minimized by mating or lowering the collection plate onto the microraft array cassette an angle ~ 45°, forcing any bubbles through the fastener’s cavity and into the outer medium chamber of the cassette. Once mated, the quad microraft array and microwell array were separated by 2 mm to minimize the microraft distance travel to the collection plate yet prevent microneedle contact with the collection plate during colony biopsy. The holder containing the EL light sheet and magnet was positioned on the collection plate and into designated slots on the microraft array cassette.

After the collection assembly was mated to the quad microraft array, colony biopsy was initiated. Colony fragment collection locations were matched back to their biopsy locations on the quad array by registering stitched images of the quad array before and after microraft release (Fig. S3) to track the microraft movement (Fig. 4A). Biopsied fragments were collected with 95 ± 3% efficiency (N=3 experimental replicates, 245 total microrafts) and with an average total XY displacement of 192 ± 73 μm (N=3 experimental replicates, 168 total microrafts). Thus, most microrafts moved into the microwell directly above and adjacent to their release site. These displacements correspond to a travel angle of −5 ± 53 degrees from normal (N=3 experimental replicates, 168 total microrafts) with the displacement tending to be outward from the center of the quad array, suggesting that the microrafts followed the radial magnetic field lines of the overlying disk magnet. The developed microwell collection plate addresses challenges of current microarray platforms by eliminating the need to transfer individual colony fragments into discrete culture vessels enabling automatic collection of all fragments simultaneously while preserving the ability to match the fragments back to their mother colony.

Fig. 4.

Automated micro-colony biopsy and cell collection into a microwell collection plate for immunofluorescence assays. (A) Microraft X-Y travel trajectory from biopsy location on quad microraft array to collected location in the microwell collection plate. Total length of arrow from tip-to-tail is representative of the total distance traveled by one microraft. Scale bar is 1 cm. (B) Microraft X-Y travel trajectory from initial collected position in the microwell collection plate to the post assay location in the microwell array top-piece. Scale bar is 1 cm. (C) Representative quad colony site pre-biopsy and the vacant microraft location post-biopsy on the quad microraft array (left two panels). Image of the biopsied microraft immediately after collection into and after the assay into the microwell array (right two panels). In the post-biopsy and post-collection images, EL light sheet illumination (due to light obstruction by the magnet) was needed for visualization, under which the cell and microwell walls were not visible. Scale bar is 400 μm.

3.5. Immunofluorescence Assay of Biopsied Cells and Resampling of Mother Colonies

The RBL-2H3 leukemic cell line is a commonly used model cell line in the study of inflammation and allergy since they have many mature mast cell behaviors.²⁵ For example, these cells possess an immunoglobulin E (IgE) receptor FcεRI, which when cross-linked by an antigen activates a plethora of downstream intracellular signaling pathways, including the STAT3 signaling cascade.²⁶ Inhibition of STAT3 signaling in mast cells has been shown to impair FcεRI-mediated upstream and downstream signaling, as well as reduce secretion of inflammatory mediators. The phosphorylation of STAT3 on Serine727 and Tyrosine705 residues following FcεRI cross-linking is known to impact secretion of granule contents by modulating granule exocytosis, as well as to promote mast cell proliferation (and in the case of tumor cells, uncontrolled cell growth).^26,27 To demonstrate the feasibility of collecting cells into the microwell collection plate and then performing assays on the cells, clonal RBL colonies were biopsied from the quad microraft array, cultured to permit cell growth off of the microrafts, and the biopsied cells assayed for the presence of pSTAT3 and STAT3.

After fragment collection, and with the collection plate/magnet and quad microraft array still mated, the cells were cultured for 2 days within the collection plate microwells. Cells were cultured post-collection to enable the cells to expand off the microrafts and onto the PDMS-on-glass microwell bottom, since the microrafts exhibit high fluorescence at some wavelengths. As a result, cells became adhered to the microwell bottom prior to detachment of the collection plate. After cell expansion, the collection plate was detached from the main assembly so that the plate resided above the magnet (Fig. 3C). The microwell collection array containing the expanded colony fragments was removed and the cells fixed in 60% methanol. Since the immunoassay protocol involved multiple washing steps, it was important to quantify the extent of microraft movement from the start to the end of the immunoassay. The majority of the microrafts (82%) remained in their original microwell traveling 24 ± 22 μm within the 183 μm-wide microwell relative to their initial location at the start of the immunoassay (Fig. 4B and C). The remaining 18% of collected microrafts traveled 321 ± 233 μm (corresponding to ~1–3 microwells) during the assay, likely due to minor fluid flow in the mated system. Despite the microraft movement, 100% of the collected colony fragments were re-located and matched back to their initially collected location, and in turn, their biopsy location (mother colony) on the microraft array. Because all biopsied microrafts were ≥1070 μm distant from adjacent biopsied microrafts, the microraft displacements were relatively small and did not prevent successful tracking and mother colony matching to the post-assayed biopsy cells.

To enhance imaging of the assayed biopsy fragments, an automated image analysis script was developed. Images of the microwell collection array were first scanned at 4× (NA 0.13) magnification to identify colony location. Microrafts were located 100 ± 0% of the time with 73 ± 13% specificity relative to manual identification at 4× magnification (N=3 experimental replicates, 126 total objects detected). The low specificity was due to misidentification of debris having a size, shape, and pixel intensity similar to microrafts. Debris incorrectly identified as a microraft was manually labelled and excluded. With microraft locations known, the number and location of FOVs at 10× (NA 0.3) magnification was calculated to maximize the number of microrafts per 10× FOV (Fig. 5A). Compared to whole-array imaging at 10× magnification, the combined methods of 4× position identification plus 10× FOV minimization reduced imaging time from 77 min (972 images per channel) to 8 min (90 images per channel). The average colony fluorescence for pSTAT3 and STAT3 was measured from each of the assayed biopsy fragments and the ratio of pSTAT3-to-STAT3 (pSTAT3/STAT3) calculated (Fig. 5B and 5C). The immunofluorescence assay was only possible on cells that had grown off the microraft due to the high autofluorescence of the microraft. 82% of biopsies expanded off the microraft within 2 days and were assayed. Biopsy fragments with the 3 lowest (L1, L2, and L3) and 3 highest (H1, H2, and H3) pSTAT3/STAT3 were selected as targets (Table 1). The target daughter fragments were matched back to their still expanding mother colonies on the microraft array, and the mother colonies resampled and expanded further in a well plate. Of the resampled mother colony fragments, 100 ± 0% (N=6 fragments) grew (Fig. 5D). These data demonstrate the feasibility of obtaining biopsies from micro-colonies and assaying the biopsy fragments with a cell-destructive assay followed by successful matching of the fragments back to their viable mother colony.

Fig. 5.

Immunofluorescence assay of biopsied clones and identification of clones to target mother colony resampling. (A) Image locations across the microwell collection array, where the orange circles depict the microraft and colony fragment locations identified at 4× magnification, the open black squares marked the FOVs used to image the microwell array at 10× magnification, and the blue line is the order of 10× FOV image collection identified by the nearest-neighbor imaging algorithm (scale bar is 1 cm). The right-side panels show a 10× FOV of microwells as marked on the microwell collection array, where the top FOV has 2 microrafts each in a different microwell and the bottom FOV shows a single microraft in a microwell (scale bar is 400 μm). (B) The total average immunofluorescence of pSTAT3, STAT3, and pSTAT3/STAT3 for the 25 clonal fragments collected. Data points for fragments identified as having the 3 lowest (L1, L2, L3) and 3 highest (H1, H2, H3) pSTAT3/STAT3 are plotted as solid blue squares and pink triangles respectively. The median value among the measurements is plotted as a horizontal line. (C) Representative fluorescence images of immunostained fragments in the microwell collection array for Hoechst 33342 (DNA), pSTAT3, STAT3, and a composite image of pSTAT3 and STAT3. Of the 6 target fragments identified, representative images shown are for clonal fragment L3. Scale bar is 100 μm. (D) Representative bright field images of the resampled mother colony for clone L3 immediately after collection (day 0) and the expanded colony (day 5) before and after rinsing the well with media which removes the microraft. Scale bar is 300 μm.

Table 1.

Summary of the average pSTAT3/STAT3 per cell in the biopsied fragments.

Sort Group	Biopsy Fragment	pSTAT3/STAT3
Lower Expressers	L1	0.15 ± 0.04
	L2	0.19 ± 0.03
	L3	0.20 ± 0.04
Higher Expressers	H1	0.22 ± 0.04
	H2	0.25 ± 0.09
	H3	0.29 ± 0.11

3.6. Characterization of Clonal Culture Phenotype

Single cells within a normal or leukemic cell population possess heterogeneous attributes and can have variable responses to receptor binding events such as FcεRI cross-linking.²⁸ In general, single cell differences may result from the inherent properties of growing cell populations, like cell-cell contacts, that create a breadth of microenvironmental differences to which cells adapt, combined with the non-genetic memory of phenotypic states and protein levels. With respect to mast cell degranulation specifically, variability may be due to the expression of different FcεRI densities leading to altered degrees in FcεRI aggregation by IgE/antigen complexes and clustering patterns, therefore inducing or inhibiting granule secretion.²⁵

The goal of this feasibility sort was to identify cells within a population that possessed varying levels of STAT3 phosphorylation (on Tyrosine705) to STAT3 (not phosphorylated) protein and whether this variation might predict cell behavior in response to cross-linking of FcεRI. As discussed in previous sections, clonal RBL cell colonies were cultured on quad microraft arrays for 5 days (19 ± 4 h cell area doubling time, 38 colonies over 96 h, Fig. S4), biopsied, cultured for 2 days, and their average nuclear pSTAT3/STAT3 measured in each expanded fragment at day 2 after biopsy. Clonal biopsy fragments identified as having the lowest average levels of pSTAT3/STAT3 were clones L1, L2, and L3, and those with the highest levels were clones H1, H2, and H3. Single-cell pSTAT3/STAT3 in the 6 target clonal fragments was also measured by manually segmenting cell nuclei based on Hoechst 33342 staining (Table 1). Significant differences in pSTAT3/STAT3 between individual clones was observed, as well as between clonal fragments pooled as having either low or high pSTAT3/STAT3 (p<0.0001) (Fig. 6A).

Fig. 6.

Nuclear pSTAT3/STAT3 and β-hex release among RBL-2H3 clones. (A) pSTAT3/STAT3 measured per cell for the initial fragments’ biopsied after 7 days in culture (5 days on quad microraft array and 2 days’ post-biopsy in microwell collection array): i.) for the 6 individual target clones where each data point represents a single cell’s pSTAT3/STAT3 value and the mean value is plotted as a horizontal line (7–53 cells per fragment). ii.) the total distribution of pSTAT3/STAT3 per cell in the colony fragments pooled as low (L1, L2, and L3) and high (H1, H2, and H3). (B) Distribution of pSTAT3/STAT3 measured per cell for the resampled and expanded mother colonies after 18 days in culture (18 days’ post-cell seeding on microraft arrays): i.) for the 6 individual resampled clones (2,369 – 7,072 cells per clone) and ii.) for the resampled clones pooled based on their initial sort criteria of low (L1, L2, and L3) and high (H1, H2, and H3) pSTAT3/STAT3. (C) Percent β-hex release measured after activation via IgE cross-linking of FcεRI: i) in unsorted bulk RBL-2H3 cells without treatment and with activation of FcεRI cross-linking (N=12 wells per condition, 100,000 cells per well), and ii) in the expanded mother clones with activation of FcεRI cross-linking after 24–28 days in culture (N=12 wells per colony type, 100,000 cells per well). For each violin plot, the solid horizontal line represents the median value, while dotted horizontal lines below and above the medium value represent the 25^th and 75^th percentile values, respectively. p values are represented as p<0.05 = *, p<0.01 = **, p<0.001 = ***.

The mother colonies matched back to the 6 target clones were biopsied again at day 9 of growth on the quad microrafts, the biopsy fragment expanded for 9 days in a well plate (day 18 of culture) and the expanded cells assayed again for pSTAT3/STAT3. The cells were additionally expanded for 6–10 more days (day 24–28 of culture) (16 ± 3 h cell number doubling time, N=5 experimental replicates per clone over 48 h, Fig. S5) and assayed once more for pSTAT3/STAT3. Across each of the expanded clones, the average nuclear pSTAT3/STAT3 per cell was re-measured via automated image segmentation (Fig. S6) and significant differences in pSTAT3/STAT3 per cell were observed at day 18 (Fig. S7) and day 24–28 (Fig. 6B) which were not present in the initial colony fragment biopsied at day 7. Variability in expression is likely due to the inherent properties of single cells within growing cell populations.²⁹ At both day 18 and 24–28, there were significant differences between the initially low (L1, L2, L3) and the high (H1, H2, H3) pSTAT3/STAT3 groups (p<0.0001), suggesting that the clonal cultures maintained their pSTAT3/STAT3 signature over-time.

To investigate whether the low and high pSTAT3/STAT3 cells possessed different immune signaling responses, β-hexosaminidase (β-hex) secretion by the resampled and expanded RBL mother colony fragments was measured following cross-linking of the FcεRI receptor. β-hex is an enzyme localized in the granules of mast cells and, thus, is commonly used as a measure of mast cell degranulation or activation.³⁰ As a control assay, β-hex release from bulk cells activated by antigen-receptor cross-linking was measured and found to be significantly different from that of non-activated bulk cells (10 ± 1 % versus 22 ± 2%, p=0.002, N = 6, 100,000 cells per replicate) (Fig. 6C). β-hex release from the activated expanded clones at 24–28 days was statistically significantly different between clone L1 and clones L3 (p<0.0001) and H1 (p=0.039), and between clone L2 and L3 (p=0.0007) (N = 12, 100,000 cells per replicate) (Fig 6C). However, upon FcεRI cross-linking, no statistically significant differences in β-hex release was measured between the pooled clones initially sorted into low (L1, L2, L3) or high (H1, H2, H3) pSTAT3/STAT3 groups. These results indicate that heterogeneity in FcεRI cross-linking in the expanded clones is not likely due to STAT3 phosphorylation alone. Other drivers for β-hex are identified in literature, such as sphingolipid and glycosaminoglycan metabolism,^31,32 which may be more dominant drivers of β-hex expression or activity than that due to the STAT3 pathway. However, it is also possible that rare pSTAT3 subpopulations exist that were not sampled by this experiment. Ultimately, the semi-automated microraft platform enabled the paired measurements of STAT3 intracellular proteins, as well as subsequent functional live cell phenotypes.

Conclusion

In this paper, the automated quad microraft array technology was enhanced to permit viable cell microcolony sorting based on a cell-lethal assay. The new pipeline enabled detection of clonal colonies from bright field microscopy images, biopsy and collection of colony fragments into known locations on a collection plate, quantification of intracellular protein levels via immunostaining, and retrieval of mother colonies based on a characterized daughter fragment. To achieve these milestones, a microwell collection plate was developed to mate with a one-to-one spatial correspondence directly above clonal colonies cultured on a quad microraft array. The collection plate permitted collection of fragments biopsied from the quad array into defined locations and the immunoassay of all collected fragments simultaneously. Uniquely, the platform eliminated the need to individually transfer cell samples into separate culture vessels for assays, thus increasing assay throughout and decreasing manual labor and use of consumables. For detecting rare colony phenotypes, the pool of clonal colonies could be increased by culturing cells at lower density or increasing the number of quad microraft sites. The pipeline may permit colony screening and sorting by other cell-destructive assays, such as fluorescence in situ hybridization to study spatiotemporal patterns in gene expression within the cells. The collection plate could also be used for identifying cell colonies based on other cell-modifying assays, such as differentiation. Overall, the established pipeline demonstrates the potential to characterize and sort viable cell colonies based on cell-lethal assays.