Thin Layer Chromatography Goes Ultrasmall to Assay Sphingosine Kinase Activation in Single Primary Leukemic Cells

Abstract

Cell-to-cell heterogeneity in lipid signaling underlies variations in response and recurrence for many cancers including leukemias. A highly parallel, miniaturized thin-layer chromatographic platform capable of assaying single cells was developed. Ultrasmall volumes (50 pL) of standard fluorescent lipids were separated with excellent repeatability, reproducibility, and limits of detection. Sphingosine-cyanine 5 (Sph-Cy5) was loaded into cells and the single- cell contents separated to identify Sph-Cy5, and two metabolites, Sph-1-phosphate-Cy5 (S1P-Cy5) and hexadecanoic acid Cy5 (HA-Cy5). In leukemic cells, the CD34+ blast cells demonstrated significantly greater conversion of Sph-Cy5 to its phosphorylated form compared to the CD34− cells. After treatment with a sphingosine kinase (SphK) inhibitor, the formation of S1P-Cy5 remained significantly greater for the inhibited CD34+ cells relative to that of the inhibited CD45− cells. Over 1200 single cells were rapidly assayed using 8 chips within 4 h. Sphingosine kinase activity in the CD34+ blast cells of 3 patients with acute myeloid leukemia was assayed with and without inhibitors. The patient cells displayed inter-tumor and intra-tumor heterogeneity, and subsets of cells with distinct enzymatic activities and products, highlighting the diversity of the cells within a clinical sample and patient.

Graphical Abstract

Ultrasmall-scale thin-layer chromatography (pTLC) was utilized for assay of a lipid-modifying enzyme in single cells. Cells were loaded with fluorescent enzyme substrates, deposited onto the pTLC microband, separated by application of organic solvent and the fluorescence analytes quantified.

Introduction

Lipids play crucial roles in cell function including participation in signal transduction cascades.¹ Disrupted lipid signaling has also been linked to many diseases including cancer, neurodegenerative disorders, and viral infections.^{2, 3} Sphingolipids (SLs) which possess an eighteen carbon amino-alcohol backbone are exemplars with some members acting as key nodal points in chemical cascades driving cell behavior. Sphingosine (Sph) is one such lipid and can be phosphorylated by sphingosine kinase (SphK) to form sphingosine-1-phosphate (S1P), which serves as a molecular messenger to promote cell migration, growth, and survival.^{4, 5} Alternatively, metabolism of sphingosine to other products e.g. ceramide promotes cell death. SphK activity is often altered in diseases such as cancers to drive proliferation by increasing S1P and subsequent downstream pathways.^{5, 6} Alterations in SL metabolism are prominent in leukemias where formation of S1P drives tissue invasion and cell proliferation, and mutations can increase SphK activity promoting drug resistance.^7-9 For example, studies in multi-drug resistant acute myeloid leukemia (AML) cell lines have shown that SphK inhibitors decrease S1P production and restore sensitivity to chemotherapeutic compounds to enable apoptotic cell death.^{8, 9} Similar outcomes have been shown in primary leukemic cells.^{7, 10} For this reason, extensive studies have been conducted to investigate SphK activity and its inhibition as a therapeutic strategy for leukemias.^10-12

Cytogenetic studies at the single-cell level have revealed extensive cell-to-cell genetic heterogeneity among the cells of patients with AML.^{13, 14} Such findings have led to the current thinking that cellular heterogeneity underlies the variation in response to therapy and recurrence observed in AML as well as other cancers.^{15, 16} However, cytogenetics even at the single-cell level have proven limited in predicting therapeutic responsiveness and patient outcome.¹⁷ The concept that a specific mutation in a leukemic cell acts as a faithful predictor of response to treatment is now recognized as too simplistic since signaling networks can be modulated by crosstalk, epigenetic changes, and other mechanisms in the presence or absence of oncogenic mutations in those pathways.^{17, 18} It is therefore expected that quantifying the heterogeneity of metabolic processes and apoptotic/proliferative signaling on a cell-by-cell basis will enhance prognosis and rational selection of targeted therapeutics in AML.^19-22

The appreciation of the importance of cellular biochemical heterogeneity for designing targeted treatment regimens for an individual patient along with the practical issue of performing tests on clinical specimens are driving the development of assays capable of such analyses. To date, measurement of critical lipid second messengers such as Sph and its metabolites has been challenging due to the small volume of human cells (~1 pL) coupled with the low concentration of these lipids amid high concentrations of structural lipids.^{10, 23, 24} Innovations in mass spectrometry (MS), Raman spectroscopy, and fluorescence-based approaches have permitted a some number of lipids to be quantified in single cells especially those lipids at high concentration.^{1, 25} MS is a powerful and high-throughput method for lipidomics, but it remains challenged for measurement of sphingosine metabolites in the individual cells of leukemia patients by its limited sensitivity and complex, high-cost instrumentation.^26-29 Raman scattering is capable of imaging cellular lipids but is too low in sensitivity and specificity to reveal Sph metabolites in single cells.^30-32 Immunofluorescence assays are plagued by the poor immunogenicity of Sph metabolites as well as nonspecific binding and high cross-reactivity of antibodies to lipid antigens.^{33, 34} Microscopy using fluorescently labeled Sph can evaluate its localization within a cell, but cannot distinguish Sph from its metabolites, limiting usefulness.^{1, 35, 36} Microelectrophoresis-based assays combined with fluorescently labeled Sph have proven useful to measure SphK activity when paired with fluorescent Sph analogues (or reporters), but suffer from low throughput, the need for significant technical expertise and advanced instrumentation, and so have not become broadly available.^{10, 19, 37}

We have previously described a chemical cytometry technique termed picoliter-scale thin-layer chromatography (pTLC) for TLC-based analysis of picoliter-to-nanoliter volume samples, including single mammalian cells.³⁸ The microfluidic device is comprised of an array of monolithic silica microbands on a glass substrate. Samples of fluorescent lipids or single cells loaded with fluorescent lipid-based enzyme substrates e.g., reporters are deposited at the beginning of each microband. Organic solvent driven by surface-tension driven flow sweeps across the deposited cells extracting the cellular lipids and separating the fluorescent lipid metabolites based on their partitioning between the organic solvent and silica surface as occurs in macro-TLC. Notably the device is operated without active fluidic controls or valves, and the separation is rapidly terminated by solvent evaporation leaving the fluorescent lipids fixed in place on the microband for detection. Analytes are identified by their retention times while the amount or moles is quantified from the fluorescence peak area when compared to that of a known standard. The ratio of substrate to products on these chromatograms acts as a measure of enzyme activation (SphK in this case). While this miniaturized format created a new strategy for single-cell lipid enzyme assays and demonstrated proof-of-concept, significant challenges remained including a very low assay throughput, an unknown inter-chip and intra-chip reproducibility, and limited sensitivity. Multiplexing with other assays such as immunocytochemistry was not possible due to the assay format. These aspects ultimately prevented examination of cells in clinical samples in meaningful numbers particularly when examining cellular responses to inhibitors.

In the current work, we developed and characterized an automated pTLC device capable of assaying SphK activation in statistically relevant numbers (100’s) of single cells. Two different SphK substrates (aka reporters) Sph-fluorescein (Sph-F) and Sph-Cyanine-5 (Sph-Cy5), and their phosphorylated products (S1P-F and S1P-Cy5) were used for separations, optimization, and enzyme activation measurements. The reproducibility and reliability of separations of the fluorescent lipids, analyte detection sensitivity and assay throughput were evaluated across different pTLC microbands and batch of devices. The percentage of the fluorescent S1P peak area relative to the total fluorescent analyte peak area (Sph + S1P + other metabolites) was used as a measure of the activation of SphK in single cells of tissue-cultured cell lines as well as the peripheral blood of patients with AML. The activation of SphK in hundreds of single cells as well as the heterogeneity in the single-cell responsiveness to four different inhibitors of SphK was measured by pTLC. Immunofluorescence labeling of the hematopoietic stem-cell marker CD34 combined with pTLC separation was used to correlate SphK behavior with a subpopulation of stem-like cancer cells. The heterogeneity in single-cell SphK activity in non-inhibited and inhibited cells in three patients with and without therapeutic blockade was also examined to demonstrate the power of pTLC.

Materials and Methods

Materials and Reagents.

Sphingosine fluorescein (#S-100F), sphingosine-1-phosphate fluorescein (#S-200F), sphingosine Cy5 (#S-100C), sphingosine-1-phosphate Cy5 (custom synthesized) were purchased from Echelon Biosciences (Salt Lake City, UT). RPMI 1640 media with L-glutamine (#SH30027.FS) was obtained from Cytiva (Marlborough, MA). Heat-inactivated fetal bovine serum (FBS, #10438026), Hanks' balanced salt solution (HBSS, with calcium and magnesium salts, # 14025092) were from Gibco (Grand Island, NY). K145 (#11691), FTY720 (#10006292), SKI2 (#10009222), PF-543 (#17034) were procured from Cayman Chemical Company (Ann Arbor, MI). Extracellular buffer (ECB) was defined as: 135 mM NaCl, 5 mM KCl, 1 mM MgCl2, 1 mM CaCl2, 10 mM HEPES, 10 mM glucose and pH 7.4.

Fabrication of pTLC Microchip.

The pTLC microchips were fabricated as previously published.³⁸ Briefly, soda lime glass slides (75 mm × 25 mm) possessing an array of 40 open microfluidic channels (width 80 μm, depth ~13 μm, length 10 mm, center-to-center inter-channel gap 250 μm) were fabricated by photolithography and hydrofluoric acid (HF) wet etching. Silica sol-gel precursor (5.6 mL tetramethyl orthosilicate, 1.4 g polyethylene glycol, 0.9 g urea, and 10 mL of 10 mM acetic acid solution) was used to fill the channels, followed by polymerization at 40°C for 24 h, baking at 121°C for 90 min, and calcination at 330°C for 24 h. Careful consideration was given in determining appropriate feature dimensions of the microbands. The distance between adjacent microbands was 100 μm for ease of fabrication while still maximizing the number of microbands in a given area. To minimize lateral sample diffusion, a channel width of 80 μm was employed. Microband length was set at 10 mm to provide maximum flexibility and performance in lipid separations.

Cell Preparation.

Kasumi-3 cells (human acute myeloid leukemia; #CRL-2725) and K-562 cells (human chronic myeloid leukemia; #CCL-243) were obtained from the American Type Culture Collection (ATCC). Kasumi-3 and K-562 cells were grown in RPMI-1640 media supplemented with 10% FBS. Cells were maintained in a humidified atmosphere of 37°C and 5% CO₂ and subcultured into fresh media every 3 days. Cells were not used beyond passage #10 from the original ATCC stock. Just prior to use, cells (1 million/ml) were transferred to a 1.5-mL microtube and centrifuged (300×g for 4 min) and the culture medium aspirated. For assay, cells were resuspended in 100 μL cell culture media containing 100 nM or 10 μM Sph-Cy5 in the presence of DMSO control or a SphK inhibitor and incubated for 30 min at 37°C.

Cryopreserved de-identified AML patient samples were obtained from Fred Hutch/UW Hematopoietic Diseases Repository. The samples were stored in liquid nitrogen. Primary cells were used immediately upon thawing. Cells were resuspended in 100 μL cell culture media containing 100 nM Sph-Cy5 in the presence of a DMSO control or a SphK inhibitor and incubated for 30 min at 37°C.

After incubation, cells were centrifuged, and supernatant was carefully aspirated. The cell pellet was resuspended in 200 μL cell culture media with or without an SphK inhibitor (at the concentration indicated in the text), transferred to a microcentrifuge vial, and centrifuged (300×g for 4 min) to remove the supernatant. The cells were rinsed with 200 μL ECB for an additional two times before being fixed with 4% glyoxal in ECB for 5 min at 25°C. For Kasumi-3 cells and patient samples, the cells were stained with a PE-conjugated mouse anti-human CD34 antibody (BDB560941; Fisher Scientific) and Hoechst 33342 (1 μg/ml). Fixed cells were stored in the dark at 4°C. All samples were analyzed within 2 h of fixation.

The following molecules modulating the SphK pathway were employed. K145 is a competitive inhibitor of sphingosine kinase 2 (SphK2) that induces apoptosis.³⁹ SKI2 is as an inhibitor of sphingosine kinases 1 and 2 and has been extensively used to study the inhibition of SL pathway for disease treatment.⁴⁰ PF543 is potent competitive inhibitor of SphK1.⁴¹ FTY720 (Fingolimod) is a substrate for SphK and a competitive inhibitor of SphK at high concentrations.^{42, 43} Clinically, FTY720 has a different mechanism of action in that the phosphorylated form has a high affinity for the S1P receptor resulting in receptor down regulation. FTY720 has been approved by FDA for clinical treatment of multiple sclerosis and is in clinical trials for treating various cancers.^{42, 43} The reported IC50 levels for these inhibitors are: K145 (IC₅₀: 4.3-10μM), SKI2 (IC₅₀: 0.5-10μM), FTY720 (IC₅₀: 10μM) and PF543 (IC₅₀: 2nM; Fig 8).^42-45

Fluorescence Imaging of Cells.

Sph-F and S1P-F were imaged at peak excitation 500 nm/emission 520 nm, Sph-Cy5 was imaged at peak excitation 650 nm/emission 670 nm, PE-CD34 was imaged at peak excitation 565 nm/emission 578 nm, and Hoechst 33342 was imaged at peak 350 nm/emission 450 nm. The imaging pixel size was 3.5 μm x 3.5 μm, and the integration time was 2 μs/pixel - 20 μs/pixel. The laser power was optimized at 1-5% for all fluorophores. Measurements were performed using an Olympus FluoView FV3000 confocal laser scanning microscope equipped with the FV3000 software. An 4x/0.16 objective was used to yield an imaging depth of 15 μm.

pTLC Development, Imaging and Data Analysis.

Control solutions and single cells were deposited on pTLC microbands using a commercial piezoelectric print head (MicroFab Inc., Plano, TX) and a microfluidic controller (#OB1-C-4, Elveflow, Paris, France). For separations, the pTLC microchip was placed face down in a custom-constructed development chamber. Extraction of lipids from cells and lipid separation was initiated by addition of solvent to an absorbent pad in contact with the microband inlet closest to the deposited cells.³⁸ After separation, the chip was air-dried and the fluorescence was measured using an Olympus FluoView FV3000 confocal laser scanning microscope.³⁸ The 4× imaging objective possessed a numerical aperture of 0.16. The chromatograph (i.e., fluorescence intensity profile along each microband) was generated using ImageJ's plot profile function (Version 1.53t). Percent phosphorylation was defined as [S1P peak area/(S1P peak area + Sph peak area + other peak areas)]. The resolution ($\mathrm{R}$) is defined as the distance between two peak centers ($\mathrm{d}$) divided by the width ($\mathrm{W}$) of the peaks $R=\frac{2d}{W1+W2}$. The theoretical plate number was calculated as N = 16(x/w)²/L where w is the width of the peak base, x is the distance from the origin to the center of the peak, and L is the distance from the origin to solvent front.⁴⁶ Data was analyzed using GraphPad Prism.

Identification of Cellular Metabolites using Ultra Performance Liquid Chromatography/Mass Spectrometry (UPLC/MS).

To determine the identity of the additional peaks on the separation traces, 20 million K562 cells were incubated with Sph-Cy5 reporter for 30 min (10 μM Sph-Cy5 added to 1 million cells per vial, 20 vials total). The cells were then extensively washed, transferred to a single tube, and lysed in 200 μL acetonitrile. Analysis of Sph-Cy5 and its metabolites within the cell lysate was performed with an UPLC-MS system consisting of a Water’s Acquity I-class UPLC connected to a Waters Synapt XS TOF with Masslynx V4.2 SCN1015 software (Waters, Milford MA) and ACQUITY Premier Fluorescence (FLR) detectors. Separations were performed using a 2.1 mm × 100 mm ACQUITY UPLC BEH C8 column with 1.7 μm particles (Waters, part #186002878). The mobile phases were 0.1% formic acid in water with 10 mM ammonium acetate (solvent A) and 0.1% formic acid in acetonitrile (solvent B). Following sample (10 μL) injection, the mobile phase mixture was 70% solvent A, 30% solvent B (0.3 mL/min) for 1 min followed by a linear gradient to 60% (solvent B) over 17 min. The buffer composition was then held at 60% (solvent B) for 2 min. Fluorescence was detected with excitation set at 649 ± 10 nm and fluorescence emission recorded at 670 ± 10 nm. Prior to injection of the next sample, the column was re-equilibrated in the initial conditions for 4 min. Two major fluorescent peaks were observed on the time vs fluorescence trace (Fig. S5A). The molecular weight of the species in each peak was measured by mass spectrometry. Mass spectrometry was performed using positive mode electrospray ionization (ESI) MS with the following settings: capillary voltage 3.0 kV, cone voltage 80 V, source temperature 120°C, desolvation temperature 350°C, and a nebulizing gas pressure 6.5 bar.

Results

pTLC Enables Reproducible Separation of a Signaling Lipid from Its Phosphorylated Form.

Repeatability and reproducibility are essential for validating an analytical technology. Band-to-band and chip-to-chip variation in the separation of a mixture of 2 model lipids (Sph-F and S1P-F, 50 amol each in a total volume of 50 pL) was measured. An automated piezoelectric printing system was used to print droplets of standards near the origin of each microband on a 40 microband array (Supplemental Data Fig. S1 & S2). The control lipids were separated by application of solvent, and the separation was stopped at 2 min. The chips were imaged by fluorescence microscopy (Fig. 1A-1C). The resolution across 5 chips (10 microbands each) was 1.7 ± 0.3 (Fig. 1D). No significant differences were observed (p=0.8125, one-way ANOVA test) among the 5 tested chips, and the resolution of all channels was greater than 1, indicating that the two peaks were well-separated. The measured percent phosphorylated lipid (average 50 ± 3 %) displayed repeatable measurements and device performance did not differ significantly amongst the batches (p=0.52, one-way ANOVA; Fig. 1E). The standard deviation of the percent phosphorylation was below that recommended for preclinical and clinical assays (standard deviation σ <0.15) providing reasonable power for bias detection.^{47, 48} The peak area and width (normalized to that of Chip 1) across the 5 devices was 1.03 ± 0.09 (area) and 0.98 ± 0.14 (width) for S1P-F, and 1.04 ± 0.08 (area) and 0.95 ± 0.09 (width) for Sph-F. The peak areas (Fig. 1F) and peak widths (Fig. 1G) were not significantly different across the chips for either analyte. The theoretical plate number was 9.7 ± 1.3 mm⁻¹ (S1P-F) and 87.4 ± 17.5 mm⁻¹ (Sph-F), with no significant difference across the microdevice batches. After usage, the microdevices were incubated in sulfuric acid (1 h) and deionized water (12 h). The device was then air-dried without visible damage occurring to the microbands. In addition, the % CV across these 5 chips (10 microbands each) was 13% for the resolution, 9% (S1P-F) and 8% (Sph-F) for peak area, 6% for percent phosphorylation, and 12% (S1P-F) and 10% (Sph-F) for peak width at half height. These results suggest the pTLC chips generate reproducible data when a standard sample is assayed. Furthermore, the microdevices could be reused >30 times over several months.

Figure 1. Characterization of pTLC separation of the standards Sph-F and S1P-F.

(A) Photograph of a microfabricated pTLC chip. (B) Comparison of the microband under dry and wet conditions. (C) Fluorescence image of 10 microbands after 8 min of separation. Each microband was printed with 50 pL of an equimolar mixture of Sph-F and S1P-F. The standards were separated with 1-propanol:water (v:v = 4:1). (D) Resolution across 5 chips after 2 min of separation. (E) The mean and standard deviation for the percent S1P-F was plotted for 5 different pTLC chips. The percent S1P-F was defined as ([peak area of S1P-F]/[peak area of Sph-F + peak area of S1P-F]). Each data point represents a separation on a microband of the indicated chip. (F) Normalized peak area. (G) Peak width at half height. Panels D-G display the data for 5 chips (N= 10 channels/chip). The error bars represent a single standard deviation.

The limit of detection (LOD) for Sph-Cy5 was measured after spotting 0.04 amol to 4 amol of a standard onto a pTLC microband. The Cy5 labelled version of Sph was used due to its brighter fluorescence and greater resistance to photodegradation relative to that of fluorescein. After application of solvent, the signal-to-noise ratio (SNR) of the fluorescent peak was measured on each chromatogram (Supplemental Data, Fig. S3). The LOD for Sph-Cy5 was of 0.04 amol.⁴⁹ The LOD of pTLC exceeds that of commercially available high-performance thin-layer chromatography (~10⁶ amol) as well as mHPLC (~10 amol) for lipid analytes.^{50, 51} While advanced capillary electrophoresis offers an improved fluorescent lipid LOD (~10⁻³ amol), this methods remains extremely challenging to pair with single cells.^{10, 51-53} Additionally, an LOD near 0.01 amol suggests that pTLC will be suitable for measurement of the metabolism of fluorescent reporter lipids within key cellular signaling pathways.

Measurement of Sphingosine Metabolism in Single Cells by pTLC.

K562 cells, a chronic myelogenous leukemia cell line, were incubated with the SphK substrate, Sph-Cy5, to load the reporter into the cells (Fig. 2A). The cells were then fixed to halt the intracellular reactions with the reporter. The automated droplet printer was used to place cells at the origin of each microband (Fig. S4). Separation was performed by application of solvent followed by fluorescence imaging and quantification of the area under each peak (Fig. 2B). The average percent phosphorylation within the cell population was 48% ± 11% (n = 600) with a wide range of behaviors observed across the cells. The range of percent phosphorylation among individual cells was 13% - 69% (Fig. 2C-J). No significant difference was observed between the four experiments performed (p = 0.24, one-way ANOVA), suggesting that the experiments conducted on cells using different pTLC chips at different times were repeatable for single-cell assays. The majority of cells possessed 1-2 peaks with retention times corresponding to that of Sph-Cy5 and/or S1P-Cy5 (Fig. 2C,D). However, 10.7% of cells possessed up to 2 additional peaks (Fig. 2E,F). To identify the additional peaks, a bulk lysate of cells loaded with Sph-Cy5 was purified, separated by UPLC, and the molecular weight of fluorescent peaks measured by mass spectrometry. Two major peaks were present on the fluorescence trace (Supplemental Data Fig. S5). The initial peak was comprised of two major species of m/z, 859.6 and 732.6, corresponding to the expected m/z for S1P-Cy5 and its downstream metabolite hexadecanoic acid-Cy5 (HA-Cy5), respectively. Hexadecanoic acid has been observed previously in cells metabolizing fluorescent sphingosine.^{4, 7, 54} HA is formed by the action of fatty aldehyde dehydrogenase (FALDH) on hexadecinal, a metabolite of S1P in cells.⁵⁵ The second major fluorescence UPLC peak possessed a single species of m/z 779.6 corresponding to the expected m/z for Sph-Cy5 (Fig. S5B). When the fluorescent UPLC peaks were collected and separated by pTLC, the retention time of the m/z 732.6 peak (HA-Cy5) matched that of the unidentified peak migrating immediately after S1P-Cy5 (Fig. S5). Rare cells (2.4%) possessed an additional unidentified fluorescent peak (marked as “U”). However, only 3 distinct m/z species were observed by HPLC-MS of bulk lysates and consequently the identification of the U peak remains unknown (Fig. 2E,F). It was likely that the concentration of U was present at too low a concentration in the bulk lysate since so few cells formed the metabolite. The products identified by pTLC match that previously identified in single cells or bulk cell lysates by capillary electrophoresis and mass spectrometry.^{4, 54, 56-61} After pTLC performance, unidentified fluorescent residua were observed at the original cell deposition site (Fig. 2B) and accounted for 2.5 ± 0.9% of the observed fluorescence on the microband.

Figure 2. Measurement of Sph-Cy5 metabolism in single K562 cells.

(A) Shown is a schematic of the assay steps. (B) Image of the contents of a single cell separation on a pTLC microband at varying times after the addition of solvent (1-propanol:water, v:v = 4:1). (C-F) Chromatographs of the analytes from single cells. Shown is the relative fluorescence intensity plotted against the distance from the microband origin at the termination of the separation. A single cell possessing a mixture of S1P-Cy5 and Sph-Cy5 (C), Sph-Cy5 with minimal S1P-Cy5 (D), or S1P-Cy5, Sph-Cy5, HA-Cy5 and another metabolite, marked as “U” (E,F). (G-J) “SuperPlots” showing the percentage S1P-Cy5 from single cells with and without inhibition by one of four SphK pathway modulators (E) 10 μM K145, (F) 10 μM SKI2, (G) 1 μM PF543, and (H)10 μM FTY720.⁶³ The “SuperPlots” identify different replicates by color. In these experiments, each color represents the data points from the 3 biologic replicates (N=50 cells each replicate). The triangle represents the average percent phosphorylation in each corresponding experiment. The p values represent a paired two-tailed t test of the average phosphorylation from each repeat in DMSO and inhibitor-treated groups.

Separation performance metrics for the lipids derived from the single cells were significantly improved relative to that obtained for the standards. For example, the theoretical plate number for a Sph-Cy5 standard (418 ± 50 mm⁻¹, n=50) was significantly less than that for Sph-Cy5 from cells (475 ± 52 mm⁻¹, n= 50 cells, p < 0.001). Similarly, the theoretical plate number for a S1P-Cy5 standard (38 ± 3 mm⁻¹, n=50) was significantly less than that for S1P-Cy5 from cells (92 ± 15 mm⁻¹, n= 50 cells, p < 0.001). The enhancement in separation efficiency was most likely due to a decrease in injection dispersion when using the small cells (1 pL, 15 μm in diameter) compared to the larger printed droplet for control samples (50 pL, 100 μm printed diameter).⁶² The average migration speed during a 4 min separation for Sph-Cy5 was 647 ± 59 μm/min, faster than that of S1P-Cy5, 393 ± 34 μm/min. The Sph-Cy5/S1P-Cy5 separation reached a resolution of 1 in 34 ± 9 seconds (Figure 2B). The resolution for Sph-Cy5 and S1P-Cy5 was 1.9 ± 0.1.

Inhibition of Sphingosine Kinase Measured in Single Cells by pTLC.

Pharmacologic inhibition of SphK at the single-cell level was readily detected by pTLC. Single cells loaded with Sph-Cy5 were incubated with a panel of drugs utilized in pre-clinical or clinical investigations to modulate SphK signaling.^39-43 K145, SKI2, PF543 are all direct inhibitors of SphK I and/or SphK II while FTY720 is a competitive inhibitor of the S1P receptor that modulates S1P production by an as yet unknown mechanism.^42-44 These pharmacologic agents were used at their reported IC50 value or K_M for SphK.^{39, 41-43} All 4 pharmacologic agents significantly decreased the percentage of S1P-Cy5 formed relative to that of the control samples (Fig. 2G-J). A significant difference in the percentage of S1P-Cy5 was observed among the different inhibitors (p < 0.0001, one-way ANOVA), indicating that the efficacy of the inhibitors varied under these conditions. PF543 was the most potent inhibitor, corroborating prior reports.⁴¹ There was no significant difference between K145 and FTY720 groups (p=0.44, two-tailed t-test); however, both were more effective than SKI2 at reducing formation of S1P-Cy5. Heterogeneity among individual cells in response to the inhibitors was readily observed (Fig. 2). The percent of inhibitor-treated cells that possessed peaks in addition to that of Sph-Cy5 and/or S1P-Cy5 varied between the inhibitors (Table 1). Significantly more cells (11.3 ± 1.2 %) in the DMSO control group possessed additional peaks compared to the cells exposed to PF543 (p = 0.04). This was likely due to blockade of SphK, reduced production of S1P-Cy5 and consequently a decrease in downstream metabolites such as HA-Cy5. No significant difference was observed between DMSO and other inhibitor-treated groups. Importantly, assay of the pharmacologic agents at the single-cell level provided better granularity into cell signaling behavior than that obtained from a population of cells. Many cells in each of the drug-treated groups continued to produce large quantities of S1P in the face of potent inhibitors. In this initial work, over 1200 single cells were rapidly assayed using 8 chips within 4 h with excellent throughput (>5 cells/min) demonstrating that pTLC is suitable for screening of Sph-K activation and its inhibition within individual cells.

Table 1.

Cell Population with Additional Peaks

% Population	DMSO	PF543	K145	SKI2	FTY720
PeakU	1.3 ± 1.2%	not observed	4.7 ± 4.2%	6.7 ± 3.1%	4 ± 2%
HA-Cy5	4.7 ± 1.2%	2.7 ± 3.1 %	2.7 ± 1.2%	not observed	not observed
U + HA-Cy5	5.3 ± 1.2%	not observed	not observed	2.1 ± 3.5%	6.7 ± 1.2%

Measurement of SphK Activation in Stem-like Cancer Cells (CD34+ Cells).

CD34 is a hematopoietic stem-cell marker that labels a “stem-like” subclass of leukemic blasts associated with therapeutic resistance, upregulated SphK, and poor clinical outcome.^{10, 64} Consequently, both SphK activation and CD34 status have therapeutic implications. Kasumi-3 cells (an AML cell line) were assessed for both SphK activity as well as CD34 expression. Cells were loaded with Sph-Cy5, incubated with or without the SphK inhibitor K145 and then fixed. Cells were stained with phycoerythrin (PE)-labelled anti-CD34 antibody and Hoechst 33342 (DNA) (Fig. S6). After deposition of cells on pTLC microbands, the presence of CD34 was measured by fluorescence microscopy and the microband location recorded (Fig. 3A,B). Consistent with prior reports, 91% ± 2% of Kasumi-3 cells were CD34+.^{65, 66} pTLC was initiated and the fluorescence of Sph-Cy5 and S1P-Cy5 quantified for both CD34+ and CD34− cells. The CD34+ cells demonstrated significantly greater conversion of Sph-Cy5 to its phosphorylated form compared to the CD34− cells (Fig. 3C). There was not a simple linear relationship between CD34 intensity and S1P-Cy5 formed (Fig. S7). Both populations showed a significant decrease in phosphorylation of Sph-Cy5 in the presence of the inhibitor, K145. However, the formation of S1P-Cy5 remained significantly greater for the K145-inhibited CD34+ cells relative to that of the K145-inhibited CD45− cells. Substantial heterogeneity in the percent phosphorylation was displayed by all groups. For example, phosphorylation of Sph-Cy5 varied from 9.4% to 57.1% in the control single cells. Even in the inhibitor-treated groups, the range of phosphorylation was broad (3.7% to 47.4%). In the presence of a potent SphK inhibitor, many of the cells continued to produce considerable amounts of S1P-Cy5 illustrating the challenge faced by targeted therapeutics.

Figure 3. Measurement of SphK Activity and CD34 Expression in Kasumi-3 cells.

(A) The schematic of the assay steps. Cells were loaded with Sph-Cy5, incubated with or without the SphK inhibitor K145 and then fixed, immunostained with PE-conjugated anti-CD34 antibody and Hoechst 33342, followed by deposition on pTLC microbands. (B) Fluorescence microscopy of 5 cells spotted on microbands. (C) “SuperPlots” showing the percentage S1P-Cy5 from single cells with and without inhibition by 10 μM K145 in CD34− and CD34+ cells. “SuperPlots” identify different replicates by color. In these experiments, each color represents the data points from the 3 biologic replicates (N=35 cells each replicate). The triangle represents the average percent phosphorylation in each corresponding experiment. The p values represent a paired two-tailed t test of the average phosphorylation from each repeat in DMSO and K145-treated groups. (D) Percent of phosphorylated Sph-Cy5 (S1P-Cy5/[Sph-Cy5+S1P-Cy5+All Others]) plotted against the total amount of reporter loaded into a cell (moles of Sph-Cy5 + S1P-Cy5 + others) for cells treated with DMSO and 10 μM K145.

The total moles of Sph-Cy5 loaded into a cell calculated from the total area of all fluorescent peaks varied amongst the population suggesting that different cells might have different concentrations of Sph-Cy5 loaded initially. These reagent loading differences in cells are commonly observed for exogenously added reagents such as ion indicators, peptide sensors and therapeutic drugs, and are likely due to variability in uptake and export pumps, metabolic enzymes, cell volume and stochastic features.^{1, 38} When sufficient numbers of cells are examined, a plot of moles of reporter vs percent phosphorylation could be constructed (Fig. 3D). Remarkably for SphK, a plateau or decreasing amount of phosphorylation with increasing amounts of loaded Sph-Cy5 was not observed suggesting that the cellular enzyme was not saturated even at very high intracellular Sph-Cy5 concentrations. Thus, some cells possessed a remarkable capacity to convert sphingosine to its phosphorylated form likely to the advantage of the tumor cell since S1P activates signaling cascades promoting cell proliferation and migration.

Measuring SphK Activity in AML Cells from Patients.

To demonstrate the ability of pTLC to perform assays on primary cells, peripheral blood mononuclear cells (PBMC) from patients with high levels of circulating AML blasts (30.7-69.0% of PBMC) were assayed for the presence of CD34 and SphK activity (Table 2). Cells were loaded with Sph-Cy5 and incubated with DMSO (control) or K145. Cells were fixed, immunostained for CD34, spotted onto the microbands and the CD34 immunofluorescence measured (Fig. S8). As expected, the measured number of CD34+ cells were less than the reported blast count for each patient since the blasts were measured by flow cytometry and needed only one of the following markers (CD34, CD36, CD44, CD123 and CD135) to be defined as a blast (Table 2). In these samples, the CD34− cells are expected to be predominantly normal mononuclear blood cells but with some CD34− leukemic cells. The vast majority of CD34+ cells in these samples are expected to be leukemic cells since <1% of normal PBMCs are CD34+.⁶⁷ After separation the fluorescence of the single-cell lipid reporter and its metabolites was quantified. The average percentage of S1P-Cy5 formed in each cell ranged from 9% - 33% across the different patients (Table 2). When CD34− cells amongst the patients were compared, there were significant differences in S1P-Cy5 (p=0.0002, one way ANOVA, Fig. 4A-C). A similar observation was made for the CD34+ cells amongst the patients (p=0.0187, one way ANOVA), indicating signaling heterogeneity between patients’ cells. In patient 1, the single-cell phosphorylation rate ranged from 0.2% to 78.3% in CD34+ cells and from 0.1% - 62.9% in CD34− cells; however, no significant difference in the formation of S1P-Cy5 was observed between CD34+ and CD34− cells. Similarly, the single-cell phosphorylation rate varied from 1.0% to 82.0% in CD34+ cells and 0.1% - 61.7% in CD34− cells for patient 2, and from 2.1% - 71.3% in CD34+ cells and 1.1% - 52.0% in CD34− cells for patient 3. Notably, in patients 2 and 3, CD34+ cells converted significantly greater Sph-Cy5 to its phosphorylated form than the CD34− cells consistent with the typically more aggressive nature of CD34+ leukemic cells (Table 2 and Fig. 4A-C). There was no simple linear relationship between CD34 intensity and percentage of S1P-Cy5 formed (Fig. S7). An advantage of this approach was that the assay stop time is known and all cells have similar incubation times (30 min in this case). The average velocity (V) of the SphK reaction ranged from 0.01 – 1.89 mM/min (a single human leukemic cell is ~1 picoliter in volume.⁶²). The V of SphK was significantly higher in CD34+ cells (0.31 ± 0.29 mM/min) than in CD34− cells (0.17 ± 0.21 mM/min) in patient 2 (p=0.004, two-tailed t-test. The same trend was not apparent in patients 1 and 3. As with the tissue cultured tumor cell lines, the majority of the patient cells possessed 1-2 peaks with retention times corresponding to that of Sph-Cy5 and/or S1P-Cy5 (4D-F); however, 8.2% of the cells of patient 2 possessed an additional peak of HA-Cy5 (Fig. 4D-F). Interestingly, all the cells with 3 peaks were CD34+ cells, in which the average percent S1P-Cy5 formation was 14.1% while the average percent of HA-Cy5 was 30.0%. The cells from all 3 patients were incubated with K145 to assess inhibition of S1P-Cy5 formation in primary cells. K145 significantly decreased SphK activity in both the CD34+ and CD34− cells of patients 1 and 3. Remarkably, the inhibitor did not significantly decrease formation of S1P-Cy5 in either the CD34+/− cells in patient 2 (Fig. 4G-I). Given the very high blast count of this patient (69%), it is likely that the majority of the CD34− cells were leukemic cells so that both the stem-like (CD34+) and more differentiated (CD34−) cell phenotypes were not impacted by SphK inhibition. Next the capacity of SphK in the cells to become saturated with substrate was examined by plotting the loaded substrate amount vs the percentage S1P-Cy5 formed for each cell (Fig. S9,10). Neither the CD34− nor CD34+ patient cells were readily saturated even at very high intracellular Sph-Cy5 concentrations (up to ~100 amol or ~100 mM for a 1 pL cell) suggesting that the cells possessed a large capacity to form the phosphorylated product, an attribute likely beneficial to tumor cells.

Table 2.

Clinical Data

Patient Sample Data	Clinical Report	% Phosphorylation		pTLC Microband % Cell CD34+	% Phosphorylation		pTLC Microband % Cell CD34+
	% Blasts	+DMSO CD34−	+DMSO CD34+	DMSO Control	+K145 CD34−	+K145 CD34+	K145 Group
Patient #1	30.7	26.2 ± 20.2	33.1 ± 19.9	12.6 ± 2.4	11.5 ± 7.9	14.5 ± 12.7	11.9 ± 2.4
Patient #2	69.0	12.5 ± 11.3	25.3 ± 19.8	22.5 ± 5.4	14.6 ± 16.4	25.4 ± 24.5	21.3 ± 8.3
Patient #3	59.0	17.4 ± 13.3	24.4 ± 15.4	21.9 ± 5.2	9.4 ± 8.1	14.1 ± 11.4	20.6 ± 6.3

Figure 4. Measurement of SphK Activity in Single Cells from AML Patients.

(A-C) Violin plots showing the percent of phosphorylated reporter for the DMSO-treated control and K145-treated group of the CD34 negative and positive cells from AML patients (A - patient 1; B - patient 2; C - patient 3; n=55 cells in each subgroup). (D-F) Chromatographs of the contents of single cells with 3 peaks from patient 2. Shown is the relative fluorescence intensity plotted against the distance from the microband origin at the termination of the separation. Unidentified metabolites are marked as “U.” (G-I) Percent of phosphorylated reporter (S1P-Cy5/[Sph-Cy5+S1P-Cy5+HA-Cy5]) plotted against the CD34 immunofluorescence for DMSO- and K145-treated cells from the patients (G - patient 1; H - patient 2; I - patient 3).

Conclusion and Discussion

When combined with a highly sensitive detection technique (fluorescence), the pTLC platform enabled measurement of as little as 0.04 amol of fluorescent lipids, providing a platform for investigating the heterogeneity of lipid signaling and drug response in single cells loaded with fluorescent lipid reporters. Compared to capillary electrophoresis, pTLC required significantly less expertise with reduced preparation time and increased ease of use. Fabrication of large numbers of microchannels on a single device holds the potential for highly parallel assays with excellent throughput given the fast separation time. Analyte separation in pTLC needs no accessory flow controls (electrical or pressure-driven) since the solvent flows through the matrix by capillary action. Sample detection need not be performed in “real time” since the separation can be stopped and the analyte bands fixed in place by simply evaporating the solvent eliminating diffusion. This greatly simplifies assay readout so that straight forward strategies can be used for high-sensitivity detection of analytes. The picoliter-to-nanoliter sample requirements support assay of reactions formed in ultrasmall volumes such as screening of chemical libraries. The utility of pTLC in assaying leukemic cells opens the opportunity to assay enzyme activity and screen therapeutic drugs in clinical samples with few cells, such as aspirates or biopsies. The compatibility of pTLC with conventional immunofluorescence enables multiplexed assays to identify cell subpopulations. A multitude of fluorescent or clickable lipid substrates are commercially available and are suitable for pTLC assays. In summary, pTLC has the power to reveal normal and aberrant lipid metabolism in small samples of clinical interest in a high-throughput, simple-to-use fashion.

Supplementary Material

An automated platform for single-cell assay on pTLC microchip (Figure S1); pTLC workflow for parallel single-cell assay (Figure S2); Limit of detection using lipid standards (Figure S3); Single cell capture rate (Figure S4); Identifying cellular metabolites using UPLC/MS (Figure S5). Measurement of CD34 expression in single Kasumi-1 cells (Figure S6); Simple linear regression analysis of Kasumi-3 cells (Figure S7); Measurement of CD34 expression in single patient cells (Figure S8); Phosphorylation in single cells (Figure S9); Simple linear regression analysis of patient samples (Figure S10).