Label-free Mass Cytometry for Unveiling Cellular Metabolic

Jun 26, 2019 - Comprehensive analysis of single-cell metabolites is critical since differences in cellular chemical compositions give rise to speciali...
0 downloads 0 Views 3MB Size
Article pubs.acs.org/ac

Cite This: Anal. Chem. XXXX, XXX, XXX−XXX

Label-free Mass Cytometry for Unveiling Cellular Metabolic Heterogeneity Huan Yao, Hansen Zhao, Xu Zhao, Xingyu Pan, Jiaxin Feng, Fujian Xu, Sichun Zhang,* and Xinrong Zhang* Department of Chemistry, Tsinghua University, Beijing 100084, P.R. China

Downloaded via GUILFORD COLG on July 24, 2019 at 15:07:21 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

S Supporting Information *

ABSTRACT: Comprehensive analysis of single-cell metabolites is critical since differences in cellular chemical compositions give rise to specialized biological functions. Herein, we propose a label-free mass cytometry by coupling flow cytometry to ESI-MS (named CyESI-MS) for highcoverage and high-throughput detection of cellular metabolites. Cells in suspension were isolated, online extracted by sheath fluid, and lysed during gas-assisted electrospray, followed by real-time MS analysis. Hundreds of metabolites, including nucleotides, amino acids, peptides, carbohydrates, fatty acyls, glycerolipids, glycerophospholipids, and sphingolipids, were detected and identified from one single cell. Discrimination of four types of cancer cell lines and even three subtypes of breast cancer cells was readily achieved using their distinct metabolic profiles. Furthermore, we screened out 102 characteristic ions from 615 detected peak signals for distinguishing breast cancer cell subtypes and identified 40 characteristic molecules which exhibited significant differences among these subtypes and would be potential metabolic markers for clinical diagnosis. CyESI-MS is expected to be a new-generation mass cytometry for studying cell heterogeneity on the metabolic level. introducing fluorescence probes or mass tags to the targets. Many small molecule metabolites are difficult or impossible to measure by the two techniques because there is no easy way to keep the small molecules and a binding agent associated with the cell.19 Therefore, cellular metabolites, which are end products and essential mediators of cellular behavior,19 are largely excluded. Metabolic profiles of cells directly reflect realtime cellular states and have been proved to be specific with cell phenotypes; for example, reprogrammed energy metabolism has been regarded as one of the hallmarks of cancer progression.25,26 Despite the valuable genomics, transcriptomics, or proteomics information provided by the well-established workflows, an efficient high-throughput and high-coverage approach is in urgent demand to unveil the downstream metabolomics information on single cells as a reflection of the real-time cell physiological state. Single-cell metabolomics analysis using mass spectrometry has attracted increasing interest in the past decade.27−29 A number of analytical techniques coupled with MS have been proposed to realize single-cell metabolite analysis without labeling. ESI-MS is the common approach for single-cell metabolite analysis. A series of techniques derived from ESIMS were suitable for ionization of small-volume samples that

D

ifferent types of cells play specific roles in the life activities of a complex living organism.1 Unveiling cell heterogeneity provides deep insights of types,2 cycle stages,3 degrees of differentiation,4,5 and fates5,6 of cells, which is expected to contribute to the prognosis, diagnosis, and therapy of metabolism-related diseases.7,8 Methods for the characterization of phenotypes and functions of single cells, such as fluidic platforms derived from flow cytometry,9 single-cell sequencing,10−14 and RNA fluorescence in situ hybridization,15 have revealed heterogeneity of cell populations which were once considered homogeneous. Among these methods, fluorescence-based flow cytometry and its derivatives have been widely used in analyzing immunophenotypes of heterogeneous cell populations,15−17 owing to its advantages of high-throughput detection of single cells. However, due to the commonly occurring spectral overlaps between fluorescence channels, it is hard to detect a large number of features from samples and only 10−20 specific antigens can be analyzed simultaneously from one cell.18 Mass cytometry,19,20 which combines the high-throughput single-cell analysis feature of flow cytometry with the simultaneous acquisition of multiplexed information via inductively coupled plasma mass spectrometry (ICP-MS), has become the new choice for indepth characterization of heterogeneous cell populations to interrogate multiple levels of cellular metabolism.21−24 Both fluorescence-based flow cytometry and ICP-MS-based mass cytometry analyze cellular components indirectly by © XXXX American Chemical Society

Received: March 20, 2019 Accepted: June 26, 2019 Published: June 26, 2019 A

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry was required for single-cell analysis, such as nano-ESI-MS,30−34 desorption-ESI-MS,35 laser ablation-ESI-MS, 36−38 single probe,39 and T-probe.40 Moreover, capillary electrophoresisESI-MS41−44 could realize online separation and detection of single-cell metabolites. Our group has also made efforts in developing single-cell MS techniques. Probe ESI-MS was employed in single-cell metabolite analysis of Allium cepa cells.45 Pulsed-dc-ESI-MS was developed for picoliter-volume sample analysis in order to obtain MS2 information from single plant and mammalian cells.46 Droplet extraction combined with Pico-ESI-MS was used to extract and analyze metabolites with desired properties (hydrophobic or hydrophilic) from single cancer cells.47−50 These approaches made it possible to study cell-to-cell metabolic heterogeneity and have already revealed that cells of different types or even with identical genotype have their specific metabolic profiles which were closely related to their functions. However, the detection throughput of most ESI-MS based single-cell metabolite analysis methods was severely limited by complex single cell sampling processes (e.g., extracting cellular contents manually). Several methods have also been proposed to improve the speed of single-cell metabolite analysis, but the excess dilution of cellular contents compromised the number of metabolites analyzed, as no more than 20 lipids have been detected.51,52 Therefore, a technique with the capability of high-throughput and high-coverage metabolic profiling is highly demanded for differentiating cell types as well as unraveling cell heterogeneity. In this work, we propose a label-free mass cytometry by employing ESI-MS as the detector of flow cytometry (named CyESI-MS). Label-free analysis of hundreds of single-cell metabolites, including nucleotides, amino acids, peptides, carbohydrates, fatty acyls, glycerolipids, glycerophospholipids, and sphingolipids, was achieved in a high-throughput way (∼38 cells per minute). In addition, a systematic single-cell metabolite analysis platform was established to process the data and unveil cellular metabolic heterogeneity. Discrimination of four cancer cell types (HeLa, MCF7, A549, and HepG2) as well as subtypes of breast cancer and liver cancer cells was realized using single-cell metabolic profiles. Forty characteristic metabolites were screened out as the indicators of different subtypes of breast cancer cells, which were potential metabolic biomarkers for discrimination.

Research Biotechnology Co. Ltd. (Shanghai, China). Dulbecco’s modification of eagle’s medium (DMEM), DMEM/F12, RPMI 1640 (w/o Hepes), Leibovitz’s L-15 medium (L-15), Dulbecco’s phosphate buffered saline (DPBS), fetal bovine serum (FBS), Trypsin-EDTA (0.25%), penicillin-streptomycin (100 U·mL −1 ) were all purchased from Gibco Life Technologies (Carlsbad, CA, USA). Construction of CyESI-MS. Figure 1A shows the schematic of CyESI-MS, which was a combination of

Figure 1. (A) Schematic of label-free mass cytometry (CyESI-MS) system and (B) photo of CyESI-MS showing its basic construction.

commercially available components and did not require complicated mechanical processing. The central part of CyESI-MS was three coaxial capillaries constructed by connecting different sizes of fused silica capillaries using metal tees and peek pipes. The inner capillary (150 μm O.D., 50 μm I.D.) was named the cell injection channel. Cell suspension was introduced into the channel by a microfluidic flow control system (MFCS-EZ, Fluigent, France). The inbetween capillary (350 μm O.D., 200 μm I.D.) was used to transfer the sheath fluid supplied via a syringe pump (Syringe Pump 11 Elite, Harvard Apparatus, Canada). The outlet end of the in-between capillary was 5 mm longer than the inner one. Cells were focused and isolated by the surrounding sheath fluid and flowed toward the outlet along the axis, and the contents of the cells were online-extracted by the sheath fluid. DC voltage power (purchased from Tianjin Dongwen High Voltage Power Supply Co., Tianjin, China) was transmitted to the sheath fluid through the conductive tee to generate electrospray. The outer capillary (750 μm O.D., 530 μm I.D.), coaxially wrapped outside the in-between capillary, was used as the carrier gas tube and was 2 mm shorter than the in-between capillary. Nitrogen was used to assist solvent evaporation of charged droplets for better ionization efficiency. The outlet end of the coaxial three-layer capillary was 5 mm away from the inlet of MS. The high flow of the carrier gas also ensured



EXPERIMENTAL SECTION Reagents and Materials. Methanol (HPLC-grade) was purchased from Sigma-Aldrich (St. Louis, MO, USA). ULCMS-grade formic acid and LC-MS grade ammonia solution (≥25% in H2O) were obtained from Aladdin Chemicals (Shanghai, China). Ultrapure water (resistance ≥18 MΩ· cm−1) was prepared from a Milli-Q water purification system. Fused silica capillary tubes were purchased from Polymicro Technologies (Molex, USA), and peek columns were purchased from VICI Jour (Switzerland). Metal tee connectors were purchased from Nanjing Haosheng Experimental Instrument Co., Ltd. (Nanjing, China). Nitrogen (≥99.999%) was supplied by Liquefied Air Co., Ltd. (Tianjin, China). MS calibration solutions for both ion modes were purchased from Thermo Scientific (San Jose, CA, USA). HeLa, A549, HepG2, MCF7, MCF 10A, and L-02 cells were purchased from National Infrastructure of Cell Line Resource (Beijing, China). BT-474, MDA-MB-468, HCCC-9810, and SMMC-7721 cells were obtained from Shanghai Enzyme B

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

searching characteristic product ions in MS2 spectra obtained from population cells (Figure S2). A few numbers of isomeric metabolites that could not be distinguished by existing methods were all listed out tentatively. Lastly, machine learning based on the t-distributed stochastic neighbor embedding (t-SNE), the linear discriminant analysis (LDA), and the Kruskal−Wallis nonparametric hypothesis test (K-W test) was used to mine the complex metabolic data sets. The t-SNE algorithm is a powerful technique for nonlinear reduction of high-dimensional single cell metabolic data sets which can realize discrimination of subtle groups qualitatively and visualize the difference in the two dimensional plane. The LDA algorithm based on the clustering results of t-SNE was used to quantitatively characterize the results of the discrimination and predict cell types affiliation. The K-W test was used to evaluate the level of statistical significance (P-value) for single-cell metabolic information between cell subtypes.

that sample ions enter MS coaxially which reduced the loss of sample. A microscope (MV-EM120M, MicroVision, China) was used to observe the flow of cells in the capillary and their morphological changes at the interface with the sheath fluid. The entire device was fixed to a custom-built three-dimensional operation platform. A photo of CyESI-MS at work is provided in Figure 1B. Preparation of Cell Suspensions. All cells used in this study were adherently cultured in vitro under optimum conditions which are listed in Table S1. Cell suspension was freshly prepared prior to CyESI-MS analysis. Cells were trypsinized using Trypsin-EDTA (0.25%) and subsequently dispersed into DPBS. Then, DPBS containing residual culture medium was removed by centrifugation (1000 rpm, 5 min), and cells were resuspended into 140 mM ammonium formate aqueous solution (pH = 7.3). Details of solvent selection for cell suspension are described in Experiment S1, Figure S1, and Tables S6, S7, S10, and S12. Cell concentrations were determined by a hemocytometer. Cell Injection, Metabolite Extraction, and Ionization. Prior to cell injection, the cell injection channel was thoroughly washed with 140 mM aqueous ammonium formate. The sheath fluid, methanol, was used to clean the in-between capillary. When the background total ion chromatogram (TIC) was stable, cell suspension was introduced into the device at a flow rate of 1 μL/min. In the positive ion mode, the sheath fluid was doped with 1% formic acid by volume, and a high voltage of +2.3 kV was applied. In the negative ion mode, 1% ammonia solution by volume was added and a voltage of −2.3 kV was applied. The pressure of the carrier gas was adjusted to 0.6 MPa in order to maximize signal intensity. Mass Spectrometry. All mass spectrometry detection was performed on a QE-Orbitrap mass spectrometer (Thermo Scientific, San Jose, CA). The optimal MS parameters were set as follows: capillary temperature = 320 °C, resolution = 35000, AGC target = 106, maximum inject time = 10 ms, microscans = 1 in both positive and negative ion modes. QE-Orbitrap MS was calibrated using commercially available calibration solutions prior to all measurements. Data Analysis. A systematic metabolite data analysis platform was developed to perform raw data processing, background elimination, single-cell determination, peak assignment, as well as machine learning. Multistep pretreatments were employed on the raw MS data to extract information on single-cell metabolites. Cell-related ion signals were first extracted. Among them, ions whose signal-to-noise ratio (S/ N ratio) was greater than three and occurrence frequency in all cell events was greater than 20% were considered as detected signals. These selected ions, along with their average relative intensities in each pulse, constituted the metabolic MS profile of each event. After that, a histogram gated on two classes of lipids was used to discriminate single-cell events. Lipid metabolites were the major components of membranes whose contents in the cell were relatively stable and were detected with high sensitivity in CyESI-MS. The signal intensity would be significantly larger or smaller when a multicellular event or cell debris was detected, which was the basis for extracting single-cell events. Besides, the extracted ions were assigned to hundreds of cellular metabolites through matching the accurate mass with the standards in online databases including the Human Metabolome Database (http://www.hmdb.ca/) and METLIN metabolite database (https://metlin.scripps.edu/) as well as



RESULTS AND DISCUSSION Configuration of CyESI-MS. CyESI-MS was composed of four sections: a cell suspension transfer system, sheath fluid for dispersing cells and extracting the metabolites, electrospray ionization interface, and an Orbitrap MS as the detector. Cell suspension was pumped into the device at a constant flow rate, and cells were isolated and linearly focused by the sheath fluid (Figure S3). Cellular contents were online extracted when cells contacted with the sheath fluid. The interaction time during flowing cells extracted by sheath fluid could determine the extraction efficiency and single-cell resolution. When the flow rates of cell suspension and sheath fluid were fixed, the interval distance between the outlet of cell injection channel and the in-between capillary became the only factor of the interaction time. After detailed optimization (Experiment S2, Figure S4), 5 mm was chosen as the optimal contact distance, which meant the interaction time was ∼0.86 s. A 2.3 kV DC voltage was applied to the sheath fluid via an electrode to form a stable electrospray at the outlet of the capillary. Nitrogen was supplied between the outer and inbetween capillaries to facilitate the desolvation process. Cells were lysed by gas-assisted electrospray ionization, which ensured more complete metabolite extraction (Figure S5). Multiple ionized cellular molecules from one cell were detected by Orbitrap mass analyzer simultaneously. We evaluated the performance of CyESI-MS in single-cell metabolite analysis. A mixed cell suspension composed of three types of cancer cell lines (HeLa, MCF7, and A549) was introduced into the device at a flow rate of 1 μL/min. As shown in Figure 2A, pulselike signals appeared in TIC when the suspension was introduced into the device. Each pulse contained metabolite information from a cell-related event (Figure S6). After multistep data processing, three distinctive types of single-cell metabolic profiles were extracted from raw MS data and assigned to HeLa, MCF7, and A549 cell lines, respectively (Figure 2C−E) by comparing with the home-built cancer cell metabolite profile database using the LDA algorithm, which would be further explained in the following sections. It should be mentioned that the metabolite profiles acquired by CyESI-MS not only reflect the original cellular metabolic level but also include the metabolic change introduced by environmental stimuli, such as shear stress. However, the effect of the external stimuli was proved to be negligible in the current CyESI-MS for unveiling the

C

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry

observing cell movements in the in-between capillary under a microscope (Figure S3). At a 10 μL/min rate, the average intensity of m/z 804.5760 was also the highest (Figure 3B). Overall, 10 μL/min was chosen as the optimal sheath flow rate for subsequent experiments. Further experiments were carried out to verify that cells were detected by MS individually. Cells suspensions at different concentrations (5 × 103, 1 × 104, 2 × 104, 4 × 104 cells/mL) were examined under the optimal flow rate by CyESI-MS. Based on the flow velocity of cell suspension (1 μL/min) and cell concentration, cell numbers in a given time could be calculated. As shown in Figure 3C and Table S2, the number of pulse signals in 1 min was proportional to the cell concentration and the number of detected cell events was substantially the same as the calculated number, indicating that most of the cells were monodisperse and detected individually. The detection throughput could reach 38 cells/min when cell concentration was 4 × 104 cells/mL. These results indicated that CyESI-MS performed well in the high-throughput assay of single cells. Profiling Single-Cell Metabolites. The metabolite coverages of CyESI-MS were assessed in both positive and negative ion modes. In the range of m/z 100−1000, 421, and 425, ion signals related to cellular metabolites were picked out in positive and negative modes, respectively (Figure 4A, Figure S8), of which 174 and 117 cellular metabolites were assigned based on accurate mass measurements as well as MS2

Figure 2. Mixed cell suspension composed of three cancer cell lines analyzed by CyESI-MS. (A) TIC and (B) EIC of m/z 804.5760 acquired in the negative ion mode. Three distinct single-cell metabolic profiles were extracted from the raw MS data and assigned to (C) HeLa, (D) MCF7, and (E) A549 cells, respectively.

metabolite heterogeneity between cells (Experiment S5, Figure S7). The peak at m/z 804.5760 (assigned to PS 37:0 which was a component of cell membranes) was chosen as a marker of the presence of cells. The frequency of m/z 804.5760 matched that of TIC, and the intensity of m/z 804.5760 dropped to zero sharply between two pulses in the extracted ion chromatogram (EIC) (Figure 2B), which indicated that almost no overlap existed between cell events. Over 400 cell events were detected in 20 min. The results indicated that the CyESI-MS had the ability to capture single-cell metabolite signals in a flow cytometry way. The flow rate of the sheath fluid was one of the important factors to isolate single cells. For optimization, a HeLa cell suspension with concentration of 2.5 × 104 cells/mL was constantly introduced into the device at a flow rate of 1 μL/ min. As shown in Figure 3A, when the flow rate of the sheath

Figure 3. Effect of the flow rate of the sheath fluid on single-cell detection. (A) Relationship between the flow velocity of the sheath fluid and the number of pulse signals per minute. (B) Changes of the intensity of m/z 804.5760 with the flow velocity of the sheath fluid with other parameters being constant. (C) The number of pulses was proportional to the cell concentration.

fluid was set as 10 μL/min, the frequency of the pulse signals was 24.25/min (average of four parallel experiments). When the flow velocity increased, the frequency of the pulse signals decreased because of the cells slipping away during the interval between two times of the MS scan. At a flow velocity of 6 to 8 μL/min, the frequency of cell occurrence decreased because there were some overlaps between pulse signals. There were no stable pulselike ion signals observed when the flow rate was less than 4 μL/min, which indicated that cells were not separated properly. This result was further confirmed by

Figure 4. Cellular metabolite assignment using accurate mass combined with MS2 information from population cells. (A) Average normalized mass spectrum of single-cell metabolites of HeLa in the positive ion mode. Abundant metabolites were labeled on the mass spectra in three selected regions, including (B) m/z 100−300, (C) m/ z 300−650, and (D) m/z 690−840. (E) Classifications of detected metabolites in both ion modes. D

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry information obtained from population cells (Table S8−9). Abundant metabolites assigned in positive mode were labeled on the mass spectra of selected m/z regions (Figure 4B−D). In m/z 100−300, 14 common amino acids (proline, valine, threonine, leucine, aspartate, glutamine, lysine, methionine, histidine, phenylalanine, arginine, glutamic acid, tyrosine, and tryptophan) and their derivatives as well as small-molecular lipids (including acyl carnitines, glycerophosphocholines and fatty amides) were detected. Some amino acid analogues, peptides, nucleotides, nucleotide sugars, acyl carnitines, and lyso-phospholipids lay in m/z 300−650. In m/z 650−1000, most detected metabolites were assigned to lipids, for example, phosphatidylcholines (PC), phosphatidylethanolamines (PE), phosphatidylglycerols (PG), phosphatidylserines (PS), phosphosphingolipids, as well as some glycerolipids. Figure 4E summarized the detected cellular contents, which covered most groups of important functional metabolites. The details of detected metabolite classification are shown in Tables S3 and S4. It should be mentioned that the metabolite coverage was related to the selection of scan range (Experiment S3, Figure S9, and Tables S4, S5, S9, and S10); for example, 120 lipids were detected using the scan range m/z 300−1000, which was 39.5% more than that of m/z 100−1000. These results confirmed that CyESI-MS was effective in detecting and identifying a broad range of metabolites in single cells. The composition of sheath fluid was critical since it determined the coverage and signal intensities of extracted metabolites. Several solvents with different properties were examined, including methanol, ethanol, acetonitrile, acetone, chloroform, dimethylformamide, H2O, as well as their mixtures (Experiment S4, Figure S10). Comparison of the numbers and classification summaries of assigned metabolites were performed using methanol and methanol:H2O = 1:1 solution as sheath fluid (Figure S11 and Tables S6, S7, S10, and S11). The results indicated that pure methanol exhibited the highest intensity of all referred cellular metabolites as well as the highest coverage of detected metabolites owing to its modest polarity that promised good solubility of most metabolites. Therefore, all cells in this work were analyzed using methanol as sheath fluid solvent. It should be mentioned that due to the inherent selective properties of the extraction process which mainly related to the permeability of the cell membrane under different solvents and the heterogeneous nature of the cellular metabolites, the substances online-extracted by CyESI-MS were just part of all cellular metabolites even under the optimal extraction procedure. Therefore, CyESI-MS was used to unveil the metabolic heterogeneity by comparing the relative intensities of metabolite ions, rather than the absolute concentrations. Discrimination of Cancer Cell Types. Four types of cancer cell lines, including HeLa, A549, MCF7, and HepG2, were used to verify that single-cell metabolic information was sufficient for cell differentiation. Figure 5A showed the relative intensities of metabolite ions from each cell. The complex metabolic information on different cancer cells was visualized in the two-dimensional plane by the t-SNE algorithm. As shown in Figure S12, separation of HeLa (n = 204), A549 (n = 167), MCF7 (n = 121), and HepG2 (n = 132) was clearly observed, indicating that single-cell metabolite profiles between different cancer types were significantly heterogeneous. The above experiment was repeated using the same experimental setup, and cells with different subculture times

Figure 5. Results of the discrimination of cancer cell types. (A) Heat map of the single-cell metabolite profiles. The color map shows the relative intensity of each metabolite ions (row) in each single cell (column). (B) The t-SNE results of two different batches experiments were comparable. (C) t-SNE analysis of five samples. HeLa cells (orange points), A549 cells (blue points), MCF7 cells (purple points), HepG2 cells (green points), and a mixture of the first three cell types (gray points). (D and E) The discrimination of four cancer cell types was realized by two specific MS regions, m/z 300−650 and m/z 650−1000, respectively.

and nearly identical clustering results were achieved. After that, the first set of data on the four cancer cell types (204 HeLa cells, 167 A549 cells, 121 MCF7 cells, and 132 HepG2 cells) were used to train the LDA algorithm to build the cancer cell metabolite profile database. The other set of data (230 HeLa cells, 312 A549 cells, 235 MCF7 cells, and 227 HepG2 cells) was treated as the test set. LDA result of the discrimination of four cancer cell types yielded a sensitivity of at least 96.9% and specificity of at least 99.1% (Table 1), indicating that CyESIMS possessed good repeatability and little batch effect. Table 1. LDA Result of Discriminating Four Types of Cancer Cell Lines

The mixed cell suspension mentioned in the section Configuration of CyESI-MS was also evaluated by the homebuilt cancer cell metabolite profile database. Among 406 detected cells, 137 (33.74%) HeLa cells, 61 (15.02%) A549 cells, and 208 (51.23%) MCF7 cells were assigned by the LDA algorithm. The result could also be roughly visualized by t-SNE clustering analysis (Figure 5C). Overall, our results suggested that the combination of CyESI-MS and the LDA algorithm would become a new strategy for quantitative prediction of the composition of a heterogeneous cell population using metabolic information. It should be mentioned that when applying CyESI-MS to cancerous tumors, the LDA algorithm E

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry should be trained with several samples from different patients and tissue regions considering the effect of microenvironment heterogeneity on metabolism, to make the data processing platform more robust for cell typing. Benefited from the ability of high-coverage detection, the metabolic heterogeneity within a specific class could be studied. Ions in m/z 650−1000 were mainly assigned to lipids, including PE, PS, phosphatidylinositols (PI), phosphatidylglycerols (PG), as well as some sphingolipids, while most ions of metabolites related to energy metabolism lay in the region of m/z 300−650. Metabolic information in the selected two m/z regions was used for discrimination of four cancer cell lines by t-SNE analysis (Figure 5D, E). The Euclidean distance representing the degree of separation was 8.9968 for m/z 300− 650, which was larger than m/z 650−1000, whose separation distance was 7.3565. The t-SNE clustering results illustrated that the metabolic heterogeneities of energy-metabolism related molecules as well as lipid metabolites both existed among four cell types, and the difference was more apparent in the former group. Our study provided a preliminary sketch that heterogeneity of cellular energy-metabolism related small molecules was more amenable as a discrimination marker for different cancer cell lines, which was also consistent with previous studies that reprogramming energy metabolism was one of the marks of cancer.27 These four types of cancer cell lines were chosen as models in this work to illustrate the feasibility and stability of CyESIMS and to demonstrate the ability of CyESI-MS to distinguish cell types in a heterogeneous cell population. After method establishment, CyESI-MS was applied to study some biologically meaningful models, such as cancer cell subtypes. Metabolite Heterogeneity of Cancer Cell Subtypes. Comparing with the conventional classification method based on immunohistochemical markers, discrimination of different subtypes on the metabolic level would give downstream information which reflected real-time cellular state. We further evaluated the capability of CyESI-MS workflows in classifying subtypes of cancer cells. As shown in Figure 6A and Figure S13, discrimination of three subtypes of breast cancer cells (MCF7, MDA-MB-468 and BT474), three subtypes of liver cancer cells (HepG2, HCCC-9810, and SMMC-7721), and their corresponding normal cell lines (MCF 10A and L-02) was feasible based on the distinct single-cell metabolite profiles acquired by CyESI-MS. Furthermore, differentiation analysis was performed for screening out characteristic metabolites which were potential biomarkers for distinguishing cancer cell subtypes. The correlations between P-values (evaluated by K-W test) and fold changes in relative abundances for all detected singlecell metabolic information (both assigned and unassigned) within breast cancer cells and the corresponding normal cell lines were visualized by volcano plots (Figure 6B and Figure S14). Orange and blue data points represented that the corresponding molecules displayed large intensity alterations (up-regulated or down-regulated) along with statistical significances (P < 0.05 and fold change >2.0). The heat map in Figure 6C contained 102 characteristic metabolite ions with significant difference in the relative intensities. Figure 6D showed that subtypes of breast cancer cells and the corresponding normal cell line could be separated using these specific metabolite ions, in which 40 metabolites, including some lipid species and metabolites of glycolysis (e.g., UDP-sugars, AMP, NAD) were assigned (Table S13). As

Figure 6. Discrimination results of cancer cell subtypes. (A) t-SNE clustering results of three subtypes of breast cancer cell lines including MCF7 (blue points), MDA-MB-468 (yellow points), and BT474 (purple points) as well as the corresponding normal cell lines MCF 10A (orange points). (B) Volcano plot displayed the correlations between P-values and relative intensities (fold changes) for all detected signals within MCF7 and MCF 10A. Orange and blue data points represented for the characteristic ions with significant difference (P < 0.05 and fold change >2.0) between MCF7 and MCF 10A. (C) Heatmap of 102 characteristic metabolite ions and their relative intensities in each cell. (D) t-SNE analysis of these four subtypes based on 102 distinctive ions.

a result, these characteristic metabolites would become potential metabolic biomarkers for discrimination.



CONCLUSION In conclusion, CyESI-MS, a label-free mass cytometry, has been developed to realize online extraction and real-time ESIMS analysis of single cells in a high-throughput way. Different from the fluorescence or ICP-MS-based flow cytometry, hundreds of cellular metabolites could be simultaneously collected in a label-free way to predict cell types using machine learning based on the t-SNE clustering algorithm. Discrimination of four types of cancer cell lines, and even three cell subtypes of breast cancer, was readily achieved with this highcoverage method. Characteristic metabolites of different breast cancer cells were picked out and identified. We anticipate that the principle of CyESI-MS would serve as a paradigm for new generation mass cytometry and contribute to biological sciences and clinical studies.



ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.analchem.9b01419. Solvent and flow rate optimizations of cell suspension and sheath fluid, metabolite identification, single-cell metabolite profile, cell type/subtype discrimination, metabolites assigned from a single HeLa cell, and classification of assigned metabolites and potential metabolic biomarkers (PDF) F

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX

Article

Analytical Chemistry



(19) Spitzer, M. H.; Nolan, G. P. Cell 2016, 165, 780−791. (20) Han, G.; Spitzer, M. H.; Bendall, S. C.; Fantl, W. J.; Nolan, G. P. Nat. Protoc. 2018, 13, 2121−2148. (21) Lin, D.; Maecker, H. T. Methods Mol. Biol. 2018, 1678, 37−47. (22) Korin, B.; Dubovik, T.; Rolls, A. Nat. Protoc. 2018, 13, 377− 391. (23) Brown, H. M. G.; Arriaga, E. A. Anal. Chem. 2018, 90, 13315− 13321. (24) Bendall, S. C.; Davis, K. L.; Amir, E.-a. D.; Tadmor, M. D.; Simonds, E. F.; Chen, T. J.; Shenfeld, D. K.; Nolan, G. P.; Pe’er, D. Cell 2014, 157, 714−725. (25) Hsu, P. P.; Sabatini, D. M. Cell 2008, 134, 703−707. (26) Hanahan, D.; Weinberg, R. A. Cell 2011, 144, 646−674. (27) Rubakhin, S. S.; Romanova, E. V.; Nemes, P.; Sweedler, J. V. Nat. Methods 2011, 8, S20−S29. (28) Comi, T. J.; Do, T. D.; Rubakhin, S. S.; Sweedler, J. V. J. Am. Chem. Soc. 2017, 139, 3920−3929. (29) Zhang, L.; Vertes, A. Angew. Chem., Int. Ed. 2018, 57, 4466− 4477. (30) Mizuno, H.; Tsuyama, N.; Harada, T.; Masujima, T. J. Mass Spectrom. 2008, 43, 1692−1700. (31) Fujii, T.; Matsuda, S.; Tejedor, M. L.; Esaki, T.; Sakane, I.; Mizuno, H.; Tsuyama, N.; Masujima, T. Nat. Protoc. 2015, 10, 1445− 1456. (32) Zhang, L.; Vertes, A. Anal. Chem. 2015, 87, 10397−10405. (33) Jarmusch, A. K.; Pirro, V.; Baird, Z.; Hattab, E. M.; CohenGadol, A. A.; Cooks, R. G. Proc. Natl. Acad. Sci. U. S. A. 2016, 113, 1486−1491. (34) Zhu, H.; Zou, G.; Wang, N.; Zhuang, M.; Xiong, W.; Huang, G. Proc. Natl. Acad. Sci. U. S. A. 2017, 114, 2586−2591. (35) Gonzalez-Serrano, A. F.; Pirro, V.; Ferreira, C. R.; Oliveri, P.; Eberlin, L. S.; Heinzmann, J.; Lucas-Hahn, A.; Niemann, H.; Cooks, R. G. PLoS One 2013, 8, No. e74981. (36) Shrestha, B.; Vertes, A. Anal. Chem. 2009, 81, 8265−8271. (37) Stolee, J. A.; Shrestha, B.; Mengistu, G.; Vertes, A. Angew. Chem., Int. Ed. 2012, 51, 10386−10389. (38) Stolee, J. A.; Vertes, A. Anal. Chem. 2013, 85, 3592−3598. (39) Pan, N.; Rao, W.; Kothapalli, N. R.; Liu, R.; Burgett, A. W.; Yang, Z. Anal. Chem. 2014, 86, 9376−9380. (40) Liu, R.; Pan, N.; Zhu, Y.; Yang, Z. Anal. Chem. 2018, 90, 11078−11085. (41) Nemes, P.; Knolhoff, A. M.; Rubakhin, S. S.; Sweedler, J. V. Anal. Chem. 2011, 83, 6810−6817. (42) Nemes, P.; Rubakhin, S. S.; Aerts, J. T.; Sweedler, J. V. Nat. Protoc. 2013, 8, 783−799. (43) Onjiko, R. M.; Moody, S. A.; Nemes, P. Proc. Natl. Acad. Sci. U. S. A. 2015, 112, 6545−6550. (44) Onjiko, R. M.; Portero, E. P.; Moody, S. A.; Nemes, P. Anal. Chem. 2017, 89, 7069−7076. (45) Gong, X.; Zhao, Y.; Cai, S.; Fu, S.; Yang, C.; Zhang, S.; Zhang, X. Anal. Chem. 2014, 86, 3809−3816. (46) Wei, Z.; Xiong, X.; Guo, C.; Si, X.; Zhao, Y.; He, M.; Yang, C.; Xu, W.; Tang, F.; Fang, X.; Zhang, S.; Zhang, X. Anal. Chem. 2015, 87, 11242−11248. (47) Zhang, X. C.; Zang, Q.; Zhao, H.; Ma, X.; Pan, X.; Feng, J.; Zhang, S.; Zhang, R.; Abliz, Z.; Zhang, X. Anal. Chem. 2018, 90, 9897−9903. (48) Zhang, X. C.; Wei, Z. W.; Gong, X. Y.; Si, X. Y.; Zhao, Y. Y.; Yang, C. D.; Zhang, S. C.; Zhang, X. R. Sci. Rep. 2016, 6, 24730. (49) Wang, R.; Zhao, H.; Zhang, X.; Zhao, X.; Song, Z.; Ouyang, J. Anal. Chem. 2019, 91, 3667−3674. (50) Feng, J.; Zhang, X.; Huang, L.; Yao, H.; Yang, C.; Ma, X.; Zhang, S.; Zhang, X. Anal. Chem. 2019, 91, 5613−5620. (51) Huang, Q.; Mao, S.; Khan, M.; Zhou, L.; Lin, J. M. Chem. Commun. 2018, 54, 2595−2598. (52) Chen, F.; Lin, L.; Zhang, J.; He, Z.; Uchiyama, K.; Lin, J.-M. Anal. Chem. 2016, 88, 4354−4360.

AUTHOR INFORMATION

Corresponding Authors

*Prof. Sichun Zhang e-mail: [email protected]. *Prof. Xinrong Zhang e-mail: [email protected]. ORCID

Hansen Zhao: 0000-0003-2978-5756 Fujian Xu: 0000-0001-8797-4788 Sichun Zhang: 0000-0001-8927-2376 Notes

The authors declare no competing financial interest.



ACKNOWLEDGMENTS We acknowledge Qi Li (Tsinghua University, China) for his generous helpful suggestions on the construction of the device and Ruihua Wang (Beijing Normal University, China) for her efforts on culturing cells. This research was supported by the Ministry of Science and Technology of China (Grant 2016YFF0100301) and the National Natural Science Foundation of China (Grants 21621003 and 21727813).



REFERENCES

(1) Rozenblatt-Rosen, O.; Stubbington, M. J. T.; Regev, A.; Teichmann, S. A. Nature 2017, 550, 451−453. (2) Plass, M.; Solana, J.; Wolf, F. A.; Ayoub, S.; Misios, A.; Glazar, P.; Obermayer, B.; Theis, F. J.; Kocks, C.; Rajewsky, N. Science 2018, 360, eaaq1723. (3) Ly, T.; Whigham, A.; Clarke, R.; Brenes-Murillo, A. J.; Estes, B.; Madhessian, D.; Lundberg, E.; Wadsworth, P.; Lamond, A. I. eLife 2017, 6, No. e27574. (4) Frisz, J. F.; Choi, J. S.; Wilson, R. L.; Harley, B. A.; Kraft, M. L. Anal. Chem. 2012, 84, 4307−4313. (5) Ng, N. H. J.; Teo, A. K. K. Cell Death Dis. 2018, 9, 222. (6) Biddy, B. A.; Kong, W.; Kamimoto, K.; Guo, C.; Waye, S. E.; Sun, T.; Morris, S. A. Nature 2018, 564, 219−224. (7) Iriyama, N.; Miura, K.; Hatta, Y.; Uchino, Y.; Kurita, D.; Takahashi, H.; Sakagami, H.; Sakagami, M.; Kobayashi, Y.; Nakagawa, M.; Ohtake, S.; Iizuka, Y.; Takei, M. Med. Oncol. 2016, 33, 87. (8) Bandura, D. R.; Baranov, V. I.; Ornatsky, O. I.; Antonov, A.; Kinach, R.; Lou, X.; Pavlov, S.; Vorobiev, S.; Dick, J. E.; Tanner, S. D. Anal. Chem. 2009, 81, 6813−6822. (9) Ng, E. X.; Sun, G.; Wei, S. C.; Miller, M. A.; DasGupta, R.; Lam, P. Y. P.; Chen, C. H. Anal. Chem. 2019, 91, 1277−1285. (10) Rantalainen, M. Briefings Funct. Genomics 2018, 17, 273−282. (11) Alemany, A.; Florescu, M.; Baron, C. S.; Peterson-Maduro, J.; van Oudenaarden, A. Nature 2018, 556, 108−112. (12) Andrews, T. S.; Hemberg, M. Mol. Aspects Med. 2018, 59, 114− 122. (13) Enge, M.; Arda, H. E.; Mignardi, M.; Beausang, J.; Bottino, R.; Kim, S. K.; Quake, S. R. Cell 2017, 171, 321−330. (14) Artegiani, B.; Lyubimova, A.; Muraro, M.; van Es, J. H.; van Oudenaarden, A.; Clevers, H. Cell Rep. 2017, 21, 3271−3284. (15) Porichis, F.; Hart, M. G.; Griesbeck, M.; Everett, H. L.; Hassan, M.; Baxter, A. E.; Lindqvist, M.; Miller, S. M.; Soghoian, D. Z.; Kavanagh, D. G.; Reynolds, S.; Norris, B.; Mordecai, S. K.; Quan, N.; Lai, C.; Kaufmann, D. E. Nat. Commun. 2014, 5, 5641. (16) Hagberg, C. E.; Li, Q.; Kutschke, M.; Bhowmick, D.; Kiss, E.; Shabalina, I. G.; Harms, M. J.; Shilkova, O.; Kozina, V.; Nedergaard, J.; Boucher, J.; Thorell, A.; Spalding, K. L. Cell Reports 2018, 24, 2746−2756. (17) Kopeikina, E.; Dukhinova, M.; Ponomarev, E. D. Methods of Study of Neuron Structural Heterogeneity: Flow Cytometry vs. Laser Interferometry. In Cellular Heterogeneity: Methods and Protocols; Barteneva, N. S., Vorobjev, I. A., Eds.; 2018; pp 155−166. (18) McKinnon, K. M. Multiparameter Conventional Flow Cytometry. In Flow Cytometry Protocols, 4th ed.; Hawley, T. S.; Hawley, R. G., Eds.; 2018; pp 139−150. G

DOI: 10.1021/acs.analchem.9b01419 Anal. Chem. XXXX, XXX, XXX−XXX