Whole Cell MALDI Fingerprinting Is a Robust Tool for Differential Profiling of Two-Component Mammalian Cell Mixtures
- 507 Downloads
MALDI fingerprinting was first described two decades ago as a technique to identify microbial cell lines. Microbial fingerprinting has since evolved into an automated platform for microorganism identification and classification, which is now routinely used in clinical and environmental sectors. The extension of fingerprinting to mammalian cells has yet to progress partly due to compartmentalization of eukaryotic cells and overall higher cellular complexity. A number of publications on mammalian whole cell fingerprinting suggest that the method could be useful for classification of different cell types, cell states, and monitoring cell differentiation. We report the optimization of MALDI fingerprinting workflow parameters for mammalian cells and its application for differential profiling of mammalian cell lines and two-component cell line mixtures. Murine fallopian tube cells and high-grade ovarian carcinoma cell lines and their mixtures are used as model mammalian cell lines. Two-component cell mixtures serve to determine the method’s feasibility for complex biological samples as the ability to detect cancer cells in a mixed cell population. The level of detection of cancer cells in the two-component mixture by principle component analysis (PCA) starts to deteriorate at 5% but with application of a different statistical approach, Wilcoxon rank sum test, the level of detection was determined to be 1%. The ability to differentiate heterogeneous cell mixtures will help further extend whole cell MALDI fingerprinting to complex biological systems.
KeywordsWhole cell MALDI fingerprinting Mammalian cell lines Two-component cell line populations
Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI -TOF MS) is a powerful analytical technique widely used for analysis of proteins in a broad range of biological samples, which vary from lysed cells extracts [1, 2, 3, 4, 5, 6, 7, 8, 9, 10] to tissue sections [11, 12, 13, 14, 15, 16, 17, 18]. Whole cell MALDI fingerprinting is based on the mass spectral analysis of a whole cell without additional preparatory steps, such as fractionation or extraction. Minimal sample processing circumvents complex and time-consuming experimental steps necessary for cell proteomic analyses. MALDI protein spectral profiles generated during analyses are species specific and serve as unique cellular ID fingerprints [19, 20, 21]. Speed, simplicity, sensitivity, and specificity for large biomolecules in complex matrices confer whole cell MALDI fingerprinting as a robust, facile, and inexpensive technique. The most advanced application—identification and classification of microorganisms—is an automated platform, which is routinely used in clinical [22, 23, 24, 25, 26], biodefense , and environmental [28, 29, 30] laboratories.
Mammalian fingerprinting has yet to be broadly deployed in clinical applications in part because of the overall complexity of eukaryotic cells and a lack of standardized workflows for mammalian fingerprinting. Mammalian cell compartmentalization results in numerous regulatory pathways for cellular networking and multiple cell cycle states, which affect the resulting mammalian cellular fingerprint. The lack of standardized workflows in mammalian fingerprinting was recently addressed by Munteanu et al. . The review comprehensively covers all aspects of the fingerprinting workflow necessary for method standardization, such as sample preparation (type of matrix, matrix/solvent composition/additives, sample application and cell density), instrument parameters (type of MS instrument, laser wavelength/number of shots, acquisition parameters), and statistical assessment models [principal component analysis (PCA), partial least square analysis (PLSA), hierarchical clustering analysis (HCA), Pearson coefficient]. For example, it is known that in bacterial cells, different matrices and crystallization methods produce different MALDI spectra of the same sample [32, 33], and given the complexity of eukaryotic cells, there is a high need for rigorous standardization of sample preparation/instrument parameters. Also noted is a lack of extensive databases for mammalian cellular fingerprints that would require universal automated workflows and classification algorithms.
Despite the obvious challenges of mammalian fingerprinting, numerous reports on mammalian cells characterization have been published. Zhang et al. demonstrated the ability to differentiate three mammalian cell lines after cell lines were lysed in 2,5-dihydroxybenzoic acid matrix solution . Another study identified 66 cell lines representing 34 species from insects to primates based on MALDI analysis of formic acid/acetonitrile extractions of cultured cells . More advanced applications were aimed at distinguishing different cell types originating from the same cell lineage: identification of two different pancreatic hormone-secreting cell lines , the comparison of primary human blood cells and blood cell lines [37, 38], molecular phenotyping of central nervous system (CNS) glial cells (astroglial, microglial, and oligodendroglial) , and MALDI-MS fingerprinting of different melanoma cell lines . Further applications of mammalian fingerprinting have focused on physiological changes of a single cell, reflecting its specific cell states or cell transformations, such as differentiation of human colon carcinoma  or leukemia  cell lines, multifaceted activation of human macrophages , identification of resting and activated human monocyte subsets , rapid detection of apoptosis/necrosis signature , and monitoring of histone deacetylase drug target engagement . Regardless of the scope of the aforementioned studies, no consistency in method parameters was observed (such as matrix, cell density, cell media, sample application technique, and laser frequency/number of shots) for either cell authentication [35, 36, 37, 38, 39, 40] or close monitoring of a single cell change application [41, 42, 43, 44, 45].
Based on literature, mammalian fingerprinting has the potential to discriminate between different cell types and cell states, but there is a scarcity of publications on application of this technique towards heterogeneous samples. Heterogeneity is a hallmark of cancer, and there is unmet medical need for sensitive detection of cancerous cells in a complex environment of biological samples or detection of relevant biomarkers using emerging liquid biopsy techniques. As a cancer diagnostic tool, MALDI fingerprinting has been successfully applied towards clinical fine-needle aspirates of lung cancer cells  and oral mucosa brush biopsies  to obtain cancer cell specific protein profiles, which differentiate tumor samples and non-tumor controls. Unfortunately, the sample heterogeneity or the percentage of cancer cells detected was not reported, limiting the utility of these reported methods. We used a reduced model system for workflow optimization consisting of two-component cell line mixtures with known concentrations of cancer cells. Method parameters were optimized for whole cell MALDI fingerprinting workflow and validated using defined cell line mixtures. Optimized method parameters allowed for the discrimination between non-cancerous and cancer mammalian cell lines as well as between two-component cell line mixtures with the minimum threshold for cancer cells to be 1% in an otherwise non-cancerous “healthy” cellular background.
Acetonitrile (ACN, HPLC grade) and trifluoroacetic acid (TFA, LC-MS grade) was purchased from Sigma-Aldrich (St. Louis, MO, USA). Milli-Q water (ddH2O; Millipore) was prepared in-house. Sinapinic acid (SA) (matrix substance for MALDI-MS, ≥ 99%) was obtained from Sigma-Aldrich (St. Louis, MO, USA). α-Cyano-4-hydroxycinnamic acid (CHCA), 2,5-dihydroxybenzoic acid (DHB), and protein calibration standards I and II were purchased from Bruker Daltonics (Bremen, Germany).
The human ovarian cancer cell line (OVCAR3) was purchased from the American Type Culture Collection. OVCAR8 cells expressing red fluorescent protein (OVCAR8-RFP) were a gift from Sharon Stack at the University of Notre Dame. OVCAR4 and OVCAR8 were obtained from the NCI 60 Cell Panel Cell Bank Repository. OVCAR4-RFP was generated with lentiviral transduction according to the manufacturer’s protocol (GenTarget #LVP023). OVCAR3 was grown in RPMI 1640 supplemented with 20% fetal bovine serum (FBS), 1% penicillin/streptomycin (P/S), and 10 μg/mL insulin. OVCAR8 and OVCAR8-RFP were grown in DMEM with 10% FBS and 1% P/S. OVCAR4-RFP was grown in RPMI 1640 supplemented with 10% FBS, 1% P/S, and 1% l-glutamine. Our non-cancerous cell line was murine oviductal cells (MOE) obtained from Dr. Barbara Vanderhyden at the University of Ottawa and were maintained in α-modified Eagle’s medium supplemented with 10% FBS, 1% P/S, 1% l-glutamine, 2 μg/ml epithelial growth factor, 5 μg/ml insulin, 5 μg/ml transferrin, 5 ng/ml sodium selenite, 1 mg/ml gentamycin, and 20 ng/ml β-estradiol. Cultured cells were maintained in a humidified incubator at 37 °C in 5% CO2. Cells were passaged a maximum of 30 times. Cell lines were validated by short tandem repeat analysis and tested mycoplasma-free in 2017.
Primary Human Cells
Human fallopian tube specimens were obtained from benign gynecologic surgeries performed at the University of Illinois Hospital after consent and de-identification. Fallopian tube epithelium was mechanically isolated and incubated in MOE media with 1 mg/mL collagenase and 0.05% trypsin for 20 min at 37 °C with occasional vortexing. Dissociated cells were allowed to attach and grow in culture plates for 5 days before FBS concentration was reduced to 1% to enrich for epithelial cells. Cells were passaged a maximum for five times thereafter. Tissue collection was approved by the UIC Institutional Review Board (IRB #2012-0539) and performed in accordance with NIH guidelines on human subject research.
Aqueous Cell Suspensions
Cells were collected by trypsinization and washed with PBS. Cell densities were measured via hemocytometer and cells were re-suspended in ddH2O to obtain a cell concentration of 10,000 cells/μL and stored as frozen suspensions at − 20 °C until use.
Live Cells in PBS
For live cell applications, live cells were collected by trypsinization and washed with PBS. Cell densities were measured via hemocytometer after which cells were re-suspended in PBS to obtain a cell concentration of 10,000 cells/μL, followed by storage on ice prior to sample preparation.
Sample Preparation for MALDI-MS Analysis
Matrices containing 10 mg/mL of SA (SA10), CHCA (CHCA10), or DHB (DHB10) were prepared in the solution of ACN/water = 70/30 containing 0.1% TFA. Matrix solution of SA with a higher aqueous composition (SA20) was prepared by dissolving 20 mg of SA in 1 mL of ACN/water/TFA = 30/70 containing 0.1% TFA.
Dried Droplet and Double Layer Sample Applications from Aqueous Cell Suspensions
Frozen cell suspensions were thawed on ice and diluted with ddH2O to desired concentrations. For dried droplet application, equal volume amounts of a cell suspension and matrix solution were mixed together and left on ice for 10 min. To avoid the problem of non-specific interactions of cellular proteins with the pipette surface, each pipette tip was saturated with matrix-analyte mixture (by pipetting the mixture three times) prior to application. Then 1–1.5 μL of the resulting mixture was spotted on 384-well ground steel target (Bruker Daltonics) in n technical replicates (n = 3, 4, 16, or 24) and air dried. For double layer application, first, 1–1.5 μL of SA in acetonitrile (1 mg/mL of ACN) was applied to 384-well ground steel plate and air dried. Then, it was followed by dried droplet application of matrix-analyte mixture as described above.
Whole Cell Applications of Live Cells from PBS Media
Live cells in PBS media were diluted with PBS to desired concentrations. The samples were kept on ice in between dilution steps and sample preparation. Before sample application, cell mixture was gently re-suspended by light agitation. Each pipette tip was saturated with cell mixture (by gentle pipetting the mixture three times) prior to application. For sample preparation, 1.5 μL of cell mixture was applied on 384-well ground steel plate in triplicates and dried at room temperature. Then, 1.5 μL of SA20 matrix (20 mg/mL of ACN/water = 30/70 + 0.1%TFA) was applied over the dried cell culture and air dried.
Dried Droplet Applications of Live Cells from PBS Media
Live cells in PBS media were diluted with PBS to desired concentrations. The samples were kept on ice in between dilution steps and sample preparation. Equal volume amounts of a cell suspension and matrix solution were mixed together and left on ice for 10 min. To avoid the problem of non-specific interactions of cellular proteins with the pipette surface, each pipette tip was saturated with matrix-analyte mixture (by pipetting the mixture three times) prior to application. Then, 1.5 μL of the resulting mixture was spotted on 384-well ground steel target (Bruker Daltonics) in triplicate and air dried. Dried droplet applications that were washed with water to remove PBS salts were prepared following all the steps described above. Then, 1.5 μL of ice-cold water was gently applied to the top of a dry crystalline preparation and briefly removed with Kimwipe trying not to disturb the sample integrity. After washing, step crystalline samples were dried at room temperature.
Mixed Cell Populations from Aqueous Cell Suspensions
Frozen MOE and OVCAR8 cell suspensions at 1 × 104 cells/μL were thawed on ice and diluted with cold ddH2O to 2500 cells/μL. Then, MOE and OVCAR8 suspensions were mixed to obtain a range of two-component cell line populations containing 50 to 10% (with 10% increment) or 9 to 1% (with 1% increment) of OVCAR8 cancer cells in the mixture. All cell suspensions were kept on ice during dilution and mixing steps. For dried droplet application, equal volume amounts of a cell suspension and SA20 matrix solution were mixed together and left on ice for 10 min. To avoid the problem of non-specific interactions of cellular proteins with the pipette surface, each pipette tip was saturated with matrix-analyte mixture (by pipetting the mixture three times) prior to application. Then, 1.5 μL of the resulting mixture was spotted on 384-well ground steel target (Bruker Daltonics) in 16 or 24 replicates per population and air dried.
MALDI-MS and Data Analysis
All mass spectra were acquired on an Autoflex Speed LRF MALDI-TOF mass spectrometer (Bruker Daltonics) equipped with Smartbeam™-II laser (355 nm) in positive linear mode within the mass range of m/z 5000 to 20,000 Da with a low mass gate at 3300 Da. Protein spectra can be acquired manually or automatically using the AutoXecute function of the FlexControl 3.4 software with 2-kHz Smartbeam laser, acceleration voltage setting at 19.5 kV, and delayed extraction at 170 ns. Four thousand laser shots were accumulated from 20 different positions of the MALDI target chosen by random walk for manual acquisition or by “spiral_large” measuring raster for automated runs. All spectra were externally calibrated from calibration spots adjacent to the sample spots and processed by FlexAnalysis 4.2 (Bruker Daltonics) software using Top Hat algorithm for baseline subtraction. PCA analysis was performed using ClinProTools 3.0 software. The following CPT processing parameters (adapted from ) were used: resolution, 800; top hat baseline subtraction with 10% minimal baseline width; mass range, 5000–20,000 Da. All spectra were automatically normalized by their total ion count by CPT software. Mass spectra were further smoothed using the Savitsky-Golay algorithms (5 cycles with m/z = 6 Da). Null spectra and noise spectra exclusion with a noise threshold of 1.00 were enabled. Spectra were recalibrated, allowing a maximal shift tolerance of 1000 ppm match on 25% of the peaks. Peak picking was performed on average spectra using S/N > 5, and peak calculation was performed using peak intensities.
For statistical evaluation profile, mass spectra were converted to mzML format.
To detect differentially expressed features between cell lines, mass spectra in the mzML format were pre-processed using the MALDIQuant software  following the workflow suggested in the software documentation (https://cran.r-project.org/web/packages/MALDIquant/MALDIquant.pdf). Protein features across all samples were assembled into a matrix and queried for features, which were differentially expressed in the mixed cell samples versus the non-cancerous cell line (MOE). Utilizing the abundances for each feature, a Wilcoxson rank-sum test  was applied to assess whether the abundance differences between the mixture and controls were statistically significant. The Wilcoxson rank-sum test was chosen because it makes very few assumptions about the distribution of the data, e.g., normality in the case of t tests. Bonferroni-correction  was applied to the p value of each test to account for multiple hypothesis testing and a threshold of 0.01 was applied on the corrected p value for each test; other methods were also considered and the results are reported in Figure S21. For each feature, we determined its minimum level of detection L, such that all mixed samples with cancer cell population ≥ L% show statistically different abundance for that feature. For example, if a feature is deemed as differentially expressed in sample with 2 and 5–50% of cancer cells samples, its limit of detection was determined at 5%.
Results and Discussion
Optimization of Matrix/Solvent Composition and Cell Density Using Aqueous Cell Suspensions
The basis of whole cell MALDI fingerprinting can be defined as the cellular protein profile with the best ionization efficiency (highest number of characteristic and/or intense features). Sample ionization is dependent on many factors and therefore required optimization to identify the combination of matrix/solvent composition and plating cell density for generating unique cellular fingerprints. Two to twenty kilodaltons has historically been popular for mammalian fingerprinting and SA is widely used [38, 39, 40, 41, 43, 46, 51, 52, 53, 54], whereas consensus on the matrix/solvent formulation and sample application technique has varied across the literature. For the initial round of optimization, we chose three different matrices (CHCA, DHB, and SA) for simultaneous optimization of matrix/solvent composition and plating cell density using dry droplet sample application. For side to side comparison, all three matrices were prepared at 10 mg/mL in 70% ACN(aq) (CHCA10, DHB10, and SA10). For the matrix with higher aqueous composition, we used the combination of 20 mg/mL of SA in 30% of ACN(aq) (SA20). The matrix solvent acidity was kept low (0.1% of TFA), since higher TFA concentrations were reported to produce fewer characteristic signals .
Optimization of Mass Detection Range Using Aqueous Cell Suspensions
Sample Application and Cell Media
Previous reports [52, 53] using SA matrix for dried droplet preparation relied on samples with high cellular densities (10,000 cells/spot). In our case, higher cell densities (10,000 or 5000 cells/μL) yielded modest improvement for intensities/additional peaks in the 6000–12,000 region for whole cell and dried droplet applications from PBS media for OVCAR3 cells (Figure S11A and S11B), but suppression of protein signals above 14 kDa occurred. Aqueous cell suspensions provided the most efficient and uniform protein ionization coupled with the dried droplet application. All further sample preparations utilized this combination.
Optimization of Plating Volume
For plating volume optimization, the statistical distribution of different cell densities (5000, 2500, 1250, 625, and 312 cell/μL) at 1- and 1.5 μL plating volumes was compared. As seen from PCA plots and ClinProTool Peak statistics data for MOE cells (Figure S12, Tables S1 and S2), 1.5 μL volume resulted in more ordered distribution and tighter clustering of different MOE cell density data, which could be the result of overall improved values for statistical parameters: p values, standard deviations, and coefficients of variation for peak intensities. Similar results were obtained for OVCAR3 cells as well (data not shown).
Comparison of Cancer and Normal Mammalian Cells Lines Fingerprints
Mixed Two-Component Cell Line Populations
Biological and Technical Reproducibility
Stability of a cellular fingerprint is defined by its biological and technical reproducibility over a period of time. During the course of method optimization and experiments with mixed cell populations, we acquired multiple protein spectra of cells harvested at different dates or spectra were repeatedly recorded on different dates from the same frozen stocks. These data became a small in-house data set and spectra acquired on different days from different batches were compared for assessment of biological and technical reproducibility of cellular fingerprints. Both MOE (Figure S16a and S16b) and OVCAR3 (Figure S17a and S17b) cells offer good biological reproducibility for three biological replicates. While some variations for peak patterns and intensities are observed in individual spectra, the unique signatures of cellular fingerprints are recognizable from peak values and peak patterns. Technical reproducibility is exemplified by four replicate spectra acquired on different days from the same frozen stocks of OVCAR3 (Figure S18a and S18b) and MOE (Figure S19a and S19b) cells. It also should be noted that cell growth media does not influence protein cellular fingerprints (Figure S20a–c).
Mammalian whole cell MALDI fingerprinting is a developing area of mass spectrometry but still requires attention in the standardization of workflows and creation of universal databases to become a valuable tool for industrial and clinical applications. The goal of this study was to approach workflow standardization based on the whole cell MALDI fingerprinting techniques published in the literature and identification of workflow parameters that provide the most robust mammalian cellular fingerprint. We report the optimization of sample preparation parameters, such as cell media for cell lines under study, sample application technique, matrix/solvent composition, matrix/analyte density, and plating volume. Additionally, our work supports the use of the medium protein range (5–20 kDa) for providing robust protein spectra for mammalian fingerprinting. In our experiments on the AutoflexSpeed LRF model (Bruker), we found that 500 Hz laser frequency with 3-medium or 4-large SmartBeam II parameter settings resulted in efficient protein ionization with appropriate laser power and detector gain adjustments and those settings can be used for both manual and automated acquisitions. We also demonstrated the extension of mammalian fingerprints to heterogeneous cell mixtures with the ability to statistically discriminate 1% of cancer cells in an otherwise non-cancerous “healthy” cell background. We aim to further expand the utility of this method to detection of aberrant cells in host organisms in the future.
The project described was supported by the National Center for Advancing Translational Sciences, National Institutes of Health, through Grant UL1TR002003 (LMS and JEB) and by Grant Number K12HD055892 from the National Institute of Child Health and Human Development (NICHD) and the National Institutes of Health Office of Research on Women’s Health (ORWH) (LMS), University of Illinois at Chicago Startup Funds (LMS), UG3 ES029073 (JEB), and F30 CA217079 (ANY).
We would like to thank Dr. Pavel Petukhov for help with mzML file conversion and Laura Rodgers Hardy for the creation of the OVCAR4-RFP cell line.
- 1.Wang, H., Kachman, M.T., Schwartz, D.R., Cho, K.R., Lubman, D.M.: Comprehensive proteome analysis of ovarian cancers using liquid phase separation, mass mapping and tandem mass spectrometry: a strategy for identification of candidate cancer biomarkers. Proteomics. 4, 2476–2495 (2004)CrossRefPubMedGoogle Scholar
- 7.Alcolea, P.J., Alonso, A., Garcia-Tabares, F., Mena, M.D.C., Ciordia, S., Larraga, V.: Proteome profiling of the growth phases of Leishmania pifanoi promastigotes in axenic culture reveals differential abundance of immunostimulatory proteins. Acta Trop. 158, 240–247 (2016)CrossRefPubMedGoogle Scholar
- 8.Anders, U., Schaefer, J.V., Hibti, F.E., Frydman, C., Suckau, D., Pluckthun, A., et al.: SPRi-MALDI MS: characterization and identification of a kinase from cell lysate by specific interaction with different designed ankyrin repeat proteins. Anal. Bioanal. Chem. 409, 1827–1836 (2017)CrossRefPubMedGoogle Scholar
- 17.Alberts, D., Pottier, C., Smargiasso, N., Baiwir, D., Mazzucchelli, G., Delvenne, P., et al.: MALDI imaging-guided microproteomic analyses of heterogeneous breast tumors-a pilot study. Proteomics Clin Appl. 12, (2018)Google Scholar
- 18.Klein, O., Hanke, T., Nebrich, G., Yan, J., Schubert, B., Giavalisco, P., et al.: Imaging mass spectrometry for characterization of atrial fibrillation subtypes. Proteomics Clin Appl. e1700155 (2018)Google Scholar
- 24.Donohue, M.J., Smallwood, A.W., Pfaller, S., Rodgers, M., Shoemaker, J.A.: The development of a matrix-assisted laser desorption/ionization mass spectrometry-based method for the protein fingerprinting and identification of Aeromonas species using whole cells. J. Microbiol. Methods. 65, 380–389 (2006)CrossRefPubMedGoogle Scholar
- 32.Vaidyanathan, S., Winder, C.L., Wade, S.C., Kell, D.B., Goodacre, R.: Sample preparation in matrix-assisted laser desorption/ionization mass spectrometry of whole bacterial cells and the detection of high mass (> 20 kDa) proteins. Rapid Commun. Mass Spectrom. 16, 1276–1286 (2002)CrossRefPubMedGoogle Scholar
- 41.Marvin-Guy, L.F., Duncan, P., Wagniere, S., Antille, N., Porta, N., Affolter, M., et al.: Rapid identification of differentiation markers from whole epithelial cells by matrix-assisted laser desorption/ionisation time-of-flight mass spectrometry and statistical analysis. Rapid Commun. Mass Spectrom. 22, 1099–1108 (2008)CrossRefPubMedGoogle Scholar
- 45.Munteanu, B., Meyer, B., von Reitzenstein, C., Burgermeister, E., Bog, S., Pahl, A., et al.: Label-free in situ monitoring of histone deacetylase drug target engagement by matrix-assisted laser desorption ionization-mass spectrometry biotyping and imaging. Anal. Chem. 86, 4642–4647 (2014)CrossRefPubMedGoogle Scholar
- 46.Amann, J.M., Chaurand, P., Gonzalez, A., Mobley, J.A., Massion, P.P., Carbone, D.P., et al.: Selective profiling of proteins in lung cancer cells from fine-needle aspirates by matrix-assisted laser desorption ionization time-of-flight mass spectrometry. Clin. Cancer Res. 12, 5142–5150 (2006)CrossRefPubMedGoogle Scholar
- 50.Dunn, O.J.: Multiple comparisons among means. Journal of American Statistical Association. 52–64 (1961)Google Scholar
- 51.Valletta, E., Kucera, L., Prokes, L., Amato, F., Pivetta, T., Hampl, A., et al.: Multivariate calibration approach for quantitative determination of cell-line cross contamination by intact cell mass spectrometry and artificial neural networks. PLoS One. 11, e0147414 (2016)CrossRefPubMedPubMedCentralGoogle Scholar
- 54.Povey, J.F., O'Malley, C.J., Root, T., Martin, E.B., Montague, G.A., Feary, M., et al.: Rapid high-throughput characterisation, classification and selection of recombinant mammalian cell line phenotypes using intact cell MALDI-ToF mass spectrometry fingerprinting and PLS-DA modelling. J. Biotechnol. 184, 84–93 (2014)CrossRefPubMedGoogle Scholar