Article
Reference
Drug-induced expression of the RNA-binding protein HuR attenuates the adaptive response to BRAF inhibition in melanoma
MERAT, Rastine, et al.
Abstract
Strategies that aim to limit the adaptive response to pathway inhibition in BRAF-mutated melanoma face the inherent limit of signaling redundancy and multiplicity of possible bypass mechanisms. Drug-induced expression of selected RNA-binding proteins, like the ubiquitously expressed HuR, has the potential to differentially stabilize the expression of many genes involved in the compensatory mechanisms of adaptive response. Here, we detect in BRAF-mutated melanoma cell lines having a higher propensity for adaptive response and in non-responding melanoma tumors, a larger proportion of HuRLow cells in the expression distribution of HuR. Using knockdown experiments, we demonstrate, through expression profiling and phenotypic assays, that increasing the proportion of HuRLow cells favors the adaptive response to BRAF inhibition, provided that the HuRLow state stays reversible. The MAPK dependency of melanoma cells appears to be diminished as the proportion of HuRLow cells increases. In single-cell assays, we demonstrate that the HuRLow cells display plasticity in their growth expression profile. Importantly, the adaptive [...]
MERAT, Rastine, et al . Drug-induced expression of the RNA-binding protein HuR attenuates the adaptive response to BRAF inhibition in melanoma. Biochemical and Biophysical Research Communications , 2019, vol. 517, no. 2, p. 181-187
PMID : 31279529
DOI : 10.1016/j.bbrc.2019.06.154
Available at:
http://archive-ouverte.unige.ch/unige:126980
Disclaimer: layout of this document may differ from the published version.
1 / 1
1 2. Materials and methods
2.1 Cell lines and drugs
All human melanoma cell lines other than the A375 cells were specifically purchased for this project, provided with a certificate of authenticity and passaged less than six months at the time of use in the experiments. The A375 cell line was purchased earlier from CLS Cell Lines Service GmbH and authentified, before being used for this project, by an independent laboratory (DSMZ, Leibniz-Institute, Germany) with DNA profiling using eight different and highly polymorphic STR loci. SK-MEL5 (HTB-70), SK-MEL28 (HTB-72), Malme-3M (HTB-64) cell lines were purchased from ATCC. The UACC-62 cell line (C0020003) was purchased from AddexBio, CA, USA. The WM-88 cell line (01-0001) was purchased from Rockland antibodies
& assays, PA, USA. The A375, SK-MEL5, SK-MEL28 cells were grown in DMEM media supplemented with 10% FBS, 2 mmol/L glutamine and 1% penicillin/streptomycin. Malme-3M cells were grown in IMDM media supplemented with 20% FBS. UACC-62 cells were grown in RPMI media supplemented with 10% FBS. WM- 88 cells were grown in RPMI media supplemented with 5% FBS. The BRAF inhibitor vemurafenib (PLX4032, RG7204) was purchased from Selleckchem and dissolved at 10 mmol/L in dimethylsulphoxide (DMSO, storing concentration, renewed every six months). Lithium carbonate (Li2CO3) was purchased from Sigma-Aldrich and stored in distilled water at 180 mM.
2.2 Flow cytometry, cell cycle, cell viability and apoptosis assays
For all three types of analysis, cells were seeded at 2 x105 per well (A375, UACC-62) or 3 x105 per well (SK-MEL5, SK-MEL28, Malme-3M, WM-88) in six well plates three days before being harvested and maintained in their respective medium containing (for adapted cells) or not (for parental cells) the final concentration of vemurafenib used in the adaptive regimen.
For flow cytometry analysis, following trypsinization, cells were counted, washed (PBS), fixed in 1.5%
formaldehyde solution (10 min at room temperature) and permeabilized with 100% ice-cold methanol (10 min).
For experiments comparing different time-points, cells were stored in methanol (-20°C) for later concomitant staining. Stainings were performed on 5.x105 cells. After being washed (x2, PBS. 1% BSA), cells were re- suspended in 100 μL PBS 1% BSA and incubated (30 min) with the FITC-conjugated anti-HuR mouse monoclonal 3A2 antibody (sc-5261, Santa Cruz, 1:20) or a similar concentration of FITC-conjugated isotype control (sc-2339, Santa Cruz), washed (x3, PBS) and re-suspended in 200 μL of PBS 1% BSA. FACS analysis was conducted on at least 10 000 events in each sample. For HuR expression distribution comparison of parental with adapted cells, cells were exposed to vemurafenib (1μM) during six hours before being harvested.
For cell cycle analysis, a PI/RNase Staining Buffer (BD Pharmigen) was used according to the manufacturer’s instructions. Cell viability and apoptosis assays were performed on cells harvested using Versene and stained using the propodium Iodide / Annexin V Apoptosis Detection Kit APC (Affymetrix eBioscience) according to the manufacturer’s instructions with a slight modification of the Annexin V binding buffer to which 2mM final concentration of CaCl2 and MgCl2 were added. For both the cell cycle and cell viability / apoptosis assays, at least 20 000 events were analysed in each sample.
All flow cytometry, cell cycle, cell viability and apoptosis assays were performed on a BD Accurri 2 flow cytometer and analysed in the FlowJo software. Cell cycle files were analysed using the Watson (pragmatic) algorithm.
2.3 siRNA short-term and shRNA long-term HuR knockdowns
For short-term knockdown experiments, Malme-3M cells were seeded at 2.2x105 per well in 12 well plates in medium free of antibiotics. The following day, transfection was performed using the siRNA transfection reagent (Santa Cruz) according to the manufacture’s instructions. For HuR knockdown, in order to minimize off-targets effects, we used, as described by others [1], a mixture of Flexitube (Qiagen) functionally verified siRNAs (SI03246887, SI03246551, SI00300139) combined with an additional siRNA from Santa Cruz (sc-35619). Each well was transfected with the equivalent amount of 20 pmol of the mixture (five pmol of each siRNA duplex).
Similar amounts of a scrambled nonspecific 19-25 (fluorescein-conjugated) siRNA was used as a control (sc- 36869, Santa Cruz).
For long-term knockdown experiments, cells were seeded in 1 mL medium free of antibiotics w/o FBS in 12 well plates (SK-MEL28 cells at 2.2x105 per well, A375 cells at 1.5x105 per well). The next day, cells were infected o/n in 1 mL polybrene containing medium (5 μg/mL) with 105 IFU of either HuR shRNA (sc-35619-V) or control shRNA (sc-108080) lentiviral particles (Santa Cruz). According to the manufacturer’s instructions,
2
after three days, stable clones were selected via puromycin dihydrochloride treatment (1 µg/mL) during three weeks. Resistant colonies were further amplified and screened for HuR expression by flow cytometry analysis.
2.4 Western blot analysis
For the initial HuR expression analysis in various melanoma cell lines, cell lysates were prepared using RIPA buffer (PBS, 1% IGEPAL®, 01% sodiumdeoxycholate, 0.1% SDS) supplemented with fresh protease and phosphatase inhibitors (easy packs, Roche). Cell lysates were quantified for protein content using Bio-Rad DC kit. Protein samples were resolved on 4-12% polyacrylamide Bis-Tris gel (NuPAGE, ThermoFisher-Scientific) and transferred to nitrocellulose membrane using an iBlot® Dry Blotting system (invitrogen, life technologies) according to the manufacturer’s instructions. After saturation (PBS, 0.1% Tween, 5% BSA), the membranes were incubated for 1 h with the anti-HuR mouse monoclonal 3A2 antibody (sc-5261, Santa Cruz, 1:1000). An anti-HSP90 mouse monoclonal antibody was used as a loading control (sc-13119, Santa Cruz, 1:1000). The membranes were incubated for 30 min with a secondary goat anti-mouse IgG conjugated to horseradish peroxidase (Bio-Rad, catalogue n°170-6516, 1:3000), washed and developed using the ECL system (ThermoFisher-Scientific) according to manufacturer’s instructions.
For the initial Western-blot analysis of the comparative response to BRAF inhibition of different cell lines in parental and adapted cells (Fig.1), with no any experimental modulation of the adaptive response, cells were washed with serum-free medium and synchronized o/n in their corresponding medium containing 0.1% serum with no addition of vemurafenib. For Western-blot analysis of experiments in which the adaptive response was potentially modulated (HuR knockdown and Li2CO3 HuR expression induction, Fig.2 and 3), cells were similarly synchronized in their corresponding medium containing ± vemurafenib at the final concentration used in the adaptive regimen. Similarly, Li2CO3 (1mM) was ± added in Li2CO3 experiments. This conditioning was justified to avoid any additional perturbation of the response to BRAF inhibition and an uncontrolled synchronization, occurring only in the adapted/addicted cells, induced by vemurafenib withdrawal. The next morning cells were treated with vemurafenib or DMSO at the indicated concentration. After six hours, cells were washed with cold PBS containing NaF (50 mM) and Na3VO4 (10mM). Lysates were collected and analysed as described above using the primary antibodies: SOX10 (rabbit, EPR4007, abcam, catalogue n°155279, 1:1000), alpha Tubulin (mouse, DM1A, abcam 1:5000). Following antibodies were all purchased from Cell Signaling and used at 1:1000: EGFR (raised in rabbit, clone D38B1, catalogue n°4267), MET (rabbit, D1C2, 8198), BRAF (rabbit, D9T6S,14814), CRAF (rabbit, D4B3J, 53745), MEK1/2 (mouse, L38C12, 4694), phospho-Ser217/221-MEK1/2 (rabbit, 41G9, 9154), ERK (rabbit, polyclonal, 9102), phospho-Thr202/Tyr204- ERK (rabbit, 20G11, 4376), phospho-Ser235/236-S6 ribosomal protein (rabbit, polyclonal, 2211), AKT (rabbit, polyclonal, 9272), phospho-Ser473-AKT (rabbit, D9E, 4060), phospho-Ser240/244-S6 ribosomal protein (rabbit, D68F8, 5364), 4EBP1 (rabbit, 53H11, 9644), DUSP4 (rabbit, D9A5, 5149). For rabbit primary antibodies, a goat anti-rabbit IgG-HRP was used as the secondary antibody (Bio-Rad, catalogue n°170-6515, 1:3000).
2.5 Short-term proliferation and long-term clonogenic assays
Short-term proliferation assays were conducted using WST-1 reagent (Roche applied Science). Parental and adapted A375 cells were plated at 2000 per well, SK-MEL28, Malme-3M and WM-88 cells at 5000 per well, SK-MEL5 cells at 10000 per well, UACC-62 cells at 4000 per well in 96-well plates in 100 µL of medium.
After 24h, cells were treated with vemurafenib or DMSO at the indicated concentrations in quadruplicate. After a treatment period of 72h, WST reagent (20 µL, 10% of the final volume) was added to the wells and incubated at 37°C for 1-2h so as to stay within the linear range of the assay. The plates were read at 450(-650) nm on a V max kinetic ELISA Microplate Reader (Molecular devices). Cell proliferation was expressed, after background subtraction, as the percentage of absorbance compared with the excipient-treated cells.
For clonogenic assays, cells were plated at 1x103 per well in six-well plate. For Li2CO3 experiments, Li2CO3
was added or not to cells when being seeded at the indicated concentrations. After 24h, cells were treated with vemurafenib at the indicated concentrations or DMSO in duplicates. The medium with corresponding treatments was changed twice a week. After 25 days for SK-MEL28 cells or 11 days for A375 cells, cells were washed with PBS, fixed with 4% PFA for 10 min and stained with 0.5% (w/v) crystal violet in 70% ethanol for 15 min.
The wells were washed with distilled water, dried and photographed with a macro lens.
2.6 Immunocytochemistry (ICC)-based single-cell automated quantification of HuR expression
A375, A375 sh Ctrl, A375 sh HuR, SK-MEL28 sh Ctrl, SK-MEL28 sh HuR cells were plated in triplicates at 5000 per well (so as to be sparse enough at staining after 48h) in 96-well flat clear bottom black polystyrene plates (Corning n°3603) and treated with 0, 0.5 and 1mM of Li2CO3 (Sigma-Aldrich). After 48h, cells were fixed in 4% paraformaldehyde for 15 min at room temperature, washed with PBS, permeabilized with PBS-
3
containing 0.5% Triton X-100 for 15 min and incubated with a blocking buffer solution (PBS containing 5%
normal donkey serum and 0.1% Triton X-100) for 1h at room temperature. Cells were subsequently incubated with the HuR mouse monoclonal 3A2 antibody (sc-5261, Santa Cruz, 1:200) during 1h, washed, and incubated with a Alexa Fluor488-conjugated donkey anti-mouse secondary antibody (Bio-Rad, catalogue n°A-21202, 1:500) for 1h at room temperature in the dark before being stained for 5 min with DAPI. Image acquisition was conducted using an ImageXpress XL scanner. Automated image quantification was further conducted using MetaXpress 5.1.0.41. The Multi Wavelength Cell Scoring setting was used to measure the HuR nuclear intensity in the DAPI staining area and the HuR cytoplasmic intensity in an automated-defined area around the nucleus having an intensity of at least 80 graylevels above the local background. The integrated fluorescent signal was obtained from at least 500 cells in each condition. Statistical analysis was conducted in MatlabR 2018.b. A two- sample Kolmogorov-Smirnov test with unequal tails was used to confirm the significance of the Li2CO3 induced shift. Nuclear variation was the main contributor (≥ 70% relative contribution) to the total (nuclear + cytoplasmic) average intensity variation at the most efficient shift-inducing concentrations of Li2CO3 for each cell type (1mM for A375 cells, 0.5 mM for A375 sh Ctrl cells, see Fig.3A). Nevertheless, the total variation was reported in order to take into account any type of detectable increase in HuR expression.
2.7 Immunohistochemistry (IHC)-based single-cell automated quantification of HuR nuclear expression The study was approved by the Geneva ethics committee (study n°2017-01346). According to the Swiss Federal Law for research, a positive vote of an ethical committee in a retrospective study is sufficient for using patient data for research purposes without further need of individual informed consents. All patient-related data and samples were identified from the Division of medical oncology, the dermato-oncology unit and the Department of pathology at the University Hospital of Geneva and selected on the availability of the samples.
Standard fluorescence-based immunohistochemical staining on deparaffinised tumor sections was performed as previously described [2]. HuR staining was performed using the same mouse monoclonal anti-human HuR antibody 3A2 used throughout this study (1:100) and a secondary fluorescent Alexa 488-conjugated anti-mouse antibody described in the previous section, (1:500). S100 staining was performed to identify malignant cells using a primary rabbit polyclonal anti-S100 antibody (Dako Cytomation, Carpenteria, CA, 1:1000) and a secondary donkey Alexa 568-conjugated anti-rabbit antibody (Invitrogen, 1:500). Images were acquired using an automated Zeiss Axioscan.Z1 (0.22 µm/pixel) and were segmented using Definiens Tissue Studio 4.4.2 (Definiens AG, Munich, Germany). The nucleus Detection algorithm was used to extract the mean HuR fluorescence intensity for each cell within the region of interest (ROI). All areas of homogeneous S100 staining available within the tumor section were selected as ROI. The number of cells analysed per sample is specified in Table S1. The data processing was conducted in MatlabR 2018.b. Z-scores were defined for cells having an average nucleus intensity below two standard deviation. Samples were subsequently ranked for this value. For samples for which we had enough paraffin-embedded material left (8 samples /11), the procedure was entirely repeated to ensure the reproducibility of the ranking (small box in Fig.1G). The whole procedure was blindly performed before the tumor response to BRAF inhibition was evaluated.
2.8 Single-cell mass cytometry analysis
Although we analysed fewer than 15 markers, we used mass cytometry, considering its advantages compared to fluorescence-based flow cytometry: (i) there is no need to compensate for spectral overlap (ii) there is no background due to cell auto-fluorescence (iii) barcoding of the samples and their simultaneous analysis within the same staining reaction minimizes the staining variability among samples being compared [3]. Pretitrated, validated metal-conjugated antibodies were purchased from Fluidigm. The metal tag selection was optimized using the Maxpar panel designer (Fluidigm). The Nd144-tagged anti-HuR antibody was generated using a BSA and azide free anti-HuR antibody from Abcam (EPR17397, catalogue n° ab242410) and the Maxpar Antibody labelling kit according to the manufacturer’s instructions. This antibody was validated and titrated by infecting at various m.o.i. the A375 cell line with a previously generated adenovirus expressing HuR [4] and by confronting the CyTOF results to the ones obtained with flow cytometry analysis of the same cell line using the same un-tagged antibody. CyTOF analysis was conducted at steady state in parental and adapted (± Li2CO3) A375 cells. Cells were plated at 8x105 per well in six-well plates and grown in their corresponding medium containing or not the final concentration of vemurafenib used in the adaptive regimen. After three days, cells were washed three times with PBS, incubated with Versene for 5 min at 37°C, washed again, transferred to falcon tubes, counted and submitted to a standard Ficoll extraction protocol to eliminate dead cells. After being counted again, 4x106 cells for each sample were incubated for 5 min at 37°C in serum-free medium containing Cisplatin -194Pt (Fluidigm) at a final concentration of 1µmol/L. Cisplatin staining was quenched by washing the cells with prewarmed PBS containing 10% FBS. Subsequent to being washed, cells were fixed in 500 µL of Maxpar Fix 1 buffer for 10 min at room temperature, washed and froze in DMSO containing 10% FBS. This
4
freezing step was applied at TP1 and TP2 so as to obtain a homogeneous conditioning of samples and in order to conduct the barcoding and the final staining simultaneously on samples obtained at different time-points.
Upon being thawed, 1x106 cells from each sample were washed in 10 volumes of Maxpar cell staining buffer (CSB). For barcoding [5], Tin (Sn) isotopes (isotopes Sn119, Sn122, Sn124 used for these samples, Trace Sciences, ON, Canada, stored in concentrated 37% HCL at 30mM) were pre-diluted at 10 µM in Maxpar acquisition solution (CAS). Each sample was resuspended in 225 µL of CSB and deposited in a well of a 96- well plate containing 25 µL of a pre-diluted Sn isotope (1 µM final concentration). After 30 min incubation at room temperature, cells were washed twice with CSB, resuspended in 200 µL of CSB and all samples were mixed in a single tube. After being counted again, 3x106 cells of the mixed barcoded samples were resuspended in 50 µL of CSB and mixed with an equal volume of a master mix of metal-conjugated antibodies directed against cell surface markers, that is, 169Tm-anti-EGFR (clone D38B1, Fluidigm), 167Er-anti-MET (D1C2, Fluidigm), 156Gd-anti-PDGFRB (18A2, Fluidigm). The master mix was prepared so as to get a 1:100 final dilution of antibodies (1 µL of each pretitrated antibody in 50 µL of CSB). After 30 min of incubation at room temperature, cells were washed with CSB and chilled on ice for 10 min, before being mixed for further permeabilization with a pre-chilled 70% methanol solution. Following 15 min incubation on ice, cells were washed with CSB twice, resuspended in 50 µL of CSB and mixed with an equal volume of a master mix containing the custom made 144Nd-anti-HuR (3A2, 1:200, 0.5 µL of a 5µg/mL stock solution) and the following metal-conjugated antibodies prepared as described above (for the cell surface markers): 161Dy-anti- Ki67 (B56, DVS Sciences), 166Er-anti-phospho-S807/S811-Rb (J112-906, DVS Sciences), 142Nd-anti-cleaved Caspase 3 (D3E9, Fluidigm), 150Nd-anti-p53 (DO-7, Fluidigm), 159Tb-anti-p21 Waf1/Cip1 (12D1, Fluidigm),
171Yb-anti-phospho-T202/Y204-ERK1/2 (D13.14.4E, Fluidigm), 152Sm-anti-phopho-S473-AKT (D9E, Fluidigm), 172Yb-anti-phospho-S235/S236-S6 (N7-548, DVS Sciences), 149Sm-anti-phospho-T37/T46-4EBP1 (236B4, Fluidigm). After 30 min incubation at room temperature, cells were washed twice with CSB, resuspended in 3 mL of Fix 1 buffer solution containing a DNA intercalator (125 nmol/L final concentration, Cell-ID Intercalator 191/193Ir, Fluidigm) and following 10 min additional incubation at room temperature, were washed with CSB. Dry pellets were stored at + 4°C until acquisition. Prior to mass cytometry analysis, cells were adjusted to 2.5-5.105/mL in CAS, mixed with 1:10 diluted EQ four calibration beads (Fluidigm) and filtered with cell strainer caps. The sample was run on CyTOF 2 mass cytometer (Fluidigm). Live singlets were gated on Cis-194Pt staining (dead cell exclusion) and 191/193Ir staining (DNA content analysis for debris and doublets exclusion) in the Cytobank software and were debarcoded using the deconvolution algorithm CATALYSTLite (http://imlspenticton.uzh.ch:3838/ CATALYSTLite/). Prior to being exported to the R environment for viSNE, SPADE and cluster optimization, an equalized cell number among the samples was generated by the viSNE implementation in Cytobank Premium.
In order to define the optimal number of clusters, we first generated 100 different viSNE maps out of the combination of 20 different values for the perplexity parameter (set to be 5-10-15…up to 100) and 5 different values of iteration (1000 up to 5000). Each of these maps was used to run the SPADE algorithm with 40 different values of target number of clusters (k ꞊ 5, 15, 25 …up to 395) and for each value of k (and same viSNE parameters), the algorithm was reiterated three times. The theta value was set to zero for all conducted viSNE runs (no use of Barnes-Hut approximation of the learning gradient) [6]. These varying parametrizations allowed us to generate 12 000 combinations of t-SNE embeddings and subsequent SPADE clusterization. The error sum of squares (SSE) was then calculated for each value of k so as to fit upon non-linear minimization an optimization curve (Fig. S3). According to the knee (or elbow) method, the optimal value for k was defined as the maximum value given by the curvature function of this curve [7].
2.9 Mouse xenografts and in vivo drug studies
Animal experiments were approved by the Animal Welfare Commission of the Canton of Geneva (approval n° GE/108/18) and followed the Swiss guidelines for animal experimentation. Two million A375 cells were resuspended in 100 µl of PBS and mixed with an equal volume of Matrigel (Corning® Matrigel® Matrix High Concentration, Phenol-Red free) and injected subcutaneously into the posterior flanks of six-week-old female immunodeficient NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ mice (NSG, Charles River). Tumor formation was monitored twice a week using calliper measurements and calculated by the ellipsoidal formula: tumor volume ꞊ (length x width2) x 0.5. When tumors reached a volume of approximately 0.5 cm3, mice were assigned into two equal arms to receive or not Li2CO3 containing chow dosed at 0.25% (2.5g/kg) and subsequently were followed for two weeks before initiating the vemurafenib treatment. This initial Li2CO3 treatment follow-up period allowed us to use two arms instead of four and still assess a possible effect of Li2CO3 administrated alone. All mice received the vemurafenib treatment by oral gavage once a day; 240 mg vemurafenib tablets (for human use) were manually ground and suspended in a water containing solution of carboxymethylcellulose (CMC, 1%) and DMSO (5%). The final concentration of vemurafenib was 16.5 mg/mL. A fixed vemurafenib dosage of 100
5
mg/kg/day was administered in all mice (150 μL). To ensure a homogeneous vemurafenib treatment in both arms, the residual immediate post-gavage blood concentration of vemurafenib was measured in all mice, 6 days after initiating the treatment, using an in-house liquid chromatography-electrospray ionization-tandem mass spectrometry method as described elsewhere [8]. Four μL of whole blood were collected from a small incision on the animal tail and spotted onto a filter paper card from Whatman (Dassel, Germany). In the untreated and the Li2CO3 treated arms, the deduced residual plasma vemurafenib concentrations were respectively (mean ± SD) 2.85 ± 2.03 and 3.3 ± 2.68 μg/mL. A repeated measurement ANOVA with post-hoc comparison was used to confirm the significance of differences in tumor growth.
References
[1] A. Badawi, A. Biyanee, U. Nasrullah, et al., Inhibition of IRES-dependent translation of caspase-2 by HuR confers chemotherapeutic drug resistance in colon carcinoma cells, Oncotarget 9 (2018) 18367-18385.
[2] M.E. Truchetet, N.C. Brembilla, E. Montanar, et al., Interleukin-17A+ cell counts are increased in systemic sclerosis skin and their number is inversely correlated with the extent of skin involvement, Arthritis Rheum. 65 (2013) 1347-1356.
[3] H.G. Fienberg, G.P. Nolan, Mass cytometry to decipher the mechanism of nongenetic drug resistance in cancer, Curr. Top. Microbio.
Immunol. 377 (2014) 85-94.
[4] M. Fernandez, H. Sutterluty-Fall, C. Schwarzler, et al., Overexpression of the human antigen R suppresses the immediate paradoxical proliferation of melanoma cell subpopulations in response to suboptimal BRAF inhibition, Cancer Med. 6 (2017) 1652-1664.
[5] C. Schwärzler, M. Garcia, Alternative Mass-Cell-Barcoding Options and their Application to Control Samples, CYTO2018, 33rd Congress of the International Society for Advancement of Cytometry 2018 April-May, Prague.
[6] L. van der Maaten, Accelerating t-SNE using tree-based algorithms, J. Mach. Learn. Res. 15 (2014) 1-21.
[7] V. Satopää, J. Albrecht, D. Irwin, Finding a "Kneedle" in a haystacck: Detecting knee points in system behavior, ICDCSW 2011, 31st jjjjjjInternational Conference on Distributed Computing Systems 2011 June, Minneapolis.
[8] C.M. Nijenhuis, H. Rosing, J.H. Schellens, et al., Quantifying vemurafenib in dried blood spots using high-performance LC-MS/MS, iiiiiiBioanalysis 6 (2014) 3215-3224.
