Molecular & Cellular Proteomics · 2009 · 91 citations · 44 references
Life SciencesMolecular BiologyOncologyMolecular OncologyOncogenesProteomicsCancer ProteomicsMass Spectrometry Data AnalysisMass Spectrometry-based ProteomicsMolecular Tumor ProfilingPeptide AnalysisProteomic ProfilingActivating Point MutationsAcute Myeloid Leukemia BiologyMass Spectrometry ProteomicsMass Spectrometry-based Phosphorylation AnalysisMass Spectrometry-based Protein IdentificationMass Spectrum InterpretationPeptide Mass SpectrometryProteomic Data Analysis
The FLT3 receptor tyrosine kinase plays an important role in normal hematopoietic development and leukemogenesis. Point mutations within the activation loop and in-frame tandem duplications of the juxtamembrane domain represent the most frequent molecular abnormalities observed in acute myeloid leukemia. Interestingly these gain-of-function mutations correlate with different clinical outcomes, suggesting that signals from constitutive FLT3 mutants activate different downstream targets. In principle, mass spectrometry offers a powerful means to quantify protein phosphorylation and identify signaling events associated with constitutively active kinases or other oncogenic events. However, regulation of individual phosphorylation sites presents a challenging case for proteomics studies whereby quantification is based on individual peptides rather than an average across different peptides derived from the same protein. Here we describe a robust experimental framework and associated error model for iTRAQ-based quantification on an Orbitrap mass spectrometer that relates variance of peptide ratios to mass spectral peak height and provides for assignment of p value, q value, and confidence interval to every peptide identification, all based on routine measurements, obviating the need for detailed characterization of individual ion peaks. Moreover, we demonstrate that our model is stable over time and can be applied in a manner directly analogous to ubiquitously used external mass calibration routines. Application of our error model to quantitative proteomics data for FLT3 signaling provides evidence that phosphorylation of tyrosine phosphatase SHP1 abrogates the transformative potential, but not overall kinase activity, of FLT3-D835Y in acute myeloid leukemia. The FLT3 receptor tyrosine kinase plays an important role in normal hematopoietic development and leukemogenesis. Point mutations within the activation loop and in-frame tandem duplications of the juxtamembrane domain represent the most frequent molecular abnormalities observed in acute myeloid leukemia. Interestingly these gain-of-function mutations correlate with different clinical outcomes, suggesting that signals from constitutive FLT3 mutants activate different downstream targets. In principle, mass spectrometry offers a powerful means to quantify protein phosphorylation and identify signaling events associated with constitutively active kinases or other oncogenic events. However, regulation of individual phosphorylation sites presents a challenging case for proteomics studies whereby quantification is based on individual peptides rather than an average across different peptides derived from the same protein. Here we describe a robust experimental framework and associated error model for iTRAQ-based quantification on an Orbitrap mass spectrometer that relates variance of peptide ratios to mass spectral peak height and provides for assignment of p value, q value, and confidence interval to every peptide identification, all based on routine measurements, obviating the need for detailed characterization of individual ion peaks. Moreover, we demonstrate that our model is stable over time and can be applied in a manner directly analogous to ubiquitously used external mass calibration routines. Application of our error model to quantitative proteomics data for FLT3 signaling provides evidence that phosphorylation of tyrosine phosphatase SHP1 abrogates the transformative potential, but not overall kinase activity, of FLT3-D835Y in acute myeloid leukemia. Deregulated tyrosine phosphorylation is a well established molecular hallmark in the development of both solid tumors and hematopoietic malignancies (1Blume-Jensen P. Hunter T. Oncogenic kinase signalling.Nature. 2001; 411: 355-365Crossref PubMed Scopus (3143) Google Scholar, 2Schlessinger J. Cell signaling by receptor tyrosine kinases.Cell. 2000; 103: 211-225Abstract Full Text Full Text PDF PubMed Scopus (3537) Google Scholar). In acute myeloid leukemia (AML), 1The abbreviations used are:AMLacute myeloid leukemiaFLT3Fms-like tyrosine kinase 3ITDinternal tandem duplicationWTwild typeFLFLT3 ligandIL-3interleukin 3STAT5asignal transducer and activator of transcription 5aJAK2Janus kinase 2SHP1protein tyrosine phosphatase, non-receptor type 6SOCS1suppressor of cytokine signaling 1GAB2growth factor receptor-bound protein 2-associated binding protein 2HCDhigher energy collisional dissociationMACLmaximum approximate conditional likelihoodr.t.room temperatureiTRAQisobaric tags for relative and absolute quantitationTMTtandem mass tagLTQlinear trap quadrupoleHOAcacetic acidCADcollisionally activated dissociationERKextracellular signal-regulated kinaseMAPKmitogen-activated protein kinasepAKT, pERK, and pSTAT5phosphorylated AKT, ERK, and STAT5. mutations of the Fms-like tyrosine kinase 3 (Flt3) gene or its overexpression represent the most frequent molecular abnormalities, observed in ∼35 and ∼90% of patients, respectively (3Gilliland D.G. Griffin J.D. The roles of FLT3 in hematopoiesis and leukemia.Blood. 2002; 100: 1532-1542Crossref PubMed Scopus (1191) Google Scholar, 4Small D. FLT3 mutations: biology and treatment.Hematology Am. Soc. Hematol. Educ. Program. 2006; 1: 178-184Crossref Scopus (190) Google Scholar). Internal tandem duplication (ITD) in the juxtamembrane region occurs in ∼25% of patients, whereas another 7% of cases manifest as point mutations in the activation loop at Asp-835 (typically D835Y). Both mutation classes induce constitutive tyrosine kinase activity and confer IL-3-independent growth of the factor IL-3-dependent BaF3 and 32D cells. Interestingly, the ITD and D835Y mutants respond differently to the current suite of targeted therapeutics (5Knapper S. FLT3 inhibition in acute myeloid leukaemia.Br. J. Haematol. 2007; 138: 687-699Crossref PubMed Scopus (58) Google Scholar), and in vitro studies suggest that they can activate different downstream targets, including STAT5a (6Choudhary C. Schwäble J. Brandts C. Tickenbrock L. Sargin B. Kindler T. Fischer T. Berdel W.E. Müller-Tidow C. Serve H. AML-associated Flt3 kinase domain mutations show signal transduction differences compared with Flt3 ITD mutations.Blood. 2005; 106: 265-273Crossref PubMed Scopus (206) Google Scholar, 7Grundler R. Miething C. Thiede C. Peschel C. Duyster J. FLT3-ITD and tyrosine kinase domain mutants induce 2 distinct phenotypes in a murine bone marrow transplantation model.Blood. 2005; 105: 4792-4799Crossref PubMed Scopus (166) Google Scholar, 8Rocnik J.L. Okabe R. Yu J.C. Lee B.H. Giese N. Schenkein D.P. Gilliland D.G. Roles of tyrosine 589 and 591 in STAT5 activation and transformation mediated by FLT3-ITD.Blood. 2006; 108: 1339-1345Crossref PubMed Scopus (109) Google Scholar). Collectively, these observations motivate continued efforts to delineate oncogenic pathways in further detail to expand our understanding of the molecular basis for disease progression and to identify additional therapeutic entry points. acute myeloid leukemia Fms-like tyrosine kinase 3 internal tandem duplication wild type FLT3 ligand interleukin 3 signal transducer and activator of transcription 5a Janus kinase 2 protein tyrosine phosphatase, non-receptor type 6 suppressor of cytokine signaling 1 growth factor receptor-bound protein 2-associated binding protein 2 higher energy collisional dissociation maximum approximate conditional likelihood room temperature isobaric tags for relative and absolute quantitation tandem mass tag linear trap quadrupole acetic acid collisionally activated dissociation extracellular signal-regulated kinase mitogen-activated protein kinase phosphorylated AKT, ERK, and STAT5. Over the past decade, mass spectrometry-based proteomics has become the technique of choice for large scale identification of peptides derived from various biological contexts (9Domon B. Aebersold R. Mass spectrometry and protein analysis.Science. 2006; 312: 212-217Crossref PubMed Scopus (1621) Google Scholar, 10Mann M. Hendrickson of and by mass 2001; PubMed Scopus Google Scholar). Moreover, relative quantification measurements, a routine means to protein and as a of biological as the B. S. N. S. S. S. S. P. S. M. protein quantitation in isobaric Full Text Full Text PDF PubMed Scopus Google and J. S. J. T. R. C. mass a quantification for of protein by PubMed Scopus Google for quantitative including with the of derived from in vitro and in the to to in a and signal for peptide derived from experimental and signal for from of peptide The of and signals is for of or dissociation both and tags to that different stable the peak or of these signals a relative of peptide in biological that individual peptides to a of individual ratios from all peptides provides a for quantification at the protein B. S. N. S. S. S. S. P. S. M. protein quantitation in isobaric Full Text Full Text PDF PubMed Scopus Google Scholar, C. for quantitative proteomics 2005; PubMed Scopus Google Scholar, a for protein 2006; PubMed Scopus Google Scholar, S. D. M. protein quantification based on peptide quantification 2007; PubMed Scopus Google Scholar, C. of relative mass spectrometry data from PubMed Scopus Google Scholar, M. D. S. M. S. T. J. M. C. T. C. N. J. B. proteomics of of clinical kinase 2007; PubMed Scopus Google Scholar). is for protein studies peptides and used as evidence for the quantification of a protein. The is different for quantification of as tyrosine Here is for to a of tyrosine Moreover, phosphorylation events on the same the relative across sites as a of regulation biological a ratios from individual peptides be the of peptide evidence for a protein. In to these is that peptides to in a of to peptide ratios these peptides by is of for individual on for than or a L. J. for from tags and with and tandem mass 2005; PubMed Scopus Google Scholar, T. P. S. J. S. C. D. for for and in clinical isobaric with and tandem mass PubMed Scopus Google Scholar), or of based on characterization of individual ion C. Fischer R. T. S. L. model the of in quantitative PubMed Scopus Google Scholar). additional well is that quantification is to different of error and different data to the of based on stable of and associated with the of or stable to other of for peptide ratios based on stable by in B. H. M. by in as a and to 2002; 1: Full Text Full Text PDF PubMed Scopus Google and B. Aebersold R. of protein PubMed Scopus Google across the to and of the to the In quantification and is based on of that and to a In principle, to error that to or from and development of an for assignment of to peptide Here we and energy in tandem on an to identify and quantify peptides from that wild or D835Y FLT3 In peptides on from In we an error model that relates the variance of ratios to observed ion and provides a p value, q value, and confidence interval for every peptide demonstrate that our model is stable over time and can be applied in a manner directly analogous to ubiquitously used external mass calibration routines. Collectively, these of for quantification in phosphorylation is of signaling pathways mediated by FLT3 wild type and D835Y that tyrosine phosphorylation of SHP1 by but not by the transformative potential, but not overall kinase activity, of FLT3-D835Y in by of BaF3 FLT3-ITD and FLT3-D835Y with at in with and with 2 and BaF3 wild type FLT3 with as a of in of at and at of BaF3 with for Cell with of and at Cell with 2 of with 1 activated with at for 1 and with at for 1 in the Cell to with of at a at with to and at to on a and with acetic acid peptides and at of with of for for for and for to the with of and peptides with of the at for 1 all and in a of from Cell and of from with of protein in at for on a The protein with of for peptides with of the to with 1 the to of protein and at on a at and with of with of 7% at for at and the a 6 at 3 the with at for and at for on the directly a trap with of 1 at 2 The trap with at for and to a M. L. for and proteomics of tyrosine signaling in PubMed Scopus Google Scholar). at a of with a in in in for and in a with a peptides directly an for with a in the Orbitrap with at maximum time at and at for its signal over and its to to be for and for a and in the linear and an in the energy with energy and and in the Orbitrap at a of The mass for at The data with to and as S. Orbitrap for quantitative of Am. Soc. Mass PubMed Scopus Google Scholar). The to and all in a that of FLT3 and the protein on The as The mass for at for the and for the as of on and and of and and The data the with the same The to peptides that tyrosine in the and The to be at a of and at a of with a over to and phosphorylation of the region for peptide from the mass spectrometry data and with from the our M. T. an based for proteomics data PubMed Scopus Google Scholar). from further observed within of the in the The for all The ion and for time and in C. for quantitative proteomics 2005; PubMed Scopus Google Scholar). all for a from by the maximum by the data for of the in that peptides from of BaF3 from and as in the as all of to measurements, a is applied that the the and the the variance of the peak and as in by for the variance in Scopus Google Scholar, M. Am. Scopus Google as an peptides by as ion peak as for in protein a of the ion from these data used as a factor for of data from the The used to p the of a p in be used to q J.D. R. for 100: PubMed Scopus Google Scholar). a confidence interval by the error model at the observed of the The FLT3 BaF3 at with by or for The wild type with or with FLT3 ligand or ligand for The for protein a The protein to by or to the of protein with the for 1 with at and of the protein and at The protein in the in and the at or to a FLT3 SHP1 and from STAT5 and from AKT, and from Cell peptides with or ion trap the the as by the peptide and well used for activation activation q B. S. M. dissociation for peptide 2007; PubMed Scopus Google the of a to the on an mass provides for at used in ion trap with of mass used a for the but the to a energy of every in the linear ion trap by a higher energy in the and to a peptide from and B. S. M. dissociation for peptide 2007; PubMed Scopus Google Scholar), both confidence assignment based on with the and of the of the mass region the of that to of the these can for the quantification is based on individual peptide data the of by isobaric and the of mass in with iTRAQ-based quantification mass within the ion region to isobaric other of error and the development of an model to the associated with quantification the associated with we peptides from of murine BaF3 cells. from of with on an with to and as ratios our M. a for mass spectrometry PubMed Scopus Google and as a of ion mass spectral peak height with other large scale data S. P. for and Scopus Google Scholar), we observed a of variance on spectral peak height C. for quantitative proteomics 2005; PubMed Scopus Google Scholar, a for protein 2006; PubMed Scopus Google in proteomics but in most cases for both and data In a C. Fischer R. T. S. L. model the of in quantitative PubMed Scopus Google a variance based on of a of In we to the of by of peptides derived from of ratios that a of mass spectral peak derived an error model based on the that for all and that across the ion In we that the of the ion mass spectral peak height a variance In for the of our calibration we the of as a from a the variance of the ratios is to that of the peak height to the of and represent peak both with a and variance the and in 1 by maximum approximate conditional likelihood for the variance in Scopus Google and established a Interestingly, we observed that of in 1 the of a at with that a in as a of spectral peak height in Scholar, L. of and of data in of peak and in 1: Scopus Google Scholar, relative ion from ion mass PubMed Scopus Google Scholar, and and and in Scopus Google Scholar). we the peak height in the mass spectrometry data based on the ion to the for M. M. D. T. B. and quantification on an Orbitrap mass Full Text Full Text PDF PubMed Scopus Google Scholar). the region of the to our Here the variance at spectral peak height in with 1 In the of quantification is with other of for the Orbitrap mass spectrometer S. of mass in Orbitrap mass Am. Soc. Mass 2006; PubMed Scopus Google Scholar, J.D. J. quantification of stable peptides a linear ion mass 2007; PubMed Scopus Google Scholar). with studies in and to other as L. of and of data in of peak and in 1: Scopus Google Scholar, relative ion from ion mass PubMed Scopus Google for spectral peak as a of observed and the of data across a the in and Orbitrap we our with efforts to the of our ratios as a of and L. of and of data in of peak and in 1: Scopus Google and relative ion from ion mass PubMed Scopus Google Scholar). The the to our whereas the to the by a framework in the of studies L. of and of data in of peak and in 1: Scopus Google Scholar, relative ion from ion mass PubMed Scopus Google Scholar). the the by and we experimental data with from L. of and of data in of peak and in 1: Scopus Google and relative ion from ion mass PubMed Scopus Google and our model a for the of these data as compared with the L. of and of data in of peak and in 1: Scopus Google Scholar, relative ion from ion mass PubMed Scopus Google Scholar). further our that the data or on a different mass spectrometry and different the of our we the from 3 the observed spectral peak height to ion and the data used to the and in of both data with a of we a based on variance of the data in The of the data our of for spectral peak and the assignment of p and q for all peptide ratios within a In the variance an of confidence for all Collectively, these suggest that a variance can be established and applied to all data with in to the of external mass calibration used in mass The to for peptide ratios experimental that of is used to the signal that from of a However, the to the is not well The experimental in with a of and growth robust of in an active signaling we established murine BaF3 that point mutation or ITD FLT3 normal and with and peptides by on and by in iTRAQ-based quantitation from of within the for has in detail M. J. C. in and the and the PubMed Scopus Google Scholar). with our on tyrosine we observed in a of and the peptides from further we sites on and and In sites on across and the basis data for the of our peptide phosphorylation and quantification data of pathways activated the kinase with that overall kinase activity of to data by 1 and of a in assignment of including and phosphorylated and phosphorylated as However, of our error model regulation of tyrosine phosphorylation on signaling our proteomics data show that phosphorylation of on STAT5a by ligand not a our error model provides a p of these data with by 1 and the of our to as compared with and p for STAT5a in the of a biological of our error model that of the sites based on to be based on the large error associated with of signals the error model to a confidence interval for every and we can that the of STAT5a be and as in the of ITD or D835Y mutation on the signaling that normal we ratios for peptides derived from D835Y and ITD to the derived from with kinase Both ITD and D835Y show tyrosine phosphorylation on compared with including and is with the constitutive kinase activity for both mutants (3Gilliland D.G. Griffin J.D. The roles of FLT3 in hematopoiesis and leukemia.Blood. 2002; 100: 1532-1542Crossref PubMed Scopus (1191) Google Scholar, C. Müller-Tidow C. Berdel W.E. Serve H. transduction of oncogenic J. Hematol. 2005; PubMed Scopus Google Scholar). by in BaF3 and The of sites on in including and with in the of FLT3 (6Choudhary C. Schwäble J. Brandts C. Tickenbrock L. Sargin B. Kindler T. Fischer T. Berdel W.E. Müller-Tidow C. Serve H. AML-associated Flt3 kinase domain mutations show signal transduction differences compared with Flt3 ITD mutations.Blood. 2005; 106: 265-273Crossref PubMed Scopus (206) Google Scholar). phosphorylation of STAT5 with and of hematopoietic both in vitro and in C. Müller-Tidow C. Berdel W.E. Serve H. transduction of oncogenic J. Hematol. 2005; PubMed Scopus Google Scholar, R. D. FLT3 signaling to a in myeloid 2005; PubMed Scopus Google Scholar). in the of the phosphorylation of STAT5a is in ITD at a to that in with and is in D835Y these observations in BaF3 by an for on STAT5 and data with that activation of downstream of FLT3 mutants (6Choudhary C. Schwäble J. Brandts C. Tickenbrock L. Sargin B. Kindler T. Fischer T. Berdel W.E. Müller-Tidow C. Serve H. AML-associated Flt3 kinase domain mutations show signal transduction differences compared with Flt3 ITD mutations.Blood. 2005; 106: 265-273Crossref PubMed Scopus (206) Google Scholar). used a of FLT3 kinase activity C. P. D. T. Gilliland D.G. Griffin J.D. of FLT3 in leukemia by the tyrosine kinase 2002; 1: Full Text Full Text PDF PubMed Scopus Google Scholar), to STAT5a phosphorylation to constitutive activation of the receptor by of on STAT5a in the of the whereas phosphorylation on the same in the of both ITD and D835Y mutations and the ITD is with a clinical our data demonstrate that overall kinase activity is higher for the D835Y is with for mutations as phosphorylation not correlate with oncogenic of the loop of 2006; 103: PubMed Scopus Google Scholar). to the of activity and transformative is that and pathways activated in the of the D835Y Interestingly, we observed of tyrosine phosphorylation on in to STAT5a in the of D835Y as compared with FLT3 Moreover, phosphorylation of is to be for downstream activation J. T. of activity phosphorylation of in the kinase activation PubMed Scopus Google Scholar). directly for we ratios for peptides from D835Y and ITD mutants in respectively a and In the of D835Y we observed tyrosine phosphorylation on including and In in hematopoietic to signaling by protein tyrosine Full Text Full Text PDF PubMed Scopus Google Scholar), the activity of including type tyrosine by the H. M. T. with and of kinase by the protein tyrosine phosphatase PubMed Scopus Google Scholar, of to the receptor of and of Full Text PDF PubMed Scopus Google Scholar, P. R. S. of regulation of of in cytokine 103: PubMed Scopus Google Scholar). sites on SHP1 and all as compared with the ITD and with a p of SHP1 by that protein in the of FLT3-ITD and and SHP1 phosphorylation inhibition of FLT3-D835Y with C. P. D. T. Gilliland D.G. Griffin J.D. of FLT3 in leukemia by the tyrosine kinase 2002; 1: Full Text Full Text PDF PubMed Scopus Google Scholar). Collectively, our suggest as compared with constitutive activation by D835Y to phosphorylation of the phosphatase SHP1 as well as of that a of linear ion trap with higher energy an means to peptides on an Orbitrap the mass in the ion region from that quantification a robust error model that of p value, q value, and confidence interval on from a a relative to for quantitative proteomics in that we a error model from a large of variance P. P. R. J. model for mass spectrometry based PubMed Scopus Google Scholar), all within a rather than characterization of individual mass spectrometry C. Fischer R. T. S. L. model the of in quantitative PubMed Scopus Google Scholar, L. of and of data in of peak and in 1: Scopus Google Scholar, relative ion from ion mass PubMed Scopus Google Scholar, M. S. H. C. for for protein spectrometry of PubMed Scopus Google Scholar). The of our for of peptides on an to of for data on by and in on an mass spectrometer the of our In our framework is directly analogous to the of external mass calibration and provides the that the experimental for calibration is to that for quantification The to variance and the of directly proteomics that to biological to rather than large scale of peptides that be to the at In our that overall phosphorylation downstream of the point as compared with the FLT3-ITD a and Here the of ratios from and these data not to by L. J. for from tags and with and tandem mass 2005; PubMed Scopus Google Scholar, J. M. peptide identification mass and protein PubMed Scopus Google Scholar, protein quantification ion trap or mass and with 2006; PubMed Scopus Google Scholar, of and by mass spectrometry 2007; PubMed Google that internal to a large that of our error model be for quantitative of or clinical or in in of and biological not In we applied quantitative proteomics in with our error model to identify of signaling downstream of FLT3 mutants in murine BaF3 cells. kinases in principle, clinical to for in the case of inhibition by for that become to therapeutics Griffin J.D. of for the of myeloid 2007; PubMed Scopus Google Scholar). observations the need for continued of oncogenic pathways to identify and molecular targets. In the of FLT3 signaling in BaF3 we sites on of phosphorylation events the oncogenic mutants and across that FLT3-ITD and FLT3-D835Y signaling as for by differences in phosphorylation on and The plays a role in of signaling for a of receptor tyrosine kinases of receptor tyrosine kinase signaling by protein tyrosine Cell 2001; Full Text Full Text PDF PubMed Scopus Google Scholar, J. Roles of the tyrosine phosphatase in the regulation of 2000; PubMed Scopus Google Scholar). In SHP1 has in a of malignancies C. M. L. The of the protein tyrosine phosphatase in PubMed Scopus Google Scholar). from P. M. P. J. D. mutation signaling of 2005; Full Text Full Text PDF PubMed Scopus Google Scholar), in the of leukemia that of by a in phosphatase activity, a for of kinase activity and oncogenic FLT3 and and other studies (6Choudhary C. Schwäble J. Brandts C. Tickenbrock L. Sargin B. Kindler T. Fischer T. Berdel W.E. Müller-Tidow C. Serve H. AML-associated Flt3 kinase domain mutations show signal transduction differences compared with Flt3 ITD mutations.Blood. 2005; 106: 265-273Crossref PubMed Scopus (206) Google Scholar, 7Grundler R. Miething C. Thiede C. Peschel C. Duyster J. FLT3-ITD and tyrosine kinase domain mutants induce 2 distinct phenotypes in a murine bone marrow transplantation model.Blood. 2005; 105: 4792-4799Crossref PubMed Scopus (166) Google Scholar, C. Brandts C. J. Tickenbrock L. Sargin B. Berdel W.E. Müller-Tidow C. Serve H. of STAT5 by oncogenic 2007; PubMed Scopus Google and additional detail that differences observed FLT3-ITD and that tyrosine phosphorylation at and the of SHP1 signaling by of P. R. S. of regulation of of in cytokine 103: PubMed Scopus Google Scholar). data that SHP1 can signaling pathways a of activity or data demonstrate that SHP1 is phosphorylated at the in the of to be phosphorylation of these sites from with FLT3 or is mediated a downstream our with the that activation of SHP1 tyrosine phosphorylation the transformative of the point as compared with Collectively, our data a of signaling downstream of oncogenic FLT3 mutants and suggest for further for and of the In we for of the with
44
Quantitative analysis of complex protein mixtures using isotope-coded affinity tags
Steven P. Gygi, Beate Rist, Scott A. Gerber et al. · Nature Biotechnology · 1999 · 4.7K citations