Now, More Than Ever, Proteomics Needs Better Chromatography

From plant research to biomedicine, proteome analysis plays a critical role in many areas of biological inquiry. Steady improvement in mass spectrometer (MS) technology has transformed the speed and depth of proteome analysis. Proteomes of simple organisms can now be sequenced to near completion in just over an hour. Comparable coverage of mammalian proteomes, however, still requires hours or even days of analysis. Here we ask why current technology fails to achieve comprehensive and rapid analysis of the more complex mammalian proteomes. We propose that further advancements in MS technology alone are unlikely to solve this problem and suggest that concomitant improvements in peptide separation technology will be critical. From plant research to biomedicine, proteome analysis plays a critical role in many areas of biological inquiry. Steady improvement in mass spectrometer (MS) technology has transformed the speed and depth of proteome analysis. Proteomes of simple organisms can now be sequenced to near completion in just over an hour. Comparable coverage of mammalian proteomes, however, still requires hours or even days of analysis. Here we ask why current technology fails to achieve comprehensive and rapid analysis of the more complex mammalian proteomes. We propose that further advancements in MS technology alone are unlikely to solve this problem and suggest that concomitant improvements in peptide separation technology will be critical. Advancing myriad applications—from basic biological studies to precision medicine—will require technologies for rapid and comprehensive proteome profiling. Dominating the field, liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) delivers increasingly promising results (Riley et al., 2016Riley N.M. Hebert A.S. Coon J.J. Cell Syst. 2016; 2: 142-143Abstract Full Text Full Text PDF PubMed Scopus (33) Google Scholar). The key tenets of this strategy, however, have been in place for nearly 20 years (Pandey and Mann, 2000Pandey A. Mann M. Nature. 2000; 405: 837-846Crossref PubMed Scopus (1943) Google Scholar). Proteins are digested into smaller, more manageable peptides, which are separated using chromatography and then presented to the MS. The eluting peptide ions are converted from the condensed phase to the gas phase using electrospray ionization. Following injection into the MS, their overall mass-to-charge ratio is recorded. In order of abundance, each unique peptide species is isolated and subjected to fragmentation, either by collisions, ion/ion reactions, or photons. The resulting product ions are then measured to create a tandem MS (MS/MS) scan. These MS/MS scans are then searched against a database of in silico generated spectra, scored, and used to generate a list of confidently identified peptides and consequently proteins. Twenty years ago, state-of-the-art MS systems could acquire up to one MS/MS scan per second. In this scenario, the rate-limiting step was the collection of MS/MS spectra (Link et al., 1999Link A.J. Eng J. Schieltz D.M. Carmack E. Mize G.J. Morris D.R. Garvik B.M. Yates 3rd, J.R. Nat. Biotechnol. 1999; 17: 676-682Crossref PubMed Scopus (2071) Google Scholar). That is, a 60 min LC-MS/MS experiment could not possibility produce more than 3,600 peptide identifications, as only this many MS/MS spectra could be collected during that period (i.e., 60 MS/MS per minute × 60 min). To achieve greater proteomic analysis depth, the field moved toward multi-dimensional separations like MudPit (Link et al., 1999Link A.J. Eng J. Schieltz D.M. Carmack E. Mize G.J. Morris D.R. Garvik B.M. Yates 3rd, J.R. Nat. Biotechnol. 1999; 17: 676-682Crossref PubMed Scopus (2071) Google Scholar, Washburn et al., 2001Washburn M.P. Wolters D. Yates 3rd, J.R. Nat. Biotechnol. 2001; 19: 242-247Crossref PubMed Scopus (4077) Google Scholar). Fractionation of a proteome digest into 80 samples, each of which could be subjected to a 60 min LC-MS/MS experiment, allowed one to collect considerably more MS/MS spectra (nearly 300,000 in this example). Proteome depth increased as a consequence—but so too did the duration of the proteomics experiment. Mass spectrometer performance has since steadily improved in nearly every category, including mass accuracy, mass resolving power, scan speed, and sensitivity. With these improvements and other advances, a modern mass spectrometer now sequences up to 19 peptides per second, and near complete coverage of the whole yeast proteome can be achieved in just over 1 hr of analysis (Hebert et al., 2014Hebert A.S. Richards A.L. Bailey D.J. Ulbrich A. Coughlin E.E. Westphall M.S. Coon J.J. Mol. Cell. Proteomics. 2014; 13: 339-347Crossref PubMed Scopus (411) Google Scholar). Deep sequencing of more complex mammalian proteomes, however, still requires two-dimensional chromatography, mandating considerably extended analysis time (Richards et al., 2015Richards A.L. Merrill A.E. Coon J.J. Curr. Opin. Chem. Biol. 2015; 24: 11-17Crossref PubMed Scopus (74) Google Scholar). Because these experiments take several hours or even days to conduct, systems-wide studies of mammalian proteomes are not routine. We reason that to permit the practical analysis of hundreds or even thousands of mammalian proteomes we must substantially reduce the amount of analysis time required. This rather obvious goal will not raise controversy. However, there is no general consensus on which technological innovations should be pursued to achieve it. Here we present a largely overlooked but, in our opinion, primary obstacle to achieving rapid, whole mammalian proteome analysis— peptide separations. As MS systems have evolved to enable faster scanning, those seeking to further increase throughput have pursued single-shot proteomics, where the complex mixture of peptides is separated in only one dimension and analyzed via MS. Figure 1A highlights some prominent single-shot studies. First, we note the chromatographic format and performance is primarily uniform across these works; however, the MS scan rate, defined as the number of MS/MS scans taken per second, is steadily increasing over time, as new technology is released and adopted. What is not as obvious from the figure is an exciting concomitant trend of increasing proteomic sampling depth over this same period. Specifically, the 2011 work of Thakur et al. detected just under 3,000 yeast proteins from an 8-hr analysis (Thakur et al., 2011Thakur S.S. Geiger T. Chatterjee B. Bandilla P. Fröhlich F. Cox J. Mann M. Mol. Cell. Proteomics. 2011; 10: 003699Crossref Scopus (270) Google Scholar). The very next year that same lab reduced the separation duration to 4 hr and detected nearly 4,000 yeast proteins (Nagaraj et al., 2012Nagaraj N. Kulak N.A. Cox J. Neuhauser N. Mayr K. Hoerning O. Vorm O. Mann M. Mol. Cell. Proteomics. 2012; 11: 013722Crossref Scopus (304) Google Scholar). This remarkable improvement was primarily afforded by the adoption of a new MS platform offering an increase in MS/MS acquisition rate, high resolution, and accurate mass. Two years later, by using an even faster scanning Orbitrap hybrid (20 Hz), our laboratory further increased throughput for yeast proteomics by identifying 4,000 proteins in just over 1 hr of analysis (Hebert et al., 2014Hebert A.S. Richards A.L. Bailey D.J. Ulbrich A. Coughlin E.E. Westphall M.S. Coon J.J. Mol. Cell. Proteomics. 2014; 13: 339-347Crossref PubMed Scopus (411) Google Scholar). A similar upward trend is observed among the studies aimed at human proteomes, although even the latest works report identification of only about half of the expressed proteins (Pirmoradian et al., 2013Pirmoradian M. Budamgunta H. Chingin K. Zhang B. Astorga-Wells J. Zubarev R.A. Mol. Cell. Proteomics. 2013; 12: 3330-3338Crossref PubMed Scopus (106) Google Scholar, Scheltema et al., 2014Scheltema R.A. Hauschild J.P. Lange O. Hornburg D. Denisov E. Damoc E. Kuehn A. Makarov A. Mann M. Mol. Cell. Proteomics. 2014; 13: 3698-3708Crossref PubMed Scopus (231) Google Scholar). Complementing the increase in scan rate of all instruments are improvements in analysis sensitivity, sample preparation techniques, and data processing. The relative contributions of these factors must not be overlooked, but they are more difficult to assess and likely secondary to the astounding progression in MS performance. Based on this trajectory, we fully anticipated that further gains in single-shot proteomics would be readily achieved by the next-generation MS systems featuring faster data acquisition, unprecedented sensitivity, and ion transfer efficiency. Upon acquirement and use of the newest Orbitrap hybrid instrument (Fusion Lumos), capable of collecting high quality MS/MS spectra at a rate of over 40 Hz, we were dismayed to discover that our prediction did not prove true. Single-shot analysis of both human and yeast proteomes were minimally improved—far less than one would expect upon doubling MS/MS acquisition rate. To understand why this latest technology did not continue the trend toward performance improvement, we took a closer look at the literature and our data. Specifically, we wondered how many eluting peptides were presented to the MS system, what percentage was sampled, and whether all of the system’s sequencing capacity was utilized. In 2011 Mann and co-workers pondered these questions and reported that a standard LC-MS analysis produced ∼100,000 peptide-like features and estimated that an MS/MS rate of 25 Hz was needed for their complete sampling (Michalski et al., 2011Michalski A. Cox J. Mann M. J. Proteome Res. 2011; 10: 1785-1793Crossref PubMed Scopus (476) Google Scholar). To evaluate this theoretical prediction, we measured the percentage of MS/MS scanning capacity utilized at various scan rates (Figure 1B). We reasoned that if insufficient precursors were available for sampling, then MS/MS scans would not be triggered and the cumulative number of acquired MS/MS scans would be lower than the maximum value achievable at a given scan rate. Overall we observed a strong correlation with reduced utilization of MS/MS capacity as scan rate increased, even at a very low dynamic exclusion setting of 5 s, which permits redundant sampling of many peptide-like precursors. These data both confirm the Mann prediction and explain our unexpected results. We therefore conclude that to exploit the capabilities of new MS technology and, ultimately, to continue improving single-shot proteomic analysis, we must increase the number of detectable peptide features. We believe that the most promising way to achieve this aim is through enhancing the quality of peptide separations. To survey the quality of single-shot proteomic separations in the published literature we obtained raw data and calculated chromatographic peak capacity (nc; Neue, 2005Neue U.D. J. Chromatogr. A. 2005; 1079: 153-161Crossref Scopus (310) Google Scholar) for many published proteomic experiments, including those shown in Figure 1A (refer to Table S1 for the complete list). These experiments had nc values ranging from 350 to over 1,000 (Figure S1), with the highest values drawn from the milestone works displayed in Figure 1A. By varying column length and particle diameter, we fabricated capillary columns that performed across this range (Table S2). Note our highest performing chromatography setup delivered separations with nc of ∼1,025, commensurate with the highest nc observed in the milestone single-shot articles (∼1,015). Using these columns, we analyzed a whole yeast proteome digest to characterize the dynamic interplay between MS/MS scan rate and separation quality. Figure 1C shows that as the scan rate increases so does sensitivity of the analysis to separation quality. For example, at lower MS/MS acquisition rates, high-quality separations provide little benefit. When collecting MS/MS spectra at <10 Hz, a cutting-edge rate just a few years ago, the significance of chromatographic quality is greatly diminished with doubling of peak capacity resulting in detection of virtually no additional peptides. The most recent quadrupole-ion trap-Orbitrap hybrid is, however, capable of >40 Hz MS/MS acquisition, and operation in this mode exhibits the strongest performance correlation with nc. The number of identified sequences increases by greater than 2-fold exclusively due to chromatographic improvements. Several benefits of the higher quality peptide separations help put these observations in perspective (Figure 1D). First, the narrower peptide elution profiles serve to greatly reduce redundant sampling. Second, as the chromatographic peak area remains relatively constant, a reduced peak width is accompanied with increased ion intensity, boosting sensitivity of the analysis. Further, because ionization suppression diminishes ion flux, reducing the number of co-eluting peptides increases dynamic range and lowers the occurrence of problematic “chimeric” spectra (Michalski et al., 2011Michalski A. Cox J. Mann M. J. Proteome Res. 2011; 10: 1785-1793Crossref PubMed Scopus (476) Google Scholar). Lastly, poor separations adversely impact the ability of the mass spectrometer to identify precursor monoisotopic peaks, obstructing their selection for a tandem scan (McAlister et al., 2014McAlister, G., Canterbury, J., Remes, P., Huguet, R., Eliuk, S., Zabrouskov, V., Senko, M., 2014. Exploring the uncharted depths of the complex human proteome using the Orbitrap Fusion mass spectrometer. 62nd Annual Meeting of the American Society of Mass Spectrometry and Allied Topics, Baltimore, MD.Google Scholar). In 2004, the field of liquid chromatography underwent a revolution: commercial ultra-high pressure pumps (UPLC, up to 10,000 PSI) became available and afforded use of fully porous sub-2 μm separation particles (Jorgenson, 2010Jorgenson J.W. Annu. Rev. Anal. Chem. (Palo Alto, Calif.). 2010; 3: 129-150Crossref Scopus (93) Google Scholar, Gritti and Guiochon, 2012Gritti F. Guiochon G. J. Chromatogr. A. 2012; 1228: 2-19Crossref Scopus (152) Google Scholar). Now over a decade old, this technology still remains at the core of proteomic analyses and was used by all aforementioned landmark single-shot proteome studies. Meanwhile, as we have discussed, considerable gains in MS acquisition rate have caught up with and outpaced the relatively constant front half of the LC-MS/MS experiment, so that only recently has the problematic mismatch between the MS and chromatographic performance fully emerged. The general importance of separations is recognized by other researchers. Many improved and alternative approaches have been introduced in the last years, including capillary electrophoresis (Fonslow and Yates, 2009Fonslow B.R. Yates J.R. Capillary electrophoresis applied to proteomic analysis.J. Sep. Sci. 2009; 32: 1175-1188Crossref Scopus (80) Google Scholar), superficially porous and sub-micron particles (Blue and Jorgenson, 2015Blue L.E. Jorgenson J.W. J. Chromatogr. A. 2015; 1380: 71-80Crossref Scopus (42) Google Scholar), use of slower flow rates (Köcher et al., 2014Köcher T. Pichler P. De Pra M. Rieux L. Swart R. Mechtler K. Proteomics. 2014; 14: 1999-2007Crossref Scopus (18) Google Scholar), HPLC systems with higher-pressure capabilities (i.e., >20,000 psi) (Shen et al., 2005Shen C. et al.Automated 20 kpsi RPLC-MS and MS/MS with chromatographic peak capacities of 1000-1500 and capabilities in proteomics and metabolomics.Anal. Chem. 2005; 77: 3090-3100Crossref Scopus (223) Google Scholar), and monolithic columns (Iwasaki et al., 2012Iwasaki M. Sugiyama N. Tanaka N. Ishihama Y. J. Chromatogr. A. 2012; 1228: 292-297Crossref PubMed Scopus (52) Google Scholar, Horie et al., 2014Horie K. Kamakura T. Ikegami T. Wakabayashi M. Kato T. Tanaka N. Ishihama Y. Anal. Chem. 2014; 86: 3817-3824Crossref Scopus (50) Google Scholar, Zhang et al., 2015Zhang Z. Sun L. Zhu G. Yan X. Dovichi N.J. Talanta. 2015; 138: 117-122Crossref Scopus (42) Google Scholar), among others. Due to a variety of drawbacks, such as robustness, load capacity, ease of implementation, and availability, these technologies have not been broadly adopted by the proteomics community or provided superior performance for single-shot analysis. We and others view the use of longer columns (Liu et al., 2007Liu H. Finch J.W. Lavallee M.J. Collamati R.A. Benevides C.C. Gebler J.C. J. Chromatogr. A. 2007; 1147: 30-36Crossref PubMed Scopus (64) Google Scholar, Fountain et al., 2009Fountain K.J. Neue U.D. Grumbach E.S. Diehl D.M. J. Chromatogr. A. 2009; 1216: 5979-5988Crossref Scopus (151) Google Scholar, Pirmoradian et al., 2013Pirmoradian M. Budamgunta H. Chingin K. Zhang B. Astorga-Wells J. Zubarev R.A. Mol. Cell. Proteomics. 2013; 12: 3330-3338Crossref PubMed Scopus (106) Google Scholar), packed with smaller particles, as the most direct path toward improved peptide separations. This, however, demands LC systems and fittings that can withstand very high pressures (Grinias et al., 2016Grinias K.M. Godinho J.M. Franklin E.G. Stobaugh J.T. Jorgenson J.W. J. Chromatogr. A. 2016; 1479: 60-67Crossref Scopus (33) Google Scholar), currently unavailable to non-specialists. Understanding the interplay between separations and MS sampling rate impacts proteomics at every level. When collecting data on older, slower systems, one need not be overly concerned with separation quality. However, those acquiring faster systems will best leverage their investment by careful implementation of state-of-the-art separations. In our opinion, quality of chromatographic separations is currently the key but underappreciated bottleneck limiting the speed and depth of single-shot proteomic analysis. Our hope is that researchers across multiple disciplines—i.e., separation scientists and proteomic practitioners—will become aware of this roadblock and employ their respective expertise to enable rapid and comprehensive analysis of human proteome at last. E.S. fabricated columns, prepared samples, and performed mass spectrometric analyses, data searches, and calculations. E.S. and A.S.H. designed the study and interpreted results. J.J.C. directed and supervised all aspects of the study. All authors contributed to writing and critical review of the manuscript. We thank Graeme McAlister and Mike Westphall for helpful discussions. We gratefully acknowledge support from NIGMS grants GM118110 and GM108538 (awarded to J.J.C.). Download .pdf (.32 MB) Help with pdf files Document S1. Supplemental Methods, Figure S1, and Tables S1 and S2