Proteome sequencing goes deep

Abstract

Advances in mass spectrometry have transformed the scope and impact of protein characterization efforts. Identifying hundreds of proteins from rather simple biological matrices, such as yeast, was a daunting task just a few decades ago. Now, expression of more than half of the estimated ~20,000 human protein coding genes can be confirmed in record time and from minute sample quantities. Access to proteomic information at such unprecedented depths has been fueled by strides in every stage of the shotgun proteomics workflow – from sample processing to data analysis – and promises to revolutionize our understanding of the causes and consequences of proteome variation.

Introduction

Spurred by the advent of the soft ionization methods, i.e., electrospray [1] and matrix-assisted laser desorption ionization [2], in the late 1980s, mass spectrometry (MS) has become the central method for protein analysis. Since this time, the depth and rate at which a proteome can be characterized has steadily improved so that today comprehensive analysis of most proteomes is within reach. The shotgun method, outlined in Figure 1, has proven the most useful tool for such applications. Here, proteins are extracted from lysed cells, enzymatically digested, and chromatographically separated prior to MS analysis. The MS records the masses of eluting peptide cations every second or so. In between these so-called MS¹ scans the system isolates selected peptide precursors, dissociates them using collisions or chemical reactions, and records the masses of the pieces (i.e., MS² or tandem MS). Modern MS systems can measure peptide masses accurately to thee decimal places while at the same time collecting tandem mass spectra at a blazing rate of 20 Hz. The hundreds of thousands of spectra generated from one of these experiments are then analyzed in silico using spectral matching algorithms.

Figure 1.

Workflow for “shotgun” or “bottom-up” proteomics. (a) Preparing proteomic samples for LC-MS/MS analysis requires protein extraction, proteolysis, and, optionally, peptide-level fractionation. (b) Online LC separation of complex peptide mixtures introduces analytes into the mass spectrometer for precursor and fragment ion mass analysis. (c) Tandem mass spectra are matched to theoretical spectra generated in silico to garner peptide sequences that are used for protein inference.

Mammalian proteomes are complex [3]. The human proteome contains ~20,300 protein-coding genes; however, non-synonymous single nucleotide polymorphisms (nsSNPs), alternative splicing events, and post-translational modifications (PTMs) all occur and exponentially increase the number of distinct proteoforms [4–6]. Detection of ~5,000 proteins in a proteomic experiment was a considerable achievement just a few years ago [7–9]. More recently, two groups identified over 10,000 protein groups in a single experiment. Through extensive protein and peptide fractionation (72 fractions) and digestion with multiple enzymes, Nagaraj et al. identified 10,255 protein groups from HeLa cells over 288 hours of instrument analysis [10•]. A comparison with paired RNA-Seq data revealed nearly complete overlap between the detected proteins and the expressed transcripts. In that same year, a similar strategy enabled the identification of 10,006 proteins from the U2OS cell line [11•].

A more comprehensive analysis of the human proteome can be achieved by applying similar technologies to large-scale comparisons of multiple cell lines and tissues [12,13,14•,15•]. Kim and co-workers analyzed 30 human tissues and primary cells over 2,000 LC-MS/MS experiments, resulting in the detection of 293,000 peptides with unique amino acid sequences and evidence for 17,294 gene products [16••]. Wilhelm et al. amassed a total of 16,857 LC-MS/MS experiments from human cell lines, tissues, and body fluids. These experiments produced a total of 946,000 unique peptides, which map to 18,097 protein-coding genes [17••]. Together, these two studies provide direct evidence for protein translation of over 90% of human genes (Figure 2). Despite providing the deepest coverage to date, the latter study required non-stop operation of a mass spectrometer for four straight years! New developments in mass spectrometer technology have increased the rate at which proteomes can be analyzed. Using such a device, we recently described a method that characterizes nearly every protein in yeast in just over one hour (4,000 of the 4,500 expressed yeast proteins) [18••]. In this review, we describe developments in sample preparation, MS instrumentation, and bioinformatics that have been key to obtaining comprehensive proteomic coverage. Further, we consider how access to such proteomic detail will impact genomic research.

Figure 2.

Chromosomal coverage of the human proteome (reproduced with permission from ref. [17]). In one of two recent large-scale investigations of the human proteome, Wilhelm and coworkers identified 18,097 proteins, covering over 90% of all but three chromosomes (11, 21, and Y). The density of proteins covered in any particular chromosomal region is indicated by the blue bars.

Advances in proteomic sample preparation

For any proteomic method, proteins must first be liberated from their host cells, via mechanical and/or chemical disruption, often into a denaturing solution. Reduction of disulfide bonds and alkylation of cysteine residues disrupts protein structure, leaving proteins amenable to site-specific cleavage with one or more proteases. This initial step – protein extraction and solubilization – is paramount, as it dictates which proteins will be accessible for eventual MS detection. Strong detergents, such as sodium dodecyl sulfate (SDS), are exceptional denaturants, but their removal, a requirement for efficient proteolytic digestion and sensitive mass-spectrometric analysis, is challenging. Standard filtration devices can be employed as proteomic reactors (filter-aided sample preparation, FASP), allowing dissolution of proteins in high concentrations of SDS which are then depleted before digestion [19]. Alternatively, an unbiased proteomic characterization can be achieved without SDS by digesting unclarified cellular lysate, a tactic that improves coverage of proteins harbored in poorly soluble membrane and nuclear organelles [18••,20,21••].

The maximum coverage obtainable for a protein is theoretically determined by its amino acid sequence and the cleavage specificity of the chosen proteolytic enzyme, typically trypsin. A straightforward and long-recognized approach for boosting protein and proteome coverage is to digest a sample separately with multiple proteases [22]. Recently, digestion with α-lytic proteases, semi-specific enzymes that preferentially cleave after aliphatic residues, increased trypsin-only protein identifications and sequence coverage by 24% and 101%, respectively [23]. Some downsides of such multi-protease strategies include heightened sample quantity and analysis time demands. Digesting with multiple enzymes sequentially instead of in parallel, however, can afford better coverage without the extra requirements [24,25].

Even following efficient solubilization and proteolysis, many proteins are only represented following detection of a few unique peptides [6]. This mainly stems signal suppression during the electrospray ionization – that is, peptides having higher overall basicity tend to preferentially ionize rendering the more acidic peptides undetected. The most successful approach to curb this problem is to reduce the number of unique peptide sequences present in the ionization source at one time. To this end, separation chemistries and implementations thereof are central to proteomic analysis. The extreme separation resolution provided by some of these platforms, such as the automated coupling of three physicochemically orthogonal stages of chromatography [26] and high-resolution isoelectric focusing [27••], is key to achieving genome-scale coverage of the proteome.

Overall, recent developments in sample processing for shotgun proteomics have emphasized simplification and scalability [21••]. Furthermore, a robust workflow with minimal sample loss and contamination opens the door for applications with limited starting material. Using current technology, 9,500 proteins can be identified from just 100 nL of formalin-fixed and paraffin-embedded (FFPE) tissue [28•]. Although it is impossible to ionize and sequence every peptide, efficient sample preparation, coupled with advances in MS instrumentation [29–31], separation methodology [32–35], and fragmentation techniques [36–39] have vastly increased the observable portion of the human proteome.

Advances in peptide separation and MS instrumentation

Modern hybrid mass spectrometers couple highly accurate MS1 scans with ultra-fast MS/MS sequencing rates. A recent study used a linear ion trap (LIT)-Orbitrap hybrid mass spectrometer [40], which, compared to the previous generation instrument, achieves approximately twice the resolving power at the same scan speed, to analyze eleven human cell lines [14•]. Across all cell lines, 11,731 proteins were identified, with an average of 10,361 proteins identified per cell line. The number of identified protein groups is comparable to a previous study from HeLa lysate [10•], but is generated in a fraction of the time (3 versus ~ 12 days). A newly released mass spectrometer combining a quadrupole mass filter, a collision cell, a dual cell LIT and an Orbitrap mass analyzer, operates at a maximum MS/MS acquisition speed of 20 Hz [41], doubling the number of tryptic yeast peptides identified per second as compared with the Orbitrap Elite (19 versus 10 peptides/second) [18••]. Ion mobility coupled with MS has also been explored as an option for decreasing sample complexity and improving identification efficiency [42]. Traveling wave ion mobility spectrometry (TWIMS) significantly improved the duty-cycle of a time-of-flight (TOF) instrument, identifying ~7,500 protein groups from HeLa cells in one day of analysis time [31].

The sequencing speed of modern mass spectrometers is best harnessed when coupled to efficient, online peptide separation. HPLC systems operating at high pressures (>8,000 psi) [33] and longer columns packed with small particles have become standard (< 2 μm) [18••,34]. The linear relationship between the number of identified peptides and peak capacity, the number of resolvable peaks across an elution, has been demonstrated [32]. Many recent workflows have focused on optimizing chromatographic separations rather than extensive fractionation for whole proteome analysis [43]. Forgoing sample pre-fractionation in favor of long columns packed with 2 μm particles, a recent study identified 4,825 protein groups from the A375 cancer cell line over a three hour LC-MS/MS experiment [44]. A comparison of column lengths revealed that, for this particular instrument platform, a 50 cm column allowed identification of more proteins than either a 15 or 25 cm column at all gradient lengths tested, although a decrease in cumulative identified protein groups after three hours was reported. Note that the combination of long columns and small particles significantly raises column backpressure, necessitating either a UHPLC system capable of operating at pressures > 10,000 psi, or a column heater. Silica monolithic columns, which can achieve separation efficiencies similar to traditional packed columns without a substantial increase in back pressure, have also been used [35].

Advances in computational proteomics

The scale of proteomic data generated by streamlined sample processing pipelines and high-resolution mass spectrometry now approaches that of analogous genomic and transcriptomic technologies. Given that proteins more closely resemble phenotype than their encoding nucleic acid counterparts, they harbor unique biological details that can inform larger biological processes.

For organisms with sequenced genomes, peptides detected by mass spectrometry can assist in refining prediction-based gene annotations, a primary goal of the emergent proteogenomics field [45]. In addition to validating predicted genes, deep proteomic coverage can suggest novel protein-coding loci [46], N-terminal signal peptides [47], splice sites [48], and nonsynonymous variants [49]. A long-term objective of proteogenomic mapping is to associate certain variations in protein sequence with disease states. One recent study combined a customized protein database with in-depth transcriptome and proteome profiling of livers from two inbred rat strains [50•]. Interestingly, the results associated a genomic variant in the promoter region of a mis-annotated gene with the observed hypertensive phenotype of one strain, illustrating the advantages of such integrated approaches. Proteogenomic endeavors match tandem mass spectra to a database containing, ideally, all possible protein sequences encoded by an individual genome. This poses a computational challenge for large genomes with low protein-coding content, requiring extensive search-space reductions to boost sensitivity. A fresh strategy enabled unbiased proteogenomic mapping against the full human genome, along with deep proteome coverage, by blending isoelectric focusing for high-resolution peptide separation with accurate isoelectric point prediction for rational reduction of the search space [27••].

Systems-level analyses have also benefitted from the growing robustness and availability of informatics tools for label-free quantification strategies. Even highly fractionated proteomes can now be accurately compared in the absence of stable isotope labels [51]. Aided by highly accurate mass measurements, the confident transfer of peptide identifications between matching runs provides a 25–30% boost in the number of proteins quantified across multiple samples [14]. This feature makes label-free approaches very attractive for deep proteome quantification, though stable isotope labeling strategies are still more straightforward for the comparative analysis of low-abundance PTMs [52]. Furthermore, statistical analysis of signatures at the peptide-level can reveal information regarding the presence and expression patterns of one or more proteomes, an approach that will be greatly empowered by high protein sequence coverage [53].

Conclusions and outlook

Breakthroughs in every stage of the shotgun proteomics workflow have collectively ushered in a new era of proteomics, one in which identification and quantification of complete proteomes can be routinely achieved [54]. Beyond propelling basic research, this age holds great potential for personalized medicine [55]. Earlier this year, two independent efforts reported evidence of protein translation for 90–95% human genes, an impressive display of progressing technologies for proteome characterization [16••,17••]. As deep cataloguing of protein expression becomes widespread, the spotlight will shift to extensive functional mapping of proteoforms and determining how their expression is regulated by genomic elements [56]. To this end, the complementary benefits of top-down [57], targeted [58], and antibody-based [59] approaches must be harnessed and effectively integrated. Finally, in light of the surging trend of deep proteomics, it is important to remember that, for some systems, meaningful biological insight can still be drawn from moderate depths of proteome coverage [15•], which are becoming ever more accessible to proteomic researchers of all experience levels.

• Recent advances in mass spectrometry have transformed the depth of coverage of the human proteome

• Expression of over half of the estimated human protein coding genes can be verified in record time

• Here, we highlight the impact of advances in mass spectrometry sample preparation, instrumentation, and data analysis on the characterization of the human proteome