Advancedadvanced biology

Proteomics: Measuring the Protein Complement at Scale

Learning objectives

  • Explain the bottom-up proteomics workflow from protein extraction to peptide identification.
  • Compare label-free, SILAC, and TMT quantification strategies.
  • Describe how post-translational modifications are detected and why they matter.
  • Contrast top-down and bottom-up approaches and their trade-offs.
  • Evaluate interaction proteomics and emerging single-cell methods.

The bottom-up workflow: from proteins to spectra

Proteomics measures what genes only promise: proteins are the cell's actors, and their abundances correlate imperfectly with mRNA (typical correlations around 0.4-0.6) because of translation control and regulated degradation. The dominant strategy is bottom-up: proteins are extracted, digested with trypsin into peptides, separated by liquid chromatography, ionized, and fragmented inside a tandem mass spectrometer (LC-MS/MS). Each peptide's fragmentation spectrum is a fingerprint matched against theoretical spectra from protein databases, identifying the peptide and inferring its parent protein. The genius of the approach is that peptides -- small, uniformly behaved -- are far easier to separate and fragment than intact proteins. Modern instruments identify 5,000-10,000 proteins from a mammalian cell line in a single run. False discovery rates are controlled by target-decoy searching: spectra are also matched against reversed sequences, and identifications are thresholded so decoy hits stay below 1%. The workflow's limitation is inference: shared peptides ambiguate which protein isoform was present, the protein inference problem.

Quantification: label-free, SILAC, and TMT

Identification is half the battle; quantification -- how much of each protein -- drives biology. Label-free quantification compares peptide ion intensities or spectral counts across runs: simple and unlimited in sample number, but vulnerable to run-to-run variation, mitigated by match-between-runs algorithms and data-independent acquisition (DIA/SWATH), which systematically fragments everything rather than selecting peaks. SILAC (stable isotope labeling by amino acids in cell culture) grows cells in heavy-isotope media, mixing samples before processing so ratios are measured within one run -- precise but limited to culturable cells. TMT (tandem mass tags) chemically labels peptides from up to 18 samples with isobaric tags that release reporter ions upon fragmentation, enabling multiplexed comparison -- powerful for clinical cohorts but suffering ratio compression, where co-isolated peptides dilute measured fold-changes. The choice depends on the question: deep precise ratios for mechanism (SILAC), large cohorts for biomarkers (TMT), or broad surveys (label-free DIA). Each trades accuracy, throughput, and cost differently.

Post-translational modifications: the real regulatory layer

Much of cellular regulation happens after translation: phosphorylation, ubiquitination, acetylation, glycosylation, and hundreds of rarer modifications. Phosphoproteomics -- enriching phosphopeptides with titanium dioxide or IMAC before MS -- routinely quantifies 10,000+ phosphorylation sites, mapping signaling networks in action: EGF stimulation time courses revealed the order of pathway activation directly at the protein level. Ubiquitinomics captures the degradation code, distinguishing K48 linkages (proteasomal destruction) from K63 (signaling scaffolds). The analytical challenge is combinatorial explosion: a protein with n modifiable sites has 2^n possible states, and MS typically measures site occupancy, not full proteoforms. Top-down proteomics addresses this by fragmenting intact proteins, preserving modification combinations -- technically demanding but conceptually the right level, since biology acts on proteoforms, not average site occupancies. Modified peptides also fragment differently, requiring specialized search strategies; the field's dirty secret is that many reported modification sites rest on single spectra demanding manual validation.

Interaction proteomics: mapping the social network

Proteins work in complexes, and proteomics maps these social networks. Affinity purification-MS (AP-MS) pulls down a bait protein and identifies co-purifying partners -- powerful but biased toward stable complexes and plagued by contaminants, controlled with quantitative scoring (SAINT, CompPASS) against negative controls. Proximity labeling (BioID, TurboID, APEX) instead tags neighbors in living cells: a fused enzyme biotinylates proteins within nanometers over minutes, capturing transient interactions and mapping spatial proteomes -- organelle inventories built by targeting the label to mitochondria, ER, or synapses. Cross-linking MS freezes interactions with chemical linkers, even yielding distance restraints for structural modeling. These approaches built the first draft human interactomes (BioPlex) and revealed that most proteins moonlight in multiple complexes depending on context. The emerging view is dynamic: interactions assemble and dissolve with signaling state, cell cycle, and stress, so static maps are giving way to condition-dependent networks.

Frontiers: single cells, spatial, and clinical proteomics

Proteomics is pushing toward the scales genomics already reached. Single-cell proteomics, using isobaric carriers (SCoPE-MS) or ultra-sensitive DIA, now quantifies 1,000-3,000 proteins per cell -- fewer than transcriptomics but measuring the molecules that actually act, revealing post-transcriptional regulation invisible to RNA. Spatial proteomics (CODEX, MIBI, Deep Visual Proteomics) maps proteins within tissue architecture, critical for tumors where location predicts behavior. Plasma proteomics hunts biomarkers in blood, where the dynamic range challenge is extreme: albumin outweighs cytokines by ten orders of magnitude, requiring depletion or nanoparticle enrichment. Clinically, proteomics is closest to phenotype -- protein levels and modifications are what drugs target -- and proteogenomics personalizes databases with patient-specific variants. Open problems remain formidable: complete proteoform characterization, routine single-cell depth matching bulk, and the computational integration of proteomics with other omics. The field's trajectory mirrors genomics a decade earlier: from catalogs to mechanisms to medicine.

Why proteins, not just RNA

mRNA levels predict protein levels only moderately (r ~ 0.4-0.6). Translation rates, protein half-lives, and modifications -- all invisible to RNA sequencing -- often dominate the actual functional state of a cell.

Key points

  • Bottom-up LC-MS/MS digests proteins to peptides, fragments them, and matches spectra to databases at 1% FDR.
  • Quantification strategies (label-free DIA, SILAC, TMT) trade precision, multiplexing, and sample compatibility.
  • Phosphoproteomics maps signaling; ubiquitinomics decodes degradation; proteoforms are the true functional units.
  • AP-MS finds stable complexes; proximity labeling (TurboID/APEX) captures transient, spatial interactions.
  • Single-cell and spatial proteomics are emerging; plasma proteomics faces extreme dynamic range.
  • Protein measurements sit closest to phenotype, making proteomics the omics layer drugs actually target.

Common misconceptions

✗ Myth: More mRNA means more protein.

✓ Truth: Correlation is only moderate. Translation efficiency varies 100-fold between transcripts, protein half-lives range from minutes to days, and regulated degradation decouples the two -- which is exactly why proteomics is necessary.

✗ Myth: Proteomics identifies proteins directly.

✓ Truth: Bottom-up proteomics identifies peptides and infers proteins. Shared peptides create the protein inference problem: the data often cannot distinguish which isoform or family member was present.

✗ Myth: A list of interacting proteins is a pathway.

✓ Truth: AP-MS hits include contaminants and indirect associations; scoring against controls is essential. Even validated interactions need functional follow-up -- physical association does not prove regulatory relationship.

Frequently asked questions

What is the difference between DDA and DIA mass spectrometry?

Data-dependent acquisition (DDA) selects the most intense peaks for fragmentation, giving deep but stochastic coverage with missing values across runs. Data-independent acquisition (DIA/SWATH) systematically fragments all peptides in windows, yielding reproducible quantification across many samples at the cost of more complex analysis.

Why is phosphoproteomics so informative about signaling?

Because phosphorylation is signaling's currency -- kinases act in seconds, far faster than transcription. Measuring thousands of phosphorylation sites over a stimulation time course reveals pathway order, feedback, and cross-talk directly at the level where regulation happens.

What limits single-cell proteomics compared to single-cell RNA-seq?

No amplification: proteins cannot be PCR-amplified, so sensitivity is bounded by instrument physics. Current methods reach 1,000-3,000 proteins per cell versus whole transcriptomes by RNA-seq, though they measure the more functionally relevant molecules.

What is ratio compression in TMT experiments?

When the mass spectrometer co-isolates background peptides with the target, their reporter ions dilute the measured fold-change toward 1:1. Synchronous precursor selection (SPS-MS3) reduces this by adding a purification fragmentation step before reporter quantification.