LSMA

LONDON SCHOOL OF MULTIOMICS & AI

Free course

Free course

0 of 73 lessons

Lesson 9 of 10

Module 1 · Understanding omics

Proteomics and metabolomics

Proteomics and metabolomics

Proteomics and metabolomics

Proteins do the work and metabolites are the chemistry they run. These are the layers closest to what a cell is actually doing, and most drugs act on them.

⏱ 11 min readModule 1 · Understanding omicsFree
Lesson at a glance3 steps · about 11 min
  1. Why measure proteins directly
  2. How mass spectrometry reads proteins
  3. Metabolomics

Why measure proteins directly

Across genes, RNA and protein levels correlate only moderately. Proteins are made and broken down at different rates, and many are switched on or off by post-translational modifications such as phosphorylation, which no RNA measurement can show.

"Correlate only moderately" means that knowing a gene's RNA level gives you a rough guide to its protein level, not a reliable one. Two genes with the same amount of RNA can end up with very different amounts of protein. One protein may last for days once it is made, while another is destroyed within minutes. The cell can also slow down or speed up translation of a particular RNA.

A post-translational modification is a chemical change made to a protein after it has been built. Phosphorylation, the most studied, adds a phosphate group to particular amino acids. The enzymes that add phosphates are called kinases, and many cell signals work as chains of kinases, each switching on the next by phosphorylating it. Many cancer drugs are kinase inhibitors. Whether a kinase is switched on at a given moment is invisible in its RNA; you have to look at the protein and its phosphate groups.

In Bulk transcriptomics you turned RNA into a count table and asked which genes changed. This lesson asks the next question: did those changes reach the proteins, which do most of the work in a cell? There is a practical reason too. Most drugs act on proteins, and most clinical tests that guide treatment measure proteins, so a protein-level result is often closer to a decision than an RNA one.

A widely cited large-scale comparison of RNA and protein is Schwanhäusser and colleagues (2011, Nature), measured in mouse cells. The review by Liu, Beyer and Aebersold (2016, Cell) sets out when RNA predicts protein well and when it does not.

Be careful which correlation a paper reports. Comparing many genes within one sample (are high-RNA genes also high-protein genes?) usually gives a higher correlation than following one gene across many samples (does a rise in its RNA bring a rise in its protein?). Measurement noise in either layer also lowers the figure. So quote the type of comparison, not just the number.

How mass spectrometry reads proteins

A mass spectrometer measures the mass of charged molecules with great precision. Whole proteins are large and awkward to measure, so the usual approach, sometimes called bottom-up proteomics, cuts them into short pieces first:

  1. Digest proteins into peptides, usually with trypsin, which cuts after lysine (K) and arginine (R), except when proline (P) comes next.
  2. Separate the peptides over time by liquid chromatography.
  3. Measure each peptide's mass-to-charge ratio (m/z), then break it and measure the fragments.
  4. Identify it by matching fragment patterns to a database of protein sequences.

What each step does

  • Digest. A peptide is a short chain of amino acids. Trypsin is a digestive enzyme that cuts protein chains at predictable places. Because the cut sites are known, the computer can work out in advance every peptide that each known protein should produce.
  • Separate. A digested sample contains many thousands of different peptides, far too many to measure at once. In liquid chromatography the mixture is pumped through a thin column packed with material that peptides stick to. Each kind of peptide sticks with its own strength, so they come out at different times, spread over an hour or more.
  • Measure. As peptides leave the column they are given an electrical charge and enter the instrument, which measures their mass-to-charge ratio. It then selects a peptide, breaks it into fragments and measures those too. Measuring twice like this is called tandem mass spectrometry, or MS/MS.
  • Identify. The fragments of a peptide form a pattern that depends on its sequence of amino acids. Software compares each observed pattern with the patterns predicted for every peptide in the database and reports the best match, with a false discovery rate, as for genes in Bulk transcriptomics.

Identifying peptides is only half the job; the other half is quantification, working out how much of each protein is present. There are two broad approaches. In label-free quantification, each sample is run separately and the signal strength of the same peptide is compared across runs. In labelled quantification, peptides from several samples are tagged with chemical labels of slightly different mass, mixed and measured in one run, so the samples are compared directly. Labelling reduces run-to-run differences but limits how many samples fit in one batch.

On real data, the result looks much like an RNA count matrix: one row per protein, one column per sample, with a measured intensity in each cell. A common pitfall is how to treat empty cells. A peptide too faint to be picked for measurement in one run may simply be missed there, so a missing value often means "low or not detected", not "absent". How missing values are handled can change which proteins appear to differ, so a paper should say what it did.

Phosphorylation, the switch described above, needs an extra step. Phosphorylated peptides are a small share of all peptides, so they are first enriched, pulled out of the mixture with materials that bind phosphate groups, before measurement. This branch of the field is called phosphoproteomics, and it is how kinase activity is read across thousands of sites at once.

Proteomics by mass spectrometry
Follow one protein across the two rows: it is digested into peptides, the peptides are separated by liquid chromatography, ionised, and weighed as mass-to-charge ratios, and the resulting spectrum is matched to a known protein. Schematic, not to scale. Created in BioRender.com.

Worked example · mass-to-charge

peptide mass 1,500 Da, charge 2+ (each proton adds about 1.007 Da)
m/z = (1,500 + 2 × 1.007) ÷ 2 = about 751.0

The instrument sees this peptide at 751, not 1,500.

Step by step. A dalton (Da) is the unit of molecular mass, close to the mass of one hydrogen atom. The peptide picks up its charge by gaining protons, and each proton adds its own small mass. With 2 protons: 2 × 1.007 = 2.014 Da. Add that to the peptide: 1,500 + 2.014 = 1,502.014 Da. The instrument measures mass divided by charge, so divide by 2: 1,502.014 ÷ 2 = 751.007, about 751.0.

What the answer means. The instrument never measures mass on its own, only mass divided by charge. The same peptide carrying 3 protons would appear at (1,500 + 3 × 1.007) ÷ 3 = 1,503.021 ÷ 3, about 501.0. Software works out each peptide's charge from the spacing of its peaks and converts back to the true mass before searching the database. If you see the same peptide reported at two different m/z values, this is usually why.

In the clinic, protein is often measured with antibodies instead: immunohistochemistry (IHC) for HER2 or PD-L1 is protein measurement on a tissue slide.

An antibody is a protein of the immune system that binds one particular target very tightly. In immunohistochemistry, an antibody against the protein of interest is applied to a thin slice of tumour on a glass slide, and a chemical reaction makes the bound antibody visible as a coloured stain, usually brown. A pathologist then judges under the microscope how many cells are stained and how strongly. IHC measures one or a few proteins, but it shows exactly which cells carry them, and it can be done on the stored tissue blocks that hospitals already keep. Mass spectrometry measures thousands of proteins at once, but from ground-up tissue, without positions.

IHC has its own pitfalls. Scoring is partly a judgement, and results depend on the antibody, the staining protocol and how long the tissue sat before it was fixed. For this reason, clinical scores use fixed categories. HER2 in breast cancer, for example, is scored from 0 to 3+, and an uncertain 2+ result is usually checked with a second test that counts copies of the HER2 gene. That second test reads the genome, a reminder that the layers are often used together.

Go deeper: which protein did that peptide come from?

Identifying a peptide is not the same as identifying a protein. Many peptides occur in more than one protein: in isoforms of the same gene, or in closely related genes from an old duplication. If every peptide you saw is shared, the data cannot say which of the candidates was really there. This is called the protein inference problem, and software handles it by reporting protein groups: sets of proteins that the evidence cannot tell apart.

The false discovery rate for peptide matches is usually estimated with a decoy trick. The software also searches a fake database of reversed or shuffled protein sequences, which cannot truly be in the sample. Any match to a decoy is wrong by design, so counting decoy hits estimates how many real-looking hits are wrong too. Elias and Gygi set out this target-decoy approach in 2007.

Metabolomics

Metabolites are small molecules such as sugars, amino acids and lipids. They are measured by mass spectrometry or nuclear magnetic resonance (NMR). Cancer rewires metabolism in characteristic ways:

  • Many tumours turn glucose into lactate even when oxygen is plentiful, the Warburg effect. Their high glucose uptake is what FDG-PET scans detect.
  • Mutations in the enzymes IDH1 or IDH2 produce a new metabolite, 2-hydroxyglutarate, which builds up to very high levels and changes the epigenome. You will follow this story in the case study.

Making sense of the two examples

Lipids are fats and fat-like molecules, such as those that make up cell membranes. The metabolome is the whole set of small molecules in a sample, and it changes within minutes, for example after a meal or a drug (The central dogma, as data). Studies are either targeted, measuring a chosen list of known molecules accurately, or untargeted, detecting as many signals as possible and identifying them afterwards.

In untargeted work, identification is the hard part. Unlike peptides, metabolites are not built from a fixed alphabet, so there is no sequence database to predict them from. Signals are matched against libraries of measured standards, and molecules with the same mass but a different structure can be hard to tell apart. It is normal for a large share of detected signals to remain unidentified, and a careful paper says how confident each identification is.

Because the metabolome changes so fast, handling matters even more than for RNA. Whether the patient had fasted, how long a sample waited before freezing, and how many times it was thawed can all shift metabolite levels. Studies that compare groups need these steps to be the same for every sample.

Normally, a cell with enough oxygen breaks glucose down completely, which releases far more energy per molecule than turning it into lactate. Tumours that make lactate anyway take up glucose very fast to compensate, and the pathway also supplies building blocks for new cells. In an FDG-PET scan, the patient is injected with fluorodeoxyglucose (FDG), a glucose look-alike carrying a weakly radioactive label. Cells take it up like glucose but cannot break it down fully, so it stays inside them, and the scanner shows where it has gathered. Tumours with high glucose uptake light up. So does the brain, which normally uses a great deal of glucose, which is one reason the scan is read by a specialist.

The IDH example is unusual: a single change in DNA gives an enzyme a new job, its product reaches very high levels, and that product changes the epigenome. It links three layers, and the case study at the end of this module asks you to trace it.

PC1 (41% variance)PC2 (17% variance)NormalSubtype ASubtype B
Samples placed by their overall protein or metabolite profiles. Two tumour subtypes separate from normal tissue. Illustrative data.

This kind of plot uses principal component analysis (PCA). Each sample is described by thousands of measurements; PCA finds the few directions along which samples differ most and places each sample on the first two. The first axis, PC1, carries the most variation, and the second, PC2, the next most. Samples that sit close together have similar overall profiles. PCA is not told which samples are tumour, so groups that separate on their own are a sign of a strong, consistent difference. It is also the first place a batch effect (Bulk transcriptomics) tends to show up, as samples grouping by processing date instead of biology.

One caution when reading such a plot: PCA shows the largest sources of variation, whatever they are. If one technical factor, such as sample storage time, is larger than the biology, it will dominate PC1. Values are usually log-transformed and scaled first, so that a few very abundant proteins or metabolites do not decide the picture on their own.

Key point

An RNA level says a protein could be made. Protein measurements say it is there, and measuring its phosphorylation shows whether it is active; metabolite measurements show what it is doing.

A second example: a kinase inhibitor can switch a kinase off within minutes of reaching the cell. The kinase's RNA does not change, and neither does the amount of the kinase protein. What changes is the phosphorylation of the proteins it acts on, and then the metabolites those pathways produce. An experiment that measured only RNA after a short treatment could wrongly conclude that the drug had done nothing.

Further reading · Module 1Sources for this module and the rest of the course are gathered on one page. · All reading for this course

Check your understanding

1. Trypsin cuts proteins after which amino acids?

2. Why can't protein activity be read off RNA-seq?

3. What is the Warburg effect?

Up next

Case study

A tumour across the layers

Ask BioTutor

Create a free website with Framer, the website builder loved by startups, designers and agencies.