Analytical and Computational Tools — Microscopy, Spectroscopy, Assays, Bioinformatics and Systems Biology

UV–visible spectroscopy measures light absorbed by a sample. Beer–Lambert law, A = epsilon·c·l, makes absorbance A proportional to concentration c (epsilon the molar absorptivity, l the path length), the basis of quantitation. Nucleic acids absorb at 260 nm and proteins at 280 nm (from aromatic residues), so the A260/A280 ratio reports purity: about 1.8 for clean DNA and 2.0 for clean RNA, lower when protein contaminates. Fluorescence spectroscopy reads emitted light and is far more sensitive than absorbance. Circular dichroism (CD) measures the differential absorption of left- and right-circularly polarised light by chiral molecules; in the far-UV it reports protein secondary structure (alpha-helix, beta-sheet) and is a standard probe of folding.

Infrared (IR) and FT-IR spectroscopy probe molecular vibrations, so an IR spectrum is a fingerprint of the functional groups present (a protein’s amide I band reports its secondary structure); FT-IR collects the whole spectrum at once from an interferogram, which is faster and more sensitive. Mass spectrometry (MS) ionises molecules and separates the ions by their mass-to-charge ratio (m/z), giving accurate mass; soft ionisation — ESI (electrospray) and MALDI — made MS work for proteins and peptides, and tandem MS (MS/MS) fragments a peptide to read its sequence, the heart of proteomics. Nuclear magnetic resonance (NMR) uses the behaviour of nuclear spins (¹H, ¹³C, ¹⁵N) in a strong magnetic field; chemical shifts, couplings and nuclear Overhauser effects give atom-level structure and dynamics, and NMR can solve the three-dimensional structure of a protein in solution.

Worked example (Beer–Lambert). A DNA solution in a 1 cm cuvette gives A260 = 0.30. Using the convention that an A260 of 1.0 corresponds to about 50 micrograms per mL of double-stranded DNA, the concentration is 0.30 × 50 = 15 micrograms per mL. If this reading came from a tenfold dilution, the stock is 150 micrograms per mL. The A260/A280 ratio of the same sample tells whether that DNA is protein-free.

Electrophoresis separates charged molecules in a gel under an electric field. Agarose gel electrophoresis sorts nucleic acids by size — DNA is negatively charged and migrates to the anode, and small fragments move faster, so log(size) is roughly linear in migration distance; it is visualised with an intercalating dye. SDS-PAGE separates proteins by size because the detergent SDS coats every protein with a uniform negative charge and denatures it, so migration depends on mass alone; isoelectric focusing separates by pI, and 2-D gel electrophoresis combines IEF then SDS-PAGE to resolve thousands of proteins. Microarrays hold thousands of DNA probes in a grid; labelled sample hybridises to them, so a DNA microarray reads the expression of thousands of genes at once (now largely superseded by RNA-seq) and genotyping arrays read many SNPs.

Immunoassays use an antibody to detect and measure a specific antigen. In an ELISA (enzyme-linked immunosorbent assay) an antibody is linked to an enzyme that turns a substrate to a coloured product read in a plate reader; the sandwich ELISA captures the antigen between two antibodies for high specificity, and the indirect ELISA detects antibodies in a serum (as in antibody tests). RIA (radioimmunoassay) uses a radiolabel instead of an enzyme and is extremely sensitive. Immunohistochemistry (IHC) and immunofluorescence apply labelled antibodies to tissue sections to show where an antigen is. Immunoblotting (western blot) detects a protein separated by SDS-PAGE and transferred to a membrane. Flow cytometry passes single cells in a stream past a laser and reads their scatter and fluorescence, counting and sorting cells by surface or internal markers (FACS physically sorts them). Enzymatic assays quantify an enzyme by the rate at which it converts substrate (Section 4 of the paper develops the kinetics).

Sequencing-based methods read genomes and their interactions directly. Whole-genome sequencing reads an organism’s entire genome (NGS, Chapter 1) and underlies genomics, resequencing and variant detection. ChIP sequencing (ChIP-seq) finds where a protein binds DNA across the whole genome: chromatin is cross-linked, the DNA is sheared, the protein (a transcription factor or a modified histone) is pulled down with an antibody (chromatin immunoprecipitation), and the co-precipitated DNA is sequenced, so the binding sites appear as peaks. It is the genome-wide version of the DNA–protein interaction assays in Chapter 1.

Bioinformatics resources and databases. The primary sequence archives are GenBank (NCBI), ENA (EMBL-EBI) and DDBJ, which mirror each other; UniProt is the main protein-sequence database and the PDB holds experimentally determined 3D structures. Data move in standard file formats — FASTA (a header line beginning with ">" and the sequence), FASTQ (a read plus per-base quality scores, for NGS), GenBank flat files, and PDB/mmCIF for structures. Searching a sequence against a database is done with BLAST, which finds local similarities fast and reports an E-value — the number of hits of that score expected by chance in a database of that size — so a small E-value means a significant, probably homologous, match.

Sequence alignment lines up residues to reveal similarity. A pairwise alignment can be global (Needleman–Wunsch, end-to-end) or local (Smith–Waterman, the best matching sub-region); both use dynamic programming and a scoring scheme of match/mismatch scores and gap penalties. For proteins the match scores come from a substitution matrix — PAM (built from closely related proteins and extrapolated) or BLOSUM (built directly from blocks of aligned sequences; BLOSUM62 is the common default) — which scores a pair of amino acids by how often they substitute for each other in real alignments. A multiple sequence alignment (MSA) aligns many sequences at once (progressive methods such as Clustal and MUSCLE) and reveals conserved motifs and columns.

Phylogeny builds a tree of evolutionary relationships from aligned sequences. Distance methods (UPGMA, neighbour-joining) turn the alignment into a matrix of pairwise distances and cluster from it; character methods work from the columns directly — maximum parsimony chooses the tree needing the fewest changes, maximum likelihood and Bayesian methods choose the tree most probable under an explicit model of substitution. A tree’s branching is tested by bootstrap resampling, whose values say how often a clade recurs. The same molecular-evolution reasoning appears in the Genetics, Cellular and Molecular Biology section of this paper.

The omics disciplines study whole classes of molecule at once. Genomics is the study of the entire genome — its sequence, structure and variation; functional genomics and transcriptomics (RNA-seq) read which genes are active. Proteomics is the large-scale study of all proteins in a sample, dominated by mass spectrometry (Section 2) often after 2-D gel or liquid-chromatographic separation, and it reports identity, abundance and post-translational modification — information a genome cannot give because one gene makes many protein forms. Metabolomics profiles the small-molecule metabolites, using MS and NMR, and is the closest readout of a cell’s actual physiological state. Together these feed systems biology (Section 6).

Gene prediction finds the genes in a raw genome sequence. In prokaryotes it looks for open reading frames — stretches from a start codon to a stop codon long enough to be real — and for promoters and ribosome-binding sites. In eukaryotes it is harder because genes are split by introns, so predictors combine signals (splice sites, start/stop, codon-usage statistics) with evidence from aligned ESTs, cDNA and RNA-seq reads and from homology to known genes. Functional annotation then assigns a predicted gene a likely function — by homology (BLAST to characterised proteins), by conserved domains (Pfam, InterPro) and by controlled vocabularies such as Gene Ontology and pathway databases (KEGG).

Structure prediction infers a protein’s shape from its sequence. Secondary-structure prediction classifies each residue as helix, strand or coil. 3D (tertiary) structure prediction was long split between homology (comparative) modelling — building a model on a known structure of a related protein — and ab initio folding when no template exists; deep-learning methods have recently made accurate prediction routine for many proteins. Knowledge discovery in biochemical databases — data mining across sequence, structure, pathway and literature databases — turns these accumulated data into new hypotheses, for example predicting a protein’s function or an enzyme’s substrate from patterns across many entries.

Metagenomics sequences DNA taken directly from an environment — soil, gut, ocean, a bioreactor — without first culturing the organisms in it. Because most microbes cannot yet be grown in the lab, metagenomics is the way to survey the genes and species of a whole microbial community, to find novel enzymes and pathways, and to compare communities (for example a healthy against a diseased gut). It couples the sequencing of Chapter 1 with the bioinformatics of Section 4 (assembly, binning, taxonomic and functional annotation).

Metabolic engineering is the deliberate redesign of a cell’s metabolic network to improve or create the production of a compound — redirecting flux by overexpressing, deleting or introducing enzymes, removing feedback inhibition and balancing cofactors. Flux balance analysis (FBA) models the network at steady state (production equals consumption for every internal metabolite) and uses linear programming to find the flux distribution that maximises an objective such as growth or product yield, within the constraints of the stoichiometry. Engineered microbes made this way produce drugs, biofuels and chemicals (the artemisinic-acid yeast is a landmark). Systems biology is the larger programme: understanding how genes, proteins and metabolites behave as a network rather than as isolated parts, by combining omics data with computational models, so that the behaviour of the whole can be predicted and, in metabolic engineering, redesigned.