A protein is a large biological molecule made of one or more chains of amino acids linked by peptide bonds. The sequence of those amino acids is specified by genes, and the chain folds into a defined three-dimensional structure that determines what the protein does. Proteins are the most functionally diverse class of macromolecule in cells: they act as enzymes that catalyze chemical reactions, as structural and motor components, as transporters, hormones and receptors, and as antibodies in the immune system. Because almost every biological process depends on them, the study of proteins links biochemistry, molecular-biology, medicine and biotechnology.
This entry explains proteins from the chemical level upward. It covers the amino acids and the peptide bond, the four levels of protein structure, and the folding process that connects sequence to shape. It then describes how proteins are made through translation and modified afterward, and surveys their major functional classes. The later sections address how structures are determined and predicted, how proteins are measured at scale, their role in nutrition and metabolism, the history of their discovery, their evolution, their applications, and unresolved questions such as misfolding and intrinsically disordered regions.
Definition and Classification
A protein is a macromolecule built from one or more unbranched chains of amino acids joined by peptide bonds. The words peptide, polypeptide and protein are related but not interchangeable. A peptide is a short chain, conventionally a few dozen residues or fewer. A polypeptide is any single chain regardless of length. "Protein" is usually reserved for a polypeptide, or an assembly of polypeptides, that has folded into a defined and functional form. Some proteins are called simple proteins because they yield only amino acids on complete hydrolysis. Others are conjugated proteins, which carry a non-amino-acid component that is often essential to their function. Glycoproteins carry covalently attached carbohydrate, and many secreted and membrane proteins are glycosylated. Lipoproteins, such as the apolipoproteins that transport lipids in blood plasma, carry lipid components. Metalloproteins bind metal ions. Hemoglobin is a related case: its iron-containing heme group is a non-protein part that is required for oxygen binding.
Protein sizes span a very wide range. Small proteins and peptide hormones contain tens of residues, while a typical cellular protein has several hundred. At the upper end, titin, the largest known human protein, has roughly 34,000 residues. Many functional units are multi-subunit complexes. RNA polymerase II has twelve subunits in its core form, and the 26S proteasome is built from dozens of distinct subunits.
Counting Genes and Proteins
Estimates of the number of human protein-coding genes have fallen steadily since the genome was first sequenced. With the publication of the final draft of the Human Genome Project, the number of protein-coding genes was revised downwards again to between 20 000 and 25 000. Current reference catalogues cluster near 20,000 but do not fully agree. Manual annotation teams at both the National Center for Biotechnology Information (NCBI) and Ensembl have curated the list to arrive at about 20,500 protein-coding genes. Other annotation resources give somewhat different totals. Gencode, Ensembl and RefSeq do not agree on the core number of genes, with only about 19,000 common to all three databases.
The number of protein molecules is a different quantity from the number of genes. Alternative splicing allows one gene to produce several mRNA variants, and post-translational modifications add further diversity, so more proteins are found in humans than the gene count would suggest. The total therefore depends on what is counted: individual splice isoforms, chemically modified forms (often called proteoforms), or both. Counting conventions also matter for small proteins. Proteogenomics studies have predicted thousands of novel protein-coding regions, including low-molecular-weight proteins encoded by small open reading frames in untranslated regions of messenger RNAs and in non-coding RNAs. Whether these predictions are included changes the tally, and most of them rely on bioinformatic analysis and ribosome footprint data rather than direct observation of the proteins.
Amino Acids and the Peptide Bond
The twenty standard residues
Roughly 500 amino acids occur in nature, but only twenty are found in protein molecules in humans. These twenty are the residues the genetic code specifies, and all the standard amino acids are L-α-amino acids. Each shares one backbone pattern: a central α-carbon bonded to an amino group, a carboxyl group, a hydrogen atom and a variable side chain (R group). Because the α-carbon is chiral in all but one case, nineteen of the twenty have a chiral α carbon. The exception is glycine, whose side chain is a single hydrogen.
Side-chain chemistry is the usual basis for grouping. The IMGT classification places alanine, cysteine, glycine, isoleucine, leucine, methionine, phenylalanine, proline, tryptophan and valine in the nonpolar class. The polar class contains both uncharged residues (asparagine, glutamine, serine, threonine, tyrosine) and charged ones (arginine, histidine, lysine, aspartate, glutamate). Some assignments are subtle. Methionine's thioether is somewhat polar but cannot form hydrogen-bonding interactions with water molecules, so its side chain is also classified as hydrophobic. Tryptophan has an N–H group that can hydrogen-bond to water, but the tryptophan side chain, as a whole, is still classified as hydrophobic.
Two residues are special cases. Cysteine's sulfhydryl group allows it to form disulfide bridges with other Cys residues, which is the basis of disulfide bonds. Proline differs because its side chain loops back onto the backbone nitrogen, making it an imino acid, and this rigid ring affects chain flexibility throughout the protein.
Acid–base behavior and the isoelectric point
Amino acids are amphiprotic. At the pH of a typical living organism, the amino acid is a zwitterion, meaning its amino group is protonated and its carboxyl group is deprotonated. The α-carboxyl group typically has a pKa near 2, and the α-amino group about 9 to 10. Basic residues have side chains with pKa values above about 6, so they are typically positively charged under physiological conditions. Histidine is a partial exception, since its imidazole ring has a pKa close to 6 and can be either charged or neutral near neutral pH.
The pH at which an amino acid carries no net charge is its isoelectric point (pI). For amino acids without ionizable side chains, the pI is calculated from the pKa values that surround the neutral species; for residues with ionizable side chains, the pI is the average of the two pKa values flanking the neutral form. Thus alanine, with no ionizable side chain, has a pI of roughly 6, whereas acidic residues such as aspartate have a pI set by the α-carboxyl and side-chain pKa values. Net charge changes systematically with pH, so the pI matters practically: at the pI, amino acids are least soluble in water and do not migrate in an electric field. This property underlies separation methods such as gel electrophoresis.
Peptide bond formation
Amino acids join through a covalent link formed by condensation: the α-carboxyl group of one amino acid reacts with the α-amino group of another, releasing a molecule of water in a condensation (dehydration) reaction. The resulting amide linkage is not favored on thermodynamic grounds alone. In cells, the reaction is coupled to ATP hydrolysis during protein biosynthesis, a process described in the translation section.
The peptide bond has partial double-bond character. Resonance between the lone pair on the nitrogen and the carbonyl group gives the bond partial double-bond character. Estimates place this at about 40% double-bond character. As a result, six atoms involved in the peptide bond (C, O, N, H, and the two α-carbons) lie in the same plane, and the group is usually found in the trans configuration. Rotation is blocked around the C–N bond, but the bonds at each α-carbon remain free to rotate. These rotations define the backbone torsion angles φ and ψ, which determine local folding. Isomerization of the peptide bond is slow, with an activation energy of roughly 80 kJ/mol. The reaction is accelerated by enzymes called peptidyl prolyl isomerases, which catalyze cis–trans isomerization of X–Pro bonds. The bond is also remarkably stable against spontaneous hydrolysis: one estimate gives a half-life of 350–600 years at room temperature and at pH 4–8.
Because each residue contributes one –NH–CHR–CO– unit, the chain has a free amino end (N-terminus) and a free carboxyl end (C-terminus). Sequences are conventionally written from the N-terminus to the C-terminus, which gives the chain a fixed direction that is preserved in every protein.
Primary Structure
Definition and direction
The primary structure of a protein is the linear order of its amino acid residues, joined end to end by peptide bonds. Each residue contributes a backbone amino group and carbonyl group, so the chain has two chemically distinct ends: a free amino group at the N-terminus and a free carboxyl group at the C-terminus. By convention, sequences are written from the N-terminus to the C-terminus, and residues are numbered from the first one reached at the amino end. Because peptide bonds have a fixed orientation, a sequence read backward describes a different molecule. Numbering conventions can also vary: in many proteins the initiating methionine is removed after synthesis, so a residue's position depends on whether that starting residue is counted.
Sequencing methods
Classical sequencing relied on Edman degradation. The N-terminal residue is labeled with phenyl isothiocyanate under mildly alkaline conditions and then cleaved off under acidic conditions as a derivative that can be identified chromatographically, leaving the rest of the chain intact for the next cycle. Each cycle reports one residue, and the efficiency of each cycle falls as the chain is shortened, so reliable reads are generally limited to roughly 30 to 50 residues. Longer proteins are therefore cut into shorter peptides with proteases or chemical reagents, and the overlapping fragments are then assembled into a complete sequence. Sanger's work on insulin in the 1950s used this kind of fragment-overlap strategy.
Modern mass spectrometry-based sequencing usually begins with enzymatic digestion, most often with trypsin, which cleaves after lysine and arginine residues except when the next residue is proline. The resulting peptides are ionized, selected and fragmented in a tandem instrument. The mass differences between successive fragment ions correspond to specific residue masses, and the measured spectra are matched against theoretical spectra computed from sequence databases. This allows thousands of proteins to be identified in a single experiment. Some limits remain: leucine and isoleucine have identical masses and must be distinguished by other means, and modifications such as phosphorylation shift masses in ways that must be modeled explicitly. Much of the sequence data in current databases is instead inferred by translating genes, as described in DNA sequencing work, which reveals the sequence but not the modifications present in the living protein.
Sequence comparison
Alignment algorithms such as Needleman–Wunsch (global alignment, 1970) and Smith–Waterman (local alignment, 1981) score matches, mismatches and gaps, and substitution matrices such as BLOSUM, derived from conserved blocks of aligned sequences, weight substitutions by chemical similarity. Positions that remain identical across distantly related species are usually under functional or structural constraint, whereas conservative substitutions, such as isoleucine for valine or aspartate for glutamate, are typically less damaging than changes that alter charge or size drastically.
Conserved stretches form sequence motifs. The Walker A (P-loop) motif, GxxxxGK[ST], wraps around the phosphate groups of nucleotides in many ATP- and GTP-binding proteins. The N-glycosylation sequon N-X-S/T, in which X is not proline, marks asparagine residues that can receive N-linked sugars. Sequence similarity over a long alignment is evidence of homology, that is, shared ancestry, and it supports transferring functional annotation between proteins. A short motif alone is weaker evidence, because such patterns can arise by chance or through convergence.
Single residue substitutions
A single amino acid change can alter function when it affects a buried, catalytic or interface position. The best-known example is sickle hemoglobin. In the β-globin chain of hemoglobin, a glutamate at position 6 is replaced by valine (Glu6Val), a change caused by a single nucleotide substitution in the gene. Pauling and colleagues showed in 1949 that the abnormal hemoglobin migrated differently in an electric field, and they described sickle-cell anemia as a "molecular disease." The substitution replaces a charged side chain with a hydrophobic one. In the deoxygenated state, the exposed valine at position 6 fits into a hydrophobic pocket on a neighboring β chain, promoting the assembly of hemoglobin into fibers that distort red blood cells. People who inherit two copies of the variant have sickle-cell disease, while those with one copy carry the sickle trait.
Not all substitutions are harmful. Changes at exposed, variable positions often have little effect, and the same residue type may be tolerated at one site and disruptive at another. Deletions can be equally consequential: the loss of a single phenylalanine at position 508 of the CFTR protein impairs folding, a defect that underlies the most common form of cystic fibrosis.
Secondary Structure
Secondary structure describes the local, regularly repeating conformations of the polypeptide backbone, considered apart from the overall fold of the chain. These patterns arise mainly from hydrogen bonds between backbone carbonyl oxygens and amide hydrogens, so the side chains of the amino acid residues shape the pattern only indirectly, through their effect on which conformations are favored.
Alpha helix
In the alpha helix, the chain coils into a right-handed rod. The carbonyl oxygen of residue i hydrogen-bonds to the amide hydrogen of residue i+4, and these bonds run roughly parallel to the helix axis. Each turn contains about 3.6 residues, so the side chains project outward at a steady rotational spacing, while the backbone groups inside the coil are all satisfied by partners within the same helix. Left-handed helices are rare because they are energetically less favorable. Many globular proteins are built largely from helices, and α-keratin, the main component of hair and nails, is almost entirely helical. At the helix C-terminus, a proline residue can interrupt the regular i to i+4 pattern and be followed by a tighter i to i+3 hydrogen-bonded turn, the 3₁₀ helix.
Beta sheet
A β-strand is an extended stretch, often five to ten residues long, in which the backbone zigzags instead of coiling. Strands that lie side by side are linked by hydrogen bonds between their backbone groups, forming a beta sheet. Adjacent strands can run in the same direction (parallel) or in opposite directions (antiparallel), and the antiparallel arrangement is generally the more stable of the two. Successive residues in a strand point their side chains alternately above and below the plane of the sheet, giving the pleated look that the name refers to. Real sheets are often twisted or curved. The strands in a sheet can come from segments that are far apart in sequence yet neighbors in three-dimensional space.
Turns, loops and coils
Not all of the backbone is regular. β-turns reverse the direction of the chain over roughly four residues and are usually stabilized by a hydrogen bond between the first and last residues of the turn. Loops and coils lack a repeating pattern, but they connect helices and strands and often form binding or catalytic surfaces. Glycine, whose side chain is only a hydrogen atom, can adopt a much wider range of backbone angles than other residues, which makes it common in tight turns and in flexible regions. Proline behaves in the opposite way. Its side chain loops back to the backbone nitrogen, which restricts rotation about the N–Cα bond to a narrow range, and its nitrogen carries no hydrogen that could donate in a hydrogen bond. As a result, proline is disruptive inside a helix but is tolerated in the first turn at the helix N-terminus.
Backbone torsion angles and the Ramachandran plot
Because the peptide bond is nearly planar, backbone conformation is set by rotation about two bonds in each residue: the N–Cα bond (phi, φ) and the Cα–C bond (psi, ψ). Plotting φ against ψ produces a Ramachandran plot. In 1963, G. N. Ramachandran and colleagues used computer models of small polypeptides to map sterically allowed regions. The most favored regions correspond to β-sheet conformations, to the right-handed α-helix and to a polyproline-type helix, and a less favored region corresponds to the left-handed helix. The right-handed α-helix sits near φ ≈ −57° and ψ ≈ −47°, while extended β-strands fall in a separate region with large positive ψ values. Glycine can occupy a far broader part of the plot than other residues, so validation plots are often drawn only for non-glycine and non-proline residues. Such plots are used to check the quality of experimentally determined and predicted structures.
Propensities in structure analysis
Each amino acid tends to favor certain conformations, and this tendency, called its propensity, can be exploited to infer structure from sequence. In 1974, Chou and Fasman derived propensity tables from proteins of known structure. In their scheme, alanine, glutamate, leucine and methionine favor helices, whereas proline and glycine break them. Their algorithm summed propensities over a sliding window and assigned a secondary-structure state where the sum passed a threshold. The Garnier–Osguthorpe–Robson method of 1978 extended this idea by using a window that spans eight residues on either side of the position being predicted. Such rule-based approaches reached only modest accuracy and tended to over-predict structure, so they were later refined. Propensity analysis remains a useful intuition for why a sequence is likely to form a helix or a strand, but it is no longer the basis of the most accurate predictors.
Tertiary and Quaternary Structure
Tertiary structure and domains
Tertiary structure is the overall three-dimensional arrangement of a single polypeptide chain, after secondary elements such as alpha helices and beta sheets have packed together. It is defined by the positions of every amino-acid side chain and backbone atom in space, and it is what determines where the active sites, binding surfaces and interaction faces of a protein are located. Tertiary structure is not uniform across proteins. Many chains of a few hundred residues fold into a single compact globular unit, as in myoglobin, whose first atomic model was determined by X-ray crystallography in the late 1950s. Longer chains often fold as a set of domains: compact, semi-independent units of roughly 50 to 200 residues that can often fold and remain stable on their own. Domains are frequently connected by flexible linkers, which allows them to move relative to one another and to be reused in different proteins. Immunoglobulin-type domains, for example, recur throughout antibodies and many cell-surface receptors.
Forces that stabilize the folded state
The folded structure is maintained by many weak, non-covalent interactions rather than by a few strong bonds. The dominant contribution is the hydrophobic effect: nonpolar side chains cluster in the interior of the protein, away from water, which increases the entropy of the surrounding solvent. Burial of hydrophobic residues is therefore the main force that drives a chain into a compact core. Hydrogen bonds form between backbone carbonyl and amide groups in secondary structure, and between polar side chains and the backbone or with each other, often inside the protein where they are shielded from water. Salt bridges between oppositely charged side chains, such as lysine or arginine paired with aspartate or glutamate, contribute to specific contacts, although their effect is weaker on the protein surface, where water screens the charges. Van der Waals forces act at very short range but are numerous when side chains pack tightly together, and they help define the precise fit of the core. Because the net stability of a typical globular protein is only modest, on the order of several to about 15 kcal/mol, the folded state is a balance of many small contributions against the entropic cost of ordering the chain.
Some proteins gain additional stability from disulfide bonds, which are covalent links formed when the thiol groups of two cysteine residues are oxidized to create an S–S bond. Disulfides restrict the conformations of the unfolded chain and so add stability to the folded state. They are common in extracellular and secreted proteins, where the environment is oxidizing, and they are formed in the endoplasmic reticulum with the help of oxidoreductases such as protein disulfide isomerase. Cytosolic proteins usually contain few or no disulfides, because the cytoplasm is a reducing environment. Insulin is a well-known example: its two chains are held together by two interchain disulfides, and its A chain contains one intrachain disulfide.
Quaternary structure and cooperativity
Quaternary structure describes how several polypeptide chains, called subunits, assemble into a single functional complex. Subunits may be identical, as in homodimers and in many enzymes, or different, as in heterodimers and larger assemblies. Some complexes contain dozens of subunits, such as viral capsids and the chaperonin GroEL, which forms a double ring of seven subunits each. Subunit assembly can allow a protein to form binding sites or channels that a single chain could not form, and it allows the activity of one subunit to influence another.
hemoglobin is the classic example of cooperative behavior in a quaternary structure. Adult hemoglobin is a tetramer of two alpha and two beta chains, each binding one heme group and one oxygen molecule. Its oxygen-binding curve is sigmoidal rather than hyperbolic, meaning that binding of the first oxygen increases the affinity of the remaining sites. This cooperativity arises because the tetramer exists in a low-affinity, deoxygenated (tense) state and a high-affinity, oxygenated (relaxed) state; oxygen binding shifts the subunits toward the relaxed conformation, and the transition is coordinated across the complex. The effect is modulated by allosteric factors such as protons and carbon dioxide (the Bohr effect) and by 2,3-bisphosphoglycerate, which binds in the central cavity of the deoxygenated form and lowers oxygen affinity. The same quaternary architecture is sensitive to single substitutions: in sickle-cell hemoglobin, a glutamate-to-valine change at position 6 of the beta chain creates a hydrophobic patch on the surface of the deoxygenated tetramer, promoting aggregation of fibers that deform red blood cells in sickle-cell-disease.
Protein Folding and Misfolding
Christian Anfinsen tested whether a sequence alone is enough to specify a fold. He used ribonuclease A, unfolding it with urea, which at high concentrations disrupts the noncovalent bonds of proteins; and mercaptoethanol, which reduces disulfide bonds. When the denaturants were removed, the enzyme refolded to its native, functional structure, even in the absence of cellular machinery. Anfinsen shared the 1972 Nobel Prize for Chemistry for his work on the structure of the enzyme ribonuclease A. The result was formulated as the thermodynamic hypothesis: the native structure is the state in which it has a minimum of free energy. In thermodynamic terms, folding is spontaneous when the change in Gibbs free energy is negative, , and the stability of a folded protein is usually modest, because the favorable interactions that stabilize the fold are only slightly larger than the entropic cost of ordering the chain. The hypothesis has exceptions: It is only true for some proteins. For other proteins, chaperones are needed.
Cyrus Levinthal's 1968 argument highlighted the search problem. A random search over all accessible conformations would take longer than the age of the universe, whereas, in reality, most proteins fold within milliseconds to seconds. The resolution is that folding is not a random walk. Protein folding can be understood as a multidimensional optimization problem within a funnel-like energy landscape. The term "funnel" was introduced by Ken A. Dill in 1987. In the folding funnel model, local interactions form early and narrow the conformational space, so the chain descends toward the native state. Dill and Chan later described a "golf-course" landscape, where a random searching for the native states would prove impossible, and a rugged pathway can channel the chain into a directed route. Partially folded intermediates and transition states have been detected experimentally, which supports this picture. In cells, folding often begins before synthesis is complete, since folding commonly begins while the protein is being produced in the ribosome.
Folding can also go wrong because a chain may become kinetically trapped. In the process of folding it is possible for proteins to become trapped in intermediate states, a locally optimal state, but not a globally optimal state. Molecular chaperones address this problem in vivo. They assist in the folding process but do not determine the native structure of a protein. Their job is to bind exposed hydrophobic surfaces of unfolded or partly folded chains, which would otherwise associate with other chains. Among the best-studied are Hsp70, chaperonins (GroEL/ES), PDI, PPI. The concentrated cytoplasm, where macromolecules occupy a large fraction of the volume, makes such nonspecific intermolecular contacts more likely, which is why chaperone assistance is important in cells even for proteins that can refold in dilute buffer.
When folding fails, proteins may form insoluble aggregates. Misfolded proteins form insoluble amyloid aggregates. Amyloid fibrils are long, ordered assemblies rich in cross-β structure, and they are associated with neurodegenerative conditions. Other amyloid diseases, including Alzheimer's disease and Parkinson's disease, are also exceptions to Anfinsen's dogma. Prion disorders show a self-propagating mechanism. Pathogenic PrP^Sc induces misfolding of normal PrP^C. Misfolding can also cause disease by a simpler route: in cystic fibrosis, the ΔF508 mutation in CFTR leads to premature degradation, so the protein is lost rather than gaining a toxic activity. The two general pathogenic mechanisms are loss of function and gain of toxic function.
Folded proteins are destabilized by heat and by chemical denaturants. Heating increases thermal motion, which weakens hydrogen bonds and the hydrophobic packing that hold the fold together, and high heat often leads to aggregation rather than reversible unfolding. Chemical denaturants act in a similar way: in the classic ribonuclease experiment, denaturant conditions such as 8 M urea and β-mercaptoethanol completely unfold the molecule and result in a random coil. Ribonuclease contains 124 amino acid residues and 4 disulfide bridges, so reducing the disulfides is a necessary part of unfolding it completely. Insulin behaves differently: a teaching question often raised is why active insulin is not recovered when the mature form is subjected to the same procedures used by Anfinsen with RNase A, which reflects the particular disulfide arrangement and chain processing of insulin.
Protein Synthesis and Gene Expression
Gene expression is the process by which the information in a gene is turned into a functional product. For protein-coding genes the route runs from DNA to RNA to protein. Transcription copies a gene into an RNA transcript, which in eukaryotes must be processed before it can serve as a messenger RNA (mRNA). Translation then reads that mRNA and assembles the amino acid chain.
From pre-mRNA to mature mRNA
In eukaryotes the primary transcript carries introns and unprotected ends. Capping occurs first, when the nascent transcript is only 20–30 nucleotides long. The 3′ end is then generated as the polyadenylation signal is recognised by cleavage proteins that cut the transcript a short distance downstream from the signal. The resulting tail promotes mRNA stability, export, and translation initiation. Between these steps, the spliceosome removes introns and joins exons. Because the choice of exons can vary, alternative splicing allows a single gene to code for multiple mRNA transcripts and protein isoforms, contributing to protein diversity. Estimates of how many protein isoforms arise from a given gene set differ between sources, so the figures should be read as approximate.
Decoding the message on the ribosome
Translation takes place on the ribosome, which has three sites. E, P, and A. The E-site is where uncharged-tRNAs exit the ribosome. The transfer RNA (tRNA) molecules that enter these sites each carry a specific amino acid. The enzymes that attach amino acids to tRNAs use ATP, since the energy to make this bond comes from ATP. Each mRNA codon is read through base pairing with the tRNA anticodon, which is the central rule of the genetic code. The third codon position is more flexible, a feature known as wobble. The ribosomes "read" the mRNA in the 5′ to 3′ direction. Consequently, protein synthesis proceeds from the N-terminus to the C-terminus of the protein. In a cell, several ribosomes often work on one message at once, since initiation can happen multiple times on a single mRNA, forming the polyribosome, or polysome.
The start signal differs between domains of life. In bacteria, a sequence upstream of the first AUG codon, called the Shine-Dalgarno sequence (AGGAGG), interacts with the rRNA molecules that compose the ribosome. This interaction anchors the 30S ribosomal subunit at the correct location on the mRNA template. In eukaryotes, the eukaryotic initiation complex recognizes the 7-methylguanosine cap at the 5' end of the mRNA.
Initiation, elongation and termination
Initiation assembles the ribosome on the message. Once bound, the 30S subunit-mRNA complex is further stabilized by initiation factor 1 (IF1), initiation factor 3 (IF3), and initiator tRNA (fMet-tRNA). The initiator tRNA then pairs with the start codon, and in the last step of initiation, the large ribosomal subunit binds to this complex.
Elongation is a repeated cycle. In total, elongation consists of three major steps: aa-tRNA binding, transpeptidation, and translocation. The peptide bond itself is made by an RNA-based catalytic center: the formation of each peptide bond is catalyzed by peptidyl transferase, an RNA-based enzyme that is integrated into the 50S ribosomal subunit.
Termination occurs because no tRNA recognizes a stop codon. The ribosome actually stalls and the translation slow-down is just long enough for a protein termination factor to enter the A site. When protein synthesis is terminated, release factor proteins bind to the stop codons, GTP hydrolysis occurs, and peptidyl transferase activity is stimulated, causing release of the protein from the tRNA. Afterwards, ribosome recycling factor (RRF) prompts the separation of mRNA, small subunit, and large subunit and recycling of the ribosome.
Energy cost
Translation is among the most energy-demanding processes in the cell. Charging tRNAs with amino acids is the costliest step, and translation is perhaps the most energy-intensive job a cell must do, beginning with the attachment of amino acids to their tRNAs. Guanosine triphosphate (GTP) powers the protein factors. GTP is required for translation initiation and for elongation and termination. For example, a charged tRNA with anticodon complementary to the next codon will enter the A site of the ribosome, escorted by elongation factor EF-Tu. Again, GTP hydrolysis provides the energy for this process. In standard accounting, each residue added thus costs about two ATP-equivalents, one for charging the tRNA and one for GTP-driven steps. Translation of mRNA into protein is therefore linked to the cell's energy supply, which is why energy-limited cells slow protein synthesis.
Post-translational Modification and Turnover
After a polypeptide is synthesized on the ribosome, many proteins are chemically altered again. Post-translational modifications (PTMs) are covalent changes made after translation, and they extend the functional range that the genetic-code alone specifies. Many PTMs are enzymatic and reversible, so a cell can switch a protein's activity, location or binding partners on and off quickly. Other modifications tag a protein for destruction, which sets how long it persists.
Covalent modifications
Phosphorylation adds a phosphate group, most often to serine, threonine or tyrosine side chains. Protein kinases transfer the phosphate from adenosine-triphosphate, and phosphatases remove it. The added negative charge can change a protein's shape, catalytic activity or affinity for partners, which makes phosphorylation a central switch in cell signaling. Glycosylation attaches sugar chains either to asparagine residues (N-linked) or to serine and threonine residues (O-linked). In secreted and membrane proteins, these glycans influence folding, stability and recognition by other cells. Acetylation of lysine side chains removes their positive charge and alters how a protein interacts with DNA or with other molecules. Methylation of lysine and arginine residues can create binding sites for proteins that read such marks. Lipidation attaches fatty acids, such as myristate or palmitate, or isoprenoid groups, which help anchor proteins to the cell-membrane.
Proteolytic processing of precursors
Some enzymes and hormones are made as inactive precursors, called zymogens or proenzymes, that become active only after a specific peptide bond is cut. This cleavage is irreversible and does not require ATP, so it can take place outside the cell. Pepsinogen is converted into the digestive enzyme pepsin in the stomach. Blood clotting proceeds through a cascade in which one zymogen after another is cut, with prothrombin giving rise to thrombin. The same logic applies to hormone precursors. Proinsulin is cut at pairs of basic residues by prohormone convertases inside secretory granules, and the connecting C-peptide is removed, leaving insulin's two chains joined by disulfide bonds. Some proprotein convertases, such as furin, are themselves activated only after they remove their own inhibitory prodomain.
Ubiquitination and proteasomal degradation
Ubiquitin is a small protein of 76 residues that is attached to substrate proteins through an enzymatic cascade. An E1 activating enzyme prepares the ubiquitin, an E2 conjugating enzyme carries it, and an E3 ligase usually selects the substrate. Human cells contain several hundred E3 ligases, and more than a thousand proteins have been reported to regulate ubiquitination. A single ubiquitin can be added (monoubiquitination), or ubiquitin chains can form through different lysine residues on ubiquitin itself, and each linkage pattern carries a different signal. Chains linked through lysine 48 typically direct a protein to the 26S proteasome, an ATP-dependent complex that unfolds and cuts substrates into short peptides. Monoubiquitination of some membrane receptors, such as the epidermal growth factor receptor, instead sends them into the endosomal pathway toward lysosomes. Because ligases and deubiquitinating enzymes continually add and remove these tags, the steady-state level of each protein and its half-life are tuned to the needs of the cell.
Lysosomal degradation and autophagy
Damaged, aggregated or surplus proteins and whole organelles are removed largely through autophagy. In macroautophagy, a double-membrane vesicle called the autophagosome forms around cytoplasmic cargo and delivers it to the lysosome. In selective autophagy, ubiquitin tags on the cargo help to mark it for capture. The autophagosome fuses with a lysosome to form an autolysosome, where acidic hydrolytic enzymes break down the contents, and the fusion depends on proteins such as syntaxin-17 and the small GTPase Rab7. Chaperone-mediated autophagy and endocytic routes also deliver proteins to lysosomes. Impaired proteostasis has been associated with ageing, cancer, autoimmune diseases and neurodegenerative disorders, with the buildup of normally degraded proteins a common feature. Ubiquitin-proteasome and autophagic pathways are both regulated by ubiquitination, so the two systems are linked rather than separate.
Enzymes and Catalysis
Most enzymes are proteins that speed up chemical reactions without being consumed. They work by lowering the activation energy of a reaction, which is the free-energy barrier between substrate and transition state. A catalyst does not change the free-energy difference between reactants and products, so it leaves the equilibrium position unchanged and accelerates the forward and reverse reactions alike. According to transition-state theory, the rate depends exponentially on the height of this barrier, so a modest reduction in activation energy produces a large increase in speed. Reported rate enhancements range from a few-fold to about 10<sup>17</sup>-fold; the decarboxylation of orotidine 5′-monophosphate is among the most dramatic examples. Catalysis takes place at the active site, a cleft or pocket lined with residues that bind the substrate with high specificity and position the reacting groups precisely.
Catalytic mechanisms
Enzymes achieve their effect through several overlapping strategies. The most important is transition-state stabilization: the active site binds the transition state more tightly than it binds the substrate, which lowers the energy required to reach it. Catalytic groups contribute through acid-base catalysis, in which side chains donate or accept protons, and through covalent catalysis, in which a transient covalent bond forms between the enzyme and the substrate. Serine proteases such as chymotrypsin use a catalytic triad of serine, histidine and aspartate, in which the serine hydroxyl attacks the peptide carbonyl to form an acyl-enzyme intermediate. Metal ions can polarize bonds or stabilize negative charges, as zinc does in carbonic anhydrase, one of the fastest known enzymes. Holding two reactants in the correct orientation and creating a local electrostatic environment that favors the transition state also contribute to rate enhancement.
Substrate binding models
Emil Fischer proposed in 1894 that enzymes and substrates fit together like a lock and key, with a rigid, pre-formed complementarity. Daniel Koshland's induced-fit model, put forward in 1958, holds that the binding of the substrate induces a conformational change in the enzyme, which realigns catalytic groups and may exclude water from the site. Structural and kinetic studies support elements of both views, and modern accounts often combine induced fit with conformational selection, in which the enzyme samples several shapes and the substrate preferentially binds one of them.
Many enzymes require non-protein components for activity. Metal ions such as Zn<sup>2+</sup>, Fe<sup>2+</sup> or Mg<sup>2+</sup> serve as cofactors. Organic coenzymes, many of them derived from vitamins, act as carriers of electrons, hydrogen atoms or chemical groups; examples include NAD<sup>+</sup>, flavin adenine dinucleotide and pyridoxal 5′-phosphate. A tightly bound cofactor is called a prosthetic group. The protein component without its cofactor is the apoenzyme, and the complete active form is the holoenzyme.
Enzyme kinetics
For many single-substrate enzymes, the initial reaction rate follows the Michaelis-Menten equation, derived from a mechanism proposed by Leonor Michaelis and Maud Menten in 1913:
At low substrate concentration the rate rises roughly in proportion to [S]. At high concentration the enzyme becomes saturated and the rate approaches the maximum velocity, . The Michaelis constant is the substrate concentration at which the rate is half of ; a low generally indicates high apparent affinity, although is not strictly a dissociation constant for complex mechanisms. The turnover number gives the number of substrate molecules converted per active site per second, and the ratio is commonly used to compare catalytic efficiency. Lineweaver-Burk plots, which linearize the rate data as reciprocals, were widely used before computer fitting became routine.
Inhibitors are classified by how they interact with the enzyme. Competitive inhibitors bind the active site and raise the apparent while leaving unchanged. Uncompetitive inhibitors bind only the enzyme-substrate complex and lower both and . Noncompetitive inhibitors bind at a different site with equal affinity for free enzyme and complex, reducing without changing . Mixed inhibitors alter both parameters. Irreversible inhibitors form stable covalent bonds with the enzyme; aspirin, for instance, acetylates a serine in cyclooxygenase.
Regulation of enzyme activity
Allosteric effectors bind at sites distinct from the active site and alter its conformation and activity. Activators and inhibitors may be the substrate itself (homotropic effects, which often produce sigmoidal kinetics through cooperativity) or other molecules (heterotropic effects). The concerted (Monod-Wyman-Changeux) and sequential (Koshland-Némethy-Filmer) models, both proposed in the 1960s, describe how such cooperative transitions occur. Feedback inhibition is a common form of allosteric control in which an end product of a pathway inhibits an early, committed enzyme. In Escherichia coli, CTP inhibits aspartate transcarbamoylase, the first committed step in pyrimidine synthesis, while ATP acts as an activator, balancing purine and pyrimidine supplies.
Covalent modification provides a faster and often reversible means of control. Protein kinases add phosphate groups to serine, threonine or tyrosine residues, and phosphatases remove them. In glycogen metabolism, phosphorylase kinase phosphorylates glycogen phosphorylase at a specific serine, converting the enzyme into its active form and mobilizing glycogen in response to hormonal signals. Some enzymes are also activated irreversibly by proteolytic cleavage of an inactive precursor, a mechanism that is discussed elsewhere in the context of protein processing.
Structural, Motor and Transport Proteins
Structural proteins
Collagen is one of the most abundant proteins in mammals and is the main load-bearing protein of connective tissue. Each collagen molecule is a triple helix of three polypeptide chains. The chains carry a repeating Gly-X-Y pattern in which glycine occupies every third position, so that its small side chain fits into the crowded central axis of the helix. Positions X and Y are frequently occupied by proline and hydroxyproline. After the secreted precursor (procollagen) is processed, the molecules assemble into fibrils, and enzymatic cross-links between them give tendon and bone their tensile strength. Type I collagen predominates in bone, skin and tendon, while other types occur in cartilage, basement membranes and other tissues, and fibrils are then organized into larger fibers.
Keratins belong to the intermediate-filament family. Alpha-keratins form coiled-coil dimers, which associate into higher-order filaments roughly 10 nm in diameter. In hair and nails, the keratin filaments are embedded in a matrix rich in cysteine, and disulfide bonds between these residues make the tissue rigid and chemically resistant. The keratins of the epidermis are a different set of paired types, which is one reason hair and skin differ in their mechanical properties.
Cytoskeletal and motor proteins
Actin exists as globular monomers that polymerize into a helical filament about 7 nm across. Filaments are polar, with a faster-growing barbed end and a slower-growing pointed end, and polymerization depends on bound ATP. Tubulin forms heterodimers of α- and β-subunits, each of which binds GTP, and these heterodimers assemble into hollow microtubules built from typically 13 protofilaments. Microtubules alternate between growth and rapid shrinkage, a behavior termed dynamic instability, which allows the cytoskeleton to reorganize quickly.
Motor proteins use the energy released by ATP hydrolysis to generate directed movement. In skeletal muscle, myosin II forms thick filaments whose heads bind actin, undergo a conformational change known as the power stroke, and then release and reattach, pulling actin filaments past one another in the sliding-filament mechanism. Kinesin-1 is a dimer whose two heads alternate along a microtubule in a hand-over-hand manner, advancing about 8 nm per step toward the microtubule's plus end. Dynein, which is a different motor, moves cargo in the opposite direction.
Transport proteins
Hemoglobin is a tetramer of two α and two β subunits, each carrying a heme group that binds one oxygen molecule. Binding is cooperative: the affinity of the remaining sites increases as oxygen binds, which lets the protein load oxygen efficiently in the lungs and release it in tissues with lower oxygen tension. Hemoglobin also carries a fraction of the body's carbon dioxide and protons.
Membrane transporters move solutes across the cell membrane against or along their gradients. ATP-binding cassette transporters couple ATP hydrolysis to the export of substrates, while secondary transporters, such as the glucose transporters of the GLUT family, use the energy stored in ion or concentration gradients. Ion channels instead form water-filled pores that open in response to voltage, ligands or mechanical force. The structure of the bacterial potassium channel KcsA, solved by Roderick MacKinnon in 1998, revealed how a selectivity filter discriminates between potassium and sodium ions. MacKinnon shared the Nobel Prize in Chemistry in 2003 with Peter Agre, who discovered the aquaporin water channels.
Storage proteins
Ferritin sequesters iron in a form that is soluble and non-toxic. Its shell is made of 24 subunits of two types, heavy (H) and light (L), and the cavity inside the shell can hold a mineral core of ferric oxyhydroxide containing up to several thousand iron atoms. Casein is the major protein of mammalian milk. Casein molecules associate with calcium phosphate into colloidal micelles, which keep high concentrations of calcium and phosphate soluble and supply the newborn with amino acids and minerals.
Signaling, Receptors and Immune Proteins
Protein hormones and the insulin receptor
Many hormones are proteins or peptides. Because they are large and hydrophilic, they cannot pass through the lipid bilayer of the cell membrane, so they act by binding receptors on the cell surface. insulin is the classic example. Frederick Sanger's sequencing of the molecule showed that it consists of two chains, A and B, joined by disulfide bonds. Pancreatic β cells secrete it in response to rising glucose. Its receptor is a heterotetramer of two extracellular α subunits and two membrane-spanning β subunits. Insulin binding to the α subunits activates the tyrosine kinase domains of the β subunits, which phosphorylate one another and then recruit intracellular adaptor proteins. The resulting signal, which runs through the PI3K–AKT pathway, moves the glucose transporter GLUT4 to the plasma membrane and so increases glucose uptake.
Receptor families and intracellular relay
Extracellular signals are converted into intracellular responses by a small number of receptor architectures. G protein-coupled receptors (GPCRs) traverse the membrane seven times with α helices. Ligand binding alters the receptor's conformation so that it activates a heterotrimeric G protein. The Gα subunit exchanges GDP for GTP and separates from the Gβγ dimer, and both parts then act on effectors such as adenylyl cyclase, which generates the second messenger cAMP. GPCRs form the largest family of human receptors and are the targets of roughly a third of approved drugs.
Receptor tyrosine kinases (RTKs) are single-pass transmembrane proteins. Ligand binding brings two receptor molecules together, and each phosphorylates tyrosine residues on the other. These phosphotyrosines become docking sites for proteins that carry SH2 or PTB domains, which connect the receptor to the Ras–MAPK and PI3K–AKT cascades. The epidermal growth factor receptor and the insulin receptor follow this pattern.
Cytokine receptors have no intrinsic kinase activity. Instead, they associate with Janus kinases (JAKs). After ligand binding, JAKs phosphorylate the receptor, and STAT proteins are recruited, phosphorylated, and paired. The STAT dimers then move into the nucleus and regulate gene expression. This JAK-STAT signaling pathway carries the responses to interferons, many interleukins and erythropoietin, and it is the target of several approved JAK inhibitors.
Antibodies and T-cell receptors
Antibodies are Y-shaped glycoproteins. A typical IgG molecule is about 150 kDa and contains two identical heavy chains and two identical light chains, linked by disulfide bonds. Each arm carries a Fab region, in which the variable domains of the heavy and light chains form the antigen-binding site. The three hypervariable loops in each variable domain, called complementarity-determining regions, contact the antigen. The Fc region, which is formed from the constant domains, mediates effector functions by engaging Fc receptors and complement. Antibody diversity arises from V(D)J recombination in developing B cells, followed by somatic hypermutation and class switching after antigen exposure.
T-cell receptors are heterodimers, most often of α and β chains, that resemble a Fab fragment in their variable and constant domains. They do not recognize free antigen. Instead, they bind peptide fragments displayed by the major histocompatibility complex (MHC). Class I MHC presents peptides to CD8⁺ cytotoxic T cells, and class II MHC presents them to CD4⁺ helper T cells. Signal transduction after binding depends on the associated CD3 complex, whose cytoplasmic tails carry immunoreceptor tyrosine-based activation motifs.
Cytokines in immune coordination
Cytokines are small secreted proteins that act as local messengers among immune cells and their neighbors. They include interleukins, interferons, tumor necrosis factor and chemokines. Their effects are often pleiotropic and overlapping. Interleukin-2 drives the proliferation of activated T cells, interferon-γ activates macrophages, and tumor necrosis factor-α promotes inflammation. Chemokines guide leukocytes to sites of infection. Because the response to a given cytokine depends on the receptors present on the target cell, the same molecule can produce different outcomes in different tissues. Excessive cytokine release contributes to the systemic inflammation known as a cytokine storm.
Classification, Families and Databases
Structural classification schemes group protein domains by three-dimensional fold and architecture rather than by sequence alone. The two most widely used are SCOP and CATH. In SCOP, domains are assigned to one of four classes (mainly α, mainly β, alpha beta, or few secondary structures) on the basis of composition, secondary-structure contacts and the proportion of parallel and antiparallel sheets. Below the class, SCOP's tiers narrow through fold and superfamily to family, and the fold tier pools superfamilies whose secondary-structure makeup and connection order match, while superfamilies hold domains that are homologous or structurally very similar (The hierarchical levels of SCOP from bottom to top are family, superfamily, fold and class. In general, families and superfamilies consist of domains that are homologous or structurally very similar.).
CATH uses four major levels in this hierarchy which are Class, Architecture, Topology, and Homologous superfamily. Its architecture tier has no counterpart in SCOP (The architecture (A) level is unique to CATH.). The two schemes also differ in how they are built. CATH is largely algorithmic, although expert curation is used for the "architecture" level and to resolve ambiguity or disagreements in the algorithms' output, whereas in SCOP the classification hierarchy is completely defined by expert curators, with automation used only to assign newly characterized protein structures to the hierarchy. As a result, the same protein can receive different labels. CATH's geometric emphasis places several Rossmann-type folds in one group that SCOP separates into distinct folds, and for one comparison set SCOP recognized 286 folds where CATH recognized 447 (CATH on the other hand focuses on the geometric aspects of structural similarity, and thus encompasses all these SCOP folds into one large fold family; SCOP classifies pset3 into 286 separate folds, whereas CATH uses 447 folds to classify the same set.). Because homology takes precedence over fold similarity in both systems, neither classification is strictly hierarchical anymore, since in some families a small percentage of homologues may have different folds and even occasionally architectures.
Protein families and domains are defined differently in Pfam. Pfam is a large collection of protein multiple sequence alignments and profile hidden Markov models. Each entry is built around a seed alignment of representative sequences, and the full alignment collects every member detected by the profile model (The seed contains representative members of the family, while the full alignment contains all members of the family as detected with a profile hidden Markov model (HMM) constructed from the seed alignment). Pfam-A holds curated families with checked seed alignments, while Pfam-B holds families generated automatically from the remaining sequences (Pfam-A is curated and contains well-characterized protein domain families with high quality alignments, which are maintained by using manually checked seed alignments and HMMs to find and align all members. Pfam-B contains sequence families that were generated automatically). The 37.0 release, issued in June 2024, contains 21,979 families. Profile models are useful because they can place a sequence in a known family even if the homology by alignment appears to be low.
Structural and sequence data are archived separately from these classifications. The Protein Data Bank was established in 1971 with just seven protein structures, and since 2003 it has been run by the wwPDB consortium, which manages the PDB archive according to the FAIR principles: Findability, Accessibility, Interoperability and Reusability. The RCSB PDB homepage reports 260,626 3D structures from the PDB and 1,062,058 computed structure models from AlphaFold DB and ModelArchive. The wwPDB also adds value-added annotation, such as secondary structure, quaternary structure descriptions and information about ligand-binding sites. UniProt is the central sequence counterpart: its manually reviewed Swiss-Prot section and its automatically annotated TrEMBL section differ in annotation depth, so records vary in how much curated evidence supports them. Pfam itself draws on this sequence universe, being based not only on the UniProtKB sequence database, but also on NCBI GenPept and on sequences from selected metagenomics projects.
Family and fold membership is therefore an inference from sequence and structural similarity, not a direct observation of shared ancestry. Pfam's non-domain annotations, such as transmembrane and signal-peptide regions, are pre-computed over the sequence database by the following third party programs, so their accuracy depends on those predictors. Annotation quality also varies between resources: a family assignment is only as reliable as its seed alignment and the curation behind it, and SCOP and CATH can disagree on the same domain. Comparing these resources, and checking them against experimental structures, is the main way to judge whether a classification is robust.
Structure Determination and Proteomics
For several decades, X-ray crystallography was the main route to atomic-resolution protein structures, and it remains the most productive method in the Protein Data Bank. Its central limitation is the need for an ordered crystal. The protein must first be purified, typically to better than 90–95 percent, and then brought out of a supersaturated solution under carefully chosen conditions of pH, temperature, precipitant and protein concentration. Crystallization is still largely a trial-and-error process. The resulting crystals are loosely packed and contain roughly 40–60 percent solvent by volume. Diffraction quality determines how accurately the atoms can be placed, and high-resolution data usually require crystals large enough to survive the X-ray beam. Radiation damage is a particular problem for membrane proteins, even at cryogenic temperatures. Serial crystallography at X-ray free-electron lasers has partly addressed this by using many microcrystals, each exposed only once.
Nuclear magnetic resonance (NMR) spectroscopy determines protein structure in solution, without crystals. It detects the magnetic behaviour of atomic nuclei, and the resulting spectra report on local geometry, inter-atomic distances and molecular motion, which makes NMR especially useful for flexible regions and for studying dynamics near physiological conditions. Structural NMR of proteins usually depends on isotopic labelling (for example with ¹⁵N and ¹³C) and on relatively concentrated, pure samples. As molecular size increases, spectral signals overlap and relax more quickly, so conventional NMR is practical mainly for smaller proteins, although specialized labelling schemes extend the accessible size range.
Cryo-electron microscopy (cryo-EM) images individual protein molecules that have been rapidly frozen in a thin layer of vitreous ice. Computational averaging of many thousands of particle images then yields a three-dimensional map. Because no crystal is needed, the method suits large assemblies, flexible complexes and membrane proteins that resist crystallization. Advances in electron detectors, image-processing software and microscope stability have raised resolution to near-atomic levels, and small proteins of around 60 kDa can now be analysed. The 2017 Nobel Prize in Chemistry was awarded to Jacques Dubochet, Joachim Frank and Richard Henderson for developing cryo-EM for high-resolution structure determination of biomolecules in solution. Cryo-EM typically requires far less sample than X-ray crystallography or NMR.
Identifying and quantifying proteins in complex mixtures relies heavily on mass spectrometry (MS) combined with separation methods. A classic workflow used two-dimensional gel electrophoresis: proteins were first separated by isoelectric point through isoelectric focusing and then by molecular mass through SDS-PAGE. Protein spots were excised, digested with a protease such as trypsin, and identified by MS, using peptide masses or peptide fragmentation (tandem MS). Relative abundance could be estimated from spot intensities, but absolute quantification was difficult, and very acidic or basic proteins, very large or very small proteins, and membrane proteins were under-represented. Modern workflows usually couple liquid chromatography to MS, separating peptides before they are ionized. Quantification then uses isotope-labelled tags that differ in mass, or label-free approaches based on spectral counts or ion intensities. Because different peptides ionize with different efficiencies, label-free comparisons between samples require careful normalization.
Proteomics is the large-scale study of proteins in a cell, tissue or organism. It addresses protein abundance, post-translational modifications (such as phosphorylation, enriched by specific peptide-capture methods), subcellular localization, and interaction networks, as well as the activities and functions of proteins. The field expanded rapidly as high-throughput MS, improved separation techniques and computational data analysis matured. Its results are commonly deposited in shared databases and integrated with structural and sequence resources such as UniProt, allowing observed protein levels and modifications to be interpreted in the context of known structures and functions.
Structure Prediction and Protein Design
Before deep learning, computational prediction relied on two main strategies. Rule-based methods used residue-level propensities for helices and sheets to flag likely secondary-structure segments, but they rarely produced a complete three-dimensional model. Homology modelling copied the backbone of a related protein of known structure onto a target sequence. It worked well when a suitable template existed and became unreliable as sequence identity fell. Proteins without close relatives in the structure databases stayed largely out of reach, and physics-based ab initio folding was too expensive for all but small chains.
Deep learning and AlphaFold
The decisive change came from neural networks trained on known structures. AlphaFold, developed by Google DeepMind, takes multiple sequence alignments as input and entered the blind CASP14 competition in 2020. A key aspect of AlphaFold's success lies in its use of co-evolution data, which identifies evolutionary correlations that suggest how residues interact within the folded protein. One of its outputs is a probability profile of residue–residue distances, from which atomic coordinates are assembled. Its predictions reached a median domain GDT_TS of 92.4 in the assessment, a level of average accuracy that earlier methods had not approached. The AlphaFold team has since predicted structures for almost all of the 200 million proteins scientists know today.
Limits of the approach
Accuracy depends on the alignment. Performance degrades when homologous sequences are sparse or altogether absent. Even later versions still cannot predict high-accuracy structures, or even the correct fold, for some proteins that have very few or no homologous sequences in existing databases. The CASP16 assessment judged that the problem of single-domain protein fold prediction is nearly solved, but it noted that challenges remain in accurately modeling truncated sequences, irregular secondary structures, and interaction-induced conformational changes. Multi-chain assemblies are less settled. One benchmark reported a success rate of approximately 85% for general heterodimer complexes and 25% for immune-related complexes or repebody-antigen complexes. A predicted structure is also a single static model, and it remains unclear whether such models can capture protein dynamics or multiple conformations.
De novo design and directed evolution
Protein design approaches the problem from the opposite direction. Instead of predicting the structure of a natural sequence, de novo design writes new sequences intended to fold into chosen shapes. David Baker's group has dedicated its work to designing and constructing proteins that are not, and even can not, be found in nature. A limitation is that a model trained on alignments may perform poorly on designed sequences that have few natural relatives, a question that remains to be seen for many design problems.
Directed evolution offers an experimental route to improved proteins. By mutating and recombining the gene that encodes the enzyme followed by artificial selection for the desired traits, protein engineers can "breed" biomolecules. Frances Arnold pioneered the method in the early 1990s as a bioengineering approach for creating new and better enzymes in the laboratory, and it was recognised by the 2018 Nobel Prize in Chemistry. It is now widely used to create enzymes for making everything from fuels to medicines, chemicals, and consumer goods.
Recognition and open problems
The 2024 Nobel Prize in Chemistry was divided: one half was awarded to David Baker "for computational protein design", the other half jointly to Demis Hassabis and John Jumper "for protein structure prediction". Several questions remain open. Structure prediction has largely solved single-domain folds but is less reliable for complexes, flexible regions and conformational change. Design of proteins with new chemistries and functions is still being tested against experiment, and the link between predicted static structures and the dynamic behaviour of proteins in cells is not yet fully resolved.
Nutrition and Metabolism
Dietary protein is digested into small absorbable units before the body can use its amino acids. Digestion begins in the stomach, where hydrochloric acid denatures the protein, and denaturation results in the loss of the protein's function. The same acid converts inactive pepsinogen to its active form, pepsin. Pepsin's cleavage is selective: gastric pepsins catalyze specific cleavage at phenylalanine or tyrosine, which produces shorter polypeptides rather than free amino acids.
Most of the work is done in the small intestine. The pancreas supplies trypsin, chymotrypsin, elastase, and the carboxypeptidases, which cleave the polypeptides into oligopeptides and amino acids. These pancreatic enzymes are themselves released as inactive zymogens; trypsin is first activated by enteropeptidase and then activates the other pancreatic proenzymes. The intestinal lining adds aminopeptidases located on the brush border and other peptidases located within the cells. Absorption then proceeds by two routes: about half of the digested protein enters enterocytes as free amino acids through specific transporters, while the other half enters as di- and tripeptides through the PepT1 and PepT2 cotransporters. Peptides are hydrolysed inside the cell, and the amino acids are then passed into the portal blood.
Amino acids the body cannot synthesize are called essential. These cannot be synthesized in the body and must be present in the diet. Examples include valine, isoleucine, leucine, lysine, methionine, threonine, tryptophan and phenylalanine. Histidine is also normally counted as essential, and the classification of arginine is more nuanced: arginine is synthesized by mammalian cells, but at a rate that is insufficient to meet the growth needs of the body, and most of what is synthesized is cleaved to form urea. Non-essential amino acids are made mainly through transamination. Once their nitrogen has been removed, the carbon skeletons are used for making glucose, ketone bodies, or energy production. The body draws on amino acids for protein synthesis and also for nitrogen-containing molecules: nitrogen is used in DNA and RNA, neurotransmitters, and hormones.
When amino acids are present in excess of these needs, the amino group must be removed. Transamination transfers the amino group to a keto acid, and because 2-oxoglutarate participates in numerous transaminations, glutamate is a prominent intermediate in nitrogen elimination. Deamination then releases the nitrogen: glutamate is oxidatively deaminated by liver glutamate dehydrogenase, forming ammonia, which is incorporated into urea. Ammonia is toxic, and the urea cycle handles it: the cycle incorporates two amino groups into urea, a highly water-soluble waste product. The liver is the only organ capable of eliminating nitrogen from amino acids via urea synthesis, and urea is excreted by the kidneys into the urine. Mammals excrete nitrogen as urea, whereas birds excrete it as uric acid.
Nitrogen balance is the accounting framework for this process. The relationship between nitrogen entering the body and nitrogen leaving the body is called the nitrogen balance. Intake and loss in equilibrium define a normal state. A positive nitrogen balance, where intake exceeds loss, occurs in growth and pregnancy, as more protein is required for growth. A negative balance, where intake is less than loss, occurs with trauma, malnutrition, or infection, and can also be seen during short-term fasting or with ageing. Because the amount of urea excreted in urine correlates with the rate of protein metabolism, nitrogen balance serves as an indicator of nutritional status.
Inadequate protein intake is described under the heading of protein-energy undernutrition, which was previously called protein-energy malnutrition. The Merck Manual describes it as an energy deficit due to deficiency of all macronutrients, but primarily protein. Causes are classified as primary, where the condition is caused by inadequate nutrient intake, or secondary, where it results from disorders or drugs that interfere with nutrient use. Severity ranges widely: from subclinical deficiencies to obvious wasting with edema, hair loss and skin atrophy, up to starvation, and multiple organ systems are often impaired. In the severe forms, body fat and eventually visceral tissue are lost, immunity is impaired, and organ function slows.
The two classic clinical patterns reflect different deficits. In marasmus, an insufficient energy intake to match the body's requirements leads the body to draw on its own stores, resulting in emaciation. In kwashiorkor, adequate carbohydrate consumption and decreased protein intake lead to decreased synthesis of visceral proteins, and the resulting hypoalbuminemia contributes to extravascular fluid accumulation. Diagnosis typically relies on laboratory testing, including serum albumin, together with measures such as body mass index. The WHO and FAO defined the condition in 1973 as a range of pathological conditions arising from coincident lack, in varying proportions, of protein and Calories, occurring most frequently in infants and young children, and commonly associated with infections.
History of Discovery
The word "protein" entered chemistry in 1838. The Dutch chemist Gerardus Johannes Mulder reported elemental analyses of several animal and plant substances, including fibrin and egg albumin, and found that they shared nearly the same composition. He concluded, incorrectly, that they were built from one common substance. In a letter of 10 July 1838, the Swedish chemist Jöns Jacob Berzelius proposed the name, derived from the Greek proteios, "of the first rank," because he regarded these substances as the principal material of animal nutrition. Mulder also identified leucine among the products of protein breakdown, and leucine had been isolated as early as 1819. Through the nineteenth century, chemists in Europe, notably in Germany, the Netherlands and Sweden, worked to separate these materials and to find the building blocks that make them up.
The chemical logic of protein structure was developed around the turn of the twentieth century. Emil Fischer, a German chemist, synthesized amino acids and then joined them into short chains. In 1901, working with Ernest Fourneau, he prepared glycylglycine, generally regarded as the first free dipeptide. At a meeting in Karlsbad in 1902, Fischer introduced the terms "peptide" and "peptide bond" for the amide linkage that joins successive amino acids. In the same year, Franz Hofmeister independently proposed that polypeptides are chains of amino acids linked by such bonds. Fischer's 1902 Nobel Prize in Chemistry was awarded for his work on sugars and purines rather than for peptides, but his peptide studies established the chain model on which later protein chemistry rested. He also proposed the "lock and key" picture of enzyme specificity, which is discussed in the section on enzymes.
The idea that every protein has a defined sequence was demonstrated by Frederick Sanger at Cambridge. He chose insulin because it could be obtained in high purity and was medically important. Sanger broke the molecule into fragments with specific enzymes and partial acid hydrolysis, labelled the N-terminal residue of each fragment with 1-fluoro-2,4-dinitrobenzene (the "Sanger reagent"), and separated peptides by paper chromatography. He then assembled the overlapping fragments into complete sequences. The sequences of the two insulin chains, A and B, were established in the early 1950s, and the full structure, including the disulfide bonds that link the chains, was reported in 1955. Insulin contains 51 residues in two chains, A (21 residues) and B (30 residues). The work showed that a protein has a specific, reproducible sequence rather than an amorphous composition. Sanger received the 1958 Nobel Prize in Chemistry for it, and he later won a second Nobel Prize, in 1980, for DNA sequencing methods.
Structural studies followed. John Kendrew determined an atomic-resolution model of myoglobin in 1958 using X-ray crystallography, and Max Perutz worked out the structure of hemoglobin in the late 1950s and early 1960s. Kendrew and Perutz shared the 1962 Nobel Prize in Chemistry. Christian Anfinsen then addressed how a sequence becomes a folded structure. In the early 1960s, his group showed that chemically denatured ribonuclease A could regain its native structure and enzymatic activity when the denaturant was removed, which indicated that the amino acid sequence contains the information needed for folding. He shared the 1972 Nobel Prize in Chemistry with Stanford Moore and William Stein for this work. These results framed the central questions of protein science, namely how sequence determines structure and how structure determines function, which later sections of this article examine in more detail.
Evolution and Protein Families
Proteins that share a common ancestry are called homologous. Walter Fitch introduced the distinction between two kinds of homologs in the context of reconstructing species phylogenies. Orthologs are genes that diverged when a speciation event split their ancestral population, while paralogs diverged after a gene duplication within a genome. Because orthologs descend from the same gene in the last common ancestor of the species being compared, they usually retain similar functions, whereas paralogs, once duplicated, are often free to change function. The distinction is practically important: it guides protein function annotation and the reconstruction of gene trees. Gene trees and species trees can disagree, and mixing paralogous and orthologous sequences in one analysis can yield a phylogeny that is correct for the sequences but misleading for the species.
Gene duplication is a principal route to new protein functions. After a duplication, one copy can keep its original role while the other accumulates changes, a process often described as subfunctionalization when the copies divide the ancestral roles and neofunctionalization when one copy acquires a genuinely new one. Point mutations supply the raw variation, and natural selection retains the variants that improve fitness. Domain-level rearrangements, in which existing folding units are combined into new multidomain architectures, add a further mechanism, because a domain that folds independently can be recruited into a different protein without being rebuilt. A recurring observation is that structure is more conserved than sequence. Two proteins can share a fold or a functional site while their amino acid sequences have diverged so far that pairwise identity is low. Comparative analyses have shown that functional sites are often more structurally conserved than the global fold around them, which is why structure-based searches can detect relationships that sequence searches miss.
Convergent evolution produces a different kind of resemblance: unrelated proteins arrive at similar folds or catalytic arrangements. A classic case is the pair of serine proteases chymotrypsin, a vertebrate pancreatic digestive enzyme, and subtilisin, from the bacterium Bacillus subtilis. Their overall tertiary structures bear no resemblance, yet their active-site residues occupy nearly identical positions, and both enzymes evolved the same catalytic function independently. Systematic surveys suggest that such cases are not rare. A study of enzyme active sites found that transformational analogues, enzymes catalysing the same reaction with different folds, were identified for 45 of the 951 four-digit EC numbers in the Catalytic Site Atlas, and about half of these also showed convergence of their active sites. Structural classification schemes such as SCOP and CATH generally place homologous domains in a superfamily and exclude similar folds judged to have arisen by convergence, so classification itself depends on an evolutionary judgement. Because functional constraints, particularly the chemistry of catalysis, can limit the spatial arrangement of residues, convergence in active sites is a plausible outcome rather than a coincidence, although the extent to which convergence versus divergence explains a given similarity remains debated.
The early evolution of proteins is among the least certain areas of this subject. The RNA world hypothesis proposes that RNA molecules once carried both genetic information and catalytic function before DNA and proteins took over these roles. Support comes from the discovery of ribozymes and from the fact that the catalytic centre of the ribosome is RNA, which suggests that protein synthesis itself grew out of an RNA-based system. Harold White argued in the 1970s that many coenzymes, such as coenzyme A and ATP, contain nucleotide-like components, which he read as remnants of an RNA-dominated past. Some researchers have proposed that translation-related protein architectures expanded after the RNA world but before the establishment of a DNA genome, and that this expansion was driven by a feedback loop in which better proteins improved the translation system that made them. Other workers have questioned whether a functional water-soluble globular protein could have arisen by chance from an RNA-only system, and they have emphasised the physicochemical properties of amino acids and oligopeptides as drivers of early selection. Laboratory work continues to test individual steps; for example, a 2025 study reported a simple, selective way to link amino acids to RNA in neutral water. Such experiments make specific steps plausible, but they cannot reconstruct the actual sequence of events, and how the genetic code and the DNA genome arose remains unresolved.
Applications in Medicine and Industry
Proteins are used as drugs, vaccine components, catalysts and analytical reagents. Most of this use depends on recombinant DNA technology, in which a gene is inserted into a host organism that then produces the encoded protein in quantity.
Recombinant insulin was the first such product. Before it, people with diabetes were treated with insulin purified from pig and cattle pancreases, which occasionally caused serious immune reactions. In 1978, a Genentech team led by David Goeddel expressed the insulin A and B chains separately in *Escherichia coli* and combined them. Genentech and Eli Lilly then worked together on scale-up, and the first rDNA insulins were marketed in 1982 under the name Humulin. Later manufacturing also used yeast (Saccharomyces cerevisiae), and the precursor is folded and processed differently in each host. Recombinant insulin is now one of the main insulin sources worldwide, and engineered analogues with altered absorption profiles are widely used.
Monoclonal antibodies form the largest class of protein therapeutics. The first one approved by the US Food and Drug Administration, muromonab, dates from 1986. Early rodent-derived antibodies were often recognized as foreign by patients and provoked anti-antibody responses, which limited their use. Protein engineering addressed this in stages: chimeric antibodies combine mouse variable domains with human constant domains, humanized antibodies graft the mouse antigen-binding loops (CDRs) onto a human framework, and fully human antibodies are selected from phage-display libraries or transgenic mice. Grafting often lowers affinity, so a few key framework residues are restored to preserve the original loop conformation. By September 2020, 93 therapeutic monoclonal antibodies had been approved by the FDA, and antibody-based products accounted for roughly half of biopharmaceutical sales. Such monoclonal antibodies are used for cancer, autoimmune disease and infection, and related formats such as antibody fragments and antibody–drug conjugates extend the approach.
Protein-based vaccines generally use purified subunits rather than whole pathogens. Because they contain no live organism, they are considered safer and more homogeneous, and they can be produced with standard expression systems. An early example is the hepatitis B vaccine, whose surface antigen is made in yeast. Subunit antigens are often weakly immunogenic on their own, so they are commonly formulated with adjuvants or displayed on particles such as virus-like particles or self-assembling nanoparticles, which present repeated epitopes to the immune system. Antigen design increasingly relies on structural data from X-ray crystallography, electron microscopy and NMR. Engineered antigens can keep the conformation of an epitope, or combine epitopes from different proteins into a single chimeric antigen; in one meningococcal B study, such constructs elicited antibodies against both component proteins in mice. Several SARS-CoV-2 vaccines based on the spike protein or its receptor-binding domain were approved during the COVID-19 pandemic.
Industrial enzymes are used in large volumes. Proteases, lipases and amylases are added to detergents to remove protein, fat and starch stains at lower wash temperatures. In food processing, enzymes modify starch, clarify juices and beverages, and assist in dairy and baking. In biofuel production, cellulases and hemicellulases break down plant cell-wall polysaccharides into fermentable sugars. Industrial enzymes are typically optimized by directed evolution or rational design to tolerate high temperature, extreme pH, organic solvents or high substrate loads, and to retain activity over long storage periods. Improvements in stability often trade off against catalytic rate, so the choice of variant depends on the process conditions.
Proteins also serve as recognition elements in biosensors and diagnostic assays. Enzyme-based electrodes, such as those using glucose oxidase, underlie the glucose meters used by people with diabetes. Antibody-based immunoassays, including the enzyme-linked immunosorbent assay (ELISA), detect protein biomarkers such as hormones, cytokines and cardiac troponin in serum or other clinical samples. Sandwich-format ELISAs with two antibodies are also used for quality control, for example to measure the antigen content of a vaccine batch. Lateral-flow tests apply the same principle on paper strips for rapid point-of-care use. Their performance depends on antibody specificity, cross-reactivity and the sample matrix, and results are usually confirmed by laboratory methods before diagnosis.
Disordered Proteins and Open Questions
Intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs) do not settle into one stable three-dimensional fold, yet they carry out essential biological roles. Reviews describe them as dynamic conformational ensembles rather than fixed structures, a property that lets them form high-specificity, low-affinity interactions, multivalent regulatory functions, and context-dependent binding. Many IDPs also fold only when they bind a partner, a behavior known as coupled folding and binding, and some keep an open, flexible conformation even in the bound state, so one disordered region can contact several partners at once (a single IDR can bind several partners simultaneously). Because the conformational state is tuned by post-translational modifications, the same segment can be ordered in one cellular setting and disordered in another. IDPs therefore challenge the classical structure–function paradigm, which holds that function follows from a specific fold determined by sequence, and they have become a major topic in protein folding research. Their flexibility also carries risk: proteins linked to neurodegeneration, such as alpha-synuclein and tau, are both functionally disordered and prone to aggregation.
Predicting how disordered and folded proteins reach their final states remains difficult. Experimental work on unbound IDPs is hard because they are not captured by approaches designed for well-ordered crystals, and a bound-state structure alone does not explain the regulatory mechanism. Computational work faces a related problem. AlphaFold and similar systems predict one static model per sequence, and proteins such as α-synuclein, Tau, and the p53 transactivation domain are returned as fragmented, low-confidence predictions in these tools. Newer diffusion-based models can also introduce spurious order into disordered regions, and the same model typically predict[s] static structures as seen in the PDB, not the dynamical behaviour of biomolecular systems in solution. Several groups have tried to coax AlphaFold into producing multiple conformations by altering its input alignments, but their success has varied by target, and reviewers note that the gap between predicted structures and the continuous motion of proteins in living cells remains wide.
Membrane proteins raise their own open questions. Cryo-EM has made it possible to observe membrane proteins in near-native states and to capture several conformations of a single protein, which has clarified allosteric regulation and drug action. Interpretation is still uncertain, however: detergent or lipid densities in the maps are largely unclear in how they should be modeled, and the molecular identity of lipid-like densities is often difficult to establish. Predictive tools also struggle here, since membrane proteins depend on lipid interactions and electrostatic environments that are poorly represented in their training data. Similar gaps apply to proteins in the crowded cytoplasm, where crowding, cofactors and modifications shape behavior in ways that single-chain models do not capture. Methods that combine molecular dynamics, NMR relaxation data and structure prediction are being developed to describe these ensembles more fully.
The design of proteins with new chemistries, which lies beyond natural amino acid sets, is another frontier. Current design methods are most reliable for folds and functions that resemble natural examples, and extending them to unfamiliar catalytic chemistry or dynamic behavior is an active research goal.
Debate also continues over how much misfolding contributes to disease. Aggregates are clearly associated with several neurodegenerative conditions, but researchers disagree about whether aggregation is usually a cause, a consequence, or a protective response. Structure-based models of these processes are hard to validate, because the relevant intermediates are transient and heterogeneous, and because computed ensembles are difficult to compare directly with experimental observables. Progress therefore depends on better experimental measurements of conformational dynamics and on tests that link predicted ensembles to behavior in cells and tissues.
References
- Multiple evidence strands suggest that there may be as few as 19 000 human protein-coding genesacademic.oup.com
- Identification of Small Open Reading Frame-encoded Proteins in the Human Genomencbi.nlm.nih.gov
- Sandwalk: How many protein-coding genes in the human genome? (2)sandwalk.blogspot.com
- How many protein-coding genes are contained in the human ...vaia.com
- Human genomesciencedaily.com
- Amino Acidswisc.pb.unizin.org
- Q4. Classify the 20 standard amino a... [FREE SOLUTION]vaia.com
- IMGT classes of the 20 common amino acidsimgt.org
- Variant LovD polypeptides and their usesimage-ppubs.uspto.gov
- Biochemistry Study Guide: Amino Acids, Peptide Bonds & Propertiespearson.com
- Biochemistry Proteins Study Guide: Essential Concepts & Structurepearson.com
- Biochemistry Study Guide: Amino Acids & Peptide Bondspearson.com
- Hotspot-Driven Peptide Design via Multi-Fragment Autoregressive Extensionarxiv.org
- Structure and Properties of Peptides: Biochemistry Study Notespearson.com
- Peptide Bond - an overviewsciencedirect.com
- Amino Acids, Peptides, and Protein Structure (5D)varsitytutors.com
- Peptide bonden.wikipedia.org
- 20 Amino Acids: Structure, Classification & Abbreviationproteinstructures.com
- Biochemistry Protein Secondary Structure Study Guidepearson.com
- Alpha helix - Proteopedia, life in 3Dproteopedia.org
- Secondary Structure and Backbone Conformationswissmodel.expasy.org
- A Comprehensive System for Secondary Structure Analysis of Protein Modelsarxiv.org
- Conformation of Proteins: Secondary Structure - CSIR NET LIFE SCIENCE COACHINGletstalkacademy.com
- Directed evolution of proteinsimage-ppubs.uspto.gov
- Secondary Structurednastar.com
- Improved Chou-Fasman method for protein secondary structure predictionbmcbioinformatics.biomedcentral.com
- Prediction of protein secondary structure based on residue pairsarxiv.org
- Chou fasman algorithm for protein structure predictionslideshare.net
- Protein Misfolding and Degenerative Diseasesnature.com
- Levinthal's paradoxen.wikipedia.org
- Anfinsen's dogmasimple.wikipedia.org
- Levinthal%27s paradoxen.wikipedia.org
- How Do Proteins Fold?arxiv.org
- Folding funnelen.wikipedia.org
- Lecture 9. Protein stability and foldingguweb2.gonzaga.edu
- Protein Folding and Chaperones — NEET-PG Revision Notes 2026getoncourse.ai
- The concept of protein folding/unfolding and its impacts on human health - ScienceDirectsciencedirect.com
- RNA Processing — Explain RNA processing (capping, splicing,...varsitytutors.com
- Gene Expression: RNA Processingedu.abi.am
- mRNA Metabolism in Health and Diseasencbi.nlm.nih.gov
- Modeling ribosome dynamics to optimize heterologous protein productionimage-ppubs.uspto.gov
- Multivalent macrolide antibioticsimage-ppubs.uspto.gov
- 11.4: Translation - Biology LibreTextsbio.libretexts.org
- Steps of Translationcourses.lumenlearning.com
- Biology, Genetics, Genes and Proteins, Ribosomes and Protein Synthesisoertx.highered.texas.gov
- Translation in prokaryotesrotel.pressbooks.pub
- Continued Protein Synthesis at Low [ATP] and [GTP] Enables Cell Adaptation during Energy Limitation - PMCpmc.ncbi.nlm.nih.gov
- What is Translation in Biology: AP® Biology Reviewalbert.io
- Ubiquitin modificationsncbi.nlm.nih.gov
- Ubiquitination Is a Novel Post-Translational Modification of VMP1 in Autophagy of Human Tumor Cellsncbi.nlm.nih.gov
- Protein degradation Guide to the proteasome and autophagy systemsresources.revvity.com
- Post-Translational Modification of Cellular Proteins by Ubiquitin and Ubiquitin-Like Molecules: Role in Cellular Senescence and Ageing - Madame Curie Bioscience Database - NCBI Bookshelfncbi.nlm.nih.gov
- Structural Biochemistry/Enzyme Regulation/Proteolytic activation - Wikibooks, open books for an open worlden.wikibooks.org
- Proinsulin - an overviewsciencedirect.com
- Autocatalytic Activation of the Furin Zymogen Requires Removal of the Emerging Enzyme's N-Terminus from the Active Sitencbi.nlm.nih.gov
- Proc. Natl. Acad. Sci. USA Vol. 73, No. 11, pp. 3825-3832, November 1976pnas.org
- A systematic comparison of protein structure classifications: SCOP, CATH and FSSP: Structurecell.com
- Impact of structure space continuity on protein fold classificationnature.com
- CATH - an overviewsciencedirect.com
- Origin and Evolution of Protein Fold Designs Inferred from Phylogenomic Analysis of CATH Domain Structures in Proteomesjournals.plos.org
- The value of protein structure classification information—Surveying the scientific literatureonlinelibrary.wiley.com
- A systematic comparison of protein structure classifications: SCOP, CATH and FSSP - ScienceDirectsciencedirect.com
- Class Architecture Topology Homology - an overviewsciencedirect.com
- Pfam Protein Families Databaseacademic.oup.com
- (PDF) Pfam: Multiple sequence alignments and HMM-profiles of protein domainsresearchgate.net
- Pfamen.wikipedia.org
- Transgenic plants with enhanced agronomic traitsimage-ppubs.uspto.gov
- Protein Data Bank: A Comprehensive Review of 3D Structure Holdings and Worldwide Utilization by Researchers, Educators, and Studentsncbi.nlm.nih.gov
- Worldwide Protein Data Bank biocuration supporting open access to high-quality 3D structural biology datancbi.nlm.nih.gov
- RCSB PDB: Homepagercsb.org
- The Pfam protein families database - PubMedpubmed.ncbi.nlm.nih.gov
- Protein Data Bank: the single global archive for 3D macromolecular structure datancbi.nlm.nih.gov
- Transgenic plants with enhanced agronomic traitsimage-ppubs.uspto.gov
- Protein structure determination by X-ray crystallographyelearning.unite.it
- Sample delivery methods for protein X-ray crystallography with a special focus on sample consumptionnature.com
- Growing Crystals for X-ray Free-Electron Laser Structural Studies of Biomolecules and Their Complexesncbi.nlm.nih.gov
- Structural Biochemistry/Proteins/X-ray Crystallography - Wikibooks, open books for an open worlden.wikibooks.org
- X-Ray Powder Diffraction from Sub-Micron Crystals of Photosystem-1 Membrane Proteinarxiv.org
- The 2017 Nobel Prize in Chemistry: cryo-EM comes of ageresearchgate.net
- Frontiersfrontiersin.org
- Membranes under the Magnetic Lens: A Dive into the Diverse World of Membrane Protein Structures Using Cryo-EMpubs.acs.org
- A cold, clear view of life wins chemistry Nobelscience.org
- Isotopically-labeled proteome standardsimage-ppubs.uspto.gov
- Highlights on the capacities of "Gel-based" proteomicsncbi.nlm.nih.gov
- Quantitative Mass Spectrometry-based Proteomicsintechopen.com
- Quantifying Proteins by Mass Spectrometrychromatographyonline.com
- Whole gel processing procedure for GeLC-MS/MS based proteomicsproteomesci.biomedcentral.com
- System and methods for navigating and visualizing multi-dimensional biological dataimage-ppubs.uspto.gov
- Nobel Prize in Chemistry 2024: Revolutionising Protein Structure Prediction and Designkinesismagazine.com
- (PDF) Applying and improving AlphaFold at CASP14researchgate.net
- Work on protein structure and design wins the 2024 chemistry Nobelsciencenews.org
- AlphaFold2 at CASP14: Breakthrough in Protein Foldingemergentmind.com
- Improving AlphaFold2-based Protein Tertiary Structure Prediction with MULTICOM in CASP15biorxiv.org
- CASP16 Protein Monomer Structure Prediction Assessmentncbi.nlm.nih.gov
- Refinement of AlphaFold-Multimer structures with single sequence inputbiorxiv.org
- Nobel Prize in Chemistry 2024nature.com
- CASP14: what Google DeepMind’s AlphaFold 2 really achieved, and what it means for protein folding, biology and bioinformaticsblopig.com
- Nobel Laureate Frances Arnold Explains the Breakthrough Process of Directed Evolution — Caltech Magazinemagazine.caltech.edu
- Frances Arnold Wins 2018 Nobel Prize in Chemistry - www.caltech.educaltech.edu
- Press release: The Nobel Prize in Chemistry 2024 - NobelPrize.orgnobelprize.org
- Frontiersinfrontiersin.org
- Evaluation of Deep Neural Network ProSPr for Accurate Protein Distance Predictions on CASP14 Targetsncbi.nlm.nih.gov
- Molecule of the Month: Directed Evolution of Enzymes - PDB-101pdb101.rcsb.org
- 6.2: Digestion and Absorption of Proteins - Medicine LibreTextsmed.libretexts.org
- Pepsin A - an overviewsciencedirect.com
- Protein Digestion and Amino Acid Absorptionbasicmedicalkey.com
- NITROGEN METABOLISM Nitrogen Fixation and Assimilation Amino Acid Biosynthesisuomustansiriyah.edu.iq
- Nitrogen Metabolism and the Urea Cycle - The Medical Biochemistry Pagethemedicalbiochemistrypage.org
- Proteins: Metabolismopen.oregonstate.education
- Protein & Amino Acid Metabolism: The Urea Cyclesimplemed.co.uk
- Nitrogen Metabolism - an overviewsciencedirect.com
- Common mechanisms of amino acids conversion, deamination, transamination, nitrogen balance - WikiLectureswikilectures.eu
- Protein-Energy Undernutrition (PEU) - Nutrition - Merck Manual Professional Editionmerckmanuals.com
- Protein-Energy Malnutrition: Background, Pathophysiology, Etiologyemedicine.medscape.com
- The Young Child: Protein-Energy Malnutrition - Springer Naturelink.springer.com
- Protein Digestion and Absorptionopenoregon.pressbooks.pub
- Chemical Digestion and Absorption: A Closer Lookcourses.lumenlearning.com
- Protein Digestion and Absorptionmedia.lanecc.edu
- Protein Energy Malnutrition in India: The Plight of Our Under Five Children - PMCpmc.ncbi.nlm.nih.gov
- PROTEIN ENERGY MALNUTRITIONslideshare.net
- Protein, the business - by Ms. An Daofullhousesnotes.substack.com
- Protein - New World Encyclopedianewworldencyclopedia.org
- Jamanetworkjamanetwork.com
- Proteinen.wikipedia.org
- Gerardus Johannes Mulderen.wikipedia.org
- History of Protein Discoveryslideshare.net
- Protein Chemisty Timelineumass.edu
- Franz Hofmeisteren.wikipedia.org
- Hermann Emil Fischer - New World Encyclopedianewworldencyclopedia.org
- Emil Fischerbritannica.com
- Emil Fischersciencehistory.org
- One hundred years of peptide chemistryresearchgate.net
- Emil Fischeren.wikipedia.org
- Sanger Determines the Structure of Insulinebsco.com
- Professor Frederick Sangerwhatisbiotechnology.org
- Frederick Sangeren.wikipedia.org
- Insulin's Primary Amino Acid Structureinnovation.world
- Unraveling Insulinscience.org
- The first sequence: Fred Sanger and insulinresearchgate.net
- Inferring Orthology and Paralogylink.springer.com
- The Theory of Gene Family Historiesarxiv.org
- Disentangling homology, orthology, paralogy and similarity with BLASTsequenceserver.com
- Paralogy - an overviewsciencedirect.com
- Detecting evolutionary relationships across existing fold space, using sequence order-independent profile–profile alignmentspnas.org
- Convergent Evolution of Enzyme Active Sites Is not a Rare Phenomenon - ScienceDirectsciencedirect.com
- Convergent evolution of enzyme active sites is not a rare phenomenon - PubMedpubmed.ncbi.nlm.nih.gov
- A Structure-Centric View of Protein Evolution, Design and Adaptationarxiv.org
- Convergent Evolution - an overviewsciencedirect.com
- The evolution and functional repertoire of translation proteins following the origin of lifencbi.nlm.nih.gov
- Cofactors are Remnants of Life’s Origin and Early Evolutionncbi.nlm.nih.gov
- Pseudo-Replication of [GADV]-Proteins and Origin of Lifencbi.nlm.nih.gov
- Early Life: Embracing the RNA World - ScienceDirectsciencedirect.com
- Origin Of Life Breakthrough: How RNA Might Have Started To Make Proteins On Early Earth - Astrobiologyastrobiology.com
- History of insulin - PMC - NIHpmc.ncbi.nlm.nih.gov
- Equivalent Recombinant Human Insulin Preparations and their Place in Therapytouchendocrinology.com
- Downstream processing of recombinant human insulin and its analogues production from E. coli inclusion bodies - PMCpmc.ncbi.nlm.nih.gov
- Insulin Production -Dr. Ekta Khare Department of Microbiologygyansanchay.csjmu.ac.in
- Recombinant DNA the Bio-Revolution, Between Promise, Hurdles, and Achievementsncbi.nlm.nih.gov
- Humanization of antibodies using a machine learning approach on large-scale repertoire databiorxiv.org
- Humanized antibodiesimage-ppubs.uspto.gov
- Methods for antibody engineeringimage-ppubs.uspto.gov
- Subunit Vaccine Design Services - Creative Biolabscreative-biolabs.com
- The Versatile Manipulations of Self-Assembled Proteins in Vaccine Designncbi.nlm.nih.gov
- Structure-based design of chimeric antigens for multivalent protein vaccinesnature.com
- Development of an ELISA Assay for the Determination of SARS-CoV-2 Protein Subunit Vaccine Antigen Contentncbi.nlm.nih.gov
- Advances in the structure-based design of protein vaccines for respiratory infectious diseases - PMCpmc.ncbi.nlm.nih.gov
- Beyond the structure-function paradigm: A comprehensive review of intrinsically disordered proteins - PMCpmc.ncbi.nlm.nih.gov
- (PDF) Beyond the structure-function paradigm: A comprehensive review of intrinsically disordered proteinsresearchgate.net
- Advantages and Limitations of AlphaFold in Structural Biology: Insights from Recent Studieslink.springer.com
- Accurate structure prediction of biomolecular interactions with AlphaFold 3nature.com
- Automated simulation-based membrane-protein refinement into cryo-EM databiorxiv.org
- Single-particle Cryo-EM structure determination for membrane proteins - PMCpmc.ncbi.nlm.nih.gov
- Intrinsically disordered proteins: modes of binding with emphasis on disordered domainsncbi.nlm.nih.gov
- Intrinsically Disordered Proteins in a Physics-Based Worldncbi.nlm.nih.gov
- Modeling protein conformational ensembles by guiding AlphaFold2 with Double Electron Electron Resonance (DEER) distance distributionsnature.com