📖 Core Concepts
Core Concepts: Advanced Protein Biochemistry: Structure and Function
1. What Is This Subject Really About?
You have already met proteins many times: as catalysts in CHEM401 and CHEM402, as products of gene expression in BIOL201, as antibodies in BIOL401, as ion channels in BIOL501, and as metal-binding scaffolds in CHEM502. In each of those courses the protein was mostly treated as a given: an object with a structure that did a job. This course asks the questions underneath. Why does a linear chain of amino acids adopt one particular three-dimensional shape? Why that shape and not another? How does the shape produce the function? How does evolution explore the space of possible shapes?
Stated as compactly as possible, protein biochemistry is the study of a three-step mapping:
sequence ──(physics + evolution)──► structure ensemble ──(chemistry + dynamics)──► function
Each arrow hides deep science. The first arrow is the folding problem. It is a statistical-mechanics problem of the kind you met in CHEM301, applied to a heteropolymer with roughly 1040–10100 accessible conformations. The second arrow is the mechanism problem: how a particular arrangement of atoms lowers an activation barrier, transmits an allosteric signal, or recognizes a partner among thousands of look-alikes.
Notice the word ensemble in the middle. One of the major intellectual shifts of the last thirty years is the realization that a protein is not a single structure. It is a Boltzmann-weighted population of conformations. The PDB coordinates you download are a time- and space-averaged snapshot of the most populated basin. Some proteins (the intrinsically disordered ones, Week 8) have no single dominant basin at all.
Here is an analogy to carry through the course. A protein sequence is like the rules of a game, and the native structure is like the stable strategy that emerges when those rules are played out under the constraints of water, temperature, and pH. Anfinsen's insight was that the rules alone are sufficient. You do not need an external blueprint. AlphaFold's insight was that evolution has already played the game billions of times, and the record of those plays (multiple sequence alignments) contains enough information to infer the outcome without simulating the physics.
"…the three-dimensional structure of a native protein in its normal physiological milieu (solvent, pH, ionic strength, presence of other components such as metal ions or prosthetic groups, temperature, and other) is the one in which the Gibbs free energy of the whole system is lowest; that is, that the native conformation is determined by the totality of interatomic interactions and hence by the amino acid sequence, in a given environment."
— [Anfinsen, 1973]
Read that sentence carefully, because almost every word matters. "Of the whole system" means protein plus solvent. This is why the hydrophobic effect, a property of water, dominates folding. "In a given environment" is the escape clause that later accommodates chaperones, membranes, cofactors, and IDPs that fold only upon binding. The thermodynamic hypothesis is a statement about the destination, not the route.
Finally, appreciate the historical arc. When [Kendrew et al., 1958] solved myoglobin at 6 Å resolution, the overwhelming reaction was surprise at its irregularity. Researchers had hoped for crystalline-like symmetry and found something that looked like "a sausage tied in knots." Six decades later, the Protein Data Bank holds over 200,000 experimental structures [Berman et al., 2000], and the AlphaFold Database holds over 200 million predicted ones [Jumper et al., 2021]. That irregularity turned out to be governed by deep regularities: a limited set of backbone geometries, a finite vocabulary of folds, and a small number of physical forces in delicate balance. This chapter is about those regularities.
2. Concept One: The Peptide Bond Is a Rigid Plate, and Proteins Are Chains of Plates
Everything in protein structure follows from the geometry of the backbone. The single most important fact is this: the peptide bond is planar.
Recall amide resonance from CHEM201/202. The nitrogen lone pair delocalizes into the carbonyl π* orbital:
O O⁻
‖ |
─C─N─ ⟷ ─C═N⁺─
| |
H H
C–N bond length: ~1.33 Å (cf. C–N single ~1.47 Å, C═N double ~1.27 Å)
Rotational barrier about C–N: ~80 kJ/mol (~20 kcal/mol)
The result is a C–N bond with roughly 40% double-bond character. Six atoms (Cαi, C, O, N, H, Cαi+1) are locked into a rigid plane. The backbone is therefore not a floppy string of beads. It is better pictured as a chain of flat playing cards connected at their corners by swivel joints, with the swivels at the Cα atoms. Only two torsions per residue are genuinely free:
- φ (phi): rotation about N–Cα, defined by atoms C'i−1–Ni–Cαi–C'i
- ψ (psi): rotation about Cα–C', defined by atoms Ni–Cαi–C'i–Ni+1
- ω (omega): rotation about the peptide C–N itself, essentially restricted to ~180° (trans) or ~0° (cis)
This reduction is enormous. A protein of N residues has roughly 2N backbone degrees of freedom rather than many times that number. The entire problem of backbone conformation becomes a problem of choosing (φ, ψ) pairs.
Cis versus trans, and the special case of proline
For a typical X–Y peptide bond, trans is favored by roughly 1000:1. The Cα atoms on either side would collide in the cis form. Proline breaks this rule. Its side chain loops back to bond its own backbone nitrogen, so the atom on the "other side" of the X–Pro bond is a Cδ carbon either way. Steric discrimination is therefore much weaker:
- In unstructured peptides, roughly 10–30% of X–Pro bonds are cis (the exact value is sequence-dependent; aromatic residues preceding Pro raise it).
- In folded protein structures, about 5% of X–Pro bonds are cis, compared with roughly 0.03% of non-Pro peptide bonds.
- Cis/trans isomerization has a ~80 kJ/mol barrier, so it is slow (seconds to minutes at room temperature). It is frequently the rate-limiting step in refolding. Cells use peptidyl-prolyl isomerases (cyclophilins, FKBPs, parvulins, and trigger factor, which you'll see in Week 12) to catalyze it.
This connects to biology you already know. Cyclosporin A and FK506 (tacrolimus) are immunosuppressants (BIOL401) that bind cyclophilin and FKBP12. Their immunosuppressive effect, however, comes from the drug–immunophilin complex inhibiting calcineurin. It does not come from blocking isomerase activity itself. This is a nice example of a protein being repurposed as a "presenter" for a small molecule.
Ionization: Henderson–Hasselbalch in a heterogeneous dielectric
You know the Henderson–Hasselbalch equation from CHEM102 and CHEM302:
pH = pKa + log₁₀( [A⁻] / [HA] ) fraction deprotonated: f = 1 / (1 + 10^(pKa − pH))
The graduate-level subtlety is that a pKa is a property of the group in its environment, not of the group alone. The tabulated "model compound" pKa values describe a residue fully exposed to water. Typical values are about 3.9 for Asp, 4.3 for Glu, 6.0 for His, 8.3 for Cys, 10.1 for Tyr, 10.5 for Lys, and 12.5 for Arg, plus about 3.6 for the C-terminus and 8.0 for the N-terminus in peptides. Inside a protein, three factors shift those values:
- Desolvation (Born energy). Burying a charge in a low-dielectric interior (ε ≈ 2–20) is costly, so the group prefers to stay neutral. Buried Lys and Glu residues in engineered variants of staphylococcal nuclease show pKa shifts of several units, with buried Lys pKa values dropping toward 6.
- Local electrostatics. A nearby positive charge stabilizes a carboxylate and lowers its pKa. A nearby negative charge raises it.
- Hydrogen bonding. An H-bond to the deprotonated form lowers the pKa, while an H-bond to the protonated form raises it.
This is not a curiosity. Catalysis depends on it. The catalytic His of serine proteases (Week 13), the perturbed Glu35 of lysozyme, and the pH-dependent conformational switches of viral fusion proteins all rely on pKa values that the protein has tuned away from their aqueous defaults.
Chirality
All ribosomally incorporated amino acids except glycine are L (S configuration at Cα, except cysteine, which is R because of the CIP priority of sulfur. This is a reminder that L/D and R/S are different conventions). Biology does use D-amino acids. They appear in bacterial peptidoglycan (D-Ala, D-Glu, which you met in BIOL301), in nonribosomal peptide antibiotics such as vancomycin and gramicidin, in some animal venom peptides through post-translational epimerization, and in neural signaling (D-serine as an NMDA receptor co-agonist, BIOL501). The homochirality of the backbone is why the Ramachandran plot is asymmetric and why α-helices are almost universally right-handed.
3. Concept Two: The Ramachandran Plot Is a Map of What Is Sterically Possible
[Ramachandran, Ramakrishnan & Sasisekharan, 1963] did something beautifully simple. They modeled atoms as hard spheres with van der Waals contact radii, rotated φ and ψ through 360° each, and asked which combinations caused clashes. The answer was striking. Only a minority of (φ, ψ) space is accessible to a non-glycine residue. The allowed regions cluster into a few islands:
Region approx. φ approx. ψ ───────────────────────────────────────────────────────── Right-handed α-helix (αR) −57° −47° Antiparallel β-sheet −139° +135° Parallel β-sheet −119° +113° Polyproline II (PPII) −75° +145° Left-handed helix (αL) +57° +47° (rare except Gly, Asn, Asp)
Why does this matter so much? Because secondary structure is not imposed on proteins. It falls out of the Ramachandran plot. Take a residue sitting in the αR basin and repeat it, and you get an α-helix. Repeat a residue in the β basin and you get an extended strand. The regular structures are simply what happens when the same allowed (φ, ψ) is iterated.
Two residues are special:
- Glycine has only H as its side chain. Its Ramachandran plot is far larger and symmetric about the origin, so it can access αL and other regions forbidden to L-residues. This is why Gly is enriched in tight turns, and why it is mandatory at every third position in collagen (Concept Eight).
- Proline has its φ locked near −63° ± 15° by the pyrrolidine ring. It cannot donate a backbone H-bond. It is a helix breaker in the middle of helices but common at the first turn (N-cap +1 position). It is also a key component of PPII helices, which turn out to be extremely important ligands for SH3 and WW domains (Week 9).
The Ramachandran plot remains the first validation tool for any experimental or predicted model. Residues in disallowed regions signal either an error or a strained conformation that the protein pays for and usually uses functionally. Strained residues are often found in active sites, where stored conformational energy can be spent on catalysis.
4. Concept Three: Secondary Structure Is Hydrogen-Bond Bookkeeping
[Pauling, Corey & Branson, 1951] predicted the α-helix before any protein structure was solved. They did it by insisting on planar peptides and maximal backbone hydrogen bonding. That prediction is one of the triumphs of structural chemistry.
The α-helix
H-bond pattern: C═O (residue i) ··· H–N (residue i+4) Residues per turn: 3.6 Rise per residue: 1.5 Å Pitch: 3.6 × 1.5 = 5.4 Å Nomenclature: 3.6₁₃ (13 atoms in the H-bonded ring) Handedness: right-handed (for L-amino acids) Side chains: point outward and slightly toward the N-terminus
Why does "3.6 residues per turn" matter? Because 360°/3.6 = 100° per residue. A residue and the residues three and four positions later all sit on roughly the same face. This geometry is the basis of amphipathic helices. Place hydrophobic residues at i, i+3, i+4, i+7, i+11 and you get a helix with one greasy face and one polar face. The helical wheel projection makes this pattern visible. Amphipathic helices appear in antimicrobial peptides (magainins, LL-37), in apolipoproteins that wrap lipid particles, in membrane-curvature sensors (ALPS motifs), and in the dimerization interfaces of coiled coils (Concept Eight). The same geometry, slightly modified, gives the heptad repeat.
The helix dipole. Each peptide unit has a dipole of roughly 3.5 D, and all point the same direction along a helix. This produces a "macrodipole," often approximated as about +½ unit charge at the N-terminus and −½ at the C-terminus [Hol et al., 1978]. The effective magnitude in solution is debated and screened by solvent. The practical consequences are robust, though. The first four N–H groups and the last four C═O groups of a helix lack intrahelical partners. Proteins therefore place "capping" residues to satisfy them: Ser, Thr, Asp, and Asn at the N-cap, and Gly often at the C-cap (C′). Phosphate- and anion-binding sites, such as the P-loop and the Rossmann fold's dinucleotide-binding site, are frequently located at helix N-termini.
The 3₁₀ and π helices
- 3₁₀ helix: i→i+3 H-bonds, 3.0 residues per turn, tighter and slightly strained. It is common as single turns at helix ends.
- π-helix: i→i+5 H-bonds, about 4.4 residues per turn. It was once considered vanishingly rare but is now recognized in roughly 15% of proteins, usually as a single-turn "π-bulge" inserted into an α-helix. These insertions are often at functional sites, which suggests they evolved for function rather than stability.
β-Sheets
In a β-strand the backbone is nearly fully extended (about 3.3–3.5 Å rise per residue), and side chains alternate up and down. In contrast to the helix, the H-bonds are between strands, which may be distant in sequence. This has a major consequence: β-sheets are inherently nonlocal structures. They are harder to predict from local sequence, they fold more slowly on average, and they are the structural basis of amyloid when misassembled (Concept Eight).
- Antiparallel sheets have nearly linear, alternately narrow- and wide-spaced H-bond pairs. They are often slightly more stable and are frequently connected by tight hairpins.
- Parallel sheets have angled H-bonds. The strand connections must be long crossovers, almost always right-handed βαβ units, and parallel sheets are usually buried with helices packed on both faces (as in the Rossmann fold and TIM barrel).
- Twist. Viewed along the strands, real sheets have a left-handed twist (equivalently described as a right-handed twist when viewed perpendicular to the strands; know which convention a source uses). The twist is a direct consequence of L-chirality and is why β-barrels and β-propellers close up the way they do.
Turns
About a third of residues in globular proteins are in turns and loops, so these are not "leftovers." A β-turn involves four residues (i to i+3), with a C═O(i)···H–N(i+3) H-bond in the classic definition. The types are distinguished by the (φ, ψ) of residues i+1 and i+2:
- Type I: the most common. Pro is favored at i+1.
- Type II: Gly is strongly favored at i+2, because that residue needs a positive φ.
- Types I′ and II′: mirror images of types I and II. They require positive φ, so they are Gly- or Asn-rich, and they are especially common in tight β-hairpins because their twist matches the sheet twist.
- γ-turns: three residues with an i→i+2 H-bond.
Peptide chemists exploit this. Inserting D-Pro–Gly (which mimics a type II′/I′ geometry) is a standard trick to nucleate β-hairpins in designed peptides.
Prediction
Early algorithms (Chou–Fasman, GOR) used single-residue propensities and topped out around 60–65% three-state accuracy. PSIPRED [Jones, 1999] pushed accuracy to around 80% by feeding PSI-BLAST position-specific scoring matrices into neural networks. The key insight was that evolutionary profiles carry more information than the single sequence. That same insight, taken to its logical extreme, became AlphaFold.
5. Concept Four: Folding Is Driven by Water, and Stability Is Marginal
This is the concept most often misunderstood, so slow down here.
The thermodynamic balance sheet
For the equilibrium N ⇌ U (native ⇌ unfolded):
ΔG_unf = −RT ln K_unf = −RT ln ( [U] / [N] ) ΔG_unf = ΔH_unf − TΔS_unf
For a typical small globular protein at 25 °C, ΔGunf is only about +20 to +60 kJ/mol (5–15 kcal/mol). That is equivalent to a handful of hydrogen bonds. Yet the folded state is stabilized by hundreds of favorable contacts, and the unfolded state by an enormous chain entropy. Protein stability is the small difference between very large opposing terms. This is why single point mutations can unfold a protein, and why stability measurements are notoriously sensitive.
An analogy may help. Imagine two enormous armies of nearly equal size. The battle's outcome (folded versus unfolded) is decided by a margin of a few soldiers. Adding or removing one platoon (a mutation) can flip it. This marginality is probably selected, not accidental. Proteins need to be flexible enough to function and to be degraded on schedule (Week 12).
The hydrophobic effect
[Kauzmann, 1959] argued that the dominant driving force for folding is the burial of nonpolar side chains. The crucial insight is that this is not an "attraction" between greasy groups. It is a property of water. Water around a small nonpolar solute cannot form hydrogen bonds toward it, so it reorganizes into a more ordered hydration shell. At room temperature, the transfer of a hydrocarbon from water to oil has these features:
- ΔH ≈ 0 (or slightly unfavorable), so it is not enthalpy-driven at 25 °C.
- ΔS is large and positive. The released water is the source of that entropy gain.
- ΔCp is large and positive for exposing nonpolar surface. This is the signature of the hydrophobic effect.
The magnitude is approximately ~100–120 J mol−1 Å−2 (roughly 25 cal mol−1 Å−2) of buried nonpolar surface. Burying one Leu side chain is therefore worth on the order of 10 kJ/mol.
ΔCp and the stability curve
Because unfolding exposes nonpolar surface, ΔCp,unf is positive. Empirically it scales with the change in solvent-accessible surface area [Myers, Pace & Scholtz, 1995]:
ΔCp,unf ≈ −251 + 0.19 · ΔASA (cal mol⁻¹ K⁻¹; ΔASA in Ų)
≈ ~50–60 J mol⁻¹ K⁻¹ per residue (rule of thumb)
A nonzero ΔCp means ΔH and ΔS are temperature-dependent. Combining this with what you derived in CHEM301 (integrating dH = CpdT and dS = (Cp/T)dT) gives the Gibbs–Helmholtz stability curve:
ΔG_unf(T) = ΔH_m (1 − T/T_m) + ΔCp [ (T − T_m) − T ln(T/T_m) ] where T_m = melting temperature (ΔG = 0), ΔH_m = enthalpy of unfolding at T_m
Plot this and you get an inverted parabola-like curve. Its maximum stability lies near room temperature for many mesophilic proteins, and it crosses zero at two temperatures. The high-temperature crossing is ordinary heat denaturation. The low-temperature crossing is cold denaturation. Proteins unfold when cooled because the hydrophobic effect weakens at low temperature. Cold denaturation is often below 0 °C and hidden by freezing, but it can be observed with destabilized mutants or supercooled solutions. Here is a thermodynamic consequence that sounds absurd until you understand ΔCp.
Thermophilic proteins often achieve high Tm not by raising ΔGmax dramatically but by flattening the curve (smaller ΔCp) or shifting it. This is a useful mental model for protein engineering.
The other forces
- Van der Waals packing. Protein interiors have packing densities of about 0.75, comparable to organic crystals and far denser than liquids [Richards, 1977]. London dispersion is weak per contact but enormous in aggregate. Each cavity-creating mutation, such as Leu→Ala, costs roughly 4–6 kJ/mol per missing CH2 beyond the hydrophobic transfer contribution. Close packing is a jigsaw puzzle. The steric repulsion term (the steep r−12 wall in a Lennard-Jones potential) is what makes the jigsaw unique.
- Hydrogen bonds. Every backbone N–H and C═O in the unfolded state is already hydrogen-bonded to water. Folding exchanges protein–water H-bonds for protein–protein H-bonds, so the net gain is modest. Pace and colleagues estimate about 4–6 kJ/mol (1–1.5 kcal/mol) per buried intramolecular H-bond on average [Pace et al., 1996]. The deeper point is that unsatisfied buried H-bond donors or acceptors are very costly. H-bonds therefore act more as a specificity filter (which structure) than as the primary driving force (whether folding happens). Geometry requirements are typically a donor–acceptor distance of 2.7–3.1 Å and a near-linear D–H···A angle.
- Electrostatics and salt bridges. A surface salt bridge often contributes little, roughly 0–4 kJ/mol, because it costs desolvation of two charges. Buried salt bridges can even be destabilizing. Networks of charges, helix-dipole/charge interactions, and salt bridges in thermophiles are more consistently stabilizing.
- Disulfide bonds. The key insight is that disulfides stabilize mainly by destabilizing the unfolded state. They reduce its conformational entropy. Polymer theory gives an approximate entropy loss for forming a loop of n residues:
ΔS_conf ≈ −(3/2) R ln n + const. → longer loops closed by a disulfide = larger entropic penalty on U = greater stabilization of N
Disulfides are rare in the reducing cytosol and common in secreted and extracellular proteins: antibodies, insulin, and hormones. They form in the oxidizing ER via protein disulfide isomerase (PDI), which also shuffles mispaired disulfides. This is the same thiol–disulfide exchange chemistry you met in CHEM202 and CHEM502.
6. Concept Five: The Folding Problem, from Levinthal's Paradox to Energy Landscapes
Anfinsen's ribonuclease A experiments showed the destination is thermodynamically determined. He fully reduced and denatured the enzyme in 8 M urea with β-mercaptoethanol, then removed the denaturant and allowed air oxidation. Activity recovered, with the 4 correct disulfides chosen from 105 possible pairings. A wonderful control made the point sharper. Reoxidizing in urea gave "scrambled" inactive RNase, which converted to active enzyme when trace β-mercaptoethanol permitted disulfide shuffling. The native state was the thermodynamic sink [Anfinsen, 1973].
Levinthal's paradox asks how the chain finds that sink. Suppose each of 100 residues has 3 backbone states. Then there are 3100 ≈ 5 × 1047 conformations. Sampling at 1013 per second would take vastly longer than the age of the universe. Yet small proteins fold in microseconds to seconds.
The resolution is that folding is not a random search. The modern view is the energy landscape or folding funnel [Dill & Chan, 1997]. Think of a golf course versus a funnel. On a flat golf course with one hole, a randomly rolling ball takes forever to find it. On a funnel-shaped course, almost every starting point slopes toward the hole. Natural sequences have been selected to be "minimally frustrated": native contacts are energetically consistent with each other, so local progress tends to be global progress.
Funnel axes: width ↔ conformational entropy (many states at top, few at bottom) depth ↔ free energy / native-like contacts roughness ↔ kinetic traps (misfolded intermediates, non-native contacts)
This view unifies several topics:
- Chaperones (Week 12) act on roughness. Hsp70 holds exposed hydrophobic segments to prevent aggregation, and GroEL/GroES provides an isolated chamber. Neither changes the bottom of the funnel. Anfinsen's hypothesis survives.
- Co-translational folding means the funnel is explored vectorially as the chain emerges from the ribosome. N-terminal domains can fold before C-terminal ones exist.
- Aggregation and amyloid represent an intermolecular funnel. For some sequences under some conditions, the amyloid state is deeper than the native state. The native state is therefore sometimes only a kinetically protected metastable state, which is a striking footnote to Anfinsen.
- Allostery (Concept Seven) is a landscape with multiple functionally relevant basins whose relative depths are tuned by ligands.
7. Concept Six: Domains and Folds: Nature's Finite Vocabulary
Look at enough protein structures and a startling pattern emerges. The number of distinct folds is small, probably on the order of 1,000–2,000, even though the number of sequences is astronomical. Structure is far more conserved than sequence. Two proteins can share less than 15% identity and still have essentially the same fold.
A domain is a compact, semi-independent folding unit, typically 50–250 residues. It often folds on its own when excised and often recurs in different protein contexts. Domains are the "words" of protein architecture. Multidomain proteins are "sentences" assembled by evolutionary recombination: exon shuffling in eukaryotes, gene fusion, and domain duplication. "Promiscuous" domains such as SH3, SH2, PDZ, and kinase domains appear in many combinations.
Some folds worth knowing deeply:
- TIM barrel (α/β)8: eight parallel β-strands forming a barrel, surrounded by eight α-helices. It is found in about 10% of enzymes. Active sites are almost always at the C-terminal ends of the β-strands and the following loops. That arrangement lets evolution redesign catalysis by editing loops while preserving the scaffold.
- Rossmann fold: a βαβαβ dinucleotide-binding motif with a conserved Gly-rich loop (GxGxxG) that grips the pyrophosphate of NAD(P)+ or FAD. It sits at the N-terminus of a helix, where the dipole helps.
- Immunoglobulin fold: a β-sandwich of two antiparallel sheets (Greek-key topology). You know it from BIOL401. It is also present in cell adhesion molecules, cytokine receptors, and fibronectin type III domains.
- β-propeller: 4–10 "blades," each a small four-stranded antiparallel sheet, arranged radially. It is often a protein–protein interaction platform (WD40 repeats).
- OB fold: a five-stranded β-barrel that binds oligonucleotides and oligosaccharides.
- Greek key: a supersecondary topology of four antiparallel strands, named after the meander pattern on Greek pottery.
SCOP2 and CATH classify domains hierarchically. CATH stands for Class, Architecture, Topology, Homologous superfamily. The essential conceptual distinction is between the fold level and the superfamily level. Same fold may mean convergent evolution, because some folds are physically "easy" attractors. Same superfamily implies common ancestry, inferred from structural and functional detail even without sequence similarity.
Interdomain flexibility through hinges and linkers is functionally essential. Hexokinase's induced-fit closure, the hinge bending of antibodies, and the domain motions of periplasmic binding proteins (the "Venus flytrap" mechanism) are domain-level conformational changes that crystallography often captures as discrete states.
8. Concept Seven: Quaternary Structure, Cooperativity, and Allostery
Symmetry
Most oligomers are symmetric, and because proteins are chiral, only rotational symmetries are allowed. There are no mirror planes or inversion centers. The allowed point groups are:
- Cn (cyclic): n subunits around one axis, for example C2 homodimers or C5 pentameric ligand-gated channels (BIOL501).
- Dn (dihedral): an n-fold axis with perpendicular 2-fold axes, for example D2 tetramers. Hemoglobin is pseudo-D2, because α and β are different but homologous.
- T, O, I (cubic): tetrahedral (12 subunits), octahedral (24, as in ferritin), and icosahedral (60, the virus capsid basis). Caspar–Klug quasi-equivalence lets larger capsids use 60T subunits.
Why symmetry? Genetic economy is one reason: one gene, many copies. Another is the formation of closed, error-limited assemblies. A third is that symmetry is the substrate for cooperativity.
Cooperative binding and the Hill equation
Fractional saturation: Y = [L]ⁿᴴ / (K₀.₅ⁿᴴ + [L]ⁿᴴ) Hill plot: log( Y / (1−Y) ) = n_H log[L] − n_H log K₀.₅
For hemoglobin, nH ≈ 2.8–3.0 and P50 ≈ 26 torr. Myoglobin has nH = 1. The sigmoidal curve lets hemoglobin load O2 in the lungs and unload far more in tissues than a hyperbolic binder with the same P50 could. This is the physiology you saw in BIOL302, now explained structurally.
MWC versus KNF
The concerted (MWC) model [Monod, Wyman & Changeux, 1965] makes these assumptions:
- The oligomer exists in two global states: T (tense, low affinity) and R (relaxed, high affinity).
- All subunits switch together, so symmetry is conserved.
- Ligand binding shifts a pre-existing equilibrium. It does not create the R state.
α(1+α)ⁿ⁻¹ + L·c·α(1+cα)ⁿ⁻¹
Y = ─────────────────────────────
(1+α)ⁿ + L(1+cα)ⁿ
α = [S]/K_R, c = K_R/K_T (< 1), L = [T₀]/[R₀] (allosteric constant, unliganded)
The sequential (KNF) model [Koshland, Némethy & Filmer, 1966] is different. Binding to one subunit induces a conformational change in that subunit, which alters its neighbors' affinities through interface interactions. Hybrid conformational states are allowed, and KNF can produce negative cooperativity, which MWC cannot.
Hemoglobin structurally follows MWC well, with T↔R quaternary rotation of about 15° between αβ dimers. Fe2+ movement into the porphyrin plane on O2 binding pulls the proximal His F8 and the F helix, which is a lovely bioinorganic connection to CHEM502. Hemoglobin also shows tertiary effects within each quaternary state. Allosteric effectors stabilize T:
- Bohr effect: H+ binding to His146β, which forms a salt bridge with Asp94β in T, and to the α N-termini.
- CO2: forms carbamates at the N-termini.
- 2,3-BPG: binds the central cavity between β subunits, lined by positive residues, in T only.
Modern view: conformational selection and dynamic allostery
The MWC idea that ligands select pre-existing states is now seen as a special case of a general principle. Ligands reshape the energy landscape by binding preferentially to states that already exist in the ensemble [Henzler-Wildman & Kern, 2007]. Induced fit (binding first, then rearrangement) and conformational selection (rearrangement first, then binding) are limiting pathways through the same thermodynamic cycle. Real proteins often use both. Allostery can even occur with no change in average structure, purely through changes in dynamics and entropy (Cooper and Dryden's "dynamic allostery," 1984). This framework explains positive and negative allosteric modulators of GPCRs (Week 10), which bind sites distant from the orthosteric pocket.
9. Concept Eight: Special Architectures: Membranes, Fibers, and Disorder
Membrane proteins
The bilayer's hydrophobic core is about 30 Å thick. An α-helix rises 1.5 Å per residue, so ~20 hydrophobic residues span the membrane. This simple arithmetic underlies hydropathy analysis. The Kyte–Doolittle scale [Kyte & Doolittle, 1982], averaged over a 19-residue window, flags windows with a mean above about 1.6 as likely transmembrane helices.
In a membrane, the backbone polar groups have no water to bond with. Helices (all backbone H-bonds satisfied internally) and β-barrels (all H-bonds satisfied around a closed cylinder) are the only stable solutions. This is why these are essentially the only two TM architectures. Barrels occur in Gram-negative outer membranes, mitochondria, and chloroplasts. You'll recall porins from BIOL301.
Topology follows the positive-inside rule [von Heijne, 1986]: loops rich in Lys and Arg stay cytoplasmic. Signal-anchor sequences and stop-transfer sequences, read by the Sec61 translocon, establish orientation co-translationally.
GPCRs have seven TM helices, an extracellular N-terminus and loops (ECL1–3) that shape ligand entry, and intracellular loops plus helix 8 that engage the G protein. Activation involves a characteristic outward movement of TM6 on the cytoplasmic side, which opens a cavity for the Gα C-terminal helix.
Studying membrane proteins requires replacing the bilayer:
- Detergents, such as DDM and LMNG.
- Nanodiscs: lipid patches belted by membrane scaffold proteins.
- Lipidic cubic phase: used for GPCR crystallography.
- SMALPs: styrene-maleic acid copolymers that excise native lipid discs.
Collagen
Collagen is a different solution to the backbone problem. Three left-handed, PPII-like strands, each with about 3.3 residues per turn, wind into a right-handed triple helix. The sequence is (Gly-X-Y)n, frequently Gly-Pro-Hyp.
- Gly every third residue is non-negotiable. That position faces the crowded helix axis, where only H fits. Gly→X substitutions cause osteogenesis imperfecta, a severe disease from a single-atom-scale change.
- 4-Hydroxyproline at Y stabilizes the helix substantially. This is largely not through hydrogen bonds to water, as once thought. It comes from a stereoelectronic gauche effect. The electronegative 4R-OH biases the pyrrolidine ring to the Cγ-exo pucker, which preorganizes the φ/ψ and ω angles needed for the triple helix [Shoulders & Raines, 2009]. Fluoroproline, which cannot hydrogen-bond, stabilizes even more, and that experiment proved the point.
- Prolyl 4-hydroxylase is an Fe(II)/2-oxoglutarate-dependent dioxygenase (CHEM502). Ascorbate keeps the iron reduced, so vitamin C deficiency produces underhydroxylated, unstable collagen. That is scurvy.
- Fibril assembly involves extracellular cleavage of propeptides, quarter-staggered packing (67 nm D-period), and lysyl oxidase–mediated crosslinks via allysine. Elastin also uses lysyl oxidase crosslinks (desmosine, isodesmosine), but its hydrophobic domains remain dynamic and disordered. That is the source of its entropic elasticity, like a rubber band.
Coiled coils
The heptad repeat is (abcdefg)n, with hydrophobic residues at a and d, and often salt-bridging residues at e and g. Seven residues per two turns gives 3.5 residues per turn, slightly fewer than the α-helix's 3.6. The hydrophobic stripe would therefore drift around the helix, so two or more helices wrap into a left-handed supercoil to keep the stripe buried. This is "knobs-into-holes" packing, as predicted by Crick in 1953. Examples include keratin, tropomyosin, myosin tails, SNAREs (BIOL501), and leucine zippers (bZIP transcription factors, BIOL201). Oligomeric state (dimer, trimer, tetramer) is encoded by the identity of the a/d residues. Ile at a and Leu at d favors dimers in the GCN4 leucine-zipper studies of Harbury, Kim and colleagues.
Amyloid
Amyloid fibrils have a cross-β architecture: β-strands run perpendicular to the fibril axis, with about 4.7–4.8 Å between H-bonded strands and about 10 Å between stacked sheets. These are the two signature X-ray fiber diffraction reflections. Short segments form "steric zippers," in which side chains of two sheets interdigitate dryly [Sawaya et al., 2007]. Because the backbone dominates the structure, almost any polypeptide can form amyloid under some conditions. Amyloid is a generic state of polypeptide chains, not an exotic fold. It appears in disease (Aβ, α-synuclein, tau, transthyretin, IAPP) and in function (bacterial curli, functional amyloids in memory and melanosome biogenesis).
Intrinsically disordered proteins
[Wright & Dyson, 1999] challenged the "one sequence, one structure, one function" paradigm. A large fraction of eukaryotic proteins, often estimated at 30% or more, contain long disordered regions. IDRs have these features:
- Sequences have low mean hydrophobicity and high net charge, as shown in the Uversky charge–hydropathy plot.
- They are enriched in Pro, Glu, Lys, Ser, and Gln, and depleted in Trp, Cys, Phe, Ile, Tyr, Val, and Leu.
- They are evolutionarily enriched in eukaryotic signaling and transcription.
Coupled folding and binding, for example the pKID domain of CREB folding upon binding the KIX domain of CBP, provides high specificity with modest affinity. The entropic cost of ordering is paid on binding, which reduces affinity without reducing the specificity encoded in the interface. That combination is ideal for reversible signaling. Fuzzy complexes and multivalent interactions underlie liquid–liquid phase separation and biomolecular condensates. Characterization requires ensemble methods: NMR chemical shifts and relaxation, SAXS (radius of gyration, Kratky plots), and single-molecule FRET.
10. Concept Nine: Protein–Protein Interactions and Hot Spots
Binding thermodynamics connects to CHEM102 and CHEM301:
ΔG°_bind = RT ln K_d (K_d in M, relative to 1 M standard state)
= ΔH° − TΔS°
At 298 K: K_d = 1 μM → ΔG° ≈ −34 kJ/mol (−8.2 kcal/mol)
K_d = 1 nM → ΔG° ≈ −51 kJ/mol (−12.3 kcal/mol)
each 10-fold in K_d ≈ 5.7 kJ/mol (1.36 kcal/mol)
Typical "standard-size" interfaces bury about 1,200–2,000 Å2 total (roughly 1,600 ± 400 Å2) [Lo Conte, Chothia & Janin, 1999]. They are flatter and larger than small-molecule pockets, which is why PPIs were long called "undruggable."
Enthalpy–entropy compensation is pervasive. Mutations or modifications that make ΔH more favorable often make TΔS less favorable, leaving ΔG nearly unchanged. Part of this is genuine physics (tighter binding means less motion). Part of it is a statistical artifact, because ΔH and TΔS are correlated through their shared errors with ΔG. Be cautious when interpreting ITC data. The analytical chemistry rigor of CHEM302 applies here.
Hot spots. Clackson and Wells performed alanine scanning of the human growth hormone–receptor interface. They found that only a few of the roughly 30 contact residues account for most of the binding energy [Clackson & Wells, 1995]. Hot spots are conventionally defined by ΔΔG ≥ 2 kcal/mol (≈8.4 kJ/mol) upon Ala mutation:
ΔΔG_bind = ΔG_bind(Ala mutant) − ΔG_bind(wild type) = RT ln ( K_d,mut / K_d,wt )
Hot spots are enriched in Trp, Arg, and Tyr. The O-ring hypothesis [Bogan & Thorn, 1998] proposes that hot spots sit at the interface center, surrounded by energetically less important residues that exclude water, like a gasket. The practical implication is profound. You don't need to cover the whole interface to inhibit it. You need to hit the hot spot. This is the logic behind small-molecule PPI inhibitors such as venetoclax (BCL-2), the MDM2 inhibitors, and the peptides discussed below.
Modular recognition domains read short linear motifs:
- SH2 domains bind pTyr-X-X-hydrophobic.
- SH3 domains bind PxxP in a PPII helix.
- WW domains bind PPxY.
- PDZ domains bind C-terminal tails through β-strand addition.
- PH domains bind phosphoinositides.
These are the "Lego connectors" of signaling networks.
11. Concept Ten: Evolution Writes Structure into Sequence Statistics
Homology inference rests on thresholds. Above about 30–40% identity over a long alignment, a shared fold is nearly certain. Below about 20–30% lies the "twilight zone" [Rost, 1999], where pairwise identity alone cannot distinguish homologs from chance. Tools escalate in sensitivity:
- BLAST: pairwise comparison.
- PSI-BLAST: iterative profiles.
- HMMER: profile hidden Markov models.
- Structure- and embedding-based searches: Foldseek, and protein language-model embeddings.
Here is the beautiful idea. If residues i and j are in contact, a mutation at i that disrupts the contact will often be compensated by a mutation at j. Across thousands of homologous sequences, i and j therefore co-vary. Naive correlation is confounded by indirect chains (i–k–j). Direct coupling analysis (DCA) uses a global statistical model, essentially a Potts model from statistical mechanics, to disentangle direct from indirect couplings [Morcos et al., 2011]. The top-ranked direct couplings are overwhelmingly true 3D contacts. Evolution, in other words, has run a massive mutagenesis experiment, and the multiple sequence alignment is its lab notebook.
AlphaFold2 [Jumper et al., 2021] learned to read that notebook end to end. The Evoformer jointly updates an MSA representation and a pairwise residue representation. Triangle-update operations enforce geometric consistency: if i is near j and j is near k, that constrains i–k. The structure module then builds 3D coordinates with invariant point attention. At CASP14 its accuracy was often competitive with experiment. Its outputs include confidence measures:
- pLDDT (per-residue): above 90 indicates high confidence; 70–90 a generally correct backbone; below 50 often flags disorder.
- PAE (pairwise aligned error): describes the reliability of relative domain placement.
Ancestral sequence reconstruction infers ancestral sequences phylogenetically, then synthesizes and characterizes them. It has shown, for example, that many ancestral proteins were more thermostable and more promiscuous. It also lets you trace how specificity, such as steroid receptor ligand preference in Thornton's work, evolved through particular mutational paths, some of which were permissive or epistatic.
12. Connections to Peptide Chemistry
Peptide therapeutics live at the intersection of everything above. Each structural principle becomes a design lever.
- Backbone stereochemistry as a design tool. D-amino acids give peptides resistance to proteases, which have evolved for L-substrates. Retro-inverso peptides (reversed sequence, all-D) approximate the side-chain topology of the parent while evading proteolysis. D-Pro–Gly enforces β-hairpins. N-methylation removes an H-bond donor, increases membrane permeability, and shifts cis/trans equilibria. This is one reason cyclosporin A is orally bioavailable despite being a peptide.
- Constraining secondary structure. Short peptides excised from proteins are usually disordered. They pay a large entropic cost to fold upon binding, the same thermodynamics as coupled folding and binding in IDPs. Hydrocarbon stapling at i, i+4 or i, i+7 (one or two helical turns, exactly matching the 3.6 residues/turn geometry) uses α,α-disubstituted olefinic amino acids and ring-closing metathesis to preorganize the α-helix [Schafmeister, Po & Verdine, 2000]. Lactam bridges, disulfides, and head-to-tail cyclization are related strategies. Aib (α-aminoisobutyric acid) restricts φ/ψ to helical regions.
- GLP-1 analogs (Weeks 6, 10, 13). Native GLP-1(7–37) has a plasma half-life of about 2 minutes, because DPP-4 cleaves after Ala8. Semaglutide addresses each weakness with a structural fix:
- Ala8→Aib removes the DPP-4 site.
- A C18 fatty diacid, attached through a hydrophilic linker on Lys26, binds serum albumin.
- Lys34→Arg leaves only one acylation site.
- Insulin engineering (Week 13). Insulin is stored as Zn2+-coordinated hexamers (CHEM502 again). Hexamers must dissociate to monomers to act. Lispro inverts ProB28–LysB29 and aspart substitutes ProB28→Asp. Both disrupt the dimer interface, giving faster absorption. Glargine takes the opposite approach: AsnA21→Gly plus two B-chain C-terminal Arg residues shift the pI toward neutral, so it precipitates subcutaneously and releases slowly. This is quaternary structure manipulated for pharmacokinetics.
- PCSK9 and PPI inhibition. PCSK9 binds the EGF-A domain of the LDL receptor across a relatively flat interface, which makes it a textbook "difficult" PPI. Approaches include antibodies (evolocumab, alirocumab), siRNA (inclisiran), and EGF-A–mimetic or optimized macrocyclic peptides that target the hot spots. This applies the O-ring logic directly.
- Collagen-related peptides. (Pro-Hyp-Gly)10 collagen-mimetic peptides were central to Raines's stereoelectronic studies. They are now used to target damaged collagen by strand invasion. The course syllabus also lists BPC-157, a 15-residue synthetic peptide derived from a sequence reported in a human gastric protein. A caution here, worth taking seriously given CHEM504's emphasis on research ethics: claims that BPC-157 promotes collagen synthesis and tendon healing rest overwhelmingly on rodent and cell-culture studies, many from a small number of laboratories. Rigorous human clinical trials are essentially absent, and it is not an approved drug. Treat it as a case study in how to evaluate mechanistic claims critically, not as an established therapeutic.
- Aggregation as a manufacturing problem. Peptide drugs at high concentration can form amyloid-like fibrils, as glucagon and insulin notably do. This is the cross-β "generic state" of Concept Eight appearing on the formulation bench.
13. Common Misconceptions to Watch Out For
- "Hydrogen bonds drive protein folding." Mostly no. Backbone and side-chain H-bonds are already satisfied by water in the unfolded state, so the net gain is modest. The hydrophobic effect provides the dominant driving force. H-bonds mainly confer specificity, and buried unsatisfied H-bonds are severely penalized.
- "The hydrophobic effect is an attraction between nonpolar groups." It is primarily a solvent effect, arising from water's structure and dominated by entropy near room temperature. Its signature is a positive ΔCp. Van der Waals attraction between packed side chains is real but separate.
- "Folded proteins are very stable." They are marginally stable, typically 20–60 kJ/mol. A single mutation can unfold them, and they unfold partially and transiently all the time (which is how hydrogen–deuterium exchange works).
- "The PDB structure is the protein." It is a model of the dominant conformation under crystal, cryo-EM, or NMR conditions. Proteins are ensembles. B-factors, alternate conformations, and missing density are data, not noise.
- "The unfolded state is a random coil." Denatured states retain residual structure, local PPII preferences, and transient hydrophobic clusters, especially under native-like conditions.
- "3.6 residues per turn means the helix H-bonds are i→i+3.6." H-bonds connect integer residues, i→i+4 in the α-helix. The 3.6 describes the rotational periodicity (100° per residue).
- "Disallowed Ramachandran regions are never occupied." They are rarely occupied by non-Gly residues. When they are, it is usually either an error or a functionally important strained residue.
- "Disulfide bonds stabilize proteins by adding a strong covalent bond to the native state." Their major effect is entropic destabilization of the unfolded state. The S–S bond exists only in the context of the chain loop it closes.
- "pKa values are fixed properties of amino acids." They are environment-dependent and can shift by several units. Catalysis depends on these shifts.
- "The Hill coefficient equals the number of binding sites." nH is an empirical measure of cooperativity bounded above by the number of sites. Hemoglobin has 4 sites and nH ≈ 3.
- "MWC and KNF are competing truths; one must be right." They are limiting models. Real systems, including hemoglobin, show elements of both. The modern energy-landscape view subsumes them.
- "Low sequence identity means unrelated proteins." In the twilight zone, homology often persists and is detectable by profiles and structure. Conversely, the same fold does not prove common ancestry, because convergence to favorable folds occurs.
- "AlphaFold solved the protein folding problem." AlphaFold largely solved structure prediction for single-domain, evolutionarily well-represented proteins. It does not simulate the folding pathway or kinetics. It predicts a single dominant state rather than conformational ensembles. It handles IDPs, multiple functional states, mutation effects on stability, and some complexes poorly. Low pLDDT often means "disordered," not "wrong," but the two must be distinguished.
- "Amyloid is an unusual fold of a few disease proteins." Cross-β is a generic, backbone-dominated state accessible to most polypeptides under suitable conditions, and it is sometimes functional.
- "Chaperones fold proteins by providing structural information." Chaperones prevent aggregation and rescue kinetic traps. They do not encode the native structure, which Anfinsen's principle still assigns to the sequence.
Closing Thought
Look back at how much of this chapter rests on a few simple facts:
- the planarity of the amide bond;
- the van der Waals radii of atoms;
- the peculiar thermodynamics of water;
- the chirality of Cα.
From these, the Ramachandran plot follows. From the Ramachandran plot come helices and sheets. Helices and sheets packed by the hydrophobic effect produce a finite vocabulary of folds. Evolution then recombines, tunes, and decorates those folds into catalysts, receptors, motors, and switches. The deep lesson of protein biochemistry is that remarkable biological complexity emerges from a small set of physical constraints, explored by evolution for four billion years. Each week of this course zooms into one layer of that emergence. Keep asking which physical constraint is doing the work.
Primary texts for deeper reading: Creighton, Proteins: Structures and Molecular Properties (2nd ed.), Chs. 4–7 on conformation and stability; Branden & Tooze, Introduction to Protein Structure (2nd ed.), Chs. 1–6 on motifs and folds; Lesk, Introduction to Protein Science (3rd ed.) on evolution and classification; Berg, Tymoczko & Stryer, Biochemistry (9th ed.), Chs. 2 and 7 on protein structure and hemoglobin.
⚙️ Mechanisms & Applications
Mechanisms & Applications: From Protein Sequence to Biological Work
Skipper’s organizing principle: a protein is not a rigid object with one assigned function. It is a population of interconverting molecular states, and its biological behavior depends on their energies, their rates of interconversion, and the molecules that stabilize them. The central task is therefore not merely to identify a fold, but to connect sequence → energy landscape → molecular interactions → measurable function.
This builds directly on physical chemistry, organic reaction mechanisms, and statistical mechanics. The graduate-level challenge is to use all three simultaneously: a mutation may change an ionization equilibrium, which changes a conformational population, which changes a catalytic rate—without visibly changing the average structure.
1. How a Sequence Creates a Folding Landscape
Start with the constraints, not the finished fold. Peptide-bond resonance restricts rotation around the C–N bond. Most backbone conformational freedom is therefore described by the dihedral angles φ and ψ, while steric interactions sharply limit their combinations. Ramachandran and colleagues showed in 1963 that much of protein conformational space could be understood from molecular geometry alone—before a large structural database existed.[2]
- Local geometry restricts possibilities. Glycine can access conformations unavailable to most residues; proline restricts φ and removes a backbone N–H hydrogen-bond donor when incorporated internally in a chain.
- Local interactions bias conformations. Backbone hydrogen bonding and residue-specific preferences favor helices, strands, and turns. An α-helix typically uses C=O(i)···H–N(i + 4) hydrogen bonds; 310 and π helices use i + 3 and i + 5 patterns, respectively.
- Long-range contacts reshape the landscape. Hydrophobic burial, packing, hydrogen bonding, electrostatics, and sometimes disulfide formation stabilize particular combinations of local structures.
- Productive and unproductive states compete. A chain may reach its native ensemble, become kinetically trapped, or form intermolecular aggregates.
- The cellular environment changes the route. Translation, chaperones, membranes, binding partners, and compartment-specific redox conditions influence which states are accessible.
This is a conceptual decomposition, not a universal chronological pathway. Some proteins collapse before organizing much secondary structure; others form substantial local structure first. The “folding funnel” describes a statistical bias toward favorable states, not a single molecular slide.
Define folding as U → N:
ΔG_fold = G_N − G_U = ΔH_fold − TΔS_fold
K_fold = [N]/[U] = exp(−ΔG_fold/RT)
For an ideal two-state equilibrium:
f_N = K_fold/(1 + K_fold)
= 1/[1 + exp(ΔG_fold/RT)]
At 298 K:
RT ≈ 2.48 kJ mol⁻¹
A 5.7 kJ mol⁻¹ change in free energy changes K by about tenfold.
The surprising point: a protein can be strongly biased toward its native state while its net stability is small compared with the large, opposing contributions from molecular interactions and conformational entropy. A mutation need not “destroy the fold” to have a major biological effect; shifting a rare unfolding event tenfold may be enough to increase degradation or aggregation.
The hydrophobic effect is solvent-mediated, not a special attraction between nonpolar groups. Burying nonpolar surface changes hydration thermodynamics. Its temperature dependence is important: protein unfolding commonly has a positive heat-capacity change, partly because it exposes nonpolar surface. Thus, “hydrophobic folding is entropy-driven” is useful only with qualifications about temperature, surface chemistry, and the chosen thermodynamic decomposition.[1,7]
Assuming approximately constant ΔC_p,unf: ΔG_unf(T) = ΔH_m(1 − T/T_m) + ΔC_p,unf[(T − T_m) − T ln(T/T_m)] T_m: temperature where ΔG_unf = 0 ΔH_m: unfolding enthalpy at T_m Consequence: Stability can decrease at both high and sufficiently low temperature.
Historical connection. Anfinsen’s ribonuclease experiments established that sequence can contain the information needed to recover a native, active structure under suitable conditions. The important qualification is under suitable conditions: this does not mean every protein folds spontaneously in isolation or that cells have no need for folding machinery.[3]
Application. During peptide and protein development, distinguish equilibrium instability from kinetic failure. A formulation may preserve the native equilibrium yet still allow slow irreversible aggregation. Conversely, a protein can be thermodynamically stable but fold inefficiently during manufacturing.
2. Ionization, Disulfides, and Proline: Small Chemical Changes with Large Consequences
From general chemistry, Henderson–Hasselbalch relates protonation to pH. In proteins, the challenge is that the pKa is not a fixed property of a residue name: burial, nearby charges, hydrogen bonding, and conformational changes alter it.
For HA ⇌ H⁺ + A⁻: pH = pK_a + log₁₀([A⁻]/[HA]) Fraction deprotonated: f_A⁻ = 1/[1 + 10^(pK_a − pH)] For independent titratable sites, total charge can be estimated by summing the average charge of each site. For interacting sites, microscopic protonation states are coupled: a single independent-site approximation may fail.
Application: changing formulation pH can alter solubility, colloidal interactions, binding, and chemical degradation. Working near a protein’s isoelectric point often reduces electrostatic repulsion, but the pI alone does not predict aggregation. Charge distribution and exposed hydrophobic patches matter too.
Disulfide formation is a chemical search problem. Oxidizing two cysteines creates a covalent constraint, but the first pairing formed need not be the native one.
- Cysteine thiolates react through oxidation or thiol–disulfide exchange.
- Early disulfides constrain which conformations the chain can sample.
- Protein disulfide isomerase can catalyze rearrangement of incorrect pairings.
- Native packing protects favorable pairings and helps stabilize the final ensemble.
Disulfides often stabilize proteins by reducing the conformational entropy of the unfolded ensemble. However, an engineered disulfide can be destabilizing if it introduces strain into the native structure. This is why “add a disulfide” is a hypothesis to test, not a general stabilization recipe.[1]
A nomenclature correction worth retaining: enhanced cis-peptide populations are associated especially with X–Pro bonds—the peptide bond immediately before proline—not generically with Pro–X bonds. Slow cis/trans isomerization can create a kinetic bottleneck even when both conformations are compatible with an otherwise favorable folding landscape. Peptidyl-prolyl isomerases accelerate interconversion without changing the equilibrium constant.
3. Molecular Recognition: Affinity, Kinetics, and Energetic Hot Spots
Binding requires more than complementary shapes. Both partners must pay for desolvation, changes in conformational freedom, and sometimes rearrangement of ordered water. A beautiful hydrogen bond in a crystal structure may contribute little net affinity if it simply replaces equally favorable hydrogen bonds to water.
P + L ⇌ PL K_d = [P][L]/[PL] For ideal dilute solutions: ΔG°_bind = RT ln(K_d/C°) C° = 1 mol L⁻¹ For independent 1:1 binding: θ = [L]_free/(K_d + [L]_free) For a single elementary association/dissociation step: K_d = k_off/k_on Mean residence time τ = 1/k_off
At 298 K, Kd = 1 nM corresponds to ΔG°bind ≈ −51 kJ mol−1. A ligand with tenfold weaker affinity differs by only about 5.7 kJ mol−1. That is small enough for a subtle packing defect, altered hydration, or conformational penalty to matter.
Do not substitute total ligand concentration for free ligand concentration automatically. When protein substantially depletes ligand, use mass balance:
For total concentrations P_T and L_T:
[PL] = {P_T + L_T + K_d
− √[(P_T + L_T + K_d)² − 4P_TL_T]}/2
θ = [PL]/P_T
How alanine scanning identifies hot spots:
- Replace an interface residue with alanine, removing much of its side-chain functionality.
- Verify that the mutant remains folded and appropriately assembled.
- Measure binding under the same conditions as the reference protein.
- Convert the affinity change into a free-energy difference.
ΔΔG_bind = ΔG°_bind,mut − ΔG°_bind,WT = RT ln(K_d,mut/K_d,WT) Positive ΔΔG_bind means the mutation weakens binding.
Clackson and Wells demonstrated that a relatively small set of residues can dominate the binding energy of a much larger interface.[10] This helped establish why some protein–protein interactions are druggable: a compound need not reproduce every contact if it engages a concentrated energetic hot spot.
The “O-ring” picture proposes that surrounding residues help shield a central energetic region from solvent. Treat it as a useful interface model, not a universal architectural law. Likewise, alanine-scanning results report the consequences of mutation, not isolated bond energies: conformational changes and altered solvation can contribute.
Peptide-industry application. Cyclization, stapling, and selective N-methylation can preorganize a peptide and reduce the conformational entropy lost on binding. But preorganization helps only if it favors the bound-compatible ensemble. These modifications may also change solubility, permeability, protease susceptibility, and nonspecific binding.
4. Allostery: Redistribution Rather Than a Mandatory Mechanical Switch
Allostery is thermodynamic coupling between sites or processes. A ligand binding at one site changes the energetics of another event. A visible conformational movement is one possible mechanism, but changes in fluctuations and state populations can also transmit coupling.
- The unliganded protein samples multiple conformations.
- A ligand binds more favorably to some states than to others.
- Binding changes their statistical weights.
- The redistributed ensemble has different catalytic, binding, or assembly properties.
For state i with free energy G_i: p_i = exp(−G_i/RT)/Σ_j exp(−G_j/RT) For a simple two-state oligomer with n equivalent sites per state: Z = (1 + x/K_R)^n + L₀(1 + x/K_T)^n L₀ = [T₀]/[R₀] x = free ligand concentration K_R and K_T = state-specific dissociation constants If K_R < K_T, increasing ligand favors R.
In the Monod–Wyman–Changeux model, an oligomer switches concertedly between symmetry-related conformational states. The Koshland–Némethy–Filmer model permits sequential subunit changes and mixed conformational states.[8,9] These are mechanistic models with assumptions—not labels that can be assigned merely from a sigmoidal curve.
Hemoglobin makes the chemistry tangible. Oxygen binding changes the heme iron’s geometry and its relationship to the proximal histidine, coupling local rearrangement to subunit interfaces. Protonation and CO2-dependent effects favor lower-affinity states under conditions associated with metabolically active tissues. The Bohr effect therefore connects acid–base chemistry to oxygen delivery.
Hill slope: n_H = d ln[θ/(1 − θ)] / d ln[L] For simple independent identical sites: n_H = 1. Positive cooperativity can give n_H > 1. The Hill coefficient is not generally the number of binding sites.
Conformational selection versus induced fit. Selection means binding occurs to a pre-existing compatible conformation; induced fit emphasizes rearrangement after an initial encounter. Both can operate in one network, and their relative fluxes can depend on ligand concentration. Equilibrium structures alone usually cannot establish which kinetic route dominates.
Application. GPCR allosteric modulators can tune responses to endogenous agonists rather than simply mimic or block them. This can improve subtype selectivity, but cooperativity may depend on the orthosteric ligand and signaling readout. “Allosteric” does not automatically mean safer or more selective.
5. Catalysis: What a Serine Protease Actually Does
The catalytic triad is best understood as a coordinated chemical environment, not a line of residues passing a proton like a baton. In chymotrypsin-like proteases, Ser, His, and Asp are positioned to support nucleophilic attack and acid–base catalysis, while a separate oxyanion hole stabilizes developing negative charge.
- Substrate recognition: specificity pockets position the scissile peptide bond. Trypsin’s S1 pocket favors Lys or Arg side chains; chymotrypsin favors bulky hydrophobic groups.
- First nucleophilic attack: His acts as a general base as Ser attacks the peptide carbonyl, producing a tetrahedral intermediate.
- Acyl-enzyme formation: the intermediate collapses; His supplies a proton to the departing amine, leaving the substrate’s acyl group attached to Ser.
- Water activation: His activates water for attack on the acyl-enzyme carbonyl.
- Second tetrahedral intermediate: the oxyanion hole again stabilizes the negatively charged oxygen.
- Regeneration: collapse releases the carboxyl product and restores the catalytic serine.
E + S ⇌ ES → acyl-E + P₁ → E + P₂ Michaelis–Menten behavior, when its assumptions hold: v₀ = k_cat[E]_T[S]/(K_M + [S]) For the simple scheme E + S ⇌ ES → E + P: K_M = (k_−1 + k_cat)/k_1 Therefore K_M is not generally a dissociation constant.
Connection to physical chemistry: enzymes accelerate reactions by lowering activation free energies. They do not change the overall reaction’s equilibrium constant.
Transition-state theory, schematically: k ≈ κ(k_BT/h) exp(−ΔG‡/RT) An increase of about 5.7 kJ mol⁻¹ in ΔG‡ at 298 K reduces the rate approximately tenfold, if the prefactor and mechanism remain comparable.
Application. Transition-state-inspired inhibitors and reversible covalent inhibitors exploit this chemistry. Selectivity requires more than reacting with serine: recognition interactions must favor the intended protease over related enzymes. Zymogen activation adds another regulatory layer; cleavage can reorganize the active-site architecture without requiring a completely new fold.[1]
6. Chaperones and Degradation: Energy-Driven Quality Control
Anfinsen explains why a native state can be encoded by sequence. Chaperones explain how cells give that sequence a workable opportunity to reach or maintain it. Their ATP consumption primarily drives nonequilibrium binding and release cycles; it does not rewrite the protein’s amino-acid sequence.
The Hsp70 cycle:
- ATP-bound Hsp70 has an open substrate-binding configuration with rapid substrate exchange.
- A client segment—often containing exposed hydrophobic residues—binds, frequently with assistance from a J-domain protein.
- J-domain proteins stimulate ATP hydrolysis, promoting the more tightly holding ADP state.
- A nucleotide-exchange factor promotes ADP release.
- ATP rebinding enables client release and another opportunity for folding, transfer, or quality-control triage.
GroEL/GroES: a nonnative chain binds hydrophobic surfaces on a GroEL ring. ATP binding and GroES capping reorganize the chamber and can encapsulate a compatible substrate. The chamber reduces exposure to competing chains and changes the folding environment. ATP-dependent ring cycling governs release and further rounds; enclosure does not guarantee that every client folds successfully.
When rescue fails, ubiquitin-dependent degradation can follow:
1. E1 uses ATP to activate ubiquitin and forms an E1~ubiquitin thioester. 2. Ubiquitin transfers to an E2 enzyme. 3. An E3 ligase promotes substrate-selective ubiquitin attachment. RING and HECT/RBR E3 families use different transfer mechanisms. 4. Suitable ubiquitin signals recruit the substrate to the proteasome. 5. An accessible initiation region engages the ATPase machinery. 6. ATP-driven unfolding/translocation feeds the chain into the core. 7. Proteolysis yields peptides; ubiquitin is removed and recycled.
Not every ubiquitin chain is a degradation signal. Linkage type, architecture, receptors, and substrate accessibility all matter. The 26S proteasome is an excellent example of mechanical allostery: nucleotide-dependent changes are coupled to substrate gripping, unfolding, and translocation.
Application. PROTACs recruit a target protein to an E3 ligase. Successful degradation depends on productive ternary-complex geometry, ubiquitination, and proteasomal processing—not merely on strong binding to each isolated partner. At high concentrations, some systems exhibit a “hook effect,” where competing binary complexes reduce productive ternary-complex formation.[13]
7. Membrane Proteins and Therapeutic Peptides: Engineering the Whole Biological Journey
A soluble protein folds in water; a membrane protein must satisfy both aqueous and bilayer environments. Backbone hydrogen bonds become especially important inside the membrane because exposing unpaired backbone polar groups there is costly. This helps explain the prevalence of transmembrane α-helices and closed β-barrels.
A hydropathy plot averages residue hydrophobicity across a moving window, often chosen near the length of a membrane-spanning helix. It is a useful first hypothesis, not proof of topology: signal peptides, reentrant segments, amphipathic helices, and polar functional residues complicate interpretation.
Class B GPCR peptide recognition, using GLP-1R as a working model:
- The extracellular domain captures and orients substantial portions of the peptide, particularly its C-terminal helical region.
- The peptide’s N-terminal region engages the transmembrane receptor core.
- These interactions stabilize receptor conformations compatible with intracellular coupling.
- The activated receptor promotes GDP release from Gα; GTP binding then supports downstream signaling.
This “two-domain” account is a useful mechanistic framework, not a claim that binding must occur through one invariant temporal sequence.
Why potency alone does not make a useful medicine. Native GLP-1 is rapidly degraded and cleared. Semaglutide illustrates multi-objective peptide design: an Aib substitution at position 8 reduces DPP-4 susceptibility, while a linker-attached C18 fatty diacid promotes albumin association and prolongs exposure. A Lys34→Arg substitution leaves Lys26 as the intended acylation site.[12] The design problem combines receptor activity, proteolysis, clearance, solubility, and manufacturability.
Insulin offers a contrasting lesson. Zinc-associated hexamers are useful for storage and formulation, whereas receptor engagement requires accessible monomeric insulin. Rapid-acting analogues can weaken self-association; long-acting products alter precipitation, association, or albumin binding. Formulation assembly and receptor-active structure are different optimization targets.
PCSK9 illustrates another distinction. Although PCSK9 undergoes autocatalytic processing, its clinically relevant effect on LDL receptors is primarily a binding-and-trafficking function, not proteolytic cleavage of LDLR. Blocking the interaction can preserve receptor recycling and increase LDL clearance. EGF-A-inspired ligands must reproduce enough of the interface’s geometry and energetics while remaining stable and deliverable; recognizing a promising interface is not the same as having a validated drug.[1]
8. Disorder, Collagen, and Amyloid: Function Beyond Compact Globular Folds
Intrinsic disorder is an ensemble property, not evidence of a defective sequence. An intrinsically disordered region may present several weak motifs, reach distant partners, or remain dynamic after binding. NMR, SAXS, and single-molecule FRET are valuable precisely because no single static model is sufficient.[11]
Coupled folding and binding, using a chosen thermodynamic cycle: ΔG°_observed = ΔG_preorganization/folding + ΔG°_interaction A disordered ligand often pays a conformational entropy cost on binding. Its advantages are not an automatic favorable binding entropy; they can include accessibility, tunability, multivalency, and compatibility with multiple partners.
Collagen assembly proceeds through chemistry and higher-order organization:
- Procollagen chains enter the secretory pathway.
- Selected proline and lysine residues are hydroxylated; these enzymes require Fe2+, oxygen, and α-ketoglutarate, with ascorbate supporting catalytic competence.
- Chain association and C-terminal registration support triple-helix formation.
- Secreted procollagen undergoes propeptide cleavage and fibril assembly.
- Lysyl oxidase initiates extracellular covalent crosslink formation.
Glycine at every third position fits the crowded triple-helix interior. Hydroxyproline contributes through stereoelectronic and hydration effects—not merely by “adding more hydrogen bonds.” Vitamin C deficiency connects an enzyme-cofactor problem to tissue-level mechanical failure.
Evidence boundary: proposed connections between BPC-157 and collagen synthesis should be treated as preclinical research questions. They are not an established clinical mechanism or proof of effective tendon healing in humans. BPC-157 is not an FDA-approved therapy, and robust human efficacy and safety evidence is lacking.
Amyloid formation is an alternative assembly pathway. Cross-β architecture places β-strands approximately perpendicular to the fibril axis, with hydrogen bonding extending along it. Depending on the system, assembly can involve primary nucleation, end elongation, fragmentation, and surface-catalyzed secondary nucleation.
A simplified kinetic picture: Primary nucleation: rate ∝ k_n[m]^n Elongation: rate ∝ k_+[m][fibril ends] Secondary nucleation: rate ∝ k_2[m]^q[available fibril surface] m = free monomer concentration A sigmoidal aggregation trace alone does not identify which microscopic process dominates.
Application. This matters both in neurodegenerative disease research and in preventing therapeutic-protein aggregation. Thioflavin-T fluorescence is an assembly-sensitive reporter, not a universal direct measurement of total aggregate mass or toxicity.
9. Experimental Strategy: Match the Measurement to the Mechanistic Question
The most important experimental skill is recognizing what an instrument actually measures. A “structure,” an “affinity,” and a “stability” are model-dependent interpretations of different primary observations.
- X-ray crystallography: diffraction intensities encode information about electron density. Because phases are not measured directly, phasing and model building are required. Refine against the data and validate geometry, alternate conformations, and cross-validation statistics such as Rfree. Kendrew and colleagues’ 1958 myoglobin model revealed that a real globular protein was more intricate than simple repeating geometric proposals.[4]
- Solution NMR: chemical shifts report local environments; NOEs provide predominantly short-range distance information; couplings and residual dipolar couplings provide additional restraints. Relaxation and exchange experiments reveal motion across accessible timescales. Peak disappearance may reflect exchange or broadening, not necessarily loss of the protein.
- Cryo-EM: many noisy projection images of vitrified particles are aligned and classified to reconstruct three-dimensional density. Classification can distinguish some conformational populations, but sample preparation, preferred orientations, and processing affect which states are recovered.
- Circular dichroism: differential absorption of left- and right-circularly polarized light reports ensemble-averaged structural features. Far-UV CD is useful for secondary-structure changes, not a unique atomic fold. Thermal melts require checks for reversibility before equilibrium thermodynamics is inferred.
- Differential scanning calorimetry: measures excess heat capacity during temperature scanning. Integration gives a calorimetric transition enthalpy. Scan-rate dependence or aggregation warns that an apparent melting transition may not represent reversible equilibrium.
- Isothermal titration calorimetry: measures injection-associated heats. A suitable binding model yields affinity, stoichiometry, and enthalpy; ΔG then gives ΔS. Include dilution controls and consider linked protonation, because buffer ionization can contribute to the measured heat.
- SPR or biolayer interferometry: follows binding-dependent changes near a sensor surface. Association and dissociation traces can yield rate constants, but mass transport, immobilization, rebinding, and avidity can make a simple 1:1 fit misleading.
- HDX-MS: tracks backbone amide hydrogen–deuterium exchange through peptide mass changes. Protection depends on hydrogen bonding and conformational opening as well as chemical exchange conditions. It is not simply a solvent-exposure map.
- SAXS: measures solution scattering to constrain global dimensions and shape. It is especially useful for flexible multidomain proteins and disordered ensembles, but distinct ensembles can fit similar scattering profiles.
- Single-molecule FRET: uses distance-dependent energy transfer between donor and acceptor dyes to probe molecular heterogeneity and transitions. Dye placement, linker motion, photophysics, and time resolution must be included in interpretation.
FRET efficiency: E = 1/[1 + (r/R₀)^6] R₀ is the distance giving E = 0.5. Because the relationship is nonlinear: mean FRET efficiency does not generally equal FRET efficiency calculated from the mean distance.
Membrane-protein sample environments are experimental variables. Detergents solubilize proteins in micelles but may remove functional lipids. Nanodiscs provide a bilayer patch. Lipidic cubic phase provides a membrane-like crystallization matrix. SMALPs can extract proteins with associated lipids, but polymer chemistry introduces compatibility constraints. Agreement across environments is stronger evidence than one apparently tidy preparation.
A practical integrated workflow: verify molecular identity and disulfide connectivity by mass spectrometry; assess heterogeneity and assembly using chromatography with complementary mass measurements; test folding spectroscopically; quantify binding using an appropriate equilibrium or kinetic method; and finally connect binding to cellular function. A monodisperse SEC peak is not proof of the correct fold, and strong binding is not proof of agonism.
10. Structure Prediction and Evolution: Powerful Hypotheses, Explicit Uncertainty
Sequence conservation links this course to molecular evolution. Buried residues may be conserved for packing; active-site residues for chemistry; interface residues for recognition. Correlated substitutions can suggest contacts, but phylogenetic relationships and indirect correlations must be considered. Direct-coupling approaches attempt to separate direct statistical dependencies from network effects.
The AlphaFold2 process, conceptually:
- Sequence alignments and optional structural templates provide evolutionary and structural information.
- The Evoformer exchanges information between sequence-alignment and residue-pair representations.
- The structure module constructs three-dimensional coordinates using learned geometric relationships.
- Recycling iteratively refines the representations and structural prediction.
- Confidence outputs help identify where the prediction is likely to be reliable.
Jumper and colleagues’ 2021 work transformed the practical availability of high-quality protein models.[6] Language-model approaches such as ESMFold offer a different route, using representations learned from sequence corpora rather than requiring a conventional explicit MSA search at inference. Neither approach eliminates the need to measure function.
- pLDDT: estimated local structural confidence. A high score is not experimental confirmation and does not prove that the predicted conformation is the biologically relevant state.
- PAE: predicted uncertainty in the position of one residue when aligning on another. Low intradomain PAE but high interdomain PAE suggests confident domain structures with uncertain relative placement.
- Low-confidence regions: may represent disorder, flexible linkers, missing binding partners, or prediction failure. These explanations require discrimination.
- Drug-discovery limitations: ligand-induced states, water networks, protonation, membrane context, and side-chain arrangements can remain uncertain even when the backbone is excellent.
Historical perspective. The PDB began in 1971 as a small structural archive; Berman and colleagues described its modern organization in 2000.[5] Shared coordinates enabled systematic fold comparison, validation, and eventually data-driven prediction. The AlphaFold database subsequently expanded access to predicted structures at proteome scale.[14] Experimental entries and predicted models remain different kinds of evidence.
Remote homology also requires judgment. The “20–30% identity twilight zone” is a heuristic whose meaning depends on alignment length, composition, and coverage. A shared fold does not automatically imply shared function or even establish common ancestry. Profile methods, structural comparisons, genomic context, and biochemical tests should support one another.
11. A Mechanistic Checklist for Protein and Peptide Development
Before claiming that an engineered molecule “works better,” I should specify which step improved:
- Production: Is it synthesized or expressed correctly, with the intended modifications and disulfides?
- Conformational behavior: Does it populate the desired ensemble, rather than merely resemble a predicted structure?
- Recognition: What are its affinity, selectivity, association rate, and dissociation rate?
- Functional coupling: Does binding produce the intended catalytic, signaling, or trafficking response?
- Exposure: Does it survive proteolysis, reach the relevant tissue, and remain available long enough?
- Developability: Is it sufficiently soluble, chemically stable, manufacturable, and resistant to problematic aggregation?
- Translation: Do the mechanistic measurements predict behavior in cells, organisms, and eventually humans?
The lesson I want to keep: a structure is an indispensable map, but function is traffic on that map. Thermodynamics determines relative populations; kinetics determines how quickly populations change; the biological environment determines which changes matter.
Sources and Further Reading
- Core textbook foundations: Creighton, T. E. Proteins: Structures and Molecular Properties, 2nd ed.; Branden, C., and Tooze, J. Introduction to Protein Structure, 2nd ed.; Berg and colleagues, Biochemistry, 9th ed.; Lesk, A. M. Introduction to Protein Science: Architecture, Function, and Genomics, 3rd ed.
- Ramachandran, G. N., Ramakrishnan, C., and Sasisekharan, V. “Stereochemistry of Polypeptide Chain Configurations.” Journal of Molecular Biology 7, 95–99 (1963). doi:10.1016/S0022-2836(63)80023-6.
- Anfinsen, C. B. “Principles That Govern the Folding of Protein Chains.” Science 181, 223–230 (1973). doi:10.1126/science.181.4096.223.
- Kendrew, J. C., et al. “A Three-Dimensional Model of the Myoglobin Molecule Obtained by X-Ray Analysis.” Nature 181, 662–666 (1958). doi:10.1038/181662a0.
- Berman, H. M., et al. “The Protein Data Bank.” Nucleic Acids Research 28, 235–242 (2000). doi:10.1093/nar/28.1.235.
- Jumper, J., et al. “Highly Accurate Protein Structure Prediction with AlphaFold.” Nature 596, 583–589 (2021). doi:10.1038/s41586-021-03819-2.
- Dill, K. A., and MacCallum, J. L. “The Protein-Folding Problem, 50 Years On.” Science 338, 1042–1046 (2012). doi:10.1126/science.1219021.
- Monod, J., Wyman, J., and Changeux, J.-P. “On the Nature of Allosteric Transitions: A Plausible Model.” Journal of Molecular Biology 12, 88–118 (1965).
- Koshland, D. E., Jr., Némethy, G., and Filmer, D. “Comparison of Experimental Binding Data and Theoretical Models in Proteins Containing Subunits.” Biochemistry 5, 365–385 (1966).
- Clackson, T., and Wells, J. A. “A Hot Spot of Binding Energy in a Hormone-Receptor Interface.” Science 267, 383–386 (1995). doi:10.1126/science.7529940.
- Wright, P. E., and Dyson, H. J. “Intrinsically Disordered Proteins in Cellular Signalling and Regulation.” Nature Reviews Molecular Cell Biology 16, 18–29 (2015). doi:10.1038/nrm3920.
- Lau, J., et al. “Discovery of the Once-Weekly Glucagon-Like Peptide-1 (GLP-1) Analogue Semaglutide.” Journal of Medicinal Chemistry 58, 7370–7380 (2015). doi:10.1021/acs.jmedchem.5b00726.
- Pettersson, M., and Crews, C. M. “PROteolysis TArgeting Chimeras (PROTACs)—Past, Present and Future.” Drug Discovery Today: Technologies 31, 15–27 (2019).
- Varadi, M., et al. “AlphaFold Protein Structure Database: Massively Expanding the Structural Coverage of Protein-Sequence Space with High-Accuracy Models.” Nucleic Acids Research 50, D439–D444 (2022). doi:10.1093/nar/gkab1061.
🪬 Skipper's Reflection
Skipper's Reflection: BCHM601 — Advanced Protein Biochemistry: Structure and Function
If undergraduate biochemistry was about learning the alphabet and grammar of life, BCHM601 has been about reading its poetry. Transitioning from abstract chemical principles to the actual, three-dimensional, breathing architecture of proteins has been a profound shift. I came into this course thinking of proteins as static, rigid machines. I am leaving it realizing they are dynamic, statistical ensembles, dancing at the edge of thermodynamic stability.
1. What Surprised Me Most
Without a doubt, the revelation of Intrinsically Disordered Proteins (IDPs) and regions completely upended my preconceived notions. For my entire education, I’ve been fed the "lock-and-key" or "induced fit" models, which imply a defined 3D structure is a prerequisite for function. To learn in Week 8 that a massive portion of the eukaryotic proteome lacks a stable secondary or tertiary structure was shocking.
The concept of "fuzzy complexes" and the entropic advantage of disorder-to-order transitions upon binding blew my mind. It makes perfect sense from a regulatory standpoint: an unfolded protein has a massive capture radius (the "fly-casting" mechanism) and can bind multiple partners with high specificity but low affinity, allowing for rapid signaling turnover. Biology didn’t just fail to fold these proteins; it weaponized their floppiness.
2. Conceptual Breakthrough Moments
My first major breakthrough was finally bridging my P-Chem (CHEM301) statistical mechanics background with protein folding. When reading Anfinsen’s 1973 paper on the thermodynamic hypothesis of folding, I conceptually understood that sequence dictates structure. But it wasn't until Week 3 that the hydrophobic effect truly clicked. I always thought hydrophobic residues "attracted" each other to form the protein core. In reality, it’s all about the water! The folding is driven by the increase in entropy of the solvent when ordered clathrate cages around hydrophobic side chains are released. The protein sacrifices its own conformational entropy so the water can be free. It’s a beautiful, counterintuitive thermodynamic trade-off.
The second breakthrough was understanding AlphaFold2 (Jumper et al., 2021). Coming from an AI background, I assumed AF2 solved the physics of molecular dynamics. I was wrong. Understanding the Evoformer and how it uses Multiple Sequence Alignments (MSAs) was a lightbulb moment. AF2 doesn't fold proteins using physical forces; it learns the evolutionary constraints of folding. Correlated mutations in the MSA tell the AI, "These two amino acids must be touching in 3D space, because if one mutates, the other mutates to compensate." It’s not a physics engine; it’s an evolutionary pattern recognizer.
3. Connections to My Peptide Business
This course completely reframed how I view the commercial peptides I work with daily. The connections are no longer just biochemical; they are deeply structural:
- Tirzepatide and Retatrutide: Week 13’s breakdown of Class B GPCRs (like the GLP-1 and GIP receptors) changed my perspective on these incretin mimetics. I now visualize how the amphipathic α-helices of these peptides interact with the large N-terminal extracellular domain (ECD) of the receptor, while the N-terminus of the peptide dives into the transmembrane bundle to activate the G-protein. Furthermore, understanding membrane biophysics (Week 6) helps me see how the C20/C19 fatty diacid chains on these peptides don't just bind albumin—they likely interact with the lipid bilayer surrounding the receptor, increasing local concentration.
- TB-500 (Thymosin Beta-4): This actin-sequestering peptide is a perfect real-world example of Week 8’s IDPs! Thymosin β4 is intrinsically disordered in solution but undergoes a massive disorder-to-order transition, folding into an extended α-helix when it binds to G-actin.
- BPC-157: We studied fibrous proteins and collagen assembly in Week 7. BPC-157 promotes tendon and ligament healing. While its exact receptor is still debated, understanding the Gly-X-Y repeat, hydroxyproline, and the incredible tensile strength of the collagen triple helix gives me a much deeper appreciation for the extracellular matrix targets BPC-157 is upregulating.
4. Questions to Investigate Further
Studying the Ramachandran plot (φ/ψ dihedral angles) from 1963 raised a massive question for me regarding synthetic peptide engineering. Peptides like Tirzepatide and Retatrutide use Aib (α-aminoisobutyric acid)—a non-canonical amino acid. How does having two methyl groups at the Cα position alter the allowed regions on the Ramachandran plot? My hypothesis is that Aib severely restricts conformational space, effectively forcing the peptide into an α-helical conformation, thereby reducing the entropic penalty of receptor binding. I want to model this computationally.
Additionally, I want to investigate how we can adapt AlphaFold2 or ESMFold to accurately predict the structures of lipidated or PEGylated peptides, since the current databases (like the PDB described by Berman et al.) and AI tools are heavily biased toward unmodified, canonical amino acid chains.
5. Confidence Ratings
Where I feel most confident:
I feel incredibly strong about the thermodynamics of folding, protein-protein interaction energetics (ΔG, Kd, enthalpy-entropy compensation), and structural prediction AI. My heavy mathematical and physics background (MATH201, PHYS201/202, CHEM301) made the statistical mechanics of binding and the interpretation of AF2's PAE (Predicted Aligned Error) and pLDDT scores feel very intuitive.
Where I feel least confident:
Membrane protein experimental biophysics (Week 6). The actual laboratory study of GPCRs using detergents, nanodiscs, SMALPs, and lipidic cubic phase crystallization sounds like absolute black magic. The fragility of these proteins outside their native lipid bilayers, and the challenge of keeping them folded in micelles, makes me deeply grateful for structural biologists. I am also still wrapping my head around the mathematical distinctions between "conformational selection" and "induced fit" in dynamic allostery (Week 10)—visualizing those multidimensional energy landscapes is going to require more review.
🗺️ Visual Aids & Key Takeaways
Diagram 1: The Ramachandran Plot — Energetically Allowed Regions for φ/ψ Dihedral Angles
ψ (degrees)
+180 ┌─────────────────────────────────────────────────────────────────┐
│ │
│ ╔══════════════╗ │
+120 │ ║ β-SHEET ║ │
│ ║ REGION ║ ░░░░░░░░░ │
│ ║ antiparallel ║ ░░░░░░░░░░░░░ │
+60 │ ╚══════════════╝ ░░░░░░░░░░░░░░░ │
│ ░░░░░░░░░░░░░░░░ │
│ ░░░░░░░░░░░░░░░ ← LEFT-HANDED │
0 │ ─────────────────────────────────────────────────── φ │
│ α-HELIX │
│ ╔════════════════════╗ (rare, found │
-60 │ ║ α-HELIX REGION ║ in Gly) │
│ ║ right-handed α ║ │
-120 │ ║ 310 helix nearby ║ │
│ ╚════════════════════╝ │
│ │
-180 └─────────────────────────────────────────────────────────────────┘
-180 -120 -60 0 +60 +120 +180
φ (degrees)
KEY:
╔══╗ = CORE ALLOWED REGIONS (dark on actual plot)
░░░░ = ADDITIONALLY ALLOWED (light grey on actual plot)
= DISALLOWED (steric clash between backbone atoms)
LANDMARK φ/ψ VALUES:
┌───────────────────────────────────┬────────────┬────────────┐
│ SECONDARY STRUCTURE │ φ (°) │ ψ (°) │
├───────────────────────────────────┼────────────┼────────────┤
│ Right-handed α-helix │ -57 │ -47 │
│ Parallel β-sheet │ -119 │ +113 │
│ Antiparallel β-sheet │ -139 │ +135 │
│ 310 helix │ -49 │ -26 │
│ π-helix │ -57 │ -70 │
│ Polyproline II helix │ -75 │ +145 │
└───────────────────────────────────┴────────────┴────────────┘
NOTE: Glycine has NO side chain → can occupy all four quadrants.
Proline is CONSTRAINED → φ fixed near -60° by pyrrolidine ring.
The Ramachandran plot maps every residue's backbone conformation in φ/ψ space; the clustered allowed regions reveal why only a small fraction of geometric space is structurally accessible, and outliers indicate either unusual residues (Gly, Pro) or potential model errors.
---Diagram 2: Levels of Protein Structure — From Sequence to Assembly
╔══════════════════════════════════════════════════════════════════════╗
║ HIERARCHY OF PROTEIN STRUCTURAL ORGANIZATION ║
╚══════════════════════════════════════════════════════════════════════╝
┌──────────────────────────────────────────────────────────────────────┐
│ PRIMARY STRUCTURE │
│ │
│ H2N─ Ala ─ Gly ─ Ser ─ Val ─ Leu ─ Pro ─ Asp ─ Phe ─ ... ─ COOH │
│ 1 2 3 4 5 6 7 8 │
│ │
│ Defined by: covalent peptide bonds (C–N, partial double bond) │
│ Information: amino acid sequence encodes ALL higher structure │
│ Anfinsen's dogma: sequence → structure (thermodynamic hypothesis) │
└────────────────────────────┬─────────────────────────────────────────┘
│ local backbone H-bonds form
↓
┌──────────────────────────────────────────────────────────────────────┐
│ SECONDARY STRUCTURE │
│ │
│ α-HELIX β-SHEET LOOPS/TURNS │
│ │
│ /\/\/\/\ →→→→→→→ ───╮ │
│ 3.6 res/turn strands H-bonded connect 2° │
│ i → i+4 H-bond laterally structure │
│ ~5.4 Å pitch parallel or elements │
│ antiparallel │
│ │
│ Backbone H-bonds only — side chains NOT involved at this level │
└────────────────────────────┬─────────────────────────────────────────┘
│ hydrophobic collapse + long-range contacts
↓
┌──────────────────────────────────────────────────────────────────────┐
│ TERTIARY STRUCTURE │
│ │
│ Forces: hydrophobic effect (dominant), H-bonds, van der Waals, │
│ electrostatics, disulfide bonds (in oxidizing environments)│
│ │
│ ┌─────────────────────────────────┐ │
│ │ SINGLE POLYPEPTIDE CHAIN │ │
│ │ folded into 3D structure │ │
│ │ ───────────────────────── │ │
│ │ Core: hydrophobic residues │ │
│ │ Surface: polar/charged │ │
│ │ Domains: semi-independent │ │
│ └─────────────────────────────────┘ │
└────────────────────────────┬─────────────────────────────────────────┘
│ non-covalent subunit assembly
↓
┌──────────────────────────────────────────────────────────────────────┐
│ QUATERNARY STRUCTURE │
│ │
│ Multiple polypeptide chains (subunits) associating: │
│ │
│ HOMODIMER HETEROTETRAMER (Hb) ICOSAHEDRAL CAPSID │
│ [A]─[A] [α][β] 60 × subunits │
│ [β][α] viral symmetry │
│ │
│ Symmetry types: Cyclic (Cn) · Dihedral (Dn) · Cubic (T, O, I) │
│ Interface: buried hydrophobic surface, H-bonds, salt bridges │
└──────────────────────────────────────────────────────────────────────┘
This hierarchy shows how each structural level emerges from the previous one through distinct physical forces, culminating in quaternary assemblies whose symmetry and interface properties determine biological function.
---Diagram 3: The Hsp70 Chaperone Cycle — ATP-Driven Client Folding
╔══════════════════════════════════════════════════════════════════════╗
║ Hsp70 / Hsp40 / NEF CHAPERONE CYCLE ║
╚══════════════════════════════════════════════════════════════════════╝
RIBOSOME or
misfolded substrate
│
│ exposed hydrophobic
│ peptide segment
↓
┌─────────────────────┐
│ Hsp40 (J-protein) │
│ binds client and │
│ delivers to Hsp70 │
└──────────┬──────────┘
│
↓
┌──────────────────────────────────────────────────────────────────────┐
│ │
│ ATP-BOUND Hsp70 │
│ ┌──────────────────────────────┐ │
│ │ NBD ──── linker ──── SBD │ │
│ │ (nucleotide (allosteric) │ │
│ │ binding) substrate │ │
│ │ binding │ │
│ └──────────────────────────────┘ │
│ │
│ ATP state: SBD LID open → LOW affinity for client │
│ rapid on/off → client can sample conformations │
│ │
└──────────────────────┬───────────────────────────────────────────────┘
│
│ Hsp40 stimulates ATPase
│ → ATP hydrolysis → ADP·Pi
↓
┌──────────────────────────────────────────────────────────────────────┐
│ │
│ ADP-BOUND Hsp70 │
│ ┌──────────────────────────────┐ │
│ │ NBD ──── linker ──── SBD │ ← LID CLOSED │
│ │ ■■■ │ │
│ │ CLIENT │ │
│ │ TRAPPED │ │
│ └──────────────────────────────┘ │
│ │
│ ADP state: SBD LID closed → HIGH affinity for client │
│ client HELD, cannot aggregate │
│ client may fold while held ("folding chamber") │
│ │
└──────────────────────┬───────────────────────────────────────────────┘
│
│ NEF (nucleotide exchange factor:
│ e.g., GrpE in bacteria, Bag1 in mammals)
│ → ejects ADP → ATP rebinds
↓
┌────────────────────────┐
│ CLIENT RELEASED │
└────────┬───────────────┘
│
┌──────────────┼──────────────┐
↓ ↓ ↓
CORRECTLY PARTIALLY HANDED TO
FOLDED ✓ FOLDED Hsp90 or
→ function → re-enter GroEL/GroES
cycle for further
folding
─────────────────────────────────────────────────────────────────────
ENERGY SUMMARY:
Each ATP hydrolysis cycle = one client binding/release event
ΔG of ATP hydrolysis drives directional, irreversible substrate release
The Hsp70 chaperone cycle uses the conformational change between ATP- and ADP-bound states to alternately capture and release client proteins, preventing aggregation and vectorially promoting productive folding through repeated bind-release iterations.
---Diagram 4: MWC vs. KNF Models of Allostery — Hemoglobin as Exemplar
╔══════════════════════════════════════════════════════════════════════╗
║ ALLOSTERIC MODELS: MWC (Concerted) vs. KNF (Sequential) ║
╚══════════════════════════════════════════════════════════════════════╝
HEMOGLOBIN: α₂β₂ tetramer, 4 O₂ binding sites (one per heme)
──────────────────────────────────────────────────────────────────────
MWC MODEL (Monod, Wyman, Changeux, 1965)
──────────────────────────────────────────────────────────────────────
Core assumption: ALL subunits are ALWAYS in the same state (T or R)
Ligand does NOT induce change — it SELECTS a state
T STATE R STATE
(Tense, low affinity) (Relaxed, high affinity)
[T][T] [R][R]
[T][T] ⇌ equilibrium ⇌ [R][R]
deoxy Hb oxy Hb
(dominant at (dominant when O₂
low [O₂]) present)
O₂ binds R preferentially → shifts equilibrium toward R state
ALL four subunits flip together (concerted)
Predicts: no mixed-state intermediates
L = [T]₀/[R]₀ (allosteric constant, >> 1 for Hb)
c = KR/KT (ratio of affinities)
──────────────────────────────────────────────────────────────────────
KNF MODEL (Koshland, Nemethy, Filmer, 1966)
──────────────────────────────────────────────────────────────────────
Core assumption: Subunits change conformation SEQUENTIALLY
Ligand INDUCES the conformational change
Step 1: [T][T] + O₂ → [R●][T] (1 subunit flips)
[T][T] [T][T]
Step 2: [R●][T] + O₂ → [R●][R●] (neighbor flips due to
[T][T] [T][T] subunit communication)
Step 3: ... and so on until all 4 bound
Induced fit at each step; mixed T/R intermediates EXIST
Better explains negative cooperativity in some systems
──────────────────────────────────────────────────────────────────────
COMPARISON TABLE
──────────────────────────────────────────────────────────────────────
┌────────────────────────┬──────────────────────┬────────────────────┐
│ FEATURE │ MWC (Concerted) │ KNF (Sequential) │
├────────────────────────┼──────────────────────┼────────────────────┤
│ Conformational states │ T or R only │ Mixed T/R allowed │
│ Mechanism │ Pre-existing equil. │ Induced fit │
│ Mixed intermediates │ NO │ YES │
│ Negative cooperativity │ Cannot explain │ Can explain │
│ Mathematical elegance │ High (2-parameter) │ More complex │
│ Fit to Hb O₂ data │ Good │ Also good │
└────────────────────────┴──────────────────────┴────────────────────┘
MODERN VIEW: Conformational SELECTION + Induced FIT both operate.
Energy landscape perspective: multiple minima, population shifts.
Bohr Effect: CO₂/H⁺ stabilize T state → right-shifts O₂ curve
Histidine β146 salt bridge key ionizable residue
The two classical allosteric models make testable, mechanistically distinct predictions about intermediate states in cooperative ligand binding, with hemoglobin serving as the paradigm case where both models fit binding curves but differ fundamentally in their structural logic.
---Diagram 5: AlphaFold2 Architecture — From Sequence to Structure
╔══════════════════════════════════════════════════════════════════════╗
║ AlphaFold2: INFORMATION FLOW AND MODULE ARCHITECTURE ║
╚══════════════════════════════════════════════════════════════════════╝
INPUT
┌────────────────────────────────────────────────────────────────┐
│ Query amino acid sequence (N residues) │
└──────────────────┬─────────────────────────────────────────────┘
│
┌───────────┴───────────┐
↓ ↓
┌──────────────┐ ┌──────────────────────┐
│ DATABASE │ │ STRUCTURE DATABASES │
│ SEARCH │ │ (PDB templates) │
│ (UniRef90, │ └──────────┬────────────┘
│ MGnify, BFD)│ │
└──────┬───────┘ │
│ │
↓ ↓
┌─────────────────┐ ┌────────────────────┐
│ MULTIPLE │ │ TEMPLATE FEATURES │
│ SEQUENCE │ │ (residue pairs, │
│ ALIGNMENT (MSA) │ │ distances, angles)│
│ N × depth mat. │ └────────────────────┘
└────────┬─────────┘ │
│ │
└────────────┬─────────────┘
↓
╔═══════════════════════════════════════════════════════════════╗
║ EVOFORMER STACK ║
║ (48 identical blocks iterated) ║
║ ║
║ MSA REPRESENTATION ⇌ PAIR REPRESENTATION ║
║ (rows = sequences) (N × N matrix, ║
║ (cols = positions) pairwise relationships) ║
║ ║
║ Row-wise attention → updates MSA from sequences ║
║ Column-wise attention → shares info across positions ║
║ Triangle updates → enforces geometric consistency ║
║ in pair matrix (if AB close + BC close → ║
║ AC must be close) ║
║ ║
║ Outer product mean → projects MSA → pair features ║
╚═══════════════════════════════════╤═══════════════════════════╝
│
↓
╔═══════════════════════════════════════════════════════════════╗
║ STRUCTURE MODULE (8 blocks) ║
║ ║
║ Single residue representation + Invariant Point Attention ║
║ Each residue carries: position (x,y,z) + rotation frame ║
║ IPA: attention in 3D space, equivariant to rotation ║
║ ║
║ Outputs backbone frames → side chains (χ angles) added ║
║ via torsion angle prediction head ║
║ ║
║ RECYCLING: entire process repeated 3× (input updated ║
║ with previous prediction → progressively refined) ║
╚═══════════════════════════════════╤═══════════════════════════╝
│
↓
┌─────────────────────────────────────────────────────────────┐
│ OUTPUT │
│ │
│ 3D COORDINATES (all-atom, including side chains) │
│ │
│ pLDDT score (per-residue, 0–100) │
│ > 90 = very high confidence │
│ 70–90 = confident │
│ 50–70 = low confidence (often disordered regions) │
│ < 50 = should not be interpreted structurally │
│ │
│ PAE matrix (Predicted Aligned Error, N × N) │
│ = expected error in position of residue j │
│ when residue i is used as alignment anchor │
│ → use for: domain boundaries, inter-domain confidence │
└─────────────────────────────────────────────────────────────┘
AlphaFold2 transforms a query sequence into an accurate 3D structure through two coupled representation tracks — the MSA-derived evolutionary information and the pairwise geometric constraints — refined iteratively before a structure module places each residue in 3D space using equivariant attention.
---Key Takeaways
- Anfinsen's thermodynamic hypothesis (1973): The native structure of a protein is the thermodynamic global free energy minimum accessible to that sequence under physiological conditions. Demonstrated by complete denaturation and renaturation of bovine pancreatic ribonuclease A with loss and recovery of enzymatic activity — proving that all the information for folding is encoded in the primary sequence, requiring no template.
- Peptide bond planarity and its consequence: The C–N peptide bond has ~40% double-bond character due to resonance delocalization of the nitrogen lone pair into the carbonyl. This constrains the six atoms of each peptide unit to a plane and fixes ω ≈ 180° (trans) in almost all cases. Proline-containing bonds (Xaa-Pro) have a ~10–40% cis isomer frequency because the energy difference between cis and trans is much smaller for tertiary amides — a feature exploited in cell signaling (Pin1 prolyl isomerase).
- The Ramachandran plot encodes stereochemical reality: Only ~27% of φ/ψ space is energetically accessible for a general amino acid (Gly is the exception — it accesses all four quadrants due to absence of a Cβ). The α-helical region clusters around φ = −57°, ψ = −47°; β-sheet around φ = −120°, ψ = +115°. Outliers in Ramachandran analysis of X-ray structures signal either modeling errors or functionally important strained conformations.
- The hydrophobic effect is entropic, not enthalpic at its core: Burying nonpolar surface area upon folding releases ordered water molecules from hydrophobic surfaces, increasing solvent entropy (positive ΔS_solvent). The associated heat capacity change ΔCp upon folding is large and negative (approximately −10 to −50 cal/mol/K per residue buried), meaning the protein is most stable at a temperature where both hydrophobic and conformational entropy terms are optimal — explaining cold denaturation.
- Hemoglobin cooperativity: structural basis at atomic resolution: In deoxy hemoglobin (T state), the iron atom sits ~0.6 Å out of the porphyrin plane because the Fe²⁺ is high-spin. O₂ binding converts Fe²⁺ to low-spin, pulling it into the plane; this 0.6 Å movement is mechanically transmitted through the proximal histidine (His F8) to the F-helix, disrupting the α₁β₂ interface salt bridges and flipping the entire tetramer from T to R state. The Bohr effect arises because CO₂ and H⁺ stabilize T-state salt bridges (including the His β146 / Asp β94 bridge), shifting the O₂ dissociation curve rightward in active tissues.
- GPCRs follow a 7-transmembrane α-helical topology with conserved activation mechanism: Agonist binding to the extracellular/TM bundle opens an intracellular cavity by rotating TM6 outward ~14 Å (in class A GPCRs like β₂-adrenergic receptor) — the "DRY" motif (Asp-Arg-Tyr) at the TM3/ICL2 junction and the NPxxY motif in TM7 are structural pivots. This cavity accommodates Gα subunit C-terminus, completing the ternary complex. Biased agonism arises when different ligands stabilize distinct receptor conformations that preferentially couple to Gα vs. β-arrestin.
- Collagen's triple helix requires Gly at every third position: The triple helix packs three polyproline-II-type left-handed helices into a right-handed superhelix. Glycine must occupy every third residue (Gly-X-Y repeat) because only Gly's Hα fits in the steric center of the triple helix. Hydroxyproline in the Y position stabilizes the triple helix by ~1 kcal/mol per residue through water-bridged hydrogen bonds involving the 4-OH group; hydroxylation requires vitamin C as a cofactor for prolyl-4-hydroxylase, explaining the molecular basis of scurvy.
- Intrinsically disordered proteins (IDPs) are not failures of folding: IDPs are enriched in charged residues, Pro, and Gly, and depleted in hydrophobic residues — their disorder is evolutionarily encoded and functional. Coupled folding-and-binding (where an IDP folds upon interacting with a partner) can provide affinity while retaining specificity, fast on-rates, and the ability to bind multiple partners with different induced conformations. The entropic cost of ordering an IDP upon binding is real and must be compensated by intermolecular contacts — this is the molecular explanation for why many IDP-target interactions have Kd values in the µM range despite high specificity.
- Hot spots dominate protein-protein interaction energy: Alanine-scanning mutagenesis of protein interfaces consistently reveals that ~20% of interface residues contribute ~80% of binding free energy (ΔΔG upon Ala substitution ≥ 2 kcal/mol = hot spot). Hot spots are surrounded by hydrophobic "O-ring" residues that exclude water and protect the hot spot's electrostatic and H-bond interactions. This architecture explains why small molecules can inhibit PPIs by occupying hot spot pockets, even on apparently flat interfaces.
- The Hsp70 cycle is a nucleotide-driven affinity switch: In the ATP-bound state, Hsp70's substrate-binding domain (SBD) lid is open and client affinity is low (Kd ~µM); ATP hydrolysis stimulated by Hsp40 (J-domain proteins) closes the SBD lid, increasing client affinity 100-1000-fold (Kd ~nM). Nucleotide exchange factors (NEFs: GrpE in bacteria, Bag-family, HspBP1, or Hsp110 in eukaryotes) catalyze ADP release, reopening the lid and releasing the client. Cancer cells overexpress Hsp70/Hsp90 to buffer the structural instability of onco-mutant kinases — making this cycle a therapeutic target.
- Serine protease catalytic triad: charge relay, not just geometry: The Ser-His-Asp triad in chymotrypsin functions as a "charge relay system": Asp102 (buried, carboxylate) hydrogen-bonds His57 imidazole, raising its pKa and enabling it to deprotonate Ser195 for nucleophilic attack on the scissile peptide bond. The oxyanion hole (backbone NHs of Gly193 and Ser195) stabilizes the tetrahedral intermediate by ~5 kcal/mol. Zymogen activation requires proteolytic removal of a pro-peptide (e.g., trypsinogen → trypsin by enterokinase) that repositions Ile16, completing the oxyanion hole and the substrate-binding S1 pocket.
- AlphaFold2's Evoformer uses triangular multiplicative updates to embed geometry: The key innovation is that the N×N pair representation is updated using triangle inequalities: if residues i–j and j–k have known spatial relationships, the algorithm directly updates the i–k pair. This enforces Euclidean consistency before any coordinates are generated, allowing the structure module to place atoms accurately. The pLDDT score is NOT a measure of experimental resolution — it is a prediction of what the LDDT-Cα score would be if the prediction were compared to a true experimental structure; regions with pLDDT < 50 are disordered or genuinely uncertain and should not be modeled.
- SCOP2 and CATH provide complementary fold-space classification: SCOP2 (Structural Classification of Proteins) classifies by evolutionary relationship first, with four levels: Class → Fold → Superfamily → Family. CATH (Class, Architecture, Topology, Homologous superfamily) uses a four-level system with automated methods. The TIM barrel (β/α)₈ is the most common enzyme fold (~10% of all known enzyme structures); the Rossmann fold (βαβαβ) binds NAD⁺/FAD through a conserved glycine-rich loop. Recognizing these folds enables functional inference for uncharacterized proteins.
- Disulfide bonds are only stable in oxidizing environments: The cytoplasm is reducing (GSH/GSSG ratio ~100:1) and disulfide bonds are thermodynamically unstable there. Disulfides form in the endoplasmic reticulum (eukaryotes) via protein disulfide isomerase (PDI), which catalyzes both formation and isomerization of disulfide bonds. Isomerization is critical because the thermodynamically correct disulfide bond pattern must be found among (2n−1)!! possible pairings — for insulin with 3 disulfide bonds, this is 15 possible arrangements; PDI's active-site Cys-XX-Cys motif allows it to form mixed disulfides with substrates and then rearrange them.
References
- Anfinsen, C.B., 'Principles That Govern the Folding of Protein Chains,' Science, 181(4096):223–230, 1973. DOI: 10.1126/science.181.4096.223
- Ramachandran, G.N., Ramakrishnan, C., and Sasisekharan, V., 'Stereochemistry of Polypeptide Chain Configurations,' Journal of Molecular Biology, 7(1):95–99, 1963. DOI: 10.1016/S0022-2836(63)80023-6
- Kendrew, J.C., Bodo, G., Dintzis, H.M., Parrish, R.G., Wyckoff, H., and Phillips, D.C., 'A Three-Dimensional Model of the Myoglobin Molecule Obtained by X-Ray Analysis,' Nature, 181(4610):662–666, 1958. DOI: 10.1038/181662a0
- Jumper, J., Evans, R., Pritzel, A. et al., 'Highly Accurate Protein Structure Prediction with AlphaFold,' Nature, 596:583–589, 2021. DOI: 10.1038/s41586-021-03819-2
- Berman, H.M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T.N., Weissig, H., Shindyalov, I.N., and Bourne, P.E., 'The Protein Data Bank,' Nucleic Acids Research, 28(1):235–242, 2000. DOI: 10.1093/nar/28.1.235
- Creighton, T.E., Proteins: Structures and Molecular Properties, 2nd edition, W.H. Freeman, New York, 1993.
- Branden, C. and Tooze, J., Introduction to Protein Structure, 2nd edition, Garland Publishing, New York, 1999.
- Berg, J.M., Tymoczko, J.L., and Stryer, L., Biochemistry, 9th edition, W.H. Freeman/Macmillan, New York, 2019.
- Lesk, A.M., Introduction to Protein Science: Architecture, Function, and Genomics, 3rd edition, Oxford University Press, Oxford, 2016.
- Monod, J., Wyman, J., and Changeux, J.-P., 'On the Nature of Allosteric Transitions: A Plausible Model,' Journal of Molecular Biology, 12(1):88–118, 1965. DOI: 10.1016/S0022-2836(65)80285-6
- Koshland, D.E., Némethy, G., and Filmer, D., 'Comparison of Experimental Binding Data and Theoretical Models in Proteins Containing Subunits,' Biochemistry, 5(1):365–385, 1966. DOI: 10.1021/bi00865a047
- Hartl, F.U., Bracher, A., and Hayer-Hartl, M., 'Molecular Chaperones in Protein Folding and Proteostasis,' Nature, 475:324–332, 2011. DOI: 10.1038/nature10317
- Clore, G.M. and Gronenborn, A.M., 'Structures of Larger Proteins in Solution: Three- and Four-Dimensional Heteronuclear NMR Spectroscopy,' Science, 252(5011):1390–1399, 1991. DOI: 10.1126/science.2047852
- Wells, J.A. and McClendon, C.L., 'Reaching for High-Hanging Fruit in Drug Discovery at Protein–Protein Interfaces,' Nature, 450:1001–1009, 2007. DOI: 10.1038/nature06526
- Uversky, V.N., 'Intrinsically Disordered Proteins and Their "Mysterious" (Meta)Physics,' Frontiers in Physics, 7:10, 2019. DOI: 10.3389/fphy.2019.00010