Turns, Loops, and Bends

Turns and loops are not a historical curiosity confined to crystallographic catalogs — they are actively engineered and targeted in modern biotechnology. Antibody complementarity-determining regions (CDRs), the loops that define antigen specificity in the variable domains of immunoglobulins, are built almost entirely from turn and loop segments grafted onto a conserved β-sheet framework; therapeutic antibody engineering (humanization, affinity maturation, CDR grafting) is fundamentally the art of redesigning these loops while preserving the canonical turn geometries that hold the antigen-binding surface in the correct shape. Structure-based drug design increasingly targets flexible loops directly, since loops that close over an active site upon substrate binding (the classic example being the glycine-rich "P-loop" that wraps around the phosphates of ATP/GTP in kinases and G-proteins) create transient pockets exploitable by allosteric inhibitors. In protein engineering and de novo design, β-hairpins are routinely nucleated by explicitly installing a type I′ or type II′ turn with the required Gly or Asn at the strained position — precisely the logic of the steric derivation above — because getting the turn residue identity wrong is one of the most common reasons a designed hairpin fails to fold. In structure prediction (including modern deep-learning methods), loops remain the hardest regions to model accurately, precisely because they lack the repeating (φ,ψ) signature that makes helices and strands so predictable from sequence alone; loop modeling accuracy is still a standard benchmark metric separately reported from core secondary-structure accuracy. Finally, turns are directly relevant to folding kinetics: β-hairpin turns are widely believed to nucleate folding by forming early and locally, providing a scaffold onto which the flanking strands zip up — a idea tested extensively by fast-folding kinetics experiments on isolated hairpin peptides. Understanding turns is therefore a direct bridge to how Supersecondary Structure and Motifs assemble and, ultimately, to how the full three-dimensional Tertiary Structure and the Fold of a protein is built from a limited alphabet of connectable pieces.

Ask the tutor: How does the geometry of a specific turn type dictate whether a β-hairpin twists left- or right-handed as it extends into a larger β-sheet?

Every turn faces the same geometric problem that a strand of β-sheet or a helix does not: the backbone must change direction by roughly 180° within the smallest possible number of residues, without breaking any covalent bond angle or bond length, and — for a stable turn rather than a floppy hinge — while forming a hydrogen bond that locks the reversal in place. Recall that each residue's backbone conformation is described by two dihedral angles, φ (the C–N–Cα–C rotation) and ψ (the N–Cα–C–N rotation), and that the allowed (φ,ψ) combinations cluster in a few regions of the Ramachandran plot: the right-handed helical region near (−60°, −45°), the β-sheet region near (−120°, +120°), and a small, energetically costly left-handed helical region near (+60°, +45°) that is disfavored for all residues except glycine. A β-turn works by placing residues i+1 and i+2 into specific, often unusual combinations of these regions — sometimes one of them squarely in the left-handed region — so that the vector sum of the backbone bonds swings the chain through 180° in just two peptide units. The most common β-turn hydrogen bond connects the backbone carbonyl oxygen of residue i to the backbone amide hydrogen of residue i+3, forming a ten-atom ring when you count around the loop C=O···H–N–Cα–C(=O)–N–Cα–C(=O)–N–Cα. This is why turns are sometimes still called "10-membered-ring" hydrogen-bonded structures, distinguishing them from the α-helix's 13-membered-ring (i, i+4) hydrogen bond and the smaller 7-membered-ring γ-turn (i, i+2).

Not every turn uses the same (φ,ψ) pair. Historically, turns were sorted into types I through VIII based on crystallographic surveys, but two types dominate: type I and type II, plus their "primed" mirror-related variants I′ and II′ in which the sign of both φ and ψ at position i+2 (and often i+1) is flipped. The distinction between type I and type II is small in absolute angle terms but large in consequence, because it changes whether a bulky side chain at position i+2 clashes sterically with the carbonyl of residue i+1.

The defining geometric relationship of a β-turn is a distance criterion applied to backbone atoms, together with a dihedral-angle classification. First, the distance test used to identify a "tight" reverse turn from Cartesian coordinates:

where:

• $d(\mathrm{C\alpha}_i,\mathrm{C\alpha}_{i+3})$ is the distance between the α-carbons of the first and fourth residues of the turn,
• 7 Å is the empirical cutoff (compare: a fully extended β-strand would place these Cα atoms roughly 10.5 Å apart, since each extended residue advances the chain about 3.5 Å),
• this criterion alone does not require a hydrogen bond — it only says the chain has curled back sharply.

Reading: an extended, non-turning stretch of chain advances the backbone by about 3.5 Å per residue; three peptide-bond spans without any turn would separate Cα(i) and Cα(i+3) by roughly $3 \times 3.5 = 10.5$ Å. A turn compresses that separation below 7 Å — the chain has reversed direction rather than continuing to extend. Second, the energetic/structural criterion that promotes a geometric "close turn" into a true hydrogen-bonded β-turn:

where:

• $r_{\mathrm{N\cdots O}}$ is the nitrogen–oxygen distance between the amide N–H of residue i+3 and the carbonyl O of residue i,
• the angle criterion enforces near-linear H-bond geometry, which maximizes the dipole–dipole and n→σ* donor–acceptor overlap that gives hydrogen bonds their partial covalent character,
• this is the same distance/angle logic used for α-helix and β-sheet hydrogen bonds — turns are not exempt from ordinary H-bond stereochemistry.

Third, the classification equations: a turn is assigned to a type by comparing the measured (φ,ψ) of residues i+1 and i+2 to canonical reference values.

where:

• primed types (I′, II′) are obtained by negating all four angles — the mirror image in (φ,ψ) space, corresponding to a left-handed-helix-like local conformation,
• a residue is assigned to a type if its actual (φ,ψ) lies within roughly ±30° of these canonical values (real turns are never exactly canonical; the classification is a nearest-neighbor assignment, not an exact match),
• position i+2 in type II and II′ turns requires positive φ, which is sterically forbidden for all L-amino acids except glycine (see derivation below) — hence the strong glycine preference at that position.

The energetic stabilization of a turn versus an unstructured, non-hydrogen-bonded reversal can be estimated the same way as any backbone H-bond:

where:

• $\Delta H_{\mathrm{HB}}$ is the enthalpic stabilization from the i,(i+3) backbone hydrogen bond, typically −2 to −5 kJ mol⁻¹ for an isolated backbone H-bond in water (much weaker than in vacuum, because the alternative is an H-bond to solvent),
• $\Delta S_{\mathrm{conf}}$ is the (unfavorable, i.e., $-\Delta S$ makes $-T\Delta S>0$) loss of conformational entropy in fixing four consecutive residues into one narrow region of (φ,ψ) space,
• net turn stability is therefore marginal by itself; turns are usually stabilized cooperatively by the secondary structure they connect (e.g., a hairpin's flanking β-strand hydrogen bonds) rather than standing alone.

Example 1 (foundational): classify a turn from measured dihedral angles. Suppose X-ray coordinates give residues i+1 and i+2 of a candidate turn the following backbone dihedrals:

Compute the angular distance to each canonical type. For type I, $(-60,-30,-90,0)$:

For type II, $(-60,+120,+90,0)$:

For type I′, $(+60,+30,+90,0)$:

Type I gives by far the smallest deviation (29°, with every individual angle within 11° of canonical), so this is confidently classified as a type I turn. Since φ at i+2 is −84° (negative), no glycine is sterically mandated at that position — consistent with type I turns tolerating a wide range of residues, including bulky ones, at i+2 (the steric clash discussed in the derivation only arises for positive φ).

Limiting-case check: if the measured angles were exactly the canonical type I values (−60,−30,−90,0), $D_{\mathrm I}$ would be exactly 0°, confirming the formula collapses correctly to a perfect match; real turns, as here, deviate by tens of degrees while still being unambiguously classifiable because the deviation from the correct type is always much smaller than from any competing type.

Example 2 (applied): estimating turn hydrogen-bond geometry from a hairpin loop and testing the 7-Å criterion. A designed β-hairpin has a two-residue turn connecting antiparallel strands. Backbone coordinates (in Å, arbitrary origin) place:

Compute the Cα(i)–Cα(i+3) distance:

Since 4.23 Å is well under the 7 Å cutoff (and much less than the ~10.5 Å expected for an untucked extended stretch of the same length), this satisfies the geometric turn criterion convincingly — the chain has genuinely reversed direction, not merely bent slightly. Now suppose the same structure places the carbonyl oxygen of residue i at $(12.0, 9.0, 0.3)$ and the amide nitrogen of residue i+3 at $(13.6, 11.0, 1.0)$. The N···O distance:

This falls slightly below the typical 2.9–3.2 Å window quoted for a relaxed backbone H-bond, meaning the geometry here is unusually short/compressed — plausible in a rigid designed hairpin turn where the H-bond is under some geometric strain, but it would need the N–H···O angle checked (not given here) before concluding the bond is ideally linear; a short N···O distance with a poor angle can still be a weak or geometrically distorted hydrogen bond.

Limiting-case check: if instead $\mathrm{C\alpha}_i$ and $\mathrm{C\alpha}_{i+3}$ were separated by the fully extended-chain value of $3 \times 3.5 = 10.5$ Å (three peptide spans with no reversal), the 7 Å turn criterion would correctly reject this stretch as "not a turn" — consistent with the physical picture that only genuine chain reversal, not merely three consecutive residues, satisfies the geometric test.

Example 3 (exam/competition level): free-energy comparison of forcing a bulky residue into position i+2 of a type II turn versus a type I turn. A designed loop requires a chain reversal at a fixed sequence position, and the designer must choose between engineering a type I turn or a type II turn, with a leucine side chain unavoidably present at position i+2 (say, for a downstream disulfide-bonding constraint that fixes the sequence). Estimate, using the steric-clash argument, which turn type is thermodynamically favorable, and give an order-of-magnitude free-energy penalty for forcing leucine into the disfavored type. Leucine has a branched Cβ–Cγ(CH₃)₂ side chain — bulkier than even a simple alanine methyl. Using the hard-sphere Ramachandran argument (Step 2–3 of the glycine derivation), a positive φ at a residue bearing any Cβ substituent falls almost entirely within the disallowed region: essentially every accessible (φ,ψ) combination with $\varphi>0°$ for a non-glycine residue incurs a steric clash energy on the order of several $k_BT$ (roughly 8–20 kJ mol⁻¹, comparable to a full van der Waals overlap of 0.5–1.0 Å between heavy atoms, since each 0.1 Å of overlap in a hard-sphere-like repulsive potential typically costs several kJ mol⁻¹ once atoms are already near contact). For type II, position i+2 requires φ ≈ +90° — squarely in the region disallowed for leucine. This is not a marginal preference; crystallographic surveys essentially never find leucine at this position in type II turns (occupancy near zero), which by a Boltzmann argument,

with an observed ratio on the order of $10^{-2}$ to $10^{-3}$ (a rough but defensible estimate from turn-residue propensity tables), implies

a penalty of roughly 14 kJ mol⁻¹ — squarely in the expected range for a serious but not catastrophic steric clash, consistent with the qualitative hard-sphere reasoning. For type I, position i+2 has φ ≈ −90° (negative), which is the normal, low-strain region accessible to any residue including leucine, so the type I turn should be strongly preferred by the designer, incurring no comparable steric penalty.

Limiting-case check: if the design instead required a bulky residue at position i+1 rather than i+2, the same steric argument would flip depending on turn type — type II′ requires positive φ at i+1 (again disfavoring bulky residues there) while type I′ requires positive φ at both i+1 and i+2, making type I′ turns essentially glycine-only at both central positions; this correctly predicts that type I′ turns are markedly rarer than type I in real proteins, exactly as observed.

Reverse turns as a distinct structural category emerged from the same wave of stereochemical reasoning about the peptide backbone that gave rise to the Ramachandran plot itself. G. N. Ramachandran, working at the University of Madras in the early 1960s, built physical hard-sphere models of dipeptide units and systematically mapped which (φ,ψ) combinations avoid atomic overlap — the same steric logic used throughout this article to explain why glycine dominates positive-φ turn positions. His 1963 paper with Ramakrishnan and Sasisekharan did not describe "turns" by that name, but it supplied the exact geometric grammar — the allowed and forbidden regions of the Ramachandran plot — that later crystallographers used to recognize and classify reverse turns as a distinct, non-helical, non-sheet structural class once enough high-resolution protein structures existed to survey them systematically (work extended through the 1970s–1980s by crystallographers including Venkatachalam, who in 1968 first formally proposed the type I/type II turn classification from a theoretical conformational-energy analysis of tripeptide chain reversals).