Primary Structure and the Amino Acid Sequence

Primary-structure determination is now overwhelmingly done indirectly, by sequencing the gene and translating through the genetic code, backed up and cross-checked by tandem mass spectrometry (MS/MS), which fragments peptide ions along the backbone and reads out b- and y-ion mass ladders to reconstruct sequence directly from the protein itself — essential for detecting post-translational modifications invisible to genomic sequencing alone (phosphorylation, glycosylation, proteolytic processing, disulfide connectivity). Edman degradation survives in niche roles: confirming the exact N-terminus of a recombinant therapeutic protein, checking that a signal peptide has been correctly cleaved, or verifying blocked/modified N-termini. Clinically, single-residue changes in primary structure are the direct molecular cause of a large fraction of monogenic diseases: sickle-cell disease arises from a single Glu→Val substitution at position 6 of the hemoglobin β-chain, changing the surface chemistry of deoxyhemoglobin enough to permit pathological polymerization. Cystic fibrosis, in its most common form, arises from a three-base deletion removing a single phenylalanine (ΔF508) from the CFTR chloride channel, causing a primary-structure defect that leads to a downstream folding catastrophe. Cancer genomics routinely reports point mutations in primary structure (e.g., KRAS G12D, BRAF V600E) that are directly actionable with targeted small-molecule and antibody therapies. In biotechnology, engineered primary structure is now a design target in its own right: therapeutic antibodies are affinity-matured by systematically varying residues in complementarity-determining regions; de novo protein design software (Rosetta, AlphaFold-based inverse folding tools) searches sequence space computationally to find primary structures predicted to fold into a target three-dimensional shape, directly exploiting the same $20^n$ combinatorics developed here, now navigated by machine learning rather than natural selection alone. Forensic and evolutionary biology use sequence comparison (percent identity, conservative-substitution scoring matrices like BLOSUM and PAM) to establish species relationships and protein family membership — quantitative descendants of exactly the overlap-and-compare logic Sanger pioneered by hand.

Ask the tutor: How does the partial double-bond character of the peptide bond actually constrain which secondary structures a given primary sequence can adopt?

A protein is built from L-α-amino acids, each with a central (α) carbon bearing an amino group, a carboxyl group, a hydrogen, and a distinguishing side chain (R group). In isolation, each amino acid is a small zwitterion. In a polypeptide, all but two of these building blocks have lost the chemical identity of a "free" amino acid: their α-amino and α-carboxyl groups have been consumed in forming amide linkages to neighbors. What survives in the interior of the chain is the residue — the repeating $\text{–NH–CHR–CO–}$ unit — plus the intact side chain, which is the only part of an interior residue exposed to chemistry and folding. Only at the two ends of the chain does a "whole" amino acid group remain: an unreacted α-amino group at one end (the N-terminus or amino terminus) and an unreacted α-carboxyl group at the other (the C-terminus or carboxyl terminus). This is not a bookkeeping convention. The N-terminal amino group typically carries a formal positive charge near physiological pH (pKa ≈ 8–9, so it is protonated, $\text{–NH}_3^+$), while the C-terminal carboxyl is typically deprotonated ($\text{–COO}^-$, pKa ≈ 3–4). These two ends have different reactivity, different roles in post-translational modification (N-terminal acetylation, C-terminal amidation), different targeting signals recognized by cellular machinery, and different chemistry available for sequencing reactions. A protein read backward is not the same molecule, and in general it is not even a chemically sensible molecule, because the peptide bond geometry and side-chain attachment pattern are asymmetric with respect to chain direction. The peptide bond itself forms by a condensation (dehydration) reaction: the carboxyl carbon of one amino acid attacks — mechanistically, is attacked by — the α-amino nitrogen of the next, expelling water and creating a covalent C–N bond flanked by a carbonyl. Ribosomes catalyze this same net transformation (via aminoacyl-tRNA chemistry, not free amino acids, and using peptidyl transfer rather than direct condensation) strictly in the N-to-C direction: the growing chain's C-terminus is extended by adding new residues, while the N-terminus, fixed early in synthesis, is never revisited. This built-in vectorial character of ribosomal synthesis is the deep reason the N→C convention for writing, numbering, and reading sequences is not arbitrary — it mirrors the actual chemical order of construction.

The chemical logic of the peptide bond has an important structural consequence foreshadowed here but developed properly elsewhere: the C–N bond in the amide linkage has significant partial double-bond character from resonance delocalization of the nitrogen lone pair into the carbonyl π system. That resonance — the subject of The Peptide Bond and Its Partial Double-Bond Character — is what locks the six backbone atoms of each peptide unit into a rigid, nearly always-trans planar group, which in turn is the geometric constraint that makes regular secondary structures like the α-helix and β-sheet possible. Primary structure supplies the sequence; the planarity of that same bond supplies the geometric alphabet from which secondary structure is built.

Peptide bond stoichiometry. For a chain of $n$ residues, exactly $n-1$ peptide bonds are formed (or broken), and exactly $n-1$ water molecules are involved:

where: $n$ is the number of amino acid residues in the final chain; each condensation step joins the growing chain's C-terminus to the next amino acid's amino group, releasing one water.

Reading. A protein of length $n$ has $n-1$ peptide bonds, not $n$ — the very first and very last residues each contribute only one reactive group to the backbone (the free N- or C-terminal group is "used up" by the environment, not by another residue).

Molecular mass bookkeeping. If $M_i$ is the free-amino-acid molar mass of residue $i$, the polypeptide's molar mass is

where: $M_{\text{H}_2\text{O}} = 18.02\ \text{g/mol}$; the subtraction accounts for the $(n-1)$ water molecules lost during synthesis.

Sequence space. The number of distinct possible sequences of length $n$ drawn from the 20 standard amino acids is

where: $\Omega(n)$ is the total combinatorial sequence space; this treats each position as independently able to hold any of the 20 residues (a generous upper bound, since real proteins have compositional and structural biases).

Sequence information content. Treating each residue position as drawn from an alphabet of 20 symbols with maximal (uniform) entropy gives an information content per residue of

so that a protein of $n$ residues carries at most

where: $I$ is the Shannon information per position assuming all 20 amino acids are equiprobable (an idealization — real amino acid frequencies are unequal, so the true information content per position is somewhat lower, around 4.0–4.2 bits depending on the proteome).

Derivation 1 — why the number of peptide bonds is always one less than the number of residues, and its energetic bookkeeping

Goal. Show rigorously, by induction on chain length, that an $n$-residue polypeptide contains exactly $n-1$ peptide bonds and releases exactly $n-1$ waters on formation (equivalently consumes $n-1$ waters on complete hydrolysis), then use this to compute the standard free energy of forming a specific chain from its free amino acids.

Step 1 — base case. Take $n = 2$: two free amino acids, $\text{aa}_1$ and $\text{aa}_2$. Condensation forms one amide bond between the carboxyl carbon of $\text{aa}_1$ and the amino nitrogen of $\text{aa}_2$, releasing one $\text{H}_2\text{O}$. This is a single peptide bond: $n - 1 = 2 - 1 = 1$. Base case holds.

Step 2 — inductive step. Suppose a chain of $k$ residues has exactly $k-1$ peptide bonds (inductive hypothesis). Extending the chain by one more residue at the C-terminus means forming exactly one new peptide bond between the current chain's C-terminal carboxyl and the new residue's amino group, releasing one more water. The new chain has $k+1$ residues and $(k-1) + 1 = k$ peptide bonds, which equals $(k+1) - 1$. The relation $n-1$ bonds for $n$ residues therefore holds for all $n \geq 1$ (with the trivial case $n=1$: a single free amino acid, zero peptide bonds).

Step 3 — energetics of condensation. Peptide bond formation in vitro, amino acid + amino acid → dipeptide + water, is endergonic: the standard free energy change is approximately

This positive value is why free amino acids do not spontaneously polymerize in dilute aqueous solution — hydrolysis (the reverse reaction) is thermodynamically favored, with $\Delta G^{\circ\prime}_{\text{hydrolysis}} \approx -21\ \text{kJ/mol}$ per bond. For a hypothetical direct (non-ribosomal) synthesis of an $n$-residue chain from its $n$ free amino acids, the naive total would be

Step 4 — why cells can still build proteins. Cells do not condense free amino acids directly; ribosomes couple peptide bond formation to the hydrolysis of high-energy phosphoanhydride and ester bonds (ATP consumed in amino-acid activation to aminoacyl-adenylate, then GTP hydrolyzed during translation elongation), making the overall cellular process strongly exergonic despite the intrinsically uphill chemistry of the isolated condensation step. The peptide bond itself, once formed, is kinetically very stable (uncatalyzed hydrolysis half-life of years to decades) even though it is thermodynamically "downhill" to hydrolyze — a classic separation of thermodynamic favorability from kinetic accessibility, guarded by the substantial activation energy for uncatalyzed amide hydrolysis (~ 20 kcal/mol, roughly 84 kJ/mol).

Result. An $n$-residue chain has $n-1$ peptide bonds; each costs roughly +21 kJ/mol to form directly from free amino acids (hence the reverse, digestion by proteases, is favorable), and this bond count sets both the water stoichiometry and — combined with residue molar masses — the exact molecular weight of any specified sequence.

Derivation 2 — sequence space, redundancy, and why "one sequence, one protein" is compatible with astronomical combinatorics

Goal. Quantify how large the theoretical sequence space is for a modest protein, contrast it with the number of proteins that actually exist in a genome or across all known life, and show why this gap is not a paradox but a direct illustration of natural selection acting on a combinatorial substrate.

Step 1 — set up the counting problem. Each of the $n$ positions in a polypeptide chain can independently be filled with any of the 20 standard proteinogenic amino acids (ignoring, for this idealized count, post-translational modifications, D-amino acids, and non-standard residues). By the multiplication principle of counting, the total number of distinct sequences of length $n$ is

Step 2 — evaluate for a realistic protein length. Take $n = 100$, a small globular domain (e.g., roughly the size of lysozyme's globular core). Then

Since $2^{10} \approx 1024 \approx 10^{3.01}$, we have $2^{100} = (2^{10})^{10} \approx (10^{3.01})^{10} = 10^{30.1}$, so

This is roughly $10^{130}$ — far larger than the estimated number of atoms in the observable universe (≈ $10^{80}$), and vastly larger still than the total number of protein molecules that have ever existed on Earth.

Step 3 — compare to actual biological usage. The number of distinct protein-coding genes across all sequenced life is on the order of $10^8$–$10^9$ (counting every known family and organism, generously), and within any single genome the number of distinct proteins is typically $10^3$–$10^5$ (roughly 20,000 protein-coding genes in humans, expanded by splicing and modification to perhaps a few hundred thousand distinct protein products). Even if evolution had explored $10^{20}$ sequences over 3.5 billion years of history (a wildly generous upper estimate on total polypeptide-synthesis events, since a bacterial cell makes on the order of $10^6$ proteins per generation and Earth has hosted perhaps $10^{30}$–$10^{40}$ total microbial cell-generations, most of which reuse existing sequences rather than sampling new ones), that number divided by $\Omega(100) \approx 10^{130}$ is still a vanishingly small fraction:

Step 4 — resolve the apparent paradox. This tiny fraction is not evidence that functional sequences are found "by accident" from a uniform search of $\Omega(n)$; it is evidence that evolution never performs a blind search of sequence space at all. New proteins arise overwhelmingly by duplication and divergence of existing, already-functional sequences (point mutation, insertion/deletion, domain shuffling, gene fusion), so the accessible neighborhood of sequence space explored at each evolutionary step is minuscule and local, not a uniform draw from $20^n$. Functional protein sequences form isolated, structurally constrained "islands" within the vast sea of $\Omega(n)$, and evolutionary search moves between and within these islands via small steps, not by sampling the whole ocean.

Result. The combinatorial size of sequence space, $20^n$, is genuinely enormous even for short chains, which is exactly why a specific, reproducible sequence (Sanger's finding for insulin) is so remarkable: a cell must have an information-storage-and-readout mechanism (DNA → mRNA → ribosome) precise enough to select one sequence out of $10^{130}$ possibilities, every single time, for every molecule of a given protein it makes.

Example 1 (foundational) — computing exact molecular weight and peptide bond count for a pentapeptide. Consider the pentapeptide Ala-Gly-Ser-Cys-Val (read N→C; this is a fictitious but chemically ordinary sequence, chosen with unequal, non-repeating residues to avoid masking an error with symmetry). Use standard free-amino-acid molar masses (g/mol): Ala = 89.09, Gly = 75.07, Ser = 105.09, Cys = 121.16, Val = 117.15. Step 1: bond count. With $n = 5$ residues, the number of peptide bonds is $n - 1 = 4$, and 4 water molecules are released during synthesis. Step 2: sum of free amino acid masses.

Step 3: subtract water lost.

Step 4: sanity/limiting check. If instead we asked for the mass of the same five amino acids simply mixed, unreacted, in solution (i.e., the $n=0$-bonds limit, no condensation at all), the total mass of material is trivially conserved at $507.56$ g/mol distributed over five separate small molecules — no water is lost because no bonds form. The subtraction of exactly $(n-1)M_{\text{H}_2\text{O}}$ is what distinguishes "five separate amino acids" (mass 507.56 g/mol, five molecules) from "one pentapeptide" (mass 435.48 g/mol, one molecule): the difference, $72.08$ g/mol, is precisely the mass of the four waters released, confirming the stoichiometric bookkeeping.

Example 2 (applied) — using Edman degradation cycles to reconstruct an N-terminal sequence from experimental yield data. Edman degradation removes and identifies one N-terminal residue per cycle: phenylisothiocyanate (PITC) reacts with the free α-amino group of the N-terminal residue under mildly alkaline conditions, and the resulting phenylthiocarbamyl derivative is cleaved under acidic conditions to release a phenylthiohydantoin (PTH)–amino acid, shortening the peptide by exactly one residue while leaving the rest of the chain intact for the next cycle. Suppose a decapeptide is subjected to five cycles of Edman degradation, and the PTH-amino acids identified, in order, are: Leu (cycle 1), Thr (cycle 2), Gly (cycle 3), Asp (cycle 4), Phe (cycle 5). Suppose also each cycle proceeds with 95% efficiency (5% of chains fail to react and "fall behind," a real limitation of the method that eventually causes signal-to-noise problems in long reads). Step 1 — sequence read-off. The five cycles directly establish the N-terminal five residues, in strict N→C order:

Step 2 — quantify signal decay. If the starting molar amount of peptide is $P_0$, the fraction of chains still correctly reacting at cycle $k$ (assuming independent 95% success per cycle) is

At cycle 5: $f_5 = (0.95)^5 = 0.774$, meaning about 77.4% of original peptide molecules are still contributing clean signal to the 5th PTH-amino acid identification, while the rest have accumulated one or more "skipped" cycles and now contribute background noise (out-of-register signal) rather than useful sequence information. Step 3 — practical limiting case. As the number of cycles $k$ grows, $f_k = (0.95)^k \to 0$ exponentially; solving $(0.95)^k = 0.5$ gives

so after only about 13–14 cycles, half the original signal has been lost to incomplete reactions — this is exactly why classical Edman sequencing, applied directly to intact proteins, is practically limited to reading the first 20–40 residues reliably, and why full-length sequencing of large proteins requires cleaving the chain into smaller fragments (with trypsin, chymotrypsin, or cyanogen bromide) and sequencing each fragment separately before reassembling the full sequence from overlapping reads — exactly Sanger's overlap strategy for insulin.

Example 3 (exam/competition-level) — resolving an ambiguous sequence from overlapping proteolytic fragments, and computing the probability of two unrelated random sequences matching by chance over a stretch. (a) Sequence reconstruction from overlaps. A protein is cleaved separately by trypsin (cuts C-terminal to Lys/Arg) and by chymotrypsin (cuts C-terminal to bulky aromatics: Phe/Tyr/Trp), yielding these fragment sets (each is N→C within itself, but the fragments from the two digests are given in arbitrary order): Tryptic fragments: `T1 = Ser-Phe-Tyr-Lys`, `T2 = Ala-Trp-Arg`, `T3 = Gly-Leu` Chymotryptic fragments: `C1 = Ser-Phe`, `C2 = Tyr-Lys-Ala-Trp`, `C3 = Arg-Gly-Leu` Step 1 — anchor using overlaps. `C1 (Ser-Phe)` overlaps the start of `T1 (Ser-Phe-Tyr-Lys)`, confirming Ser-Phe-Tyr-Lys as a contiguous run and telling us that this run is followed, not preceded, by material found in the next chymotryptic fragment, since chymotrypsin does not cut after Lys. Step 2 — extend. `C2 (Tyr-Lys-Ala-Trp)` overlaps `T1`'s C-terminal Tyr-Lys and continues into `T2 (Ala-Trp-Arg)`'s N-terminal Ala-Trp, establishing the contiguous run Ser-Phe-Tyr-Lys-Ala-Trp. Step 3 — close the sequence. `C3 (Arg-Gly-Leu)` overlaps `T2`'s C-terminal Arg and continues into `T3 (Gly-Leu)`, giving the final contiguous run Ala-Trp-Arg-Gly-Leu, and stitching everything together:

a 9-residue sequence consistent with every fragment from both digests (check: trypsin correctly cuts after Lys4 and after Arg7, chymotrypsin correctly cuts after Phe2 and after Trp6 — every cut site matches the specificity rule). (b) Chance-matching probability. Given this reconstructed 9-residue sequence, what is the probability that an unrelated, randomly generated 9-residue sequence (uniform over the 20 amino acids, independent positions) matches it exactly, purely by chance? Each position matches independently with probability $1/20$, so

Since $20^9 = 20^{10}/20 = (1.024\times10^{13})/20 \approx 5.12\times10^{11}$ (using $20^{10} = (2\times10)^{10}=2^{10}\times10^{10}\approx1.024\times10^3\times10^{10}=1.024\times10^{13}$), we get

As a limiting check: for a match of only 1 residue ($n=1$) the probability is trivially $1/20 = 0.05$, matching the intuitive single-position case, and the probability falls off geometrically — multiplying by $1/20$ for every additional residue of required exact overlap — which is precisely why database sequence-alignment tools (BLAST and relatives) treat short exact matches as statistically unremarkable but treat long, exact or near-exact multi-residue matches as very strong evidence of true homology rather than coincidence.

Frederick Sanger (1918–2013) is the only person to win two Nobel Prizes in Chemistry outright for distinct achievements, the first (1958) for determining the complete amino acid sequence of insulin, the second (1980, shared) for developing dideoxy DNA sequencing. Trained at Cambridge, Sanger approached insulin — chosen partly because it was one of the few proteins available in reasonably pure, abundant form (from bovine pancreas, a slaughterhouse byproduct) — with patient, unglamorous chemistry: partial acid hydrolysis to generate overlapping fragments, paper chromatography and electrophoresis ("fingerprinting") to separate them, and his own reagent, 1-fluoro-2,4-dinitrobenzene (Sanger's reagent, DNFB), which reacts irreversibly with free α-amino groups to tag the N-terminal residue of a peptide before hydrolysis, allowing its identity to be read out even after the rest of the chain is destroyed. By 1951 Sanger had shown insulin consists of two distinct chains (A, 21 residues, and B, 30 residues) linked by disulfide bonds, and by 1955 he had the complete sequence of both chains — the first complete primary structure of any protein ever determined. The achievement settled, empirically and beyond doubt, the question of whether proteins have exact, unique, genetically specified sequences (as opposed to being statistical, disordered polymers, a serious competing view at the time). Sanger's insulin work directly motivated Crick's sequence hypothesis (that the sequence of bases in DNA specifies the sequence of amino acids in protein) and the subsequent cracking of the genetic code by Nirenberg, Khorana, and others in the 1960s.