Circular dichroism and peptide higher-order structure: what a CD spectrum can and cannot settle

A certificate of analysis for a synthetic peptide converges on three numbers: an identity confirmation by mass spectrometry, a purity figure by reversed-phase HPLC, and — on the better documents — a peptide content value from amino acid analysis. All three describe covalent structure. None describes shape. For a twelve-residue linear peptide dissolved in water that is a defensible omission, because it very likely has no persistent shape to describe. For a disulfide-constrained peptide, a lipidated incretin analogue, or any material positioned as comparable to a recombinant product, the omission is real, and circular dichroism is the technique most often proposed to fill it.

CD is cheap, fast, non-destructive, and among routine biophysical methods one of the easiest to over-interpret — because it produces a smooth, confident-looking curve from samples that may be badly characterized, and because the numbers derived from that curve inherit every error in the concentration value used to scale them. This post covers what the far-UV signal physically is, what the reported units depend on, and where the technique genuinely settles a question versus where it only appears to.

What the far-UV signal is

Circular dichroism measures the difference in absorbance between left- and right-circularly polarized light as a function of wavelength. A signal requires a chromophore in a chiral environment. In the far-UV region, roughly 190 to 250 nm, that chromophore is the peptide bond itself. Two electronic transitions dominate: a weak n→π* transition centered near 220 nm and a much stronger π→π* transition near 190 nm, which splits into components when amide groups are held in a regular repeating geometry.

The consequence is the useful part: the far-UV spectrum is determined almost entirely by the backbone dihedral angles φ and ψ, and almost not at all by which side chains are attached. A right-handed α-helix produces a characteristic double minimum near 208 and 222 nm with a strong positive band around 192 nm. A β-sheet gives a single broader minimum near 216 to 218 nm and a positive band near 195 nm. A disordered chain gives a strong negative band near 198 nm and very little intensity above 210 nm.

Two structural facts follow immediately, and both constrain what the method can be asked to do. First, because the signal comes from the backbone rather than the side chains, CD is nearly blind to sequence. Two peptides differing by a single conservative substitution, or by a deamidation, or by an aspartate-to-isoaspartate rearrangement, will in most cases produce spectra that overlay within noise. CD does not confirm identity and should never be represented as doing so. Second, the spectrum is a population-weighted average over every conformer present. A sample that is uniformly 50 percent helical and a sample that is an even mixture of fully helical and fully disordered chains give approximately the same curve. The technique cannot distinguish partial order from conformational heterogeneity, which is exactly the distinction that matters when the question is whether a fraction of the material is misfolded.

Units, calibration, and the concentration term

Raw instrument output is ellipticity in millidegrees. Comparison across laboratories requires normalization to mean residue ellipticity, in units of degrees square centimeter per decimole, which divides the observed signal by path length, by molar concentration, and by the number of residues. Every one of those divisors is a place for error to enter, and the concentration term is the dominant one.

Mean residue ellipticity scales inversely with concentration, so the error propagates linearly: a peptide stock that is ten percent less concentrated than assumed yields helix fractions inflated by roughly ten percent, with no visible signature in the spectrum itself. This is where the mass on the vial label becomes an unsuitable denominator. Lyophilized synthetic peptide is not pure peptide by weight — it carries counterion, residual moisture, and any bulking agent used during lyophilization, and the gap between gross weight and peptide content routinely runs from a few percent to well over twenty. Trifluoroacetate salts sit at the high end of that range. The correct denominator is a measured peptide content value, typically from amino acid analysis or quantitative nitrogen determination, and a CD result reported without stating how concentration was established is not a comparable number.

Instrument calibration is the other half. The conventional standard is camphor-10-sulfonic acid, usually in the non-hygroscopic ammonium salt form, which produces bands of opposite sign at 290.5 and 192.5 nm. Reported differential extinction values are approximately 2.36 and −4.72 square centimeter per millimole at those two wavelengths, corresponding to molar ellipticities near 7,800 and −15,600, and the ratio of about −2.00 between them serves as a two-point check on both magnitude and far-UV performance. More recent determinations for the (1S)-(+)-ammonium salt give 2.39 ± 0.04 and −4.92 ± 0.06, within European Pharmacopoeia acceptance criteria. A spectrum from an instrument with no documented recent calibration carries an unknown scale factor.

The buffer ceiling

The far-UV region is the part of the spectrum where nearly everything absorbs. Chloride absorbs strongly below 200 nm, which puts saline and most chloride-buffered systems in direct competition with the analyte for the available photons. Residual trifluoroacetate from preparative purification absorbs in the same window. Carbonate, imidazole, high concentrations of Tris, dithiothreitol, and any aromatic excipient all contribute. The practical result is not a distorted spectrum but a truncated one: as total absorbance rises, the photomultiplier dynode voltage climbs to compensate, the signal-to-noise ratio collapses, and the data below some cutoff wavelength become noise wearing the shape of a curve.

The dynode voltage trace is therefore not optional metadata. It is the primary validity check on the measurement, and a spectrum published without it cannot be evaluated. As a working rule, data collected where total absorbance exceeds roughly 1.5 to 2 should be treated as unreliable regardless of how smooth the plotted line looks. The usual remedies are dilution into a low-absorbance buffer such as dilute phosphate or water, and shortening the path length — 1 mm cells are standard, with 0.1 mm or demountable cells extending the accessible range when concentration cannot be reduced. Counterion exchange away from trifluoroacetate, where the material permits it, removes one of the larger contributors outright.

Short peptides and the co-solvent trap

Most compounds in a research peptide catalogue are shorter than thirty residues, and short linear peptides in aqueous buffer are predominantly disordered. Their far-UV spectra converge on the same random-coil signature more or less regardless of sequence, which means that for a large fraction of catalogue material CD returns no discriminating information at all. This is not a failure of the measurement; it is an accurate report that there is no persistent secondary structure to measure.

The common response is to add a structure-inducing co-solvent, most often trifluoroethanol, and titrate until helical bands appear. The resulting spectra are real and reproducible, and they are frequently presented as evidence that the peptide is helical. They are not. A trifluoroethanol titration measures helical propensity — the capacity of a sequence to adopt a helix when the solvent environment is engineered to stabilize one — and essentially any sequence with modest helical preference will comply at high enough co-solvent fraction. The same caution applies to spectra collected in micelles or detergent above the critical micelle concentration. Such experiments answer a question about conformational tendency; reported without the aqueous control and the co-solvent fraction, that answer becomes a claim about the native state which the data do not support.

Where CD actually settles something

Stated negatively at this length, the technique can sound useless. It is not; it is narrow. Its strongest application is comparative rather than absolute. Overlaying the spectrum of a test lot against a well-characterized reference lot, under identical buffer, concentration method, temperature, and path length, is a sensitive fingerprint comparison — sensitive enough to flag gross conformational divergence, disulfide scrambling in constrained peptides, or a change in aggregation state between production campaigns. No structural model needs to be extracted for the comparison to be informative, and the concentration error that undermines absolute helix fractions largely cancels when both samples are prepared the same way.

Thermal ramps are the second genuine use. Monitoring ellipticity at a single wavelength, commonly 222 nm, while raising temperature produces an unfolding transition and an apparent midpoint that functions as a stability-indicating parameter — useful for comparing formulations, buffer systems, or excipient effects even when the transition is not thermodynamically reversible and the midpoint therefore has no rigorous physical meaning.

Where a structural model is genuinely wanted, deconvolution algorithms are well documented: SELCON3, CONTINLL, CDSSTR and VARSLEC remain available through the DichroWeb server, with BeStSel developed specifically to address the spectral variability of β-structures. Reference spectra and metadata are curated in the Protein Circular Dichroism Data Bank. These tools were benchmarked on globular proteins, and their output for short synthetic peptides should be treated as indicative rather than quantitative.

The regulatory position

Higher-order structure is an established element of physicochemical characterization for biotechnology products under ICH Q6B, where CD appears alongside NMR and related methods. For synthetic peptides the picture has been less settled. FDA’s May 2021 guidance addressing abbreviated applications for certain highly purified synthetic peptides referencing recombinant listed drugs — covering glucagon, liraglutide, nesiritide, teriparatide and teduglutide — set out expectations that included comparative higher-order structural characterization. That guidance has since been withdrawn on the stated basis that it no longer reflects the agency’s current scientific thinking, alongside a substantial revision of product-specific guidances for injectable peptide products, with a replacement framework signalled for later in 2026.

None of this framework governs research-grade material, which is supplied for laboratory use and is not the subject of an application. It is nonetheless the vocabulary in which higher-order structure claims are argued, and a supplier invoking CD data is implicitly borrowing that vocabulary. The reasonable expectation is that the borrowing be complete: buffer composition, concentration and the method used to determine it, path length, temperature, calibration date, and the dynode voltage trace.

A circular dichroism spectrum is a low-resolution report on the average backbone geometry of whatever is in the cell. It cannot confirm sequence, cannot separate partial order from a mixed population, and cannot be scaled correctly without a concentration value that most peptide characterization packages do not independently establish. Within those limits it does something no other routine method does cheaply — it detects that the shape of this lot differs from the shape of the last one. That is a narrower claim than the smooth curve suggests, and it is the claim worth making.