FTIR and Raman of Peptide Secondary Structure: the Amide I Band

The peptide backbone carbonyl is an unpromising spectroscopic target in every respect but the one that counts: there is one per peptide bond, they are chemically identical, and the frequency at which each stretches moves measurably when the hydrogen bond it accepts changes. That is the amide I band. It spans roughly 1600–1700 cm^-1 and is the signature region for C=O stretching with an N–H bending contribution, while the amide II band at 1510–1580 cm^-1 is the signature region for N–H bending coupled to C–N stretching ^[1]. Of the nine characteristic infrared absorptions generated by the polypeptide structural repeat — amides A, B and I through VII — amide I is the most prominent and the most sensitive vibrational band of the backbone, and the one that relates to secondary-structural components ^[2].

A single functional group reports on fold rather than on chemistry because the oscillators are coupled. Carbonyls held in a regular geometric array interact, and it is those vibrational couplings that are the source of the identifying features in the infrared spectrum; the same couplings are what isotope-labelling strategies exploit, whether by site-specific labelling to obtain residue-by-residue information or by isotope dilution of uniformly labelled material to identify structural folds ^[3]. Helix, sheet and coil differ in how the oscillators sit relative to one another, so they differ in where the coupled band lands. This is why amide I is a structural observable and not merely a carbonyl count.

Where the components sit

Second-derivative amide I spectra of twelve globular proteins in aqueous solution, compared against available crystal structures, established that band frequency assignments to alpha-helix, beta-sheet, unordered and turn structures are highly consistent among all proteins and agree closely with predictions from theory ^[4]. The same work fixed an asymmetry you have to design around: alpha-helix and unordered structures can each be assigned to only one band, whereas multiple bands are associated with beta-sheets and turns ^[4]. Band intensities from that treatment provide a useful method for estimating the relative amounts of different structures ^[4].

Concrete positions measured on peptide material: in solid amyloid beta (1–40), the amide I maximum sat at 1655 cm^-1 (assigned alpha-helix) and moved to 1647–1643 cm^-1 (random coil) as applied mechanical pressure increased, with a new peak at 1634 cm^-1 assigned to beta-antiparallel sheet; in the same samples the 1540 cm^-1 amide II feature shifted to 1527–1529 cm^{-1 [5]}. In the homopolypeptide poly-L-lysine, spray-drying from alpha-helix, beta-sheet or random coil feeds transformed all three initial conformations to antiparallel beta-sheet with bands at 1623 and 1690 cm^-1; the beta-sheet liquid feed itself showed a band at 1616 cm^-1, indicating a denatured, extended chain structure that also converted on drying ^[6].

That pairing is the conventional beta-sheet signature: a low-frequency component with a partner band near 1690 cm^-1. Note also that 1616 and 1623 cm^-1 were not read as the same state in that experiment. Wavenumber differences of a few units inside the low-frequency region carry structural meaning, which is exactly why the resolution treatments discussed below are not optional cosmetics.

The low-frequency component: aggregation state and strand sense

The frequency of the low-frequency component separates native beta-sheet from amyloid. Amide I′ spectra recorded in D₂O across a compiled set of amyloid fibril samples and native beta-sheet proteins split by frequency: the fibril range extends from 1611 to 1630 cm^-1, whereas native beta-sheet proteins produce amide I′ peaks clustering between 1630 and 1643 cm^{-1 [7]}. The authors relate that difference to the distribution of phi/psi dihedral angles within the Ramachandran plot, the average number of strands per sheet and possibly the beta-sheet twist, and conclude that fibril formation from a native beta-sheet protein can involve a substantial structural reorganisation ^[7]. A component below about 1630 cm^-1 is therefore an aggregation flag, not simply a beta flag.

Strand sense is a separate question, and the band can address it. Under conditions yielding well-defined amyloid beta (1–42) oligomers or fibrils, ATR-FTIR showed fibrillar material organised in a parallel beta-sheet conformation, while the oligomeric preparation displayed distinct spectral features attributed to an antiparallel beta-sheet structure; those oligomer spectra also showed striking similarities to spectra of bacterial outer membrane porins ^[8]. The practical consequence is that the two aggregation intermediates are distinguishable in a single fast measurement: ATR-FTIR gives an almost continuous structural view of protein and peptide conversion during aggregation, and infrared can differentiate oligomers from fibrils simply on their spectral features ^[9].

The high-frequency component deserves more suspicion than it usually receives. In self-assembling aromatic peptide amphiphiles, the band around 1680–1695 cm^-1 that had previously been assigned to an antiparallel beta-sheet conformation was shown, using Fmoc-dialanine and the analogous Fmc-dialanine as examples, to be in fact absorption of the stacked carbamate group in Fmoc-peptides, with ¹³C-labelled samples and density functional frequency calculations supporting the reassignment ^[10]. If your peptide carries a carbamate cap, an Fmoc group left in place, or any stacking aromatic protecting chemistry, the 1690 cm^-1 feature cannot be read as an antiparallel marker without an independent check.

Resolving the envelope, and what the maths costs

Second derivative and Fourier self-deconvolution are the two commonly used techniques to resolve the overlapped component peaks from the often featureless amide I band in the FTIR curve-fitting approach ^[11]. Their reliability is greatly affected by the omnipresent water vapour in the atmosphere ^[11]. Worse, the criteria in general use as quality controls cannot guarantee a reliable evaluation of water vapour interference: a protein absorption spectrum can still be significantly affected even though it satisfies the established criteria, a behaviour described as sample absorbance-dependent water vapour interference ^[11]. Comparing the second-derivative spectra of the protein and of liquid water has been proposed as a better criterion before either treatment is applied ^[11]. This is the noise-amplification problem in its practical form — both operations sharpen narrow features, and atmospheric rotational lines are narrow features.

Overlap is intrinsic, not an instrumental failing. One identified obstacle to better predictions is the strong overlap of bands assigned to different secondary structures; disordered and alpha-helical structures absorb almost at the same wavenumber ^[12]. That is the honest reason a fitted alpha-helix percentage is softer than a fitted beta-sheet percentage: across 85 proteins measured through a hydrogen/deuterium exchange series, beta-sheet content was in absolute terms always better predicted than alpha-helix content ^[12]. Treat second-derivative minima as band positions and treat curve-fit areas as a ranking, not as a structure determination.

H₂O, D₂O and the amide II handle

Water is the first obstacle in the amide I region. The second-derivative method in aqueous solution was only made workable once criteria were developed for evaluating how successfully the spectra due to liquid and gaseous water are subtracted from the observed spectrum in that region ^[4]. Published protocols estimate secondary-structural components in aqueous solution in both H₂O and D₂O, using second-derivative, deconvolution and curve-fitting algorithms, and call for high protein concentrations above 3 mg ml^-1, with the whole procedure typically completed in one to two days ^[2].

Deuteration is not just a way of clearing the solvent. Because the amide II region is dominated by N–H bending coupled to C–N stretching ^[1], it is directly sensitive to backbone NH → ND substitution, and the size of the effect maps accessibility rather than conformation. In bacteriorhodopsin, exposure to D₂O even after long periods caused only a partial shift of the amide II band, attributed to exchange of peripheral peptide structure alone, while the band downshifted completely when the protein was reconstituted and regenerated in the presence of D₂O — evidence that almost the entire core backbone had then exchanged ^[13]. Residual amide II intensity is thus a read-out of protected backbone amide hydrogens.

Deuteration also moves amide I components differentially, in part because exchange is much faster for disordered structures, and that differential is exploitable. Recording FTIR spectra of 85 proteins at different stages of hydrogen/deuterium exchange, prediction error fell for alpha-helix content and for the grouped category of random, turns and bends after 15 minutes of exchange, while further deuteration degraded the prediction; the beta-sheet fraction, by contrast, was better predicted in non-deuterated conditions ^[12]. In other words the choice of H₂O or D₂O is not a housekeeping decision. Pick it for the structural class you are trying to quantify, and do not splice a D₂O helix figure onto an H₂O sheet figure from the same sample.

ATR sampling: films, hydrated layers and the drying artefact

Attenuated total reflection removes the pathlength problem that makes transmission work in water so unforgiving. ATR-FTIR requires only 1–100 µg of material and spectra are recorded in a matter of minutes; the environment can be modulated so that conformation is studied as a function of temperature, pressure and pH, and in the presence of specific ligands ^[14]. Replacement of amide hydrogen by deuterium is extremely sensitive to environmental changes, and the kinetics of exchange can be followed on the same accessory; with polarised infrared, the orientation of parts of the molecule can additionally be estimated ^[14]. For aggregating peptides this combination is what buys the near-continuous structural view of the conversion process ^[9].

The sampling geometry carries its own hazard, and the poly-L-lysine result is the warning. All three starting conformations converted to antiparallel beta-sheet on spray-drying, and the authors argued the shift cannot be a simple result of forming the conformation with the strongest hydrogen bonds in the dried state, suggesting instead that steric effects arising from close approach of the molecules during drying make antiparallel beta-sheet energetically favourable in the solid state ^[6]. A dried ATR film reports the structure of a dried film. If the question is about solution conformation, keep the layer hydrated and say so in the method.

Raman as the complementary experiment

Raman reaches the same amide I region by scattering rather than absorption, and the analysis transfers: the same second-derivative peak finding followed by curve fitting has been applied to the amide I band of both infrared and FT-Raman spectra of globular proteins in aqueous solution and of hydrated zein and gluten ^[1]. The marker positions are well characterised. In alpha-synuclein, a prototypical natively unfolded protein, the predominant Raman amide I marker band was at 1654 cm^-1 in sodium dodecyl sulfate, alongside the typical double minimum in the circular dichroism spectrum; in 25% hexafluoro-2-propanol the band appeared at 1653 cm^-1 with a peak width at half-height of approximately 33 cm^-1, and in 25% methanol, where the structure is predominantly beta-sheet, it shifted to 1667 cm^-1 with a peak width at half-height of approximately 26 cm^{-1 [15]}. A three-component fit of the amide I band was used to describe the ensemble, with a higher-frequency feature resembling the Raman amide I band of ionised polyglutamate and polylysine, peptides that adopt a polyproline II helical conformation ^[15].

Raman also carries side-chain markers that the amide bands cannot supply. The disulfide stretching region is conformationally resolved: in bovine serum albumin in solution, two Raman bands at 520 and 505 cm^-1 indicated that the 17 disulfide bridges take both gauche-gauche-gauche and gauche-gauche-trans conformations, and the intensity ratio I₅₀₅/I₅₂₀ rose from 1.6 to 2.1 across the N-to-F transition and from 2.1 to 6.3 across the F-to-E transition as gauche-gauche-trans converted to the energetically more stable gauche-gauche-gauche form ^[16]. For a cystine-containing peptide that is a direct handle on whether a disulfide has been formed in one geometry or another, which amide I cannot supply.

The aromatic markers need their scope stated precisely or they will be over-read. The tyrosine doublet at 850 and 830 cm^-1 arises from Fermi resonance between the ring-breathing vibration and the overtone of an out-of-plane ring-bending vibration of para-substituted benzenes ^[17]. Its intensity ratio depends on changes in the relative frequencies of those two vibrations, which are sensitive to the nature of the hydrogen bonding of the phenolic hydroxyl group and to its ionisation, but much less so to the environment of the phenyl ring and to the conformation of the amino acid backbone ^[17]. The doublet reports on the tyrosine hydroxyl, not on the fold.

Positioning against circular dichroism

The two methods are not redundant, and the strongest argument for running both is a negative one. Using circular dichroism and infrared spectra of a purpose-built 50-protein database and multivariate statistical analysis, only the comparison of analyses run on circular dichroism and infrared spectra independently was able to identify failed solutions in the absence of a known structure ^[18]. The same study found that no smaller subset of a well-selected database contains the information needed to describe the entire set, so large databases observing stringent selection criteria are necessary for predicting unknowns; and, unexpectedly, that the amide II band has high information content and could be used alone for secondary structure prediction in place of amide I ^[18].

Practically, that sets the division of labour. Circular dichroism and vibrational spectroscopy agreeing on a peptide is a real result; disagreement is the error signal the infrared analysis alone would have hidden. The vibrational arm runs on microgram quantities in minutes and lets you modulate temperature, pressure, pH and ligand presence during the measurement ^[14], and it reads aggregating mixtures in which oligomers have to be told from fibrils ^[9]. It is also where the specific structural discriminations sit: native beta-sheet against amyloid ^[7], parallel strands against antiparallel ^[8], and, on the Raman side, one disulfide rotamer against another ^[16].

None of it licenses a residue-level model. Amide I gives you populations of secondary-structure classes, with alpha-helix and disorder overlapping badly ^[12], with the second-derivative and self-deconvolution steps that precede any fit vulnerable to atmospheric water vapour even when the usual quality-control criteria are met ^[11], and with at least one widely used marker band that means something else entirely on a protected peptide ^[10]. Report band positions, report the solvent and the sampling mode, state whether the film was dried, and let the second orthogonal method decide whether the fit was real.