On the purported "backbone fluorescence" in protein three-dimensional fluorescence spectra

In this study, several proteins (albumin, lysozyme, insulin) and model compounds (Trp, Tyr, homopolypeptides) were used to demonstrate the origin of the ﬂ uorescence observed upon their excitation at 220 – 230 nm. In the last 10 years we have observed a worrying increase in the number of articles claiming that this ﬂ uorescence originates from the protein backbone, contrary to the established knowledge that UV protein emission is due to aromatic amino acids only. Overall, our data clearly demonstrate that the observed emission upon excitation at 220 – 230 nm is due to the excitation of Tyr and/or Trp, with subsequent emission from the lowest excited state ( i.e. the same as obtained with 280 nm excitation) in agreement with Kasha's rule. Therefore, this ﬂ uorescence peak does not provide any information on backbone conformation, but simply reports on the local environment around the aromatic side chains, just as any traditional protein emission spectrum. The many papers in reputable journals erroneously reporting this peak assignment, contradicting 5 decades of prior knowledge, have led to the creation of a new dogma, where many authors and reviewers now take the purported backbone ﬂ uorescence as an established fact. We hope the current paper helps counter this new situation and leads to a reassessment of those papers that make this erroneous claim.


Introduction
Fluorescence spectroscopy is a commonly used and powerful method to study protein conformational changes, by taking advantage of the emission properties of aromatic residues, in particular tryptophan (Trp). 1,2 Among many different uorescence approaches, three-dimensional (3D) uorescence spectra, also called excitation-emission matrices or total uorescence, represent a powerful tool to analyse complex samples. In this technique, emission intensity is measured for all possible combinations of excitation and emission wavelengths, thus summarizing in a single 3D plot the information corresponding to many different excitation and emission spectra. [3][4][5][6][7] This approach is very useful in analysing samples containing multiple uorophores with overlapping spectra, such as wastewaters. 8 In a large number of papers, primarily published in the last 10 years, 3D uorescence spectra have also been used to characterize ligand-protein interactions, and specically the purported conformational changes induced by the association. Among other components, 3D uorescence spectra of proteins usually exhibit a peak with an excitation maximum at 220-230 nm, and an emission maximum at 300-350 nm, the exact peak position depending on the specic protein and the uorimeter used (see Fig. 1 for selected examples of human serum albumin (HSA), lysozyme and human insulin). Surprisingly, in a large number of recent papers this peak has been alleged to originate from the protein backbone and its properties (intensity and position) to be sensitive to conformational changes of that backbone. For example, an analysis of 176 articles published in 2012 and dealing with uorescence studies of ligand binding to albumin revealed that 24 of those, published in 14 different journals, assigned a peak in 3D uorescence spectra to backbone emission. 9 Also in 2016 many examples of papers making the same claim can be found in the literature. [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26] These latter references are solely meant as examples of the widespread nature of the backbone uorescence claim and are not an allinclusive list. More importantly, we would like to point out that 3D uorescence spectra are simply one of the many experiments reported in those articles. Here we wish to comment only on the use of this particular technique. It should also be noted that some other reports describe self-assembly induced emission in the visible region by peptides in brillar or other aggregated states. The still-debated origin of this phenomenon has been recently attributed to proton transfer through intermolecular hydrogen bonds, which could take place only in very specic structures. 27 However, here we will limit our discussion to the claim that all protein backbones emit in the UV region.
This claim is actually extremely surprising, as it contradicts half a century of prior knowledge. Already in 1952, Debye and Edwards demonstrated that protein phosphorescence is due to the aromatic residues only. 28 In 1953 Gregorio Weber postulated, on the basis of general considerations of the properties of aromatic amino acids and the positions of their absorption maxima, that protein uorescence originates from aromatic amino acids, too. 29 This hypothesis was shortly aerwards conrmed by ground-breaking experimental studies. [30][31][32] In 1967, 32 Konev summarized the pioneering years of protein uorescence studies as follows: "The exclusively aromatic nature of protein uorescence was conclusively shown by experiments in which no uorescence could be detected in proteins which contain no aromatic amino acids. It can now be regarded as established that the formation of uorescence and phosphorescence spectra of proteins involves only three aromatic amino acids which are capable of luminescence in the free state. These are tryptophan, tyrosine, and phenylalanine".
Also several more recent papers discuss uorescence obtained by far-UV excitation (220-230 nm range) in the context of excitation and emission by the Tyr and Trp residues. [33][34][35] Thus, in a historical perspective the observation of uorescence from the protein backbone is a potentially exciting new nding. Unfortunately, an examination of the references used to substantiate this claim reveals no study that directly investigated this phenomenon. If a reference is provided at all, it is to papers that make the same claim, or to papers reporting that the protein backbone absorbs UV-radiation in a broad range between 200 and 250 nm. However, also the aromatic amino acids absorb in this range. We tried to nd the original article where backbone emission was rst proposed, and identied a work by Zhang et al. 36 as a likely rst. Here, the authors speculated that, since the backbone absorbs in the far UV region, then the 3D peak "may mainly exhibit the uorescence characteristic of polypeptide backbone structures". The only data reported to support this hypothesis was the decrease in peak intensity with increasing protein concentration. This nding, which is in reality likely arising from inner-lter effects, 37 was interpreted as the result of the interaction between peptide chains. One of the authors on the current paper was also misled by the aforementioned conclusion in peak assignment. 38 It is rather remarkable that the large number of studies using 3D uorescence spectra to study structural changes of proteins upon ligand binding are based on such insubstantial evidence, against a large historical body of studies stating the opposite.
There are a number of further reasons that cast doubt on the possibility of backbone emission: (1) Very few molecules are measurably uorescent, due to the fact that non-radiative decay processes usually predominate over the emission of a photon in the relaxation of excited molecules.
Although it is still difficult to predict theoretically which molecules will exhibit uorescence, 39 uorescent compounds are aromatic, or (less oen) highly unsaturated aliphatic molecules. Delocalized electrons, formally present in conjugated double bonds, are required for providing the rigidity necessary to inhibit non-radiative relaxation processes. [40][41][42] No extended conjugated systems are present in the peptide backbone and the low wavelength region of peptide absorption at around 220 nm is due to an n / p* transition, 43 which is forbidden in the electric dipole approximation. Both these points argue against the backbone showing any uorescence. [40][41][42] (2) In their seminal 1957 article on the origin of protein uorescence, Teale and Weber 30 indicated that an important criterion to verify "whether the uorescence shown by a solution is due to a given substance present in it" is a good match between the uorescence excitation spectrum and the absorption of the putative uorophore. By contrast, the protein backbone has an absorption maximum around 190 nm, corresponding to a p / p* transition of the amide band, 44 and therefore does not match with the 220-230 nm excitation peak observed in 3D uorescence spectra.
(3) Fluorescence is always observed at wavelengths longer than those of the lowest energy absorption band. This phenomenon, called Stokes shi, 45 is largely due to differential solvation of the excited and ground states (in addition to vibrational relaxation phenomena). Therefore, the solventinduced shis of emission bands can be used to calculate dipole moments of electronically excited molecules. 46 An absorption-emission shi from 220 to 340 nm would correspond to a Stokes shi of approximately 2 eV in energy. This shi would be much larger than the record shis reported for charge transfer compounds such as 6-propionyl-2-(N,N-dimethylamino)naphthalene (PRODAN) 47 or 4-dimethylamino-4 0nitrostilbene (DANS), 48 which undergo an exceptional dipole increase following excitation.
For all the reasons explained above, it should be clear that the peak 220-230/300-350 nm peak observed in 3D uorescence spectra of proteins cannot be due to backbone emission. To make this point clear from an experimental point of view, we collected 3D uorescence spectra for different proteins and model compounds. These data will allow us to fully clarify all the features commonly observed in 3D uorescence spectra of proteins. Samples were analysed independently in different laboratories around the world (Denmark, Italy, and Malaysia). Experiments were performed on different uorimeters and using different settings, to show the robustness of the observations and of the conclusions drawn, and to illustrate the differences that can be observed in 3D uorescence spectra due to different experimental settings and instrumentation.

Materials and methods
In this section, differences between the three groups are indicated by the two letter code of the relevant country (DK, IT, MY).

Sample preparation
All samples (IT, DK) were dissolved in sodium phosphate buffer (10 mM, pH 7.4), except the insulin solution that was prepared by rst dissolving 5 mg in 80 ml of 0.1 M HCl, then neutralized using 80 ml of 0.1 M NaOH and then diluted using ca. 5 ml of phosphate buffer. This sample was then ltered through a 0.22 mm lter. All samples (MY) were prepared in Milli-Q water except for Tyr and Trp, which were dissolved in 1 M HCl, diluted with Milli-Q water and neutralized with NaOH.
The concentrations of the samples containing aromatic moieties were determined spectrophotometrically using molar extinction coefficients of 5630 cm À1 M À1 at 280 nm for Trp, 49 $35 200 cm À1 M À1 at 280 for HSA, 49 38 940 cm À1 M À1 at 280 nm for lysozyme, 50 6200 cm À1 M À1 at 276 nm for insulin, 51 and 1295 cm À1 M À1 at 278 nm for Tyr. 49 Poly-and oligo-peptide concentrations were determined by weighing the dissolved powder in the case of octa-L-arginine (DK), by absorbance at 214 nm for Poly-K (IT), using an extinction coefficient of 923 cm À1 M À1 for the monomer, 52 or by the bicinchoninic acid method 53 for Poly-A and Poly-K solutions (MY), using bovine serum albumin as a standard, whose concentration was determined spectrophotometrically, with an extinction coefficient of $43 900 cm À1 M À1 at 280 nm. 49 Concentrations used in the 3D spectra were: HSA 0.9 mM (IT).

Fluorescence spectroscopy
Fluorescence spectra were collected with a Cary Eclipse uorescence spectrometer (Agilent Technologies) (IT and DK), or a FP-6500 spectrouorimeter (Jasco) (MY). Both these instruments have also been used by several of the authors claiming the presence of backbone uorescence. Experiments were performed in 1 cm path length cells. 3D uorescence spectra were executed under the following conditions: excitation 220-350 nm; emission 220-450 nm; data interval 2 nm (IT, DK) or 5 nm in excitation and 1 nm in emission (MY); bandwidth 2.5 nm (IT), 5 nm (DK) or 10 nm (MY); scanning rate 600 nm min À1 (IT), 1200 nm min À1 (DK) or 500 nm min À1 (MY). In the Cary the detector voltage was set at 800 V (IT) or 600 V (DK), while it was xed at 250 V (MY) in FP-6500. Blank subtraction or spectral corrections were not performed. Fluorescence emission or excitation spectra were executed under the same conditions of the 3D spectra (DK) or with the following settings (IT): data interval 1 nm, excitation bandwidth 5 nm, emission bandwidth 10 nm, scan rate 30 nm min À1 , detector voltage 600 V. Excitation spectra were corrected for inner lter effects. 54 Results and discussion 3D uorescence spectra of proteins The 3D uorescence spectra of HSA collected in the three laboratories participating in the present study show that the relative intensities of the different peaks and even their exact position can vary due to inner lter effect and instrumental factors, such as the wavelength response of monochromators and detectors. Corrections for these factors are in principle possible, 1,55 but were not performed in the current study, as it is customary to report uncorrected, so-called "technical" spectra, particularly for 3D uorescence. 1 Therefore, the multiple datasets reported here serve to illustrate possible variations in the spectral shape, due to instrumental factors. Indeed, a variability in exact peak positions and relative intensities similar to those observed in our three labs can also be noticed among the papers claiming backbone uorescence. [10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26] Some of the signals in the 3D uorescence spectra are not due to uorescence, but to light scattering, and can therefore be observed also for a sample that does not contain any uorophore, such as buffer or pure water. The line labelled as 'a', corresponding to the matrix diagonal, i.e. the region with l em. ¼ l exc. , is due to elastic scattering of the excitation light by the sample solvent, the dissolved molecules, and suspended particulates. A second line, labeled as 'b', is observed at l em. ¼ 2l exc. , and is due to an instrumental artefact caused by transmission of second-order diffraction by grating monochromators. That is, when set to select a given wavelength, l, a grating monochromator also partially transmits light with wavelength l/2. 1,56 For these reasons, peak 'b' actually corresponds to light of wavelength l em. ¼ l exc. that can cross the emission monochromator, when this is set at l em. ¼ 2l exc. .
Another much weaker line is sometimes visible (depending on the uorescence intensity scale), and is due to inelastic (Raman) scattering by the solvent. In this phenomenon, photons lose an amount of energy corresponding to the energy needed to excite vibrations of the solvent molecules. Therefore, the energy difference between the incident and scattered light (as the wavelength of the rst is varied in a scan) is constant. A constant energy difference corresponds to a varying difference in wavelengths. Therefore, Raman scattering is responsible for the line labelled as 'c' in the Fig. 1, which runs close to line 'a', but is not parallel to it.
The remaining two peaks, designated as peaks '1' and '2' in the 3D uorescence spectra, are due to protein uorescence. The comparison of the three HSA 3D uorescence spectra shows that the relative intensities of these two peaks with respect to the scattering signals is dependent on protein concentration (in addition to the aforementioned experimental factors).
Peak '1' corresponds to excitation at $280 nm and emission at 330-340 nm (for HSA and lysozyme) or 304 nm for insulin. This peak corresponds to emission by the aromatic side-chains of Trp and Tyr residues in proteins. Both these amino acids have an absorbance peak at $280 nm, and are known to be signicantly uorescent, while emission from Phe can be neglected. 30 Tyr uorescence is insensitive to the polarity of its environment, and its emission maximum is always found around 303-305 nm. 1,2,57 Indeed, insulin contains four Tyr residues, and no Trp, and in its 3D spectrum peak '1' is located at l em. ¼ 304 nm. By contrast, Trp emission is very sensitive to the polarity of its environment and its uorescence peak shis from above 350 nm in water down to almost 305 nm in very nonpolar environments or when buried in a hydrophobic protein core, 1,2 like in the extreme case of azurin. 58 The quantum yields of Tyr and Trp uorescence are comparable. However, Tyr uorescence in proteins is usually rather weak, mostly for two reasons: its extinction coefficient at 280 nm is lower by a factor of four than that of Trp, and its emission partially overlaps with the absorption spectrum of Trp, so that Förster resonance energy transfer (FRET) can occur from Tyr to Trp residues. Consistent with these considerations, peak '1' in HSA and lysozyme is dominated by Trp emission, and is located at l em. values, typical for this amino acid, with a red shied maximum for lysozyme, where the Trp residues are more solvent exposed than the lone Trp residue of HSA.
Peak '2' is the feature of protein 3D uorescence spectra that has been claimed to be due to protein backbone emission. By comparing the 3D uorescence spectra of Fig. 1, we note that the wavelength of maximum emission for this peak is always essentially coincident with that of peak '1', even though the latter varies signicantly from one protein to the other (HSA MY: peak '1', 330 nm; peak '2', 331 nm; HSA DK: peak '1', 336 nm; peak '2', 340 nm; HSA IT: peak '1', 332 nm; peak '2', 334 nm; insulin: peak '1', 304 nm, peak '2', 302 nm; lysozyme: peak '1', 339 nm, peak '2', 342 nm). This spectral similarity of the uorescence is further exemplied by overlaying the emission spectra of HSA and insulin upon excitation at 220 and 280 nm in Fig. 2. Kasha's rule 1,2,59 states that the position and shape of the emission spectrum is independent of the excitation wavelength, since uorescence occurs in appreciable yield only from the lowest excited singlet state. Therefore, the observation of a coincidence of the maximum emission wavelength for peaks '1' and '2' is the rst indication that the uorescence originates from the excitation of the same uorophore at different excitation wavelengths, and immediately suggests that peak '2' is due to excitation of a higher excited state of the aromatic residues.
Peak '2' is due to aromatic residues To further show that peak '2' is due to the aromatic moieties and not the protein backbone, we measured the 3D uorescence spectra of individual aromatic amino acids, which therefore contain no amide bonds. Fig. 3 shows that peak '2' is present also for solutions of Trp and Tyr and that its position corresponds roughly to those observed for Trp containing proteins and for proteins with Tyr residues only, respectively (Trp 353 nm; Tyr 303 nm).
There is a signicant difference for the observed emission maximum of lysozyme and HSA ($330-340 nm) and that of Trp ($350 nm), but this can be explained by the different solvent exposure of Trp in the various samples. 1,2,31 Also in the case of Trp and Tyr, overlap of the normalised emission spectra obtained with excitation at 220 nm and 280 nm (Fig. 2) conrms that the emitting species is the same. Finally, Fig. 4 reports the excitation and absorption spectra for HSA and Trp, collected by measuring the emission intensity at 350 nm. For both samples, two bands are observed, peaked at approximately 280 and 220 nm, conrming that the 3D peak observed in the case of HSA and oen attributed to backbone   emission, actually originates from the uorescence of aromatic residues. 30 The slight differences between the excitation spectra of Trp and HSA can be attributed to the presence of Tyr residues in the protein and to the occurrence of Tyr-to-Trp energytransfer, which contributes to the observed uorescence.
The identity of the excited states corresponding to the two excitation peaks of Trp and Tyr (and therefore also of proteins) has been the object of extensive experimental and theoretical studies. Near UV light ($280 nm) leads to excitation of quasidegenerate excited states termed 1 L a and 1 L b in Platt's notation. Excitation in the far UV (peak at 220 nm) brings molecules in higher energy states called 1 B a and 1 B b . 60-62 A small fraction of these molecules then can undergo photoionization, but most relax very rapidly to 1 L a and 1 L b , with subsequent uorescence emission. 63,64 The protein backbone is non-uorescent As a nal conrmation that the peptide backbone does not contribute at all to protein uorescence, we measured 3D uorescence spectra for a number of homopolypeptides containing no aromatic residues (Fig. 5). In these cases, the 3D uorescence spectra correspond to the spectrum observed for the buffer alone (Fig. 1, bottom row, right), and contain only scattering signals. That is, no uorescence signal is observed in the region 300-350 nm, as is observed for the proteins containing aromatic residues.

Conclusions
Even though the polypeptide backbone absorbs in the far-UV region, we show here that it emits no uorescence. The emission peak at 300-350 nm, observed in 3D uorescence spectra upon excitation at 220-230 nm, oen claimed to be caused by backbone emission, is in reality due to the excitation of higher excited electronic states of the aromatic residues present in the protein. Once the aromatic moieties reach these states by absorption of far-UV light, they rapidly relax to the lowest excited state and from there emit at the usual wavelengths of 300-350 nm, according to Kasha's rule. Therefore, this peak reports simply on the local environment of the aromatic residues, just like the peak with excitation at 280 nm, and not on the backbone conformation or secondary structure of the protein.
As illustrated in the introduction, this conclusion should have been obvious to any expert in protein uorescence. Nonetheless, this error has persisted for more than 8 years, and we estimate that there are now well over 100 papers claiming to have observed uorescence from the protein backbone. Unfortunately, we are witnessing an alarming number of errors or misinterpretations in several published uorescence studies (for a discussion of this issue, see ref. 9, 54, 56 and 65-67). The current paper hopefully contributes to a reassessment of one of the possible pitfalls.