INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Tryptophan fluorescence is widely used to study the location, physical and dynamic properties of microenvironment of indole fluorophores, and the structural features and behavior of the protein molecule as a whole (Burstein, 1976, 1977a, 1983; Lakowicz, 1983; Demchenko, 1986). Depending on the environment of tryptophan residues in proteins, the maximum position (_m) and quantum yield (q) of tryptophan fluorescence could vary widely, from 308 to 353 nm and from 0.4 to immeasurably low, respectively.

In 1967 Konev put forward the hypothesis of the existence of two main classes of tryptophan residues in proteins, which possess discrete values of fluorescent parameters _m and q (Konev, 1967; Volotovski and Konev, 1967). One of the classes included tryptophan fluorophores inside the protein in a low-polar hydrophobic environment with a shorter-wavelength position of fluorescent maximum (_m of ~330 nm) and rather low quantum yield (0.04 to 0.07). The second class consisted of exposed tryptophan residues in a high-polar aqueous environment with long-wavelength position of spectra (_m of ~350 nm) and quantum yield equal or higher than that of free aqueous tryptophan (~0.13-0.17). This hypothesis had been based on the observation that protein spectrum shifts toward 350-353 nm upon denaturation by urea, and toward 330-332 nm upon addition of anionic detergents in acidic solutions. However, this model could not explain the existence of proteins with a high quantum yield (for example, 0.20-0.27 for serum albumins (Longworth, 1971)) and intermediate spectral maximum positions (341.5 nm) for some single-tryptophan-containing proteins, such as human serum albumin (Ivkova et al., 1971).

In 1973 one of us, with co-workers, revised and extended Konev's hypothesis of discrete classes of tryptophan residues in proteins and suggested a new model using some additional spectral parameters and approaches (Burstein et al., 1973). The spectral bandwidth at the half-maximal amplitude was taken as an additional parameter. The linear relationship was found between values of and maximum position _m for tryptophan and other C-3-substituted indole derivatives in solvents of various polarities. These spectra could be regarded as a series of "elementary" components representing the emission bands of individual tryptophan residues in proteins. Existence of a spectral shift accompanying the quenching of protein fluorescence by ionic solutes (NO₃, I, Cs⁺) indicated the multicomponent character of a protein emission spectrum. It was demonstrated that the spectra of proteins, which are shifted upon quenching, possess values exceeding those of "elementary" components. The analysis of the differences between initial spectra of such proteins and those after 20% quenching revealed that the best-quenched components have the longest-wavelength position at ~340 nm in native proteins, which is ~10 nm shorter than that postulated in the two-state model. However, the tryptophans in proteins denatured by urea or guanidinium chloride have a spectrum with _m of ~350 nm. These results allowed us to develop an extended model of discrete states (classes) of tryptophan residues in proteins, which assumed the existence of five statistically most probable classes (Burstein et al., 1973; Burstein, 1977a, 1983). According to the model, the following discrete classes of tryptophan residues were predicted to be most probable in proteins (Burstein, 1977b, 1983):

1.	Class A (m = 308 nm, structured spectra); the fluorophores, which do not form hydrogen-bound complexes in the excited state (exciplexes) (Hershberger et al., 1981) with solvent or neighboring protein groups;
2.	Class S (m = 316 nm, structured spectra) includes the buried tryptophan residues that can form the exciplexes with 1:1 stoichiometry;
3.	Class I (m = 330-332 nm,  = 48-50 nm) represents the buried fluorophores that can form the exciplexes with 2:1 stoichiometry;
4.	Class II (m = 340-342 nm,  = 53-55 nm) represents the fluorophores exposed to the bound water possessing very long dipole relaxation time, which precludes completing the relaxation-induced spectral shift during the excited-state lifetime;
5.	Class III (m = 350-353 nm,  = 59-61 nm) contains rather fully exposed fluorophores surrounded by highly mobile water completely relaxing during the excitation lifetime, which makes their spectra almost coinciding with those of free aqueous tryptophan.

At that time, there was constructed and used the calibrated diagram versus _m for estimating the contributions of three most frequent model classes (I, II, and III) of tryptophan residues in the fluorescence spectrum of a protein (Burstein et al., 1973). The application of this hypothesis was rather effective in the interpretation of protein tryptophan fluorescence data in various biophysical and molecular-biology studies. However, the idea of the existing of discrete classes of tryptophans in proteins was based on the analysis of very limited numbers of proteins. Now it seems reasonable to reappraise the hypothesis of discrete states applying more recent analytic and computing methods to the wide statistical database of protein spectra.

Such a reappraisal is possible now due to progress in protein preparation techniques and methods of component analysis of protein emission spectra. It was shown that the fluorescence spectra of tryptophan and its derivatives may be accurately described analytically by a log-normal function (Burstein, 1976), proposed initially for the absorption spectra by Siano and Metzler (1969). Thereupon, it was found that the shape of the emission spectra of tryptophan both in solutions and in a protein depends only on the spectral maximum position _m in a known manner (Burstein and Emelyanenko, 1996). This fact allowed us to develop effective algorithms of reliable decomposition of tryptophan fluorescence spectra of proteins into individual "elementary" log-normal components (Abornev and Burstein, 1992; Burstein et al., 2001).

Here we present the results of such decomposition of tryptophan emission spectra of >100 different proteins, some in various structural states (native and denatured, in complexes with ions or organic ligands, in various pH-induced conformations, etc.). The database of obtained log-normal components allowed us to check the hypothesis of the discrete classes of tryptophan residues on the basis of a representative set of proteins. Analysis of the histograms of occurrence of spectral components with various maximum positions confirmed the statistical discreteness of several states of emitting tryptophan fluorophores in proteins. The discrete classes observed here are similar to those proposed in the model of discrete states in 1973-1977.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Fluorescence spectra

The large majority of protein fluorescence spectra used in this work and listed in Table 1 were taken from the archive of experimental data of the Laboratory of Functional Biophysics of Proteins (Institute of Theoretical and Experimental Biophysics of the Russia Academy of Sciences). The spectra have been published and/or analyzed in various special publications cited in the column "References." These publications also contain the data about protein preparations and techniques of fluorescence measurements. If the spectra were measured elsewhere and given to us, they were recorrected for instrument spectral sensitivity curves using the emission spectra of aqueous tryptophan solution as a standard. Such a recorrection was necessary because the component analysis algorithms were developed based on the model spectra measured with our lab-made instrument (Burstein et al., 1973; Bukolova-Orlova et al., 1974) with a carefully estimated spectral sensitivity curve (Burstein and Emelyanenko, 1996).

The preparations of wheat germ agglutinin, inorganic pyrophosphatase (pyrophosphate phosphohydrolase) from baker's yeast, monellin from Dioscoreophyllum cummensii fruits, Escherichia coli alkaline phosphatase, and subtilisin BPN' produced by Sigma (St. Louis, MO) were kindly given to us by Dr. J. Borejdo (University of North Texas Health Science Center, Fort Worth, TX). Human and bovine carbonic anhydrases II, staphylococcal serine protease, and subtilisin Carlsberg (Sigma) were a kind gift from Dr. S. S. Lehrer (Boston Biomedical Research Institute, Boston, MA). The preparation of bovine carboxypeptidase A was from Reanal (Hungary). The yeast 3-phosphoglycerate kinase was from Sigma. Porcine pancreatic lipase was from Serva. Hen egg white ovalbumin was from Reakhim (Olaine, Latvia). Actin from dog heart was prepared by one of us (Ya. K. R.) according to Pardee and Spudich (1982).

Fluorescence measurements

Fluorescence spectra of the several proteins marked in Table 1 (in the "References" column) as "Present work" were measured by us during the last six years. Acrylamide, KCl, KI, CsCl, Tris, and ATP preparations used in these experiments were ultra-pure grade of Russian or Soviet production or Sigma.

Steady-state emission spectra were recorded on the lab-made spectrofluorimeter (Burstein et al., 1973; Bukolova-Orlova et al., 1974) with collection of emitted light from the cell front face, that allowed application of the strict correction function for screening and reabsorption inner filter effects (Burstein, 1968). The fluorescence was excited with the mercury

Analysis and representation of spectral data

The decomposition of tryptophan fluorescence spectra of proteins into log-normal components was performed for sets of spectra measured at varying concentrations of quenchers using two algorithms: SIMS (SImple fitting with Mean-Square criterion) (Abornev and Burstein, 1992; Burstein et al., 2001) and PHREQ (PHase-plot-based REsolving with Quenchers, for two-component decomposition) (Burstein et al., 2001). For the decomposition of a single spectrum (without quenchers) the modified SIMS algorithm was applied (Burstein et al., 2001). Using SIMS, the spectra were independently fitted by one, two, or three components. The criterion of attaining the solution (a minimal sufficient number of components describing the spectra) was the minimal root-mean-square differences (residuals) between theoretical and experimental spectra multiplied by the number of parameters under fitting. For several proteins two solutions with different numbers of components, which had the similar values of this criterion, were selected as the best ones.

The fact that the components of composite tryptophan emission spectra of proteins are wide (~30-60 nm at half-maximal level) compared with the differences between the values of their maximum positions (from 5 to ~50 nm), might considerably reduce the reliability of decomposition. Therefore, we present in Table 2 the averaged results (values of the maximum position, _m(i) and the contribution into the area under spectrum, S(i) and the standard deviations of averaging) obtained separately with the mathematically different algorithms: PHREQ and SIMS. The good agreement between the parameters that were obtained with these two methods demonstrate the reliability of a result. We can note that for the absolute majority of proteins the results obtained with these two methods differ less than by errors of means. It allowed us to present unified results of the component analysis for individual proteins independent of the decomposition method.

Moreover, the majority of protein spectra were measured in the presence of various quenchers, in most cases more than once, using preparations from different sources or varying solvent conditions. To extract unified values of the parameters for any given protein or its conformer, two averaging schemes were applied to the sets of decomposition results: "Averaging" and "Mean Best Fit" (Table 2). According to the first scheme, the parameters of components were calculated as averages of all solutions obtained for all individual experiments, independent of the quality of the decomposition. Within the second scheme, only the data obtained with minimal values of the decomposition functional for each experiment were averaged.

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

In Table 2 we present the results of decomposition of tryptophan fluorescence spectra of >100 different proteins, some in various structural states (native and denatured, in complexes with ions or organic ligands, and in various pH-induced states), obtained by the above-mentioned algorithms, SIMS and PHREQ, and different averaging schemes of the obtained solutions, averaging and mean best fit. The SIMS algorithm was used to the log-normal decomposition of 163 sets of tryptophan emission spectra. The PHREQ algorithm was applied to the 105 sets of fluorescence spectra because the other 58 spectra of proteins were measured without quenchers, and only the modified SIMS algorithm could be applied to decompose them. The results obtained with both SIMS and PHREQ methods for absolute majority of proteins differ less than errors of means presented in the _m(i) and S(i) columns. It allows presenting the unified results of the component analysis independent of the decomposition method. Besides the results of log-normal component analysis, Table 2 contains the maximum position values of whole, non-decomposed fluorescence spectra of proteins ("_m total," the third column). The denatured proteins are italicized in the tables.

The database of log-normal components of protein tryptophan fluorescence spectra shown in Table 2 may be regarded as a generalized representative database of fluorescence spectra of individual tryptophan residues in proteins. It allowed us to perform statistical testing of the hypothesis of discrete states, i.e., the existence of the set of several classes of tryptophan residues in proteins discretely differing in their fluorescence spectral maximum positions, which was put forward as early as 1973-1977 (Burstein et al., 1973; Burstein, 1977a, b, 1983). Such a test is performed here by constructing and analyzing the histograms of occurrence of log-normal components presented in the database of Table 2.

Before examining histograms of occurrence of individual log-normal components, it seems reasonable to consider how randomly are distributed spectral maximum positions of whole, non-decomposed emission spectra of proteins of our database. The histograms of occurrence of _m values of whole spectra are shown in Fig. 1, A and B for native and denatured proteins, respectively, constructed using the "_m total" values from Table 2. It is seen that the histogram of _m for native proteins is rather well fitted by normal distribution with the peak at 336.7 ± 0.4 nm and  = 6.0 ± 0.9 nm, which is a good confirmation of random character of the excerpt of native proteins. A worse fitting by the normal distribution was obtained for the maximum positions of fluorescence spectra of denatured proteins (the peak at 346.5 ± 0.6 nm;  = 4.1 ± 1.3 nm); however, it is expected due to the small number of proteins and, mainly, due to essentially different structural features of protein states formed under various denaturing effectors used (urea, dioxane, extreme pH, etc.).

View larger version (28K): [in this window] [in a new window]   FIGURE 1   Canonical distribution of occurrence of maximum positions of total, non-decomposed emission spectra for native (A) and denatured (B) proteins with 2-nm step. Solid lines represent the fitting by Gauss function.

The canonical histograms of occurrence of maximum position values of log-normal components (with the 2-nm steps) of emission spectra of native proteins are presented in Fig. 2. Panels A and B represent the distributions _m(i) obtained with two decomposition algorithms, SIMS (375 components) and PHREQ (249 components), respectively. Panels C and D reflect the results obtained using two schemes of averaging ("Average" and "Mean Best Fit" in Table 2), respectively. All four histograms have the well-expressed deep minimum at 335-337 nm as a common feature. This feature seems to clearly demonstrate the existence of a statistical discreteness of at least two large spectral classes of tryptophan residues in native proteins. Moreover, panels A-D demonstrate reproducible dips at 329 ± 1 and 347 ± 1 nm, which may also mark the frontiers between overlapping discrete classes of tryptophans in proteins.

View larger version (64K): [in this window] [in a new window]   FIGURE 2   Canonical distribution of occurrence of maximum positions of log-normal components obtained by applying SIMS (A), PHREQ (B) algorithms, and different schemes of averaging of the results: "Average" (C) and "Mean Best Fit" (D) for native proteins with 2-nm step.

However, the canonical form of histogram, constructed from columns, cannot reflect the accuracy of calculated maximum position values and of relative contributions of components in the whole emission spectrum. Therefore, we constructed a new kind of histogram accounting for these factors (Figs. 3-5). In such histograms, each element (the individual ith spectral component) is represented by the little Gauss distribution y(_m(i)) with maximum at _m(i), (i) equal to the standard (root-mean-square one) error of the mean _m(i) value (see Table 2), and the maximal amplitude of Gauss curves proportional to the relative contribution S(i) of the ith component to the total protein emission. Such a histogram is a sum of these elements:
Here, S(i) is expressed in decimal fractions and (i) is a multiplier equal to 1 if only one decomposition result is selected for a protein spectrum, or to 0.5 in the cases when two solutions with similar functional values are taken.

View larger version (45K): [in this window] [in a new window]   FIGURE 3   Distribution of occurrence of maximum positions of log-normal components obtained by applying SIMS (A), PHREQ (B) algorithms, and different schemes of averaging of the results: "Average" (C) and "Mean Best Fit" (D) with 1-nm step. Each curve is a sum of Gauss function (see explanation in the text) for each spectral component of native (N) and denatured (D) proteins with (i) = 1.0 and S(i) = 1.0.

In such a representation, the histograms of Fig. 2 acquire the view seen in Fig. 3 (curves N). In this case, the S(i) values are taken to be 1, and  = 1.0 for every spectral component. One can see that the main features of the canonical histograms are retained in the new, Gauss-curve-based representation. Besides the histograms for components of spectra of native proteins, Fig. 3 also contains the curves obtained for proteins denatured under various influences (see Table 2). The shorter-wavelength "continua" reflect the fact that denaturation is not completed in some cases and the emission spectra possess components belonging to the retained populations of native and/or partly unfolded molecules. However, the highest peak at ~350 nm and peaks or shoulders at 340-347 nm are due to the components belonging to tryptophans in proteins denatured by urea or guanidinium chloride and by dioxane, pH and other less unfolding denaturing factors. It is interesting that the histograms for denatured proteins also have the minima at 335-340 nm, which position coincides with the deep minima of histograms for native proteins.

We constructed two more Gauss-curve-based histograms with values of (i) and S(i) varying according to the calculation data for native (N) and denatured (D) proteins. The histograms accounting only for the differences in contributions of individual components in whole spectra are presented in Fig. 4 ((i) = 1.0 nm; S(i) = var). The resulting reduction in contributions of "small" components (S(i) < 0.01) in the histogram essentially improved the resolution of individual peaks in it. The "small" components may not only have worse determined _m values, but also may be erroneous in principle in the cases of bad quality of a spectrum under decomposition. The fact that the improved resolution of peaks in such histograms at ~326, 331, 345, and 350 nm allowed us to assume that they reflect the existence of discrete classes of tryptophan emitters in native proteins. The reduction of contributions of minor components essentially decreased the relative amplitude of a shorter-wavelength wing in the denatured-proteins' histogram. The remaining shorter-wavelength peaks (instead of a "continuum") roughly correspond to the peaks in the native-proteins' histogram. However, the 350-nm peak and the shoulder at 340-347 nm became much better-expressed.

View larger version (46K): [in this window] [in a new window]   FIGURE 4   Distribution of occurrence of maximum positions of log-normal components obtained by applying SIMS (A), PHREQ (B) algorithms, and different schemes of averaging of the results: "Average" (C) and "Mean Best Fit" (D) with 1-nm step. Each curve is a sum of Gauss function (see explanation in the text) for each spectral component of native (N) and denatured (D) proteins with (i) = 1.0 and S(i) equal the relative contribution of the component in the total protein emission (see Table 2).

To demonstrate that the resolution of peaks obtained in previous histograms (and, thus, the observed spectral discrete classes) are beyond the error of the estimation of the maximum position of spectral components, we constructed the histograms (Fig. 5) accounting for the precision of _m(i) estimation ((i) = var) and the relative contributions of components (S(i) = var). These histograms demonstrate the presence of discreteness of emission of tryptophan residues in proteins in the same manner as Figs. 3 and 4. There are seen the global minimum at ~337 nm and four maxima at ~326, 333, 344, and 350 nm for native proteins. The PHREQ algorithm well-resolves only two classes, because it was applied only to the limited number of protein spectra, possessing one or two components. For denatured proteins, accounting for (i) made peaks even more expressive. The main maxima for denatured proteins are at ~346 and 350 nm; however, the peak at ~334 nm also exists.

View larger version (46K): [in this window] [in a new window]   FIGURE 5   Distribution of occurrence of maximum positions of log-normal components obtained by applying SIMS (A), PHREQ (B) algorithms, and different schemes of averaging of the results: "Average" (C) and "Mean Best Fit" (D) with 1-nm step. Each curve is a sum of Gauss function (see explanation in the text) for each spectral component of native (N) and denatured (D) proteins with (i) equal to the standard (root-mean-square one) error of the mean m(i) value and S(i) equal to the relative contribution of the component in the total protein emission (see Table 2).

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

In 1973-1977 Burstein and co-workers postulated the hypothesis of the existence of discrete classes of tryptophan residues in proteins (Burstein et al., 1973; Burstein, 1977a, 1983). Later, algorithms of stable and reliable decomposition of composite tryptophan fluorescence spectra into log-normal components were developed (Abornev and Burstein, 1992; Burstein and Emelyanenko, 1996; Burstein et al., 2001) that allowed us to verify and revise the hypothesis. Here, we used mathematically different algorithms of decomposition of emission spectra to examine the stability of solutions. The fitting algorithm (SIMS) and the analytical ones (PHREQ) gave, in general, very similar results: obtained maximum positions _m(i) differed within an error of ±2.8 nm, and contributions S(i) of components differed within the range of ±8.6%.

First of all, we showed that the distribution of maximum positions of total, nondecomposed spectra ("_m total" in Table 2) for native proteins is rather well-fitted by a Gauss equation with the maximum at ~337 nm and dispersion of ~12 nm (Fig. 1 A), which demonstrated the randomness and, thus, representativeness of the set of the proteins under study. Then, we analyzed the distributions of the maximum position of log-normal components taking into consideration the experimental deviations of maximum position and the contributions of components to spectra of native and denatured proteins (Figs. 2-5). All obtained data confirmed the existence of several statistically discrete classes of emitting tryptophan fluorophores in proteins. The discrete classes observed here are in approximate agreement with those proposed in 1973-1977 in the model of discrete states. A very important point is that there was no a priori assumption about the discreteness in any algorithm used for component analysis of fluorescence spectra (Burstein et al., 2001).

It is evident that to find the causes for so widely differing and discrete Stokes shifts in the emission spectra of tryptophan fluorophores in proteins one has to look in a variety of combinations of interactions of individual fluorophores with their environment, occurring mainly during the lifetime of the fluorescent excited state. The physical causes for the differences among the spectroscopic classes can be analyzed in terms of various combinations of Bakhsiev's theory of specific and universal interactions of the excited fluorophores with their environment (Bakhshiev, 1972). Both kinds of interactions are induced by essential redistribution of electronic density on the atoms and chemical bonds of fluorophore after its electronic excitation. The specific interactions include changes in the nearest range interactions with neighboring groups and/or solvent molecules, i.e., resolvation, hydrogen bond formation, or other noncovalent complexing (Bakhshiev, 1972; Mataga et al., 1955). The universal interactions occur due to the dipole relaxation of surrounding dielectric continuum in response to the changes in direction and magnitude of fluorophore dipole moment at the excitation (Bakhshiev, 1972; Bilot and Kawski, 1962; Lippert, 1957; Liptay, 1965;. Mataga et al., 1956).

The formation of stoichiometric complexes (most probably hydrogen-bonded ones) of excited indole fluorophore with alcohols was demonstrated experimentally (Walker et al., 1967; Lumry and Hershberger, 1978; Hershberger et al., 1981). Such complexes with alcohols in apolar cycloheptane had discrete positions of fluorescence spectral maxima dependent on the stoichiometric ratios of alcohol/fluorophore: the maximum was at 316 nm for the 1:1 ratio and at ~330 nm at the 2:1 ratio compared with the central peak at 307 nm in the structured spectrum obtained in the absence of any polar co-solvents. The authors interpreted these excited-state complexes (exciplexes) as a result of H--bonding of alcohol hydroxyls with indolic N- and C-atoms as those possessing maximal electronic density in the excited state. Such a location of H-bonds may be revised in the light of recent quantum-mechanical calculations of electronic density distribution in the excited indole and tryptophan fluorophores, which revealed that maximal densities are located at the C3, C2, and C2 atoms in the main fluorescent ¹L_a state of the indolic ring (Callis, 1997).

In proteins, the structured fluorescence bands coinciding with those of nonexciplexed indole fluorophores occur in azurin (Burstein et al., 1977) and in bacteriorhodopsin purple membranes suspended in 2 M CsCl (Permyakov and Shnyrov, 1983). Such tryptophan residues were attributed to class A in the hypothesis of discrete states. The structured spectra similar to those of the exciplexes 1:1 occurring in several single-tryptophan-containing proteins were assigned to class S (Burstein, 1977a, b, 1983). Emission spectra of classes A and S do not undergo any spectral shift under freezing the protein solutions down to 196°C, i.e., the dipole moments of these fluorophores practically do not change at the excitation. In the histograms obtained in this work, classes A and S are reflected by peaks and/or shoulders at 305-308 and 316-318 nm, respectively (see Figs. 3-5).

In the model experiments (Walker et al., 1967; Lumry and Hershberger, 1978; Hershberger et al., 1981), the maximum position of the exciplex 2:1 spectrum was not constant under increasing alcohol concentration, which reflected the rise of Stokes shift induced by the solvent dipole relaxation in response to the change of fluorophore dipole moment in the excited state of the exciplex. Therefore, the florescence of protein tryptophan residues possessing the spectra with maxima at ~330 nm and longer were assigned to the emission of exciplexes with stoichiometry not <2:1 in the model of discrete classes (Burstein, 1977a, b, 1983). The model provided three discrete classes (I-III) for such exciplexes with stoichiometry 2:1, differed in their accessibility to extrinsic quenchers (vanishing accessibility for class I, _m at ~330 nm, and the accessibility of >30% of that of free tryptophan for classes II and III). The difference between two accessible classes was assumedly interpreted as a result of different dipole relaxation rates in their environments.

The histograms (Figs. 3-5) contain peaks reflecting all three discrete classes. However, the class I band in the histograms splits up into two peaks, at ~325 and 332 nm. Almost all histograms contain the global minimum at 336-337 nm and two evident maxima at 344-345 and 349-351 nm, which correspond to the emission of exposed tryptophan residues of classes II and III, respectively. In the model of discrete classes, it was suggested that exposed tryptophan residues in native proteins should emit at 340-342 nm, and emission at 350-353 nm is due to tryptophan residues of denatured proteins (Burstein et al., 1973). However, the extended set of proteins investigated here indicates that tryptophans in native proteins also can have spectra with maxima at ~350 nm.

Tryptophan residues of denatured proteins (italicized in Tables 1 and 2) have the main peak positions at 340-345 and 351 nm. The longest-wavelength peak corresponds to the completely unfolded proteins in urea or guanidinium chloride solutions. However, the peak at 340-345 nm belongs mainly to the fluorophores in proteins denatured by extreme pH values (e.g., pepsin at pH 6-9; see Table 2) or high dioxane concentrations (chymotrypsin, chymotrypsinogen, and trypsin; see Table 2). It is possible that such maximum positions of fluorescence spectra are characteristic of tryptophan residues in such intermediate unfolding states of proteins as molten globule (Ptitsyn, 1995). The shorter-wavelength peaks (_m < 337 nm) in histograms for denatured proteins belong, most probably, to the fluorophores in molecules retained native or partly unfolded in denaturing conditions.

The model of discrete states reflects the existence in proteins of the five most manifested classes of tryptophan residues. The discreteness of fluorescence parameters has to have a probabilistic nature. The histograms obtained in this work reflected all these classes and can be regarded as a statistical confirmation of the hypothesis of discrete classes of tryptophan residues in proteins. It could be assumed that such a situation might be realized because tryptophan residues are located in a few kinds of physically preferred environments in protein structures that provide realization of different combinations of specific and universal interactions of excited fluorophore with its environment.

There were a lot of papers published during the last several last years where it was demonstrated that tryptophan residues could have various fluorescence parameters depending on their environment in proteins (Callis, 1997; Callis and Burgess, 1997; Reshetnyak and Burstein, 1997a, b; Chen and Barkley, 1998; Kuznetsova and Turoverov, 1998; Meagher et al., 1998). Because highly resolved x-ray and NMR structures of many proteins are now available, we developed a system of describing physical and structural characteristics of the microenvironment of tryptophan residues and examined the existence of discreteness of structural parameters of the microenvironment of tryptophan residues in proteins. Moreover, that work allowed us to elucidate the main physical and structural factors determining the class of an individual tryptophan residue. The next paper of this series contains the results of that study.