*2.2. Raman Analysis of the DBD:AZ Complex*

The previous analysis on the Raman spectra of AZ and DBD proteins, isolated in solution, has provided information on their structural properties paving the way to investigate possible structural change when they are involved in the formation of a complex. The spectrum of DBD:AZ solution, obtained by mixing equimolar amounts of DBD and AZ in the 600–1725 cm−<sup>1</sup> frequency region is shown in Figure 4. We note almost the same general features displayed by the isolated protein spectra with no significant shifts in frequency for the main vibrational modes. From the Tyr peaks visible at 828 and 854 cm−1, the Fermi doublet ratio IY is 0.58 <sup>±</sup> 0.08, which is indicative of a predominant hydrophobic environment. Such a value is closer to that of AZ (IY = 0.38 ± 0.07) with respect to that of DBD (IY = 1.37 ± 0.16), suggesting some changes in the Tyr microenvironment resulting from the interaction between the two biomolecules. To further support such a hypothesis, we have analyzed the Raman spectrum obtained by directly summing the spectra of isolated DBD and AZ molecules acquired at the same concentration used to form the complex (Figure 4), with the resulting spectrum being called added spectrum (AS) in the following. The analysis of the Tyr peaks in the AS spectrum reveals an IY of 1.11 ± 0.18, which is indicative of an average hydrophilic environment for all the Tyr residues in the system, as expected for isolated proteins. Therefore, the marked differences between the IY values from the complex and AS spectra can be ascribed to changes due to the interaction between the molecules. Although the spectroscopic results alone cannot allow us to identify the Tyrs that are involved in the structural changes, they support literature data that point out the involvement of the S7–S8 loops, comprising Tyr220, Tyr234, and Tyr236 and also Tyr126 at DBD binding sites with AZ, which, in turn, is engaged through its a.a 50–77 fragment, including Tyr72 [40].

**Figure 4.** Comparison among the 532-nm-excited Raman spectra (600–1730 cm<sup>−</sup>1) of DBD:AZ complex (green) and of added spectrum AS (orange) in PBS: The principal proteins' vibrational modes are marked. Spectra were normalized in the all spectrum frequency region and baseline corrected for a better visualization.

The spectrum of DBD:AZ shows that the Trp Raman peaks are located at the same frequencies as in the isolated proteins, with an IW ratio of 1.15 ± 0.17, indicating a high hydrophobicity for the three Trps residues (AZ–Trp<sup>48</sup> and DBD–Trp91/Trp146). The found value of IW for DBD:AZ, slightly lower than that for AZ (IW = 1.54 ± 0.10) and higher than both isolated DBD (IW = 0.68 ± 0.10) and AS (IW = 0.53 ± 0.10), suggests some modifications in the environment experienced by these residues upon complex formation. The AZ–Trp<sup>48</sup> is well-known to be strongly buried in the central hydrophobic core of the AZ; therefore, these changes can be due to variations in the DBD–Trp neighboring. Additionally, since the DBD–Trp91 has been shown to be engaged with the Arg<sup>174</sup> [36], we suggest that the observed modifications of DBD as due to AZ interactions occurring within the DBD–Trp146 environment, with this being in agreement with Docking and Molecular Dynamics (MD) data showing the involvement of Trp146 in AZ-binding site [40].

Figure 5A,B shows the fitted curves of the Amide I band for the DBD:AZ and AS spectra, respectively. The results of the best fit for both experimental and AS Amide I bands, obtained by applying the same method used for isolated molecules, are reported in Table 2. DBD:AZ shows a predominant contribution from β-sheet structures (51%) and an α-helix amount of about 26%, while the random coil conformations contributes to 23% of the total Amide I band area. Best fit of AS reveals a predominant β structure (41%) with α-helices and random coils percentages of 31% and 28%, respectively. The observed changes in the secondary structure composition in DBD:AZ with respect to those of DBD and AZ can be attributed to the interaction between these proteins. Furthermore, since AZ is characterized by a highly structured conformation, the decrease of random coil structures can be mainly attributed to DBD. Such a result is supported by previously reported molecular dynamics simulations and docking studies showing that the binding of AZ at the peripheral, unstable, L1 and S7–S8 loops of DBD can enhance their stability upon restraining their flexibility [21], with this being in agreement with a reduction of the DBD disordered regions upon binding to AZ. Accordingly, it could be hypothesized that the increase of structural stability of DBD could be at the basis of the anticancer effect exerted by AZ. Since the structural dynamics and the interactions between proteins are strictly connected, a deeper characterization of the structural–functional relations is of fundamental interest for developing AZ-based drugs, of which effective action in vivo requires, however, further validations [41].

**Figure 5.** Amide I Raman band (open circles) excited at 532 nm of (**A**) DBD:AZ and (**B**) AS spectra, fitted through the Levenberg–Marquardt minimization algorithm (LMA; red line) in which the AZ total fit has been imposed as a constraint: Solid curves indicate the main structural conformations (α-helices, β-sheets, and random coils). Fitting results are summarized in Table 2.

#### *2.3. Principal Component Analysis of DBD, AZ, and the DBD:AZ Complex*

Different combinations of scores for the first three principal components (PC1, PC2, and PC3) have been used to build two-dimensional plots; in the following, the components releasing the highest structural information will be shown. Figure 6A shows the PCA scores of PC1 vs. PC2 components (providing about the 90% of the total variance) for the Fermi Doublet region relative to Tyr residues (790–870 cm−1) for the AZ and DBD isolated molecules and for the DBD:AZ complex. In the scatter plot, two distinct groupings along the PC1 axis can be identified (see the ellipses drawn as a guide). Indeed, the AZ scores (blue symbols) are located in the positive portion of the plot along PC1 with a low spread along both the axes (10 and 5 along PC1 and PC2, respectively), while the DBD and DBD:AZ scores (magenta and green symbols, respectively) occupy the negative range of PC1 values. Along PC1, a larger variability is detected for DBD with respect to AZ and DBD:AZ. Additionally, along PC2, for AZ and DBD, negative values of PC2 are obtained while positive values are detected for DBD. To correlate the position of the scores in the plot with the samples' spectral features, we have analyzed the loadings with the variables mostly contributing to the PCA scores. As shown in Figure 6B,C, high levels of variance are detected in correspondence with the peaks at 829 cm−<sup>1</sup> and 851 cm−<sup>1</sup> related to the Fermi Doublet of Tyr modes and with a weaker peak at 805 cm<sup>−</sup>1, associated with Tyr [25,32,42];

the latter provides the largest variance for PC2 loadings (see Figure 6B) [42]. These results show that Tyr vibrational modes are responsible for sample differentiation in PCA, consistent with our previous study supporting the important role played by Tyr modes as structural markers [25,42].


**Figure 6.** (**A**) Two-dimensional scores plot PC1 versus PC2 of the Raman spectra for AZ (blue squares), DBD (magenta triangles), and DBD:AZ complex (green circles) in the PBS performed on Fermi Doublet of Tyr region (790–870 cm<sup>−</sup>1): The three groupings are indicated by ellipses. The Fermi Doublet ratio for Tyr residues are also reported. (**B**) PC1 (73% of total variance) and (**C**) PC2 (18% of total variance) one-dimensional loadings plot versus frequency. The Raman markers are indicated.

We then applied PCA to the Fermi Doublet region relative to Trp residues (1310–1380 cm−1), with the PC1 and PC2 components providing about 91% of the total variance (see Figure 7A). AZ clusters at the upper side, DBD clusters at the middle, while DBD:AZ clusters at the lower region in correspondence to negative values of PC2 axis. Along PC1, DBD, and DBD:AZ, scores are mixed within an overlapped cloud, while AZ are well-clustered in a well-separated group. Concerning the loading plots, shown in Figure 8B,C, PC2 presents a broad band with a loading positive value of about 0.2. At 1360 cm−1, a single evident peak emerges as ascribed to one of the Fermi Doublet of the Trp modes. This suggests that the separation among the three groups along PC2 depends on Trp vibrational modes. Since such a frequency changes according to the different Trp side-chain environment taken into consideration [28], a different spatial arrangement of this residue should be envisaged in the DBD isolated molecule and in DBD:AZ complex.

**Figure 7.** (**A**) Two-dimensional scores plot PC1 versus PC2 of the Raman spectra for AZ (blue squares), DBD (magenta triangles), and DBD:AZ complex (green circles) in the PBS performed on Fermi Doublet of Trp region (1310–1380 cm<sup>−</sup>1): The Fermi Doublet ratio for Tyr residues are also reported. (**B**) PC1 (84% of total variance) and (**C**) PC2 (8% of total variance) one-dimensional loadings plot versus frequency. The Raman markers are indicated.

**Figure 8.** (**A**) Two-dimensional scores plot PC1 versus PC2 of the Raman spectra for AZ (blue squares), DBD (magenta triangles), and DBD:AZ complex (green circles) in the PBS performed on the Amide I band (1560–1720 cm<sup>−</sup>1): The secondary structure percentages as obtained by the curve fitting analysis are also reported. (**B**) PC1 (46% of total variance) and (**C**) PC2 (39% of total variance) one-dimensional loadings plot versus frequency. The Raman markers are indicated.

Finally, the PCA was performed on the Amide I Raman band of AZ, DBD, and the DBD:AZ complex. From the scatter plot of PC1 versus PC2 (see Figure 8A), AZ data (blue squares) cluster at the positive side of the PC1 axis, with a low variance along both of the components, while DBD (magenta triangles) and DBD:AZ data (green circles) are characterized by negative values of PC1, with some overlap between them. A significantly larger variability is detected in DBD with respect to that of DBD:AZ.

The PC1 loading curve, accounting for 46% of the variance (Figure 8B), is characterized by a very broad band from 1625 to 1730 cm<sup>−</sup>1, including all the frequencies related to the secondary structure of a protein. Such a band shows the lowest value for the 1650 cm−<sup>1</sup> frequency and the highest for the 1680 cm−<sup>1</sup> one, which are associated to the α-helices and disordered structures, respectively [19,24]. In PC2, accounting for a 39% of the variation in the spectra, the major source of variance comes from peak at 1627 cm<sup>−</sup>1, consistent with the disordered structures component (Figure 8C) [19]. This indicates that PC1 discriminates the data based on different amounts of secondary structure of the sample, while PC2 reflects the amount of conformational disorder. Indeed, the AZ scores are very close to each other, indicating a very low variability in the secondary structure within different batches of samples, with this being consistent with the AZ ordered secondary structure [32]. Additionally, the superposition of DBD and DBD:AZ data along PC1 can be explained by assuming that the intrinsic disordered nature of DBD is able to populate an ensemble of different conformations. On the other hand, DBD:AZ distribution on the plot is narrower than that of DBD, reflecting a lower degree of disorder in the complex. These results confirm that PCA is a good reporter of the different structural differences among AZ, DBD, and DBD:AZ. Moreover, PCA is sensitive to changes in the conformational heterogeneity of DBD in the presence of AZ.
