Molecular Interactions Identified by Two-Dimensional Analysis-Detailed Insight into the Molecular Interactions of the Antimalarial Artesunate with the Target Structure β-Hematin by Means of 2D Raman Correlation Spectroscopy

Abstract

A thorough understanding of the interaction of endoperoxide antimalarial agents with their biological target structures is of utmost importance for the tailored design of future efficient antimalarials. Detailed insights into molecular interactions between artesunate and β-hematin were derived with a combination of resonance Raman spectroscopy, two-dimensional correlation analysis, and density functional theory calculations. Resonance Raman spectroscopy with three distinct laser wavelengths enabled the specific excitation of different chromophore parts of β-hematin. The resonance Raman spectra of the artesunate-β-hematin complexes were thoroughly analyzed with the help of high-resolution and highly sensitive two-dimensional correlation spectroscopy. Spectral changes in the peak properties were found with increasing artesunate concentration. Changes in the low-frequency, morphology-sensitive Raman bands indicated a loss in crystallinity of the drug–target complexes. Differences in the high-wavenumber region were assigned to increased distortions of the planarity of the structure of the target molecule due to the appearance of various coexisting alkylation species. Evidence for the appearance of high-valent ferryl-oxo species could be observed with the help of differences in the peak properties of oxidation-state sensitive Raman modes. To support those findings, the relaxed ground-state structures of ten possible covalent mono- and di-meso(C_m)-alkylated hematin-dihydroartemisinyl complexes were calculated using density functional theory. A very good agreement with the experimental peak properties was achieved, and the out-of-plane displacements along the lowest-frequency normal coordinates were investigated by normal coordinate structural decomposition analysis. The strongest changes in all data were observed in vibrations with a high participation of C_m-parts of β-hematin.

Introduction

Malaria is a widespread (sub)tropical infectious disease. Almost half of the whole UN population was at risk in 2020,¹ and over 241 million incidents with almost 627 thousand deaths occurred. More than 90% of these cases happened in African countries, with more than two-thirds in children under 5 years of age.¹

Unfortunately, a rapid development and spread of resistance against common antimalarials is observed in endemic malaria regions worldwide.²⁻⁵ Thus, a better understanding of the detailed molecular interactions of the respective drugs with their target structures is urgently needed.⁶⁻¹¹

One of the most powerful “weapons” in the arsenal for the fight against the burden of malaria are drugs of the endoperoxides class with its prototype artemisinin.¹² This herbal remedy has been known from ancient traditional Chinese folk medicine since more than four millennia.¹³ Artemisinin is very poorly soluble in water or oil; therefore, semisynthetic water- and lipid-soluble derivatives have been created.¹⁴ Artesunate (AS, see Figure 1A1) shows benefits compared to the lipid-soluble relatives¹⁵ and is the most widely used derivative.¹⁶ Hemisynthetic artemisinin derivatives have several advantages compared with other commonly used antimalarials. They are extremely well tolerated,¹⁷ have a broad stage-specificity of action, and are rapid-acting antimalarials against multiresistant parasite strains.¹⁸ Without any reported form of resistance,¹⁹ but tolerant parasite strains,²⁰ they can cure all forms of malaria infections.²¹

Figure 1.

(A): Chemical structures of artesunate (A1), dihydroartemisinin (A2), and the covalent complex between heme and artesunate for a β-alkylation (A3). The atomic nomenclature and the possible alkylation meso-positions (marked with arrows) of the complex are also shown (A3). The ruffling dihedral angle (Z-[C_α–N–N–C_α]) (indicated by a red line) and the core size (C_t–N) are also highlighted in the structure (A3). B/C/D: the sum spectra of pure β-hematin (1, black) and a 2:1 artesunate/hematin complex (2, red) are shown for the excitation wavelengths 413 nm (B), 568 nm (C), and 647 nm (D). The Raman signals that show major changes during the complexation are labeled in the respective spectra (A–C).

The malaria parasite Plasmodium falciparum digests the human host hemoglobin in the food vacuole during the intra-erythrocytic part of its life cycle,^22,23 and parasiticidal heme (ferroprotoporphyrin [ferro-PP]IX)²⁴ is released.²⁵ Due to the missing heme oxygenase cycle occurring in mammals,²⁶ the protozoan oxidizes toxic heme first to α-hematin (ferri-PPIX-OH),²⁷ which is further biomineralized into the almost unreactive, nontoxic, crystalline malaria pigment hemozoin.²⁴

Due to its very short elimination half-times of only several minutes,^28,29 AS and other artemisinin derivatives are now used in artemisinin-based combination therapy (ACT). As recommended by the World Health Organization,³⁰ an endoperoxide is given in combination with a synergistic, long-acting antimalarial to enhance the effectiveness³¹ and to minimize the risk of drug resistance.^32,33

Although the mechanism of action of the endoperoxide drugs is highly disputed and still not completely understood,³⁴ it should be different from one of any other known antimalarial classes due to its unique 1,2,4-trioxane cycle pharmacophore. Various studies agree that the endoperoxide moiety is essential for their antimalarial activity¹⁴ because compounds lacking this structural feature are without any effectiveness.¹⁵ Furthermore, a two-step mechanism of action for the artemisinins is generally recognized. The first one, the activation step, occurs via the reductive scission or open peroxide model³⁵ by low-valent synthetic or biological iron sources.³⁶ Here, oxygen-centered artemisinin-based radicals are generated and immediately rearranged by homolytic C–C-cleavage^37,38 or intramolecular 1,5-H-atom-shift³⁹ to more stable,⁴⁰ carbon-centered ones with⁴¹ or without⁴² a ring opening of the trioxane cycle. Posner et al.^43,44 propose the involvement of high-valent iron species in the mechanism of action, which are formed via homolytic oxygen–carbon bond scission.^41,43,44 These Fe^IV=O-species are responsible for oxidative damage within parasitic cells.^43,44 The produced primary and secondary carbon-centered radicals were investigated in vitro and in vivo in infected mice^45,46 and detected by electron spin resonance techniques.⁴⁷⁻⁴⁹ In the second step, the generated radicals, which are highly alkylating species, bound covalently to specific vital parasite biomolecules,⁵⁰⁻⁵² synthetic metalloporphyrins,⁵³⁻⁵⁵ heme,^18,45 hemozoin,⁵⁶ reduced glutathione,⁵⁷ cysteine,⁵⁸ or other target systems.

In this report, only the interactions of AS with β-hematin, the synthetic analogue to hemozoin, will be thoroughly investigated. The buildup of covalent heme-artemisinin adducts (HemArts) was first shown with the help of mass spectrometry.^59,60 Possible binding sites were identified as the four different meso-positions (α, β, γ, δ) of the porphyrin cycle with NMR spectroscopy.^54,61 Covalent HemArt adducts were presented by Laurent et al.,⁶² and the interaction of the heme unity with artemisinin-derived carbon-centered radicals yields mono¹⁸- and di-alkylated^36,63 drug–target complexes.

In in vitro and exvivo radiolabeling experiments, Hong et al. detected that a major part of the endoperoxide was accumulated in the parasite hemozoin⁵⁶ as covalent drug-heme adducts.¹⁴

Inside the food vacuole of the parasite, the artemisinins react with free heme, which is further transformed into hemozoin.⁶⁴ The amount of hemozoin crystals decreases in the acidic environment of the food vacuole,^65,66 which could be caused by the interactions of endoperoxides with free heme monomers.⁶⁷ These HemArt complexes cannot further crystallize into the malaria pigment because the bulky artemisinyl groups bound covalently to the porphyrin moiety, and stacking of the propionate chain of one heme molecule with the iron atom of another one is almost inhibited.⁶³

The protoporphyrin macrocycle itself consists of conjugated aromatic bonds. A planar geometry is typically favored and seen in naturally occurring iron porphyrins in the absence of external perturbation.⁶⁸ Nevertheless, the heme moiety shows high flexibility,⁶⁹ when nonplanar deformations by steric crowding of substituents at the porphyrin periphery⁷⁰ occur. These nonplanar distortions of metalloporphyrins can be classified according to irreducible representations⁷¹ of the nominal D_4h point group of a square-planar porphyrin macrocycle. Depending on the specific substituents,⁶⁹ ruffling (B_1u, ruf), saddling (B_2u, sad), doming (A_2u, dom), waving (E_g, wav), and propellering (A_1u, pro)⁷² could be observed and investigated by normal coordinate structural decomposition (NSD) analysis. NSD is used to qualify and quantify the out-of-plane distortions of porphyrin macrocycles^73,74 and to characterize them in terms of a linear combination of all normal coordinates from an ideal planar reference geometry.^72,73 In this analysis, the distortions are represented in terms of the lowest-frequency out-of-plane vibrational modes of the respective symmetry race of the molecule.⁷³ The composition of normal deformations along the normal coordinates depends on the substitution pattern of the porphyrin macrocycle. For symmetrically substituted metalloporphyrins, the deformations often occur only along one of the lowest-frequency out-of-plane normal coordinates,⁷¹ whereas the nonplanar distortion tends to deform in a more complicated manner along two or more of these normal coordinates for asymmetrically substituted porphyrins.⁷⁵

The (multiple) alkylation(s) of the heme unity by AS-derived radicals cause deformations of the porphyrin core, resulting in significant changes of chemical and spectroscopic properties of the macrocycle.⁷⁰ Raman spectroscopy^76,77 is a powerful analytical technique,⁷⁸⁻⁸² which provides highly specific molecular information⁸³⁻⁸⁸ in a direct nondestructive way.⁸⁹⁻⁹⁴ The method was used to quantify the nonplanarity of metal porphyrin rings^{68,75,95−97} induced by peripheral substituents^71,98,99 and for the investigation of molecular interactions of active agents^100−102 with their specific target structures. Various types of Raman spectroscopy were used to observe and understand the complexation mechanism of common antimalarials with β-hematin,^103−110 the synthetic analogue to hemozoin,¹¹¹ whose spectroscopic properties are well known.^112−119 The most prominent changes regarding the nonplanarity of protoporphyrins, which can be observed in the Raman spectra, occur in the high-wavenumber region (1300–1650 cm^–1), where many “marker bands”^120−125 are present. The frequencies of the core-size marker bands^121−123 are known to correlate with the core size (distance from pyrrole N atoms to C_t, the projection of the metal ion onto the N₄ plane;¹²⁴ C_t–N, see Figure 1A3) of the metalloporphyrin.^121,126 For almost planar porphyrins with the same small degree of nonplanarity, these Raman lines show an inverse linear correlation with the core size,^{121−123,127−131} and a positive correlation for nonplanar porphyrins.^{71,75,131,132} The origin of those observed frequency shifts was expected to lie in minor changes in the C_αC_m-bond length,⁷⁰ and a rough correlation with the percentage distribution of C_αC_m-stretching of the respective vibrational modes¹³³ and the core-size correlations was found.¹²⁸

While the drug–target interactions between hematin systems and endoperoxides are not fully analyzed with Raman spectroscopy, significant changes, such as the appearance of new peaks,^25,134 occur in the spectra of the target structure upon binding with the antimalarial.¹⁰⁸ Artemisinin and its derivatives react poorly with hemozoin in vitro(56) because only a few ferric heme molecules at the surface of the crystals, which are not linked by iron-carboxylate bonds,¹³⁵ are available for decomposition of artemisinins.

To investigate these minor modifications in the respective Raman spectra, two-dimensional correlation analysis (2D-Corr)^136−138 was used for data analysis. This technique was developed in pioneering work by I. Noda in the 1990s^139,140 and allows a better detection and evaluation of even small changes in spectroscopic data induced by external perturbations. 2D-Corr reveals several advantages in comparison to common one-dimensional data: the enhancement of spectral resolution by introducing a second dimension and the qualitative identification of any intermolecular interaction by changes in peak properties.^139−141

Some^12,67,103 but not all^142,143 studies suggest that artemisinins act by inhibition of hemozoin. Structure–activity relationship studies have shown that the antimalarial activity of artemisinin is highly correlated with heme-binding¹⁴⁴ and porphyrin-alkylation.¹⁴⁵ Hence, endoperoxides might act as hemozoin growth inhibitors—causing the release of toxic free heme and initiating the parasite’s death.

The interactions of the most widely used endoperoxide artesunate with its target structure β-hematin are analyzed in this study with the help of Raman and 2D-Corr high-resolution spectroscopy to derive a better understanding of the molecular mode of action of artemisinin-based antimalarials.

Materials and Methods

Synthesis of β-Hematin and its Complexes with Artesunate

β-hematin and its complexes with artesunate were prepared according to an acid-catalyzed dehydration synthesis.¹⁴⁶ For generating the drug complexes, before acidification, artesunate (ChemShuttle) in molar ratios (AS/hematin) of 2:1, 1.5:1, 1:1, 1:1.33, 1:2, 1:2.85, and 1:5 was added. After synthesis, the dark precipitate was filtered through a nylon filter (pore size: 22 μm, Millipore). A more detailed description can be found in the Supporting Information (SI).

Sample Preparation

Ultrapure deionized water (κ > 0.06 mS cm^–1) from an ultrapure water feed system (SG Water GmbH) was used for the preparation of aqueous solutions. For the Raman spectroscopic experiments, pieces of the filters containing the dried samples (see the SI) were cut and fixed on a glass slide.

Raman Spectroscopy

For the experiment, a Raman setup (HR LabRam, Horiba/Jobin-Yvon) with excitation wavelengths 413, 568, and 647 nm of a krypton-ion laser (Coherent Innova 300C) with laser powers at the samples of 125, 273, and 450 μW, respectively, was used. More detailed information about the Raman setup used, including the excitation wavelengths 413, 568, and 647 nm, can be found in the SI.

Data Preprocessing and 2D Correlation

All preprocessing and analysis of the raw Raman data were done in statistical programming GnuR (version: 3.3.3).¹⁴⁷ The packages “signal”,¹⁴⁸ “peaks”,¹⁴⁹ “EMSC”,¹⁵⁰ and “corr2D”¹⁵¹ were utilized, and their functions were complemented. After standardizing the wavenumber grid of the Raman spectra for each excitation wavelength (1 cm^–1), the data were reduced to the wavenumber region of interest (200–1650 cm^–1). Afterward, a Savitzky–Golay smoothing¹⁴⁸ was applied, and the resulting spectra were background corrected using the SNIP algorithm.¹⁴⁹

For 2D correlation (for its workflow, see Figure 2), the sum of all individual spectra of the respective ratios was formed for each of the three excitation wavelengths used in order to increase the signal-to-noise ratio. Those were then matched to the total mean spectrum of all generated sum spectra by using an extended multiplicative scatter correction (EMSC).¹⁵⁰ The 2D correlation of the preprocessed Raman data was achieved using an approach based on Fourier transformation.¹⁵¹

Figure 2.

General workflow of the experimental approach for generating 2D correlation spectra.

In addition, the most promising bands observable as intense auto peaks in the respective synchronous 2D correlation spectra (see Figure 3B,D,F) were fitted with a Gaussian peak profile.

Figure 3.

Selected regions of the asynchronous (A, C, E) and synchronous (B, D, F) 2D Raman correlation spectra of β-hematin with the respective ratios of added artesunate (see text). The resonance Raman spectra plotted at the top and the left or right are the respective reference spectra of untreated β-hematin. Regions with red color indicate positive peaks, and blue areas show negative peaks. Resonance Raman spectra at excitation wavelength λ_exc = 413 nm (A, B) are shown in the wavenumber region between 1340 and 1650 cm^–1, the chosen range of the Raman spectra in the case of λ_exc = 568 nm (C, D) was between 1265 and 1650 cm^–1, and the 2D correlation Raman spectra derived with λ_exc = 647 nm (E, F) are presented in the range of 300 to 390, 710 to 780, and 1365 to 1650 cm^–1.

The exact workflow of data preprocessing with R software and the resulting 2D correlation are described in the SI.

Density Functional Theory (DFT) Calculation and NSD Analysis

Quantum chemical calculations based on DFT were performed using Gaussian 16¹⁵² software. Ground-state geometries of the alkylation complexes between the hydrolysis product of AS,¹⁵³ dihydroartemisinin (see Figure 1A2), and α-hematin were optimized. The calculations were performed with Truhlar’s and Zhao’s Minnesota global hybrid functional with Hartree–Fock exchange (M06-2X).¹⁵⁴ The double-ζ LANL2DZ basis set with effective core potential^155,156 augmented by polarization functions of d symmetry and diffuse functions of p symmetry (LANL2DZpd¹⁵⁷) was used for the optimization of ground-state geometries for all atoms except iron, for which the LANL2TZ^158,159 basis set augmented by diffuse functions (LANL2TZ+¹⁵⁹) was utilized. All structures were modeled in the high-spin state (S = 5/2; M_S = 6) for the ferric iron center,¹⁶⁰ using unrestricted (u)M06-2X functional, and the stability of their resulting wavefunctions was checked. As starting geometry for hematin, the crystal structure^161,162 was used, and the dihydroartemisinyl-rests were added at respective meso-position(s). For the reference structure without any meso-alkylation, after successful convergence, the dihydroartemisinyl-rests were exchanged by hydrogen atoms, and an optimization and frequency calculation was started.

NSD analysis of the derived energy-minimized structures of the complexes was done with the NSD program (version 2) originally written by Jentzen et al.^73,74 and modified by Liptak et al.¹⁶³

The methodology and the level of theory used for DFT calculations, as well as the NSD evaluation of the generated structures, are accessible in the SI.

Results and Discussion

Raman Spectra of β-Hematin and Its Complexes with Artesunate

Artesunate is a weak Raman scatterer,^164,165 absorbs light only at low wavelengths (∼200 nm), and has no distinct ultraviolet–visible (UV–vis) spectra or fluorescent properties in the region of interest.¹⁶ The Raman spectrum of artesunate shows no significant overlap with one of β-hematin (see Figure S1), which allows the investigation of the interactions through small changes in the Raman peak properties. The sum Raman spectra of β-hematin and its 1:2-complex with AS at the respective excitation wavelengths were derived (Figure 1B–D). The band assignments, local symmetry coordinates, terms, and spectroscopic notation for β-hematin can be found elsewhere.^113,133,166 The mean Raman spectra of all synthesized drug complexes (Figure S2) show small deviations of maximum intensities in between the spectra for all applied excitation wavelengths. Hence, for two-dimensional (2D) correlation, the sum of all individual spectra of the respective AS-Hematin ratios was formed for each of the three excitation wavelengths used in order to increase the signal-to-noise ratio.

The interaction of AS with β-hematin causes small changes of several Raman bands. For λ_exc = 413 nm and λ_exc = 568 nm, the major changes are visible in the high-wavenumber region, whereas for λ_exc = 647 nm also, mid- and low-wavenumber signals are affected (Figure 1B–D). Explicitly spoken, in the case of λ_exc = 413 nm, the ν₁₀, ν₄, ν₃, and ν₂¹¹³ vibrations are influenced by the drug interplay. For excitation with the 568 nm laser wavelength, the ν₁₀-, ν₂₁-, ν₄-, and ν₂-Raman bands change mostly, and for λ_exc = 647 nm, the ν₁₅, ν₁₀, and ν₈ are influenced due to the interaction of artesunate with the synthetic malaria pigment (Figure 1B–D). In order to identify the small changes in the Raman spectra and to get a better understanding of these changes, a 2D correlation of the Raman data set was performed (see Figure 2).

2D Correlation Raman Spectra

Synchronous correlation spectra Φ(ν_A,ν_B) consist of auto- or diagonal peaks (always positive, see Figures 3B,D,F and S3B,D,F), localized on the main diagonal and off-diagonal cross-peaks, which can be either positive or negative. The intensity of the synchronous correlation spectra represents the dynamic intensity difference of the signal at a specific wavenumber ν_i. The higher this signal, the stronger are the changes during the applied perturbation.¹⁴⁰ From the shape of the resulting auto peaks, additional information about changes of other important peak parameters (vide supra) can be drawn based on simulations.^{138,150,167,168}

For 413 nm, the auto peak of ν₄ has the shape of a peak that is shifting to lower wavenumbers combined with a negative intensity change. The pattern in the asynchronous spectrum was examined to verify whether the observed downshift in the 1D data for ν₄ (slightly shifted from (1374.9 ± 0.4) to (1374.5 ± 0.9) cm^–1 and intensity reduction, see Figure 1B) is a pure peak position change or a superposition of multiple signals, which change its intensity in the opposite direction.¹⁶⁹ No butterfly pattern^167,169 was observed (see Figure 3B), which would be characteristic for a pure shift in band position. Thus, the peak shift is not caused by a change of a single peak position but by a combination of at least two signals changing their intensities in the opposite direction. This difference should be pointed out because it was observed for all other peaks, which might shift to red or blue during drug treatment according to the 1D data. However, this “shift” results from an overlap of multiple signals changing their intensities in a different direction. The same four-leaf-clover/angel pattern^167,169 or more scientific “quadrupolar profile” could also be observed for the auto peak of ν₂ (see Figure 3B), but the “head of the angel” is pointing toward higher wavenumbers (see Figure 3B), and thus the intensity of the signal is reduced. The peak position of ν₂ is shifted to higher wavenumbers, from (1573.5 ± 0.5) to (1577.3 ± 0.7) cm^–1 (see Figure 1B). In the asynchronous 2D-Corr spectrum (Figure 3A), the observed pattern is changing the sequence of its signs (first positive, then negative). The same intensity change, but no position shift, happens to the ν₁₀-band at 1630 cm^–1 (see Figure 3B). The intensity of ν₃ appears slightly enhanced, and no peak shift is observed (see Figure 1B).

For λ_exc = 568 nm, the strongest auto peak is Φ(ν₄,ν₄) (Figure 3D). This band shows a negative peak position shift from (1373.4 ± 0.6) cm^–1 to (1371.0 ± 0.5) cm^–1, and its intensity is decreased (see Figure 1C), which is visualized by the shape of the pattern (see Figure 3C,D). The ν₁₀-signal has a similar pattern (see Figure 3D), and the band is shifted from (1621.5 ± 1.4) cm^–1 to (1619.9 ± 0.6) cm^–1 in combination with almost the same intensity (see Figure 1C). A positive shift of the peak position happens to ν₂₁ from (1307.6 ± 0.4) cm^–1 to (1309.5 ± 0.2) cm^–1, and its intensity is increasing (see Figure 1C). The same peak shape in the asynchronous and synchronous spectra is also present for the Raman signal at ν₂ (see Figure 3C,D), which is also increasing in intensity (see Figure 1C).

In the case of λ_exc = 647 nm, the auto peak at ν₁₀ shows a distorted four-leaf-clover pattern (Figure 3F) and is shifted to lower wavenumbers from (1623.2 ± 0.4) to (1615.1 ± 1.6) cm^–1 and shows a decreased intensity during the drug treatment (see Figure 1D). The signal at ν₁₅ shows similar changes in its peak properties: a shift from (754.8 ± 0.2) to (753.3 ± 0.3) cm^–1 and an intensity decline (see Figure 1D). The signal at ν₈ has a strong negative peak position shift from (343.2 ± 0.2) to (334.6 ± 1.2) cm^–1, and its intensity is reduced due to the interactions with the drug (see Figure 1D).

Looking at Figure 1B again, one can clearly see that ν₁₀, ν₄, and ν₂ have a reduced intensity after the drug treatment, and only ν₃ has increased intensity. Looking at the synchronous 2D-Corr spectrum in Figure 3B (λ_exc = 413 nm), one can identify the cross-peaks and their signs: Φ(ν₁₀,ν₄) > 0, Φ(ν₁₀,ν₃) < 0, Φ(ν₁₀,ν₂) > 0, Φ(ν₄,ν₃) < 0, Φ(ν₄,ν₂) > 0, and Φ(ν₃,ν₂) < 0. This indicates that ν₁₀, ν₄, and ν₂ undergo the same changes in intensity, whereas the intensity of ν₃ changes in opposite toward the other three vibrations. Only at this excitation wavelength (λ_exc = 413 nm), intensive cross-peaks at 852 cm^–1 with ν₁₀, ν₄, ν₃, and ν₂ could be detected (Figure S3B). Kapetanaki et al. assigned this band at 850–852 cm^–1 to a ν(Fe^IV=O) vibration of ferryl-oxo-porphyrin complexes at resonant excitation wavelengths in the range of 407–460 nm.^134,170,171 This signal could not be observed in the 1D data because no such dramatic experimental conditions (e.g., oxygen matrices or external heating devices) were used in our experiments for generation of the high-valent iron species. According to the 2D data, this signal grows during drug treatment, because Φ(ν(Fe^IV=O),ν₁₀) < 0, Φ(ν(Fe^IV=O),ν₂) < 0, Φ(ν(Fe^IV=O),ν₃) > 0, and Φ(ν(Fe^IV=O),ν₄) < 0. The peaks of ν₁₀, ν₄, and ν₂ lose intensity during increased AS concentration, but ν₃ increases to higher intensity (vide supra; see Figure 1B1,B2). No auto peak was seen in the 2D-Corr spectra because, in Kapetanaki’s experiment, an 80 times higher laser power and over 30 times longer total accumulation time^134,172 were used, but their peak still has a signal-to-noise ratio below two.¹³⁴ This clearly shows the benefits of the 2D correlation technique with higher resolution and sensitivity.

For λ_exc = 568 nm, a different intensity situation is observed in the synchronous spectrum (Figure 3D). Beside several positive cross-peaks [Φ(ν₂₁,ν₁₀), Φ(ν₂₁,ν₂), and Φ(ν₂,ν₁₀)], where the affected Raman signals change their intensity the same way, also some negative cross-peaks [Φ(ν₂₁,ν₄), Φ(ν₄,ν₂), and Φ(ν₄,ν₁₀)] can be identified. This indicates that ν₂₁, ν₁₀, and ν₂ gain intensity and ν₄ loses intensity during the drug treatment, as may be guessed from Figure 1C.

In the case of red excitation (λ_exc = 647 nm), all cross-peaks of the investigated signals [Φ(ν₁₅,ν₁₀), Φ(ν₁₅,ν₈), and Φ(ν₁₀,ν₈)] are positive (Figure 3F). Thus, all bands have a reduced intensity after drug complexation (see Figure 1D).

Asynchronous correlation spectra Ψ(ν_A,ν_B) show only cross-peaks (Figures 3A,C,E and S3A,C,E),¹³⁸ which represent independent fluctuations of the system that occur separately during the disturbance.¹⁷³ This enhances the spectral resolution because the position of heavily overlapping peaks with only little variations in their intensities can still be identified.¹⁷⁴ To find out which Raman band of the spectrum is most affected by the applied perturbation, one can determine the “sequential order” of the peaks with the help of (simplified) “Noda’s rules”.^169,175 In doing so, the product of the signs of the synchronous Φ(ν_A,ν_B) and asynchronous Ψ(ν_A,ν_B) cross-peaks of two spectral bands at ν_A and ν_B should be calculated. If it is positive, the intensity changes at ν_A occur earlier than at ν_B or vice versa for negative values. In this case, “earlier” means that the intensity of the signal at its respective position reaches its half-intensity during the perturbation faster than the other one,¹⁷⁶ and the overall sequential order is only an average of the whole data set.¹⁷⁷ This means that events (vibrations) that occur first are more influenced by the applied drug concentration than those that were observed second and following. In the case of λ_exc = 413 nm, the asynchronous cross-peaks at Ψ(ν₁₀,ν₄), Ψ(ν₁₀,ν₃), Ψ(ν₁₀,ν₂), and Ψ(ν₄,ν₃) are positive (Figure 3A), whereas the ones at Ψ(ν₄,ν₂) and Ψ(ν₃,ν₂) have a negative value. In combination with the signs of the respective synchronous cross-peaks (Figure 3B), the determination of a sequential order of the analyzed Raman signals is possible. The spectral changes of ν₃ occur before those of ν₁₀, which occur before ν₂ and ν₄.

For λ_exc = 568 nm, several positive [Ψ(ν₂₁,ν₁₀), Ψ(ν₂₁,ν₂), Ψ(ν₄,ν₂), and Ψ(ν₂,ν₁₀)] and negative [Ψ(ν₂₁,ν₄) and Ψ(ν₄,ν₁₀)] cross-peaks can be identified in the asynchronous 2D correlation spectrum (Figure 3C). With the help of the known values (positive or negative) of the synchronous cross-peaks (Figure 3D), the spectral changes of ν₂₁ appear before ν₂, which occur before ν₄ and ν₁₀.

Because all asynchronous cross-peaks for λ_exc = 647 nm [Ψ(ν₁₅,ν₁₀), Ψ(ν₁₅,ν₈), and Ψ(ν₁₀,ν₈)] are negative (Figure 3E), a sequential order can be determined as ν₁₀ > ν₁₅ > ν₈.

Other spectral changes can also be determined by studying the behavior of other peak properties during increasing AS concentration. Relatively high FWHMs, derived from the peak fits of the porphyrin skeletal vibrational modes, are indicators for conformational inhomogeneities^71,125 in the structure of the drug–target complexes. For excitation wavelength λ_exc = 413 nm, the modes ν₄ [from FWHM = (27.2 ± 0.9) to (30.0 ± 1.3) cm^–1], ν₃ [from FWHM = (17.1 ± 0.4) to (18.4 ± 0.5 cm^–1)], and ν₂ [from FWHM = (31.4 ± 1.1) to (32.7 ± 1.1) cm^–1] show a small broadening with increasing AS ratio (see Figure 1B). The same trend was observed for λ_exc = 568 nm for ν₂₁ [from FWHM = (20.0 ± 1.9) to (23.4 ± 0.8) cm^–1] and ν₄ [from FWHM = (20.8 ± 0.5) to (22.8 ± 1.1) cm^–1] (see Figure 1C).

In contrast to those findings, by applying excitation wavelength λ_exc = 647 nm (see Figure 1D), the widths of several peaks change dramatically. This laser line was chosen because the Raman signals, which are selectively enhanced in this case, are found to be dependent on the crystallinity of the hematin sample.¹¹⁵ These morphology-sensitive bands at 344 and 372 cm^–1 (ν₈) decrease strongly in intensity with increasing the AS concentration (vide supra, Figure 1D). This clearly indicates a loss in crystallinity for the drug–target complexes, and these changes in morphology can furthermore be proven and confirmed by the high degree of broadening [from FWHM = (13.7 ± 0.3) to (47.4 ± 6.0) cm^–1] for the ν₈ mode in the Raman spectra. Thus, the alkylated heme complexes induce defects in the usually well-ordered crystal structure^103,178,179 of β-hematin,¹¹¹ which causes these differences in the peak’s height and width of the morphology-sensitive marker bands. The FWHM of other bands, such as ν₁₅ [from FWHM = (14.3 ± 0.1) to (17.0 ± 0.3) cm^–1], broadens less (see Figure 1D).

The band ν₁₀ broadens from FWHM = (20.1 ± 0.5) to (43.0 ± 3.6) cm^–1. The value of the broadening of ν₁₀ is also higher than those of all other bands in the case of excitation wavelengths 413 and 568 nm (see Figure 1B,C). It was found that the width of the ν₁₀-band results from multiple nonplanar conformers.^125,180 Indeed, in NMR experiments, four different coexisting structures, namely, the (α, β, γ, and δ) meso-alkylated heme complexes with composition ratios of 25, 28, 14, and 33% (without consideration of regioselectivity) could be observed.⁶¹ Based on probability values, the composition for the possible six di-alkylated complexes can be easily calculated as 19.2% (α, β-HemArt), 9.6% (α, γ-HemArt), 22.6% (α, δ-HemArt), 10.7% (β, γ-HemArt), 25.3% (β, δ-HemArt), and 12.6% (γ, δ-HemArt). With these numbers, it was possible to construct all structural and vibrational properties of the single- and bis-alkylated HemArt species out of the found values for the respective alkylation complexes (vide supra).

NSD Analysis of Nonplanar Relaxed DFT Structures

To sum up all results found so far, we see that for almost all excitation wavelengths, the most prominent intensity changes in the respective Raman spectra appear for vibrational modes with a high ratio of C_αC_m-stretching of C_mH bending vibrations. This clearly indicates that an alkylation of the heme unit by AS-derived radicals takes place because these bands are especially sensitive to the nature of the meso-substituent. In the case of mono-alkylation, four or, for di-alkylation, up to six coexisting HemArt structures are possible, which yield a high broadening of the ν₁₀- and ν₈-mode. For λ_exc = 413 nm, basically, no shift in peak position was observed for ν₁₀-, ν₃-, and ν₄-Raman mode, and the ν₂-vibration was shifted to higher wavenumbers. In the case of an excitation wavelength of 568 nm, ν₂ is shifted to higher wavenumbers, whereas ν₄ and ν₁₀ show a peak position shift to lower wavenumbers. By using λ_exc = 647 nm, the ν₁₀-vibration is also shifted to lower wavenumbers. By analyzing the behavior of changes in peak position for the DFT-calculated Raman spectra, a strong negative shift for ν₁₀ from 1671 cm^–1 for hematin to 1662 cm^–1 for single alkylation and finally down to 1646 cm^–1 for bis-alkylation was found (see Table S1). The same trend was observed for ν₃ (1563, 1533, and 1531 cm^–1) and ν₂ (1644, 1636, and 1633 cm^–1, see Table S1). The peak center of ν₄ shows a much smaller change and has a constant value for both alkylated species (1416, 1411, and 1411 cm^–1, see Table S1). The observed change to a higher peak center of the ν₂-band could be explained by the fact that this band is an oxidation-state marker band,¹²⁰ which is typically shifted to higher wavenumbers with increased oxidation number. As β-hematin has an oxidation number of three,¹⁸¹ the interaction with AS yields ferryl-oxo intermediates with an oxidation number of four¹⁷¹ for the central iron metal.

To quantify the level of nonplanarity of the porphyrin unit induced by possibly multiple alkylations, NSD analysis was performed with the DFT-calculated relaxed ground-state geometries of all ten HemArt complexes and hematin as a reference structure (Figure S4). Based on the composition probability values for mono- and di-alkylated drug–target complexes (vide supra), the out-of-plane displacements along the lowest-frequency normal coordinates can be easily constructed (Figure 4B,C). Here, hematin itself shows a low displacement along the sad-mode of −0.088 Å and an even lower displacement along the ruf- (0.013 Å) and dom-mode (0.073 Å) in the case of using minimal base. Analyzing the NSD values for the single-alkylated HemArt complex, a strong increase of the sad- and a minor one for the ruf-displacements was observed (Table S1). For a possible bis-alkylation, the displacement along the sad-mode is more increased than the ruf-displacement in contrast to mono-alkylation. To describe the out-of-plane distortions more accurately, all higher-order lowest-frequency normal modes of each symmetry type were included (complete basis).⁷³ In doing so, no changes in the trend neither for ruf-displacements (hematin < mono- < double-alkylated HemArts) nor along the sad-mode (hematin < bis- < single-alkylated HemArt complexes) were observed.

Figure 4.

Illustration of the static displacements of the lowest-frequency out-of-plane eigenvectors, represented by an 1 Å deformation for each symmetry type (A, adopted from ref (70)): saddling (B_2u, 1); ruffling (B_1u, 2); doming (A_2u, 3); waving (x) (E_x, 4); waving (y) (E_y, 5); and propellering (A_1u, 6). The NSD analysis of the out-of-plane displacements along the lowest-frequency normal coordinates is presented for hematin (I), mono- (II), and di-alkylated (III) HemArt complexes, using minimal (B) and complete basis (C).

The alkylation induces an increase in the nonplanarity of the heme macrocycle. Therefore, first hints can be drawn from the (C_α–N–N–C_α) dihedral (ruffling,¹⁸² see Figure 1A3) and the (N–Fe–N) angle (Table S1) in the molecular structures. The torsion angle increases strongly in value, and the ∠(N–Fe–N) decreases while ongoing alkylation of the hematin-moiety (Table S1). Those differences change the core size of the porphyrins. In our particular case, a C_t-contraction of 1.0 ± 0.2% leads to a prolongation of all outer ring bonds¹²⁵—the bond length of C_αC_m shows an increase by 0.9 ± 0.4%, whereas the other ring bonds (C_βC_β, C_αC_β, and C_αN) show only slight changes (Table S1). In contrast to other transition-metal porphyrin complexes,^{71,75,97,125,183} these small changes occur typically for five-coordinated ferrous porphyrin complexes, where only weak core-size correlations were found in previous studies.¹⁸⁴ Thus, only small changes in the peak position of high-wavenumber Raman modes can be observed.

In both experiment and theory, the ν₁₀-mode shows most of these frequency changes, whereas other high-wavenumber Raman peaks (ν₂, ν₃, and ν₄) have almost no differences in peak position. A reason for that behavior could be that the ν₁₀-band, which can be mainly assigned to an asymmetrical stretching vibration of the C_αC_m-bond,¹¹³ is over 3 times (3.1) more affected (increase in bond length) by the alkylation than the C_βC_β-bond, where the ν₂-vibration¹¹³ is mainly located. The displacement vectors of the normal modes ν₁₀ and the remaining high-frequency Raman vibrations (ν₂, ν₃, and ν₄) of the investigated HemArt complexes can be found in Table 1.

Table 1. Displacement Vectors of the Normal Modes ν₁₀, ν₂, ν₃, and ν₄ of the Optimized Starting Geometry (Crystal Structure) of Hematin and all HemArt Complexes Computed by DFT at the (u)M06-2X/LANL2DZpd/LANL2TZ+ Level of Theory^a.

^aFor better visualization, the hematin unit is shown in tube, and the added dihydroartemisinyl rest is presented in wireframe style. The respective wavenumber positions of the vibrations are listed in Table S1.

Conclusions

The vibrational (resonance) Raman spectra of in total seven different AS/hematin ratios were analyzed utilizing three visible excitation wavelengths. The properties of individual Raman peaks were investigated, as they are most affected by the increasing AS concentration. Comparing the Raman spectra of β-hematin at the respective excitation wavelength (Figure 1B1,C1,D1) with the spectra of a 2:1-AS/hematin complex under the same conditions (Figure 1B2,C2,D2), various changes in peak properties of distinct Raman bands with a high ratio of C_αC_m-stretching of C_mH-bending vibrations arose because alkylation of the hematin unit takes part at the C_m-positions. For λ_exc = 413 nm and λ_exc = 568 nm, these changes are mainly observed at peaks in the high-wavenumber region between 1650 and 1350 cm^–1 (see Figures 1B,C, and 3A–D). The ν₂-vibration shifted toward higher wavenumbers at these two excitation wavelengths. A possible explanation is a change of the ν₂-band, which is sensitive to the oxidation state of the central iron, due to the appearance of high-valent ferryl-oxo (Fe^IV=O) species. The first evidence for this hypothesis could be detected in the form of cross-peaks of the ν(Fe^IV=O) at ∼852 cm^–1 with the high-frequency vibrations ν₁₀, ν₂, ν₃, and ν₄ in the synchronous 2D-Corr spectrum (Figure S3B). Thus, the benefits of the high-resolution and high sensitivity of 2D-Corr spectroscopy could be demonstrated. Much higher laser powers and longer accumulation times were needed in conventional 1D Raman spectroscopy to observe such peaks.

For λ_exc = 568 and 647 nm, a strong downshift of the ν₁₀-band was observed due to changes in the planarity of the porphyrin core (see Figures 1C,D and 3C–F). To support this finding, the relaxed ground-state geometries of in total ten different HemArt complexes and hematin as a reference structure were computed with the help of density functional theory (Figure S4).

To derive a value for the nonplanarity of the different complexes, NSD analysis was performed with the calculated structures (Figure 4). A strong increase along the saddling mode and an increase in the ruffling displacement could be observed during single- and double-alkylation of the hematin unit (Figure 4B,C). Another reason for this alkylation-induced frequency reduction of the ν₁₀-band could be that the C_αC_m-bond (where the ν₁₀-vibration is mainly located) is more affected (increase in length) by the alkylation than any of the other outer ring bonds (Table S1). A strong change of the ν₁₀ band of hematin from 1671 to 1662 cm^–1 for mono-alkylation and finally down to 1646 cm^–1 for bis-alkylation was found in the peak position of the calculated Raman spectra of HemArts (Table 1), which supports the experimental findings. Another interesting fact about the peak properties of this band is that in the case of mono-alkylation four and in the case of di-alkylation, up to six coexisting HemArt structures are possible, which yield a high broadening of the ν₁₀-mode in the resonance Raman spectra for an excitation wavelength of 647 nm.

At λ_exc = 647 nm, the low-frequency bands at 344 and 372 cm^–1 (ν₈), which are dependent on the morphology of the sample, decrease dramatically in intensity with increasing AS concentration (Figure 1D). This clearly indicates a loss in crystallinity for the HemArt complexes and can be furthermore confirmed by a broadening of the ν₈-mode in the Raman spectra (Figure 1D).

These results provide new insights into the interaction of endoperoxide antimalarials with β-hematin and confirm existing possibilities for this type of interaction. Very small changes in the peak properties could be identified with the help of high-resolution and highly sensitive 2D correlation spectroscopy.

This technique is very interesting for further drug–target interaction experiments to analyze the molecular interaction of other antimalarials with heme targets with high selectivity and to derive a better understanding of their molecular mode of action. With the help of these investigations, differences in efficiency between common antimalarial drugs or drug combinations, in the case of ACTs, can be verified and understood in order to develop new, even more efficient antimalarials in the future.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.analchem.3c01415.

• Experimental details, materials, and methods; collection of structural parameters from DFT-calculated structures of investigated HemArt complexes as well as (scaled) Raman frequencies, saddling and ruffling displacements (Table S1); comparison between mean Raman spectra of pure artesunate and pure β-hematin (Figure S1); mean Raman spectra of pure β-hematin (1), and synthesized complexes with hematin/artesunate ratios of 5:1 (2), 2.85:1 (3), 2:1 (4), 1.33:1 (5), 1:1 (6), 1:1.5 (7), and 2:1 (8) for excitation wavelengths 413 nm (A), 568 nm (B), and 647 nm (C) (Figure S2); whole (range 200–1650 cm^–1) asynchronous (A, C, E) and synchronous 2D Raman correlation spectra (B, D, F) of β-hematin with the respective ratios of added artesunate for excitation wavelengths of 413 nm (A, B), 568 nm (C, D), and 647 nm (E, F) (Figure S3); and results of the NSD analysis of the out-of-plane displacements along the lowest-frequency normal coordinates is presented for all investigated HemArt complexes, using minimal (A) and complete basis (B) (Figure S4) (PDF)

The authors declare no competing financial interest.