Analytical and Bioanalytical Chemistry

, Volume 392, Issue 7, pp 1277–1282

Spatial distribution of heme species in erythrocytes infected with Plasmodium falciparum by use of resonance Raman imaging and multivariate analysis

Authors

    • Department of Materials and Natural ResourcesUniversity of Trieste
  • Sara Finaurini
    • Department of Public Health-Microbiology-VirologyUniversity of Milan
  • Christoph Krafft
    • Bioanalytical ChemistryDresden University of Technology
  • Silvia Parapini
    • Department of Public Health-Microbiology-VirologyUniversity of Milan
  • Donatella Taramelli
    • Department of Public Health-Microbiology-VirologyUniversity of Milan
  • Valter Sergo
    • Department of Materials and Natural ResourcesUniversity of Trieste
Technical Note

DOI: 10.1007/s00216-008-2414-0

Cite this article as:
Bonifacio, A., Finaurini, S., Krafft, C. et al. Anal Bioanal Chem (2008) 392: 1277. doi:10.1007/s00216-008-2414-0

Abstract

The multivariate algorithm hierarchical cluster analysis is applied to sets of resonance Raman spectra collected from human erythrocytes infected with the malaria parasite Plasmodium falciparum. The images obtained yield information about the distribution of hemoglobin and hemozoin (or malaria pigment) within the parasitized cells and about their molecular structure. This method has the advantage of conveying more information than other imaging approaches based on resonance Raman spectroscopy, and it is a promising tool to study the hemozoin formation process and its interaction with antimalarial drugs within unstained, well-preserved parasites.

Keywords

MalariaPlasmodium falciparumHemozoinImagingResonance RamanMultivariate analysis

Introduction

Malaria is a parasitic disease transmitted by mosquitoes occurring in tropical and subtropical regions of the world. Among the four species causing human malaria, Plasmodium falciparum is the most aggressive and lethal, causing up to two million deaths every year [1].

In the human body, the parasites multiply in the liver and then inside host erythrocytes, where they digest a major proportion of red cell hemoglobin in an acidic digestive vacuole [2]. Free heme [ferriprotoporphyrin IX, or Fe(III)PPIX], a product of hemoglobin digestion, is dangerous for the parasite because it can cause harmful oxidant species, eventually leading to irreversible damage of biological membranes. The parasites detoxify Fe(III)PPIX by converting it into an insoluble crystal named “hemozoin,” or “malaria pigment.” The mechanism of hemozoin formation is still not completely clear and many hypotheses have been proposed [3].

The detoxification of Fe(III)PPIX into hemozoin is an important drug target: 4-aminoquinoline antimalarials are believed to form a π−π stacking with the planar aromatic structures of Fe(III)PPIX, resulting in inhibition of hemozoin formation and consequent buildup of toxic heme [4, 5]. A better understanding of the mechanisms of hemozoin formation and of drug action would be extremely valuable for the design of new drugs to increase efficacy and overcome resistance, which is presently the major drawback of antimalarial chemotherapy [6, 7].

Resonance Raman (RR) spectroscopy proved to be a useful tool in malaria research, in particular to study the structure and properties of hemozoin both in situ and ex situ, its structural analogy with the synthetic compound β-hematin, and its interaction with antimalarial drugs [8]. This nondestructive technique is based on the inelastic scattering of laser radiation whose wavelength is in resonance with an electronic transition of the molecule studied (i.e., a chromophore); hence the name “resonance Raman effect” [8, 9]. In RR spectra the intensities of bands corresponding to vibrational modes of a chromophore (e.g., heme or hemozoin) are enhanced over vibrations due to other molecules which may be present in the sample. Since chromophore vibrations are related to the molecular structure, information about heme species such as hemozoin (or its synthetic analogue β-hematin) may be readily inferred from RR data, even when these species are found in complex mixtures such as those present in a cellular environment [8, 1014].

The use of microscope optics to focus a laser on the sample and collect the scattered light (RR microspectroscopy) permits one to couple the structural and chemical information contained in Raman spectra with spatial information, with a lateral resolution which is in principle only limited by light diffraction. Spectral characteristics can be “mapped” on the sample, in the approach named “Raman mapping,” by superimposing a two-dimensional grid of points over the sample and collecting a whole spectrum from each point. Unlike traditional histopathological methods (e.g., Giemsa staining), this approach allowed imaging of the distribution of hemozoin within unlabeled, undyed parasitized erythrocytes, by mapping the spectral intensity at frequencies characteristic of the malaria pigment [10].

The chemical and structural information in an image based on a Raman map can be increased upon processing data according to multivariate data analysis. Cluster analysis (CA) has been successfully employed to process nonresonant Raman maps in order to visualize simultaneously different subcellular structures and their chemical components in single cells [1517]. CA classifies spectra into groups (clusters) according to their spectral similarity, and the resulting images are “chemical maps” in which the sample is divided into areas (i.e., group of pixels) having different colors, where each color corresponds to a different cluster.

In this paper, the distributions of hemoglobin and hemozoin are shown simultaneously in RR maps of parasitized erythrocytes processed with the hierarchical cluster analysis (HCA), a particular CA method [18]. The advantages of HCA applied to RR imaging are discussed in view of future applications.

Materials and methods

Parasite cultures and sample preparation

P. falciparum culture of D10 strain was maintained at 5% hematocrit (human type A-positive red blood cells) in RPMI 1640 medium (EuroClone, Celbio, Milan, Italy) with the addition of 1% AlbuMAXII (Gibco, Invitrogen, Italy), 0.01% hypoxanthine (Sigma-Italia, Milan, Italy), 20 mM N-(2-hydroxyethyl)piperazine-N′-ethanesulfonic acid, and 2 mM glutamine (EuroClone, Celbio), as described in [19]. All the cultures were maintained at 37 °C in a standard gas mixture consisting of 1% O2, 5% CO2, and 94% N2. Smears of normal or parasitized red blood cells at the trophozoite stage were prepared on glass microscope slides, fixed in methanol (Sigma-Italia, Milan, Italy), and dried under ambient conditions. The parasitemia in the samples was between 2 and 5% (reproducing that of severe malaria in adults).

Human hemoglobin (lyophilized powder) was purchased from Sigma, whereas β-hematin was synthesized from hemin (Sigma) dissolved in methanol (2 mg/ml) by the addition of 1 molar equiv of acetic acid under stirring. After overnight incubation at 70 °C, the precipitate was washed with distilled water and unreacted hemin was removed by extracting the precipitate twice for 2 h in 0.1 M sodium bicarbonate buffer at pH 9.1. The final product was washed twice with distilled water and desiccated.

Raman imaging/spectroscopy data collection

Raman spectra and images were collected using an inVia Raman system (Renishaw, Wotton-under-Edge, UK). The laser (514.5 nm argon-ion laser, LaserPhysics, West Jordan, UT, USA) was focused on the sample by a ×100 objective (0.85 numerical aperture).

RR maps were acquired by moving the sample in steps of 1 μm with a ProScanTMII motorized stage (Prior, Cambridge, UK) coupled to the microscope to sequentially illuminate with the laser a grid of points, while collecting a Raman spectrum (850–1,700 cm-1) for each point under control of the Renishaw software program Wire 2.0. The exposure time to the CCD detector for each point was 30 s, and the laser power at the sample was 1 mW throughout the measurements. The CytoSpecTM software package (http://www.cytospec.com) was used for data preprocessing and analysis.

Spectra of β-hematin and hemoglobin were acquired from small quantities of the compounds deposited on glass microscope slides, using a laser power at the sample of 1 mW and exposure times of 300 and 100 s, respectively.

Data preprocessing and hierarchical cluster analysis

All sets of spectra were preprocessed before HCA [18]. Preprocessing involved the elimination of spikes generated by cosmic rays and the vector normalization of spectra (see the electronic supplementary material).

HCA was performed on the spectral region 1,300–1,650 cm-1 (where more intense and characteristic bands are present) by applying Euclidean distances to calculate the dissimilarity coefficients and Ward’s algorithm as a clustering algorithm to separate Raman spectra into clusters [18, 20]. As results of HCA, average spectra, dendrograms, and maps colored according to cluster membership were automatically produced with the CytoSpec software package. The average spectra were calculated using the original unnormalized data on the basis of the HCA.

Results and discussion

The application of CA tells us how many different kinds of spectra are observed in the RR map and how these clusters are spatially distributed over the two-dimensional map. Since Raman spectra convey information about chemical structure and composition, CA shows how different chemical species are distributed in a sample, at the same time giving information about their molecular structure through the average spectrum (or “centroid”) evaluated for each cluster. In HCA, spectra having a certain degree of similarity are merged into groups (clusters) whose average spectra constitute a new data set, and this agglomerative process is reiterated in several steps until all spectra are grouped into one cluster.

This process performed on a RR map of a sample containing an erythrocyte parasitized with P. falciparum at the trophozoite stage is outlined schematically in a “dendrogram” (Fig. 1), a treelike chart in which the number of branches represents the number of clusters. Depending on the number of clusters considered, different features can be detected in the sample: the erythrocytes are simply separated from the background in the two-cluster map, whereas in the five-cluster map both the erythrocytes and the background are differentiated into several areas. Assuming that a HCA map should convey as much useful chemical and structural information as possible, one should choose the number of clusters to achieve the maximum number of clusters having clearly different average spectra. This criterion allows the mapping of the maximum number of chemical species having distinct Raman spectra. In Fig. 1, the most appropriate choice appears to be the four-cluster map, since in the five-cluster map the average spectra of clusters 4 and 5 are alike and can be identified with absence of signal (i.e., noise or background). In the four-cluster map, clusters 2 and 3 have similar average spectra in terms of band frequencies and relative intensity, although the absolute intensities are very different, as can be inferred from the dissimilar signal-to-noise ratio. Considering that the spectra were normalized in the preprocessing step (see “Materials and methods”), the fact that HCA distinguishes between intense and weak spectra is somewhat unexpected. Although the clustering process itself is unsupervised, the choice of the number of clusters to be represented in an image is subjective, and it is usually made by the researcher on the basis of her/his judgement of the average spectra in relation to her/his knowledge of the system studied.
https://static-content.springer.com/image/art%3A10.1007%2Fs00216-008-2414-0/MediaObjects/216_2008_2414_Fig1_HTML.gif
Fig. 1

Outline of the hierarchical cluster analysis (HCA) applied to a set of resonance Raman spectra from parasitized erythrocytes. The dendrogram summarizing the clustering process, HCA maps for various numbers of clusters (two to five clusters), and average spectra for each cluster are shown together with the original bright-field microscope image of the sample

In RR maps of parasitized erythrocytes, hemoglobin and hemozoin are the most abundant heme species which are expected to be observed when using 514.5-nm excitation, since they absorb light at that wavelength [8]. The spectral features (position and relative intensities of the bands) of average spectra of cluster 1 (red) and cluster 2 (blue) in the three-, four-, and five-cluster maps of Fig. 1 are present in the spectra of all the infected erythrocytes studied and indeed are to be attributed to hemozoin and hemoglobin, respectively. This interpretation is based on the comparison of the average spectra with the spectra of purified β-hematin and hemoglobin (Fig. 2), and is in agreement with spectra reported in the literature [12, 21, 22]. The blue cluster average spectrum (Fig. 2, spectrum c) closely resembles the hemoglobin RR spectrum (Fig. 2, spectrum d), and the intense bands at 1,637, 1,584, 1,557, and 1,371 cm-1 are readily assigned to the ν10, ν19, ν11, and ν4 vibrational modes for the six-coordinated low-spin oxygenated heme iron [9, 21]. On the other hand, average spectrum of the red cluster (Fig. 2, spectrum a) differs from the that of the blue one, and has frequencies, relative intensities, and background matching those of β-hematin (Fig. 2, spectrum b). The bands around 1,630, 1,570, 1,550, and 1,370 cm-1, corresponding to the ν10, ν19, ν11, and ν4 normal modes of vibration for a five-coordinated high-spin Fe(III)PPIX [9], are detected in both red cluster and β-hematin spectra. β-Hematin consists of dimers of five-coordinated high-spin Fe(III)PPIX in which the two heme units are bonded via reciprocal iron–carboxylate interactions; in the crystal, these dimers are linked together in chains by hydrogen bonds between carboxylate groups [23]. Such a structure is thought to be identical to that of hemozoin [23, 24], and the RR spectra of the two compounds were reported to be very similar for many excitation wavelengths.
https://static-content.springer.com/image/art%3A10.1007%2Fs00216-008-2414-0/MediaObjects/216_2008_2414_Fig2_HTML.gif
Fig. 2

Resonance Raman spectra of a hemozoin inside a parasitized erythrocyte, b β-hematin (powder), c hemoglobin inside a parasitized erythrocyte, and d hemoglobin (human, lyophilized powder). The most intense bands are labeled as in [9]

The assignment of red and blue clusters to hemozoin and hemoglobin, respectively, is consistent with the spatial distribution of these clusters in the maps. In fact, hemoglobin is the most abundant species in erythrocytes and is assumed to be homogeneously distributed throughout the cell. On the other hand, hemozoin is localized in the food vacuole of the parasite, and therefore it is concentrated in one specific region inside the cell. According to this interpretation, the erythrocyte in Fig. 1 might be infected by two parasites, a feature characteristic of P. falciparum, since hemozoin is clearly observed in two separated areas. Since in Raman spectroscopy the intensity is roughly proportional to the concentration, the green cluster might correspond to areas of the sample containing low quantities of hemoglobin. This assignment is consistent with the fact that in Fig. 1 most of the green pixels are adjacent to blue pixels. They can indicate a transition from the erythrocyte to the background, as they are bordering the upper edge of the cell.

This rationale has been used to interpret the HCA maps of other erythrocytes reported in Fig. 3, and indeed the results are consistent with the interpretation proposed. Maps a and b in Fig. 3 of normal unparasitized erythrocytes are better represented considering only three clusters, since hemozoin is absent. In these maps, the cells are filled with a homogeneous distribution of hemoglobin (blue), and the cluster representing a lower content of hemoglobin (green) is clearly visible at cells borders. Hemozoin (red) is detected in all the parasitized cells, and is often surrounded by or close to areas (as in maps c–e and g) that very likely correspond to inner areas of the erythrocyte containing less hemoglobin because they are occupied by the parasite cytoplasm (green). This heterogeneity in the intracellular distribution of hemoglobin is seen at different levels in most of the infected erythrocytes analyzed, and exceptions (e.g., Fig. 3 maps f and h) could be explained with a particular orientation of the parasite cytoplasm, which could be hidden by the strong hemozoin signal.
https://static-content.springer.com/image/art%3A10.1007%2Fs00216-008-2414-0/MediaObjects/216_2008_2414_Fig3_HTML.gif
Fig. 3

Bright-field microscope images (left) of human erythrocytes not parasitized (a, b) or parasitized (ch) with Plasmodium falciparum, corresponding HCA maps (right) showing three (a, b) or four clusters (ch) and average spectra. Red hemozoin, blue hemoglobin (strong), green hemoglobin (weak), gray background

Besides Raman mapping, images can be obtained by using the unfocused laser to illuminate at once a large sample area, acquiring at the same time on a CCD camera the image formed by the Raman photons scattered from the sample (global Raman imaging [11]). Different from mapping, in global Raman imaging no spectra are collected: the image is obtained directly from the sample with the use of filters that let only the scattered Raman light of a specific frequency reach the CCD (similarly to fluorescence microscopy). However, global Raman imaging requires the species to be imaged to have an intense, spectrally isolated band, which should be characteristic of that species only. In fact, images obtained with this approach usually show the distribution of only one species at a time (i.e., in one image). The same limitation occurs in images showing the distribution of Raman intensity at a certain wavenumber which are built using Raman maps (intensity mapping [10]). Moreover, both global Raman imaging and intensity mapping may have problems in separating the contributions from other species which have bands overlapping with that of the species to be imaged (for an example of this limitation in intensity mapping, see the electronic supplementary material). Since HCA relies on differences in a whole spectral region, and not only at one wavenumber, it discriminates between different species better than global imaging and intensity mapping, and allows the spatial distribution of more than one species to be simultaneously shown in one single image. Moreover, spectral averaging in HCA-processed maps allows one to obtain spectra with reasonable quality even if spectra from single points have poor signal-to-noise ratio. It should be stressed that this averaging operation in HCA-processed maps, in which spatial resolution is unaffected, is different from the averaging between adjacent data points or pixels (i.e., binning) in which spatial resolution is lost.

In summary, RR mapping and HCA has been proven to be a useful tool in malaria research, and since polarization-resolved RR spectra of hematin are altered upon interaction with chloroquine [22], in the near future HCA maps based on polarized RR measurements could produce images showing the distribution of drugs in drug-treated infected erythrocytes as further experimental evidence for the direct drug–heme interaction.

Acknowledgements

V.S. and A.B. acknowledge partial financial support from BINASP/INFRAEUR and IRCCS Burlo Garofolo, respectively. The work by D.T. and S.P. was generated in the context of the AntiMal project, funded under the 6th Framework Programme of the European Community (contract no. IP-018834). S.F. is a PhD student of the AntiMal International PhD Programme run in collaboration with the European Molecular Biology Laboratory (Heidelberg). The authors thank Nicoletta Basilico and Diego Monti for useful suggestions and critical review of the data.

Supplementary material

216_2008_2414_MOESM1_ESM.pdf (101 kb)
ESM(PDF 100 kb)

Copyright information

© Springer-Verlag 2008