All methods discussed in this paper probe intrinsic molecular vibrational frequencies of bonds between molecules present in a sample; these absorption bands are displayed as a function of wavenumber [cm-1].
A typical (vibrational) spectrum of a biological sample shows several absorbance bands associated with the main macromolecules found in most biological materials, namely amide bands from proteins, membrane lipids (CHx stretching and deformation), nucleic acids (phosphate stretches), carbohydrates (CO stretching) and phospholipids (Table 2).
Tab. 2 Most common IR bands found in the FTIR spectra of RBCs and P. falciparum-infected RBCs [25-27]
Band
|
Assignment
|
Major contributions
|
2953
|
νasCH3
|
Lipids, Proteins
|
2920
|
νasCH2
|
Lipids
|
2864
|
νsymCH3
|
Lipids, Proteins
|
2852
|
νsymCH2
|
Lipids
|
1742
|
νC=O ester carbonyl
|
Lipids
|
1704-1710
|
νC=O
|
Hz (hemozoin)
|
1660-1640
|
Amide I
|
Proteins
|
1545-1525
|
Amide II
|
Proteins
|
1452
|
|
Lipids
|
1391
|
νCOO2 of fatty acids and amino acid
side chains
|
Lipids/
Proteins
|
1310−1250
|
amide III
|
Proteins
|
1244
|
νas PO2
|
RNA
|
1207- 1220
|
νC−O
|
Hz (hemozoin)
|
960−1200
|
νC−O, νC−C, νP−O
|
DNA, saccharides, lipids
|
The biochemical composition (and therefore molecular signatures that are extracted by spectroscopic techniques) of P. falciparum-infected RBCs (iRBC) are bound to be different as a result of parasite development (for example, synthesis of Plasmodium specific proteins, haemoglobin degradation, peptide and/or amino acid transportation, heme biocrystalization, etc [12]) and/or arising from metabolism (such as iron deficiency, inflammation [28]). Hence, some differences within collected spectra for iRBC in comparison to the spectra collected for healthy cells should be expected. These differences can be manifested in the shape changes such as occurrence of shoulder features in the existing bands and positional shifts of the main absorbance bands in the collected spectra from much localized areas.
In our analysis, infected cells appeared as heterogenous objects making it important to scan multiple areas within each single cell, to check if the existence of parasite “is felt” over the whole cell to the same extent, before any conclusion is drawn.
In order to comparatively present the advantages and limitations of the methods adopted in this work, results from experiments performed for the same infected cell were considered.
First, it is necessary to stress that parasite-infected cells are in the minority (our experiment used samples containing ~ 5% infected cells) as can be seen in the representative image (Fig. 1), taken by Leica Optical Microscope with 40x magnification. The diameter of the cells ranges between 8 to 10 µm.
As mentioned earlier, FTIR microspectroscopy was carried out using FPA detector (128 pixel by 128 pixel); a single pixel element defines a spot size of 2.7 µm2, though the actual achievable spatial resolution is limited by the wavelength dependent Rayleigh Criterion, which with the microscope configuration used in this study and for the main lipids peaks is about ~5 microns and up to ~15 microns of spatial resolution for the saccharide bands at around 1000 cm-1. A full field of view (345 x 345 µm2) is analysed in one measurement, as this is the dimension of the full field of view with 15x magnification objective attached to the microscope, it cannot be changed. Fig. 2a depicts the screenshot from the OPUS software (provided with the Bruker instrument); as can be seen, the area selected for experiment (red square) contains not only RBCs, but also empty spaces. Essentially, the data cube of 16384 spectra includes both the biochemistry of the cell and the substrate. A potentially more efficient and higher spatial resolution method to analyse samples that are quite spatially sparsely distributed, might be more “single point” techniques such as O-PTIR and AFM-IR spectroscopies, where analyses can be performed on single cells, deliberately chosen for analysis. Fig. 2c presents an optical image of the sample with the cell of interest in the centre taken with O-PTIR under high magnification 40x. Finding the cell of interest was a simple and direct procedure for that method. For AFM-IR a cantilever is used to perform analysis and measurements can be done on single cell, but localizing it is not straightforward as the instrument field of view is more limited. Even if the characteristic region of the sample is found (as it is shown in Fig. 2e) it is challenging to ensure that the cell of interest would be analysed as it is hidden under the cantilever shadow. Therefore, generating an initial AFM image of the sampling area is a pre-requisite for capturing and confirming the location of the selected cell (Fig. 2f).
In order to select the FTIR spectra collected for chosen iRBC, digital zooming is needed. It enables display of the position of single detectors of the FPA, as can be seen in Fig 2 b. The cell of interest is covered by approximately 9 elements of the FPA, some of them extending beyond the cell area into the substrate. As stated in the introduction, O-PTIR and AFM-IR spectroscopies enable analysis of area much smaller than 2.7 µm2.
As a direct result of the sub-diffraction limited infrared techniques of O-PTIR and AFM-IR, spatially resolved analysis reveals localised chemistry from the same infected cells in far greater detail than the nine (9) FPA-FTIR pixels could. Instead of one FTIR spectrum recorded from the spot of up to ~15 microns, around several hundred spectra can be obtained by means of O-PTIR spectroscopy and thousands of spectra for AFM-IR.
It can be argued whether such detailed analysis is needed; the answer depends on the scale of biochemical spatial heterogeneity of the analyzed sample. Fig. 3 depicts the comparison of spectra collected during experiments by means of FTIR microspectroscopy, O-PTIR and AFM-IR.
As previously discussed, the process of data collection varies significantly among these methods, however their resultant spectra can be used to gain information about biochemical constituents of the cells.
FTIR microspectroscopy is represented by nine spectra, which completely span the area of the whole iRBC; while for O-PTIR and AFM-IR six spectra have been shown, they were taken from the points marked in Fig. 2c and 2f. The points for analysis were randomly selected, the distance between two points lies in the range of 400 - 900 nm. O-PTIR spectra were collected from areas of 500 nm2, whereas AFM-IR spectra were collected from even smaller areas approximately 40 nm2.
Looking at the spectra collected for iRBC (Fig. 3), it can be seen that the most diverse are spectra obtained by means of either O-PTIR or AFM- IR spectroscopies. Band intensities, their ratios and even shapes and shifts in the positions are clearly visible and highly connected to local biochemical profile.
A noteworthy observation is that all nine FTIR spectra are consistent, with positions and ratios for amide I and II bands are almost the same. This finding is expected as in our work, a FPA detector (128 by 128) and 15x objective were used, so the spatial resolution of that system ranges from 9 to 15 µm across the fingerprint region. Pretty much all the spectra from the cell should look similar from averaging a large sampling volume within the cell. Therefore, in this case, the lack of obvious differences indicates that any unique information about biochemical characteristics are attenuated by averaging with neighbouring matters.
To better illustrate the differences, if any, among the collected spectra, hierarchical cluster analysis (HCA) has been performed. Prior to HCA, all spectra were normalized (min-max normalization), otherwise it would be difficult or even impossible to calculate the differences between each pair of the spectra.
As can be seen, the dendrogram in Fig. 4 clearly shows a hierarchical relationship between the spectra that were chosen for comparison. Spectra were allocated by the algorithm into three groups according to the type of experiments they were part of. FTIR spectra are linked together at very low height that implicates high correlation (or similarity). For O-PTIR and AFM-IR spectra prominent differences are noticed as representative spectra are joined together further apart in comparison to FTIR spectra.
In vibrational spectroscopic measurements, spectra are the most basic yet most informative component for studying biochemical changes in samples. Apart from spectral information, many scientists rely also on imaging. Spectroscopic imaging does not require usage of additional dyes; different chemical components are visualized based on inherent molecular vibrations arising from them.
In case of FTIR microspectroscopy, a single experiment using a FPA detector can generate a hyperspectral object with 16,384 (array of 128 by 128) single spectra. Each spectrum contains information about all absorbance bands present in the sample. By calculating the area under selected bands (for example corresponding to proteins Amide I) image of the particular infrared absorption in space is created. This methodology is quite powerful, taking into account that one experiment can generate a complete range of spectra, which can be later used to obtain spectroscopic images for any selected wavenumber (simply by performing mathematical calculations). Indeed, the speed of measurements could be regarded as the significant asset if only spatial resolution was not so important.
In the case of O-PTIR and AFM-IR spectroscopies, imaging is slower comparing to FTIR microspectroscopy. Due to characteristics of the IR source (QCL), single wavenumber images (also known as discrete frequency images) are captured in the form of 2D area; their size can be set by the analyst, which is very convenient as only the area of the interest is analyzed. Every single point of the image has a unique value corresponding to O-PTIR amplitude or tip oscillation amplitude (in the case of AFM-IR), proportional to the absorbance for the particular wavenumber and the quantity of the material at this spot. In order to obtain a chemical image for another wavenumber, QCL must be tuned to the new wavenumber and the whole procedure must be repeated. Software provided by the manufactures allows the analyst to collect automatically different images at different wavenumbers as a sequence.
Fig. 5 depicts the distributions of Amide I, obtained during analysis of the same cell by means of FTIR microspectroscopy (a), O-PTIR (b) and AFM-IR (c). While we are aware that such a comparison is approximated and simplistic – in the case of FTIR microspectroscopy area under bands of Amide I is presented whereas for O-PTIR and AFM-IR discrete frequency images for 1660 cm-1 are taken into account, this representation is meaningful for the purpose of our work.
FTIR microspectroscopy provides an excellent way to visualize the spatial distribution of cellular constituents (lipids, phospholipids, DNA, etc) [15, 16, 23]. However, looking at the distribution of Amide I from the FTIR measurement presented in Fig. 5 a, it seems that the image is just blurred. Even if the area around the cell of interest can be digitally zoomed in, (Fig. 5 b) it will not help to better visualize the Amide I distribution as it is limited by diffraction limited spatial resolution of the method. The wavelength dependence on spatial resolution can provide some improvement in distribution quality, but only for higher wavenumber values (3000-2800 cm-1).
In the case of O-PTIR analysis, the significantly enhanced spatial resolution manifests its presence in an image showing a heterogeneous single cell structure (Fig. 5c). More details are visualized even better in image obtained through AFM-IR analysis (Fig. 5d), owing to the nanoscale resolution of the method. It is worth noting that the protein distribution looks quite different between images obtained by means of O-PTIR and AFM-IR. This fact can be explained not only by differences in spatial resolution, but also by the way of performing experiments: AFM-IR is more surface specific method, whereas O-PTIR gets signals also from deeper layers of the cell.
Of course, the interpretation of single frequency images can be complicated due to the contribution of thickness variations over the red blood cell at various sampling locations. To see the true molecular contrast, usually the ratio of single frequency images is calculated to normalise for sample thickness. For this article, we focus on comparing the outputs from these three analytical methods. Presenting outcomes of further analysis (spectral differences discussion and their interpretation as well as band ratios calculation) exceeds the scope of this contribution and would be published separately.