Identification of Oxidative Stress in Red Blood Cells with Nanoscale Chemical Resolution by Infrared Nanospectroscopy

During their lifespan, Red blood cells (RBC), due to their inability to self-replicate, undergo an ageing degradation phenomenon. This pathway, both in vitro and in vivo, consists of a series of chemical and morphological modifications, which include deviation from the biconcave cellular shape, oxidative stress, membrane peroxidation, lipid content decrease and uncoupling of the membrane-skeleton from the lipid bilayer. Here, we use the capabilities of atomic force microscopy based infrared nanospectroscopy (AFM-IR) to study and correlate, with nanoscale resolution, the morphological and chemical modifications that occur during the natural degradation of RBCs at the subcellular level. By using the tip of an AFM to detect the photothermal expansion of RBCs, it is possible to obtain nearly two orders of magnitude higher spatial resolution IR spectra, and absorbance images than can be obtained on diffraction-limited commercial Fourier-transform Infrared (FT-IR) microscopes. Using this approach, we demonstrate that we can identify localized sites of oxidative stress and membrane peroxidation on individual RBC, before the occurrence of neat morphological changes in the cellular shape.


Introduction
Red blood cells (RBCs), or erythrocytes, play a crucial role in the transport of oxygen via the circulatory system. Mature human RBCs are flexible biconcave disks without a nucleus [1][2][3][4][5][6]. They are roughly 8 µm in diameter, 1 µm thick at the rim and hundreds of nanometres thick in the central biconcavity. Compared to the usual complexity of nucleated cells, RBCs have unique characteristics, such as a simplified architecture and metabolism, exceptional mechanical properties and elasticity that are necessary to ensure their correct functioning in vivo. Indeed, erythrocytes are required to squeeze through capillaries with diameters in the sub-micrometre range and ensure oxygen delivery to all this approach has enabled the direct measurement and correlation of the morphological, chemical and secondary/quaternary structural properties of protein and for characterizing a wide range of systems in life science at the nanoscale [25,28]. More specifically, the method has excelled in studying the properties of heterogeneous samples, such as protein micro-droplets for drug delivery [31], the process of protein aggregation and amyloid formation [25,[32][33][34][35], localization of glycogen structure with specific IR signature inside Candida albicans fungi cells [36], viruses inside infected bacteria [37], polymers inside Rhodobacter capsulatus cells [38], distribution of protein-rich material in E. Coli and human HeLa cells [39], presence of exogenous elements such as metal-carbonyl compounds in human cells, hair composition [40] and variations in the primate osteonal bone composition [41,42].
Here, we use AFM-IR nanospectroscopy to acquire nanoscale resolved IR spectra, morphology, chemical and stiffness 3-D maps of RBCs. Our aim is to correlate at the nanoscale the nanomechanical, structural and chemical properties of the sub-compartments of cells to identify the first signs of oxidative stress and membrane peroxidation. We demonstrate that infrared nanospectroscopy is capable of identifying the chemical signs of oxidative stress in RBCs, related to aging, before of the occurrence of a morphological change in their native biconcave cell structure.

Nanoscale AFM-IR of Individual RBCs
In this study, we aimed at correlating with nanoscale chemical resolution the morphological and chemical modifications that RBCs undergo during aging by exploiting the capabilities of AFM-IR. Here, the ageing path of erythrocytes has been followed by monitoring the behaviour of the most typical morphologies that indicate the progress of cell senescence: namely, biconcave, spherocytes and echynocytes. Each of these morphological phenotypes characterize a given ageing time and results from a specific protein structure and cytoskeleton-to-membrane interactions [43]. Figure 1a shows a schematic illustration of the AFM-IR setup used in this study. A tunable infrared laser is focused on the bottom of a ZnSe prism, where the beam is reflected by total internal reflection. At the interface of the reflection, an evanescent wave interacts with the sample on the substrate. If the wavenumber of the exciting laser radiation pulse matches one of the molecular vibrational energy transition levels of the sample, the light is absorbed. This absorption causes a thermal heating and expansion, which is detected with nanoscale resolution (≈10 nm) by the AFM tip in contact with the sample. At each cycle of the pulsed laser, the cantilever is kicked out of contact from the sample and it rings down at its natural resonant frequencies. The peak-to-peak amplitude of the oscillation is measured in real time by deflecting a second laser from the top surface of the cantilever. Performing a fast Fourier transform (FFT) on the raw cantilever deflection signal produces a plot of the ringdown responses, at the cantilever normal modes of vibration. The IR absorbance at each wavenumber is proportional to the peak-to-peak amplitude of the raw deflection signal or the peak amplitude, termed IR amplitude, of one of the individual ringdown frequencies in the FFT plot [25,27]. Multiple ringdown pulses can be co-added to improve the signal-to-noise ratio (S/N) before stepping to the next laser wavenumber. While acquiring morphology, IR absorption maps are obtained scanning in a raster way the AFM cantilever on the sample surface maintaining the laser wavelength fixed. Spectra are further obtained by sweeping the laser wavelength maintaining the position of the AFM cantilever fixed. and (d) cantilever contact-resonance maps of individual cells, which are monotonically related to the cell stiffness; the histogram on the right represents the distribution of values of the tip-sample contact resonance within the red (border) and blue (centre) circles within the cells. (e) After acquisition of the maps is completed, the AFM tip can be positioned (green star in panel b) with nanoscale resolution (≈20-50 nm) on the cell to acquire an IR spectrum, averaged from 5 independent spectra. (f) We focused our spectroscopic investigation in the IR lipid region (2800-3050 cm −1 ) and amide band I, related to the secondary and quaternary structural conformation of proteins.
The system enables the simultaneous measurement of conventional morphology and the mapping of chemical and mechanical properties of the samples. RBCs are ideally suited for AFM-IR studies since, when dehydrated, their height is generally smaller than 1-1.3 µm, thus allowing a regime of linearity of the IR signal as a function of the thickness of the sample and of the intensity of the laser, which was kept constant for all the measurements [29,44]. This enables the study of the chemical heterogeneity of individual cells in the acquired IR maps at fixed wavenumber. In particular, in this work, we focused our attention on the antisymmetric methylene stretching (CH 2 asymmetric) of lipids at 2930 cm −1 , which can be related to lipids biophysical properties and chemical state within the cell membrane, and in particular to lipids structure, chain length, oxidative stress and membrane peroxidation [23,[45][46][47][48]. In Figure 1b-d are shown three images of two RBCs dried onto the surface of a ZnSe substrate recorded simultaneously: (b) an AFM morphology 3-D map with the relative cross-sectional dimensions of the two cells; (c) an IR absorbance image and two cross-sections of the signal, recorded at the CH 2 antisymmetric stretching wavenumber of lipids (2930 cm −1 ), where it its evident the internal chemical heterogeneity of the cells; and (d) a contact resonance peak frequency images (indicating the relative mechanical stiffness) of the two RBCs. The sample-tip contact resonance is monotonically correlated with the stiffness of the sample, where higher contact resonance is related to a higher stiffness [31,32]. A histogram distribution of the contact resonance in the center and at the border of the cells shows that the border is significantly stiffer than the centre, as observed previously ( Figure 1d) [1,2,6]. Once the acquisition of the AFM-IR maps was completed, we placed the AFM tip on a specific location of the cells to acquire nanoscale resolved IR spectra in the ranges between 1400-1800 cm −1 , with particular focus in the Amide I (1600-1700 cm −1 , hereafter named the protein region) and C=O stretching (1700-1730 cm −1 ) vibrations, and 2800-3050 cm −1 (methylene and methyl symmetric and antisymmetric stretchings, hereafter named the lipid region) (Figure 1e). The positions of the main spectroscopic assignments used in this study are summarized in Figure 1f.

Sub-cellular Investigation of Individual RBCs
The morphology map in Figure 1b shows biconcave-shape cells, which indicates their healthy state. However, this conventional AFM topography map does not provide any insights into the biophysical properties of the cells. The IR absorbance image, on the other hand, shows a high degree of heterogeneity in the lipid-related signal ( Figure 1c). In particular, the map at 2930 cm −1 shows a higher absorption at the centre of the cells and a lower contact-resonance frequency, which indicates that the cell is softer in the centre. Thus, we aimed at investigating the chemical differences between the central and border region of individual RBC with nanoscale resolution.
In order to confirm the observed structural heterogeneity within a cell, one of the two biconcave RBCs in Figure 1 was examined in more detail and higher resolution ( Figure 2). In Figure 2a,b, we show a 3-D representation of the morphology and IR absorption map of the cell at 2930 cm −1 , corresponding to the antisymmetric methylene (CH 2 ) stretching of lipids. The height of the cell ranges within 0.8-1.2 µm (Figure 2c), thus granting the linearity of the IR absorption signal as a function of the sample thickness. The chemical map shows a significant difference in absorption between the centre and the border of the cell, which show areas with markedly lower lipid absorption (dark areas in Figure 2b). In order to evaluate the spectroscopic behaviour independently from the sample thickness, we took advantage from the linearity of the signal and calculated the ratio maps between the IR absorption maps and the corresponding morphology map, both in the CH 2 antisymmetric stretching region of lipids and the amide band II of proteins (Figure 2d,e). Both ratio maps confirm that the centre of the cell has a major IR amplitude absorption than the border, where there are dark regions corresponding to a lower IR absorption, as confirmed by a quantification of the IR signal in the absorption map at 2930 cm −1 (Figure 2f  To further investigate the biophysical and spectroscopic differences between central and border regions of the cell, we acquired nanoscale resolved spectra at 500 nm intervals along the RBC, from locations indicated by the coloured stars in the maps (Figure 2d,g). Each spectrum derives from the average, smoothing and normalisation to one of at least five spectra acquired from each position (n > 40). In good agreement with the IR absorption and the ratio maps, the spectra at the centre of the cell have higher IR signal in the lipid region in the centre of the cell than in its border ( Figure 2g). Then, we calculated the average of the spectra in the protein and lipid region and we normalised them to their maximum to compare their chemical properties. We focused our attention on the amide band I (Figure 2h) and the C-H stretching lipid region (Figure 2i). The protein region is related to the conformation of the proteins within the cell, and in particular quaternary and secondary structure conformation. While, the lipid region is intimately related to the structure and chemical state of lipids and cell membrane. The broad, intense amide I band has a peak wavenumber of 1654 cm −1 , assigned as C=O stretching vibration and related to the α-helix secondary structure of the main protein component of the cell, which is hemoglobin. In the amide band, there is also visible a clear shoulder at approximately 1625 cm −1 , which can be related to the intermolecular β-sheet content of the cytoskeleton of the cell. Remarkably, the shape of the amide I band shows that the right edge of the cell possesses a decreased content of intermolecular β-sheet secondary structure (Figure 2h). The lipid region shows methyl antisymmetric and symmetric stretching bands at 2955 and 2870 cm −1 , respectively, while the more intense antisymmetric methylene band has a peak wavenumber of 2930 cm −1 (Figure 2i). The intensity ratio of the 2930 cm −1 (CH 2 antisymmetric stretching) to 2955 cm −1 (CH 3 antisymmetric stretching) bands is different between the central and border region of the cell. In particular, the intensity of the CH 2 antisymmetric band 2930 cm −1 is significantly more intense relative to the 2955 cm −1 CH 3 band at the border of the cell. This difference indicates there are longer lipid chains at the border of the cell than in its central region [23,47,48]. As well as previously reported [45,47,48], these data demonstrate that the edge of the cell is showing evidence of oxidative stress [23,46,47]. Indeed, the outer layer of the RBC is mechanically stiffer, has a higher concentration of long chain CH 2 groups and has a lower content of intermolecular β-sheet structure, which suggests a weakening of the cell membrane-skeleton structure [21].

Comparison of Biconcave and Echinocyte
After identifying the signs of oxidative stress within biconcave RBCs, we aimed at investigating if similar changes occur in RBCs with altered morphologies. In Figure 3b, AFM-IR morphology ( Figure 3a) and chemical maps in the CH 2 antisymmetric stretching (Figure 3b, 2930 cm −1 ) for biconcave and echinocyte RBCs are shown. Within both phenotypes, the chemical mapping shows a high heterogeneity of the IR absorption at 2930 cm −1 , related to the CH 2 antisymmetric band of lipids. The biconcave cell showed a lower lipid absorption at its edge, as observed above in Figure 2; however, the echinocyte cell had internal regions of higher and lower absorption that could not be correlated with any specific pattern ( Figure 3c).
Next, we focused our attention on the echinocytes and we divided the IR signal by the morphological height in order to rule out the dependence of the signal from the cell thickness (Figure 3c). In the ratio image of the echinocyte cell, we could observe the presence of not evenly distributed dark and bright regions of IR absorption. Following the observation of a heterogeneous chemical response related to lipids, we acquire nanoscale localised IR spectra in the protein (Figure 3d) and lipid (Figure 3e) regions. The spectrum of the dark region (position 2, Figure 3d,e) showed a decreased content of intermolecular β-sheet and an increased ratio between the CH 2 and CH 3 antisymmetric bands when compared to the bright region of IR absorption (position 1, Figure 3d,e). The observed spectroscopic differences between the regions of the echinocyte cell indicated a different state of oxidation that correlates with the spectroscopic changes observed within the biconcave RBC ( Figure 2). However, for these morphologically altered cells, it was also possible to observe a change in the ratio between the less intense CH 2 and CH 3 symmetric bands and a small but significant shift at lower wavenumber of the two CH 3 bands, which can be related to a state of oxidative stress, as previously reported in literature [45,48]. These spectral differences demonstrated the presence of higher oxidative stress within the dark area of cell, as indicated by the presence of longer lipid chains and a likely degraded cell membrane-skeleton structure (Figure 3d,e). After studying at the subcellular level the echinocytic cell oxidation state, we aimed at comparing its average chemical state with the biconcave cell on the right in Figure 3a,b. We acquired several spectra at random positions on the two cells (n = 35 and n = 45 for the two cells). Then, we smoothed, averaged and normalised the spectra with respect to the amide band I (Figure 3f). At first, it was possible to observe that the ratio between lipids and protein absorption differs for the two phenotypes. Indeed, the echinocyte had a significantly lower IR signal in the C-H lipid region of absorption than the biconcave one, which might be related to the lipid loss during membrane peroxidation in aged RBCs [1,4,46,48]. Furthermore, the observation of the lipid C-H region clearly showed that the echinocyte cell is on average more oxidated than the biconcave cell ( Figure 3f). Indeed, the echinocyte cell showed, on average, an increased ratio between both the CH 2 and CH 3 symmetric and antisymmetric bands, which indicate oxidative stress and are related to the presence of longer lipid chains (Figure 3g) [23]. Remarkably, on average it was not possible to observe any significant difference between the two cells in the amide band I signal. Thus, at the average single cell level, for these two cells, it was not possible to infer any information regarding the different state of the membrane-skeleton structure (Figure 3f).

Membrane Peroxidation of RBCs
The observed similarities of the oxidative stress modifications between biconcave and echinocyte RBCs suggested that these modifications could further lead to membrane peroxidation, which is the typical hallmark of RBCs cellular aging. Figure 4a,b, shows a 3-D representation of the morphology and IR absorption maps of RBC deviating from the native biconcave shape. In particular, the two cells on the top of the images showed an expansion of their volume at their edge, while the two cells on the bottom had a spiky morphology. As observed in the previous examined cells (Figures 1-3), we observed a high heterogeneity of the IR absorption at 2930 cm −1 . In order to analyse the chemical state of the cells and in particular any possible membrane peroxidation, several spectra were acquired from different locations in the dark areas of IR absorption of biconcave and echinocyte cells (coloured stars, n = 30, in Figure 4a,b). The spectra were acquired in the protein and lipid regions and then averaged, smoothed and normalised (Figure 4c,d). The biconcave cells (position 1-3, Figure 4b) showed a spectral signature similar to that of previously analysed cells (Figures 1-3). On the contrary, the echinocyte evidenced a higher methylene to methyl intensity ratio, indicating they possess longer chain hydrocarbons and higher oxidative stress at the locations with lower IR absorption examined in these RBCs (position 4-6, Figure 4b). In particular, the spectra collected at positions 5 and 6 showed evidence of a clear signal of membrane peroxidation, as demonstrated by the appearance of the ester carbonyl stretching (C=O) band at 1734 cm −1 . Remarkably, the appearance of the ester C=O band directly correlated with a continuous change of the ratio between the intensities of both symmetric and antisymmetric methylene and methyl stretching, which are related to oxidative stress and lipid chain length. In addition, the spectra in position 4 to 6, also showed a net shift of the CH 2 bands to lower wavenumber, suggesting that the lipid chains are more ordered than in any of the other spectra. Moreover, in average, the spectra showed a lower IR signal assignable to the spectroscopic signature of intermolecular β-sheet content than the spectra collected on the biconcave cells (position 1-3). Altogether, the data in Figure 4 indicate that the early chemical modifications observed in the protein and lipid regions of spectra acquired on biconcave RBCs (Figures 1-3) can ultimately lead to membrane peroxidation and deviation from the native biconcave morphology of the cells, which is likely play a role in inducing the echinocyte morphology. In addition, the oxidative state of the cell and its membrane peroxidation directly correlate at the sub-cellular scale with a loss of intermolecular β-sheet structure. Thus, oxidative stress and membrane peroxidation could be related to a degradation of the membrane-skeleton network.

Discussion
In this work, we exploit the capabilities of AFM-IR to demonstrate the possibility of acquiring nanoscale resolved chemical and mechanical maps onto erythrocytes. This approach has enabled the localization of oxidative stress and membrane peroxidation at the sub-cellular scale in healthy biconcave cells, before any morphological change related to oxidative stress occurred. In particular, we quantified the degree of oxidative stress within an individual RBC as a function of the relative amounts and in the molecular order of long chain CH 2 groups. The difference in lipid chains lengths correlated with the nanomechanical properties of sub-compartments within an individual RBC, indeed the border of biconcave erythrocytes was stiffer and had longer lipid chains than its centre; these biophysical properties can be correlated to different composition and chemical state of lipids within the cell, such as their fluidity. Furthermore, we could correlate a higher oxidative stress to lower content of intermolecular β-sheet structure. These spectroscopic changes could be related to the membrane-skeleton integrity of the cells, through a mechanism that involves a rearrangement of the protein 4.1, which is the most important cytoskeleton protein with a significant β-sheet content in RBCs.
Furthermore, our data suggest that in biconcave erythrocytes the external edge might be the region of the cell, which is more susceptible to oxidative stress. When compared to biconcave cells, echinocytes show an even lower IR absorption in the spectroscopic region of the C-H stretching of lipids and of intermolecular β-sheet structure. The early chemical changes at the border of biconcave cells led directly to the appearance of the ester carbonyl stretching (C=O) band at 1734 cm −1 in echinocyte demonstrating, at the nanoscale, the presence of membrane peroxidation. These observations suggest that the early chemical modifications observed at the border of biconcave cells are continuously ongoing during cellular aging. In this regard, it has been previously found that cell aging and oxidative stress in RBCs correlates with general morphological changes in the cellular shape and in the membrane roughness. Since in this work we observe variation in the β-sheet content also within healthy normal shaped cells, we demonstrate that the detection of membrane peroxidation and membrane-skeleton alterations is possible before a clear morphological change of the cellular structure has occurred. These latter data are particularly interesting in the light of recent findings evidencing that the morphological changes actually occur as consequence of environmental stimuli that act by modulating the metabolic pathways with the mediation of oxidative stresses [11]. The approach developed in this work can be very promising in order to improve our understanding of RBCs viability and function, which can be very important for the development of new treatments or the screening of pathologies at early stage of development, as well as to improve the methodology of blood conservation.

RBC Sample Preparation
Blood samples were obtained from healthy donors as previously described [49]. Briefly after venipuncture the blood was immediately diluted threefold with an isotonic pH 7.35 buffer solution (140 mM NaCl, 10 mM Potassium Phosphate, containing 1 mM EDTA disodium salt as anticoagulant). Three centrifugation cycles were performed (12 min each at 3200 rpm at 4 • C) to eliminate plasma enriched supernatant and the leukocytes layer. RBCs were re-suspended and washed four times under sterile conditions in glucose-free PBS medium (potassium phosphate 4 mM, EDTA 1 mM, and NaCl 140 mM), adjusted with NaOH to pH 7.35. To avoid proteolytic degradation, a protease inhibitor (phenylmethylsulfonyl fluoride, 1 mM) was added. After purification of erythrocytes from, approximately, 2 mL of blood, the cells were re-suspended in a 20% hematocrit. However, the smears on the substrate for the infrared nanospectroscopy investigation required the addition of 3 µL of plasma to 2 µL of the RBC suspension, in order to protect the erythrocytes and avoid cell-cell adhesion during the smear. Thus, the smears were performed with a 7-8% hematocrit, which is an optimal dilution to obtain a good number of properly spaced cells across the entire substrate.

AFM-IR Measurements
Samples for AFM-IR analysis were prepared by spreading a RBC suspension diluted in plasma (ratio 1:5) onto the surface of a ZnSe prism (internal reflection element, or IRE) and allowing the cells to dry in air. A number n = 25 of red blood cells were then examined using a nanoIR™ microspectroscopy system (Anasys Instruments, Santa Barbara, CA, USA) to support the conclusions of the manuscript. We used a scan rate of 0.02 Hz per line in contact mode. A silicon AFM cantilever (AppNano, CA, USA) with a nominal tip radius of 10 nm and an elastic constant of~0.5 Nm −1 was used for all measurements. All images have a resolution of at least 500 × 100 pixels. All AFM height, IR absorbance amplitude, and contact resonance peak frequency images were first-order flattened using Analysis Studio software (Anasys Instruments, Santa Barbara, CA, USA). Statistical and cross-sectional analysis was performed by SPIP (Image metrology, Hørsholm, Denmark) and Origin Pro (OriginLab, Northampton, MA, USA) softwares.
The AFM-IR spectra were collected with a data point spacing of 2 cm −1 and 256 ringdown co-averages at each wavenumber position over the spectral range 1400-3100 cm −1 . We used an OPO pulsed laser with a pulse width of 10 ns. The laser is tunable in the range of 1000-3300 cm −1 and in this range had a power between 2-7 mW (8% of OPO laser power). The spectral resolution is determined by the natural linewidth of the laser and is about 4-8 cm −1 over this range. All AFM-IR spectra were smoothed by a Fourier Transform filter and Savitzky-Golay filter (second order, seven points) using Origin Pro software (OriginLab, Northampton, MA, USA). The peak locations were determined using a three-point centre of mass method, rounded to the nearest wavenumber, and plotted from Essential FT-IR software (Operant). The experiments were performed at room temperature.

Conflicts of Interest:
The authors declare no conflict of interest.