Mapping blood biochemistry by Raman spectroscopy at the cellular level

We report how Raman difference imaging provides insight on cellular biochemistry in vivo as a function of sub-cellular dimensions and the cellular environment. We show that this approach offers a sensitive diagnostic to address blood biochemistry at the cellular level. We examine Raman microscopic images of the distribution of the different hemoglobins in both healthy (discocyte) and unhealthy (echinocyte) blood cells and interpret these images using pre-calculated, accurate pre-resonant Raman tensors for scattering intensities specific to hemoglobins. These tensors are developed from theoretical calculations of models of the oxy, deoxy and met forms of heme benchmarked against the experimental visible spectra of the corresponding hemoglobins. The calculations also enable assignments of the electronic transitions responsible for the colour of blood: these are mainly ligand to metal charge transfer transitions.


ELECTRONIC PROPERTIES OF DEOXY-HB
Here, we compare experimental optical absorption (left) and circular dichroism (right) spectra of deoxy-Hb with calculated spectra for deoxy-heme we extracted from 2HHB entry as reported in the Structural Bioinformatics Protein Data Bank using the two indicated theory levels. Using UB3LYP: 6-311++g(d,p)/LANL2DZ we optimize structure and calculate optical absorption and circular dichroism. Then, we repeat calculations of optical properties but using SDD:D95 theory for the structures optimized with UB3LYP: 6-311++g(d,p)/LANL2DZ. The main purpose of application of SDD:D95 is to see if this high level theory would provide confirmation that the dominant resonances identified using the lower level of theory would not shift significantly. From the data presented it can be seen that both theories provide comparable optical absorption spectra. Circular dichroism calculated using UB3LYP: 6-311++g(d,p)/LANL2DZ resembles the experimental result reasonably well save the amplitudes of the 7 and 10 th resonances are slightly overestimated. These could be accounted for if we would adopt spectrally broader line-shapes for these resonances upon spectral convolution. In Fig. S1 we show spectral dispersions using the same line-shapes, though for the same frequency fluctuation dynamics, a line-shape in the lower frequency range is expected to be broader. Amplitudes of circular dichroism transitions using SDD:D95 do not resemble the experimental.
Here, we see that changing the level of theory for the structure optimized under UB3LYP: 6-311++g(d,p)/LANL2DZ does not help to predict the relative components of electric and magnetic transitions. The contribution of the NTO pair 14 may be assigned to describe the colour of deoxy hemoglobin in blood. Here it is important to stress that d-d transitions of the iron do not play a significant role in the transition. In Fig. S2 we present images of the selected NTO pairs for this structural case. We can clearly see that ligand electrons and a small contribution of ligand-to-metal charge transfer of resonance 14 determines the colour of blood in veins. Further, we list the main localized  and β electronic contributions in the imaged pairs. In a case  and β contribute into a pair, the image includes both contributions.

Fig. S1
. Optical spectroscopy in the visible spectral range and results of TD-DFT studies for deoxy-Hb. Upper panels: optical absorption (left) and circular dichroism (right) spectra experimental and calculated for deoxy-heme we extracted from 2HHB entry as reported on the Structural Bioinformatics Protein Data Bank using UB3LYP: 6-311++g(d,p)/LANL2DZ and SDD:D95 theory level. Numbers indicate NTO pairs. Lower panels: Experimental (coloured lines) and theoretical (black lines) optical absorption and circular dichroism spectra for the system at SDD:D95 theory; NTO pairs that account for the main spectral features in the visible spectral range. Here for simplicity, we sum contributions of α and β electrons. Electronic properties of deoxy-heme. NTO pairs for deoxy-heme using UB3LYP: 6-311++g(d,p)/LANL2DZ. Pairs 1-4 contribute to optical absorption at 6805, 5381, 1312 and 747 nm. Pairs 11, 14 and 16 contribute to resonances at 575, 551 and 489 nm that determine optical absorption in the visible spectral range. In a case where  and β electrons participate in the optical transition, we sum their contributions properly scaled for simplicity in visual content. However, in the text, below, we quantify the contributions of  and β electrons.

ELECTRONIC PROPERTIES OF OXY-HB
Here, we compare experimental optical absorption (left) and circular dichroism (right) spectra of oxy-Hb with calculated spectra for oxy-heme structure we extracted from 1GZX entry as reported in the Structural Bioinformatics Protein Data Bank using the indicated theory levels. Using UB3LYP: 6-311++g(d,p)/LANL2DZ we optimize structures of the singlet and of the biradical, and calculate optical absorption and circular dichroism spectra. Here, finding a stable wavefunction for the biradical provides additional energy optimization of 15.9 kcal/mol. Then, we repeat calculations of optical properties but using SDD:D95 level of theory for the structures optimized with UB3LYP: 6-311++g(d,p)/LANL2DZ. The main purpose of application of SDD:D95 is to see if this higher level theory would confirm that the dominant resonances would not shift significantly.

Fig. S3.
Optical spectroscopy in the visible spectral range and results of TD-DFT studies for oxy-Hb. Optical absorption (left) and circular dichroism (right) spectra; experimental and calculated for oxy-heme we extracted from 1GZX entry as reported in the Structural Bioinformatics Protein Data Bank using UB3LYP: 6-311++g(d,p)/LANL2DZ and SDD:D95 theory level. Numbers indicate NTO pairs.
To explore how possible structural variance may affect the electronic properties of oxy-Hb, in Fig. S4 we compare experimental optical absorption (left) and circular dichroism (right) spectra of oxy-Hb (i) where the oxygen molecule points toward histidine to form a hydrogen bond (Pauling geometry); (ii) the Pauling geometry biradical case; (iii) the Pauling geometry where Optical absorption (left) and circular dichroism (right) spectra experimental and calculated for oxy-heme under the Pauling geometry, biradical under the Pauling geometry, the Pauling geometry case where the histidine next to the oxygen molecule is double protonated using UB3LYP: 6-311++g(d,p)/LANL2DZ. At the bottom we present optical spectra calculated under SDD:D95 theory using the structure of the Pauling geometry case where the histidine next to the oxygen molecule is double protonated as optimized under UB3LYP: 6-311++g(d,p)/LANL2DZ. Numbers indicate NTO pairs. Lower panels: Experimental (coloured lines) and theoretical (black lines) optical absorption and circular dichroism spectra for the biradical system at SDD:D95 theory; NTO pairs that account for the main spectral features in the visible spectral range. Here for simplicity, we sum contributions of α and β electrons. Electronic properties of oxy-heme. NTO pairs for the biradical oxy-heme structure extracted from 1GZX.pdb using UB3LYP: 6-311++g(d,p)/LANL2DZ. Pairs 1-4 contribute to optical absorption at 6805, 5381, 1312 and 747 nm. Pairs 11, 14 and 16 with resonances at 575, 551 and 489 nm determine optical absorption in the visible spectral range. In a case where  and β electrons participate in the optical transition, we sum their contributions properly scaled for simplicity in visual content. However, in the text, below, we quantify the contributions of  and β electrons. the histidine next to the oxygen is double protonated. Here, finding a stable wavefunction for the biradical provides additional energy optimization of 14.8 kcal/mol. These three systems were optimized and computed using UB3LYP: 6-311++g(d,p)/LANL2DZ theory. Additionally, we conducted SDD:D95 theory computations of optical properties for the third structural case optimized under UB3LYP: 6-311++g(d,p)/LANL2DZ.
Comparing theoretical and experimental spectra in Fig. S3 and S4 we can clearly state that theoretical results for the biradical case (spectra in the lower panels in Fig. S3) computed either with UB3LYP: 6-311++g(d,p)/LANL2DZ or SDD:D95 theory show good agreement for the optical absorption and circular dichroism. Considering that (a) the structure is very close to that extracted from the X-ray reported 1GZX.pdb, (b) the energy of the bi-radical according to the stability of the wavefunction test, has the lowest value, (c) both theory levels show good agreement with the experimental, we can conclude that the NTO pairs 17 and 19 are the ones to describe the colour of oxy-Hb in blood in the visible spectral range. UB3LYP: 6-311++g(d,p)/LANL2DZ theory predicts the pairs to provide optical absorption at 522.6 and 513.1 nm, respectively: see Fig.  S3. It is important to stress that d-d transitions of the iron do not play any significant role in both transitions.
In Fig. S5 we present selected NTO pairs for this structural case. We can clearly see that resonance 17 is dominated by the ligand electrons with a small contribution of ligand-to-metal charge transfer. At the same time, ligand-to-metal charge transfer dominates in pair 20 which is computed to provide absorption at 503.2 nm: see Fig. S3.

ELECTRONIC PROPERTIES OF MET-HB
Here, we compare experimental optical absorption (left) and circular dichroism (right) spectra of met-Hb with calculated spectra for deoxy-heme we prepared adding ligand water next to iron in Fig. S6. Optical spectroscopy in the visible spectral range and results of TD-DFT studies for met-Hb. Upper panels: Optical absorption (left) and circular dichroism (right) spectra, experimental and calculated for met-heme using UB3LYP: 6-311++g(d,p)/LANL2DZ and SDD:D95 theory levels. Numbers indicate NTO pairs. Lower panels: Experimental (coloured lines) and theoretical (black lines) optical absorption and circular dichroism spectra for the system at SDD:D95 theory; NTO pairs that account for the main spectral features in the visible spectral range. Here for simplicity, we sum contributions of α and β electrons. the structure extracted from 2HHB.pdb. Using UB3LYP: 6-311++g(d,p)/LANL2DZ we optimize structure and calculate optical absorption and circular dichroism spectra. Then, we repeat calculations of optical properties but using SDD:D95 theory for the structures optimized with UB3LYP: 6-311++g(d,p)/LANL2DZ.

Fig. S7.
Electronic properties of met-heme. NTO pairs for met-heme using UB3LYP: 6-311++g(d,p)/LANL2DZ theory. Pairs 1-5 are to contribute optical absorption at 1582, 1522, 1456, 1404and 1193 with resonances at 693, 603, 577, 525 and 522 nm to determine optical absorption of the met-Hb in the visible spectral range. In a case if  and β electrons participate in the optical transition, we sum their contributions properly scaled for simplicity in visual content. However, in the text, below, we quantify the contributions of  and β electrons.
The calculated optical properties using UB3LYP: 6-311++g(d,p)/LANL2DZ resemble the experimental dispersions, save the main resonances in the visible spectral range demonstrate a relatively broader spread. We may suggest the calculated resonance of the 7 th , 12 th , 20 th , 28 th and 30 th NTO pairs to represent the experimentally observed resonances at 650, 575, 540 and 500 nm, respectively. Such assignment is tentative because the level of theory used may not be high enough to describe well the nature of the numerous and dense electronic states that theory predicts for met-Hb in the visible spectral range. To confirm this we recalculated optical dispersions applying SDD:D95 theory level for the structure optimized using UB3LYP: 6-311++g(d,p)/LANL2DZ.

RAMAN MICROSCOPY STUDIES
Motivated and instructed by the outcome of the comparative (experimental and theoretical) studies of resonant Raman responses of different forms of hemoglobin in solution and of model heme systems, here we approach the sampling of biochemistry in vivo under microscopy resolution. To do this, first, in Fig. S31 we select such spectra sampled in different spatial regions of single discocyte and echinocyte cells such that comparatively the differences would help to "contrast" resonances which we suggest to be specific to met, oxy and deoxy forms of hemoglobin according to the results of the preceding studies. Accordingly, in Fig. S32, S33 and S34 we plot the differences of Raman maps reconstructed using Raman scattering intensities to explore how consistent (or not) the patterns would express for different resonances under the initial assignment.
Here it is important to stress that we reconstruct images at resonances of interests without assistance of linear algebra eigen-component data sorting. In our studies we take Raman scattering at the resonances of interests as detected. At the same time we tested extraction of intensities using deconvolution. Due to the relatively high efficiency and adequate spectral resolution (down to 1 cm -1 ) of the detected resonant Raman responses from red blood cells, both approaches yield analogous results.
In Fig. S32 we compare Raman difference maps (we subtract images reconstructed for resonances considered specific to oxy-Hb from such of deoxy-Hb) using specific resonances as indicated in the caption. There is a general consistency, which, however, is expected to be possibly compromised since we sample from living dynamic cells where different complexities are expected. In particular, it is reported that under oxidative stress conditions, different forms of hemoglobin would associate with membrane Band 3 complex with different efficiency and the conversion rate between the main forms of hemoglobin would depend on how proximal they are distributed next to the membrane (S15,S16). Hence we may anticipate possible anisotropy in distributions of different forms of hemoglobin in dependence on 1) where we would sample the signal, 2) what would be the local shape of the membrane envelope, 3) how stressful or pathologic a selected cell would "feel". In this respect, mathematical modelling of possible hidden anisotropy in typical red blood cell envelopes as we conducted recently (S17) is promising and very possible since in the current study we extract Raman tensors.
In the main text of the article we develop a discussion to articulate certain aspects of red blood cell biochemistry. Here we would like to notice that exploring live dynamics in patterns of Raman difference maps to correlate with external environment (using microfluidic devices), in correlation to fluorescence microscopy imaging of actin networks may become a paradigm approach for fast patient blood responses at the level of cellular biochemistry what would allow a better intelligence in intensive care when fast but quality interventions are often required.

Fig. S32.
Raman microscopy studies. Top: replicas of bright field image of a healthy discocyte (left 4 images) and of a leaky echinocyte (right 4 images) for eye guide to review Raman difference images (below). (A -D), differences of Raman images IM DR1 -IM OR1 , IM DR1 -IM OR2 , IM DR1 -IM OR3 , IM DR1 -IM OR4 , respectively, specific to the selected discocyte. (E -H), IM DR2 -IM OR1 , IM DR2 -IM OR2 , IM DR2 -IM OR3 , IM DR2 -IM OR4 , respectively, specific to the discocyte. Here, IM DR1 and IM DR2 indicate contributions of images reconstructed using Raman at 1545 and 1605 cm -1 , respectively, which we adopt as signatures of deoxy-Hb. Similarly, IM OR1 , IM OR2 , IM OR3 and IM OR4 indicate contributions of images reconstructed using Raman scattering at 1377, 1399, 1588 and 1636 cm -1 , respectively, which we adopt as signatures of oxy-Hb. Panels (AA -HH) present the same content as in panels (A -H), but for the selected echinocyte.
Here, DR1 and DR2 indicate that IM DR1 and IM DR2 images were reconstructed using Raman scattering at 1545 and 1605 cm -1 , respectively, which we adopt as signatures of deoxy-Hb. Consistently IM OR1 , IM OR2 , IM OR3 and IM OR4 indicate images reconstructed using Raman scattering at 1377, 1399, 1588 and 1636 cm -1 , respectively, which we adopt as signatures of oxy-Hb. Panels (EE -HH) present the same content as in panels (E -H), but for the selected echinocyte. The black contour lines added to the images are to represent the silhouettes of the cells we extract using images at the top of the plate.  Fig. S32. Here, IM DR1 and IM DR2 indicate contributions of images reconstructed using Raman at 1545 and 1605 cm -1 , respectively, which we adopt as signatures of deoxy-Hb. Similarly, IM MR1 , IM MR2 , IM MR3 and IM MR4 indicate contributions of images reconstructed using Raman scattering at 747, 991, 1364 and 1541 cm -1 , respectively, which we adopt as signatures of met-Hb. Panels (AA -HH) present the same content as in panels (A -H), but for the selected echinocyte. respectively, specific to the selected discocyte. Here, IM OR1 , IM OR2 , IM OR3 and IM OR4 indicate contributions of images reconstructed using Raman scattering at 1377, 1399, 1588 and 1636 cm -1 , respectively, which we adopt as signatures of oxy-Hb. Similarly, IM MR1 , IM MR2 , IM MR3 and IM MR4 indicate contributions of images reconstructed using Raman scattering at 747, 991, 1364 and 1541 cm -1 , respectively, which we adopt as signatures of met-Hb. Panels (AA -PP) present the same content as in panels (A -P), but for the selected echinocyte.