2.1 Patient samples

Patient serum samples were obtained from the Centre for Eye Research Ireland (CERI). Blood samples were drawn from a total of 62 patients who were voluntary participants in the European Nutrition in Glaucoma Management (ENIGMA) trial, registered at ClinicalTrials.gov, identifier NCT04460365 [36]. Recruitment was carried out at the Mater Misericordiae University Hospital and Mater Private Hospital (Dublin, Ireland), whereas study visits were conducted at the Centre for Eye Research Ireland, a dedicated academic clinical trial centre at Technological University Dublin. All necessary research ethical approval was obtained from the Mater Misericordiae Institutional Review Board and from Technological University Dublin Research Ethics and Integrity Committee [36]. Written consents were obtained from participants, and the study adhered to the tenets of the Declaration of Helsinki.

The ENIGMA study [36], was a randomised, placebo-controlled, double-masked clinical trial, designed to evaluate the Macular Pigment response of patients with open angle glaucoma (OAG) to supplementation with lutein (10 mg), zeaxanthin (2 mg), and meso-zeaxanthin (10 mg) over an 18-month period. The supplement ratio was the same as that of commercially available macular pigment supplementation, Macushield, provided by Thompson & Capper Ltd, Runcorn, UK. The dose was also deemed safe, with renal, liver, lipid, hematologic, and inflammatory biomarkers all unaffected by supplementation at these concentrations [40].

Forty two participants randomly received the carotenoid supplements, while 20 received the placebo, which was as a soft gel capsule, identical in appearance to the active supplement, but containing only sunflower oil. Blood samples were taken at the beginning of the study (baseline) and at 6-monthly intervals over 18 months. Macular pigment measurements [41] and other optometric parameters were also recorded.

The main outcome of the trial was a statistically significant increase in MPOD volume (significant time effect: F (3,111) = 89.31, mean square error (MSE) = 1656.9; p < 0.01). A 60% mean increase in MPOD volume was observed over the 18-month trial duration among those randomised to receive the macular carotenoid supplement, whereas MPOD remained relatively unchanged among those assigned to placebo.

For the present study, blood serum samples obtained at baseline (before supplementation) and at 18 months, from patients who had received the carotenoid supplements, were used for Raman spectroscopic analysis. Samples from placebo patients were not examined, as the original study reported no observed effects [36]. Blood samples were drawn and processed to obtain serum by personnel from CERI. Samples were stored at −80°C and thawed in a water bath at 37°C, after which Raman spectroscopic measurements were immediately carried out.

In total, serum samples from 40 patients of the ENIGMA study were made available for the Raman spectroscopic analysis. Baseline samples were available for 25 patients and 18 months supplemented samples available for 28 patients. Both baseline and 18-months supplemented samples were available for eight patients, and this set was used for the matched patient analysis (Section 3).

2.2 Reference sample preparation

Beta carotene and bovine serum albumin (BSA) powders were purchased from Sigma Aldrich (Arklow, Ireland). The reference samples were prepared by adding 20 μL of ultra-pure water (Millipore) to 1 g of the powder and mixed until a thick paste was formed. The pastes were measured immediately on a glass slide.

2.3 Raman spectroscopic analysis

Raman spectral measurements [41] were carried out using a Horiba Jobin-Yvon LabRam HR800 spectrometer with a 16-bit Peltier cooled CCD detector, coupled to Olympus 1X71 inverted microscope and a ×60 water immersion objective (LUMPlanF1, Olympus) was employed for serum samples. A 532 nm laser line of 12 mW was used in taking measurements with a 300 lines/mm grating throughout the study, yielding a spectral dispersion of ~1 cm⁻¹ per pixel. The spectral range employed was 400–4000 cm⁻¹ and the back scattered Raman signal was typically accumulated for 5 × 4 s. Depending on the measurement, 4–6 spectra were acquired from different spots on each sample to confirm that the response was stable and there was no discernible degradation occurring as a result of the 532 nm laser.

Raman measurements of the reference samples, water, beta carotene and BSA were measured with ×60 water immersion objective in an upright position. In order to achieve as much as possible water free spectra of the beta carotene and BSA, as references, while minimising any spectral distortions due to local crystal field effects, and so forth. In the powder form, a wet paste of the powder of known concentration was measured [10]. Serum measurements were performed by focussing the objective lens into the samples contained in a cover slip glass bottom 96-well plate (Matek), [10]. Note that the spectra were not corrected for the differing instrument responses, including source power, and responses over the different spectral ranges [42]. However, the uncorrected spectra better represent the quality of the spectral signal collected.

2.4 Data processing and analysis

All data processing and analysis was conducted using the MATLAB platform. The initial Raman spectroscopic analysis was carried out on the entire data set, including both baseline and 18 month-supplemented blood serum. Blood serum is very complex, typically containing more than 10 000 different proteins [43], as well as other components such as metabolites, peptides, sugars, and lipids [44]. Notwithstanding this, carotenoids (even though found in small quantities in the blood) dominate the spectrum of blood serum, especially at resonant or near resonant conditions [28, 29]. A typical Raman spectrum of serum taken with a 532 nm laser source therefore shows a spectral profile of mixed signals from carotenoids, water, serum albumin (as the dominant serum protein), other serum proteins, and stray light scattering [45, 46]. The background correction (BC) algorithm adapted by Kerr and Hennelly [47] was used to improve the spectra by performing background subtraction of water. It was essential to remove water because it makes up 91%–92% of serum [48], so strongly dominates the spectra of blood serum. This was done by inputting a known spectrum of water into the BC algorithm [49]. Before background correction, all spectra were smoothed using the Savitsky-Golay algorithm (polynomial order of 5, window of 9) to reduce the noise. As an extra precaution, the spectrum of glass was also subtracted from the serum spectra. The glass spectrum can contribute to the spectra in the inverted geometry, especially when using very low volume samples [47, 50]. Some of the baseline spectra showed interference from the glass peak at 950 cm⁻¹ (figure not shown). It was therefore important to remove any possible glass signal from both baseline and the supplemented serum spectra; hence it was subtracted alongside water in the BC algorithm by inputting the spectra of the empty glass 96 well plate into the algorithm. The algorithm also fits and subtracts a polynomial baseline to account for any dark and optics related background. The corrected spectra were also water normalised by the co-efficient of the subtracted water, as an internal standard, using a further modification of the method by Parachalil et al. [10, 28].

Figure S1a is the 532 nm spectrum of the glass 96 well plate used for measuring the serum samples. Glass based substrates typically show significant Raman scattering contributions in the region of ~500–1000 cm⁻¹, originating from SiO₂ vibrations [47, 50]. Figure S1b is the water spectrum, used in the BC algorithm to subtract water, which constituted a major part of the background. The spectrum of water shows characteristic peaks at ~1640 cm⁻¹ from the HOH scissors, and a broad band from ~3000–3600 cm⁻¹, from the OH symmetric and antisymmetric stretch [33].

Figure S1c is the spectra of pure beta carotene (paste) at 532 nm, which was used as reference for the BC correction. The spectrum clearly exhibits the characteristic carotenoid peaks at ~1519, ~1158 and ~ 1004 cm⁻¹ [1]. Also used as reference for the BC correction was the spectrum of BSA, in Figure S1d. BSA exhibits typical protein features such as the sharp phenyl alanine peak at ~1004 cm⁻¹, the broad Amide I band at ~1620 cm⁻¹, which can be observed in the region of 400–1800 cm⁻¹ and the broad CH band at 2900 cm⁻¹ [51].

2.5 Principal components analysis

The application of multivariate analysis methods better enables visualisation of the variance of the spectral dataset, and for example to identify clustering of the responses for individual patients. Principal Components Analysis (PCA) is a multivariate analysis technique commonly used to analyse multi-dimensional data sets by reducing the number of variables in data sets with multi dimensions without altering the major variations within the data set [52]. The order of the principal component describes the relevance to the data set. PC1 is expected to describe the highest variation in the data, followed by PC2, PC3 and so on. The first three PCs will generally provide up to 99% variance in the data set, giving the best visualisation of the differentiation in the cluster sets. [52, 53].

In this study, PCA was used to explore the differences in both carotenoid content and structure of the baseline and 18 months supplemented blood serum spectral data sets. The data used for PCA was already corrected for interferents (water and glass), noise and normalised for water content. PC1 and PC2 loadings were sufficient to show majority of the variation in the data set. The spectral loading of PC1 was used to highlight the difference in the concentration of carotenoids, while PC2 highlighted the difference in carotenoid structures [33].

The spectra are mean centred as an integral part of the PCA algorithm in Matlab, and the effect of vector normalisation in addition to the normalisation to water content on the analysis is also explored.

3.1 Raman spectroscopic analysis of baseline and 18 months supplemented patient serum

Figure 1A,C show the uncorrected Raman spectra from 28 patient baseline and 25 supplemented patient serum samples, respectively. Both figures show the same features, the Raman bands from the carotenoids dominating the fingerprint region and water dominating the high wave number region of the spectra. Carotenoids characteristically are recognised by three bands at ~1004, 1158, and 1519 cm⁻¹. These bands are clearly enhanced in the Raman spectra of serum, especially when using resonant or near resonant laser wavelengths sources [17, 21], as in the case of this study. The bands are associated with the carbon–carbon double and single bond stretches of the polyene backbone and methyl bends of the carotenoid structure. The 1519 cm⁻¹ peak is the most significant and corresponds with the v₁band, originating from vibrations of conjugated CC stretching modes of the double bonds [32, 54]. The 1158 cm⁻¹ peak corresponds with the v₂ band, originating from stretching vibrations of CC single bonds coupled with CC in-plane bending modes [31, 32]. The 1004 cm⁻¹ peak corresponds to the v₃ band, originating from C-CH₃ bonds found between the main carbon chain and the side methyl groups in the carotenoid molecule [31, 55]. Other weaker bands exist, especially in the region >2000 cm⁻¹, but are overtone and combination bands which arise from the total symmetric character of v₁, v₂ and v₃ [56, 57]. They are also mostly visible in the resonant or near resonant vibrational state [56, 57]. Notable serum features are the amide band of proteins at ~1620 cm⁻¹, while the strong, sharp characteristic band of phenyl alanine residues at ~1004 cm⁻¹ is obscured by the carotenoid peak in the same region. Serum albumin is the dominant protein in blood serum and contributes the major Raman features of blood serum [58, 59].The solubility of carotenoids in serum albumin complex has previously been demonstrated, as well as the protocol for subtraction of albumin and water [33], which greatly enhances the carotenoid features in the serum spectrum. Also clearly visible is the broad band in the region of 2900 cm⁻¹, which derives from the CH bonds of lipids and proteins [45, 60].

The spectra were improved by performing background correction, inputting known references (BSA and beta carotene) into the algorithm and subtracting the water and glass spectra (Figure S1). A polynomial of 7 was chosen as the best order for the correction and the spectra were further normalised for water [10]. Figure 1B,D shows the corrected spectra for baseline and supplemented serum, respectively. The spectral range has been restricted to 2800 cm⁻¹ to eliminate the serum band at ~2900 cm⁻¹ which appeared to overwhelm the spectra after BC of water. Therefore, the restriction was done to accentuate the region of the carotenoid bands at 1004, 1158 and 1519 cm⁻¹, which is the focus of the study. Some peaks from serum also become clearly visible, for example, the amide band at ~1620 cm⁻¹ and the collagen band at ~1448 cm⁻¹ [61]. Also, important to note is the clear difference in intensity of the carotenoid peaks in baseline and supplemented corrected serum spectra. The supplemented serum spectra show a much higher intensity in the carotenoid peaks, about twice the intensity of the baseline carotenoids.

3.2 Differential analysis of baseline and supplemented patient serum

To properly analyse the carotenoid content in the two sets of data (baseline and supplemented serum) using Raman spectroscopy, it was important to consider whether they could be differentiated. Since this analysis is carried out in patient serum, many factors come into play in differentiating the baseline and supplemented samples, such as the difference in the overall carotenoid content, as well as composition of the samples. Critically, the degree of intra and inter patient variability of the content and composition of the serum should initially be considered. PCA was therefore employed to make detailed comparison of the Raman spectra from baseline and 18 months supplemented serum, by first looking at variations within the individual data sets, then in a matched data set (of serum samples of individuals taken before and after supplementation) for a more accurate comparative analysis. PCA comparison of the full baseline and 18-month cohorts is shown in the Supplementary material (Figure S6). A further PCA comparison of two extreme patient serum sample signals, the patient with the lowest carotenoid Raman signals in both baseline and supplemented serum spectra (ET022) and the patient serum with the highest signal (ET037) is also shown in the supplementary material (Figure S7).

3.3 Baseline samples

Figure 2A is the baseline scatter plot. There is a wide spread of points across the plot from the negative to the positive side, with very loose clustering and differentiation between individual patients. The results of the PCA indicate that the spread of the points in the scatter plot is primarily attributed to the variation of PC1 (68.71%). Figure 2B,C are the PC1 loading plots for the baseline group, with and without vector normalisation, respectively. Remembering that the BC algorithm employed in this study normalises the spectra of all samples based on their water content, the fact that the loading of PC1 (unnormalised) is dominated by the carotenoid spectrum is indicative of a strong variance in the overall carotenoid content of the samples. After vector normalisation, PC1 now represents differences in the serum content itself, and distinguishes the carotenoids from serum peaks with a dominance of carotenoid peaks in the positive side and serum peaks on the negative side of the plot. This, in essence, describes the relative amounts of the carotenoids compared to the serum, although there is no obvious indication of differences in the composition of carotenoids present, which would result, for example, in a shifting of the dominant peaks [33]. The corresponding loadings of PC2 are shown in Figure 2D,E Unnormalised (Figure 2D), the loading shows a strong resemblance to the normalised PC1 loading (Figure 2C), although inverted, differentiating between the carotenoids (on the negative side) and serum peaks (on the positve side) on each side of the plot. Again, there is no strong indication of variability in the composition of carotenoids present in the baseline samples. The vector normalised PC2 loading (Figure 2E) does not show any obvious strong peaks from carotenoids or serum.

3.3.1 A 18 months supplemented samples

Figure 3A is the scatter plot which shows a spread of points across the plot from the negative to the positive side, with a greater density towards the positive side. There is a more defined clustering and differentiation between patients indicating a high interpatient variability in the Raman spectra measurements. Again, the spread of the points in the scatter plot is primarily attributed to the variation of PC1 which accounts for majority of the variance (76.69%). Figure 3B,C shows the PC1 loadings for this group, without and with vector normalisation, respectively. Without the vector normalisation, the PC1 loading plot is completely dominated by the carotenoid peaks on the positive side. Also, similar to the unnormalised PC1 loading of the baseline samples, there is no indication of the presence of serum peaks. Again similarly, when vector normalised (Figure 3C), PC1 distinguishes the relative amounts of carotenoids from serum, with a dominance of carotenoid peaks in the positive side and serum peaks on the negative side of the plot. There is also no obvious indication of variability in the composition of carotenoids present from PC1. Figure 3D,E shows the PC2 loadings without and with vector normalisation, respectively. In the unnormalised plot, the PC2 loading is similar to that of PC1 in the normalised dataset (Figure 3C), carotenoid peaks dominating the positive side, while serum peaks appear on the negative side. Notably, the 1519 cm⁻¹ carotenoid peak now shows a derivative-like profile across the positive and negative side of the loading, possibly indicating a variance in the carotenoid composition [33]. When vector normalised, the PC2 loading (Figure 3E) clearly shows a derivative line-shape coming from the 1519 cm⁻¹ peak. An additional carotenoid peak at 1158 cm⁻¹ can also be seen on the negative side of the loading. Such derivative-type line-shapes in the PC loadings have previously been observed in the comparative analysis of lutein, zeaxanthin and beta-carotene [33], and are a strong indication that the PCA is highlighting variability in the relative carotenoid composition, as well as overall content, of the supplemented patient samples. While the differences in overall carotenoid content are immediately apparent from the PC1 of the unnormalized data sets, normalising accentuates any structural differences in the carotenoid composition, particularly evident in PC2.

3.4 Matched baseline and supplemented patients' serum

Figure 4A compares the mean spectra of the baseline and supplemented serum samples, showing a higher intensity in the supplemented samples, but no obvious shifts in the peaks. However, on closer examination of the 1519 cm⁻¹ peak from the vector normalised mean spectra (Figure 4B), there is a subtle but evident shift in the supplemented samples peak from high to low frequency (1522 to 1519 cm⁻¹), which may indicate structural differences between the carotenoids in both groups of serum and suggest the potential to evaluate the relative contributions of the carotenoids in the serum groups.

The scatter plot of PC2 versus PC1 (Figure 4C) scores shows a high degree of clustering of the spectra of each patient sample (e.g., rings for patients ET001, ET0013, ET0022, and ET037) from baseline and 18-months supplemented. The plot also shows differentiation between the groups, most of the spectra from the baseline samples clustering on the negative side of PC1, while those from the 18-month supplementation samples are clustered on the positive side. The distribution of the points in the scatter plot is primarily attributed to the variation of PC1, which accounts for majority of the variance.

Figure 4D,E shows the PC1 loadings from the PCA, without and with vector normalisation, respectively. The PC1 loading without vector normalisation is similar to the PC 1 loading of the individual groups (Figures 2C and 3C), carotenoid peaks completely dominating the positive side of the loading and no diffrentiation between baseline and supplemented samples. When normalised, the PC1 loading reveals differences in the relative amounts of carotenoids compared to serum, carotenoid peaks appearing on the positive side and serum peaks on the negative side of the loading. There is much higher carotenoid content in the positive side (supplemented serum) than the negative (baseline samples). Relative serum content is also much higher in the baseline than in the supplemented samples. In the unnormalised PC2 loading (Figure 4F), a much clearer differentiation can be observed, as there is a distinction in the relative amounts of serum and carotenoids, but also indications of structural differences in the carotenoid content. In the normalised PC2 (Figure 4G), this structural difference becomes more prominent, as seen in the derivative line-shape at ~1519 cm⁻¹. This variance comes from both baseline and supplemented serum samples. Again, carotenoid content is much higher in the 18 months supplemented samples. Another major variance in carotenoid structure can be seen in the 1158 cm⁻¹ peak which comes from the baseline samples.

It is notable that the scatter plot also shows a pattern in the distribution of the groups associated with different individual patients, from those clustering around the zero mark to those on the extreme negative (baseline) and positive (supplemented) ends. Generally, the closer the points are to zero, the lower the carotenoid concentration, and vice versa. For instance, in the case of patient ET001 and ET022 (highlighted), the carotenoid baseline and supplemented levels are very low, compared to the other patients. Because of the low carotenoid levels, the baseline and 18 months supplemented points for these patients, cluster close to the zero line, with the baseline points slightly past the negative (baseline) side of the plot for patient ET022, indicative of low carotenoid levels and a smaller variance gap between baseline carotenoid and supplemented carotenoid content. On the other hand, in the case of ET037 and ET013 (highlighted), the points cluster much further away from the zero line on the negative (baseline) and positive (supplemented) side of the plot, indicative of high carotenoid content on both sides and a much larger difference in serum carotenoid content before and after the supplementation regime. Further examination of these extreme cases, showing their relative carotenoid compositions in the serum before and after the period of supplementation is shown in the supplementary material (Figure S7).

EXTERNALLY HOSTED SUPPLEMENTARY FILE

https://arrow.tudublin.ie/datas/25/.