3.2.1 Cell preparation for Raman spectroscopy
This study was approved by the School of Medicine Ethics Committee, University of St Andrews (Ethics statement is provided in appendix A). Samples were obtained from healthy donors after obtaining informed and written consent. Participation in- formation sheets and consent forms were also approved by the School Ethics Com- mittee.
CD4+ T cells, CD8+ T cells, NK cells, pDCs, mDCs, B cells and monocytes were isolated and purified. Negative depletion isolation kits were used to separate cells in an untouched manner. This method was chosen to avoid the use of labels which may alter the activation state of cells and change their behaviour. Full details of this method are provided in appendix B. This section also shows the results of analy- sis by flow cytometry and functional assay characterisation to analyse the purity of samples.
Cells were prepared on quartz slides for analysis by Raman spectroscopy. A thick quartz slide (25.4 mmx25.4 mm,1 mmthickness, SPi Supplies, UK) was used, forming a chamber by placing a vinyl spacer of80µmthickness on top. Live cells were suspended in phosphate buffered saline (PBS) and20µlwas placed in the well. A second thin quartz slide (25.4 mmx25.4 mm,0.15to0.18 mmthick) was placed on top to form a seal. The chamber was inverted, allowing the cells to settle on the thin slide for 30 minutes, which would prevent any movement caused by optical forces during measurements. The sample was placed on the confocal microscope with the thinner slide towards the objective.
It is considered an important aspect of biological studies that there is a large sam- ple size. Biological material have an inherent variability; a larger sample size can provide an averaging effect and produce a more accurate representation. The num- ber of cells analysed here is therefore only constrained by practical considerations. For example, cells remain healthy in the quartz chamber (outside of an incubator) for approximately 2 hours, during which time approximately 60 cells can be analysed by WMRS. This in turn depends on the concentration of cells; dendritic cells are quite rare in the blood and can be more difficult to find, therefore fewer data are recorded
for these cells. Additionally there is a limited lifetime for cells extracted from fresh blood, meaning all data must be acquired within three days after isolation. It is a combination of these factors that limit the number of cells which may practically be analysed. Blood samples were taken from three healthy donors to make the study more clinically relevant and to ensure the method is robust against different donors.
3.2.2 Performing Raman spectroscopy
Details of the instrumentation used for the Raman spectrometer are provided in sec- tion 2.7.3. 150 mWpower was provided to the sample plane. For WMRS measure- ments the wavelength was tuned over a total range of ∆λ = 1 nm. Five spectra were acquired at five equidistant wavelengths within this range where each single spectrum was acquired for5 s; in total an acquisition time of25 swas required for a WMRS point spectrum of a single cell. The method of processing WMRS data to ob- tain a single differential spectrum is detailed in section 2.5. All Raman spectra were normalised according to the area under the curve, to account for any fluctuations in power before processing the WMRS data.
Cells were irradiated continuously for 5 minutes and no change was observed in the Raman spectrum; this indicates there was no photodamage at this laser dosage on the time scales used. Raman spectra were also collected on different days to confirm the stability of the system.
Hoboro et al [164] have compared the use of either single point Raman spec- tra or Raman imaging for the analysis of immune cells; it was reported that whilst single point spectra do not acquire information from the whole cell, they are still largely representative of the cell as a whole. There are some spectrum to spectrum differences due to variations in position in the cell, which highlights the importance of basing these studies on a large number of cells. Furthermore, when acquiring a Raman image of a whole cell the acquisition time required increases dramatically (several minutes per cell), limiting throughput of cells analysed. Reducing the spec- tral quality can improve the throughput rate of Raman images but decreases the signal-to-noise ratio achieved and subsequently the discrimination efficiency. This comparison validates that the approach of acquiring single point Raman spectra is a robust and appropriate method.
3.2.3 Statistical analysis on Raman spectra
A detailed description of statistical analysis methods used is given in section 2.8. Specific details pertaining to this study will be outlined in this section. Raman spec- tra were analysed in the region of600 cm−1 −1800 cm−1. A parametric student’s
t-test was used to highlight regions of significant difference between the mean spec- tra of any two cell subsets. This indicates the Raman peaks which vary most, and hence any biochemical differences between two different cell populations. A sig- nificance level of p< 10−7 was used. PCA was applied to the data set to reduce the dimensionality and the first 7 PCs were selected, which accounted for the major variance across the data set. Cluster plots were produced using the first 3 PCs to vi- sualise trends in the data. The discrimination efficiency of this training data set was assessed by means of LOOCV. The left-out spectrum was defined according to the nearest neighbour algorithm. This was repeated for each spectrum and correct and incorrect cell classifications were summarised in a confusion matrix. Sensitivities and specificities were then calculated in a pairwise manner for each two cell subsets.
3.2.4 Daily procedure
Before acquiring Raman spectra the performance of the Raman system was assessed and optimised. As a first step the output power of the laser was measured. If the power had fallen from the expected power output, small adjustments to the out- put mirror and cavity high reflector was normally sufficient to regain the optimum power output. Secondly, the beam shape was assessed by viewing the reflection pattern from a glass surface in the sample plane. Tuning above and below the focal point should reveal a series of circular confocal rings expanding around a constant centre. If the optical axis were misaligned or the beam shape had lost its circular symmetry, realignment was achieved by ’beam walking’. As a next step Raman spec- tra were acquired from5µmpolymer beads; the acquired spectrum will be most in- tense when the bead occupies the confocal volume (figure 2.7), at this point the bead should appear in focus on the CCD camera. By comparison to a reference spectrum one can ensure optimal performance; any significant drop in the maximum signal
intensity may be due to the collection efficiency, which can be optimised by adjust- ing the position of the confocal pinhole. The region of the spectrometer CCD camera on which Raman scattered light is detected can be observed. The acquisition region was cropped to the appropriate pixels to minimise any dark noise contributions. Once the system was optimised, cells were prepared on quartz slides as previously described, and data collection could begin. To account for any drift in laser wave- length throughout the day calibration spectra were acquired from polymer beads every 2 hours, that is before and after each batch of cells (see section 2.8.2).
3.2.5 Potential problems
There are a range of potential issues that may arise and this section aims to address how they are managed. The laser cavity is sensitive to any temperature fluctuations in the lab which may cause thermal expansion and consequently misalignment of mirrors. Typically this can affect the output power, beam quality, and laser noise, and in extreme cases loss of lasing. The laboratory temperature is controlled to min- imise these effects and regular laser maintenance can avoid the total loss of lasing. The system also experiences a drift in wavelength over time and although spectra are calibrated (section 2.8.2), caution must be taken to ensure the data are robust. Data were therefore acquired from each cell type on different days and PCA was employed to validate that no significant variations were introduced due to the Ra- man system. Considering WMRS measurements, each wavelength shift produces a spectrum with a different output intensity. This is due to the wavelength dependent parameters of the laser cavity and filters used in the Raman system. Each spectrum is normalised according to the area under the curve before further processing how- ever if the variation is large it can affect the cosmic ray treatment and analysis by PCA (see section 2.8). It is vital that the centre wavelength corresponds to the centre of the transmission linewidth of the line filter, otherwise the first or last spectrum may have a significant drop in intensity.