Fatty-Acid Uptake in Prostate Cancer Cells Using Dynamic Microfluidic Raman Technology.

It is known that intake of dietary fatty acid (FA) is strongly correlated with prostate cancer progression but is highly dependent on the type of FAs. High levels of palmitic acid (PA) or arachidonic acid (AA) can stimulate the progression of cancer. In this study, a unique experimental set-up consisting of a Raman microscope, coupled with a commercial shear-flow microfluidic system is used to monitor fatty acid uptake by prostate cancer (PC-3) cells in real-time at the single cell level. Uptake of deuterated PA, deuterated AA, and the omega-3 fatty acids docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA) were monitored using this new system, while complementary flow cytometry experiments using Nile red staining, were also conducted for the validation of the cellular lipid uptake. Using this novel experimental system, we show that DHA and EPA have inhibitory effects on the uptake of PA and AA by PC-3 cells.


Introduction
It has been known for many years that metabolic reprogramming, resulting in increased production of intermediates for the synthesis of proteins, nucleic acids and lipids, is a well-established signature of cancer development [1,2]. This process is a prerequisite for cells to break-free from the primary tumour, survive in an energy/nutrient deficient environment, and migrate to a metastatic niche [3][4][5][6]. Within the last decade or so, it has become increasingly clear that the function of cellular lipids is not just energy storage, but these biomolecules play a crucial role in cellular regulation associated with cancer progression. Cells in healthy mammalian tissues satisfy their lipid requirements through the uptake of fatty acids (FAs) from the bloodstream and this is an important and tightly controlled process [7]. Metabolic reprogramming initiates a shift in the balance between exogenous FA uptake and de novo FA synthesis. The deregulation of FA uptake in cancer metabolism and consequences for cancer progression; however, are still not well understood [8,9].
FA metabolism is a complex and dynamic process and there is a real need to develop new analytical methodologies that enable the study of such systems in real time under changing dynamic . This is the Raman shift region that was used for relative quantification of the D31-PA uptake by the PC-3 cells by calculating the areas under the curve in this study.
As shown in Figure 1b, the baseline of the CD stretching region is not flat which means the background signal was not able to be removed completely. Nevertheless, the symmetric -CD 2 stretching (ν s CD 2 ) signal from D31-PA can be clearly detected at 2106 cm −1 . Figure 1c shows just the region 2082-2136 cm −1 with a linear baseline correction. The areas under these curves were subsequently used for monitoring the relative, rather than absolute, D31-PA uptake.

Competative Fatty Acid Uptake.
Areas under the ν s CD 2 peak were calculated for each cell to monitor the relative D31-PA competitive uptake. D31-PA is the only isotopically tagged material used in the treatment and all Raman spectra were acquired from cells in pure DPBS. Therefore, the ν s CD 2 peak at 2106 cm −1 indicates only D31-PA uptake by the cells. Figure 2 shows and example box and whisker plot for cells treated with D31-PA under normoxic conditions. The trend plot shows that there is a continuous increase in the D31-PA uptake as a function of exposure. There is, however, significant variation in the amount of D31-PA taken up by individual cells as a function of time, indicative of cell population heterogeneity.  Figure 3a shows competitive fatty acid uptake under normoxic conditions. Please note that the heterogeneity in uptake is similar to that observed in Figure 2, but for visual clarity just the trend lines are shown. The cells treated with D31-PA alone have the highest uptake rates (the red line). The trend line suggests that the uptake of D31-PA started as soon as the treatment was administered and a notable increase of D31-PA signal can be seen 45 min post treatment. This uptake remains at a high rate but starts to slow down after 90 min. After 24 h of treatment, the uptake of D31-PA was still on-going but at a slower rate. suppression effect can be seen throughout the rest of the time points in the EPA and D8-AA experiments. However, cells treated with D31-PA and DHA behave differently compared to the other experiments. Although it was expected that DHA will have the similar suppression effects on the uptake of D31-PA by the PC-3 cells, the results suggest DHA reduces uptake but it is not as effective as similar concentrations of AA or EPA.  Lipid competition experiments were carried out by treating the cells with D31-PA and one other FA: DHA, EPA or D8-AA under normoxic conditions. The trend lines of the lipid competition experiments are also plotted in Figure 3a where suppression of the uptake of D31-PA by the PC-3 cells due to the presences of DHA, EPA or D8-AA can be seen to varying degrees. Since D31-PA is the only reagent used that will give signal at the 2106 cm −1 ν s CD 2 peak, inhibitory effects on the D31-PA uptake by omega-3 FAs can be reflected if a lower rate of uptake of D31-PA is observed. There are no significant uptake differences between all experiments in the first 45 min of treatment, yet suppression of the D31-PA uptakes by DHA, EPA and D8-AA were observed from t = 60 min. This suppression effect can be seen throughout the rest of the time points in the EPA and D8-AA experiments. However, cells treated with D31-PA and DHA behave differently compared to the other experiments. Although it was expected that DHA will have the similar suppression effects on the uptake of D31-PA by the PC-3 cells, the results suggest DHA reduces uptake but it is not as effective as similar concentrations of AA or EPA.
Data show the suppression of D31-PA uptake in the presence of D8-AA under normoxic conditions, suggestive of the essential nature of AA in contrast to the non-essential PA. In Figure 3a, however, the uptake trend of D31-PA by the cells treated with D31-PA only is slightly different to other cases. The uptake of D31-PA maximises at t = 180 min and starts decreasing afterwards in the lipid competition experiment with D8-AA, where the uptake of D31-PA gradually increases after t = 180 min in other cases. It is believed this phenomenon is related to the cellular adaptation to the environment.
The coupled system used in this study can also be used for studying similar mechanisms under different conditions. To demonstrate this flexibility, uptake of D31-PA by PC-3 cells was also monitored under chemically induced hypoxic conditions. The majority of the research into the relationship between the progression of PCa and dietary FAs are carried out under the "normoxic" atmospheric conditions. However, it has been reported that many prostate tumour lesion are hypoxic, with an average pO 2 of 6.7 mm Hg as compared with blood (75-100 mm Hg) and muscle (30 mm Hg) [33].
These changes in conditions may affect the metabolism of cells but not necessarily make any difference to the uptake of lipids. The trends for the uptake of D31-PA under normoxic and hypoxic conditions are compared in Figure 3b and the result suggests that the uptake of D31-PA by PC-3 cells under hypoxic condition started earlier than cells under normoxic conditions. D31-PA signal can be detected at 15 min of treatment and continues to increase until reaching a maximum at 180 min. The trend of the uptake of D31-PA under hypoxia drops at 24 h.
All results have been statistically tested by comparing the D31-PA uptake by cells treated under normoxic conditions with the competition experiments using the two-tailed t-test. The relative information and statistical results can be found in Table 1. The null hypothesis is that the uptake of D31-PA by PC-3 cells stays the same under different conditions. The p value for each case was calculated and any p value that is greater than the critical value of 0.05 or 0.01 will reject the null hypotheses, i.e., there are significant differences between the two sets of data. Table 1 shows the calculated p values for all cases at each time point. The results show that the uptake of D31-PA is supressed when omega-3 fatty acids were administered to the cells along with the D31-PA. Furthermore, a significant difference of D31-PA uptake between the control experiment and the cells treated with D31-PA only was observed. Further Student's t-tests were performed on the data to confirm that there are statistical differences of the D31-PA ν s CD 2 peak area, between the experiments using untreated and treated cells. The table showing the p values and further information can be found in Section Appendix A.3.

Distribution of Data Points
It was observed that there are larger variances in D31-PA uptake between cells at later time points. This is unlikely to be caused by the system's stability, as the system had been tested non-biologically with polystyrene beads (details can be found in Section 4.1.3). To rule out a relationship between the variance in measured uptake and the position in the sequence of 45 cell measurements, the relationship between the areas under the ν s CD 2 peak and the corresponding sequence number was investigated for three time points. As can be seen in Figure A2a-c, no trend exists between the level of uptake and the order in which the cells were measured A more likely cause of variation in the data is the localisation of lipid droplets within the cells [34,35]. The Raman laser spot size is smaller than the whole cell, so it is possible that large differences will be seen if the Raman laser hits a lipid droplet rather than an area with low lipid concentration. To investigate this, cells were treated with 20 µM D31-PA but in this case five spectra were obtained for each cell at different locations, which means that the whole cell was completely sampled, see Figure A2d,e.
The areas under the peaks measured at the five locations in each cell are different, reflecting the localisation of the lipid droplets. This may be one of the contributions to the large variance observed at later time points in the lipid uptake experiments, where the Raman laser might have just hit or missed the lipid droplet. In addition to the variations within each cell, variations between different cells were also observed.

Discussion
In this study, the BioFlux microfluidic system was coupled with a Raman microspectrometer for the first time, and therefore several system optimisation and method development experiments were required. Cells had to be seeded into the microfluidic device before performing any experiments. Coatings, such as poly L-lysine or fibronectin, are widely employed for improved cell attachment in microfluidic channels [36,37]. Although the coating is only a monolayer of molecules, it may still be detected by Raman spectroscopy, which is undesired. Therefore, it was decided to pre-seed the cells for 2 days in uncoated wells before starving so that the cells had enough time to settle down and attach. Cells seeded with this method allowed us to perform the experiments with no significant cell detachment at a sheer flow rate of 0.15 dyn cm −2 .
To ensure a consistent FA baseline, serum-free media was used in the FA uptake experiments. This is because serum-containing culture media will lead to a reduction in the D31-PA uptake compared with cells cultured in serum-free medium, specifically due to the presence of foetal calf serum (FCS) which contains a range of FAs [38]. A FA treatment concentration of 20 µM was selected based on the study by Brown et al., which suggested that cell viability decreases in the presence of FA at a concentration greater than 50 µM [30]. In the work present here, a total FA concentration of either 20 µM (single FA), or 40 µM (competitive uptake) was employed.
One of the aims of this study was to monitor a specific FA uptake which normally would require a tag on the FA. However this may affect the uptake and metabolism of the FA as FAs are highly sensitive to structure changes [20]. Deuteration overcomes this problem and has the added benefit of the resultant Raman signal of CD stretches appearing in the region of 2000-2400 cm −1 where no other biochemical information is observed. Furthermore, deuteration has minimum influence on the metabolic pathway of the FAs. D31-PA gives the strongest signal at 2106 cm −1 which is attributed to the ν s CD 2 . D8-AA was also used in one of the experiments and the = CD stretch was expected to show at 2250 cm −1 . This is because D8-AA was deuterated at the unsaturated carbons in the molecule, where D31-PA was deuterated at the saturated carbons. This results in the increase in vibration frequency of the CD stretches in the D8-AA compared with the D31-PA [39]. However, no D8-AA signal was detected in the Raman spectra acquired in this study. This is suspected to be due to the low Raman sensitivity of, ν(= CD), compared to the ν s CD 2 signal [40]. Since these two peaks are very close to each other and a flat baseline was not achieved in the CD stretching region, the weak ν(= CD) signal from D8-AA appears to have been obscured by the dominant ν s CD 2 .
Brown et al. demonstrated that omega-3 FAs have an inhibitory effect on the invasive ability of PC-3 cells induced by AA. The invasive effect could be recovered again by the addition of downstream AA metabolites [30,41]. However, it was not conclusive that the inhibition was due to metabolic competition or competitive uptake. With PA, also a FA that induces metastatic effects of PC-3 cells [42,43], it is hypothesised that the uptake could be inhibited by omega-3 FAs such as DHA and EPA. The Raman results acquired here suggest that the omega-3 FAs do inhibit the uptake of PA by the PC-3 cells. However, the uptake of PA by the cells treated with both D31-PA and DHA shows a different trend to the other cases. The suppression of the D31-PA uptake is much lower, and the trend line suggests an increase in uptake after 24 h (1440 min) of treatment. This suggests the inhibitory effect of the D31-PA uptake by DHA is less effective than EPA or AA, which is consistent with the invasion findings by Brown et al. [30]. Similar inhibitory effects on the absorption of saturated FAs, such as PA, brought about by EPA and DHA have also been observed by Yang et al. [44].
From the results presented here, D31-PA can be clearly detected using Raman microspectroscopy and trends in D31-PA uptake can be obtained under different environmental conditions. However, large variations in the ν s CD 2 signal detected were found in the live-cell experiments, especially in the later time points. Relative standard deviations (RSDs) were calculated from the cell spectra acquired. Despite the RSD associated with D31-PA uptake at time points 90, 180 and 1440 min being 60-70%, the RSD for the whole spectrum range is only~4.5%. (The corresponding tables can be found in the Section Appendix A.3.) System repeatability using polystyrene beads was determined to be greater than 99% (Section 4.1.3), and therefore it is believed the variations observed in uptake are purely because of the biological variances between single cells. This can be further demonstrated by the results obtained from the experiments shown in the Section Appendix A.2, in which the entire cell was measured. Although variations of D31-PA signal were observed due to lipid localisation, there is also significant variation of the CD signal between cells.
The uptake of D31-PA displayed in Figure 3 generally shows a decrease after treatment of 180 min, which is believed to be attributed to the adaption of the cell to the environment as discussed in Section 2.2. The cells were starved for 24 h in serum-free media before performing the uptake experiment; thus depriving the cells of lipids required for survival and proliferation and ensuring a comparable lipid baseline for all the experiments. PA can be synthesised by the cell from carbohydrates or protein (available resources from the media) while AA can only be obtained from diet [45], cells will only uptake the essential FAs (AA in this case) after adapting to the environment.
PA uptake by PC-3 cells under chemically induced hypoxic conditions was also studied. Some differences in the uptake of D31-PA were observed at 30 min and at 24 h with an increase of hypoxic uptake at 30 min but a decrease after 24 h. Bensaad et al. reported that hypoxia can induce greater uptake of lipids from the extracellular environment where there is an increase in lipid droplets upon HIF1-α induction without an increase in intracellular production [46]. However, this needs further investigation.

Instrumentation
In this study, the BioFlux 200 shear flow assay system (Fluxion Biosciences Inc., Ca, USA) was coupled with a customosed Raman microspectrometer and several optimisation and method development experiments were carried out. The final system configuration is shown in Figure 4. Figure 4a,b shows the layout of the coupled BioFlux and Raman microspectrometer and 4c,d shows the configurations of the 48-well plates used in this study.
The BioFlux 200 contains a controller, a pressure interface, a vapour trap, pipelines for connecting the units, and a heating plate. The controller regulates the air pressure applied to the wells and also the heating plate temperature. Experiments were conducted in the BioFlux well plates coupled to a Raman microspectrometer, which consists of an inverted Nikon Eclipse TE300 microscope (equipped with a Nikon Plan Fluor 100 × oil immersion objective, Nikon UK Ltd., Surbiton, UK), a Horiba Scientific iHR-320 Raman Imaging Spectrometer (focal length of 320 mm, f/4.1 aperture), coupled with a Horiba Syncerity CCD camera (Horiba Scientific, Northampton, UK). A beam splitter (Edmund Optics NT64-286), was used for guiding the 532 nm Raman laser to the sample stage and also separating the reflected excitation laser and the Raman signal going to the spectrometer.
Molecules 2019, 24, x FOR PEER REVIEW 9 of 23 f/4.1 aperture), coupled with a Horiba Syncerity CCD camera (Horiba Scientific, Northampton, UK). A beam splitter (Edmund Optics NT64-286), was used for guiding the 532 nm Raman laser to the sample stage and also separating the reflected excitation laser and the Raman signal going to the spectrometer.

Microfluidic Device
The BioFlux well plates required for the system are the size of standard 24-or 48-well plates, but with microfluidic channels built underneath the wells. The main body of the well plates are made of polystyrene, while the top and the walls of the microfluidic channels are made of polydimethylsiloxane (PDMS). Since a high numerical aperture is used for the Raman spectroscopy, the laser beam shape is no longer collimated. Thus, there is a chance of getting some PDMS signal scattered from the top and walls of the channel. Although PDMS is highly Raman sensitive and could potentially interfere with the experiment, this is not a problem since the CD signals from the deuterated lipids are located in the 'silent area' of a Raman spectrum in which PDMS does not contribute. The dynamic flow process is controlled through the BioFlux software, for which manual and automatic modes are available. The air pressure can be set from 0 to 20 dyn cm −2 and the length

Microfluidic Device
The BioFlux well plates required for the system are the size of standard 24-or 48-well plates, but with microfluidic channels built underneath the wells. The main body of the well plates are made of polystyrene, while the top and the walls of the microfluidic channels are made of polydimethylsiloxane (PDMS). Since a high numerical aperture is used for the Raman spectroscopy, the laser beam shape is no longer collimated. Thus, there is a chance of getting some PDMS signal scattered from the top and walls of the channel. Although PDMS is highly Raman sensitive and could potentially interfere with the experiment, this is not a problem since the CD signals from the deuterated lipids are located in the 'silent area' of a Raman spectrum in which PDMS does not contribute. The dynamic flow process is controlled through the BioFlux software, for which manual and automatic modes are available. The air pressure can be set from 0 to 20 dyn cm −2 and the length of operating time can be pre-set for the automatic mode. In Figure 4c, the purple shading on the well plate indicates one operation unit, with inlet at the left hand side and outlet at the right hand side. The air pressure can be used to control flow rate and can be applied on either the inlet or outlet side in each unit and the four units can be operated at the same time. Moreover, due to the pump design in the main controller, the air pressure applied has to be the same for each of the two operation units (Column 1-4 and Column 5-8). However, length of operation time can be varied.
A sheer flow of 0.15 dyn cm −2 was chosen for the experiment, which is high enough to enable a continuous supply of both fresh media and FAs to the cells in the microfluidic channels but is not high enough to cause detachment of cells from the channel surface.
A capillary structure consisting of a series of bends allows laminar flow to be developed in the microfluidic channels as shown in Figure 4b. The solution takes a certain length of time to flow from the inlet well to the viewing chamber (dead volume), and the time required will vary depending on the flow rate of the solution. To determine this time, the Raman signal changes were monitored as RPMI 1640 cell culture medium was replaced by Dulbecco's Phosphate-Buffered Saline (DPBS, Sigma-Aldrich Inc., Poole, UK). DPBS is a physiological buffer (pH 7.1-7.5 Osmolality 275-304 mOs Kg −1 ) that gives negligible fluorescence background for Raman measurements. Due to the large background intensity differences between RPMI 1640 cell culture medium and DPBS, the total area under the Raman curves (AUC) can be used as an indicator for determining the solution flowing in the microfluidic channels. The results used to obtain the dead time are shown in the Section Appendix A.4.

Raman Microspectroscopy
To minimise the potential photodamage to the cells, it is important to consider the length of time the cells are exposed to the laser. To obtain a reasonable signal to noise ratio (SNR), the maximum power available from the 532 nm Raman laser at 110 mW was chosen. However, the laser power diminishes to 41.5% of the power at source when it reaches the sample stage on the microscope. To avoid compromising SNR and signal saturation, several integration times have been tested for acquiring cell spectra at 110 mW at laser source. The final integration time used for all cell spectra was 10 s. A plot of measured power density and laser output power can be found in the Section Appendix A.5. Please note that all measurements were carried out in a DPBS background, i.e., cells in DPBS. The corresponding cell viability test results can be found in the Section Appendix A.6. With the integration time of 10 s, the time required for acquiring one set of data at a time point was 15-20 min. The test results suggest that cell viability remains greater than 90% after 60 min, which is longer than the time required for all Raman spectral acquisition at each time point. After the measurement, cells were then exposed to culture medium again to provide materials essential for their survival.

System Repeatability
The repeatability of the system was tested using polystyrene beads. All measurements were taken with Raman power of 55 mW at laser source (to avoid detector saturation) and 1 s integration time. Two sets of experiments were conducted to test the repeatability of the system with 20 µm and 25 µm diameter polystyrene beads: (i) Repeat measurements on the same bead 45 measurements on the same polystyrene bead were acquired sequentially. Two more replicates were obtained by moving the laser beam away from the bead for a few minutes, then returning to the same bead again for another 45 measurements. This experiment was repeated for three times i.e., three replicates.
(ii) Repeat measurements on different beads 45 different beads were measured sequentially. Three replicates were taken, i.e., the experiment was repeated three times, but beads measured are not necessarily the same beads among the three replicates. Figure 5 plots all the mean spectra of the three replicates of the experiments. In each case, there were 45 measurements and therefore there are 540 measurements in total. As can be seen there is almost perfect overlap between spectra. To observe marginal spectral differences, two sections have been significantly expanded and are shown in the insert in Figure 5 Standard deviations (SD) of the AUCs, mean AUC and relative standard deviations of the AUCs were then calculated. An average RSD of 0.94% was obtained from all 12 experiments indicating that the Raman microspectrometer system produces stable performance and gives high repeatability (>99%). This confirms that variances obtained in experiments involving cells are caused by biological differences.
Molecules 2019, 24, x FOR PEER REVIEW 11 of 23 almost perfect overlap between spectra. To observe marginal spectral differences, two sections have been significantly expanded and are shown in the insert in Figure 5 Standard deviations (SD) of the AUCs, mean AUC and relative standard deviations of the AUCs were then calculated. An average RSD of 0.94% was obtained from all 12 experiments indicating that the Raman microspectrometer system produces stable performance and gives high repeatability (>99%). This confirms that variances obtained in experiments involving cells are caused by biological differences.

Cell Culture
The PC-3 cells were first cultured in standard culturing flasks in Ham's F12 cell culture medium (Sigma-Aldrich Inc.), 10% foetal bovine serum (FBS, Sigma-Aldrich Inc.), 1% L-glutamine (Sigma-Aldrich Inc.) and 1% penicillin/streptomycin (Sigma-Aldrich Inc.) until 70% confluency. Cells were pre-seeded into the channels at a concentration of 2 × 10 7 cells mL −1 two days prior to the experiment to allow adhesion before treatment. Cells were seeded from the outlet side of the channel to prevent any contamination of the inlet channel at a sheer flow of 2 dyn cm −2 for 5 s. During these primary cell culture times, the cells were incubated at 37 °C with 5% CO2 supply in air.
Before applying any treatment to the cells under normoxic conditions (oxygen concentration in a normoxic incubator is 18.6%, at sea level [32]) in microfluidic channels, the cells were starved in serum-free RPMI 1640 medium (Sigma-Aldrich Inc.), 1% L-glutamine and 1% penicillin/streptomycin for 24 h. During the starvation period, the cells were also incubated at 37 °C with 5% CO2 supply in air.
In this work hypoxia was chemically induced with cobalt (II) chloride (CoCl2, Sigma-Aldrich Inc., Poole, UK) dissolved in distilled water [47]. 100 µM of CoCl2 was added to the serum-free medium and the cells were starved and incubated with the CoCl2 for 24 h before the lipid uptake experiment. The induction of hypoxia in these cells was validated using Image-iT™ Red Hypoxia

Cell Culture
The PC-3 cells were first cultured in standard culturing flasks in Ham's F12 cell culture medium (Sigma-Aldrich Inc.), 10% foetal bovine serum (FBS, Sigma-Aldrich Inc.), 1% L-glutamine (Sigma-Aldrich Inc.) and 1% penicillin/streptomycin (Sigma-Aldrich Inc.) until 70% confluency. Cells were pre-seeded into the channels at a concentration of 2 × 10 7 cells mL −1 two days prior to the experiment to allow adhesion before treatment. Cells were seeded from the outlet side of the channel to prevent any contamination of the inlet channel at a sheer flow of 2 dyn cm −2 for 5 s. During these primary cell culture times, the cells were incubated at 37 • C with 5% CO 2 supply in air.
Before applying any treatment to the cells under normoxic conditions (oxygen concentration in a normoxic incubator is 18.6%, at sea level [32]) in microfluidic channels, the cells were starved in serum-free RPMI 1640 medium (Sigma-Aldrich Inc.), 1% L-glutamine and 1% penicillin/streptomycin for 24 h. During the starvation period, the cells were also incubated at 37 • C with 5% CO 2 supply in air.
In this work hypoxia was chemically induced with cobalt (II) chloride (CoCl 2 , Sigma-Aldrich Inc., Poole, UK) dissolved in distilled water [47]. 100 µM of CoCl 2 was added to the serum-free medium and the cells were starved and incubated with the CoCl 2 for 24 h before the lipid uptake experiment. The induction of hypoxia in these cells was validated using Image-iT™ Red Hypoxia Reagent (Thermo Fisher Scientific UK Ltd. Hemel Hempstead, UK), a reversible fluorescent dye for measuring hypoxia in live cells. This reagent works based on the intracellular oxygen level which is non-fluorescent when the intracellular oxygen level is above 5% and becomes fluorescent (in red) when the intracellular oxygen level is below 5%. The results of these experiments are described in the Section Appendix A.7.

Fatty Acid Treatments
Several treatments were applied in different experiments to study the competition between the D31-PA with other fatty acids (FA). All FAs used in this project were obtained in either powder or neat oil form. Ethanol (≥ 98%, Thermo Fisher Scientific UK Ltd. Hemel Hempstead, UK) was chosen as the vehicle to make the FAs available to the cells. The concentrations used for each FA was 20 µM in serum-free RPMI 1640 cell culture medium with 1% L-glutamine and 1% penicillin/streptomycin.

Data Processing and Analysis
One-to-one background subtraction was applied to each cell spectrum (SynerJY version 3.5.7.20). All the cell spectra were imported into MATLAB 2017a for further processing including removal of any contributions from cosmic rays. 3-point smoothing with a moving mean method (chosen to minimise the effect of peak shifts), linear baseline correction and vector normalisation were applied.
With the 1200 lines per mm grating used in this project, the Raman shift is available from range 671 to 2430 cm −1 . However, the Raman shift range is purposely reduced to 2082-2136 cm −1 for monitoring the ν s CD 2 stretch from the D31-PA. The relative uptake of the D31-PA is quantified by calculating the area under the curves in this range after 2-point rubber band correction at first and last point of the spectra. Uptake curves of the experiments were plotted with the mean spectral area under this peak.

Conclusions
This work demonstrates the novel combination of coupling the BioFlux microfluidic system with a Raman microspectrometer in order to spectroscopically study biological systems under dynamic flow. With the increase in importance of studying the biology of living cells under their natural environment, development of dynamic flow systems help to simulate physiological conditions. The use of isotopically labelled fatty acids enabled monitoring of the uptake of a specific type of fatty acid by analysis of CD peaks in the Raman spectra. With the combination of this instrumentation and isotopically labelled reagents, continuous chemical monitoring in living cells at the single-cell level can be performed.
This study also demonstrates analysis of cells under different conditions such as normoxic and hypoxic states. This is a key factor in single cell studies since physiological conditions, particularly in tumours, are often highly hypoxic but most studies are carried out under in-vitro normoxic conditions.  Figure A1. During the measurements, the Raman laser was positioned onto the centre of each cell to acquire single cell spectra.
Molecules 2019, 24, x FOR PEER REVIEW 13 of 23 conceptual design of the fatty acid uptake experiments and contributed to the writing of the manuscript. R.D.S. was involved in experimental design and contributed extensive input into Raman instrument configuration. B.A.H. contributed training, technical support and scientific input into hypoxia experiments. A.R. provided the loan of the BioFlux instrument. All authors contributed to the writing of the paper and approve this version.

Funding: This research received no external funding
Acknowledgments: Lebtech International Ltd. and Fluxion Biosciences, Inc. are acknowledged for loan of the BioFlux200 and technical support.

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

Example Cell Image Taken under the Microscope
Cells under the microscope are shown inFigure A1. During the measurements, the Raman laser was positioned onto the centre of each cell to acquire single cell spectra.

Additional Data on the 5-Point Measurements
As described in Section 4.1.3, the system repeatability was confirmed non-biologically using polystyrene beads and it is believed that the variations observed are purely because of the biological variances between single cells. This was further investigated with 'five-point measurements' and the results suggest that the more likely cause of variation in the data is the localisation of lipid droplets within the cells. Results of the 'five-point measurements' are shown in, where cells were treated with 20 µM D31-PA. The five measurements at each location ensured that the whole cell was completely sampled.
The areas under the CD peak, plotted against the corresponding sequence number, are shown in Figure A2a-c. A representative plot of the D31-PA signal acquired at each point in a cell at t = 60 min is shown in Figure A2d. The locations of the five spectra obtained for each cell are indicated in Figure A2e.

Appendix A.2 Additional Data on the 5-Point Measurements
As described in Section 4.1.3, the system repeatability was confirmed non-biologically using polystyrene beads and it is believed that the variations observed are purely because of the biological variances between single cells. This was further investigated with 'five-point measurements' and the results suggest that the more likely cause of variation in the data is the localisation of lipid droplets within the cells. Results of the 'five-point measurements' are shown in, where cells were treated with 20 µM D31-PA. The five measurements at each location ensured that the whole cell was completely sampled.
The areas under the CD peak, plotted against the corresponding sequence number, are shown in Figure A2a-c. A representative plot of the D31-PA signal acquired at each point in a cell at t = 60 min is shown in Figure A2d. The locations of the five spectra obtained for each cell are indicated in Figure A2e. During the experiment, five spectra per cell were taken with 10 s of integration time. Taking the operation time into account, the total laser exposure time per cell was~1 min. As shown inFigure A2d, the area under the peak measured at the five locations in each cell are different, which reflects the localisation of the lipid droplets. This may be one of the contributions to the large variance observed at later time points in the lipid uptake experiments where the Raman laser might have just hit or missed the lipid droplet. In addition to the variations within each cell, variations between different cells were also observed. From Figure A2b, it is clear that some of the cells give larger ν s CD 2 signal than the others indicating that different cells uptake the D31-PA at different rates.

Appendix A.3 Statistical Significance for the Uptake Experiments
Results obtained were statistically analysed where the relative standard deviations (RSD) were calculated for assessing the variance. The calculated values are shown Tables A1 and A2 for the area under the 2106 cm −1 CD peak and the area under the full spectrum respectively. Two-tailed t-test was also used to evaluate the statistical significance of the uptake of D31-PA by PC-3 cells compared to the control experiment. The null hypothesis is that the uptakes of D31-PA by PC-3 cells are the same between treated and control experiments. The p value for each case was calculated and any p value that is greater than the critical value will reject the null hypotheses i.e., there are significant differences between the two sets of data. p values for all cases are shown in Table A3. The intensity differences between the Dulbecco's Phosphate-Buffered Saline (DPBS) and the RPMI 1640 media are shown in Table A3a. Figure Figure A3b against time and shows that the actual time required for replacing one solution in the channel with other solution is 105 s. The time-limiting step of the whole process is to remove the dead volume in the capillary structure. The area under the curve flattens out and remains constant at 450 s in Figure A3a, which indicates the media filled channel is fully replaced with DPBS at this point. This was taken into account when performing the time point experiments.

Relative Raman Laser Power Ratio Measurement Results
To obtain a measure of how much power loss occurs between the laser source and the objective, the relative power was measured at the two locations using an EG & G 550-1 Radiometer/Photometer equipped with an EG & G 550-2 multiprobe detector. For health and safety reasons and in order to make sure that the detector was operating in the linear range, an optical density (OD) 2.0 filter was placed in front of the laser source for the power density measurement. Please note that, since the area of the detector is significantly larger than the laser beam, it is just the relative power between the two locations that is of significance. Data were recorded up to a laser source setting of 40 mW as shown in Figure A4. Extrapolating this data to 110 mW we obtain a power at the objective of ~45 mW. Although this is quite high, the possibility of photodamage has been investigated in a similar Raman setup previously by Harvey et al. [48]. The results show that using 27.5 mW of 514.5 nm focused laser irradiation, caused minimal damage to cells up to an exposure time of 30s. Therefore, in this current study, the selected integration of 10 s suggests minimal photodamage to the cells using this system.

Appendix A.5 Relative Raman Laser Power Ratio Measurement Results
To obtain a measure of how much power loss occurs between the laser source and the objective, the relative power was measured at the two locations using an EG & G 550-1 Radiometer/Photometer equipped with an EG & G 550-2 multiprobe detector. For health and safety reasons and in order to make sure that the detector was operating in the linear range, an optical density (OD) 2.0 filter was placed in front of the laser source for the power density measurement. Please note that, since the area of the detector is significantly larger than the laser beam, it is just the relative power between the two locations that is of significance. Data were recorded up to a laser source setting of 40 mW as shown in Figure A4. Extrapolating this data to 110 mW we obtain a power at the objective of~45 mW. Although this is quite high, the possibility of photodamage has been investigated in a similar Raman setup previously by Harvey et al. [48]. The results show that using 27.5 mW of 514.5 nm focused laser irradiation, caused minimal damage to cells up to an exposure time of 30s. Therefore, in this current study, the selected integration of 10 s suggests minimal photodamage to the cells using this system.

Cell Viability in Room Temperature DPBS
During the five-point measurements, cells were in DPBS for an hour. The cell viability was tested with a trypan blue assay and pictures were taken under a microscope every 30 min with a 5 × objective, which are shown in Figure A5. During the five-point measurements, cells were in DPBS for an hour. The cell viability was tested with a trypan blue assay and pictures were taken under a microscope every 30 min with a 5 × objective, which are shown in Figure A5.

Cell Viability in Room Temperature DPBS
During the five-point measurements, cells were in DPBS for an hour. The cell viability was tested with a trypan blue assay and pictures were taken under a microscope every 30 min with a 5 × objective, which are shown in Figure A5.  Cell viability after 60 min room temperature DPBS exposure was counted by using Trypan blue assay. Automatic cell counter (Cellometer Auto 1000 Cell Viability Counter, Nexcelom Bioscience Ltd., Manchester, UK) was used and the results are shown in Table A4. Experiments were conducted to ensure the cells were under hypoxia after treating them with 100 µM cobalt (II) chloride (CoCl 2 , Sigma-Aldrich, Inc., Poole, UK) for 24 h. A solution of 5 µM ImageIT (concentration suggested by the supplier's standard procedures, Thermo Fisher Scientific UK Ltd. Hemel Hempstead, UK) was used for monitoring the changes in cell condition. This florescence dye works based on the intracellular oxygen level of the cells which will turn red if the intracellular oxygen level is below 5%. The dye is reversible and is transparent when intracellular oxygen level is increased to 5% or higher.
Cells were seeded into the BioFlux plate two days beforehand allowing them to attach and develop. Standard procedures used for seeding cells into the well plate are described in Section 4.2. Serum-free FluoroBrite with both ImageIT and CoCl 2 added as supplement were used for treating the cells. Images were taken at the 0 time point and 24 h time point. Figure A6 shows the example images taken at 0 and 24 h of treatment of one of the microfluidic channels. It can be seen that most of the cells turned hypoxic after 24 h of treatment. Bright field images are also included for locating the cells. Cell viability after 60 min room temperature DPBS exposure was counted by using Trypan blue assay. Automatic cell counter (Cellometer Auto 1000 Cell Viability Counter, Nexcelom Bioscience Ltd., Manchester, UK) was used and the results are shown in Table A4.

Monitoring the Change of States of Cells Using Hypoxic Reagent
Experiments were conducted to ensure the cells were under hypoxia after treating them with 100 µM cobalt (II) chloride (CoCl2, Sigma-Aldrich, Inc., Poole, UK) for 24 h. A solution of 5 µM ImageIT (concentration suggested by the supplier's standard procedures, Thermo Fisher Scientific UK Ltd. Hemel Hempstead, UK) was used for monitoring the changes in cell condition. This florescence dye works based on the intracellular oxygen level of the cells which will turn red if the intracellular oxygen level is below 5%. The dye is reversible and is transparent when intracellular oxygen level is increased to 5% or higher.
Cells were seeded into the BioFlux plate two days beforehand allowing them to attach and develop. Standard procedures used for seeding cells into the well plate are described in Section 4.2. Serum-free FluoroBrite with both ImageIT and CoCl2 added as supplement were used for treating the cells. Images were taken at the 0 time point and 24 h time point. During the Raman measurements, DPBS was used, and therefore it was important to ensure the cells did not switch back to normoxic state during the measurement period.
With 10 s of integration time for each cell and 50 cells per measurement, total DPBS exposure time for one time point was approximately 20-25 min. The fluid in the microfluidic channels was then switched back to the serum-free RPMI containing CoCl 2 to ensure that the hypoxic condition was maintained. To monitor the effect of DPBS on the cells condition, experiments were performed by exposing cells under hypoxia (treated with CoCl 2 for 24 h prior) in DPBS for 30 min. ImageIT reagent was used to monitor the condition of cells throughout the experiment. Figure A7 shows the dark field image of the same area in a channel of cells exposed to DPBS for 0, 10, 20 and 30 min. It is clear that the cells had not switched back into a normoxic state after 30 min.
Although measurements were expected to be completed within 30 min per time point, one further time point at 60 min was imaged. A dark field microscope picture is shown in Figure A8 which shows that the cells were still in a hypoxic state after 60 min of fresh DPBS exposure.