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
]. 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
]. 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
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 flow conditions. Uptake studies have been carried out using a range of biological assays including fluorescent imaging [10
]. Fluorescent labels can either be general or specific for a given FA, but the use of such labels is not always straightforward since the label itself can affect the uptake and biochemical activity of the labelled molecule [11
]. Detailed chemical information can potentially be gained from mass spectrometry (MS)-based studies but typical lipidomic experiments involve both extraction of lipids from the biological sample and chromatographic separation prior to analysis [12
]. Importantly, MS-based lipidomic experiments cannot be done at the single cell level and although single cell lipid imaging using time of flight secondary ion mass spectrometry (ToF-SIMS) can be achieved, it is not high throughput and generally only images from a few cells are reported [13
]. More recently, studies have been carried out using optical techniques such as infrared (IR) or Raman spectroscopy. Unlike MS, these methods are non-destructive and can measure the overall global biochemistry of a sample on a cell by cell basis. FA uptake studies have mostly sampled fixed time points under static rather than dynamic flow conditions [14
]. Infrared studies by Gazi et al. followed the uptake of both deuterated arachidonic and palmitic acid but the samples were analysed after being fixed and dried since strong IR absorption by water significantly hinders live cell analysis [14
]. Raman spectroscopy is more suited to the study of cells in aqueous environments since the water gives a relatively weak Raman spectrum. Lipid molecules modified with an alkyne tag have been employed to the study specific lipid uptake [17
]. The alkyne tag can show a much higher Raman signal intensity in the spectra and hence enables easier monitoring of the uptake of the tagged molecule. The uptake and metabolism of fatty acid molecules, however, are highly sensitive to structure changes, which raises a question of whether the tagging could affect the metabolic pathway [20
To study FA uptake under more realistic conditions it is desirable to couple spectroscopy with a microfluidic device such that real time measurements under dynamic/switchable conditions can be measured. Coupling Raman to a dynamic flow system is a relatively new development and has been used to investigate several biological systems including identification of circulating tumour cells, drug cell interaction and lipid uptake [21
]. Most studies use surface enhanced Raman spectroscopy (SERS) to increase the sensitivity of the system but again the addition of the SERS tag to the molecule in question has the potential to influence the outcome. Often these studies use home-made bespoke microfluidic devices that are tailored specifically to a given task and are thus not readily available [21
]. In the work presented here, a commercially available shear-flow assay system (BioFlux200, Fluxion Biosciences Inc., Ca, USA) is used to provide a dynamic flow environment coupled to a Raman microspectrometer. This represents the first demonstration of this particular combination of instrumentation. In this study, we use the prostate cancer cell line PC-3 to monitor the uptake of various fatty acids.
Prostate cancer (PCa) is the third most common cancer in the UK and the second most common cause of cancer death among men [26
]. Metastases to the bone marrow stroma (BMS), a rich source of lipids stored within adipocytes, is the main cause of morbidity and mortality in PCa patients. Several studies have demonstrated how dietary fatty acids relate to the progression of PCa and in recent years, it has been shown that specific types of FAs have a particularly detrimental effect on patient outcome [27
]. Moreover, research has shown a strong correlation between different types of dietary fatty acids (FAs) intake and PCa progression, where high-levels of palmitic acid (PA) and arachidonic acid (AA) intake tend to have a greater impact on cancer behaviour [27
]. Brown et al., however, found that omega-3 polyunsaturated fatty acids (PUFAs) can induce inhibitory effects on the uptake of AA [30
]. This finding has implications for competitive lipid uptake between different types FAs. In the work reported here, isotopically labelled FAs (deuterated palmitic acid (D31-PA) and deuterated arachidonic acid (D8-AA)) were used in combination with two non-deuterated omega-3 PUFAs, docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA) to study competitive uptake in PC-3 cells.
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
]. 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 νs
. 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 νs
]. 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 νs
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
]. 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
], 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 νs
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.
4. Materials and Methods
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 4
a,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.
4.1.1. 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 4
c, 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 4
b. 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
4.1.2. 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.
4.1.3. 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:
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.
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.
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.
4.2. 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 × 107 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 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
4.3. 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. The cells were treated with either: D31-PA only, D31-PA and DHA, D31-PA and EPA, and D31-PA and D8-AA in each experiment (all FAs are from Sigma-Aldrich Inc., Poole, UK).
The FA-containing RPMI 1640 cell culture medium (with 1% L-glutamine and 1% penicillin/streptomycin for all experiments, and 100 μM CoCl2 for hypoxia experiment) is supplied to the cells by constantly applying 0.15 dyn cm−2 sheer flow to the inlet during the experiment, except during the Raman spectra measurements. At each time point, the solution was then switched to DPBS at the same flow rate until the viewing chamber was fully replaced with DPBS. After the measurement, the solution was then switched back to the FA-containing medium for further treatment. Approximately 45 cells were sampled at each time point: t = 0, 15, 30, 45, 60, 90, 180 and 1440 min.
4.4. Data Processing and Analysis
One-to-one background subtraction was applied to each cell spectrum (SynerJY version 22.214.171.124).
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 νsCD2 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.