Introduction

1H Nuclear Magnetic Resonance (NMR) spectroscopy, originally an analytical chemistry technique, is now widely used for non-invasive detection of metabolites both in preclinical and clinical settings [1]. In particular, NMR spectroscopy has gained acceptance in cancer imaging by improving presurgical diagnostic accuracy and tumor grading, offering guidance for surgical intervention and determining efficacy of therapy [2].

One group of chemical species used in cancer is the NMR detectable lipids. Lipid NMR resonances have been detected in vitro from different cancer cells and tumors [3, 4] and are commonly present in malignant human brain tumors. Preclinical and clinical observations have shown that lipid signals are related to tumor class, grade and prognosis [4, 5], suggesting the potential of these signals to serve as a diagnostic tool and as a monitor for therapeutic effectiveness.

It has become evident that lipid droplets (LDs) in the cytoplasm are the main source of the NMR detectable lipids in situ [6, 7]. A study on a myeloma cell line showed that the appearance of lipid resonances detected by NMR was associated with formation of cytoplasmic LD in the presence of lipids in the culture medium [8]. In the past two decades, a growing body of evidence suggests that NMR lipid signals arise from lipid species that are associated with several important cellular processes, such as cell proliferation, activation, apoptosis and necrosis [6, 9]. Accumulation of cytoplasmic LD in cells and tissues undergoing apoptosis or necrosis has been reported, and this accumulation is accompanied with a sharp increase in the 1H NMR lipid signals [4, 10].

LDs have been regarded as generic cellular structures storing neutral lipids. However, recent studies revealed that LDs are dynamic organelles involved in multiple cellular functions [11, 12]. Lipid accumulation into LDs is a common observation in several human diseases, such as atherosclerosis [13], diabetes and obesity [14]. Recently, increasing evidence suggests that LDs are also implicated in other pathogenesis including inflammation [15], virus infections [16] and cancer [17]. An increased number of LDs displaying size variation have been reported in human colon [18] and brain tumour tissue [4].

It has been suggested that the NMR detection of lipids depends largely on the mobility of the detected molecules due to their slow transverse relaxation rate (R2) [6]. Therefore, the question has been raised as to whether an increase in NMR lipid signals is due to an increase in overall lipid amount in these LDs or due to an increase in lipid mobility associated with greater LD size [19]. NMR lipid signal is known to be related to grade and survival in brain tumours, but the variation and its relation to LDs in a panel of cell lines has not been fully explored. In addition, most studies on LDs using NMR have exploited the induction of LD accumulation in target cells through various procedures, thereby potentially complicating the interpretation of the origin of lipid NMR signals. Therefore, it is attractive to investigate viable and unexposed cells by NMR in order to gain an understanding of the relationship between the size of LDs and lipid NMR signals for quantitative interpretation of cellular lipids. In the current study, the LD size and NMR lipid signal were measured in five nervous system cancer cell lines to examine their relationship.

Materials and methods

Cell culture and harvest

The cancer cell lines used in the study were: BE(2)M17, a human neuroblastoma, BT4C, a rat glioblastoma, U87-MG, a human glioblastoma, PFSK-1, a human supratentorial primitive neuroectodermal tumor (ST-PNET) and DAOY, a human medulloblastoma cell line. These cell lines were cultured in 75 cm2 flasks with filter-vented caps (IWAKI, UK) and maintained in 20 ml Dulbecco’s modified Eagles medium (DMEM F:12), without l-glutamine (GIBCO, Invitrogen Corporation, UK) supplemented with 10 % (v/v) foetal calf serum (PAA, UK), 1 % 200 mM l-glutamine (100×) (GIBCO, Invitrogen Corporation, UK), and 1 % MEM non-essential amino acid solution (100×) (Sigma Aldrich, UK). The cells were incubated at 37 °C in a humidified atmosphere (5 % CO2, 95 % air) and harvested at around 90 % confluence. After washing with 10 ml ice-cold phosphate buffered saline (Invitrogen Ltd., Paisley, UK) 3 times, cells were then removed from the flask using a rubber policeman and centrifuged at 250 g for 6 min to form a pellet. Samples to be investigated with NMR were snap-frozen in liquid nitrogen and stored at −80 °C.

High-resolution magic angle spinning NMR (HR-MAS)

Prior to HR-MAS, frozen cells were defrosted and 36 μl was pipetted into a wide-mouthed zirconium sample tube (Varian Inc, Palo Alto, CA, USA); 4 μl 10 mmol/l trimethylsilylproprionate d4 (TSP) in D2O was added as a chemical shift standard. HR-MAS was performed on a Varian 600-MHz (14.1 T) vertical bore spectrometer using a 4-mm gHX nanoprobe (Varian NMR Inc) with a three channel Inova console running VNMRj software. The probe temperature was set to 0.1 °C, which equated to a temperature inside the rotor of 6.7 °C, determined by calibration using methanol. 6.7 °C was used to minimise the degradation of lipids. A rotor speed of 2,500 Hz was used for all experiments. The pulse sequence used for lipid investigation consisted of a single 90° pulse preceded by a 1 s duration water presaturation pulse. The receiver bandwidth was 7,200 Hz with 16 K complex points in the free induction decay. A total of 256 scans were acquired with a repetition time of 3.3 s giving a 14-min acquisition time.

Nile red and DAPI staining

Approximately 3.5 ~ 5 × 104 cells were spun onto one slide right after the harvest and stained with 4 μg/ml Nile red in PBS (made from 1 mg/ml Nile red stock solution in acetone, Sigma-Aldrich, Dorset, UK) for about 15 min under dark conditions. After Nile red staining, cells were stained with 0.4 μg/ml DAPI for 15 min. The slides were visualized with a Nikon Eclipse E600 microscope using a 100× objective lens and images were taken using a DXM1200 digital camera (Nikon, Japan). The green fluorescence of Nile red was observed with FITC (B-2A) filter set, with excitation wave length of 465–495 nm and the emission wave length of 550 nm. A DAPI UV (UV-2A) filter with excitation wavelength of 340–380 nm was used to detect DAPI stained nuclei. LD sizes were measured using the image analysis program ImageJ (National Institute of Health, USA). The diameter of all the visible LDs within one cell was measured and the results were analyzed with SPSS. The resolution limit of light microscopy is 200 nm. An Inter Pixel Stepping (IPS) technique was used with the DXM1200 digital camera and ACT-1 software to increase the resolution of captured images by a factor 9.

Post-processing of spectra and lipid peak analysis

FIDs were zero filled to 16 K data points prior to Fourier transformation, phased, referenced to the creatine peak at 3.03 ppm, water suppressed and normalized to the macromolecular peak at 1.68 ppm. The phased data were then transformed back to the time-domain and the TARQUIN algorithm as described in [20] was used to measure the lipid quantities by fitting a series of simulated individual lipid signals to the experimentally acquired data.

The lipid peak variability was demonstrated using at least 3 independent repetitions prepared from each cell line A relative value of lipid intensity, a ratio of lipid to macromolecular signal at 1.68 ppm, was used in the TARQUIN algorithm for normalisation so that the final values would be comparable across different cell lines.

Results

Size of LDs

The Nile red and DAPI co-stained images of five different cells are presented in Fig. 1. The LDs appeared as green and the blue stain represents cell nuclei. The size bar marks a reference for 5 μm. Qualitative microscopy suggested that LD size and number varied between the five cell lines studied and therefore, LD diameter and number were determined in three typical cells for each cell line.

Fig. 1
figure 1

Nile red and DAPI staining of five nervous system and brain tumor cell lines. The diameter of all visible LDs in the cell indicated by a red arrow was measured using Image J. The size bar represents 5 μm

Full size image

The distribution of LD diameters in our five experimental cell lines is shown in Fig. 2. BE(2)M17 was the only cell line that showed LDs larger than 1 μm, while the largest LD in DAOY cells was 0.4 μm. The mean diameter of LDs was similarly around 0.2 μm for each cell line (Fig. 2a). BE2M17 had the largest LDs while PFSK-1 contained the highest number of LDs. The middle black line in the box-plot (Fig. 2b) representing the median in each cell line was around 0.2 μm. BE(2)M17 had the largest in size and the highest in number of outliers shown as stars and dots in the box-plot. The dots and stars labelled with numbers in Fig. 2b are outliers to show the number of larger LDs and their distance to the mean value in each cell line.

Fig. 2
figure 2

Histogram (a) and box-plot (b) of the diameters (μm) of LDs to show the distribution and the number of LDs in five tumour cell lines

Full size image

The range, mean value, standard deviation of LD diameter and number of LDs are shown in Table 1. Total LD volumes are also shown for each cell line in Table 1.

Table 1 Statistical information of LDs across cell lines
Full size table

NMR lipid signal

HR-MAS spectra acquired from different cell types are presented in Fig. 3. The lipid resonances are assigned as follows, the protons contributing to the resonances shown in bold: 1-CH 3 at 0.9 ppm, 2-(CH 2) -at 1.3 ppm, 3-CH 2–CH2–C=O at 1.58 ppm, 4-CH 2–CH= at 2.02 ppm, 5-CH2–CH 2–C=O at 2.2 ppm, 6=CH–CH 2–CH= at 2.8 ppm, 7-CH=CH– at 5.4 ppm. The intensities of the 1.3 and 0.9 ppm lipid signals are highest in BE(2)M17. These lipid signals represent the methylene and methyl group from both saturated and unsaturated lipids. It should be noted that the 2.8 ppm resonance originating from poly unsaturated fatty acids was detected only in BE(2)M17 cells. Lipid signal intensities are intermediate in BT4C and U87-MG cells, while PFSK-1 and DAOY showed the lowest signal intensities. The resonance at 5.4 ppm, which arises from unsaturated lipids, is close to the value of the noise in PFSK-1 and DAOY cells.

Fig. 3
figure 3

HR-MAS spectra of five brain and nervous system tumor cell lines. The assignment of lipid signals is as follows: 1-CH 3, 2-(CH 2)n-, 3-CH 2–CH2–C=O, 4-CH 2–CH=, 5-CH2–CH 2–C=O, 6=CH–CH 2–CH=‚ 7-CH=CH–

Full size image

Correlation between LD size and NMR lipid signal intensity

Plots of the largest LD present in each cell line with the NMR lipid the methyl and methylene signals are shown in Fig. 4a, b. A plot of estimated LD volume and the lipid NMR signal intensities are shown in Fig. 4c, d. Linear correlations were evident between the size of LDs and the intensities of lipid NMR signals, as well as between LD volume and NMR signal intensities. However, the line of best fit does not pass through the origin, implying that no signal was observed from the smallest of the LDs. The lipid signal intensity was normalised to total MM, the sum of signal intensity from macromolecular peaks at 0.9, 1.68 and 3.0 ppm, as referenced to a protein signal, has been shown to be a good approach in NMR signal quantification [21].

Fig. 4
figure 4

Line fit plot for the largest LD in a single cell of each cell line and the lipid signal intensity of the methyl groups at 0.9 ppm (a) and the methylene groups at 1.3 ppm (b). Line fit plot for the estimated lipid volume of total LDs in a single cell of each cell line and the lipid signal of the methyl groups at 0.9 ppm (c) and the methylene groups at 1.3 ppm (d)

Full size image

Discussion

Intracellular LDs range in size from massive droplets 200 μm in diameter, found in mature adipocytes, to the nascent LDs of only 50 nm in diameter, found in milk-secreting cells [22]. Growth conditions, such as confluency, influence the size of LDs in cell lines [19]. In the current study, nervous system cancer cells were harvested at the log growth phase approaching confluence. In these cancer cells, an average LD diameter is approximately 0.2 μm. The difference between cell types lies mainly in the number and sizes of LDs with diameters larger than 0.4 μm. These observations indicate that the nervous system cancer cell lines studied produce LDs of similar diameter; however, some of these cell lines, most notably BE(2)M17, also contain significantly larger LDs under the culture conditions used.

The qualitative patterns of lipid NMR resonances from all five cell lines (Fig. 3) are similar, indicating that the chemical lipid species contributing to the NMR spectra are similar. Quantitative analysis shows that the NMR lipid peaks corresponding to –CH3 and –CH2– are largest in cells containing LDs larger than 0.4 μm. NMR detected lipid signals have been reported to correlate quantitatively with LDs measured with phase contrast microscopy [23] and Nile red-staining in a rat glioblastoma cell line [19, 23]. In order to give narrow NMR lines, molecules must tumble rapidly to have a slow R2 in the region of tens s-1 [6]. LDs exist in numerous types of cells [22], but lipids in LDs may generate no NMR signal due to restricted mobility caused by the dipolar effect in tiny droplets. Quintero et al. [19] proposed that LDs lower than 0.1 μm are NMR invisible. LDs in the cell lines studied were larger than 0.07 μm, PFSK-1 cells contained more LDs around 0.2 μm than any other cell line, yet the NMR lipid signals from PFSK-1 cells were weaker than in BE(2)M17 cells, a cell line with lower average LD count per cell than PFSK-1. Thus, the volume of LDs in this diameter range is a more important determinant of the NMR lipid detectability than the total number of LDs. This was demonstrated by the linear correlation between LD volume and NMR lipid signal intensities (Fig. 4). However, the line of best fit does not pass through the origin; rather, extrapolating to where the lipid signal is zero gives a value on the lipid droplet size axis of around 0.34 μm diameter, roughly in agreement with Quintero et al. It should be noted that the HR-MAS spectra in this study were acquired under 6.7 °C. NMR lipid signal was better detected by HR-MAS at 37 °C [24], therefore, the limiting droplet size for NMR lipid detection could be lower.

It should be borne in mind that macromolecules other than lipids, such as polypeptides and proteins [25], contribute to the peaks at 0.9 and 1.3 ppm. We observed that the lipid signal intensity of the –CH2– peak (Fig. 4a) in DAOY cells appears greater than that from –CH3 (Fig. 4b). The discrepancy is possibly due to themacromolecule contribution to the 0.9 ppm peak being higher than that of lipids. The relatively poor correlation between LD volume and lipid signal at 0.9 ppm in DAOY and PFSK-1 cells is most likely due to the large macromolecular contribution to this signal, as these two cell lines have a relatively higher protein/lipid level compared with other cell lines.

Conclusion

In conclusion, cytoplasmic LDs exist in large numbers prior to treatment in all the studied cell lines, but the size of the largest droplets varies greatly from one cell line to another. High NMR lipid signals are associated with the presence of large LDs and these cells also contain the greatest total volume of lipid. Some cell lines contain very large numbers of small LDs and are associated with small NMR lipid signals. The NMR signal reflects both the total amount of lipid present and the mobility of the lipids within the droplets. LDs above a certain size (here approximately 0.2 μm) have a similar rotational freedom for NMR detection and the total LD volume is a critical factor for NMR signal intensity.