2.1.

Cell Cultures

T24 and UMUC-13 (UC13) bladder cancer cell lines were provided by Dr. Monica Liebert (University of Michigan, Ann Arbor). Parental T24 cells originated from human urinary bladder carcinoma, while UMUC-13 cells were derived from lymphatic metastases of the transitional cell carcinomas of the bladder.^10,11 All cells were maintained in Dulbecco’s modified Eagle medium, high glucose [Gibco-Life Technologies (Gibco); Carlsbad, California] supplemented with 10% fetal bovine serum, $100\mathrm{U}/\mathrm{ml}$ penicillin, and $100\mu \mathrm{g}/\mathrm{ml}$ streptomycin in a humidified atmosphere of 5% ${\mathrm{CO}}_2$ at 37°C. Cells were passaged with 0.05% trypsin-ethylenediamine tetra-acetic acid (Gibco) when confluent (2 to 3 days). Both parental T24 (T24-Con) and UC13 (UC13-Con) cells were transduced with different lentiviral particles containing a GFP reporter. The lentiviral particles also contained two different constructs, shRNA A and shRNA C, which were transduced separately into both T24-Con and UC13-Con cell lines to generate the different respective knockdown cell lines: T24-A, T24-C, UC13-A, and UC13-C. Scrambled T24-Scr and UC13-Scr cell lines were also generated to rule out nonspecific target effects of the shRNA plasmid and to provide negative controls for the gene specific knockdown. All cell lines were grown on Petri dishes ($n=5$ per cell line), including two eight-well chamber LabTek II slides (Nunc, Rochester, New York) for DNA staining and fluorescence imaging.

Cell viability was determined by the trypan blue exclusion test. Briefly, $10\mu \mathrm{L}$ of the cell suspension was mixed with an equal volume of 0.4% trypan blue stain (Gibco; Carlsbad, California) in a microwell plate. The $10\text{-}\mu \mathrm{L}$ aliquots of the cell suspensions were loaded into the dual-chambered counting slide and the cells were counted using the TC20 Automated Cell Counter (Bio-Rad Laboratories). Percent cell viability was calculated using the following equation: cell viability = 100 − (100/total number of cells × number of dead or damaged cells). High cell viability values of 93% to 100% were routinely obtained for all cell lines used in this study.

2.2.

Cell Sample Preparation

Trypsinized cell pellets were resuspended in 5 mL phosphate buffered saline (PBS), centrifuged at low speed (1000 rpm for 3 mins), and then the supernatant was discarded. Cell pellets were further washed twice with 1 mL PBS to ensure all traces of spent culture media and trypsin were removed. Recovered cell pellets were diluted at a ratio of 1:3 with PBS in order to produce a suitable cell density for Raman microscopy. Aliquots of the diluted cell suspensions ($1\mu \mathrm{L}$ per cell line) were then deposited onto the gold-coated microscope slide (EMF Corporation, Ithaca, New York). The deposits were desiccated over silica beads using the procedure described elsewhere.¹² The desiccated circular deposits ($\sim 2.72\pm 0.07\mathrm{mm}$ in diameter) appeared granular under a light microscope because of the rounded shape of the detached cells.¹³

2.3.

Staining and Fluorescence Imaging

Culture media were removed from confluent cell lines cultured on the chamber slides ($n=2$ per cell line). The cells were immediately fixed with 0.2 mL ice-cold methanol (5 mins) and then rinsed twice with 0.4 mL ice-cold PBS (2 mins then 1 min). After removing PBS, the cells were incubated with 0.1 mL of $0.1\mu \mathrm{g}/\mathrm{mL}$ 4',6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich, St. Louis, Missouri) solution (3 mins). After DNA staining with DAPI, the cells were rinsed three times with 0.4 mL PBS ($3\times 5\mathrm{mins}$). The medium chamber was removed and the slide was mounted by a coverslip with a drop of mounting medium. The coverslip was sealed with nail varnish to prevent the cells from drying. The cells were examined using an Olympus BX41 microscope, equipped with an Olympus PP73 camera and CellSens image-processing software. An Olympus U-HGLGPS light guide-coupled illumination system served as the fluorescence light source for blue (DAPI) and green (GFP) fluorescence imaging. Images were acquired using a $40\times 0.75\mathrm{NA}$ objective and fixed acquisition times of $600\mu \mathrm{s}$ and 500 ms, respectively. Images were analyzed using the CellSens imaging-processing software. Areas of the images were estimated using a Formvar-coated copper grid (100 mesh; Electron Microscopy Sciences, Hatfield, Pennsylvania).

2.4.

Raman Microscopy and Spectral Analysis

The 785-nm Raman microprobe was constructed locally and is described in detail elsewhere.^14,15 The excitation laser was shaped into a line and focused through a $20\times /0.75\mathrm{NA}$ objective with a laser power output of $\sim 30\mathrm{mW}$ at the sample. The laser line was $\sim 135\mu \mathrm{m}$ in length with a lateral resolution of $\sim 15\mu \mathrm{m}$ (laser area $\sim 2025{\mu \mathrm{m}}^2$). This Raman line-scan approach enabled a large number of cells to be analyzed simultaneously compared to single cell collection approaches. The number of cells irradiated by the laser line ($\text{cells}/{\mathrm{mm}}^2$ laser area) is also estimated from the light microscope images. Spectra were collected at 6 to 8 different locations from 2 to 3 dried cellular deposits ($n=5$ per cell line). The cosmic ray removal setting was used prior to collecting the spectra using an accumulation cycle time of 7 mins ($7\times 60\mathrm{s}$).

All Raman data were processed in MATLAB^® software using locally written scripts described elsewhere.¹⁶ The script also included an automated “adaptive min-max” polynomial fitting procedure (third order, constrained) to correct for background fluorescence.¹⁷ All spectra were imported into GRAMS/AI^® software for baseline correction and normalization against the protein\lipid ${\mathrm{CH}}_2$ deformation band at $\sim 1447{\mathrm{cm}}^{-1}$. For curve-fitting, standard second derivative and constrained Gaussian deconvolution functions¹⁸ were applied to the following spectral ranges: 685 to $840{\mathrm{cm}}^{-1}$ (698, 718, 744, 757, 781, 810, and $826{\mathrm{cm}}^{-1}$), 1180 to $1400{\mathrm{cm}}^{-1}$ (amide III region: 1206, 1229, 1248, 1267, 1280, 1297, 1316, 1337, and $1356{\mathrm{cm}}^{-1}$), and 1530 to $1720{\mathrm{cm}}^{-1}$ (amide I region: 1550, 1570, 1584, 1605, 1619, 1635, 1656, 1672, 1685, and $1698{\mathrm{cm}}^{-1}$). All band intensities were nonnegative and were recorded when the best-fit curves yielded $r^2$ values of 0.98 or more.

Raman band assignments pertinent to cancer cells and the GFP chromophore were identified and their relative intensities measured. Only ratiometric Raman parameters derived from select band intensities that yielded the most significant results will be presented here. For GFP expression and biochemical changes, the following ratiometric parameters were calculated: GFP/amide (Am) I $\alpha$-helix ($1550/1656{\mathrm{cm}}^{-1}$), nucleic acid/tryptophan (Typ) ($781/757{\mathrm{cm}}^{-1}$), Typ/phospholipid ($757/718{\mathrm{cm}}^{-1}$), and phospholipid/cholesterol ($718/700{\mathrm{cm}}^{-1}$) ratios.^19–21 For changes in protein secondary structure, two ratiometric parameters were calculated: Am III disorder/order ($1248/1266{\mathrm{cm}}^{-1}$) and Am I disorder/$\alpha$-helix ($1685/1656{\mathrm{cm}}^{-1}$).^22,23 Data were presented as mean ± standard deviation (SD). Significant differences between parental and gene-silenced cell lines, including scrambled controls, were analyzed using one-way analysis of variance with statistical significance defined at $p<0.05$.

3.1.

Spectral Analysis of Dried Cellular Deposits

The mean Raman spectra of different parental and gene-silenced T24 and UC13 cell lines deposited on gold-coated substrates are shown in Fig. 1. All spectra exhibited similar compositional and structural profiles but with subtle differences that will be discussed in more detail later. The consistent quality of the spectra was attributed to several factors. First of all, the high percent cell viability levels (93% or more) obtained at each cell passage ensured that the cell populations used in the Raman spectroscopic analysis were free from dead or damaged cells.²⁴ Second, by using the laser-line collection mode, Raman scatter from a larger collection of individual cells could be obtained, which could be further enhanced if the cells were preconcentrated by centrifugation prior to deposition and analysis. Indeed, our image analyses show that the later approach achieved $\sim 13\text{cells}/{\mu \mathrm{m}}^2$ laser areas compared to $\sim 3\text{cells}/{\mu \mathrm{m}}^2$ laser areas, if cells grown on a slide had been analyzed. Figure 2 shows the granular appearance of desiccated UC13-Scr cells deposited on gold-coated substrates. The rounded morphology exhibited by UC13-Scr cells was typical for all cell lines used in this study and was similar in appearance to those reported for breast carcinoma and myeloid leukemia cells dried on ${\mathrm{CaF}}_2$ substrates.¹³ Third, cells deposited on gold-coated substrates exhibited good spectral signal quality that could be attributed to the substrate’s low background signal contribution and reflective properties when compared to transparent quartz substrates.^25,26

Fig. 1

Representative baselined-corrected Raman spectra (averaged from multiple locations) for the following T24 and UC13 bladder cancer cells desiccated on gold-coated substrates: (a) T24-Con, (b) T24-Scr, (c) T24-A, (d) T24-C, (e) UC13-Con, (f) UC13-Scr, (g) UC13-A, and (h) UC13-C.

Fig. 2

Light microscope image showing the granular appearance of detached UC13-Scr cells desiccated on gold-coated substrate. Scale bar: $20\mu \mathrm{m}$.

The use of desiccation to fix cells has been shown to increase spectral reproducibility and signal-to-noise ratios.¹² Another study showed that spectra from dry-fixed and live cells were highly correlated, implying that dried-fix cells could be used as proxies for living cells.²⁷ Providing that one fixation method is consistently maintained throughout the study, fixation-induced spectral effects can be minimized, thereby permitting useful comparisons between different cell types and disease states.²⁸ For instance, Raman microscopy in conjunction with linear discriminate analyses successfully differentiated between erythrocyte, leukocyte, acute myeloid leukemia, and breast tumor cells trapped inside microfluidic devices with high accuracies.²⁹ Moreover, the study showed that accuracies were comparable to previous studies in which the same cell types were air-dried or fixed on Petri dishes. In this current study, desiccation was chosen over solvent and chemical fixation methods in order to minimize the potential loss of cell membrane lipid components, modifications to protein secondary structure, and/or changes to nucleolar protein concentrations.^12,30 Indeed, a recent spectroscopic study with principal component analysis showed that membrane lipid was a key differentiating character trait between different cancer cell lines.³¹

The Raman spectra of cells are complex because the microscopy technique provides simultaneous information on intracellular proteins, amino acids, nucleic acids, lipids, and cholesterol molecules. The major protein-related and nucleic acid bands included the Typ band at $\sim 757{\mathrm{cm}}^{-1}$ and the DNA/RNA band at $\sim 781{\mathrm{cm}}^{-1}$, respectively.²⁰ Bands at $\sim 700$ and $\sim 718{\mathrm{cm}}^{-1}$ were assigned to cholesterol (mixed contributions from cholesterol esters) and phospholipid (mainly phosphocholine) vibrations, respectively.^21,32 For protein secondary structures, tentative band assignments were proposed for the following select deconvolved bands: amide (Am) III $\beta$-sheet/random coil ($\sim 1248{\mathrm{cm}}^{-1}$; contains minor contributions from cytosine), Am III $\alpha$-helix ($\sim 1266{\mathrm{cm}}^{-1}$), Am I $\alpha$-helix ($\sim 1656{\mathrm{cm}}^{-1}$), and Am I disordered ($\sim 1685{\mathrm{cm}}^{-1}$, mainly $\beta$-turns).^23,33,34 Other bands shown in Fig. 1 include the phenylalanine (Phe) and mixed protein/lipid ${\mathrm{CH}}_2$ deformation bands at $\sim 1002$ and $\sim 1447{\mathrm{cm}}^{-1}$, respectively.

3.2.

Spectral Assignment of the Green Fluorescence Protein Chromophore

A closer inspection of the mean spectra shown in Figs. 1(f)–1(h) revealed an additional band at $\sim 1552{\mathrm{cm}}^{-1}$ that appeared exclusively for the gene-silenced UC13-Scr, UC13-A, and UC13-C cell lines. Initially, the band was attributed to Typ because its position approximately coincided with the 1546- and $1552\text{-}{\mathrm{cm}}^{-1}$ Typ band assignments reported for bladder and lung tissue cancers, respectively.^35,36 However, the absence of a proportionally intense Typ band around $757{\mathrm{cm}}^{-1}$ in the spectra of gene-silenced UC13 cell lines, which is a characteristic of pure Typ and Typ-rich proteins,^37,38 suggested a different band assignment. A GFP band assignment was then proposed because only the gene-silenced UC13 cell lines fluoresced green when exposed to UV-light. Indeed, published Raman spectra of the anionic form of GFP and its structural variants contained a prominent band between 1537 and $1556{\mathrm{cm}}^{-1}$ that was assigned to the $─\mathrm{C}═\mathrm{C}─\mathrm{C}═\mathrm{N}─$ portion of the imidazolinone ring.^6,7 This assignment was further corroborated by our fluorescence imaging and Raman ratiometric studies. As shown by the fluorescence images of UC13 cell lines in Figs. 3(a)–3(d), the fluorescence intensity of GFP (with respect to UC13-Con cells) appeared to increase in the following order: $\mathrm{UC}13\text{-}\mathrm{A}>\mathrm{UC}13\text{-}\mathrm{C}>\mathrm{UC}13\text{-}\mathrm{Scr}$. A similar order was found when the ratio intensities of the GFP band at $\sim 1552{\mathrm{cm}}^{-1}$ to the protein Am I $\alpha$-helix band at $\sim 1656{\mathrm{cm}}^{-1}$ (GFP/Am I $\alpha$-helix ratio) were compared [Fig. 4(a)]. This is the first Raman spectroscopic study in which the GFP chromophore has been correctly identified in cancer cells transduced with the GFP reporter gene.

Fig. 3

Representative merged nuclear stained (blue, due to DAPI) and GFP (green) fluorescence images obtained for (a) UC13-Con, (b) UC13-Scr, (c) UC13-A, (d) UC13-C, (e) T24-Con, (f) T24-Scr, (g) T24-A, and (h) T24-C cell lines. The UC13-A cell line (c) exhibited the highest GFP fluorescence, followed by the UC13-C (d) and UC13-Scr (b) cell lines. GFP fluorescence was negligible for the corresponding gene-silenced T24 cell lines (f–h) and parental UC13 (a) and T24 (e) cell lines. Scale bar: $20\mu \mathrm{m}$.

Fig. 4

Relative distributions of biochemical and protein secondary structural components obtained by spectral fitting of parental (Con), scrambled (Scr), and gene-silenced (A or C) T24 and UC13 (A or C) cell lines. Plot error bars are based on SDs ($n=5$ per cell line).*$p$ values are defined at $<0.05$.

The near-absence of GFP expression from the fluorescent images obtained for the gene-silenced T24 cell lines (T24-Scr, T24-A, and T24-C) in Figs. 3(f)–3(h) suggests that either GFP transduction was low or that the cyclization of the GFP tripeptide to the fluorescent chromophore was inefficient. By comparing the GFP/Am I $\alpha$-helix ratios, the T24-C cell line exhibited some measureable GFP expression but was not statistically significant [Fig. 4(a)]. This is not entirely unexpected. Additionally, transduced cells with low GFP expression are sometimes preferred because they exhibit morphological and biochemical traits that are consistent with their parental cell line under study.⁸ Moreover, panel cell line studies have shown that UMUC-13 cells can be effectively transduced with the GFP reporter gene with negative toxicity as assessed by light microscopy at 48 h.¹¹ Regardless of whether the gene-silenced UC13 cell lines exhibited low- or high-GFP expression, the gene-silenced cells shown in Fig. 3 appeared to be healthy and exhibited cell morphologies that were similar to their parental counterparts.

3.3.

Spectral Biochemical and Protein Structural Analyses

To evaluate biochemical and protein structural differences between parental and gene-silenced cell lines, several ratiometric Raman parameters as shown in Fig. 4 were examined. Initially, nucleic acid/Typ ratios were examined because the parameter had previously been used to distinguish between normal human skin cells and tumorigenic keratinocytes.¹⁹ Changes in nucleic acid/Typ ratios can provide an insight into the differentiation status of cells, either living or dried.^20,27 The nucleic acid/Typ ratio was significantly increased for UC13-Scr cells but decreased for T24-A cells relative to their parental controls [Fig. 4(b)]. In addition, nucleic acid/Typ ratios were increased for UC13-Scr and UC13-C cells relative to T24-Scr and T24-C cells. This suggests that the parental UC13 cell line was more sensitive to shRNA-mediated gene silencing compared to the parental T24 cell line. The high nucleic acid and low protein content exhibited by UC13-Scr and UC13-C cell lines suggest a more undifferentiated cell phenotype, whereas T24-Scr and T24-C cells with their lower nucleic acid but higher protein content suggest a more differentiated cell phenotype.^20,39 Nevertheless, high nuclear DNA content was consistent with other bladder cancer cells and tissues studied by Raman spectroscopy.⁴⁰

Phospholipid metabolism, particularly phosphocholine and total choline, is known to be elevated in breast, brain, and prostate cancers because of the overexpression of choline kinase-$\alpha$ ($\mathrm{ChoK}\alpha$), the enzyme involved in their biosynthesis.⁴¹ Elevated $\mathrm{ChoK}\alpha$ and phosphocholine levels are also associated with a more aggressive phenotype for certain bladder carcinomas.⁴² As shown in Fig. 4(c), significant increases in phospholipid/cholesterol ratios were observed between parental and between some gene-silenced UC13 and T24 cell lines. This parameter was chosen because changes in phospholipid/cholesterol ratios had previously been stratified against known glioma tumor grades.²¹ In addition, a recent Raman tissue imaging study has shown that prostate cancer progression could be stratified according to accumulated cholesterol esters, which were elevated in high-grade and metastatic prostate cancers.⁴³ As shown in Fig. 4(c), phospholipid/cholesterol ratios were slightly reduced across all T24 cell lines but not across UC13 cell lines.

Although comparative literature lipogenic profiles on both UC13 and T24 cell lines are scarce, there is one study in which the de novo lipid synthesis by T24 cells were compared against other cancer cell lines.⁴² Compared to prostate (PC3M), lung (HO62), and hepatocellular (HepG2) carcinoma cells, T24 cells displayed a low lipogenic phenotype under standard cell culture conditions.⁴² This could account for the observation that T24 cell lines demonstrated a trend of lower phospholipid/cholesterol ratios compared to UC13 cells in Fig. 4(c), including why all four T24 cell lines consistently exhibited higher Typ/phospholipid ratios compared to UC13 cell lines in Fig. 4(c). Unfortunately, due to spectral overlap between the $\mathrm{C}─\mathrm{N}$ stretching mode of the polar choline group of phosphocholine and free choline components,⁴⁴ it was not possible to measure their relative intensity ratios. Although the spectral signature of free choline in cancer cells has yet to be identified directly, a recent isotope-labeled Raman imaging study showed that living cancer cells contained higher amounts of deuterated choline metabolites inside the nucleus compared to noncancer cells.⁴⁵

The surprisingly high nucleic acid/Typ and phospholipid/cholesterol ratios exhibited by UC13-Scr cells, the cell phenotype with the lowest GFP/Am I $\alpha$-helix ratio, implied that shRNA gene silencing had a more profound effect on cancer cell biochemistry than GFP expression. To determine whether gene silencing could also influence protein secondary structure, additional ratiometric parameters derived from deconvolved Am I and Am III subbands were examined. As shown in Fig. 4(e), the Am III disorder/order ratios were significantly higher for UC13-Scr cells than UC13-Con and UC13-A cells. A similar trend was found when the Am I disorder/Am I $\alpha$-helix ratios were examined, but the results were not statistically significant [Fig. 4(f)]. However, Am I disorder/Am I $\alpha$-helix ratios were found to be significantly higher for UC13-Src and T24-Scr compared to UC13-C and T24-C cells, respectively. The finding that Raman spectroscopy can detect changes in bladder cancer cell protein secondary structure is supported by other spectroscopic studies involving colorectal cancer tissues⁴⁶ and colon cancer cells.²³ Moreover, some cancer-associated proteins are intrinsically more disordered and are known to be associated with increased cancer risk.^47–49