The effect of optical substrates on micro-FTIR analysis of single mammalian cells
- First Online:
- Cite this article as:
- Wehbe, K., Filik, J., Frogley, M.D. et al. Anal Bioanal Chem (2013) 405: 1311. doi:10.1007/s00216-012-6521-6
- 1.1k Downloads
The study of individual cells with infrared (IR) microspectroscopy often requires living cells to be cultured directly onto a suitable substrate. The surface effect of the specific substrates on the cell growth—viability and associated biochemistry—as well as on the IR analysis—spectral interference and optical artifacts—is all too often ignored. Using the IR beamline, MIRIAM (Diamond Light Source, UK), we show the importance of the substrate used for IR absorption spectroscopy by analyzing two different cell lines cultured on a range of seven optical substrates in both transmission and reflection modes. First, cell viability measurements are made to determine the preferable substrates for normal cell growth. Successively, synchrotron radiation IR microspectroscopy is performed on the two cell lines to determine any genuine biochemically induced changes or optical effect in the spectra due to the different substrates. Multivariate analysis of spectral data is applied on each cell line to visualize the spectral changes. The results confirm the advantage of transmission measurements over reflection due to the absence of a strong optical standing wave artifact which amplifies the absorbance spectrum in the high wavenumber regions with respect to low wavenumbers in the mid-IR range. The transmission spectra reveal interference from a more subtle but significant optical artifact related to the reflection losses of the different substrate materials. This means that, for comparative studies of cell biochemistry by IR microspectroscopy, it is crucial that all samples are measured on the same substrate type.
KeywordsSingle-cell analysisSynchrotron radiation IR microspectroscopyIR optical substratesTransmissionReflectionPCA
Bovine serum albumin
Fetal bovine serum
Mercury cadmium telluride
Multimode InfraRed Imaging And Microspectroscopy
The use of infrared (IR) spectroscopy for studying biological cells is nowadays a wide and active area of research. Specifically, synchrotron radiation (SR) IR microspectroscopy, giving the high spatial resolution and signal-to- noise necessary for single-cell analysis, has proved to be an ideal tool for investigating the biochemical composition of biological samples at the microscopic and molecular scale [1–3]. It has been shown that synchrotron-based Fourier-transform infrared (FTIR) microspectroscopy has no cytotoxic effects on examined cells as no detectable biochemical changes between control and exposed cells have been found despite the increased power density of the SR light . IR spectral differences have been reported between cancerous and normal cells [5, 6], between cells in different growth stages [7–9], or as an effect of drugs on cells [10–13]. There is no unique substrate used in all these studies, and no comparison has been reported in literature of which one is more suitable for cell growth and IR analysis. If the substrate has any influence on the measurements, either through biochemical or morphological changes in the cells, or through systematic variation in the spectral data via optical artifacts, it is clear that these effects would need to be identified and that standardization would be required to allow direct comparison between different works.
Cell adhesion on the substrate is a key aspect for cellular morphology, proliferation, and function. Poly-l-lysine, laminin, fibronectin, collagen, and other components are used as substrate coatings to enhance cell adhesion, but they may interfere with the cell spectra. A study on cancerous cells done by Draux et al.  compared three substrates used for Raman microspectroscopy (Quartz, ZnSe, and CaF2). This study revealed that quartz and CaF2 were much better for cell growth than ZnSe, which showed a very weak cell adherence due to its toxicity. Another study by Meade et al.  on keratinocytes compared MirrIR and quartz substrates by IR and Raman spectroscopy but using three different coatings (laminin, fibronectin, and gelatin) for cell adhesion. This study showed that functional changes regarding proliferation and viability as well as spectral changes were induced and could influence the spectroscopic measurement. In a previous study by Carter et al. , Si3N4 was shown to be suitable for cell growth for FTIR and XRF analysis. There are no reports comparing a wide range of IR optical substrates and studying their direct effect on cell growth.
In general, metal substrates are used for reflection measurements and inorganic crystals for transmission. These materials all have different surface chemistries, some being highly biocompatible and others, potentially toxic. This raises a simple question: Do the different substrates interfere chemically with the cell growth? The effect from these materials with biological samples is likely to be minimal when tissue sections or cells are deposited on the surface but may be crucial when cells are grown for several hours or days and then fixed before IR analysis. There may be an interaction of the substrate material with the culture medium or even the fixatives. Thus, the underlying chemical interface layer could play an important role for viability and morphology of cells growing on these substrates.
The reflection measurement geometry has the advantages of a stronger absorption (due to the doubling of the path length) and cost-effective substrates. MirrIR- and aluminum-coated glass slides are typically used for IR reflection (also known as transflectance) measurements when studying cells. MirrIR slides, glass slides with reflective multilayer coating (Ag/SnO2), are especially popular because samples can be examined by conventional light microscopy in transmission and then scanned by IR in reflection mode.
In the transmission geometry, a wide variety of substrates can be used. These optical materials have different spectral ranges, e.g., CaF2 (0.35 to 10 μm wavelength) versus Si (1.2 to 15 μm wavelength) and also different refractive indices with associated reflection losses at the substrate interfaces.
The use of these two IR geometries, each with different substrates with different chemical and optical properties, raises further questions such as: Is there any difference of IR spectra between transmission and reflection measurements for the same type of cells, and do the cells grow well and in the same way on all of these substrates? To answer these questions, in this study, a wide range of IR optical substrates, CaF2, Si, ZnSe, BaF2, and ZnS for transmission and MirrIR and Al slides for reflection were compared using two cell lines. No additional coatings were applied to the substrates; the cells were grown directly on the surface to study the immediate effect of each IR optical substrate on the cell growth and biochemistry. The two cell lines selected are both adherent cell lines but from different origins. Chinese hamster ovary cells (CHO-K1) are epithelial-like and one of the most used mammalian cell lines in biological and medical research. Colorectal adenocarcinoma cell line (DLD1) is a human colon cancer cell line used as an example of cancerous cells.
Materials and methods
Cells were cultured in plastic culture flasks (polystyrene) using Hamm’s F12 medium (Sigma-Aldrich) for CHO-K1 and RPMI 1640 medium (Gibco) for DLD1. Both media were supplemented with 10 % FBS, 1 % l-glutamine, and 1 % penicillin/ streptomycin (all from Gibco, Invitrogen). Cells were maintained in a humidified atmosphere in a 37 °C incubator supplied with 5 % CO2. Before reaching confluence, cells were detached using trypsin–EDTA 0.25 % (Gibco) and then centrifuged. The pellet was collected, resuspended in culture media, and then seeded on different IR optical substrates for transmission, i.e., CaF2, Si, ZnSe, BaF2, and ZnS (Crystran, UK) and reflection measurements, i.e., MirrIR slides (Kevley Technologies, OH, USA) and Al slides (Thermofisher, UK). All substrates were cleaned with 70 % ethanol before being used for cell culture. Cells were seeded at a concentration of 5 × 104 cells/ml of medium. After 48 h incubation, cells were washed with NaCl 0.9 % and fixed with 4 % formalin (Sigma-Aldrich) for 30 min, washed with distilled water, and then dried before analysis under the IR microscope. For further viability comparison on different substrates, other sets of cells were fixed with ice-cold acetone before staining for epifluorescence observation.
Epifluorescence with DAPI and PI staining
For morphological and viability observation, parallel series of cells were stained with propidium iodide (PI) and 4′,6-diamidino-2-phenylindole (DAPI). After 48 h culture on different substrates, cells were rinsed with phosphate-buffered saline (PBS) 1×, fixed with ice-cold acetone at −20 °C for 10 min, and then washed with PBS. Cells were then equilibrated with 2× SSC (0.3 M NaCl, 0.03 M sodium citrate, pH 7.0; Gibco, Invitrogen). Cells were incubated with the dilute PI stain (Molecular probes, Invitrogen) for 1–5 min (500 nM solution of PI by diluting the 1 mg/ml corresponding to 1.5 mM stock solution 1:3,000 in 2× SSC). Cells were then rinsed three times in 2× SSC and mounted with the Prolong Gold antifade with DAPI reagent (Molecular probes, Invitrogen) and coverslipped. Samples were viewed using the fluorescence microscope (Zeiss Axio-imager M1) with the appropriate excitation/detection filters.
FTIR data acquisition and analysis
Single CHO-K1 and DLD1 cells grown on IR optical substrates were analyzed in the mid-IR range (4,000–600 cm−1) on the (Bruker) Vertex 80 V FTIR spectrometer available at the IR Beamline B22 (MIRIAM) in Diamond Light Source, UK . The spectra were measured using the LN2 cooled MCT broadband (>500 cm−1) detector (100 × 100 μm2 area), coupled to the Hyperion 3000 microscope and the SRIR source. The aperture size at the sample of 15 × 15 μm2 was used to collect spectra from single isolated cells at 4 cm−1 spectral resolution and 256 scans using the ×36 (0.5 NA) objective (matched with a ×36 condenser for transmission measurement).
All data acquisition was performed using OPUS 6.5 software (Bruker). Selection of spectra for data treatment was based on eliminating those with very weak absorbance (poor S/N ratio). Between 60 and 70 spectra were analyzed on each substrate for each cell line. Data analysis was performed in the Unscrambler X 10.1 software, taking second derivative spectra (Savitzky-Golay second order) to remove slowly varying baseline effects and then normalized using the standard normal variate (SNV). SNV is an analytical transformation applied to spectra to remove multiplicative interferences of scatter effects by centering and scaling each individual spectrum using only the data from that spectrum and not the mean spectrum of the set. Principal component analysis (PCA) was performed for each cell line population and then for both cell lines grouped together, using the nonlinear iterative partial least squares algorithm and leverage correction validation method.
Although the nucleic acids region (1,150–1,010 cm−1) could be identified in spectra from most of the substrates, the mid-IR spectral region (3,800–1,100 cm−1) excluding the CO2 region (2,400–2,100 cm−1) was used for PCA, since it includes most of the normal vibration modes of the common biological molecules (proteins, lipids, etc.). This choice allowed making a fair comparison between the substrates because the nucleic acid region can be strongly affected by the different spectral IR bandwidth of the materials (CaF2 cuts off at 1,000 cm−1, transmission range 0.35–10 μm wavelength).
Results and discussion
Cell growth and morphology observation
IR microspectroscopy of cells on different substrates
Amide I peak position of the average cell spectra on each optical substrate
Combined PCA of T and R cell spectra
PC1 scores for CHO-K1 and DLD1 cells on all substrates
PC1, average ± SD
PC1, average ± SD
−8.3 ± 1.8
−8.0 ± 2.1
−7.8 ± 1.1
−7.6 ± 1.5
−7.5 ± 0.8
−5.6 ± 2.2
13.8 ± 2.5
10.4 ± 3.0
9.2 ± 3.7
9.5 ± 3.7
PCA of the T cell spectra
The loadings vector of PC2 (Fig. 8c) shows a difference in lipid absorption between the two cell lines as illustrated in the graph of the second derivatives for the average spectra (Fig. 8d): Here the absorbance of DLD1 cell spectra is higher than in CHO-K1 cell spectra, thus suggesting higher lipid content in DLD1 cells. This distinction between cell lines is consistent across all substrate materials and therefore clearly reflects a true biochemical difference. It is unlikely that using different growth media in the cell culture process could influence the separation of the two cell lines. Harvey et al.  studying the IR spectral signatures of different prostate cell lines confirmed that different growth media used for culturing the cells did not significantly influence the chemometric discrimination. This was not investigated in the present work as differentiation between the two cell lines CHO-K1 and DLD1 is not the main purpose.
Reflection loss and optical artifact in cell transmission spectra
It is clear from Fig. 7 that PC1 scores discriminate the cell spectra mostly by their different substrates. CHO-K1 and DLD1 are two separate groups, but they are distributed along the PC1 axis depending on the Si, ZnS, and CaF2 material substrate, in order of decreasing refractive index.
The IR experiments in transmission had the sample illuminated through the substrate. It is expected that reflection losses related to the relative refractive indices at the substrate–sample interface play a major role. However, there are other interfaces—namely air–substrate and sample–air—whose optical contributions have to be considered.
Figure 10a shows the measured A*Substrate − A*CaF2 difference spectra for the cell samples along with the calculated −log(1-R12) spectra for BSA on each substrate, using the BSA–substrate relative refractive indices and Fresnel’s equations. Along the entire mid-IR spectral range, there is a close match between the cell absorbance difference A*Substrate − A*CaF2, for both CHO-K1 and DLD1 cell lines and the reflection loss estimation −log(1-R12). Also, the reflection loss amplitude scales correctly with the substrates shown, i.e., lower amplitude for ZnS and higher for Si. This confirms that the reflection loss and related optical artifact is clearly responsible for the substrate-dependent spectral changes.
Finally, the measured IR spectra difference needs to be compared with the findings from the principal component analysis on transmission data, namely PC1 loading vector. The substrate type discrimination based on PC1 was performed on the second derivative spectra, thus the actual PC1 loading vector has been integrated twice1 to recollect the absorbance information. The result is plotted in Fig. 10b, together with the experimental absorbance difference of cells on Si versus CaF2 substrate. Again, there is a striking match between the loading vector of PC1 and the difference spectra A*Substrate − A*CaF2 across the whole spectral range.
The derivate-like signal of this reflection loss is responsible for the shifting of the amide I peak in the T data shown in Table 1, but it also explains the shift between the T and R spectra and smearing of the reflection data across PC2 in the combined PCA of R and T data of Fig. 5a, c. In an IR reflection measurement, the signal consists mainly of the light transmitted through the sample and reflected off the mirror substrate (transflected), but there is also a detected reflection at the sample top surface  which is the −log(1-R23) term in Eq. 2. The top-surface reflected light has the same spectral features as the transmission artifact because both depend on the refractive index of the sample. It is the detection of this light that causes the peak shift between the T and R data. Further to this, cells of different thickness would show a different ratio of reflection and transflected signals, since the latter has a linear dependence with optical path in the sample via the absorbance while the former does not change. In practice, it is this that can cause the wide spreading of the reflection data in the combined PCA analysis as shown in Fig. 5.
Several optical artifacts have been treated before. Bassan et al.  discussed dispersion artifacts in transflectance IR data due to sample optical density and index variation. Attention has been particularly given in the IR literature to the Mie scattering artifact and software correction for single-cell analysis both in non-resonant  and resonant formalism , with dispersive effects shown through the scattering amplitude dependence on the relative refractive index change between scattering object and surrounding, e.g., nucleus and cell cytoplasm. Miljkovic et al.  have revised and applied the phase correction method onto general line shape distortions in IR spectra. All these approaches rely on iterative algorithms where phenomenological parameters have to be optimized, e.g., the IR signal is fitted in terms of transmission and reflection components , or refractive index and sphere radii in the extended multiplicative signal correction , or the convergence from the reference IR spectrum , or the best-phase angle . In this work, with no free parameters, our model can quantitatively account for the reflection losses in IR transmission data using Fresnel’s equations and via the Kramers–Kronig transformation of experimental reflectivity data.
Correction of the transmission spectra
For accurate spectra and absolute IR peak positions for cells on different transmission substrates, it would be necessary to correct the IR absorbance for the reflection losses as accounted by Fresnel’s equations.
First, the refractive index spectrum of the sample is obtained, which for practical purposes could be via Kramers–Kronig transformation of the reflectance IR spectrum of a thick sample of the same cell line(s) used in the experiment. Ideally, an IR measurement on CaF2 substrate will avoid any back-reflected components from the sample–substrate interface.
Secondly, the reflection losses from the sample–substrate and sample–air interfaces, i.e., −log(1-R12) and −log(1-R23) are calculated, using Fresnel’s equations (Eqs. 3 and 4), including the sample–substrate and sample–air relative refractive index spectra.
Such reflection losses are finally subtracted from the raw absorbance data before normalization, since the losses are independent of sample thickness.
Any further treatment of the data to account for, e.g., Mie scattering can then be performed as required.
The first objective of this work was to assess the cell viability on some of the most used IR substrates. The results show that BaF2 and ZnSe are not suitable for cell growth due to low cell viability. Three other IR transparent materials—Si, ZnS, and CaF2—are biochemically compatible for cell growth as they proved a high percentage (above 90 %) of viability and similar morphology of cells to standard polystyrene culture flasks. MirrIR- and Al-coated glass slides typically used for IR reflection mode are suitable for in situ cell culture.
In the analysis of IR spectra of single cells in transmission and reflection on the IR materials above and for two cell lines, no substrate-induced biochemical variations could be revealed. IR data for single cells in transflectance confirm that the standing wave artifact plays the major role, and such absorption spectra are affected by a dramatic non-linear dependence in absorption with sample thickness in between the fingerprint and the H stretching region, respectively below and above 2,000 cm−1.
In transmission, the IR spectral discrimination is dominated by an optical artifact due to the substrate reflectivity, which depends on the relative refractive index ratio sample–substrate. This is also the cause of the minimal shift of main absorption bands (such as the amide I) to higher wavenumbers with increasing substrate refractive index. A model based on Fresnel’s equations explaining in detail the phenomenon and quantifying the effect in the mid-IR spectral region is proposed and compared well with the experimental data. Out of the three transmission substrates that could be used for growing cells for FTIR microspectroscopy, CaF2 has the less reflective loss at the substrate–sample interface. If the interest is in the lower wavenumbers, e.g., DNA-RNA region below 1,000 cm−1, ZnS is preferable and offers a wider transmission range at the cost of some more spectral distortion of this kind. In general, for comparative cells studies by IR microspectroscopy in transmission when IR peak position is not crucial, it is sufficient that all samples are measured on the same substrate type. For accurate and absolute peak positions, it would be necessary to correct the transmission spectra for the reflection losses, for example, via Fresnel’s equation as used in this work. These results may help the IR biomedical community to make a proper choice of the substrate for cells experiments, ideally in view of a standardization of the FTIR protocol for all researchers interested in studying cells by FTIR microspectroscopy.
This procedure is preferred since integration does not enhance numerically the spectral noise.
We thank Dr. Stanley Botchway (CLF, Harwell, UK) for kindly providing us with the CHO-K1 cells and Dr. Giuseppe Bellisola (University of Verona, Italy) for his kind gift of the DLD1 cells. We also deeply thank Dr. Josep Sulé-Suso (Keele University, UK) for his great scientific input and Dr. Jacek K. Pijanka (formerly Diamond Light Source, UK) for his helpful assistance with the fixation procedure for the cells. This work is part of the in-house research program within the IR beamline (MIRIAM) at Diamond Light Source, UK.
This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.