Combining quantum cascade lasers and plasmonic metasurfaces to monitor de novo lipogenesis with vibrational contrast microscopy

2.1 Metasurface fabrication

The plasmonic nanoantennas were fabricated using electron beam lithography (EBL) on calcium fluoride (CaF₂) substrate using PMMA resist. A 5 nm chromium adhesion layer and a 70 nm gold layer were deposited sequentially using thermal evaporation. The plasmonic nanoantennas were then formed by a lift-off process in acetone. The resulting metallic nanoantennas were arranged in an array covering a 500 µm × 500 µm area, constituting the metasurface. To enable imaging in aqueous environments, the metasurface was attached to the bottom of a cell culture chamber superstructure (CS16-CultureWell Removable Chambered Coverglass, Grace Bio-Labs).

2.2 Cell culture, differentiation, and ¹³C labeling

Undifferentiated mouse embryonic fibroblast 3T3-L1 cells (American Type Culture Collection) were cultured in pre-adipocyte expansion medium consisting of Dulbecco’s Modified Eagle Medium (DMEM, Gibco) with 4.5 g/L glucose and GlutaMAX supplement, 10 % fetal bovine serum (FBS, Gibco), and 1 % penicillin/streptomycin (P/S, Gibco). Cells were maintained at 37 °C in a 5 % CO₂ incubator.

Upon reaching confluency, cells were incubated for another 48 h before switching to differentiation medium. The differentiation medium consisted of DMEM supplemented with 10 % FBS, 1 % P/S, 1 μM dexamethasone (Sigma-Aldrich), 500 μM 3-isobutyl-1-methylxanthine (IBMX, Sigma-Aldrich), 2 μM rosiglitazone (Sigma-Aldrich), and 2 μg/mL bovine insulin (Sigma-Aldrich) [59]. After 48 h in differentiation medium, the cells were transitioned to adipocyte maintenance medium, composed of DMEM supplemented with 10 % FBS, 1 % P/S, and 2 μg/mL bovine insulin, and cultured for an additional 5 days.

Then, cells were harvested using 0.25 % trypsin-EDTA (Gibco) and re-seeded onto metasurfaces. Prior to seeding the cells, the metasurface was sterilized using 70 % ethanol and coated with 0.2 % gelatin, treated overnight in the incubator in PBS. Cells were divided into eight samples, each exposed to varying durations of ¹³C glucose maintenance medium and ¹²C glucose maintenance medium. Samples were labeled day 0 to day 7, corresponding to the number of days in ¹³C glucose maintenance medium. For the ¹²C glucose maintenance medium, regular DMEM was used, while for the ¹³C glucose maintenance medium, glucose-free DMEM was supplemented with 4.5 g/L U-¹³C₆ glucose (Cambridge Isotope Laboratories) and the same concentrations of FBS, P/S, and insulin as in the ¹²C medium. Throughput the experiment, media were exchanged to fresh media containing either ¹²C glucose or ¹³C glucose at least every 2 days.

Seven days after seeding on the metasurfaces, cells were fixed with 4 % formaldehyde for 15 min, washed three times with phosphate buffered saline (PBS), and stored in PBS prior to IR measurements.

2.3 Inverted MIR microscope setup

The inverted MIR microscope used in this study has been described in detail elsewhere [37]. Briefly, the MIR beam from a quantum cascade laser (QCL, MIRcat-1400, Daylight Solutions) was focused onto the metasurface from below using a Black Diamond-2 infrared collimation lens with a numerical aperture (NA) of 0.71 (390093, LightPath Technologies). The focal spot size was approximately 5 µm. The reflected beam from the metasurface was collected by the same objective lens and detected by a liquid-nitrogen-cooled MCT detector (MCT-13-1.0, InfraRed Associates).

The output current from the MCT detector was sent to a lock-in amplifier (SR830, Stanford Research Systems) for signal demodulation. For imaging, the metasurface sample was mounted on a dual-axis motorized microscope stage (HLD117, Prior Scientific), and point-scanning was performed by moving the sample through the laser focus. Hyperspectral image cubes were acquired by collecting discrete frequency images in the range of 1,500–1,800 cm⁻¹ at 5 cm⁻¹ intervals, with 2 µm pixel size.

To reduce water vapor absorption in the MIR, the entire optical setup was enclosed in an optical enclosure purged with dried air.

2.4 Data processing and analysis

The acquired images were first processed using translational image registration in ImageJ to correct for sample drift and laser pointing fluctuations. Further data processing was performed in MATLAB. To correct for detector non-linearity, the signal intensity at each pixel was adjusted using a pre-measured calibration curve. Fluctuations in laser intensity were accounted for by mean-centering each row of the image.

Absorbance was calculated using the formula:

where R x , y , ω is the reflectance at a given pixel x , y and wavenumber ω, and R x 0 , y 0 , ω is the reflectance of a bare metasurface at a cell-free location. The hyperspectral image was subsequently de-noised using a minimum noise fraction (MNF) transform, retaining the first 20 components of the data for further analysis.

To quantify the intensity of specific vibrational peaks, a linear baseline subtraction method was applied to correct for shifts in the absorbance baseline caused by metasurface resonance variations. Two baseline points were selected on either side of each peak for the baseline correction. For the amide II peak, the absorbance at 1,555 cm⁻¹ was used to calculate the intensity, with baseline points at 1,500 cm⁻¹ and 1,600 cm⁻¹. For the ¹³C=O ester peak, the absorbance at 1,705 cm⁻¹ was used, with baseline points at 1,685 cm⁻¹ and 1,720 cm⁻¹. For the ¹²C=O ester peak, the absorbance at 1,745 cm⁻¹ was used, with baseline points at 1,725 cm⁻¹ and 1,760 cm⁻¹.

Additionally, the ¹³C=O ester peak exhibited significant overlap with the amide I peak of proteins and the H–O–H bending mode of water [8], [52]. To correct for these overlaps, the following model was used:

where B_13C=O is the apparent peak intensity calculated using the linear baseline method, and A_13C=O, A_amideII, and A_12C=O are the true peak intensities without spectral overlap. This model is based on two assumptions: (i) the spectral overlap in the amide II and ¹²C=O ester peaks is negligible; (ii) the overlap from the amide I peak is proportional to the number of amide groups whose combined bond strength is proportional to A_amideII, and the overlap from the H–O–H bending mode arises from the total lipid-associated displacement of water proportional to A 12 C  =  O + A 13 C  =  O .

The coefficients a = 0.21 and b = 0.045 were empirically determined. The coefficient a was chosen to ensure that A_13C=O was zero in cytosolic regions lacking LDs, while b was selected such that A_13C=O was zero in LDs from the day 0 sample, which was cultured entirely in ¹²C glucose medium.

3.1 MIRIAM imaging of adipocyte cells

MIR hyperspectral images of 3T3-L1 adipocyte cells were acquired using MIRIAM, a platform previously developed by our group [37]. A schematic of the experimental setup is shown in Figure 1a. The metasurface used in this study comprised an array of gold nanoantennas with dimensions of 200 nm in width, 1.8 µm in length, and 2.7 µm periodicity in both horizontal and vertical directions (Figure 1b). When excited with incident light polarized along the long axis of the nanoantennas, the metasurface exhibited a resonance mode at approximately 1,650 cm⁻¹, with a pronounced peak in reflectance (Figure 1c). The resonance, characterized by a full-width at half maximum (FWHM) of 300 cm⁻¹, covered multiple vibrational modes within the 1,500–1,800 cm⁻¹ range. This single-resonance metasurface design was employed to probe key vibrational modes associated with adipocytes, including amide II (1,555 cm⁻¹) and amide I (1,655 cm⁻¹) for proteins, and ¹³C=O ester (1,705 cm⁻¹) and ¹²C=O ester (1,745 cm⁻¹) for lipids.

Figure 2a–c displays the raw signal, reflectance spectra, and absorbance spectra, respectively, from LD and cytoplasm within a cell. The positions at which the spectra are obtained are shown in Figure 2d, with a representative bright field image of fixed 3T3-L1 cells on the metasurface. LDs within the cells are clearly visible due to the high refractive index contrast between lipids and water. The gold nanoantennas of the metasurface appear as a periodic pattern in the background.

Figure 2:

MIR spectra and vibrational contrast images of 3T3-L1 adipocytes acquired through MIRIAM. (a) Raw reflection signal at different locations. “Bare MS” refers to a metasurface region without cells, while “Gold Ref.” indicates a gold patch used for reflectance reference measurements. Cytoplasm and LD spectra are obtained at positions indicated by the arrow in (d). Sharp peaks in the spectra correspond to atmospheric water vapor absorption. (b) Reflectance spectra at different positions. (c) Absorbance spectra at LD and cytoplasm position. The black dotted line shows the linear baseline correction used to determine the ¹²C=O ester intensity. (d) Bright field image of 3T3-L1 cells on the metasurface. White dashed curve indicates the boundary of an undifferentiated fibroblast cells, barely visible in bright field image. (e) Phase contrast image of 3T3-L1 cells on the metasurface. (f) MIRIAM image at amide II band (1,555 cm⁻¹). (g) MIRIAM image at ¹²C=O ester band (1,745 cm⁻¹). Scale bar: 100 µm.

The intensity variation in the raw signal spectra (Figure 2a) results from the wavelength-dependence of QCL emission (slow variation) and atmospheric water vapor absorption (sharp dips). These variations are normalized by collecting a reference spectrum from a 70 nm thick gold patch adjacent to the metasurface, yielding the reflectance spectra shown in Figure 2b. Although our QCL-based measurement lacks the spectral range to fully capture the metasurface’s resonance mode (see FTIR measurement in Figure 1c for the full reflectance spectrum), variations in reflectance due to the metasurface resonance and a dip corresponding to the H–O–H bending mode of water are evident.

To isolate the vibrational signature of the cells, absorbance spectra are calculated using a reference spectrum collected from a bare metasurface position (Equation (1)). Note that since bare metasurface (in PBS) is used as a reference, all absorbance spectra shown in this work are differential absorbance from this reference. The absorbance spectra reveal vibrational modes corresponding to cellular biomolecules, including the amide I (1,655 cm⁻¹) and amide II (1,555 cm⁻¹) modes from cytoplasmic proteins, as well as the ¹²C=O ester mode (1,745 cm⁻¹) from lipids in the LDs (Figure 2c). In MIRIAM, the vibrational signals of analytes are superimposed on the metasurface resonance spectrum. This results in a varying baseline in the absorbance spectrum due to shifts in the metasurface resonance when analytes interact with the near-field. To quantify the vibrational signals, a linear baseline subtraction was applied, as described in the Experimental section. Figure 2c shows the baseline subtraction for ¹²C=O ester mode, as an example.

MIRIAM vibrational contrast images of 3T3-L1 adipocyte cells in the amide II and ¹²C=O ester bands are shown in Figure 2f and g, with the corresponding bright field and phase contrast images shown in Figure 2d and e. The amide II channel, derived from the protein-associated vibrational contrast, exhibits distinctly different spatial distribution compared to the ¹²C=O ester channel derived from the lipids-associated contrast. This demonstrates the specificity of MIRIAM for detecting distinct molecular vibrations with minimal crosstalk between channels. Owing to the limited penetration depth (∼100 nm) of the plasmonic near-field surrounding the antennas, the protein signal predominantly originates from the focal adhesion sites and the cytoskeletal layer around the cell cortex. Protein contrast highlights cells spreading on the metasurface, with individual cells clearly discernible. Often, there are undifferentiated fibroblast cells mixed with the adipocyte cells. The fibroblast cells can be thinly spread (Figure 2d–f, white dashed curve), making them difficult to identify with brightfield (Figure 2d) or phase contrast microscopy (Figure 2e). These cells are clearly visible with MIRIAM in the amide II channel (Figure 2f), highlighting the excellent surface sensitivity of MIRIAM. The noise in the MIRIAM images is estimated to be 0.001 in absorbance unit, based on the standard deviation in a cell-free region. This results in a typical contrast-to-noise ratio in the range of 20–40 in the amide II channel, depending on the adhesion strength of each cell.

The signal from ¹²C=O ester channel primarily originates from lipids in the LDs. However, many LDs visible in bright field microscopy (Figure 2d) are not seen in the MIRIAM images. This is attributed to the surface sensitivity of MIRIAM, which renders LDs positioned far from the metasurface essentially invisible.

Although previous studies have demonstrated MIRIAM’s capability to image living cells at a few discrete frequencies [37], the current acquisition speed of ∼5 min for a 500 µm × 500 µm image makes hyperspectral imaging of living cells impractical, especially when collecting data for 10 or more frequencies. Movement of living cells during the measurement introduces spectral inaccuracies, particularly for small structures like LDs. To address this limitation, this study focused on imaging fixed cells while keeping the cells hydrated in PBS to avoid disruptions in cellular structures and morphology.

To study the glucose metabolism and DNL in 3T3-L1 adipocytes, eight samples of 3T3-L1 adipocyte cells were prepared with varying exposure times to ¹³C glucose (Figure 3). Preadipocyte cells were chemically induced to differentiate into adipocytes and cultured for 5 days in a flask until LDs were visibly formed. At this point, the cells were reseeded onto metasurfaces and incubated in maintenance media with differing exposure times to ¹³C glucose (0–7 days). This design ensured that each sample are cultured in the adipocyte maintenance medium for the same amount of time and form LDs with comparable sizes, while varying the relative amounts of ¹³C=O ester and ¹²C=O ester in LDs due to differential exposure to ¹³C glucose and ¹²C glucose.

Figure 3:

Experimental timeline for differentiation and ¹³C glucose feeding of 3T3-L1 adipocytes.

Representative MIRIAM-based MIR images of samples with 0 days (no exposure) and 7 days of ¹³C glucose exposure are shown in Figure 4, and the corresponding bright field images are presented in Figure S2. Despite the limited visibility of LDs due to the shallow penetration depth of metasurface, sufficient LDs were visible to quantify ¹²C=O and ¹³C=O ester signals. In the 0-day sample, LDs exhibited no detectable ¹³C=O ester signal, as expected, with all lipids derived from ¹²C glucose. In contrast, the 7-day sample showed a marked increase in the ¹³C=O ester signal and a corresponding decrease in the ¹²C=O ester signal within LDs. The colocalization of ¹³C=O and ¹²C=O signals confirmed that both signals originated from the same LDs, with the ratio reflecting the time-dependent increase of ¹³C incorporation into the LDs with prolonged ¹³C glucose exposure.

Figure 4:

Fixed 3T3-L1 adipocyte IR absorbance images acquired by MIRIAM. (a) 0-day exposure to ¹³C glucose and (b) 7 days exposure to ¹³C glucose. Absorbance images are presented with contrast in amide II (left column), ¹²C=O ester (middle column), and ¹³C=O ester (right column) of (a, b). Increase in ¹³C=O ester and decrease ¹²C=O ester in the 7 days sample can be clearly seen in (b). Scale bar: 200 µm.

We note that the combination of the inverted QCL-based microscope and the metasurface enabled diffraction-limited imaging of sub-cellular LDs inside hydrated cells. Earlier attempts at spectrochemical imaging using commercial widefield MIR microscopes worked exclusively with dried samples and encountered coherence-related speckle artifacts in transmission-mode images [7], [8]. By using a plasmonic metasurface in addition to a widefield MIR microscope, Rosas et al. demonstrated enhanced MIR spectra of dried tissue samples in both transmission and reflection mode, but no applications to hydrated samples were reported [9]. Likewise, confocal MIR microscopy without a metasurface produced speckle-free vibrational images with diffraction-limited resolution, but only in the transmission mode [5], [6], which cannot be used for imaging hydrated cells.

3.2 Tracking ¹³C incorporation in lipids and proteins through IR spectra

To investigate the temporal evolution of ¹³C glucose metabolism, mean absorbance spectra were extracted from each sample (Figure 5), focusing on two distinct regions: LDs (Figure 5a) and the cytoplasmic region outside of the LDs (Figure 5b).

Figure 5:

Mean IR spectra from (a) LDs and (b) cytoplasm regions without LDs, for samples with varying exposure times to ¹³C glucose. Increases in the ¹³C=O ester peak (1,705 cm⁻¹) and the ¹³C amide I peak (1,615 cm⁻¹) are observed with longer exposure to ¹³C glucose medium.

In the LD spectra, increasing exposure time to ¹³C glucose led to a pronounced increase in the ¹³C=O ester peak at 1,705 cm⁻¹ and a corresponding decrease in the ¹²C=O ester peak at 1,745 cm⁻¹. This shift was most prominent in the samples with 5–7 days of exposure. Additionally, all spectra displayed variable but significant amide I and amide II peaks, likely attributed to the protein from cell cortex between the LD and the metasurface. The irregular shape of the amide I peak at 1,655 cm⁻¹ is partially explained by its overlap with the H–O–H bending mode of water. LDs displace water in their vicinity, resulting in reduced absorbance from the H–O–H bending mode relative to the bare metasurface used for background measurements.

In the cytoplasmic spectra (Figure 5b), the amide I and II peaks dominate, as expected. However, in samples with the highest ¹³C incorporation (5–7 days), a small peak emerges at 1,615 cm⁻¹, attributed to the ¹³C amide I peak [8], [60]. Additionally, the amide II peak exhibits a slight shift of approximately 5 cm⁻¹ in these samples, also reflecting the incorporation of ¹³C into amino acids. These observations suggest substantial de novo amino acid biosynthesis using glucose as the carbon source.

Glucose metabolism is known to be elevated in adipocyte cells, consistent with their role in lipid storage [61]. Our findings indicate that carbon from glucose metabolism is utilized not only for DNL but also for the synthesis of amino acids. This likely involves the tricarboxylic acid (TCA) cycle, which provides precursors such as α-ketoglutarate for glutamate and oxaloacetate for aspartate [62]. This coupling between the de novo synthesis of fatty acids and amino acids highlights a metabolic interdependence that warrants further investigation.

3.3 Temporal dynamics of de novo lipogenesis

To evaluate the temporal dynamics of DNL, we quantified the rate of lipid synthesis by calculating the ratio of the ¹³C=O peak intensity to the total lipid content (defined as the sum of the ¹²C=O and ¹³C=O peak intensities). We note that in most regions of the cells, this ratio is very noisy, since the amount of ¹³C=O and ¹²C=O is too small to be detected outside the LDs (Figure S1). However, this ratio has sufficient signal-to-noise ratio within the LDs. In order to choose the appropriate pixels for analysis, a threshold mask was applied to the ¹²C=O image to identify LDs with a strong ¹²C signal, and the ratio was calculated using only the selected pixels. Approximately 100 pixels were analyzed per sample, representing 5–10 LDs per sample. The results are presented in Figure 6.

Figure 6:

Ratio of ¹³C-labeled lipids to total lipids (¹²C + ¹³C) for different samples with varying exposure time to ¹³C glucose. The horizontal axis represents the number of days in ¹³C glucose, instead of ¹²C glucose, environment, with the total LD maturation time the same across all samples (see Figure 3). The control sample was cultured entirely in ¹²C glucose environment. The intensities of ¹³C and ¹²C lipids were extracted from MIRIAM images, using approximately 100 pixels per sample, selected from regions with the highest ¹²C=O ester intensity. These pixels represent roughly 5–15 LDs from 5 to 9 cells per sample.

The ratio of ¹³C=O to total lipids gradually increased over time, reaching a value of 0.51 ± 0.05 after 7 days of exposure to ¹³C glucose. This ratio is higher than reported in a previous study by Shuster et al. on DNL in 3T3-L1 cells [52], which may be attributed to a number of differences in the experimental protocols. Most importantly, we used longer exposure times (up to 7 days) compared to the 3-day exposure in the earlier study by Shuster et al., which likely resulted in greater ¹³C incorporation in the LDs. Even though the cells have been exposed to ¹³C glucose for a long time (7 days), there is still significant amount of ¹²C=O in the LD. This can be explained by alternate carbon sources that the cells may be using for DNL, which include glutamine and free fatty acids [50], [61], which are not labelled with ¹³C.

Interestingly, the rate of increase in the ¹³C=O ratio appeared to slow down after the first three days. The sample with 4 days of ¹³C glucose exposure exhibited significantly lower ¹³C=O ratios compared to the sample with 7 days of exposure, the difference between the two being that the 4 days sample was in ¹²C-glucose medium for the first three days after the cells were seeded on the metasurface, whereas the 7 days sample was in ¹³C-glucose during this period (see Figure 3). This observation may be explained by a decline in the rate of DNL lipogenesis as LDs mature over time. Alternatively, the observed slowing could be due to an accumulation of ¹²C=O lipids. As the LD size increases, the rate of change in the ¹³C=O-to-total-lipid ratio is expected to decrease even if the absolute rate of ¹³C=O generation remains constant. However, our technique’s limited penetration depth, determined by the plasmonic hotspot, prevents accurate measurement of individual LD sizes, and this hypothesis warrants further investigation.

We also observed considerable variability in the ¹³C=O-to-total-lipid ratio across the samples (as seen from the spread in the data in Figure 6). While Shuster et al. attributed this variability to spatial heterogeneity in ¹³C incorporation [52], we suggest that in our case, this spread is more likely due to experimental error in quantifying the ¹³C ratio. Similar variability was observed across samples with different ¹³C glucose exposure times, which suggests that the variation is not driven by biological differences. Potential sources of error include laser intensity fluctuations, environmental water vapor variability, or inaccuracies in quantifying the ¹²C=O and ¹³C=O intensities caused by spectral artifacts and distortions. Further refinements in experimental design and signal processing may help address these limitations.