Lipid Resolution in Living Cells: A New Breakthrough in Hyperspectral Mid-Infrared Photoacoustic Microscopy

Lipids are not only structural components of cell membranes and energy storage molecules, but they are also closely related to the occurrence and development of cancer, obesity, diabetes, cardiovascular diseases, and neurodegenerative diseases. However, directly observing and distinguishing different types of lipids in living cells has long faced technical challenges. Traditional fluorescent labeling methods are limited by labeling efficiency, specificity, and potential interference with cellular functions, while label-free optical techniques often struggle to distinguish lipid molecules with similar chemical structures.

Nature Methods has published a study introducing a technology called "Hyperspectral Fingerprint-region Photoacoustic Microscopy" (hyFOPM). By utilizing the single-bond vibration modes in the mid-infrared fingerprint region, this technology enables label-free detection and dynamic imaging of sphingomyelin (SM) and cholesterol (Chol) in living cells.

Technical Principles
Most label-free optical methods rely on signals in the C-H stretching vibration region (approximately 2800–3000 cm⁻¹), but the spectral bands in this region are highly similar across various lipids, making it difficult to distinguish between different types. In contrast, the mid-infrared fingerprint region (900–1730 cm⁻¹) contains more single-bond vibration information reflecting the unique structure of molecules, such as the characteristic absorption of amide bonds, ester bonds, and steroid rings.

The design of the hyFOPM system centers on this concept. It employs a tunable quantum cascade laser as the excitation source, covering the 900–2932 cm⁻¹ range with a spectral resolution of 2 cm⁻¹. Laser pulses excite the sample to generate photoacoustic signals, which are detected by an ultrasonic transducer to construct hyperspectral images. The system has a spatial resolution of approximately 4.3 μm, enabling imaging at the level of living cells.

Validation of Lipid Models
To verify the feasibility of the technology, the research team first prepared two-dimensional lipid solution models containing cholesterol (Chol), unsaturated phosphatidylcholine (DOPC), and sphingomyelin (SM).

(1)Spectral Feature Comparison

The fingerprint region spectra collected by hyFOPM are highly consistent with ATR-FTIR results. The three lipids exhibit distinguishable spectral peaks: cholesterol presents a strong absorption peak for steroid ring deformation at 1056 cm⁻¹; DOPC features C=O stretching vibration of the ester group at 1731 cm⁻¹; and sphingomyelin corresponds to the amide I band, amide II band, and fatty acid CH₂ bending vibration at 1645 cm⁻¹, 1555 cm⁻¹, and 1464 cm⁻¹, respectively.

(2)Spectral Unmixing and Classification Capability
When using only 15 wavenumbers in the fingerprint region for linear unmixing, the crosstalk between cholesterol and sphingomyelin is close to 0%, while the crosstalk for DOPC is 23%. In contrast, the crosstalk increases significantly when using 7 wavenumbers in the C-H stretching region. Further application of linear discriminant analysis (LDA) shows that the average classification accuracy reaches 96% when using either the fingerprint region or the C-H region, and reaches 97% when using all wavenumbers.

(3)Giant Unilamellar Vesicle (GUV) Models
The study prepared three types of GUVs to simulate cell membranes: Model 1, a 1:1 mixture of SM and Chol, forming a dense ordered membrane; Model 2, a 2:2:1 mixture of DOPC, SM, and Chol, coexisting in liquid-ordered and liquid-disordered phases; and Model 3, pure DOPC, forming a disordered fluid membrane. The images acquired by hyFOPM at 2852 cm⁻¹ are morphologically consistent with those obtained by Nile Red fluorescent staining. The spectral characteristics of different vesicles correspond to pure lipids, confirming that individual components in mixed membranes can be identified.

(4)Quality Control Applications
By performing spectral measurements on 10 different GUVs for each type and plotting ternary phase diagrams, the research team found that the actual lipid composition deviated from the target ratio (a discrepancy of approximately 40%). This indicates that hyFOPM can be used for quality assessment in GUV preparation.

Applications in Living Cells
The study further applied hyFOPM to living cells, observing the dynamic changes of sphingomyelin and cholesterol in two cell models, respectively.

(1)Sphingomyelin Accumulation in A549 Cells
Human lung adenocarcinoma cells (A549) were treated with the antitumor compound 2-hydroxyoleic acid (2-OHOA), which is expected to induce the accumulation of sphingomyelin. Fingerprint region spectra (1600–1400 cm⁻¹) were collected from 50 cells, showing that the 1464 cm⁻¹ peak area increased by 117% after treatment, compared to only 23% in the control group over the same period. Subsequently, imaging was performed on 3000 cells using only four wavenumbers (2852 cm⁻¹ for total lipids, 1540 cm⁻¹ for protein amide II, 1464 cm⁻¹ for sphingomyelin, and 1048 cm⁻¹ for cholesterol). The results showed that the sphingomyelin signal continued to rise at 48 and 72 hours post-treatment, while the cholesterol signal showed no significant change.

(2)Cholesterol Loading in HEK Cells
Human embryonic kidney cells (HEK293) were co-incubated with a methyl-β-cyclodextrin-cholesterol complex (MβCD-Chol) to increase cholesterol in the cell membrane. Fingerprint region spectra of 50 cells showed that the 1048 cm⁻¹ peak area increased by 161% after treatment, while the 1464 cm⁻¹ peak for sphingomyelin decreased slightly—consistent with the known property that cyclodextrin extracts some membrane lipids while delivering cholesterol. Multi-wavenumber imaging of 3000 cells further confirmed the elevation of the cholesterol signal, with a slight increase in the total lipid signal and little change in the protein signal.

Significance and Outlook
This study demonstrates the capability to distinguish lipid molecules with similar chemical structures in living cells without labeling. Compared to traditional methods relying on fluorescent or isotopic labeling, hyFOPM avoids issues such as labeling efficiency and interference with cellular functions, and its selectivity can be flexibly adapted to the spectral features of target lipids by adjusting the excitation wavenumbers.

The current system’s spectral specificity in the fingerprint region is superior to that of the C-H stretching region, which opens possibilities for distinguishing more lipid subtypes. The study also points out that combining advanced spectral unmixing techniques, such as deep learning, is expected to further improve sensitivity and specificity. In addition, mid-infrared photoacoustic microscopy can reach an imaging depth of over 150 μm in tissue, and future applications can be extended to thick samples or in vivo settings. Technological acceleration (e.g., spectral undersampling) and system miniaturization are important directions for advancing this technology toward point-of-care analysis or routine laboratory testing.