Chemical-state imaging of a mammalian cell through multi-elemental soft x-ray spectro-ptychography

Mammalian cells are structured samples in which different chemical states of materials are distributed in a complex manner. The materials in a cell are primarily composed of light elements, such as carbon, nitrogen, and oxygen. The chemical state of these elements varies depending on the class of biomolecule, such as sugars, proteins, nucleic acids, and lipids. The ratio of these classes of biomolecules depends on the intracellular structure, which allows for identification based on chemical state mapping of these elements. Soft x-ray absorption spectroscopy (sXAS) is a widely used method for measuring the chemical state of light elements.^1–3 For example, Ohigashi et.al. measured the distribution of biological materials such as DNAs and proteins by combining sXAS with scanning transmission x-ray microscopy (STXM).⁴ 

Ptychography is a high-resolution scanning coherent diffractive imaging technique,^5–9 and by combining ptychography with sXAS, we can measure the chemical states of elements in samples with a resolution of tens of nanometers.^10,11 Additionally, ptychography uses a defocused beam as illumination, a pixelated camera as a detector, and large position step sizes in contrast to STXM, which simultaneously enables rapid measurement time, low irradiation damage, high resolution, and high sensitivity. It has also been shown that ptychography is capable of measuring thicker samples compared to STXM.¹² 

For example, soft x-ray spectro-ptychography has been used to analyze the chemical state distribution of carbon nanotubes,¹³ magnetosome in bacteria,¹⁴ catalyst particles,¹⁵ and cathode materials.¹⁶ In these studies, a Fresnel zone plate (FZP)^17,18 was used as the focusing optic to increase the incident photon density and limit the illumination area. However, when measuring multiple absorption edges of light elements in cells, the large chromatic aberration of FZPs causes problems such as the need for wavelength-specific alignment. This complicates the measurement apparatus and poses a significant barrier to improving the spatial resolution and measurement time.

In this study, we measured hyperspectral images of a mammalian cell using soft x-ray spectro-ptychography with total-reflection optics using a cylindrical Wolter mirror. The achromaticity of the Wolter mirror avoids the issue faced by previous FZP-based experiments of a complex alignment procedure when scanning the photon energy.

By analyzing the spectra with probabilistic principal component analysis, we visualized intracellular structures that are difficult to distinguish with monochromatic x rays.

In ptychography, the complex transmittance of a sample is reconstructed by performing iterative phase recovery calculations using the coherent diffraction patterns measured at each point on the sample. In soft x-ray spectro-ptychography, ptychographic imaging is repeated at different energies near the absorption edge, and we can obtain spatially resolved soft x-ray spectra that contain rich information about the chemical states.

In this experiment, we used the CARROT (Coherent Achromatic Rotational Reflective Optics for pTychography) system, which is based on total-reflection mirror optics.¹⁹ The achromaticity of CARROT allows for seamless ptychographic measurements over different wavelengths. CARROT is set up at the soft x-ray beamline SPring-8 BL07LSU,^20,21 which can generate spatially coherent soft x-ray beams in the energy range from 250 to 2000 eV, which includes the K-edges of light elements from carbon to silicon.

The configuration of the optical system of CARROT is shown in Fig. 1. The incident soft x-ray beam is focused by a total-reflection cylindrical Wolter mirror^22,23 to illuminate the sample. The illumination beam size is about 15 μm and the transmitted x rays from the sample form coherent diffraction patterns on the downstream CMOS camera (Andor Marana-X). CARROT can capture a single coherent diffraction pattern at a repetition rate of approximately 75 ms. In previous research with CARROT, 50 nm resolution was achieved by imaging a test chart.¹⁹ 

The cell sample in this study is derived from a hybrid cell line of neuroblastoma from mice and nerve cells from rats (ND7/23). A visible light microscopy image of the ND7/23 cell is shown in Fig. 2(a). The cells were cultured on a 200 nm silicon nitride thin film for two days and then chemically fixed with paraformaldehyde. The thin film produces a uniform transmission and phase shift upon which the sample response is added, which can be removed by subtracting a sample-free region of the ptychographic image. Chemical fixation was chosen for this demonstration due to the relative ease of preparation compared to methods that might preserve the cellular structure and elements better such as flash-freezing and freeze-drying (FFFD) or live-cell observation.^24,25

Cells contain nitrogen-containing substances such as proteins, nucleic acids, as well as oxygen-containing substances such as sugars and proteins. So, soft x-ray spectro-ptychography measurements were performed at 105 points (395–425 eV) around the nitrogen absorption K-edge and 104 points (520–570 eV) around the oxygen absorption K-edge. The sample cells were illuminated with the soft x-ray beam on a grid of 21 × 21 points and a spacing of about 2.5 μm to get 83% overlap of neighboring 15 μm illumination areas. The overlap from 60% to 85% of the illumination area is found to be suitable for good ptychographic reconstruction.²⁶ The whole measurement process for both absorption edges required 115 min. The extended Ptychographic Iterative Engine (ePIE) algorithm⁸ was used to obtain the reconstructed images from the coherent diffraction patterns.

An overview of the data is shown in Figs. 2(b)–2(i). Panels (b)–(i) show representative reconstructed images at eight energies. In particular, we show the peak of N1s → π^* (c), N1s → σ^* (d), O1s → π^* (g), and O1s → σ^* (h). Panels (j) and (k) show the normalized absorption spectra averaged over the entire cellular body (supplementary material S2). Normalized absorbance spectra are scaled so that the absorbance sums to one over all energy points. This normalization removes the effect of cell thickness and isolates. The photon energies for panels (b)–(i) are labeled in these absorption spectra. Comparing the images measured at different absorption peaks we find small differences, but it is difficult to identify the intracellular structures from a direct comparison.

To distinguish these small differences, we applied Probabilistic Principal Component Analysis (PPCA)²⁷ to these images. To prepare the images for PPCA, we applied some preprocessing to spatially align each image and remove backgrounds. First, we used Lanczos resampling to interpolate the pixels proportionally to the wavelength at 520 eV. The images exhibit a slow positional drift, which is corrected with a phase correlation algorithm.²⁸ The maximum positional drift was found to be 160 nm.

We then applied PPCA separately to the hyperspectral images at the nitrogen and oxygen K-edges using the WAVEBASE library²⁹ developed by Toyota motor corporation. We begin by finding the principal components (PCs) of the entire dataset. The original hyperspectral image is first masked to select pixels within the cellular body, then binned in a 20 × 20 grid to produce low-noise spectra for finding the PCs. During the PPCA process, all the binned spectra are area-normalized. This normalization has the effect of removing the principal component associated with thickness. We then used the derived principal components to extract the features of the full-resolution hyperspectral image by projecting the spectrum associated with each pixel onto the PC basis set. Figures 4(a), 4(b), 4(e), and 4(f) show the distribution of the principal components for the nitrogen and oxygen K-edges. We separate some intracellular structures which are difficult to identify from the monochromatic x-ray images. At the nitrogen edge, the explained variance ratios for the first and second principal components are 42.2% and 16.6%, respectively. At the oxygen edge, the explained variance ratios for the first and second principal components are 69.2% and 23.1%, respectively (supplementary material S4). Only the first two PCs, PC1, and PC2, are considered to show biological relevance due to the low explained variance ratios and the relatively flat distribution of PC3 across the cell. Panels (c), (d), (g), and (h) show the spectral features of the principal components. The gray curves are the average of the normalized spectra of the whole cellular body. The red (blue) curves represent the variation in the average spectrum explained by the PCs, corresponding to plus (minus) one standard deviation. From these curves, we see that, for both the nitrogen and oxygen edges, PC1 is correlated with the peak ratio between the π^* and σ^* peaks, while PC2 correlates with the peak energy shift. Information about later principal components is shown in supplementary material S4.

From the spectral and spatial shape of the PCs presented in Fig. 4, we make several inferences about cellular structures. Nitrogen-containing substances in cells are mainly proteins and nucleic acids. Amino acids (components of the proteins) can be spectrally distinguished from nucleic acids by the relative strength of the π^* and σ^* peaks, where amino acids generally exhibit a stronger σ^* peak.³⁰ This allows us to distinguish regions of high (low) amino acid content as blue (red) regions of nitrogen PC1 shown in panel (a). We find that nitrogen PC1 is strongly blue at the perinuclear region, which is generally enriched with endoplasmic reticula, mitochondria, and Golgi apparatuses. In these regions, protein transport, synthesis, and folding occur, and the density of amino acids is higher than other cellular regions, which is consistent with our data.

The nitrogen PC2 mapping shows a clear separation of the cell into two parts: the nuclear and cytoplasmic regions. Nuclei typically contain nitrogen-containing nucleic acids and proteins such as histones, various types of zinc-finger proteins and enzymes, which distinguish the nuclear region from cytoplasmic regions. The characteristic π^* peaks of nucleic acid and the protein absorbance spectra³¹ can reflect the blue curve of the nitrogen PC2.

Oxygen-containing cellular substances include proteins, nucleic acids, carbohydrates, and lipids. In addition, Dulbecco's Modified Eagle Medium (D-MEM) was used in the culturing process, which contains a significant amount of oxygen. In the oxygen PC1 mapping, several strong red regions are detected outside of the cell that have small masses. These are likely crystalized salts from the buffer which formed during the drying process. The stronger σ^* peak of the red curve in Fig. 4(g) indicates a higher ratio of carbohydrates to proteins in the medium salts.

Within the cell body, the oxygen PC1 mapping reveals granular structures of size 2–3 μm² or less across the entire cellular body, appearing as red or white regions against a blue background. Such structures were not observed at the nitrogen absorption K-edge and hardly observed by general visible light microscopy. These structures could be lysosomes, other organelles, or crystalized salts, but identification is difficult in this study, and visualizing chemical state differences may possibly reveal unexpected or unknown structures.

In contrast to the granular structure of the oxygen PC1 mapping, the oxygen PC2 mapping shows a gradual change in spectrum between the nuclear and cytoplasm regions identified by the nitrogen PCs. The gradual structure in oxygen PC2 may reflect the oxygen concentration compared to carbon and nitrogen. In the red region, both π^* and σ^* peaks tend to be weaker. This suggests that there is an increase in uniform absorption without peaks, and the oxygen peak intensity becomes relatively weaker by applying normalization. Therefore, it can be inferred that the red regions are rich in carbon and nitrogen compared to oxygen. In contrast, blue regions are rich in oxygen compared to carbon and nitrogen.

In conclusion, we obtained a high-resolution hyperspectral image inside a mammalian cell at the nitrogen and oxygen absorption K-edges by soft x-ray spectro-ptychography with total-reflection Wolter mirror optics. The features of the absorption spectra in the hyperspectral image were extracted with PPCA, and based on the separated principal components, chemical state mapping within the cell was performed. The mapping results clearly resolved the intracellular structures which are difficult to recognize in a monochromatic image.

The achromatic Wolter mirror optics of the CARROT system enable seamless measurement of absorption edges of a larger number of elements. At BL07LSU of SPring-8, a wide energy range from 250 to 2000 eV can be generated, allowing measurements at additional absorption edges of biologically important elements such as carbon, sodium, and iron. Including more absorption edges in the hyperspectral image will contribute to more detailed identification of intracellular organelles and biological compounds. Three-dimensional spectroscopic imaging with tomography or laminography can also provide more accurate mapping of structure to spectra, avoiding the problem of superimposed organelles in two-dimensional imaging.^32–34 In these results, the structures are not strictly classified, and we made assumptions based on known cell biological findings. However, using PPCA to visualize the intracellular structure as differences in absorbance spectra is important as a step toward classifying biomolecules using spectro-ptychography. Methods for identifying intracellular organelles and labeling biological materials have been previously established.³⁵ In the future, our method can also be used to image such individual molecules by attaching label elements such as fluorine and boron, which are rare in biological environments.

In this study, as a first step in the development of a method for measuring biological samples, we used cells with chemical fixation which have already been established for soft x-ray microscopy. Chemical fixation is known to cause significant changes in cell structure and composition.³⁶ To overcome this problem, flash-freezing and freeze-drying (FFFD) or live-cell observation can be considered. The relatively large working distance of the Wolter mirror optics can be easily applied to such in situ/operando measurement system for cryo- or live-cell imaging. In situ ptychographic imaging of yeast cells suggests that it causes less damage than conventional x-ray microscopy.³⁷ By combining our previously demonstrated solution sample holder system^25,38,39 for x-ray microscopy with spectro-ptychography, it is possible to perform in situ quantitative elemental and chemical state mapping of living mammalian cells. Although radiation damage can still be a major issue,⁴⁰ several methods such as ghosts imaging have been proposed to reduce its effects.^24,41 A unique feature of ptychography is that high-sensitive phase-shift images can also be detected quantitatively. By avoiding the region just below the absorption edge and observing the energy dependence of the phase shift, it may be possible to make x-ray phase-shift spectroscopy measurements comparable to x-ray absorption spectroscopy, but with less damage. Combining these methods will become increasingly important in the future. The developed method has the distinctive feature of enabling high-resolution, nondestructive imaging of the unlabeled chemical state within cells. By providing a perspective significantly different from that of optical or electron microscopes, this method is expected to contribute to the advancement of cell biology in the future.