Tissue and cell samples of liver tumors in vitro

One hundred surgical resection specimens of HCC were provided by the First Affiliated Hospital of Xinjiang Medical University (XJMU) with approval from the ethics committee of XJMU. The inclusion and exclusion criteria for the experimental samples are listed in Table 1.

Histopathological examinations of the HCC tissue (HCCT) specimens revealed moderate or low differentiation in this study. The HCC cell samples were immediately extracted from small postoperative tissue blocks, digested with 0.25% trypsin for 20 min, washed, and then cultured in Dulbecco's modified Eagle's medium (DMEM). The DMEM was modified to contain 4500 mg/L of glucose, 4 mM of L-glutamine, 1 mM of sodium pyruvate, and 1500 mg/L of sodium bicarbonate. It was also supplemented with 10% calf serum, 100 U/mL penicillin, and 100 μg/mL streptomycin for 70% to 80% confluency on the bottom of the petri dish. The cells were maintained in a 5% CO2 incubator at 37 °C throughout the experimental process. All experimental samples were divided into two groups of tumor lesion areas and para-carcinoma areas. Prior to the experiments using SR-PPCT, all formalin-fixed tissue samples were processed with a graded dehydration series of ethanol solutions (GDES) and firmly packed in a centrifuge tube to avoid artifacts caused by tiny deformations or movements. For the SR-FTIR experiments, serial tissue sections with thicknesses of 10 μm were obtained using frozen section methods, carefully attached to a BaF₂ substrate, examined under a microscope, wrapped in foil, and stored at −80 °C to preserve the biochemical substances in the samples. For SR-STXM measurements, the cellular samples were fixed and dehydrated using GDES at ambient temperature and then dropped onto a Si₃N₄ window for cellular fine structural measurements. Tissue specimens were used for histopathological examination.

Experimental setup and data processing

A high-brightness light source can facilitate measurements with a high SNR. Our synchrotron-based experiments were conducted on the hard X-ray imaging, soft X-ray microscopy, and IR spectroscopy beamlines at Shanghai SR. We present a general technique flowchart for combining the SR-PPCT, SR-STXM, and SR-FTIR methods to investigate the pathological mechanisms of biomedical soft tissues via trans-scale-modality correlation analysis in Fig. 1. The 3D morphological tissue changes were observed using SR-PPCT on BL13W1, which is equipped with a complementary metal-oxide-semiconductor with a resolution of 6.5 μm/pixel and different camera lenses (×1.25, ×2, ×10) for achieving different microtomographic spatial resolutions. All SR-PPCT projections were preprocessed using background correction and phase retrieval to improve the contrast-to-noise ratio and the density resolution of weakly absorbing samples without the use of a contrast agent. The phase-attenuation-duality algorithm was implemented to retrieve a phase map ${\phi }_{\theta }$ at any single projection angle $\theta$ using the following formula:

Fig. 1Full-screen View

(Color online) Experimental setups for synchrotron-based PPCT, STXM, and FTIR at SSRF and general technique flowchart for the trans-scale and multi-modality imaging and data analysis of HCC tumors

where $I_{\text{θ}}$ is a single projection image, $\lambda$ is the X-ray wavelength, $r_{\text{e}}$ is the classic atomic radius, and ${\sigma }_{\text{KN}}$ is the Klein-Nishina total cross-section for X-ray photon Compton scattering from a single free electron. M = (R₁ + R₂) / R₁, where R₁ and R₂ are the source-to-object and object-to-detector distances, respectively. This method is used for the discrimination of 3D tissue micromorphological variations and visualization of abnormal microcirculation networks, and aids in the collection of high-precision nondestructive statistics for tissue microenvironmental geometric parameters ^{[36, 37]}.

The fine feature structures of tumor cells were measured using STXM with a spatial resolution of approximately 30 nm on BL08U1-A. A high-flux soft X-ray beam focused through the zone plate can be used to probe cellular structural information based on the absorption scanning mode ^{[38, 39]}.

Typical spectra of frozen tumor tissue sections at a spectral resolution of 4 cm⁻¹ can be collected using FTIR on BL01B, which is equipped with a Nicolet 6700 FTIR spectrometer and Nicolet Continuµm Microscope. In our experiments, a 20 μm × 20 μm aperture with a 10 μm step size was applied to acquire mapping data in the wavenumber range of 800–4000 cm⁻¹. In our SR-FTIR experiments, to improve the SNR, the step size was set smaller than the aperture size to produce pixel map overlaps and all data acquisitions were performed three times. The raw mapping spectra were preprocessed with nine-point smoothing and baseline correction to suppress the ambient noise, electronic offset, dark current, and readout noise of the IR detector further using OMIC 9.0.

SR-PPCT measurement of HCC microenvironments

The rapid proliferation and metastasis of tumors are closely associated with changes in tumor microenvironments. Compared to conventional CT, 3D trans-scale visualizations of HCC soft tissues based on SR-PPCT measurements reveal more detailed pathologic morphological characteristics and microenvironmental changes, as shown in Fig. 2. Macro-moderate or high-differentiation tumor lesions appear as abnormal gray spots or image shadows in Fig. 2(a, b). By using the SR-PPCT technique without staining, HCC-related pathological features can be effectively displayed and distinguished from para-carcinoma liver tissues (PCLT) on the micron scale, as indicated by the yellow arrows in Fig. 2(c, d, e). These results can reveal the morphological distributions of arterioles, venules, sacs, internal septa, fibrosis, microcalcification, and increasing inflammation reactions in HCCT from any perspective. Nondestructive 3D inner-structural information regarding PPCT reveals detailed changes in tumor tissues associated with tumorigenesis and metastasis, which are helpful for precise tumor diagnosis.

Fig. 2Full-screen View

(Color online) Results of medical imaging and SR-PPCT microtomography of HCC tumors. (a) Radiological image, (b) cross-section of traditional CT at the position of the dashed line in (a), (c) transverse section located in the red circle in (b), (d) coronal section located at the position of the green dashed line in (c), and (e) sagittal section located at the position of blue dashed line in (c) as reconstructed by SR-PPCT. Scale bar in (b) is 10⁴ μm and scale bars in (c, d, e) are 100 μm.

Angiogenesis and microcirculation network formation play important roles in tumor diagnosis and morphological understanding. We present 3D visualizations of microvascular distributions of HCC solid tumors using different scales and modalities in Fig. 3. The entire HCC vascular network based on conventional CT is presented in (a) and the formation and development of tumors associated with inflammation and micro-cyst clusters near blood vessels based on SR-PPCT are presented in (b, c). The inflammatory microenvironment and blood supply system can trigger the concentration of endothelial and epithelial cells to form micro-cystic clusters, which consist of irregular tumor micro-vessel networks. Based on visual inspections, the blood vessels in normal tissue areas are thicker and have higher curvature. In contrast, the blood vessels of HCC are relatively thin and stiff, and their distribution is more dispersive and glomerate. According to the comparisons of the high-resolution PPCT and pathologic examination slices in (d, e), SR-PPCT provides more abundant and detailed vascular 3D structures without staining and sample deformation, and reveals the sedimentation of calcium salts with high gray value areas in (d) around the microcysts indicated by yellow arrows. These salts are a concomitant precursor of hepatic fibrosis.

Fig. 3Full-screen View

(Color online) 3D microenvironment of an HCC liver tumor and medical examination comparisons. (a) 3D image of the entire liver and the solid HCC tumor from medical CT, (b and c) enlarged reconstructed results based on the PPCT, respectively located in the yellow circle area in (a) and red rectangle area in (b), and (d) PPCT slice and (e) HE slice of the HCC tumor micro-cyst area indicated by the blue rectangle in (c). One can observe similar HCC tumor lesions between different imaging modalities. Scale bar = 100 μm.

Joint measurements with high-resolution SR-PPCT and SR-FTIR mapping

The morphological characteristics of hepatic sinusoids were observed using SR-PPCT. The inflammatory microenvironment and formation of sinusoid capillaries in HCC tumors can be clearly visualized based on the high phase contrast and spatial resolution, as indicated by the blue arrows in Fig. 4(a, b, c). These factors promote the formation of abundant irregular micro-vessels around and inside HCC tumors, as indicated by the red arrows in (b) and (c). Furthermore, representative spectra and 2D distributions of biological macromolecules were collected via SR-FTIR measurements for association analysis ^{[40, 41]}. FTIR mapping measurements with 200 μm× 160 μm areas were performed in both the PCLT and HCCT regions, which are denoted by the red and orange rectangles, respectively, in (a). Images (d) and (h) were captured in FTIR microscopic images for the selection of PCLT and HCCT tissue regions of interest (ROIs) based on the SR-PPCT results and pathologist analysis. The measurements of each sample were repeated three times on average. Twenty samples were randomly selected for each group and three groups were selected. Chemical distributions and concentration maps were processed using normalized pseudo-color scale bars to identify lipids, proteins, and nucleic acids, which correspond to peaks at 2922, 1649, and 1074 cm⁻¹, respectively, as shown in parts (e, f, g, i, k, l). One can clearly see that the concentrations of the nucleic acid and lipid components in the HCCT areas increase significantly.

Fig. 4Full-screen View

(Color online) Microstructural and organic macromolecular feature spatial distributions in HCCT based on SR-PPCT coupled with 10× objective lens and SR-FTIR techniques. (a) 3D microenvironment of HCC and its virtual tumor block, (b and c) fine structural distribution of hepatic sinusoids (blue arrows) and the formation of sinusoid capillaries (red arrows) in different sections of (a), (d and h) optical microscopic images of PCLT and HCCT tissue layers indicated by the red rectangle in (b) and orange rectangle in (c). Univariate maps for collecting the chemical spatial distributions of lipids, proteins, and nucleic acids in the layers of PCLT and HCCT regions in different feature infrared spectra. (e and i) Lipids recorded with a feature peak at approximately 2922 cm⁻¹. (f and k) Distributions of proteins with a feature peak at approximately 1649 cm⁻¹. (g and l) Distributions of DNA at 1074 cm⁻¹. Scale bars: 50 μm in (b, c) and 30 μm in (d, h).

SR-STXM measurement of HCC cellular structures

The SR-STXM technique was also employed to measure the HCC cellular structures of liver cells, endothelial cells of the blood sinus, and Kupffer cells, as shown in Fig. 5(a–f). One can see that the HCC-related nucleoli are divided and diffused to varying degrees, as indicated by the white arrows in (b, e, f). A large amount of cellular secretions and lipid droplets emerge chaotically, as indicated by the blue arrows in (c, d, e). Compared to (a), the mitochondria and cytoplasmic volume in HCC cells increases significantly, as indicated by the red arrows in (c, f). A more detailed cellular structure of HCC-related changes can be observed without staining by using high-resolution SR-STXM measurements, as shown in (f). The large aggregation of mitochondria, endoplasmic reticulum, and glycogen is associated with the fast metabolism of HCC cells and active tumor microcirculation. Additionally, nucleosome dispersion and cell membrane morphological changes are helpful for understanding the occurrence, metastasis, and staging of HCC.

Fig. 5Full-screen View

(Color online) Features of HCC cells reconstructed by and SR-STXM techniques. Images of different hepatocytes obtained from SR-STXM: (a) stellate hepatocyte, (b) Kupffer’s cell, (c) liver plate cell, (d) endothelial cell of blood sinus, (e) epithelial cell of connective tissue, (f) enlarged Kupffer’s cell with mass of mitochondria. Blue arrows denote lipid globules or exosomes of hepatocytes, red arrows denote mitochondria, and white arrows denote nucleoli. Scale bars: 50μm in (a, e), 10 μm in (b, c, d, f, g), and 5 μm in (h).

SR-FTIR measurement of HCC biomacromolecules

The biomolecular vibration absorption peaks and characteristic IR spectral regions were successfully measured and are presented in Fig. 6(a, b), corresponding to the PCLT and HCCT functional group vibrations. The spectral range of 3500–3100 cm⁻¹ is attributed to O-H & N-H bond stretching vibrations in the cellulose and polyamide, and the range of 3050–2800 cm⁻¹ is attributed to the symmetric and antisymmetric stretching vibrations of methyl and methylene C-H bonds, which are related to the increasing lipid-related acyl chains of HCCT. The peaks at 1649 and 1485 cm⁻¹ are attributed to the protein amide I (N-H bending and C=O stretching) and protein amide II (C-N stretching and N-H bending), reflecting changes in nucleolus histones. In the chemical distribution image, one can see that the content of protein amide increases in HCCT. The spectra at 1453 and 1397 cm⁻¹ are attributed to C=H bending vibrations and COO- symmetric stretching vibrations, respectively. The range of 1209–1074 cm⁻¹ represents the symmetric and antisymmetric stretching vibrations of nucleic acids consisting of phosphorylated molecules (PO₂ phosphodiester stretch) and glycogen, respectively. In comparison, there is a significant increase in the concentration of nuclear acids between 1000 and 1300 cm⁻¹, which is associated with DNA phosphorylated carbonated nucleoli. The spectral band of 900–1000 cm⁻¹ for HCCT appears more active, which is related to RNA ribose chains.

Fig. 6Full-screen View

(Color online) IR spectroscopic results for PCLT and HCCT layers collected in the transmission mode of SR-FTIR. The vibrational absorptions of the most important biomolecules are highlighted in (a) and (b) based on the averages of n = 20 randomly selected data from the PCLT and HCCT samples, respectively.

Analysis and Discussion

For quantitative analyses of HCC microenvironmental changes, ROIs were selected from the areas of PCLT, outside areas of HCCT (HCCT__out), and inside areas of HCCT (HCCT__in) in 20 postoperative specimens, which were randomly selected from 100 experimental specimens and defined by our surgeon and pathologist. There were 60 ROIs in total (20 ROIs for each area), 150 μm× 150 μm × 150 μm voxels for each ROI, 6.5 μm³ per voxel, and a volume of 0.975 mm³ per ROI. We defined the 3D micro-vessel density (3D-MVD) as follows: 3D-MVD = (voxels of micro-vessels) / (total voxels in RoI). Statistical results for the microvascular numbers, microcyst numbers, and inflammation areas are presented in Fig. 7. The t-test statistical method was employed to compare the datasets with a normal distribution and p-value threshold of 0.05 representing statistical significance. This demonstrated that the SR-PPCT technique combined with phase retrieval enables the visualization of the boundaries of tumors and normal tissues, as well as the microvascular networks in tumor infiltration regions. We found that the smallest sinusoid capillaries abundantly scattered along the edges of HCC tumors are approximately 10 μm in diameter, which is a possible early diagnosis feature of HCC. This indicates that HCC is a typical multi-vascular solid tumor and that an inflammatory microcystic environment is the basis of tumor angiogenesis, which is characterized by uncontrolled growth and metastasis.

Fig. 7Full-screen View

(Color online) Analysis of pathological variations of different HCCT groups. (a) 3D-MVDs and (b) microvascular numbers of PCLT and HCCT respectively. (c) Inflammation areas and (d) microcyst numbers on the outside surface and inside of HCCT, which affect the development degree of HCC tumor infiltration and its cell proliferation.

This analysis is useful for the intelligent diagnosis and feature delineation of different differentiated tumors and for classifying the tumor microenvironmental changes by using machine learning based on SR-PPCT sectional images. There are 10 types of eigenvalues defined by GLCM, GGCM, and GH ^{[42, 43]}, as shown in Table 2. The eigenvalue datasets were calculated in three groups (PCLT, HCCT_out, HCCT_in, 100 samples per set, 300 samples total). Then, a BP-NN model with 10 hidden layers was employed to classify the three HCC tumor microenvironments with different degrees of differentiation, as shown in Table 2. The calculation of hidden layers (H) is H = (E + G)^1/2 + a, where E is the number of image eigenvalues (input layer), G is the group number for classification (output layer), and a is a constant ranging from 1 to 10 for estimating the number of hidden neurons. For training the BP-NN model, 70%, 15%, and 15% of the samples were randomly selected as training, testing, and verification samples, respectively. The optimal iteration times (I), network performance, gradients, and R values of linear regression fitting degrees were used to evaluate the neural models with different number of hidden layers (H). When H = 10 and I = 19, the R values for the training, validation, testing, and all samples were all above 0.90.

Based on the differences between the upper and lower quartiles of each sample eigenvalue, the distributions of tumor microenvironment morphologies in all sample spaces are plotted in Fig. 8(a), which quantitatively represents the HCC-induced micro-morphological variations between the control group (PCLT), benign group (HCCT_out), and malignant group (HCCT_in). The BP-NN model was employed to classify the three types of HCC microenvironmental morphologies, as shown in Fig. 8(b). The accuracy of the BP-NN model is as follows: 96.0% for type_1 (PCLT), 100% for type_2 (HCCT_out), 97.0% for type_3 (HCCT_in), and 97.7% for all samples. To evaluate classification accuracy, the precision and recall rates were calculated as 0.980 and 0.990, respectively. Therefore, the BP-NN model has the potential to classify and diagnose HCC micro-scale tissue variations based on an SR-PPCT database, which is helpful for the quantitative analysis of the pathological correlations between micro-environment changes and tumor development.

Fig. 8Full-screen View

(Color online) Image feature distributions of HCCT microenvironmental variations and classification results based on the BP-NN model. (a) Maps of 10 image eigenvalues in three groups, and (b) the confusion matrix of the BP-NN model, which represents the accuracies of three output classes.

Based on the experimental results presented in Fig. 5, HCC-induced cytopathic effects were measured using SR-STXM. The GGCM-based eigenvalues of cellular images were employed to evaluate cellular morphological changes intelligently by calculating the grayscale and texture characteristics of SR-STXM images. Eight rotation angles were considered for each image (0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°) to calculate four different characteristics, as shown in Fig. 9. These results effectively reveal the inner microstructural differences between the peri-carcinomatous liver cells (PCLC) and hepatocellular carcinoma cells (HCCC). Compared to the characteristic values of PCLC, the energy of HCCC is lower overall and the entropy, contrast, and correlation of HCCC are generally greater, indicating that increasing HCC-related cellular secretions and lipid droplets cause image eigenvalue changes, such as an increase in entropy and contrast, and a decrease in energy. The HCC-related changes in the cell nucleus and nuclear membrane trigger a slight energy reduction according to the experimental results. One can see that the entire eigenvalue distribution accurately reflects the HCCC morphological changes, which is helpful for determining early HCCC damage degrees and performing classification.

Fig. 9Full-screen View

(Color online) Calculated results for morphological eigenvalues based on the GGCM calculation of HCCC and PCLC. (a) Image energy, (b) image entropy, (c) image contrast, and (d) image correlation.

The PCLT and HCCT specimens were investigated using the SR-FTIR technique to analyze HCC-related biological macromolecular vibration absorptions, which is helpful for the discrimination of small changes in IR spectral positions and intensities. In Table 3, one can see that the N-H stretching vibration spectra of HCC shifts to higher wavenumbers more significantly based on the varying lengths of N-H chains. This implies that the HCC cellular chemical environment changes significantly as well. The symmetric and antisymmetric C-H stretching vibration spectra of the HCC samples, which represent the methyl and methylene related to cellular proteins and lipids, shift from a range of 2962–2861 cm⁻¹ to 3045–2920 cm⁻¹. This indicates that vigorous tumor cell metabolism plays an important role in angiogenesis, HCC proliferation, and metastasis. The absorption peaks of protein amide I and II for HCCT shift to lower wavenumbers at 1625 and 1485 cm⁻¹, which is associated with secondary protein structural changes. This causes the amide conformation to become loose and disordered. The peaks related to nucleic acids and phosphodiester bonds in HCC samples move to lower wavenumbers at 1151 and 1028 cm⁻¹ as a result of the breakage of hydrogen bonds and double strands of DNA. The relative absorption intensities of some characteristic IR spectra, including cholesterol (1467 cm⁻¹), RNA (1121 cm⁻¹), glycogen (1045 cm⁻¹), and DNA (1020 cm⁻¹), were selected to compare the differences between 20 control groups and HCC groups based on unpaired t-test analysis, as shown in Table.1. The peak height ratio of P₁₀₄₅ / P₁₄₆₇ for HCCT is greater than that for PCLT based on a greater increase in glycogen content compared to cholesterol content. The P₁₁₂₁ / P₁₀₂₀ ratio for HCCT is significantly greater than that for PCLT because it increases with the degree of malignancy of the tumor. The P₁₁₂₁ /P₁₀₂₀ (RNA/DNA) ratio in HCC groups is also much greater than that in PCLT, which indicates that the ratio of P₁₁₂₁ / P₁₀₂₀ significantly increases with the malignancy of the tumor and can be considered as a biomarker for determining the degree of tumor differentiation. The peak area ratio A_{amide I} / A_{amide II} of HCCT was lower than that of PCLT, indicating that the HCC-induced secondary structures of nuclear proteins differ from those of intermediates, which could be used as a potential criterion for identifying malignant tumors. The peak area ratio A_(CH3) /A_(CH2) increases overall, indicating that the chemical conformation of lipid macromolecules changes, including additional branches of the hydrocarbon chain associated with HCC cell division and proliferation.