The deep-sea mussels (G. platifrons) were collected during the “Kexue” cruise in 2021 from the cold seeps of the South China Sea (119°17′08.297″E; 22°06′55.435″N, temperature: 3.57°C, salinity: 34.65 psu). The samples were brought to the deck using a thermal-preserve sampler carried by a remotely operated vehicle (ROV), and rapidly dissected for further processing.

We compared samples obtained by four common preservation methods. For the samples obtained by first method, SF, paper was used to absorb the liquid around the gill block to prevent the formation of ice crystals. A metal beaker was then filled with isopentane and placed in liquid nitrogen. Next, the tissue block was placed in the metal beaker by using forceps to completely submerge the tissue in isopentane until it was completely frozen. This step increases the rate of freezing and restricts ice crystal formation. The snap-frozen gill was then placed in a −80°C freezer for long-term storage. For slicing, the gill was broken into small pieces and placed directly into a pre-chilled optimal cutting temperature compound (OCT) embedding agent for frozen sectioning, with a section thickness of 10 μm (Leica CM1950). The second samples obtained by method, SF-MeOH, follows the same steps as those involved in SF till the sectioning process. For sectioning in SF-MeOH, the sections were fixed in methanol for 10 min, washed three times with 0.01 M phosphate-buffered saline (1 × PBS), and then placed in a pre-chilled OCT embedding agent for frozen sectioning (thickness: 10 μm). For the third samples obtained by method (SF-GP), the dissected gills were directly transferred to a pre-cooled paraformaldehyde-glutaraldehyde (2%/2.5%) fixative, fixed at 4°C for 24 h, and placed in a fresh fixative. The slicing operation was the same as that conducted in SF. For the fourth samples obtained by method (PS-PA), the dissected gills were stored in a 4% paraformaldehyde fixative (PFA) for 24 h, rinsed twice with 1 × PBS, and stored in 75% ethanol at 4°C. The samples were dehydrated in different gradients of ethanol (70%, 80%, 95%, and 100%) for 1 h, and placed in xylene for 2 h. After transparency, the samples were embedded in paraffin wax at 60°C for 3 h. The embedded samples were sectioned using a Leica microtome (Leica RM 2016) to achieve a thickness of 7 μm. The samples were dewaxed with xylene before use and then rehydrated with different gradients of ethanol (100%, 95%, 80%, and 70%) before use.

Parts of the gill were selected for Raman spectral acquisition using a confocal Raman microspectrometer (alpha 300R, WITec, Ulm, Germany) equipped with a laser operating at 532 nm and a 600 grooves/mm grating (UHTS 300). The sample was placed on a calcium fluoride substrate, and the excitation laser was focused onto the sample for spectral acquisition using a ZEISS EC Epiplan (Carl ZEISS, Jena, Germany) with 50×/0.75 objectives. Before use, systematic calibration was performed using the characteristic peak of the silicon wafer at 520 cm^-1. The step size, total acquisition time, and laser power of the data set acquired by each method were optimized to obtain the best results. The laser power was 15–20 mW for one point in the X- and Y-axis directions of 310–670 nm.

For preprocessing the raw data, baseline correction and cosmic ray removal were performed using WITec Project plus, followed by principal component analysis (PCA) to improve the signal-to-noise ratio in order to retain the necessary signal. The tissue was then partitioned for imaging using clustering analysis to further obtain characteristic spectrum of the tissue. The spectra were fitted with a Gaussian using the GRAMS/AI software, and the model peak positions were chosen based on the values of the relevant material peak positions reported by different authors. A range of positions for each Raman peak was restricted for the fit, and the peak positions were fine-tuned on the basis of the fit. The result with the smallest sum of squared deviations between the curve fitting model and the experimental data was considered to obtain the relevant information such as peak position, peak intensity, peak width, and peak area.

We have established a pipeline to show the metabolic distribution feature of key metabolites of the gills of the symbiotic deep-sea mussels. A well-preserved SF sample was selected for the metabolic profile constructions. It has been demonstrated that SF is closest to the in situ state in terms of component composition because it has not been disturbed by immobilizers (Lyng et al., 2011; Butler et al., 2016). We performed classification imaging of the Raman dataset by multivariate analysis and found that the resulting image could be classified into four major categories: the nucleic acid enrichment region, protein enrichment region, phospholipid enrichment region, and the lysosome ( Figure 1A ). By analyzing the characteristic spectra of each region, the peak for the red nucleic acid region was found at 1,063 cm^-1 which can be attributed to DNA (Okotrub et al., 2015), suggesting the existence of the nucleus. The peak at 1,261 cm^-1 was the cyan protein enrichment region and is assigned to =C-H bending (protein) (Chan et al., 2006), indicating the presence of the cytoplasm containing symbionts. Moreover, a peak 1,270 cm^-1 indicated the blue phospholipid enrichment region, and can be attributed to the presence of typical phospholipids (Malini et al., 2006; Kochan et al., 2013), representing the basal membrane ( Figure 1B ). Lastly, the characteristic peak of 780 cm^-1 for the yellow region is likely ascribed to the presence of phosphatidylinositol (Xu et al., 2020), which can be used as a lipid signaling molecule for the lysosome (Hohman et al., 1982). This imaging distribution is consistent with conventional methods observed in the published literature (Ponnudurai et al., 2017; Sun et al., 2017; Wang et al., 2021). Therefore, we chose SF samples as standards to analyze the effects of different preservation methods on subsequent Raman studies and biological studies.

Using the same spectral processing method, we found that the spectral datasets obtained by all four methods could allow the visualization of the three major classes of the nucleus, cytoplasm containing symbionts, and basal membrane after imaging analysis ( Figures 1A and 2 ). However, cells at SF during sectioning are very prone to extensive fragmentation, leaving only the basal membrane, or adhering together indistinguishably, requiring many sections for local selection detection ( Supplementary Figure S2 ). The basal membrane detected upon SF was found to be significantly thicker than that through other methods via CRM imaging ( Figure 1A ); this thickness is probably due to being unfixed, resulting in scraping during sectioning, which may have an impact on the biomolecular distribution of the biomolecules. The overall morphology acquired upon SF-MeOH is more prominent than that upon SF, allowing the clearer distinction of the boundary of the gill cells and facilitating significantly less fragmentation ( Supplementary Figure S2 ). Slight internal breakage was found after imaging of the selected areas ( Figure 2 ), probably due to the rapid fixation of methanol, which maintained the overall morphology of the gill cells but did not provide good maintenance of the cell interior. SF-GP and PS-PA allowed the efficient maintenance of the overall morphology of the gill cells after sectioning due to prolonged fixation, and the degree of fragmentation was very low ( Figure 2 ). However, prolonged immobilization upon SF-GP and PS-PA changed the intensity of the Raman signals of some substances, such as that of phosphatidylinositol (780 cm^-1), restricting the Raman imaging of some important organelles such as lysosomes ( Figure 2 ). Moreover, prolonged immobilization resulted in poor preservation of the organelle structures.

Lysosomes are important for the study of symbiotic regulation and have been demonstrated in symbiotic systems of multiple taxa (Hohman et al., 1982; Nishikori et al., 2009; Castillo et al., 2015). Lysosomes are involved in regulating the homeostasis of symbiotic bacteria and determining the fate of the symbiont (Trench, 1971). In symbiotic deep-sea mussels, lysosomes play an important role in the intracellular digestion of the pasture. Here, SF-GP and PS-PA resulted in failure of imaging of the lysosomes; the spectra of the as-preserved samples did not show the characteristic peak for the lysosomes at 726 cm^-1, which is typically attributed to DNA (Talari et al., 2015) ( Figure 1B )—an important evidence of the host’s consumption of symbionts through lysosomes. Although the presence of lysosomes can also be verified via electron microscopy, the Raman signal of phosphatidylinositol is disrupted, leading to an inability to discriminate by Raman imaging. The absence of the peak corresponding to phosphatidylinositol, a membrane-forming molecule, indicates the loss of membrane integrity, which can lead to drastic changes in the distribution of some substances inside and outside the lysosome and affect the further study of substance distribution.

In summary, SF-GP and PS-PA are more effective toward the maintenance of the overall morphology; however, SF and SF-MeOH have a weaker effect on the distribution of substances due to the weak influence upon fixation, which is more favorable for some detailed imaging studies.

First, we took the total average spectra of the tissues after area scans of the gill sections obtained from the four different preservation methods ( Figure 3 ). Significant variability in the peak positions and intensities in the regions within the spectra were observed; however, most of the molecular contributions could be identified ( Table 1 ). The spectra were normalized to improve clarity (Faoláin, 2005). Considering that the signal of the Raman peaks may be influenced by the laser focusing, we discussed the differences in peak intensities together with the spectra before and after normalization for comparative analysis.

Following dewaxing, the peak in the spectra of the PS-PA samples were clearly different from those prepared by other methods at 2,881 cm^-1. Further analysis revealed that this peak was consistent with the Raman peak of paraffin wax. Further comparison with other paraffin wax main peaks indicated that paraffin wax likely has a stronger signal in this section. Other characteristic peaks of paraffin wax, e.g., those near 1,063 cm^-1 (C-C stretch), 1,131 cm^-1 (C-C stretch), 1,296 cm^-1 (CH₂ deformation), and 1,418 cm^-1 (CH₃ deformation), could be observed in the spectra (Faoláin, 2005; Tfayli et al., 2009). The combination of the average spectra obtained by the other treatments revealed that the peaks of the samples overlapped significantly with those of the biological tissues (Tfayli et al., 2009). It was difficult to distinguish their contribution and to determine whether the peak was due to the distortion of the biological components caused by extensive treatment. For example, in sections other than those preserved via PS-PA, the peak at 1,063 cm^-1 can be judged as the characteristic peak for DNA; however, in the PS-PA-preserved samples, it may also be the characteristic peak of paraffin. Although it does not affect the H&E staining observation, incomplete dewaxing can interfere with the acquisition of Raman data and analysis. However, it is possible to study spectral regions of interest that do not overlap with the paraffin signal; also, for the study of overlapping regions, further optimization of the proposed method is needed to verify that paraffin is removed completely for more effective Raman analysis.

The Raman peak at 517 cm^-1 corresponding to the S-S stretching (Van Wart et al., 1973; Maquelin et al., 2002) and at 656 cm^-1 ascribed to the C-S stretching were present in the SF and SF-MeOH samples, indicating the ability of these methods to preserve some spatial structures and chemical bonds that are not preserved via SF-GP and PS-PA. The Raman peak at 748 cm^-1 is derived from lanosterol ( Supplementary Figure S3 ) and that at 1,304 cm^-1 represents the amide II band (Fujioka et al., 2003; Pijanka et al., 2010), which are absent in the PS-PA-preserved samples, indicating that the PS-PA method alters the amino acid composition and disrupts the spatial structure of proteins. The 972 cm^-1 peak is specific to the SF and SF-MeOH samples and can be assigned to lipids (Czamara et al., 2015). This indicates that fixatives can destroy lipids and should be used with caution during the interpretation of lipid changes. The samples preserved via SF have a unique Raman peak at 1,738 cm^-1, and while those via SF-MeOH have a unique Raman peak at 1,746 cm^-1—both of which can be attributed to the C=O stretching (lipids) (Lakshmi et al., 2002; Silveira et al., 2002; Huang et al., 2003; Stone et al., 2004; Krafft et al., 2005; Dukor, 2006; Huang et al., 2011; Czamara et al., 2015; Gieroba et al., 2020); these results indicate that SF and SF-MeOH are effective for the preservation of triacylglycerols (Czamara et al., 2015). The disappearance of the C=O band in the spectra of the samples preserved by the other two methods may be due to the fixation of aldehydes and the xylene treatment, resulting in the massive removal of cellular lipids from the tissues (Lyng et al., 2011).

In addition, we found that the Raman spectra for the SF samples were significantly stronger (with several peaks at around 517, 656, 748, 972, 1,128, 1,171, and 1,586 cm^-1, which may represent the S-S stretching (Van Wart et al., 1973; Maquelin et al., 2002), C-S stretching (Sugeta, 1975), lanosterol ( Supplementary Figure S3 ), lipid (Czamara et al., 2015), carbohydrate (Wiercigroch et al., 2017), lanosterol, and amide II bands (Chen et al., 1995a; Chen et al., 1995b), respectively) than those for the other samples, suggesting that the concentration of these substances is higher when they are not contaminated with chemical fixatives. Therefore, the method associated with fixation will reduce the concentration of these substances.

Differences in the overall composition of the Raman spectra are expected; unfortunately, the spectra may cause some weaker peaks to be missed in the averaging process. Therefore, to further investigate the effects of different preservation methods on different biomolecules, we compared the enriched regions of nucleic acids, proteins, and lipids separately ( Figure 4 ). In the nucleic acid enriched region (nucleus), the peaks in the Raman spectra of the SF and SF-MeOH samples at 1,095 cm^-1 can be attributed to DNA (Malini et al., 2006), indicating that SF-MeOH is a good fixative for preserving nucleic acids. The Raman peak at 1,102 cm^-1 is present in the SF-GP and PS-PA samples and belongs to amide III (Lakshmi et al., 2002). This indicates that SF and SF-MeOH may disrupt the secondary structures of proteins and that different methods have different effects on the spatial structure of proteins. In the region of protein enrichment (cytoplasm containing symbionts), the SF-GP sample possessed unique Raman peaks near 505 cm^-1 representing S-S stretching (Sugeta, 1975; Van Wart and Scheraga, 1986), indicating that aldehyde fixation may well preserve the tertiary structure of proteins, while other methods would likely lead to tertiary denaturation. Alcohols denature proteins by disrupting their tertiary structures (Shao et al., 2012). The SF method may have led to conformational changes in the proteins (Lyng et al., 2011). In the phospholipid enrichment region (basal membrane), the spectra of the SF sample contained a peak at 896 cm^-1 corresponding to the CH₃ rocking band, which can be attributed to the presence of the fatty acid chain (Maquelin et al., 2002). In addition, the spectra of the SF sample contained a peak at 1,309 cm^-1, which is ascribed to the CH₃/CH₂ twisting or bending mode of lipid/collagen (Cheng et al., 2005), and one at 1,360 cm^-1, corresponding to the tryptophan bands (Huang et al., 2003; Cheng et al., 2005). The band at 450 cm^-1 in the spectra of the SF and SF-MeOH samples is due to the skeletal mode of carbohydrates (Wiercigroch et al., 2017). Only the SF-GP samples did not undergo the uracil ring stretching, whose peak is located at 1,395 cm^-1 (Clemens et al., 2014). These results suggest that SF-MeOH, SF-GP, and PS-PA disrupt the fatty acid chains, SF and SF-MeOH are more effective for the observation of the Raman peaks ascribed to carbohydrates, and SF-GP leads to protein destruction.

In addition to the direct analysis of the different peak positions, we can further identify the overlapping biomolecules by fitting the profiles via deconvolution analysis in order to obtain more information. We can observe the spectra of the samples obtained by the different methods to show significantly different profiles in terms of peak position, intensity, and width in the range of 1,500–1,700 cm^-1. The deconvolution analysis results ( Figure 5 and Table 2 ) confirmed the existence of vibrational modes such as β-sheet, α-helix, unordered, turn, and aromatic amino acid ring modes, and side chains of the amide band (Goormaghtigh et al., 1994; Chen et al., 1995a; Chen et al., 1995b; Pelton and McLean, 2000; Barth and Zscherp, 2002; Faoláin et al., 2005; Chan et al., 2006; Lefèvre et al., 2007; Kochan et al., 2013; Yang et al., 2015; Rivas-Arancibia et al., 2017; Gieroba et al., 2020; Kowalska et al., 2020; Talaikis et al., 2020). Among them, a peak corresponding to a mixture of lipid and protein may exist at around 1,664 cm^-1 (Kochan et al., 2013). The peak width range of 1,570-1,579 cm^-1 was fitted in the total average spectra of the nucleus and cytoplasm containing symbionts for both the SF and SF-MeOH samples; these peaks can be attributed to DNA (Chan et al., 2006; Gieroba et al., 2020) ( Supplementary Figure S4 and Table 2 ). This result reconfirmed that methanol can better preserve nucleic acids in cell nuclei, and has a stronger preservation effect for DNA in the cytoplasm containing symbionts. In terms of the protein structure, the SF-GP and PS-PA samples possessed unique turn structures, and the α-helix structure was completely disrupted in PS-PA.

To further investigate specific changes induced by the different preservation methods, we performed semi-quantitative analyses. Because the area of each peak corresponds to their conformational contribution (Kozicki et al., 2015), the area of each fraction was divided by the sum of the areas of all amide I band fractions to determine their relative contents. Table 3 summarizes the total average spectra of the different treatments and specific information on the subpeaks obtained after fitting the different sites for each method. From the results of the total average spectra fitting, the spectra of the SF and SF-MeOH samples contain the same vibrational modes, with differences in content; however, both were dominated by unordered structures, probably due to the depolymerization of proteins in the frozen sections during direct snap freezing (Lyng et al., 2011).

The turn structure accounts for the greater proportion of the cytoplasm containing symbionts region in the SF-GP samples, and the nucleus region in the PS-PA samples ( Supplementary Figure S4 and Supplementary Table S1 ). The β-sheet structures of the SF-GP and PS-PA samples were redundant with SF and SF-MeOH. The differences here indicate a significant change in the protein structure. The increased β-sheet amount indicates increased stiffness, easier tissue sectioning, and less fragmentation of the cells (Herrero et al., 2014). The SF-GP and PS-PA samples had a higher content of aromatic amino acids ring mode than the others, and the amino acids side chain structure accounted for the least. The lowest percentage of aromatic amino acids ring mode was found in the SF-MeOH sample. It may be because aldehyde fixation can cause cross-linking between functional groups in the side chains of amino acids, including cross-linking between adjacent amino acid chains and different positions of the same amino acid chain (Hobro and Smith, 2017). This process causes the protein to become insoluble. Concurrently, chemical changes occur, although it is usually more mechanically stable and suitable for subsequent processing, such as paraffin and resin embedding for sectioning.

In the basal membrane region, the content of β-sheet structures were high for all four methods in the three regions. Compared with SF, the immobilized samples showed lower contents of aromatic amino acid ring modes and amino acid side chain structures. The content of amino acid side chain structures in SF-GP was below the detection limit ( Supplementary Figure S4 and Supplementary Table S1 ). It has been suggested that the basal membrane is the most morphologically stable region in the cell and can maintain its integrity without fixation during sectioning. It is speculated that the fixation of this region is less effective for the preservation of amino acids. This result is in good agreement with the conclusion that we found a strong link between the ease of obtaining integrity sections and fixation during our experiments.