Surface‐enhanced Raman scattering spatial fingerprinting decodes the digestion behavior of lysosomes in live single cells

Lysosome, the digestive organelle in eukaryotic cells, plays an important role in the degradation and recirculation of cellular products as well as in maintaining the stability of cellular metabolic microenvironment. Surface‐enhanced Raman scattering (SERS) is a molecular fingerprint technology with high detection sensitivity and photostability, suited for revealing various intracellular molecular information by inducing endocytosis of SERS‐active nanoparticles. However, it remains challenging to selectively extract the molecular information of specific organelles (e.g., lysosomes) from a high‐dimensional spectral set. Herein, we proposed a novel paradigm by combining label‐free SERS spectroscopy with confocal fluorescence imaging to investigate the digestion behavior of lysosomes in cells. The structural similarity algorithm was innovatively introduced and exhibited its effectiveness in screening out the wavenumbers in the SERS spectral set with high correlation with the metabolic behaviors of lysosomes. With comprehensive experiments on HeLa single cells, we captured the intracellular macromolecular digestion phenomenon and discovered the changing pattern of cellular SERS spectra after starvation‐induced autophagy, and analyzed the molecular information within the lysosomes in three‐dimensional space.


INTRODUCTION
As the basic structural and functional unit of an organism, the cell continuously undergoes rapid material and energy exchange with the outside world.In other words, the metabolic activity of a single cell (or even subcellular) is a highly dynamic and heterogeneous system in space, where the lysosome is the digestive organelle of the cell.The lysosome participates in the degradation of damaged organelles and deformed macromolecules in the cytoplasm through the acidic environment 1 ; that is, autophagy plays an important role in the maintenance of the stability of the cellular microenvironment as well as in the synthesis, degradation, and recycling of cellular products. 2Abnormal lysosomal function has been reported to affect metabolic disorders, cellular homeostasis, inflammation, cancer, neurodegenerative diseases, and other physiopathological processes. 3Therefore, the lysosomal metabolic behavior is crucial and requires comprehensive investigation.Currently, transmission electron microscopy (TEM) is the most straightforward modality for observing morphological changes in lysosomes at various stages of autophagy, 3,4 but this technique is costly and time consuming.Alternatively, fluorescent probes designed in conjunction with well-recognized markers have been widely used in lysosomal imaging studies.However, they tend to focus on a limited number of target proteins 5 or localized microenvironmental (e.g., viscosity, polarity, pH) studies, 6 and are susceptible to quenching, photobleaching, and autofluorescence interference in biological systems.For the more challenging aspect of lysosomal functional research, current methods primarily rely on genomic, 7 proteomic, 8 and lipidomic 9 approaches to explore the functions and pathogenic mechanisms of certain macromolecules within lysosomes.Limited research has focused on the metabolic behaviors of molecules during lysosomal digestion.Metabolomics, as the endpoint of cellular activities, offers a direct and rapid reflection of the immediate state of cells. 10However, due to the vast diversity 11 and rapid fluctuations 12 in small molecular metabolites, amplifying their signals akin to those of genes is challenging.This challenge is particularly pronounced for detecting minute quantities of metabolites within individual cells. 13ompared to the aforementioned modalities, surfaceenhanced Raman scattering (SERS), a kind of molecular fingerprint spectroscopy, has the advantages of high sensitivity, high photostability, and direct response to molecular rotation and vibration information, 14 and its label-free approach is very suitable for the simultaneous detection of various molecular metabolic behaviors in cellular imaging. 15Taking above advantages of SERS, some studies have investigated molecular changes within cells. 16Generally, single-cell subcellular regions of interest are detected either by direct insertion using nanotips or nanopipettes or through nanoparticle (NP)-mediated cellular uptake for intracellular metabolic studies using SERS. 17These approaches enable real-time monitoring of complex intracellular biomolecules such as nucleic acids, 16c,18 metabolites, 19 drug molecules, 16b etc.However, due to the heterogeneity of cellular space, single or averaged spectrum hardly satisfies the characterization of global molecular information within the cell, and the increased number and dimensionality of spectra pose further challenges to the spectral interpretation. 20evelopment of effective methods to selectively extract molecular information from high-dimensional datasets is difficult but essential.
In this work, we innovatively utilized a label-free SERS mapping technique in live single cells combined with an image similarity assessment algorithm, structural similarity (SSIM), 21 to filter the wavenumbers that are highly correlated with lysosomal metabolic behaviors and to selectively extract molecular change information in HeLa cell models from the high-dimensional spectral sets.SSIM is a metric that measures the similarity of two images based on a combination of luminance, contrast, and structure comparisons for the image quality measures. 21We statistically captured the digestion phenomenon of intracellular macromolecules, searched for the generalized intercellular regularity of this degradation, and provided a detailed analysis of multiple molecular bond information within the cellular lysosome in three-dimensional (3D) space.Thus, our study paves the way for efficient extraction of image-based molecular information of subcellular structures using intracellular SERS hyperspectral datasets.

Evaluation system development for SERS spectra in lysosome
To establish a platform for quantitative assessment of the spatial overlap of SERS and fluorescence information (Figure 1), we designed experiments simultaneously containing cellular SERS mapping and fluorescence imaging assisted by the SSIM algorithm.It has been reported that lysosomes accumulate most of the internalized NPs with a "pulse-chase" strategy. 22pecifically, for adherent cells in grid culture dishes, we fed 50 µL of silver (Ag) colloids (4 nM) in fresh Dulbecco's modified Eagle's medium (DMEM) (Figure 1A) and cultured further for 16 h ("pulse") to allow cells to endocytose various NPs inside to enhance the intracellular molecular Raman signals.Then, the cells were washed to remove all extracellular NPs, followed by a 3 h "chase" period in fresh, particle-free culture medium for further live cellular SERS mapping.This procedure ensures that lysosomes can contain a large number of NPs. 23Furthermore, the lysosome contains a wide range of acidic hydrolases capable of digesting macromolecules (i.e., protein, nucleic acid, and lipid), 24 and its acidic environment induces the aggregation of particles, 25 easily leading to the formation of a multitude of electromagnetic hotspots to give rise to strong Raman signal enhancement, which reflects the molecular bond vibration related to lysosomal metabolism (Figure 1B).After SERS mapping, cells were fixed and labeled LAMP2A, a specific protein shuttle between lysosomal membranes, 26 with the fluorescence dye.Thereafter, with the traceback by grid positioning in the dish, we performed confocal fluorescence imaging of the same cells that underwent SERS mapping.
In order to decode the wavenumber ranges of Raman shift that are strongly correlated with lysosomal degradation, we have herein proposed the use of SSIM to measure the degree of correlation between SERS mappings and fluorescence images.Indeed, in assessing similarity for screening out wavenumbers in SERS spectra, the Pearson correlation coefficient (PCC) algorithm is widely employed. 27However, PCC calculates linear relationships between two variables, 28 thereby disregarding spatial information.In contrast, SSIM takes spatial information into account, enabling a robust evaluation of SSIM between images.Furthermore, unlike the simplistic calculation of absolute errors in mean squared error or peak signal-to-noise ratio (SNR), 29 SSIM quantifies the similarity between two images by comprehensively considering three critical components: luminance, contrast, and structure. 21The SSIM index ranges from 0 to 1, where a value closer to 1 indicates higher similarity between the compared images.In addition, SSIM is a widely used metric for assessing the perceived quality of images or signals and provides a comprehensive assessment of image similarity by considering not only global intensity differences but also local variations and structural features. 30SIM has indeed found applications in the field of SERS.
We found that SSIM has been utilized to enhance image quality in Raman imaging. 31For instance, it has been employed in combination with denoising algorithms to determine the SNR threshold for rapid cellular imaging under low SNR conditions.31b,32 SSIM has also been utilized to evaluate image quality for validating novel imaging techniques aimed at improving Raman imaging speed.31a However, it is worth noting that SSIM has not yet been applied to determine characteristic SERS wavenumbers.In our analysis system, we set an abundance of SERS hyperspectral mappings and fluorescence images (gray value of the red channel) as inputs.Each Raman wavenumber represents specific molecular fingerprint information.Therefore, for each Raman wavenumber, we gained a mapping based on the cellular positional information.This mapping intensity corresponds to the intensity of the Raman wavenumber.It was then multiplied by the binary mask derived from the fluorescence image to determine the cellular shape.The thousands of Raman wavenumber maps (400-1800 cm -1 ) for each cell constitute the SERS hyperspectral mappings.The SERS hyperspectral mappings were then sequentially compared with the fluorescence image to calculate the SSIM (0-1).A higher SSIM value indicates a stronger correlation between some Raman wavenumbers and lysosomal location (Figure 1C).It is worth noting that the molecular information we captured mirrors the metabolic activity of the lysosome in live cells.

Characterization and locations of Ag NPs in a HeLa single cell
In our work, citrate-stabilized Ag NPs were employed as transducers to enhance otherwise weak intracellular Raman signals for its high SERS activity. 33These labelfree plasmonic materials have been reported to be capable of enhancing the Raman signals of a wide variety of molecules.15c,34 TEM revealed that the Ag NPs had a diameter of about 50 nm, with a typical extinction peak at 412 nm (Figure 2A,B).The electronegative zeta potential (-37.8 mV) and hydrodynamic diameter distribution (polydispersity index, PDI = 0.248) also indicate a suitable charge and uniform size of particles for cellular endocytosis 35 (Figure 2C,D).From the brightfield image of a HeLa cell with a large number of internalized NPs after 16 h of endocytosis (Figure 2E), the NPs showed black aggregates located in the HeLa cellular cytoplasm (orange square in brightfield image, red fluorescence) rather than in the nucleus (blue square in brightfield image, blue fluorescence) limited by the channel size of the nucleopore (<30 nm). 36This situation is consistent with previous reports about the cellular location of NPs.16a, 20 Figure 2F shows the heatmaps of three SERS spectral sets collected from the square regions in Figure 2E (see the corresponding spectra in Figure S1).Considering the oxidation influence on Ag NPs, we did not conduct long-term, highpower imaging.Each pixel was scanned at a low power for 0.5 s to minimize the impact of oxidation.In addition, to minimize the photodamage from the photothermal effect of Ag NPs and the impact of spontaneous fluorescence background, we chose an excitation wavelength of 638 nm.It can be seen that there are ignorable SERS signals in the background region (green), and a high SNR of the fluctuating fingerprinting in the NP aggregation region (orange, cytoplasm) containing rich molecular information (e.g., 420-700 cm -1 , nucleobases, nucleic acids, and carbohydrates; 1500-1700 cm -1 , amino acids, proteins, and lipids). 20,37In contrast, we also found rather weak signals in the nucleus (blue) where these SERS peaks (pointed out by arrows in Figure S1) have the same spectral peak positions as those in cytoplasm but weaker signals, indicating that they probably came from the Raman scattering of cytoplasm close to the periphery of the nucleus.

Lysosomal degradation decoding in a HeLa single cell
For the single-cell analysis (see Figure 3A), we performed SERS mapping by a confocal Raman system with a 60× objective and 638 nm incident laser and collected fluorescence images by a confocal fluorescent microscope.After SERS spectral pre-processing, the SERS image of the cell (the bottom panel in Figure 3A) was reconstructed from the integral of the spectral intensity from 400 to 1800 cm -1 , which presented a cellular profile matching the lysosomal staining.This illustrates that the majority of NPs and lysosomes are in the cytoplasm.Figure 3B shows SERS spectra at some pixels indicated by the red symbols in Figure 3A with high SNRs.In contrast to the background spectrum with almost clean signal (bottom in Figure 3B), the spectra at different spatial locations of the cell showed heterogeneity.Two Raman peaks assigned to the S-S (524 cm −1 ) 38 and C-S (630 cm −1 ) 39 stretching vibrations presented stably but with different height ratios in each spectrum (see details of peak assignment in Table S1).They can be associated with the degradation of proteins in the lysosome, during which the polypeptides break down the disulfide bonds and transform into amino acids or smaller polypeptides so that the C-S hidden in the protein structure can approach the surface of the NPs.The ratio of the peaks therefore varies with the degree of protein degradation.38b,39b For instance, the spectrum of protein degradation marked by a red solid up-pointing triangle shows strong characteristic peaks assigned to exposed C-S vibrational stretch mode (630 cm -1 ), amino acid residues (960 cm -1 : Tyr, 1472 cm -1 : His, 1560 cm -1 : COO -, 1640 cm -1 : NH 3 + ), 39,40 and random coil (1253 cm -1 : amide III) 41 conformation in smaller peptides due to the acidic pH of lysosomes. 20rovided that Raman fingerprint spectra contain profuse molecular information inside the cell, we decipher the lysosome-related information therein by calculating the SSIM index from each wavenumber reconstructed Raman mapping and fluorescence image (Figure 3C).Compared with the mean spectrum from the cellular region (red curve in Figure 3C), a number of Raman bands with high SSIM values (∼0.7) were marked by dotted lines assigned to vibrational modes of macromolecular degradation (red: protein, green: nucleic acid).In order to figure out which substances are detected inside the cells, we further tested the SERS signal of HeLa cell lysates using the same Ag NPs. Figure 3D compared the averaged spectra of HeLa cells in situ (endocytosis) and ex situ (cellular lysates) and surprisingly showed that the SERS peak positions of the two measurements were nearly identical.Since most SERS signals measured in the lysates are dominated by small molecules (the signals are similar before and after the removal of macromolecules from the lysates, Figure S2), we demonstrated that some of the SERS signals measured in situ using the citrate-stabilized Ag NPs are from small molecules or residues from the breakdown of macromolecules.Additionally, some significantly stronger peaks from the in situ spectra are likely to be attributed to the pre-degradation state of macromolecules, such as disulfide bonds (524 cm -1 ) in proteins. 38To verify the correlation of the lysosomal characteristic peaks, we screened based on SSIM values in both imaging and explicitly displayed the intensity of peaks associated with proteolysis in the strong fluorescence region, the weak fluorescence region, and the background region of the same cell (Figure 3E).As a consequence of the protein degradation into fragments in the endolysosomal system, the significantly increased C-S to S-S peak area ratio indicates that with the breakdown of the disulfide bonds, C-S bonds are more accessible for binding to the surface of NPs.The same regularity was also observed for Raman bands associated with nucleic acid cleavage in Figure 3F.The results showed that the stronger the fluorescence, the stronger the Raman signal attributed to nucleic acid cleavage exposing the phosphate backbone (1090 cm -1 ) as well as the nucleobase (1175 cm -1 : T, 1300 cm -1 : A, C). 40b It is also noteworthy that some Raman shifts with high SSIM values are not yet explainable conclusively, which are guessed to result from the contribution of some lysosome-related peaks.

Regularity study of lysosomal degradation in HeLa single cells
To investigate the regularity of molecular fingerprinting in lysosomal digestion, 12 HeLa cells were examined for statistical analysis.Figure 4A shows that there was a good match of the cell profile between the total Raman intensity mapping and fluorescence image for most cells, whereas this was not the case for a few cells (e.g., cell 4 and cell 8).Because the SERS test examines the signal enhanced by NPs in the entire cell, there is indeed a phenomenon where a certain number of NPs are not located in the lysosomes in some cells.Therefore, pixels with strong SERS signals may not all come from lysosomal contributions; they may also come from molecules in other parts of the cytoplasm.In this case, a mismatch between fluorescence and SERS may occur.This is why it is necessary to further use SSIM to extract Raman wavenumbers related to lysosomes.Similar to the above process, we obtained one SSIM curve for each single cell, and the averaged SSIM curve of 12 cells with standard deviation provided more robust and comprehensive molecular fingerprinting pattern of the single-cell populations (Figure 4B).Along the average SSIM curve, some wavenumbers with high SSIM values were highlighted and annotated, which are most likely attributed to protein, nucleic acid, and lipid degradation in lysosomal digestion.For example, apart from the peaks representing protein (red dotted lines) and nucleic acid (green dotted lines) breakdown as screened in Figure 3C, peaks of lipid (blue dotted lines) breakdown were also identified.39a To further support these findings, we compared the peak intensities of these wavenumbers in the strong and weak fluorescence regions in the same cell.A significantly stronger Raman activity was found in the strong fluorescence region (Figure 4C).Moreover, the C-S to S-S peak area ratio was also significantly enhanced in the strong fluorescence region in the majority of cells, suggesting that active protein degradation processes occur in the lysosomal region of most cells (Figures S3 and 4D).To further promote the metabolic process of cellular lysosomes, we starved cells in nutrient-deficient medium, a process that accelerates the delivery of cytoplasmic material to lysosomes for degradation. 42An monodansylcadaverin (MDC) kit was used to characterize the degree of autophagy, which produces a green fluorescence after characteristically labeling the autophagosome.It is worth noting that before starvation induction, cells in the same state were exposed to equivalent concentrations of NPs.Therefore, it can be reasonably considered that the detected changes primarily result from starvation rather than from the influence of NPs.Compared with the control group, stronger green fluorescence was observed in the starvation group, indicating a more pronounced autophagy phenomenon (Figure S4).Twelve single cells were measured for statistical analysis (Figure S5), and starvation-induced perturbations of some SERS peaks were detected in autophagy group (Figure 4E).Among the lysosomal characteristic peaks, screened with SSIM, some peaks had an elevation (e.g., PO 2 − vibration of nucleic acid fragments, C-terminal vibration of residues of proteins, ═CH deformation, and C═C vibration of degradation of lipids), which implies that autophagy encourages lysosomes to degrade macromolecules and expose residues more actively.However, there is a substantial downregulation of the C-C bonds of the lipid backbone, which may be related to the phase transition of the fragmented lipid bilayers in the lysosomal acidic environment (Figure 4F). 43

Three-dimensional molecular information of HeLa single-cell lysosomal degradation
Next, we investigated the 3D spatial molecular characterization of single-cell lysosomes.The thickness of the adherent cell is typically a few micrometers, 44 so we set a step size of 2 µm in the z-direction.The cell was observed in focus under white light illumination (set z = 0), and then 3D cellular profiles were generated from a distance of -2 to +2 µm from this layer.Figure 5A shows the mappings reconstructed from the total Raman intensity from 400 to 1800 cm -1 , and layer 2 exhibited a stronger signal overall, which may be related to the fact that the NPs and/or lysosomes are mostly concentrated in this layer.To analyze the chemical composition of the labeled lysosomal pixels, Figure 5B shows Raman spectrum corresponding to labeled pixels in all three layers of this single cell in Figure 5A.The spectra of some pixels in the z-direction demonstrated obviously larger variability, while others showed only minor fluctuations.For instance, pixels marked by down-pointing triangles from three layers on the left side of the nucleus (Figure 5A) shared very similar spectral shapes while only the S-S bond has a sudden increase in layer 2. This disparate spectrum could indicate that the intact protein structure before degradation happened to fall into a strong hot spot and was captured.Next, we used the characteristic lysosomal peaks according to the SSIM algorithm on the 3D cellular area and found that layer 2 consistently exhibited higher peak intensity (Figure 5C) and (C-S)/(S-S) ratio (Figure 5D).It can be concluded that the middle layer has a more active biomolecular breakdown behavior.Typically, in most Raman imaging experiments, the z-axis information tends to be ignored. 45o examine the necessity of multilayer data acquisition on cells with thickness, we conducted ablation experiments (Figure 5E).Obviously, the more layers included, the larger the statistical and average SSIM values (Table S2), which means that more comprehensive lysosomal information was acquired.And the improved SSIM proves the importance of performing all layers' features for the global understanding.

CONCLUSIONS
In summary, we combined Raman hyperspectral imaging and immunofluorescence imaging to study the digestion behavior of lysosomes in cells that ingested and trapped NPs through a "pulse-chase" strategy, and Ag NPs substantially enhanced the Raman signals of intracellular molecules.For the first time, we innovatively introduced an image evaluation algorithm, SSIM, to identify the characteristic peaks that are strongly associated with lysosomal metabolic activities among the numerous molecular vibrational information.These peaks were found to be related to the degradation of macromolecules, which also demonstrated a certain noteworthy regularity in SERS spectra in cells after starvation-induced autophagy.More targeted molecular characterization of complex mixtures with hyperspectral features enabled characterizing molecular experiences such as the breakdown of proteins, nucleic acids, and lipids and the increase in acidity in the lysosomal region.Moreover, the 3D experiment with single-cell lysosomes demonstrated that the higher or finer spatial dimension of the spectra significantly benefited more comprehensive analysis of molecular characterization.We believe our method of screening characteristic peaks in complex fingerprint spectral datasets can be generalized to the study of molecular behaviors of subcellular structures.
We have also noticed that several key areas require more in-depth examination.Initially, the SERS-active particles utilized in this study were predominantly Ag NPs stabilized with citrate.Although these Ag NPs exhibit excellent enhancement capabilities, they are known to possess potential toxicity and susceptibility to oxidation.Consequently, it is essential to optimize the material to achieve a balance between enhancement performance, stability, and biocompatibility.Second, the homogeneous surface chemistry of the materials results in selective adsorption of biomolecules, thereby biasing the enhancement of different molecular Raman signals.To obtain a more comprehensive molecular profile, it is imperative to introduce a variety of NPs with diverse surface properties, including different materials, shapes, and charges.By analyzing the spectra enhanced by multiple materials for the same sample, we can extract additional information on metabolites.Moreover, there is potential for SERS analysis at the molecular level within cellular compartments beyond the cytoplasm, such as other organelles and the nucleus.By employing SERS probes with specific molecular affinities to target cellular structures in combination with fluorescent dyes at different targeting sites, and integrating algorithmic feature selection, our approach holds promise for expanding the study of molecular behaviors within additional subcellular structures.This expansion could facilitate a deeper exploration of the intricate networks within cells, thereby providing richer insights into disease mechanisms, drug development, and cellular biology.

Synthesis of Ag NPs
Citrate-stabilized Ag NPs were prepared according to Lee-Meisel method with slight modification. 46

Cell experiments
Human cervical cancer (HeLa) cells were obtained from Shanghai Institute for Biological Sciences.For in situ experiments, the cells were cultured in DMEM supplemented with 10% FBS and 1% penicillin-streptomycin at 37 • C in a 5% CO 2 incubator.The cells were seeded at a density of 1 × 10 5 in grid culture dish, which consisted of a thin glass bottom with grid to locate cells and allows high-quality light and fluorescence microscopy.Adherent cells were washed with PBS, 1 mL of medium spiked with 50 µL of the Ag colloids (4 nM) was added, and the cells were cultured at 37 • C in a 5% CO 2 incubator for 16 h.Then, cells were washed with PBS three times, the control group was suspended in fresh culture medium, the autophagy group was suspended in nutrient-deficient culture medium, and then they were left for 3 h in the incubator.
An MDC test kit was used to detect the occurrence of lysosomal autophagy.Control group and autophagy group were washed with PBS, 1 mL of MDC staining solution (1×) was added, and cultured at 37 • C in a 5% CO 2 incubator for 30 min in dark.Then, the staining solution was removed and cells were washed with assay buffer three times.Then, the assay buffer was removed and 1 mL of assay buffer was added.The cells were excited with 355 ± 56 nm excitation light under a fluorescence microscope.
For ex situ experiments, HeLa cells were harvested by scraping and lysed using repeated freeze-thaw cycles.The cells were subjected to three cycles of freezing process in liquid nitrogen for 30 s, followed by a thawing process at 37 • C for 3 min.Before measurement, 200 µL of the lysate was centrifuged to remove cell debris (400 g, 5 min), and when excluding macromolecules in lysate, a 3 kDa cutoff filter (Millipore, Amicon Ultra-4, PLBC Ultracel-3) was implemented under centrifugation (10,000 rpm, 10 min).

SERS measurement
For in situ experiments, a confocal Raman system with a 60× objective and 638 nm incident laser (power: 2.59 mW) was used for all Raman mapping of living cells.All spectra were obtained in a 0.5 s acquisition time in the pointwise scanning mode with a step size of 2 µm.For z-axis detection, each layer was 2 µm apart.For ex situ experiments, the lysate was mixed with Ag NPs at a volume ratio of 1:1 at room temperature.Then, 10 µL of the mixture was injected into a quartz capillary (I.D.: 1 mm, O.D.: 2 mm) for SERS measurements on a confocal Raman system with the 638 nm incident laser (power: 12.7 mW) for 1 s acquisition time per spectrum through a 10× objective lens.In total, 100 spectra were collected for each sample.

Fluorescent staining
For the cells that were Raman tested in Section 4.4 , they were washed with PBS and fixed with 4% paraformaldehyde for 15 min at room temperature, then washed with PBS three times.Then, cells were permeated with 0.1% Triton X-100 solution for 5 min at room temperature and washed with PBS three times.For lysosome labeling, cells were incubated overnight at 4 • C in the dark with 200 µL of ab282009 at a 1/100 dilution and then washed with PBS three times.Thereafter, the tablets were covered with a drop of VECTASHIELD with DAPI to stain nuclear DNA and stored at 4 • C in the dark.

Confocal fluorescent imaging
The cells stained in Section 4.5 were located again using confocal fluorescent microscope assisted with grid on culture dishes.Cellular fluorescence images were observed with a 60× objective (water immersion, N.A. = 1.2) for finer microscopy.DAPI was subsequently excited by 473 nm laser and detected by a photomultiplier tube with a 490-520 nm filter.Alexa Fluor 555 was excited by 543 nm laser and detected at 560-620 nm.Images were captured with 800 × 800 pixels at a speed of 2 µs per pixel.

Data processing and analysis
All data from our experiments were pre-processed and analyzed using Python and MATLAB R2022b.All Raman spectra were smoothed by Savitzky-Golay, a filter using a polynomial order of 3 and a frame length of 9. 47 Afterward, the baseline of every spectrum was corrected by the adaptive iteratively reweighted penalized least squares based on the order of 3 and lambda of 100. 48All spectral data from the cellular mapping were analyzed in the region of 400-1800 cm −1 .For each wavenumber, we gained a Raman mapping, which is called hyperspectral image.
Considering different resolutions of Raman images and fluorescence images, the red channel (lysosome) in the fluorescence image was resized to match the dimensions of hyperspectral images for comparison.Then, the red channel images were binarized to determine the binary mask shapes of the cell in Raman mappings (threshold = 7).Finally, the numerical similarity between the abundant Raman hyperspectral mappings and resized red channel image was measured by the SSIM, 21 which compares local patterns of pixel intensities that have been normalized for luminance and contrast, to quantitatively evaluate the co-localization level of lysosomes and Raman shifts.The normalization process helps alleviate the potential influence of differences in particle size or aggregation on the SERS signals.The SSIM of the two sets of data ranges between 0 and 1, and empirically, the SSIM index exceeding 0.7 exhibits a fairly strong overlap, indicating high correlation between the fingerprinting spectra of some molecular chemical vibrations and lysosomal metabolism.

Statistical analysis
To assess statistical significance, the two-tailed t-test and paired t-test were applied for the comparison of the two groups.Quantitative results were presented with mean values.The statistical significance indicated in the figures was assigned as not significant (ns); p >*.05; *p ≤*.05; **p ≤*.01; ***p ≤*.001.

A C K N O W L E D G M E N T S
This work was supported by grants from the National Natural Science Foundation of China (no.82272054), the Science and Technology Commission of Shanghai Municipality (no.21511102100), the Shanghai Jiao Tong University (no.YG2024LC09), and the Shanghai Key Laboratory of Gynecologic Oncology.

F I G U R E 1
Schematic illustration of using surface-enhanced Raman scattering (SERS) and fluorescence imaging to illustrate co-localization of various Raman shifts and lysosomes.(A) Experimental procedure for inducing silver (Ag) nanoparticles (NPs) (orange spheres) into adherent cells (1 and 2), followed by SERS mapping (3) and fluorescence imaging (4).(B) The mechanism of using Ag NPs to monitor the degradation of macromolecules (i.e., protein, nucleic acid, lipid) in the lysosomes.(C) Diagram of the similarity measure of abundant hyperspectral SERS mappings and fluorescence image assisted by structural similarity (SSIM) algorithm.

F
I G U R E 2 Characterization of silver (Ag) nanoparticles (NPs) in HeLa cells.(A) Transmission electron microscopy (TEM) image of Ag NPs.(B) UV/vis spectrum shows an extinction peak at 412 nm (the inset: photograph of Ag NPs).(C) The zeta potential of Ag NPs is −37.8 mV.(D) The average hydrodynamic diameter of Ag NPs is 78.8 nm (PDI = 0.248).(E) Brightfield image illustrates the majority of the NPs distribution is located in HeLa cellular cytoplasm instead of the nucleus.(F) Heatmap of a surface-enhanced Raman scattering (SERS) spectral set collected from the cytoplasm (orange), nucleus (blue), and background (green), the 1st to 16th spectra represent the SERS signal of each pixel from top left to bottom right in turn in the corresponding squares in (E).The fluctuating spectra are shown in Figure S1.

F I G U R E 3
Raman spectra and structural similarity (SSIM) of a HeLa single cell, lysosomal degradation in the same cell shows spectral heterogeneity.(A) Brightfield image, fluorescence confocal microscopy image, and total Raman signal mapping of a HeLa single cell.The fluorescence image was generated using the red channel (lysosome), and the Raman image was reconstructed from the total Raman intensity from 400 to 1800 cm −1 ; all scale bars are 10 µm.(B) Raman spectra from pixels marked in (A), Raman shifts marked by shaded areas are assigned to vibrational modes of molecules corresponding to lysosomal degradation (red: protein, green: nucleic acid).(C) SSIM in (A) and the average spectrum of the cellular region (red curve), Raman shifts marked by dotted lines are assigned to vibrational modes of lysosomal degradation (red: protein, green: nucleic acid).(D) Comparison of the averaged spectra of HeLa cells in situ (endocytosis) and ex situ (cellular lysates).Red shaded regions highlight the surface-enhanced Raman scattering (SERS) bands at the similar wavenumbers between these two spectra.(E) Pixels on the fluorescence image in (A) are divided into strong fluorescence (FL), weak FL, and background (BG) according to fluorescence intensity (20 points are taken from each), and the peak area ratio of C-S to S-S from strong FL (red), weak FL (pink), and BG (gray) indicates that the proteins are degraded into small molecules.(F) Scatter diagram of strong FL (green), weak FL (light green), and BG (gray), and the axes represent peaks that can be assigned to vibrational modes of the cleavage of nucleic acid (PO 2 − , T, A, C).

F I G U R E 4
Raman spectra and structural similarity (SSIM) of single cells.(A) Brightfield image, total Raman signal mapping, and fluorescence confocal microscopy image of 12 single cells.The Raman image was reconstructed from the total Raman intensity from 400 to 1800 cm −1 ; all scale bars are 10 µm.(B) The average SSIM curve over 12 cells in (A), Raman shifts marked by dotted lines are assigned to vibrational modes of molecules corresponding to lysosomal degradation (red: protein, green: nucleic acid, blue: lipids).The shaded areas indicate the standard deviation.(C) Pixels on the fluorescence image in (A) are divided into strong fluorescence (FL) (red) and weak FL (blue) according to fluorescence intensity (20 points are taken from each), each point corresponds to the average intensity and the connected points come from the same cell, the paired t-test of vibrational modes with different fluorescence intensity revealed significance.(D) The paired t-test for the peak area ratio of C-S to S-S indicated that the strong FL group (red) is significantly higher than the weak FL group (blue).(E) Comparison of the averaged surface-enhanced Raman scattering (SERS) spectra of HeLa cells before (black curve) and after (red curve) starvation.Shaded regions highlight the lysosomal characteristic peaks (n = 12).(F) Comparison of the characteristic features of HeLa cells before (black bar) and after (red bar) starvation (n = 12).

F
I G U R E 5 Three-dimensional spatial metabolic activity of a single-cellular lysosome, different z-axis in the same cell shows heterogeneity.(A) The heatmap of reconstructed from the total Raman intensity from 400 to 1800 cm −1 of a single cell from three layers on the z-axis (-2, 0, and 2 µm), layer 2 shows the highest intensity; the scale bar is 10 µm.(B) Raman spectra from pixels indicated by the symbols in Raman mapping in (A), Raman shifts marked by shaded areas are assigned to vibrational modes of lysosomal degradation (red: protein, green: nucleic acid, blue: lipids).(C) Mean Raman intensity of peaks that can be assigned to vibrational modes of macromolecular degradation from layer 1 (L1), layer 2 (L2), and layer 3 (L3).(D) The peak area ratio of C-S to S-S from different layers indicates that layer 2 has the most active protein degradation.(E) Ablation experiment results for the distribution of structural similarity (SSIM) calculated from different layer combinations of Raman signal and fluorescence signal.
Briefly, 12.3 mg of AgNO 3 was dissolved in 100 mL of ultrapure water.The solution was heated to boiling with constant stirring.Two milliliters of sodium citrate solution (1 wt%) was rapidly added.The mixture was kept boiling for 1 h, cooled to room temperature under stirring, and later stored at 4 • C. The theoretical concentration of Ag NPs was 0.2 nM.To concentrate, 1 mL of Ag colloids was centrifuged (5000 rpm, 8 min) and 950 µL of supernatant was removed (final concentration: 4 nM).