Using SERS Tags to Image the Three‐Dimensional Structure of Complex Cell Models

Methods to image complex 3D cell cultures are limited by issues such as fluorophore photobleaching and decomposition, poor excitation light penetration, and lack of complementary techniques to verify the 3D structure. Although it remains insufficiently demonstrated, surface‐enhanced Raman scattering (SERS) imaging is a promising tool for the characterization of biological complex systems. To this aim, a controllable 3D cell culture model which spans nearly 1 cm2 in surface footprint is designed. This structure is composed of fibroblasts containing SERS‐encoded nanoparticles (i.e., SERS tags), arranged in an alternating layered structure. This “sandwich” type structure allows monitoring of the SERS signals in the z‐axis and with mm dimensions in the xy‐axis. Taking advantage of correlative microscopy techniques such as electron microscopy, it is possible to corroborate nanoparticle positioning and distances in z‐depths of up to 150 µm. This study reveals a proof‐of‐concept method for detailed 3D SERS imaging of a complex, dense 3D cell culture model.

Methods to image complex 3D cell cultures are limited by issues such as fluorophore photobleaching and decomposition, poor excitation light penetration, and lack of complementary techniques to verify the 3D structure. Although it remains insufficiently demonstrated, surface-enhanced Raman scattering (SERS) imaging is a promising tool for the characterization of biological complex systems. To this aim, a controllable 3D cell culture model which spans nearly 1 cm 2 in surface footprint is designed. This structure is composed of fibroblasts containing SERS-encoded nanoparticles (i.e., SERS tags), arranged in an alternating layered structure. This "sandwich" type structure allows monitoring of the SERS signals in the z-axis and with mm dimensions in the xy-axis. Taking advantage of correlative microscopy techniques such as electron microscopy, it is possible to corroborate nanoparticle positioning and distances in z-depths of up to 150 µm. This study reveals a proof-of-concept method for detailed 3D SERS imaging of a complex, dense 3D cell culture model. the toxicity of a new (nano)material, [2] or to study cell metastasis in tumoral systems. [3] Furthermore, considering that the extracellular environment plays an important role in cell-to-cell communication, being also one of the key components of tumor invasion, [4,5] methods to incorporate it in cell cultures are essential toward achieving more physiologically relevant tissue models. One such method is based on allowing cells to grow in 3D, e.g., in spheroids, in which cell-to-cell interactions are closer to a real in vivo situation. [6,7] However, control over the internal structure and spatial location of the different biological components is complicated, and our ability to characterize such models is still rather limited; high penetration depth, reduced overlap between detection signals, and no/minimal photobleaching are all required. The use of nanomaterials as contrast agents [8] in advanced imaging techniques could provide the suitable level of precision required for accurate diagnostics and therapeutic monitoring. [9] As a rule, the successful spatial and temporal tracking of different cell populations in either 2D or 3D models requires a label (fluorescence, radioactive, Raman-active, etc.). Confocal fluorescence microscopy (CFM) is considered the reference technique and has been developed enormously over the past decades to include a wide range of high-resolution microscopy techniques such as spinning disk CFM, multiphoton CFM, and light-sheet CFM, all of which aim to improve resolution and contrast in the xyz-axes. [10,11] One of the main drawbacks of CFM is the difficulty in achieving long-term fluorescence cell labeling. This usually involves transfection of plasmids expressing fluorescent proteins, or the use of quantum dots, but their use is limited due to significant cytotoxicity. [12,13] An alternative method comprises the use of organic dyes or fluorescent proteins to label cells for short periods of time, normally ranging from 1 to 4 cell divisions, both of which are better tolerated and also offer more flexibility in the choice of cells and fluorophores to be used. [14] Regarding the physical limitations of fluorescence detection, whereas high levels of resolution and sensitivity can be achieved in living cells organized in monolayer structures, upon increasing the thickness in the z-axis various limitations arise, including a) inability of the light source to penetrate the sample and reach the fluorophore, b) inability of the emitted light to penetrate the sample and reach the detector, c) limitation in the fluorophores that can be used due to biological tissue absorption of light with Full PaPer

Using SERS Tags to Image the Three-Dimensional Structure of Complex Cell Models
Dorleta Jimenez de Aberasturi,* Malou Henriksen-Lacey,* Lucio Litti, Judith Langer, and Luis M. Liz-Marzán

Introduction
2D cell cultures have traditionally been used to evaluate the interaction of materials with biological systems. However, the use of 3D cell culture models offers the ability to increase complexity so that the model better represents the situation occurring in living tissues. This is a key aspect, for example, in the evaluation of promising anticancer drugs, [1] to determine wavelength in the visible spectrum, and d) overlapping signals due to broad emission bands. [15,16] Thus, when long-term cell cultures which may suffer from fluorophore chemical degradation or signal loss are considered, alternative methods to fluorescence are needed. [17] One such option is Raman scattering (RS). The ultimate advantage of RS is the small bandwidth of the detected spectral peaks (vibrational molecular fingerprint) which allows quantitative, multi-component identification. Although RS is an inherently improbable phenomenon, it can be amplified by several orders of magnitude when the molecule of interest interacts with plasmonic metal nanoparticles (NPs). [18] By matching the plasmon resonance of the nanoparticle with the excitation wavelength, this process, known as surface-enhanced Raman scattering (SERS), becomes particularly efficient. In analogy to CFM for live-cell imaging, SERS nanotags, composed of particularly "bright" labeling molecules (Raman reporters, RaRs) covalently bound to a plasmonic NP acting as an amplifier, must often be applied to label different cells, obtaining high SERS signals from each individual cell containing the different SERS nanotags. Thanks to this multiplexing capability, the different pre-labeled cells can be imaged using just one laser source. A key advantage of SERS nanotags against fluorophores is their stability against photobleaching, [19,20] which renders them excellent candidates for long-term imaging experiments. Furthermore, the precise control of related synthetic procedures has resulted in the availability of extremely bright near-infrared (NIR)-responsive SERS nanotags, [21] when irradiating with NIR laser sources, which provide access to deeper locations in biological systems than visible radiation. [22] Therefore, the localization of different SERS nanotag-labeled cells can be monitored over extended periods of time, with negligible loss of the SERS signal, thereby largely expanding the range of potential applications. [23][24][25] Furthermore, the noninvasive nature of SERS makes it compatible with in vivo experiments and avoids the need for fixing cells, which is crucial toward understanding cellular behavior under real-life conditions. [26] Even if SERS has revealed itself as a promising tool for the characterization of biological systems, [27] surprisingly few in vivo or 3D studies have been reported to date. [23] The use of both label-free NPs [28][29][30] and SERS nanotags [31,32] or cell tracking has been studied at a single-cell level, and SERS and surfaceenhanced spatially-offset Raman spectroscopy (SESORS) have been applied, without imaging, to analyze 3D spheroid cultures [33] and tissue analogues, [34] respectively. Reports on the use of SERS imaging in vivo have focused on cancer diagnostics via topical administration of SERS NPs coupled with endoscopy imaging, [35,36] or even intraoperative SERS-guided surgery. [37,38] Hurdles such as limited spatial and temporal resolution in animal studies have been overcome via the development of a fast small-animal Raman imaging (SARi) instrument, [39] which also led to improvements toward in vivo multiplexing. [40] For 3D in vitro SERS applications, more sophisticated microscopes and combined measurement techniques are emerging, [41,42] however, to the best of our knowledge, true 3D imaging of complex in vitro cell models is yet to be achieved.
We present a proof-of-concept for 3D SERS imaging and the ability to discriminate different SERS labeled cells within a complex, dense 3D cell culture model. The model is composed of a layer-by-layer multicellular system containing alternating unlabeled and SERS nanotag-labeled fibroblast cells, to mimic different cell populations within a tissue. Within this "sandwich" structure we focused on achieving the best possible 3D resolution for scans over large areas (100´s µm in axial and up to 1 mm in lateral dimensions), analyzed via in-depth data analysis of our SERS tags, and characterized and supported by correlative imaging techniques including CFM, transmission electron microscopy (TEM), and scanning electron microscopy (SEM).

Results and Discussion
In order to better understand the limitations of conventional CFM and SERS imaging regarding axial and lateral resolution, we designed a 3D cell model in which the distribution of SERS nanotags could be pre-designed. We chose to work with gold nanorods (AuNRs) and nanostars (AuNSs) as SERS enhancers, as the plasmon resonance of both NP types can be tuned to the NIR laser excitation wavelength of 785 nm, available in most Raman instruments. Furthermore, the similar NP size yet different NP morphology allows us to clearly distinguish both NPs via other microscopy methods, such as TEM.

SERS-Encoded Nanoparticles
SERS-encoded AuNRs and AuNSs, coated with a cationic layer of poly-L-arginine hydrochloride (PA), were prepared as SERS imaging probes, and used to label different cell populations, following the steps schematically described in Figure 1. [23] In brief, SERS encoded AuNSs and AuNRs were synthesized according to our previous work. [25] We selected the Raman reporters 2-naphthalenethiol (2NAT) and biphenyl-4thiol (4BPT), due to their unique SERS profile and high signal intensity upon 785 nm (NIR) laser excitation ( Figure 2C), for which light transmission in biological tissue is optimum. [22,43] The vibrations at 1282 cm −1 for 4BPT and 1381 cm −1 for 2NAT are highlighted in Figure 2C as the characteristic peaks chosen for SERS mapping. Surfactant-free AuNSs [44] and CTAB-capped AuNRs [45] were encoded by ligand exchange with 4BPT and 2NAT, respectively. [25] The corresponding dispersions in chloroform (CHCl 3 ) showed a slightly red-shifted localized surface plasmon resonance (LSPR), as compared to the initial dispersions in water, due to the higher refractive index of the organic solvent (Figure 2A.1-B.1). Subsequent addition of the amphiphilic polymer, dodecylamine-modified polyisobutylenealt-maleic polymer (PMA) (Section 2, Figure S1, Supporting Information) [46] led to water-soluble NPs with negative surface charge, as indicated by the measured zeta-potential values (Table S1, Supporting Information). Since it is well-known that cellular uptake is enhanced for positively charged NPs, we carried out an additional coating step with the cationic polymer PA [47] . After washing and redispersion in water, an LSPR blueshift was observed as expected, though still partly red-shifted with respect to the initial samples because of the coating shells (  particles with no signs of aggregation, in agreement with the Vis-NIR spectra. We additionally prepared both AuNRs and AuNSs coated with PMA that was fluorescently labeled with the dyes TAMRA (PMA-TAMRA) and amino-modified Dye-DY-633 (PMA-633) ( Figures S3-S5, Supporting Information). The fluorescence emission spectrum of AuNR-2NAT@PMA-TAMRA is shown in Figure 2D, and compared to that from PMA-TAMRA in solution. Fluorescent labeling of the polymer and coating of both AuNRs and AuNSs are described in the Supporting information (Section 2).

Cellular Uptake and Biocompatibility
Systematic studies on cell internalization and biocompatibility were carried out using dermal fibroblasts (HDF). This member of the connective tissue cell family presents the ability to form elongated bipolar cell morphologies which could exist in striated 3D structures, similar to the skin-like structure found in their natural environment. In addition to such 3D geometrical growth characteristics, which have been shown to be more realistic than those observed in 2D environments, [48,49] HDF cells have also been shown to avidly uptake NPs and to retain them for long periods of time. [50] We studied the uptake of AuNS-4BPT and AuNR-2NAT NPs by means of SERS imaging, comparing the same NPs coated with either an anionic or a cationic outer layer ( Figure 3A). SERS maps were created by plotting the specific vibrations of the RaRs, 1282 cm −1 and 1381 cm −1 for 4BPT and 2NAT respectively. The NPs with a negative surface charge displayed moderate levels of cellular uptake, whereas additional coating of the NP surface with PA resulted in considerably higher levels of endocytosis, as could be observed via both SERS and simple bright-field imaging ( Figures 3A and S6, Supporting Information). TEM and CFM further confirmed uptake of cationic NPs into intracellular vesicles, perinuclear locations and close to the cell membrane (Figures 3B,C, S7 and S8, Supporting Information).
Cationic NPs are often associated with cellular cytotoxicity, partly due to the specific molecules used to impart a positive surface charge having surfactant-like properties, but partly also due to excessively high levels of NP-uptake, which can affect the cellular properties. [51] As mentioned above, the adsorption of PA on the NP surface largely increased the level of cellular NP uptake. To verify that no overt NP-induced cytotoxicity occurred, we studied cell viability and proliferation over time, when cells were exposed to AuNS-4BPT@PMA-PA and AuNR-2NAT@PMA-PA. For HDF cells exposed to relatively high NP concentrations ([Au°] = 0.1 × 10 −3 m), we observed no direct effects on cell viability within the studied time period (≤ 72 h) ( Figure 3D). To verify that cell proliferation was not affected either, we analyzed HDF cell division over 11 days after incubation with both types of NPs, at a final concentration of [Au°] = 0.025 × 10 −3 m. As expected, HDF cells exposed to NPs successfully proliferated over time until reaching a highly confluent monolayer, matching the proliferation profile of non-NP exposed HDF cells ( Figure 3E). Furthermore, inductively coupled plasma-mass spectrometry (ICP-MS) analysis showed that cells retained the NPs over this period, suggesting that exocytosis was minimal or, if it occurred, the exocytosed NPs were again taken up by neighboring cells. The combination of these results highlights the biocompatibility of the prepared NPs, as: i) no short-term cytotoxicity, typically due to release of compounds such as CTAB, was noted; ii) no long-term effects were observed on cell division, which suggests that the NPs remain stable over time and do not affect the cell division process.

Cell Sandwich Model
Upon confirmation that HDF cells could be successfully labeled with AuNS-4BPT@PMA-PA and AuNR-2NAT@PMA-PA, we next designed a cell model in which the 3D spatial location of NPs could be pre-designed. To do so, it was important that the NP-containing HDF cells did not proliferate excessively, resulting in a loss of the defined 3D structure. A phenotype representative of cell senescence, whereby cell proliferation is limited or even stalled, was considered ideal for such a longterm preservation of a living 3D sandwich-like cell model. Senescence plays an important role in ageing and also in tumor formation, as not only is it one of the natural mechanisms that cancer cells must overcome to achieve tumor formation but it has also been shown to promote tumor formation in aged organisms. [52] Fibroblast senescence can be induced by replicative exhaustion (REP), that is, continued cell passage until a senescence-associated secretory phenotype (SASP) is observed. [53,54] This is characterized by a) expression of a proinflammatory cytokine profile; b) reduced cell proliferation; and c) morphological changes including flattening and spreading of cells. [53,54] With the aim of obtaining cell cultures with long-term stability, we proceeded to induce SASP in HDF via REP (the resulting cells being termed rHDF). Whilst both HDF and rHDF cells showed similar actin and DAPI staining profiles, cell shape and aspect ratio (AR), as well as the expression of proteins associated with SASP (namely IL-6 and p16INK4a), were altered in rHDF cells (Figures 4A and S9, Supporting Information). As expected, rHDF cell proliferation was considerably reduced, yet retaining their ability to effectively uptake both AuNS-4BPT and AuNR-2NAT ( Figure S10, Supporting Information).
In addition to the individual cellular morphology, many cell types, including fibroblasts, have been shown to display considerable macro-scale morphological plasticity, depending on the geometry of their encapsulating structures. [55,56] In this context, we studied the effect of culture well geometry on the overall 3D model obtained using rHDF cells. Our findings suggested that, square rather than circular wall geometry was essential for long term studies (>1 week); when rHDF cells were grown for 12 days in circular holders, total cell layer detachment occurred prematurely ( Figure S11, Supporting Information). Having defined the appropriate well shape we next studied the average height of one layer of rHDF cells. As the rHDF cell layers were added in sequential steps, this information gave us an estimate of the z-height we should expect to measure on the basis of the total number of cell layers added. rHDF cells, preincubated with either fluorescently labeled NPs or cell tracker www.afm-journal.de www.advancedsciencenews.com dyes, were sequentially added to a holder for fluorescence microscopy imaging. The use of unlabeled filler rHDF cells allowed us to achieve improved resolution, and hence measurements in the z-axis, of the different labels ( Figures 4B and S12, Supporting Information). We determined the average height of one layer of rHDF cells to be approximately 5 µm ( Figure S12, Supporting Information), which is in agreement with published studies. [55] However, with an increasing number of rHDF layers, the detected fluorescence signal decreased and a sufficient signal for quantification of fluorescence at the desired resolution could only be observed by increasing the laser power with increasing sample depth ( Figure 4C, Supporting Information). Although the need to increase laser power for imaging at deeper cell layers (i.e., those farther away from the objective) could be avoided by reducing the magnification and numerical aperture (NA) of the objective, this also translated into a significant reduction in lateral xy-resolution ( Figure S13, Supporting Information). Considering that high laser power can cause cytotoxicity and fluorophore photobleaching, methods working at low laser power are obviously preferred.

2D and 3D SERS Maps
Encouraged by the high levels of NP uptake into rHDF cells, the fact that electromagnetic simulations of local field enhancements of AuNRs and AuNSs ( Figure S14, Supporting Information) show enhancements as high as 10 6 , and their strong SERS signal in 2D cell studies ( Figure 3A), we next optimized the measurement conditions for 3D SERS. 3D models were built by separately incubating rHDF cells with either AuNS-4BPT@ PMA-PA or AuNR-2NAT@PMA-PA NPs, and subsequently depositing cell layers, with or without unlabeled "filler" rHDF cells to improve imaging resolution. Thanks to the flat spread nature of the cells, compact cell layers could be formed by the sequential addition of the NP-labeled rHDF cells.
Adv. Funct. Mater. 2020, 30,1909655  To check the capability of deep SERS imaging and the visualization of the layered texture of the cell culture structure, we first studied a sample where one AuNS-4BPT@PMA-PA cell layer was sandwiched between two AuNR-2NAT@PMA-PA cell layers (a "red-blue-red" scenario). SERS spectra along a 2 mm (x-axis) by 150 µm (z-axis) area were measured ( Figure S15A, Supporting Information), showing the localization of the 4BPT label (blue) sandwiched between the 2NAT labels (red) and thereby supporting the preservation of the layered structure. The observed discontinuities in the SERS spectra along the x-axis suggest a non-homogeneous distribution of the SERS labels within individual cells. An important issue in depth scanning is the ability to discriminate between signals originating from layers far-below and far-above the focal plane. To solve this problem we constructed a layered structure grown on a CaF 2 glass substrate, which gives a distinctive Raman reference signal at 322 cm −1 . The structure was simpler, consisting solely of 2 NP-labeled cell populations (AuNR-2NAT@PMA-PA on the bottom, and AuNS-4BPT@PMA-PA on the top), separated by 6 layers of unlabeled rHDF cells (resulting in a "red-blue" scenario, equivalent to approximately 40 µm of height, based on results shown in Figure S12, Supporting Information). Again, a wide area, 1.4 mm in x-axis and nearly 400 µm in the z-axis dimensions, was measured and a SERS map generated by integration of the signal of one vibration which does not interfere with any mode of the other two components; 1282 cm −1 , 1381 cm −1 , and 322 cm −1 for 4BPT, 2NAT and CaF 2 , respectively ( Figure 5, Figure S15B, Supporting Information). Whereas the two NP-containing layers are clearly recognizable, again there is some variation in their thickness and distribution which we attribute to variations in NP uptake. However, here we clearly see the signal attributed to CaF 2 (shown in green) which can be used to determine the baseline (for further discussion, see Section 4, Supporting Information). Layers and thicknesses can be also presented by the cross sectional analysis, i.e., by plotting the intensities of the selected mode of each component as a function of z-depth, in this case 4BPT (blue), 2NAT (red), and CaF 2 (green). Figure 5B-D (and Figures S15 and S16, Supporting Information) shows such profiles for three different positions taken from the xz-SERS map. The clear separation between the maxima in the cross-sections for 4BPT and 2NAT indicates the presence of the unlabeled filler cell layer in between NP-labeled cell layers. Measuring the distance between the cross section maxima, the thickness of the unlabeled layer can be roughly estimated between 26 µm (Figure 5B,C) and 34 µm ( Figure 5D).
Whilst in the intensity cross-section profiles the two distinct NP-containing layers can be seen, using the present imaging conditions it is clear that signal overlap leads to difficulties in achieving high resolution in z. In order to improve z-resolution we implemented two changes; we first switched to confocal SERS, using in this case a 13-layer sample comprising alternately loaded AuNR-2NAT@PMA-PA and AuNS-4BPT@ PMA-PA NP-labeled cells, spaced from each other by a layer of non-labeled cells (-red -blue -red -blue -red -blue -); and second, to determine the recurrence of the two SERS nanotags within the SERS maps, we used Pearson's correlation coefficients (PCCs, see Methods section for in-depth explanation) to estimate the linear correlation against the corresponding reference spectra. In case there was a perfect match between signal and reference spectrum, a value of 1 was assigned; when signals were absent, a PCC value of 0 was assigned.
The resulting multilayer sample was first observed at low lateral resolution, so as to evaluate the homogeneity of the distribution of both SERS nanotags, over an area of about 0.6 cm 2 ( Figure S17, Supporting Information). Surface mapping using 10x magnification and a 100 µm spaced grid confirmed the distribution of the SERS signals from both AuNS-4BPT@ PMA-PA and AuNR-2NAT@PMA-PA NP-labeled cells over the whole scanned area. The low resolution of this measurement in the z-axis ensured that the recorded data were obtained from the whole sample thickness. Next, the vertical stacking of the sandwich-like sample was investigated in a similar way to the data shown in Figures 5 and S15 (Supporting Information). A slice map (x 800 µm; z 100 µm) passing through the center (xy-plane) of the sample using 100x magnification and a smaller 10 µm grid spacing was measured ( Figure S18, Supporting Information). Improvements were then made by increasing the xzy-resolution from 10 µm to 5 µm, and incorporating a third dimension, thereby giving rise to a (75 × 140 × 150) µm 3 in x-, y-and z-axis, respectively (Figure S19, Supporting Information). Whereas a well-defined layered structure of approximately 10 4 µm 2 in xy area can be seen, only 4 of the 6 SERS nanotag-containing cell layers were identified. Furthermore, the 3D map becomes gradually less defined from the upper layer toward the bottom of the sample. One should recall that, during the measurements, the volume is scanned from top to bottom, as sketched in Figure 6A. Therefore, in addition to the variable SERS tags distribution discussed above, the reduction in reconstructed 3D image resolution can also be ascribed to the increasing path length that the radiation should travel through, to reach the bottom layer of the sampled volume. The 3D SERS map was therefore further analyzed to enhance the signal to noise ratio, specifically within the z-axis. The first data compression was performed by summing all the spectra within the same x-and z-coordinates. Each point in an xz-plane would then represent one spectrum (the sum of the spectra along the y-axis). Each spectrum was then reduced to just 2 values, i.e., the integrals for the 1383 cm −1 band for 2NAT and that at 1282 cm −1 for 4BPT ( Figure S20, Supporting Information). Each point in the xz-plane is therefore described by only these two band integrals. Figure 6B,C shows two vectors along the x-axis (-15 µm and -25 µm, respectively), in which the band integrals for the two SERS nanotags define a 6-layer structure along the z-axis. All the z-vectors along the x-axis (shown in Figure S20, Supporting Information) were deconvoluted in the same way and the peak positions, in z-coordinates, are summarized in Figure 6D, along the x-axis. The fitting clearly shows that, whereas not every result at each x-coordinate is well resolved, the 6-layer structure is seen in the overall representation and, when taking into consideration the "filler" non-labeled cell layer, an average thickness of roughly 5 µm per layer can be estimated ( Figure 6D). We are aware that the determination of the true layer thicknesses of deeply buried structures by depth profiling with confocal Raman microscopy using this method lacks high accuracy due to the blurring of the confocal intensity profile and its degradation with depth. [57] As this effect depends on various factors such as the real thickness of the layer, the Adv. Funct. Mater. 2020, 30,1909655  depth where the layer is located, the refractive index of the layer material (and that of upper layers), and the numerical aperture of the objective, a correction function should be determined and applied. [57,58] Because of imperfections and roughness of our biological cell layers, as well as of some degree of inhomogeneity in the SERS nanotag uptake, we refrained from doing such an analysis at the present stage.

Macro-and Micro-scale Characterization
Complementary imaging methods were used to corroborate the results obtained from SERS imaging of the 6-layer sandwich model ( Figure 6). The overall macro-scale morphological structure of the 3D system was first determined by means of fluorescence scanning microscopy and SEM. During the time period required for the whole study (approximately 12 days), the rHDF multilayered film did not detach, although we did observe a slight retraction of the edges (Figure 7A,B). The vast majority of the rHDF cells remained viable, as determined by live/dead staining using calcein (live, green) and Propidium Iodide (dead, red) fluorophores ( Figure 7A), and using SEM we could observe the sheet-like nature of the macrostructure ( Figure 7B). Upon further examination using both TEM and SEM, we indeed observed the multilayered SERS nanotag-containing rHDF cells ( Figure 7C; Figures S21 and S22, Supporting Information). Here the use of two identifiable NP shapes (rod vs. star) made the identification of each layer relatively simple, and furthermore allowed us to verify the average layer height. It should be noted that the height is smaller than that observed using SERS or fluorescence live imaging, presumably due to the severe sample preparation techniques used in electron microscopy which dehydrate the sample. Nevertheless, it should be stressed that the NP morphology (e.g., star tips) remained well preserved after such long cell culture periods, again verifying the suitable nature of these biocompatible NPs for cellular applications.

Conclusion
We have demonstrated imaging of a structurally complex 3D biological sample, composed of alternating layers of SERS tagloaded fibroblasts. We have thereby established optimal measurement conditions with high axial and lateral resolution. By taking advantage of the surface enhancing effect and the electron dense nature of AuNPs, we have shown how such SERS nanotags can be used in both SERS and TEM imaging, respectively, thereby allowing correlative imaging. We highlight the importance of using thorough and appropriate post-analysis, in this case by means of Pearson´s Correlation Coefficients or hypercubic data compression, to achieve quantitative data, which can be correlated with results obtained from TEM and fluorescence microscopy imaging. The results confirm the biocompatibility and high stability within biological environments Adv. Funct. Mater. 2020, 30, 1909655   Figure 6. A) Schematic view of the data analysis method for a 3D SERS map obtained from a 6-layer sandwich model comprising alternating rHDF cells that contain AuNS-4BPT and AuNR-2NAT NPs. Spectra with the same x-and z-coordinate were firstly summed and then reduced to just the two band integrals for 2NAT and 4BPT characteristic bands; B,C) Two examples of the band intervals of AuNS-4BPT (blue) and AuNR-2NAT (red) maps over the z-axis at x-axis coordinates of −15 µm B) and −25 µm C). D) Overall xz-axis mapping of AuNS-4BPT and AuNR-2NAT signals, corresponding to a 6-layer sandwich model. of the selected polymer-encapsulated SERS-tags, making them ideal for long-term labeling experiments. In this respect, the multistep synthesis of SERS tags, in particular offering flexibility in the choice of the Raman reporter molecule, the possibility of adding a fluorophore, and the tunable surface charge, without affecting their physicochemical properties, offers added value to these SERS tags toward multiplexing studies. It should be finally noted that the use of Au NPs (or other noble metals) also offers the possibility to apply additional analysis techniques, such as dark field microscopy or MALDI imaging. Aside from the presentation of SERS-tags and correlative imaging techniques, we propose that this work offers an important contribution toward the advancement of qualitative and quantitative 3D SERS-bioimaging for the characterization of biological tissues.

Experimental Section
NP Synthesis: AuNSs were prepared following a seed mediated growth method already reported. [25,44] The seed solution was prepared by adding AuNRs with LSPR maximum at 748 nm were also prepared. In brief, 1-2 nm gold seeds were grown in the presence of CTAB to form small anisotropic seeds of L 21 nm, W 8 nm. These provided the base for the synthesis of larger NRs with LSPR maximum at 748 nm. The exact protocol has been reported by González-Rubio et al. [45] Both AuNSs and AuNRs were labeled following a previously developed method, [25] including wrapping with an amphiphilic polymer to make them biocompatible, as described in Supporting Information (Section 2) and finally covered with PA to make then positively charged and enhance cell uptake.
NP Characterization: TEM images were collected with a JEOL JEM-1400PLUS transmission electron microscope operating at 120 kV, using carbon coated 400 square mesh copper grids. UV−vis optical extinction spectra were recorded using an Agilent 8453 UV−vis diode array spectrophotometer. The fluorescence of the NPs was measured with a Horiba Jobin Yvon Fluorolog fluorimeter upon excitation at λex = 530 nm. Z-Potential measurements were performed with a Z sizer (Nano ZS, Nanoseries; Malvern) instrument.
Cell Growth and Characterization: Adult Human Dermal Fibroblasts (HDF) were purchased from Invitrogen and grown in DMEM media supplemented with 10% FBS and 1% Penicillin-Streptomycin (PS). HDF cells were used fresh (up to 8 doublings) or purposefully passaged past the recommended doubling number via REP in order to work with cells expressing a SASP. [52,53] These cells were termed rHDF. To verify senescence status, immunofluorescence staining of fresh HDF and rHDF cells was used to study the presence of intracellular IL-6, IL-8 and p16INK4a. Non-labeled primary antibodies for IL-6 (AF206; 200 µg mL −1 ), IL-8 (MAB208; 500 µg mL −1 ) and p16INK4a (AF5779; 200 µg mL −1 ) were purchased from R&D, anti-Vinculin (SPM227; 2 µg mL −1 ) and anti-HSP-70 (5A5; 10 µg mL −1 ) and appropriate Alexa Fluor labeled secondary antibodies were purchased from Abcam. Actin-AF488 and DAPI were purchased from Invitrogen. All cell fluorescence microscopy experiments were conducted on a Zeiss 880 LSM confocal microscope. All samples used the same imaging conditions and post-processing including a 3-pixel median filter.
Holder Preparation: To obtain the 3D growth for SERS microscopy, a transparent polycarbonate (PC) square well was 3D-printed using an Ultimaker 2 extended + designed using blender 2.79 designing program ( Figure S23, Supporting Information) and was adhered to a quartz slide (EMS Labs) using biocompatible dentist glue (Picodent). For confocal Adv. Funct. Mater. 2020, 30,1909655  microscopy, cells were grown in a 96-well microplate with optical base (Ibidi). For TEM and SEM studies, an Au-Ti coated glass slide was used with a rectangular well stuck on-top with dentist glue. All noncommercial substrates were thoroughly cleaned and exposed to UV for 20 minutes to reduce the risk of microbial contamination.
NP Uptake, Colocalization, and Viability Analysis: Prior to any 3D model preparation, HDF cell viability and NP uptake was studied using brightfield, fluorescence, and electron microscopy techniques. To assay NP uptake with fluorescence microscopy, HDF cells were plated in optical bottom 96-well plates at 8 × 10 3 cells/well. After adhering overnight, the media was replaced with non-labeled, PMA-TAMRA or PMA-633 labeled AuNS-4BPT and AuNR-2NAT NPs, suspended in new media at a final concentration of [Au 0 ] = 0.05 or 0.1 × 10 −3 m. After 24 h, wells were washed with warmed media and the nucleus stained with DAPI. For colocalization studies, cells were also stained with Mitotracker Green and Lysotracker Red (both Invitrogen, 1/10,000). Cells were imaged with filters for DAPI (ex405), Mitotracker Green (ex488), PMA-TAMRA labeled NPs or Lysotracker Red (ex561), or PMA-633 labeled NPs (ex633).
Cell viability was analyzed using the Apoptosis/Necrosis detection kit (Abcam). HDF cells were seeded in 96-well microscopy plates (Ibidi) at 8 × 10 3 cells/well. Cells were allowed to adhere overnight followed by addition of NPs diluted in complete media to a final concentration of [Au 0 ] = 0.1 × 10 −3 m. After a further 24h, cells were washed and a solution of Apoptosis/Necrosis cell stains (all 1/100) was made in FBS-free DMEM. After 30 min, the cells were washed and complete media added. Images were taken with filters for CytoCalcein Violet 450 (live cells; ex405), Phosphatidylserine (PS) exposure (apoptotic cells; ex488), and 7-AAD (necrotic cells; ex561).
For TEM, cells were plated in a 12 well plate (1.5 × 10 5 cells/well) and the following day AuNS-4BPT@PMA-PA and AuNR-2NAT@PMA-PA were added at a final concentration of NPs at [Au°] = 0.05 × 10 −3 m and were incubated for 48h, followed by washing to remove non-uptaken NPs and then preparation of the NP-containing cells for TEM inspection. In brief, cells were trypsinized and washed in Sorensen's buffer, fixed with formaldehyde and glutaraldehyde, followed by embedding in agarose. Once solid, agarose embedded cells were stained with OsO 4 , dehydrated in an ethanol series, and embedded in Spurs resin. 60 nm cuts were made using an ultramicrotome and imaged with a JEOL JEM-1400PLUS TEM operating at 120 V.
Cell Division Measurements: ICP-MS analysis was used to study the doubling time of HDF and rHDF cells, and any subsequent dilution of the pre-incubated NPs in the cell culture. HDF and rHDF cells were serum starved in 0.1% FBS in DMEM for 24 h to induce cell division synchronization. Following this, cells were seeded in a 24-well plate (3 × 10 4 cells/well) and allowed to adhere. AuNS-4BPT@PMA-PA and AuNR-2NAT@PMA-PA NPs, diluted in cDMEM, were added to wells at a final concentration of [Au°] = 0.025 × 10 −3 m, 400 µL/well. The following day (D = 0) the wells were washed; one well of each NP type was trypsinized, the cell number counted, and the cell pellet frozen for ICP-MS measurements. On subsequent days 1, 4, 7, and 11, samples were again processed for ICP-MS (cells trypsinized, washed, cell number counted and pellet frozen).
3D Model Preparation and Imaging: rHDF cells were seeded in 24-well plates at a concentration of 4 × 10 4 cells/well. The following day, NSs or NRs were added to the cells by replacing the media with the NPs diluted in fresh media. For confocal studies, PMA-TAMRA-labeled AuNSs and AuNRs were used, whilst for SERS and TEM studies non-fluorescently labeled AuNSs and AuNRs were used. After overnight incubation, the excess of NPs was removed and the cells detached from the dish using trypsin. Cells were washed and resuspended with warmed fresh media. The cell number was counted. Considering that the footprint geometry of the well in all imaging modalities was the same, we seeded the same number of cells per cell layer: 2 × 10 4 cells. An initial layer of non-NPlabeled rHDF cells was added as a base layer, followed by consecutive layers of AuNS-and AuNR-labeled, and non-labeled rHDF cells. Layers were added every 1-2 days (allowing sufficient time for the cells to adhere) until at least 9 layers were achieved. To add each consecutive layer, the media lying on top of the previous layers was removed and new cells, either pre-incubated with the NPs or without, added. The final 3D model comprised alternating AuNS-and AuNR-labeled rHDF, with non-labeled filler cells between.
2D Model SERS Measurements: 2D SERS maps were recorded from live monolayer cell cultures in DMEM medium using a dip-in water immersion 63x (NA = 0.9) objective with integration time of 1 s and laser power of 10 mW (at surface) within an area of 40 µm x 40 µm. The step size was 1 µm in x-and y-direction.
3D Model SERS Measurements: The multi-layered 3D cell model was fixed using 4% formaldehyde and subsequently imaged in both wet and dry states. To avoid sample drying while measuring, the sample was placed in a plastic Petri dish, covered with parafilm leaving an aperture sufficiently large for the objective to come close to the sample. Raman maps were acquired using an inVia ReflexRaman microscope (Renishaw) exciting with a 785 nm diode laser at the nominal source power of 250 mW. 3D SERS maps were performed by acquiring several xz-planes along the sample y axis. 2D SERS profile maps (xz-plane) of 3D cell cultures in complete DMEM media with sequences 2NAT-4BPT-2NAT and 2NAT-unlabeled-4BPT were recorded using a dip-in water immersion 63x (NA = 0.9) objective with an integration time of 0.5 s (standard mode) and laser power of 5 mW (at surface) or with 1s (confocal mode) and 10 mW (at surface). The datasets were corrected by removing the cosmic rays and subtracting the baseline using WIRE 4 software. The corrected spectra were analyzed by component analysis (DCLS), build-in feature of WIRE 4 with reference spectra of 2NAT and 4BPT recorded from the pure 2D cell culture SERS measurements. For SERS mapping of the 6-layered cell culture Leica NPlan 10x (NA = 0.45) or NPlan 100x (NA = 0.9) objectives were used. Each spectrum was acquired for 1 s with a laser power of 3 mW (at surface) within a spectral range of 1200 to 1700 cm −1 . The spectral dataset was corrected removing the cosmic rays and the baseline using build-in functions in WiRE4 software and multivariate spectral analysis was performed in Matlab environment using Pearson's correlations with the SERS spectra of pristine nanostructures. Pearson's correlation coefficient is a convenient approach in big-data spectral analysis and it has been used before in targeted SERS labels based mapping analysis. [59,60] The computed coefficients should be interpreted as a measure of the linear correlation between a given spectrum (from the 3D maps, for instance) and the reference one (4BPT or 2NAT SERS reference spectra). The coefficients can range between 1 (perfect match) and −1 (perfect anti-correlation), even if, because of physical meanings, they usually space between 1 and 0, where 0 corresponds to any correlation against the reference (i.e., absence of the reference signal).
Confocal Fluorescence Imaging: The multi-layered 3D cell model was imaged without fixing using a Zeiss LSM 880 inverted confocal microscope equipped with a 561 nm laser for TAMRA excitation.
TEM/Resin-Embedded ESEM: The multi-layered 3D cell model grown on Au-Ti coated glass was fixed with Formaldehyde and Glutaraldehyde in Sorensen's buffer, followed by an ethanol dehydration series and Spurr's resin embedding. The multi-layered 3D cell model with its underlying Au-Ti coating was detached from the glass slide by immersing in liquid nitrogen. The sample was trimmed and 100 nm slices, cut along the x-z-axis using an ultramicrotome, were imaged using TEM. The remaining cut side of the sample was imaged using ESEM.
Surface ESEM: The multi-layered 3D cell model grown on Au-Ti coated glass was fixed with Formaldehyde and Glutaraldehyde in Sorensen's buffer, followed by an ethanol dehydration series and chemical drying using HMDS. The sample was imaged in an ESEM to see the surface topography of the whole multi-layered cell model.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author. www.afm-journal.de www.advancedsciencenews.com