Dark-Field Hyperspectral Microscopy for Carbon Nanotubes Bioimaging

: Carbon nanotubes have emerged as a versatile and ubiquitous nanomaterial, ﬁnding applications in industry and biomedicine. As a result, biosafety concerns that stimulated the research focused on evaluation of carbon nanotube toxicity. In addition, biomedical applications of carbon nanotubes require their imaging and identiﬁcation in biological specimens. Among other methods, dark-ﬁeld microscopy has become a potent tool to visualise and identify carbon nanotubes in cells, tissues, and organisms. Based on the Tyndall effect, dark-ﬁeld optical microscopy at higher magniﬁcation is capable of imaging nanoscale particles in live objects. If reinforced with spectral identiﬁcation, this technology can be utilised for chemical identiﬁcation and mapping of carbon nanotubes. In this article we overview the recent advances in dark-ﬁeld/hyperspectral microscopy for the bioimaging of carbon nanotubes.


Introduction
Since their first discovery in 1991 [1], carbon nanotubes (CNTs) have gained much attention from the various fields of science [2][3][4][5][6][7] and industry [8][9][10] as a material with prominent physicochemical characteristics. Carbon nanotubes are the allotropic form of carbon, having a cylindrical hollow structure up to several tens of micrometres in length and, generally, up to tens of nanometres in diameter [11]. CNTs are formed by rolling up a graphene sheet, another allotropic form of carbon, consisting of hexagonally bonded atoms arranged in a honeycomb pattern. Nanotubes could be rolled up from either a single graphene sheet and thus named single-walled carbon nanotubes (SWCNTs) or multiple graphene sheets and hence named multi-walled carbon nanotubes (MWCNTs) [12]. The unique morphology and size of CNTs provide excellent mechanical properties, large specific surface area, high electrical and thermal conductivity, flexible surface chemistry and valuable optical properties [13]. These features have prompted CNTs to be applied as the key component in various composites, sensors, coatings, and devices [11,14]. Moreover, hollow-tube-shaped structures coupled with the possibility of functionalisation make them highly engaging in biomedical applications [15]. For example, CNTs can be applied as radio-imaging probes in positron emission tomography (PET) and single-photon emission computed tomography (SPECT) [16][17][18], contrast agents in magnetic resonance imaging (MRI) [19] and Raman spectroscopy [20]. Strong optical resonances in the near-infrared region also allow the use CNTs in near-infrared and photoacoustic imaging techniques [21][22][23][24]. Carbon nanotubes could also be adapted to modulate neuronal growth as biocompatible scaffolds controlling cell adhesion, differentiation and migration [25][26][27]. The distinctive shape of the nanomaterial favours building the biocompatible drug-delivery system for cancer or gene therapy [9]. However, despite where CNTs could be effectively applied, their extensive toxicity and biodistribution analyses are required [15]. It is undoubted that physicochemical properties significantly determine the toxicity of nanomaterials. In terms of nanotubes, the important where λ is the wavelength of light illuminated from a specimen, NA is the numerical aperture of the objective and the condenser. Thus, the theoretical limit for microscopy with visible light illumination does not overcome 180-250 nm. Therefore, various strategies and tools are utilised to directly detect nanomaterials in biological samples, including photothermal microscopy [35], surface plasmon resonance microscopy, and other recently developed super-resolution techniques [36]. However, established methods, such as fluorescence or dark-field microscopy (DFM), remain applicable in nanotoxicological studies due to simplicity, low cost, and the possibility of observing hundreds of samples [37][38][39]. At the same time, there is no review of the published studies conducting dark-field bioimaging of carbon nanotubes. Therefore, the scope of this mini-review is limited by the keywords: optical dark-field microscopy, carbon nanotubes, and biological studies. In the first section, we briefly introduce the background and fundamentals of dark-field microscopy. Then, we overview in vitro and in vivo studies of CNTs where DFM had been applied. Finally, we summarise our conclusions and provide an outlook on future prospects.

Dark-Field Microscopy
Dark-field microscopy is a scattering-based imaging technique where elastically scattered light from a specimen entering the objective forms the image, while incident light is blocked ( Figure 1A). The fact that the light scattered from particles passes the objective leads to generating a clear and dark background with bright appearing objects, i.e., giving a higher signal-to-noise ratio (SNR) than in a conventional optical system [40]. In the case of nanosized objects, the scattering intensity may be approximated by the Rayleigh scattering theory, where the ratio of the refractive index of particles to the media is crucial to distinguish objects from the background [41]. Additionally, in noble metal nanoparticles, the prominent influence on the intensity provides a localised surface plasmon resonance (LSPR) effect, occurring when the electric field of incident light causes collective oscillations of electron density on the surface of particles [42]. Notably, the larger particles also scatter more intense light than the smaller ones [41,43]. Although carbon nanotubes could be directly observed through dark-field microscopy due to high aspect ratio [44] and Rayleigh scattering [45], deposition of nanoparticles with larger diameters or LSPR effect onto CNTs Appl. Sci. 2021, 11, 12132 3 of 17 is applied to visualise individual nanotubes ( Figure 2A) [46,47] and produce enhanced scattering agents for cell imaging [48]. Besides endogenous factors of materials, the exogenous parameters, such as the medium of the sample, method of illumination, angle of incidence, and numerical aperture of the objective, also influence the resulting image. Thus, the CytoViva TM enhanced dark-field illumination system (CytoViva, Auburn, Al, USA), widely used in biology and medicine, increases the SNR up to ten times by efficiently utilising these features [49,50]. In the enhanced dark-field microscopy (EDFM), the fibre optic illumination source is directly connected to the cardioid annular oil immersion dark-field condenser to reflect the aberration-free light at an oblique angle [51]. That type of condenser, in combination with high numerical aperture (NA) oil objectives, makes it possible to collect the scattered light of particles at wider angles, resolving nanoscale particles in biological samples [52,53].  Since the scattered spectrum of particles is highly affected by their physical, optical and electrical properties, dark-field microscopes may be equipped with hyperspectral sensors, thus obtaining spectral information with narrow bands at each pixel of an image ( Figure 1B) [54]. Typically, the hyperspectral sensors used in biological studies acquired data of elastic scattering in the visible near-infrared range (VNIR; 400-1000 nm wavelength range), although there are setups with sensing in the short-wave infrared range (SWIR; 900-1700) or Raman scattering [55]. The hypercubes are captured from samples as threedimensional data with two spatial dimensions (x, y) and one spectral (z). During the scan, the light from the specimen is almost entirely blocked, except the part that enters the narrow slit. Then, the collimated light is dispersed by a diffraction grating, and the detector captures its constitutive wavelengths. In the general scanning technique, called "push broom", one spatial and spectral dimension is captured at a time. Thus, the process is performed line by line for the whole image. The resolving power (R s ) of the sensor is mainly determined by the diffraction grating and, similarly to the spatial resolution, could be described by Rayleigh's criterion: where ∆λ is the wavelength interval between two spectral lines resolved by the grating. Although in realistic conditions, the slit width and sensitivity of the camera also significantly influence the resulting spectral resolution, which, however, can still reach up to several nanometres [56]. EDFM coupled with hyperspectral imaging (HSI) allows for robust discrimination of nanoparticles within samples with subsequent identification based on their spectral characteristics ( Figure 2C) [57]. For example, EDFM-HSI has been successfully used to measure the absorption capacity of CNTs ( Figure 2B,D) [58] and identify nanotubes within living organisms [59].   Figure 2A adapted from Ref. [46] under the terms of the Creative Commons Attribution License (CC-BY 4.0). Figure 2C adapted from Ref. [57] under the terms of the Creative Commons Attribution License (CC-BY 2.0). Figure 2B,D adapted from Ref. [58]. Copyright 2021 American Chemical Society.

In Vitro Studies of CNTs
Dark-field microscopy is a versatile and easy-to-use tool for label-free, non-invasive observation of nanomaterials in vitro using model cell cultures. At the moment, DFM has  Figure 2A adapted from Ref. [46] under the terms of the Creative Commons Attribution License (CC-BY 4.0). Figure 2C adapted from Ref. [57] under the terms of the Creative Commons Attribution License (CC-BY 2.0). Figure 2B,D adapted from Ref. [58]. Copyright 2021 American Chemical Society.

In Vitro Studies of CNTs
Dark-field microscopy is a versatile and easy-to-use tool for label-free, non-invasive observation of nanomaterials in vitro using model cell cultures. At the moment, DFM has been actively applied to visualise organic, inorganic and composite nanomaterials [60,61]. Carbon nanotubes possess a high aspect ratio and strong Rayleigh scattering, allowing for the detection of them in biological specimens. Table 1 summarises studies on mammalian cells treated with CNTs with the application of optical dark-field microscopy. Most of them involve the various CytoViva's enhanced dark-field microscopy setups since the unique illumination approach significantly increases signal-to-noise ratio resolving to observe even single nanotubes. For example, the synergistic effect of SWCNTs and etoposide on human pancreatic adenocarcinoma cells (PANC-1) viability were studied [62]. EDFM imaging demonstrated a time-dependent accumulation/distribution of SWCNTs in living cell samples, although it was hard to confirm CNT internalisation inside the cells. The same cell model was used in another study, where MWCNTs and polyethene glycol (PEG) composite was used for photothermal treatment of pancreatic cancer [63]. Using EDFM, dose-dependent internalisation of MWCNTs-PEG into PANC-1 cells after 1 h exposure was detected. An example of visualisation of single CNTs is reported in another toxicological study ( Figure 3A) [64]. The murine macrophage cells (RAW 264.7) exposed with MWCNTs for different time points (2-48 h) were directly observed through EDFM, confirming the time-dependent accumulation of nanotubes onto the surface as well as inside the cells. While examining the influence of the biological environment on the dispersion of CNTs, the EDFM technique was applied to confirm the bioavailability of nanotube dispersions to immortalised bronchial epithelium cells (BEAS-2B cell line) [65].
The complications arising from identifying cellular compartments affected by CNTs could be overcome using correlative dark-field imaging of nanoparticles and fluorescence microscopy. Notably, this approach has been applied in a study focused on the repair mechanism of lung epithelial tissue in the presence of pollutants [66]. The CNT-treated human lung adenocarcinoma cells (A549 cell line) were used as the epithelial tissue model with inhaled particles to observe their influence on recovery kinetics. CytoViva dual-mode fluorescence-enhanced dark-field microscopy setup was used to obtain 3D-rendered images of cells stained with 4 ,6-diamidino-2-phenylindole (DAPI) and rhodamine-phalloidin, which revealed strong attachment of CNTs to the cellular membrane and partial internalisation. The same technique was used in the study of proinflammatory and profibrotic activity of MWCNTs in A549 cells, macrophages (differentiated THP-1 cells) and fibroblasts (MRC-5; Figure 3C), that demonstrated comparatively higher uptake of CNTs in macrophages upon acute and prolonged exposure [67]. It is noted that increased uptake of nanotubes by macrophages could be due to their phagocytic activity and may cause fibrosis, which is also confirmed in the toxicogenomic analysis of macrophages treated with similar CNTs concentrations and exposure time [68]. Besides, a 3D rendering approach was applied to assess the internalisation of "long and thick" Mitsui-7 and "short and thin" Nanocyl-7000 MWCNTs on various cell types (bronchial and alveolar epithelial cells, mouse macrophages, and mesothelial cells) [69] as well as a complex organotypic model of alveolar tissue [70]. Nanocyl-7000 nanotubes appeared as densely packed aggregates in both cases, while Mitsui-7 were distinguishable as individual nanotubes. Nevertheless, the internalisation of both types of MWCNTs was observed in all cell models. Additionally, the cellular uptake of CNTs could be affected not only by dimensionality but also surface modification. Bai and colleagues investigated the influence of CNT functionalisation and protein corona formation on their biocompatibility [71]. It was discovered that carboxylation and base washing of MWCNTs increased cellular uptake. At the same time, coating nanotubes with bovine serum albumin (BSA) impacted cell absorption of only pristine nanotubes, which EDFM confirmed. Moreover, basic dye staining in a series of studies assessing the carcinogenic potential of pristine and surface functionalised CNTs on human small airway epithelial cells (SAECs) was applied to enhance cell contrast in EDFM [72][73][74]. Dark-field microscopy confirmed that carbon nanotubes either co-localised in the cytoplasm of cells or punctured the cellular or nuclear membranes. Another study based on fluorescence and dark-field imaging showed that ovalbumin (OVA) could be successfully delivered into macrophages in the complex with CNTs [75]. Siegrist et al. investigated the genotoxicity of raw, heat-treated (HT) and nitrogen-doped (ND) carbon nanotubes on BEAS-2B and SAECs [76]. Utilising fluorescence-enhanced dark-field imaging, they were able to quantitatively assess the nuclear uptake of the individual nanotubes, which revealed consistently higher partitioning of raw MWCNTs in nuclei of BEAS-2B cells ( Figure 3B). Additionally, simultaneous incubation of cells with high doses of CNTs and gold nanoparticles (AuNP) demonstrated aggregation of particles on cells which has not been observed in TEM images, pointing to the complementary role of the methods in distribution studies [77].
Apart from direct observations, distinctive spectral features of carbon nanotubes allow for mapping them in cell samples via hyperspectral imaging. One study showed that pre-treatment of SAECs with CNTs significantly enhances viral infectivity [78]. HSI mapping successfully detected SWCNTs and pandemic influenza A H1N1 virus (IAV) in cell samples as well as changes in their distribution patterns during co-exposure, suggesting the presence of nanotubes may influence IAV behaviour ( Figure 3D). In the comparative cytotoxicity analysis of clay and graphene-based nanomaterials, the hyperspectral imaging technique allowed the identification of nanoparticles in vital and paraformaldehyde (PFA) fixed A549 cells correctly [79]. It was found that graphene oxide nanosheets and MWC-NTs are penetrating cells less than nanoclays at the same concentration. HSI alongside fluorescence microscopy was also used to visualise the uptake and fate of SWCNTs and MWCNTs in a study of the fibrogenic effect of carbon nanotubes on human lung fibroblasts (CRL-1490) [80]. samples as well as changes in their distribution patterns during co-exposure, suggesting the presence of nanotubes may influence IAV behaviour ( Figure 3D). In the comparative cytotoxicity analysis of clay and graphene-based nanomaterials, the hyperspectral imaging technique allowed the identification of nanoparticles in vital and paraformaldehyde (PFA) fixed A549 cells correctly [79]. It was found that graphene oxide nanosheets and MWCNTs are penetrating cells less than nanoclays at the same concentration. HSI alongside fluorescence microscopy was also used to visualise the uptake and fate of SWCNTs and MWCNTs in a study of the fibrogenic effect of carbon nanotubes on human lung fibroblasts (CRL-1490) [80].  Figure 3A adapted from Ref. [64] under the terms of the Creative Commons Attribution License (CC-BY 3.0), scale not specified. Figure 3B adapted from Ref. [76] under the terms of the Creative Commons Attribution License (CC-BY 4.0). Figure 3C adapted from Ref. [67] under the terms of the Creative Commons Attribution License (CC-BY 4.0). Figure 3D adapted from Ref. [78] under the terms of the Creative Commons Attribution License (CC-BY 4.0), scale not specified. shows MWCNTs. (D) Hyperspectral mapping of SAECs co-treated with SWCNTs (blue pixels) and pandemic H1N1 influenza A virus (red pixels). Figure 3A adapted from Ref. [64] under the terms of the Creative Commons Attribution License (CC-BY 3.0), scale not specified. Figure 3B adapted from Ref. [76] under the terms of the Creative Commons Attribution License (CC-BY 4.0). Figure 3C adapted from Ref. [67] under the terms of the Creative Commons Attribution License (CC-BY 4.0). Figure 3D adapted from Ref. [78] under the terms of the Creative Commons Attribution License (CC-BY 4.0), scale not specified.   In addition to Rayleigh scattering, CNTs also exhibit Raman spectra that can be used to detect cellular uptake behaviour in combination with dark-field microscopy. The surfaceenhanced Raman scattering (SERS) coupled with EDFM was used when investigating the effect of size and aspect ratio parameters of hydroxylated nanotubes on biocompatibility, which attested the intracellular uptake only of "short and narrow" (SN) rather than "long and narrow" (LN) or "long and wide" (LW) CNTs [55]. Besides dark-field microscopy, intercellular carbon nanotube translocations could also be assessed using flow cytometry imaging of single cells on bright-field, dark-field, and fluorescent channels simultaneously [81]. Interestingly, the fluorescence signal of CNTs-loaded cells was less reliable than light absorbance and scattering data obtained from bright-field and dark-field images.

In Vivo Studies of CNTs
Dark-field microscopy is widely used to observe the distribution of CNTs in tissues and cells following in vivo exposure on animals. Table 2 summarises applications of DFM on animal tissue and cell extracts. Generally, for the microscopic examination of tissue samples, the examined organs are fixed with neutral buffered formalin (NBF), paraffinembedded, sectioned, stained with histological dyes after sacrification at specified postexposure time points. The staining of sections enhances the contrast between CNTs and the tissue, which is hard to distinguish due to low brightness. Notably, each study under investigation has applied CytoViva's enhanced dark-field imaging technique, increasing resolution compared to traditional DFM. For instance, Sager et al. studied double-walled carbon nanotubes (DWCNTs) pulmonary bioactivity in the C57BL/6 mouse model [82]. The particles appeared as bright tubular inclusions on the tissue processed from mice dosed with CNTs via pharyngeal aspiration 56 days post-exposure. The same features were also noted in tissue and cell samples of the other in vivo studies [83][84][85][86]. Another study investigating the effect of surface carboxylation on CNT bioactivity showed that clearance of functionalised nanotubes from lungs occurred better than that of unmodified [87].
Similarly, the influence of carboxylation degree on CNT deposition in lungs of albino BALB/c mice was assessed by EDFM supplemented with hyperspectral imaging [88,89]. However, the often-used Spectral Angle Mapper algorithm, detecting particles within biological specimens by comparing their spectra with the endmember spectrum, could not recognise CNTs in the samples due to the similarity of their spectrum to that of the background of stained tissue. Therefore, the Spectral Feature Fitting was applied, which compares sample spectra with the brightest spectrum of CNTs in tissues. The method showed that highly functionalised MWCNTs had a higher lung burden and were more dispersed ( Figure 4A). They also appeared to associate more with epithelial cells rather than alveolar macrophages, as occurs with pristine and less functionalised nanotubes. In order to avoid potential artefacts caused by instillation/agglomeration of CNTs because of a high single dose of particles, inhalation studies that closely mimic environmental conditions are performed. For example, Kim with colleagues have observed the toxicity of MWCNTs on Sprague Dawley (SD) rats after 5-day inhalation [90]. EDFM-HSI showed that nanotubes deposited in alveolar epithelium and macrophages persisted after 30 days post-exposure. The research group have pursued a study with extended exposure and postexposure times, finding that CNTs persist in the lungs after 90 days of the post-exposure period [91]. Another example of successful hyperspectral analysis is represented by the study of Smith et al., where it was applied to detect SWCNT uptake into circulating cells of severe combined immunodeficient (SCID) mice [92]. EDFM-HSI provided a map of the subcellular spatial distribution of nanomaterial, suggesting internalisation rather than binding to the membrane.  Microscopy confirmed the significant decrease in CNTs-laden macrophages by 28 th -day post-exposure; the clearance of CNTs in NADPH-oxidase-deficient mouse was less effective compared to the control group [98] Despite some difficulties raised in recognising the particles using HSI in histologically processed samples, enhanced dark-field microscopy itself provides the possibility to quantitively assess the distribution of CNTs in tissues by standard morphometric grid point counting methods. Mercer et al. have implemented this technique to measure the lung burden distribution of CNTs, determining that the majority of nanotubes (68%) are within or penetrating alveolar macrophages [93]. In a later study, the research group managed to analyse changes in the distribution of MWCNTs in the lungs over 336 days post-exposure and the clearance rate of aggregated and singular particles ( Figure 4B) [94]. The same approach with murine histology samples from extrapulmonary organs confirmed translocation and accumulation of CNTs in other tissues with time ( Figure 4C) [95,96]. In addition, it was found that total CNT content in the lungs of transgenic mice overexpressing SPLUNC1 (Scgb1a1-hSPLUNC1 TG), a protein involved in the innate immune system response in the respiratory tract region, was lower than in their wild-type (WT) littermates [97]. Moreover, the clearance rate of oxidised CNTs by macrophages in NADPH-oxidase-deficient mice measured using the point-counting technique was also decreased compared to the control group [98].
combined immunodeficient (SCID) mice [92]. EDFM-HSI provided a map of the subcellular spatial distribution of nanomaterial, suggesting internalisation rather than binding to the membrane.
Despite some difficulties raised in recognising the particles using HSI in histologically processed samples, enhanced dark-field microscopy itself provides the possibility to quantitively assess the distribution of CNTs in tissues by standard morphometric grid point counting methods. Mercer et al. have implemented this technique to measure the lung burden distribution of CNTs, determining that the majority of nanotubes (68%) are within or penetrating alveolar macrophages [93]. In a later study, the research group managed to analyse changes in the distribution of MWCNTs in the lungs over 336 days postexposure and the clearance rate of aggregated and singular particles ( Figure 4B) [94]. The same approach with murine histology samples from extrapulmonary organs confirmed translocation and accumulation of CNTs in other tissues with time ( Figure 4C) [95,96]. In addition, it was found that total CNT content in the lungs of transgenic mice overexpressing SPLUNC1 (Scgb1a1-hSPLUNC1 TG), a protein involved in the innate immune system response in the respiratory tract region, was lower than in their wild-type (WT) littermates [97]. Moreover, the clearance rate of oxidised CNTs by macrophages in NADPH-oxidasedeficient mice measured using the point-counting technique was also decreased compared to the control group [98]. Smaller arrows indicate some of the numerous singlets and small MWCNT structures distributed throughout the alveolar septa of this critical transition region between conducting airways and gas exchange regions of the lungs. (C) Enhanced dark-field image of MWCNT fibres in the brain of C57BL/6J mouse at 336 days after inhalation exposure. MWCNT fibre is bright white, cell nuclei are brownish red, and other tissue elements are green. Figure 4A adapted from Ref. [89] under the terms of the Creative Commons Attribution License (CC-BY 4.0), scale not specified. Figure 4B adapted from Ref. [94] under the terms of the Creative Commons Attribution License (CC-BY 2.0). Figure 4C adapted from Ref. [96] under the terms of the Creative Commons Attribution License (CC-BY 2.0).
Overview of in vivo studies of carbon nanotubes applied dark-field imaging technique. L-length; D-diameter; th. Microscopy revealed that CNTs generally observed within the interstitial tissue but also in condensed areas [82]  (B) Enhanced dark-field image of the transition region between a terminal bronchiole and first alveolar duct bifurcation of C57BL/6J mouse at 168 days after exposure to MWCNT. The large arrow indicates a cluster of MWCNTs (white fibres) in the ridge of the first alveolar duct bifurcation. Smaller arrows indicate some of the numerous singlets and small MWCNT structures distributed throughout the alveolar septa of this critical transition region between conducting airways and gas exchange regions of the lungs. (C) Enhanced dark-field image of MWCNT fibres in the brain of C57BL/6J mouse at 336 days after inhalation exposure. MWCNT fibre is bright white, cell nuclei are brownish red, and other tissue elements are green. Figure 4A adapted from Ref. [89] under the terms of the Creative Commons Attribution License (CC-BY 4.0), scale not specified. Figure 4B adapted from Ref. [94] under the terms of the Creative Commons Attribution License (CC-BY 2.0). Figure 4C adapted from Ref. [96] under the terms of the Creative Commons Attribution License (CC-BY 2.0).

Conclusions and Future Prospects
Although dark-field imaging, as presented in the literature, allows direct observations of carbon nanotubes, the technique per se is insufficient for cell-nanoparticle identification. Modifications of DFM, such as enhanced dark-field optics coupled with fluorescence modules and hyperspectral sensors, have significantly increased the applicability of the method for cell biology, permitting spectral identification of particles, quantitative and qualitative analysis of cellular and nuclear uptake, and assessment of drug delivery efficiency. Comprehensive in vivo distribution studies of carbon nanotubes have demonstrated benefits from the EDFM-his technique over conventional methods due to fast sampling and robust detection in tissue sections. However, the method is not without the limitations arising when accurate subcellular imaging is needed. The hyperspectral analysis is also complicated in the case of histological staining, leading to the possibility of a false-positive outcome. At the same time, the increase in the amount of heterogenous spatio-spectral information obtained using correlation microscopy, as well as its still-developing stage, leads to another challenge of efficient data processing. Recent studies have shown the power of machine learning algorithms, including neural networks in environmental studies and bacterial identification. This approach automates spectral feature selection, excluding the necessity for data pre-processing, and may help improve the performance of EDFM-HSI for nanomaterial systematic investigation in the future.