1.1.1.

Chemical staining

Chemical dyes are used to react with specific biological target and stain them, such as using Giemsa staining to observe chromosomes under an optical microscope. The staining method is easy to operate, and the result can be obtained by direct observation of the microscope. However, the resolution will be limited by the light diffraction. According to the Abbe diffraction limit formula, the resolution of the optical microscope is about half of the wavelength. Considering that the range of visible light is from 380 to 780 nm, the resolution limit of the visible light optical microscope is about 200 nm, which is difficult to observe biological structures at the nanoscale.

1.1.2.

Labeling method

In the labeling method, label molecules (such as fluorescence and isotopes) are often used to participate and track changes in physiological reactions. The use of labeling methods enables high-sensitivity molecular detection, but the addition of markers may change the original molecular structure, composition, or biological environment. Therefore, it cannot truthfully reflect the true state of the biomedical samples in the original environment and the limited luminescence time of marker limits the observation time. In addition, the lateral resolution of labeling method is also limited by the diffraction limit and cannot achieve nanoscale resolution. Although specially designed fluorescent microscopes, such as structured-light illuminated microscopy,² can achieve nanoscale spatial resolution, its axial resolution is still limited.

1.1.3.

Vibrational Spectroscopy

Vibration spectroscopy, including infrared spectroscopy and Raman spectroscopy, probes the specific molecular vibrations and rotations. It has the advantage of high sensitivity to the change of molecular conformation and low damage to biomedical samples. Mid-infrared and far-infrared terahertz (THz) are the most commonly used spectral ranges in infrared vibrational spectroscopy. The mid-infrared wave is an electromagnetic radiation with a wavelength range from 2.5 to $25\mu \mathrm{m}$, which corresponds to the fundamental frequency absorption band of most organic substances, and can differentiate biomedical samples by identifying specific functional groups. The THz wave, spanning wavelengths from 30 to $3000\mu \mathrm{m}$, corresponds to the rotational and vibrational energy level transitions of molecules, and can be applied to chemical identification and characterization of many substances. However, the resolution of these methods is also limited by the light diffraction, which only reach the micrometer level. In addition, different molecules may have the same bonds, and so it is difficult to distinguish them. Therefore, these methods are difficult to analyze more microscopic and complex biomedical samples.

1.2.1.

Electron microscopy

Advanced methods based on cryoelectron microscopy, such as single particle analysis and microcrystal electron diffraction, have made important contributions in the study of biological structures. Using these techniques, the atomic-level structure information of biomolecules can be obtained, which is helpful for understanding the structural change and regulation mechanism of macromolecules.

1.2.2.

Scanning probe microscopy

Scanning probe microscopy (SPM), including atomic force microscopy (AFM), is a general term for a series of microscopies developed on the basis of scanning tunneling microscopy, with nanoscale resolution. AFM is most commonly performed in biological detection due to its abilities of analyzing the morphological and mechanical properties of biomolecules, such as detecting the surface stiffness of cancer cells³ and observing protein folding.⁴ In addition, specific recognition can be achieved by chemical modification of AFM probes. However, not every type of biomedical substance can find the suitable corresponding reactant. So the application scope of this specific detection is limited.

The combination of SPM and vibrational spectroscopy, namely, scanning near-field optical microscopy (SNOM), can realize specific biological detection with a wider range of adaptability while maintaining nanoscale spatial resolution. The common vibrational spectroscopies include Raman spectroscopy and infrared spectroscopy. The development of tip-enhanced Raman spectroscopy, which combines AFM with Raman spectroscopy, is hindered by the low spontaneous Raman scattering cross sections of biomolecules. The techniques combining AFM and infrared spectroscopy, namely, infrared near-field techniques, can be roughly divided into two categories according to the different detection signals: one based on optical signals and the other based on mechanical signals. The scattering-type SNOM (s-SNOM) belongs to the first category, which detects the signal of the scattered light of organic samples after infrared irradiation. The atomic force microscope-based infrared (AFM-IR) spectroscopy, peak force infrared (PFIR) microscopy, and photo-induced force microscopy (PiFM) belong to the second category. AFM-IR is a mainly photothermal technique with near-field tip enhancement. It detects the probe oscillation signal caused by thermal expansion of biomedical samples due to infrared light absorption. PFIR is an AFM-IR method. It operates in peak force tapping mode, while the traditional AFM-IR operates in contact mode. The change of AFM working mode reduces the lateral destruction to the sample by the tip, making it more suitable for rough and sticky surfaces. In contrast, PiFM works in tapping mode and detects the change of near-field electromagnetic force. Since these methods use the tips of AFM as near-field probes, the spatial resolution obtained is comparable to the size of the tip, usually in the tens of nanometers. Noteworthy, the PFIR has a higher resolution because the contact area between the tip and the sample is smaller than the radius of the tip in tapping mode. These infrared near-field techniques all have the capacity of nondestructive recognition without labels.

The full-text logical framework is shown in Fig. 1. This review starts with the complexity of the environment, in which biomedical samples are located (from simple to complex including dry environment, liquid environment, and orthotopic environment) and the microscopic biomolecules with different characteristics (including biomacromolecules, cells, and tissues) are dicussed separately. The technical characteristics, unique advantages, and limitations of several near-field techniques (including s-SNOM, AFM-IR, PFIR microscopy, and PiFM) in biomedical detection of microscopic biomolecules are described in detail. Finally, their challenges and potential possibilities in biomedical detection are discussed. This review will pave the way for future studies into microscopic biological characteristics.

Fig. 1

The framework of this review. This review focuses on the recent biomedical research progress of several near-field techniques (including s-SNOM, AFM-IR, PFIR, and PiFM) from the perspective of different environments (including nonliquid environments and liquid environments) and different biological samples. The possibility of in vivo detection is also discussed. Adapted with permission from Refs. 56.7.–8. Adapted with permission from Ref. 9. © 2018 American Chemical Society. All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license.

2.1.1.

Treatment methods for biomacromolecules samples

When the analyte is a biomacromolecule, such as protein, it is important to purify the sample to minimize the interference of other components to the detection results. High-purity samples can be obtained by conventional protein purification methods. A sample with higher surface flatness will be prepared by spin-coating the liquid containing purified protein onto the substrate. High flatness of samples makes infrared near-field spectra and images have a better quality. When the analyte is a macromolecular complex, such as cell membrane and chromatin, it is necessary to enrich the analyte as much as possible by centrifugation. Casting, rather than spin-coating, is more often applied to these samples. One possible reason is that the size of such samples is larger than that of proteins, which is easier to be observed, so the use of low-concentration solution casting can avoid overlapping between samples.

2.1.2.

Test results and analysis

The basic composition of protein is amino acid. The peptide chains synthesized by the dehydration reaction of amino acids are spatially folded, and the proteins with specific functions are formed. The diversity of peptide chain spatial structure makes proteins have wide and various functions. It is worth noting that the secondary structure of the protein describes the spatial orientation of the main body of the peptide chain and can reflect the main structure of the protein. Therefore, the purpose of studying protein function can be achieved by studying the related changes of protein secondary structure. Infrared vibration spectroscopy is often used to nondestructively characterize the secondary structure of peptides and proteins.¹² Similarly, in infrared near-field detection, proteins are often studied from three aspects: the classes of secondary structure of proteins, factors causing secondary structure changes, and morphological changes caused by changes in secondary structure. Compared with the conventional infrared vibration method, infrared near-field detection techniques can detect proteins in the nanoscale range due to the use of a nanoscale tip, that is, it can realize the signal detection of individual proteins, rather than the average signal detection of a large area of proteins. For the complexes, such as cell membrane, the properties of the membrane are often speculated by analyzing the secondary structure of the proteins on the membrane. For chromatin, infrared near-field detection can obtain the characteristic signal of methylated groups on chromatin, so as to identify the local methylation of chromatin. This is not possible with conventional infrared vibrational spectroscopy methods.

Classes of secondary structure of proteins

First, the studies about secondary structure of protein are introduced. The secondary structure of protein can be obtained by infrared near-field detection techniques. Proteins with clear secondary structure, such as ferritin complex, lysozyme, $\gamma$-globulin, and catalase, are often selected as the research objects to verify the reliability of near-field systems for detecting proteins. Amenabar et al.¹³ proved that s-SNOM could identify the secondary structures of the individual protein complexes. Mueller et al.¹⁴ obtained the spectrum of lysozyme through AFM-IR. Bechtel et al.¹⁵ combined synchrotron infrared radiation with s-SNOM and obtained the subzeptomole sensitivity spectra of $\gamma$-gobulin. O’Callahan et al.¹⁶ used s-SNOM to study catalase (a protein widely existing in organisms) and obtained its spectra of secondary structure at nanoscale. Their results showed that there were mixed responses of multiple vibration bands. It should be noted that during the experiment, the drying process of drop cast may lead to crystal disorder and disordered structure. The infrared near-field techniques can not only be used to recognize individual protein molecules but also be used to distinguish different proteins. Fused in sarcoma (FUS), an RNA binding protein that is helpful for phase separation, has also attracted attention because its methylation degree affects the process of phase separation. Qamar et al.¹⁷ conducted an AFM morphology test and an infrared spectral test on three kinds of FUS by AFM-IR and qualitatively analyzed their secondary structures during phase separation. In the second derivative spectra, as shown in Fig. 2(a), compared with asymmetric demethylated-FUS, hypomethylated-FUS has a higher content of antiparallel $\beta$-sheet, random coil and $\beta$-turn, a lower content of $\alpha$-helical and $\beta$-sheet, and a lower degree of methylation. Compared with the first two FUSs, the specially prepared FUS + 16R had lower content of antiparallel $\beta$-sheet and higher content of parallel $\beta$-sheet. This meant that the secondary structure and methylation of FUS protein could be characterized. Ji et al.²¹ used PiFM to characterize the single protein particles of three biotoxin with molecular weights as low as 30 kDa and established a database model based on their photoinduced force spectra. Waeytens et al.²² proposed a method to predict the secondary structure of protein from AFM-IR spectra. Kallas et al.²³ studied the protein coverage of biomaterials in different bodily fluids by PiFM. Not only natural proteins but also synthetic proteins have been analyzed. Rosu et al.²⁴ characterized composite particles containing hydrophobic polypeptides and explored the protein distribution based on the PiFM measurements. Cristie-David et al.¹⁸ used PiFM to acquire the chemical composition of the self-assembling protein cage, as shown in Fig. 2(b). The obtained infrared spectra showed amide I band and amide II band, which represented the characteristics of protein, indicating that the obtained particles were indeed composed of proteins. Abrego-Martinez et al.²⁵ proved the uniform distribution of S-proteins on aptamer-based biosensors by PiFM.

Fig. 2

The application of near-field detection of biological macromolecular samples. (a) The average IR spectra and the second derivative spectra obtained from three kinds of FUS. Adapted with permission from Ref. 17. (b) PiFM images of particles composed of proteins at different frequencies. Adapted with permission from Ref. 18 © 2019 American Chemical Society. (c) Spectra comparison between AFM-IR and FTIR with two proteins. Adapted with permission from Ref. 19. (d) AFM-IR ratio map between two integrated absorption bands and the results of immuno-staining. Adapted with permission from Ref. 5. (e) s-SNOM results of influenza virus. Adapted with permission from Ref. 20.

The secondary structures of proteins fibrils at different aggregation stages have also been studied. Ruggeri et al.²⁶ observed the formation process of amyloid fibrils with AFM-IR and proved that their nanomechanical properties were related to their secondary structure, especially the content of $\beta$-sheet. Adamcik et al.²⁷ revealed the chemical structure transition from amyloid fibrils to amyloid microcrystals using AFM-IR. Their results showed that from fibrils to microcrystals, the content of protein secondary structure $\beta$-sheet increased, and the content of random coil decreased. Banerjee et al.²⁸ proved for the first time that the structure of early stage amyloid aggregates was heterogeneous.²⁸ Wang et al.²⁹ identified FapC fibrils with different formation stages from PFIR results. Banerjee et al.³⁰ found that tau fibrils with different aggregation times had similar morphology but had significant differences in secondary structure. In the early stage of fibrils, there were more parallel $\beta$-sheet structures, while in mature fibrils, there were more antiparallel $\beta$-sheet structures. Factors affecting fibril aggregation have also been investigated. Henry et al.³¹ detected the interaction between $\mathrm{A}\beta 42$ and different lipid membrane models by AFM-IR and explored the effect of cholesterol on the formation of amyloid fibrils. Since the formation of amyloid plaques was related to the electrostatic interactions between amyloid fibers, and the electrical characteristics of biological samples could be obtained through Kelvin probe force microscopy (KPFM), Jakob et al.³² revealed the correlations between residual charges and secondary conformation in amyloid fibers by PFIR-KPFM. Their results showed that there were more positive charges in the regions that were rich in $\beta$-sheets and random coil structures.

To further improve the detection sensitivity, nanostructures can be used. O’Callahan et al.³³ used gold nanorods as half-wavelength antennas in 2019 and combined with s-SNOM to realize nanoimaging of horse ferritin. The secondary structure of ferritin is mainly $\alpha$-helical. After using gold nanorods, there was plasma resonance near $1650{\mathrm{cm}}^{-1}$, which was consistent with the resonance frequency of the secondary structure of protein. Compared with the detection limit of $6\times {10}^7\mathrm{kDa}$ protein obtained by conventional surface-enhanced infrared absorption (SEIRA) detection using gold nanorods, SEIRA combined with s-SNOM could detect at least seven protein complexes, that was, the total protein quality of $3\times {10}^3\mathrm{kDa}$ and enhanced sensitivity $\sim \mathrm{10,000}$ times.³⁴ In addition, the normalized signal-to-noise ratio obtained by this method was $7.6\sqrt{\mathrm{Hz}}/{\mathrm{cm}}^{-1}$. Without gold nanorods, the normalized signal-to-noise ratio was $0.63\sqrt{\mathrm{Hz}}/{\mathrm{cm}}^{-1}$. Because the nanoantenna could enhance the resonance signal, the sensitivity was an order of magnitude higher than that of the simple s-SNOM. An optimized system can also be employed to enhance signal sensitivity. When using AFM-IR to study the structure of amyloid fibrils, Waeytens et al.³⁵ suggested equipping a bottom-up lighting system with s-polarization and silicon tip. Ruggeri et al.¹⁹ obtained the secondary structure information of single thyroglobulin protein and apoferritin protein using off-resonance AFM-IR, as shown in Fig. 2(c), with high sensitivity and signal-to-noise ratio ($\sim 10$ to 20). These applications illustrate the ability of using infrared near-field techniques to recognize the secondary structure of proteins. This will offer new approaches for identifying proteins with no significant morphological differences but different secondary structures.

2.2.1.

Treatment methods for cell samples

In general, the physical size of cells is much larger than that of proteins, for example, the diameter of human red blood cell is about $10\mu \mathrm{m}$, whereas the diameter of IgG protein is 10 nm. The cells embedded with embedding medium such as resin can retain their microstructure. The embedded samples are often sliced for ensuring the surface flatness, so that the morphology, structure, and composition of each organelle can be detected more accurately. In addition to the resin embedding method, the method of directly drying cells is also often used. Considering that cells contain a large amount of free water, the loss of free water during the drying process will lead to significant changes in the morphology of the cells, resulting in a different distribution of organelles inside the cells from that before drying. Test results obtained for these samples are more suitable for analyzing whether proteins or lipids are produced in cells, and how they are distributed, rather than for analyzing the function of a specific organelle.

2.2.2.

Test results and analysis

The infrared absorption peaks of lipid, nucleic acids, and proteins, which most cell samples contain, can be obtained by infrared near-field techniques. The components information of a cell can be identified by analyzing the nanospectra obtained at the specific location of the cells. The exogenous compound in the nucleus was detected for the first time in 2011.⁵⁷ And the spatial distribution of various substances can be characterized according to the nanoabsorption maps at specific infrared frequencies. Freitas et al.⁵⁸ studied the nanoscale structure of $\mathrm{A}\beta 42$ proteins on the surface of single primary neurons, as shown in Fig. 3(a), and demonstrated the feasibility of s-SNOM for studying amyloid structures on cell surfaces. The composition and distribution of proteins in the cells will change when cells are exposed to external stimuli, including the use of inducers or nanomaterials, exposing them to oxidative stress conditions and parasitic infections. These differences can be detected by infrared near-field detection. For example, macrophages would exhibit different polarization properties and had different protein conformations under the action of two inducers;⁶² nanomaterials such as nanodiamond could induce the secondary structure of protein in hepatoma to change from $\alpha$-helical to $\beta$-sheet or $\beta$-turn;³⁶ after oxidative stress, the probability of protein aggregation size $<0.6\mu \mathrm{m}$ was 35% higher than that in HeLa cells;⁶ and red blood cells infected with Plasmodium falciparum exhibited aggregation of hemozoin.⁶³ In addition to directly analyzing the near-field images, the near-field spectral data can also be used to reconstruct the refractive index of cell for disease detection. Tranca et al.⁶⁴ gained the complex refractive index of a single red blood cell by s-SNOM data. Microvesicles secreted by cells have also been studied. Kim et al.⁶⁵ used resonance enhanced AFM-IR to analyze the heterogeneity of a single extracellular vesicle and between the same population of extracellular vesicles. Xue et al.⁶⁶ detected the secondary structures of proteins in disease-related small extracellular vesicles by s-SNOM in 2022. They explored the relationship between protein secondary structure changes and tumor malignancy and successfully distinguished two small extracellular vesicles from different stages of disease development. When stimulated by external factors, the microbubbles secreted by cells are also different. For example, after oxidative stress, there was heterogeneity between the vesicles from two types of mesenchymal stem cells, which had different resistance to oxidative stress.⁶⁷ At the same time, there was distribution heterogeneity of components such as proteins, nucleic acids, and lipids, between different positions of a single vesicle and between different vesicles of the same type, according to the analysis of infrared near-field images and nanospectra.

Fig. 3

The application of near-field detection of cell samples. (a) AFM-IR maps of single primary neurons. Adapted with permission from Ref. 58. (b) s-SNOM results of the genetically modified E. coli. (c) The phase spectra obtained on two spots in (b). (b), (c) Adapted with permission from Ref. 59. (d) The phase images of Chlamydomonas reinhardtii cell obtained by the s-SNOM. (e) The reconstruction of cellular spectroscopic tomography from (d). (d), (e) Adapted with permission from Ref. 60. (f) PFIR images of zymosan particles. Adapted with permission from Ref. 61 © 2020 American Chemical Society.

In addition to cells, bacteria or fungi, which have cellular structures, are often mentioned, and these studies focus on proteins within bacteria. Dazzi et al.^68,69 obtained the absorption spectra of proteins and DNA of Escherichia coli (E. coli) and realized the localization of phages inside E. coli, proving the possibility of AFM-IR in single cell detection. Similarly, Meszaros et al.⁵⁹ used s-SNOM to detect transgenic E. coli and test whether the protein of interest was produced. As shown in Fig. 3(b), E. coli has an absorption signal in the amide I region ($1660{\mathrm{cm}}^{-1}$). In order to verify that E. coli only produced the desired protein internally, nanoabsorption spectra were collected at both its internal and external positions. The results are shown in Fig. 3(c). The spectrum inside the E. coli shows the absorption peak of the protein, whereas the spectrum outside the E. coli does not show an absorption peak, which indicates that the proteins exist only inside E. coli. Otzen et al.⁷⁰ identified amyloid in E. coli, Pseudomonas, and Archaea Methanosaeta in situ by the AFM-IR absorption peaks corresponding to the $\beta$-sheet structure of amyloid. To provide further insight into the spatial distribution of proteins or lipid, relevant works have been carried out. The distribution of silaffin protein and silica in genetically modified E. coli was detected by PiFM.⁷¹ Kanevche et al.⁶⁰ performed layer-by-layer analysis of sliced Chlamydomonas reinhardtii and E. coli by s-SNOM. For the phase images of the same slice of Chlamydomonas reinhardtii at different frequencies, the nucleolus and thylakoid could be clearly distinguished, and the cell wall could be obvious displayed. The spatial distribution of proteins in cells could be obtained by reconstructing the s-SNOM images layer by layer, as shown in Figs. 3(d) and 3(e). Deniset-Besseau et al.⁷² realized the localization of lipid inclusions inside bacteria by AFM-IR. Research related to vesicles in bacteria has also been conducted. Vitry et al.⁷³ used AFM-IR to detect vesicles inside Streptomyces bacteria. As a reliable source of biofuel, Streptomyces can store excess carbon in triacylglycerols, so a reliable method is needed to detect the density of triglycerides in Streptomyces. The approximate location of triglycerides could be determined from the nanoabsorption map of AFM-IR. They preliminarily determined the location of the vesicles using the specific infrared absorption of triglycerides at $1740{\mathrm{cm}}^{-1}$. However, since the optical absorption signal was also related to the height of the vesicles, the signal generated by the vesicles located just below the cell membrane would be blurred. Therefore, they also used mode synthesizing AFM and combined with the detection results of the two methods for judging the size, position, volume, and other parameters of the vesicles. In addition, Mayet et al.^74,75 characterized the distribution of polyhydroxybutyrate (PHB) in a single bacterium Rhodobacter capsulatus by AFM-IR. Rebois et al.⁷⁶ demonstrated that the PHB in the Rhodobacter sphaeroides vesicles could be detected using AFM-IR, and the crystallinity of PHB could be estimated by deconvolution of the spectrum. The cell walls of fungi were also studied. Bakir et al.⁷⁷ identified the nanoscale structure and chemical components of the cell wall of the gene deletion strains by s-SNOM. Li et al.⁶¹ used PFIR with a resolution of 6 nm to reveal the chemical composition and distribution of yeast cell wall particles. Figure 3(f) shows the high-resolution PFIR images at 1030 and $1630{\mathrm{cm}}^{-1}$. They observed that there were more $\beta$-glucans on the cell wall and more chitin at the edges of the bud scars.

The specific infrared absorption peaks of proteins, DNA, and lipids inside the cell can be obtained by infrared near-field detection. The obtained nanochemical maps can be used to characterize the distribution of various components inside the cells, including protein aggregation. Therefore, infrared near-field techniques can directly detect the components of cells, analyze the biochemical behavior inside cells, and evaluate whether individual cells are abnormal. Similarly, infrared near-field detection of vesicles used for cell-to-cell communication can obtain information about their internal components and be applied to early diagnosis of diseases.

2.3.1.

Treatment methods for tissue

In general, biology and pathology use sections to study tissue sample. Similarly, when performing infrared near-field detection, cryosection is the most common sample processing method. In addition to cryosection, electron microscopy sample preparation methods, such as epoxy-resin-embedded sectioning, are also commonly adopted. The purposes of slicing include: (a) maintaining the original tissue state as much as possible and making the distribution information of different components more intuitively, and (b) providing the nanoscale surface flatness and minimizing the detection error caused by height.

2.3.2.

Test results and analysis

Through the detection of tissue samples, it is possible to identify specific components in the tissue, study the changes in the internal components of the tissue, and distinguish different types of cells. In pathological examinations, sections are often stained to identify a specific substance from a tissue sample. In infrared near-field detection, the composition can be characterized by identifying the infrared characteristic signals of specific substances. At the same time, it can also provide the distribution of specific substances in space at the nanoscale and realize the qualitative analysis and spatial positioning of substances. Johnson et al.⁷⁸ used rotating frame s-SNOM to detect ultrathin proteins in oyster shells in 2019. The oyster shells are composed of ${\mathrm{CaCO}}_3$ crystals and contain a small amount of protein distributed in bulk. Figure 4(a) shows the s-SNOM phase images of rotating frame in the prismatic region containing the organic sheath, which are used to reconstruct the maps of protein content. Because the content of organic components in the shell was $<5\%$, the signal acquisition time of s-SNOM based on rotating frame was 60 times faster than that of traditional s-SNOM. It is worth noting that the near-field phase diagram of the rotating frame s-SNOM has a blueshift of $20{\mathrm{cm}}^{-1}$ compared with the traditional s-SNOM, as shown in Fig. 4(b). So spectral accuracy still needs to be improved. Related research based on melanin and lipid in hair has been proposed. Amenabar et al.⁷ studied human hair with s-SNOM, realizing the labeling of melanin molecules with a spatial resolution of 30 nm. From the cross section of human hair, the hair is composed of epidermal layer, cortical layer, and medulla from outside to inside. The epidermal layer is about $0.5\mu \mathrm{m}$, occupying most of the hair space. The secondary structures of keratins in the cortical layer are mainly $\alpha$-helical. There is also melanin in the cortical layer. Melanin can determine the color of human hair. Through the analysis of the near-field image, it was observed that there were disk-shaped areas about 300 nm with lower infrared absorption, which could correspond to the electron microscope results of melanin particles. However, further spectral analysis found that the spectra obtained in the areas considered to be rich in melanin were different from those in the areas considered to contain melanin. The peaks of 1760 and $1532{\mathrm{cm}}^{-1}$ in the spectrum were used to represent the keratins. There were four peaks in the spectrum obtained in the melanin-enriched region, which were 1290, 1454, 1563, and $1638{\mathrm{cm}}^{-1}$, respectively. However, a spectrogram of some areas that identified as melanin in morphology was quite different from that obtained from the area with rich melanin. Their spectra were more like the spectrograms of keratins, although the peak position could not completely correspond to the keratins. Since it could not be directly proved that the region similar in morphology must be melanin, unsupervised hierarchical cluster analysis was used to analyze the characteristics of different regions, as shown in Fig. 4(c). The areas enriched in melanin and the areas suspected to be melanin show homogeneity. Further analysis speculated that the spectral response of the suspected melanin region was actually a linear superposition of the spectral response of the melanin region and the spectral response of the keratin region. Stanic et al.⁷⁹ characterized the chemical composition of keratin and melanosome in hair and to distinguish eumelanin and pheomelanin based on their s-SNOM spectral signatures. As shown in Fig. 4(d), the spectra of pheomelanin and eumelanin are significantly different. Bildstein et al.⁸¹ used resonance-enhanced mode AFM-IR to localize the lipid distribution in hair. The heterogeneity between tissues, including kidney deposits,⁸² different skin layers,⁸³ osteoid nodules,⁸⁴ and sorghum stem wax,⁸⁵ could be revealed using infrared near-field techniques. Esteve et al.⁸² carried out AFM-IR spectral analysis of abnormal deposits in the kidney and acquired their chemical composition, including pure vancomycin and organic materials with phosphate. Kemel et al.⁸³ obtained chemical components of different skin layers using AFM-IR. Rammal et al.⁸⁴ used AFM-IR for precise chemical composition analysis of bone-like nodules. They found clear spectral differences between the cell-rich and collagen-rich regions.⁸⁴ Farber et al.⁸⁵ used AFM-IR to assess the heterogeneity of internal and external chemical components of sorghum stem wax.

Fig. 4

The application of near-field detection of bio-tissues. (a), (b) The phase images and spectra of organic sheath. Adapted with permission from Ref. 78. (c) Near-field results of three feature locations in a human hair cross section. Adapted with permission from Ref. 7. (d) Infrared s-SNOM measurements of pheomelanin and eumelanin in sheep black hair. Adapted with permission from Ref. 79. (e) Nanospectra obtained from seven positions for awns of wheat. Adapted with permission from Ref. 80 © 2020 American Chemical Society.

In addition to the identification of single components, the process of tissue state change can also be explored, including the process of the increase of lenses’s opacity,⁸⁶ tumor development changes,⁸⁷ bone maturation process,^88,89 tooth mineralization,^90,91 and plant silicification.⁸⁰ Paluszkiewicz et al.⁸⁶ used AFM-IR to discriminate proteins in clear and opaque (cataractous) human lenses. The secondary structure of proteins in the clear lenses was mainly $\beta$-turn, whereas the secondary structure of proteins in diseased lenses was mainly antiparallel $\beta$-sheet.⁸⁶ They also characterized pleomorphic adenomas and their marginal tissue by AFM-IR.⁸⁷ Pleomorphic adenoma is a common salivary gland tumor. Although there have been related applications of spectroscopic detection, the change process of tumors needs to be intensively investigated. They detected the structure changes in pleomorphic adenomas and found the presence of amyloid fibrils inside the cells. Imbert et al.⁸⁸ studied the composition and structure of tissues with different maturities in sheep cancellous bone using AFM-IR. They found that the mineral-to-matrix ratio increased in mature bone tissues and observed a nonmineralized layer.⁸⁸ Ahn et al.⁸⁹ focused on the change of matrix/mineral ratio in bone tissue with age. They found that the mineral content increased and the matrix/mineral ratio decreased with age.⁸⁹ The matrix/mineral transition areas with a size range of 100 to 200 nm were also found in the bone tissue.⁸⁹ To understand the mechanism details of tooth mineralization, Sereda et al.⁹⁰ characterized the details of demineralization and remineralization of dentin, and Huang et al.⁹¹ detected the differences in chemical composition between peritubular and intertubular dentin in human teeth. According to the analysis of the obtained spectrograms and nanoimages, there were more minerals in the peritubular dentin and more organic matter in the intertubular dentin.⁹¹ In order to study the process of plant silicification, Zancajo et al.⁸⁰ observed four plant samples, such as wheat awn by s-SNOM in 2020. Since the silica had an obvious absorption peak at $1100{\mathrm{cm}}^{-1}$, this absorption peak could be used in the s-SNOM absorption spectra to analyze whether the wheat awn was silicified, as shown in Fig. 4(e).

When multiple substances within the tissue are of interest, special data-processing methods can be taken to distinguish the different substances. Tranca et al.⁹² studied zebrafish retinal fragments with s-SNOM. The retinal fragments of zebrafish contain a variety of components, and the various components cannot be distinguished by microscopic observation. For such mixed samples, special data-processing methods can be used to make the data more visible. The phasor method was used to sort out the data obtained by s-SNOM, including near-field amplitude and near-field phase, and the phasor diagrams were recoded after color coding⁹³. The results after treatment showed that four regions corresponding to the four components could be well distinguished.

Using infrared near-field techniques, specific substances within tissues can be identified, including their distribution. The characterization of specific substances contributes to the understanding of tissue heterogeneity and physicochemical properties of biological samples. Therefore, it can be better used to study the development process of tumors and biomineralization mechanism of organisms.

2.4.1.

Treatment methods for samples in liquid environment

In order to achieve effective detection in the liquid environment, three methods are used, namely, designing a special liquid cell, improving the optical path structure, and using a liquid environment with low infrared absorption. The purpose of preparing special liquid cells is to isolate the liquid environment from the probe. The liquid cell cover is selected carefully, and it needs to transmit infrared light and prevent liquid overflow at the same time. Graphene, with its atomic thickness and infrared transparency, is suited for a lid.^95,96 However, because graphene may cause a high-pressure state in the internal liquid environment, other types of materials, such as hydrophilic SiN membrane, are also selected.^8,94 The optical path is designed as a total internal reflection mode and equipped with solid immersion lens⁹⁷ to minimize the liquid absorption of infrared light. In addition, the use of liquids with weak infrared absorption such as ${\mathrm{D}}_2\mathrm{O}$ can distinguish the infrared absorption of the biomedical samples from the infrared absorption of the liquid.⁹

2.4.2.

Test results and analysis

Diphenylalanine (FF) is a short peptide formed by dehydration and condensation of two phenylalanines and is the core recognition module of amyloid protein in Alzheimer’s disease. Studying the structure of this kind of dipeptide can have a deeper understanding of the process related to amyloid protein. Ramer et al.⁹ used AFM-IR in a liquid environment to detect self-assembled FF aggregates and FF aggregates modified by tertbutoxycarbonyl (Boc-FF) and distinguished the two aggregates through different spectrograms. The morphologies of the two aggregates were similar, so they could not be distinguished directly, as shown in Fig. 5(a). Because the secondary structures of FF and Boc-FF are different, the corresponding infrared spectral response will also be different. As shown in Fig. 5(b), the infrared spectrum of the Boc-FF polymer was different from that of the FF polymer. Due to the existence of additional carboxylic group and tertbutyloxycarbon group, there were absorption peaks at 1730 and $1660{\mathrm{cm}}^{-1}$ in the spectrum of Boc-FF, and molecules had greater conformational freedom and new $\alpha$-helical structures. The absorption peak of antiparallel $\beta$-sheet related to FF in Boc-FF ($1690{\mathrm{cm}}^{-1}$) overlapped with the absorption peak of a carbamate group in Boc-FF ($1696{\mathrm{cm}}^{-1}$), which made the peak move to high frequency. In addition to the above polypeptide aggregates, human serum albumin with relatively simple structure has also been used in research. Meireles et al.⁹⁶ combined with synchrotron radiation accelerator and near-field optics and used a special liquid cell covered with graphene to conduct nanoscale infrared analysis of human serum albumin. Graphene acted as a barrier to water and an optical window. Because human serum albumin would adhere to the graphene layer, there was a liquid-free protein region below the graphene. There were two peaks at 1650 and $1540{\mathrm{cm}}^{-1}$, which corresponded to amide I and amide II bands. This work achieved a spatial resolution of 25 nm and detected 1120 human serum albumin molecules within a thickness of 50 nm. In addition, the dynamic assembly process of protein at nanometer level can be analyzed. Zhao et al.⁸ used s-SNOM and a liquid cell closed by graphene membranes to characterize the self-assembly process of S-layer proteins. By providing a variety of buffer conditions, they demonstrated that the self-assembly strength of proteins increased nonlinearly with time, and the assembly was inhomogeneous. Figure 5(c) shows the changes of nanospectra with time in different concentrations of calcium ions. As plotted in Fig. 5(d), the integral intensity of the amide I and II bands increases gradually with time. They proved that the concentration of calcium had an impact on protein assembly.

Fig. 5

The application of near-field detection in liquid environment. (a), (b) AFM-IR results of the FF and Boc-FF in ${\mathrm{D}}_2\mathrm{O}$. Adapted with permission from Ref. 9 © 2018 American Chemical Society. (c), (d) The changes of nanospectra of S-layer proteins with time. Adapted with permission from Ref. 8. (e)–(g) Near-field third harmonic amplitude (e) and phase (f) and AFM topography (g) of purple membranes in water. Adapted with permission from Ref. 97 © 2020 American Chemical Society. (h)–(j) Li-PFIR microscopy results of yeast cell. Adapted with permission from Ref. 98 © 2021 American Chemical Society.

In addition to proteins, cell structures, such as cell membranes and cell walls, have also been studied. Pfitzner and Heberle⁹⁷ chose a solid immersion lens as infrared transparent material in 2020 and studied the purple film of bacteria in a liquid environment, as shown in Figs. 5(e)–5(g). They used the silicon hemisphere as a solid immersion lens. Whether in a dry environment or a liquid environment, the signal diagrams under different wavenumbers were obviously different. At $1660{\mathrm{cm}}^{-1}$, obvious contrast could be seen, but at $1711{\mathrm{cm}}^{-1}$, there was no obvious contrast, indicating that this method could detect the bacteriorhodopsin, which had the secondary structure of $\alpha$-helical. By improving the illumination light path and detection light path, the imaging quality in liquid was optimized, and the signal-to-noise ratio of the system in water was improved from 3.9 to 11.8. Their work provided a new idea for liquid measurement and also gave a method to optimize the spectral quality. Wang et al.⁹⁸ used liquid-phase PFIR, with a spatial resolution of 10 nm, to confirm the chemical composition of the budding site of yeast cell wall particles, as shown in Figs. 5(h)–5(j). They proved that there was protein around the budding site. Viruses in liquids have also been studied. Khatib et al.⁹⁵ took the graphene as a lid and identified the proteins of individual viruses by s-SNOM. The dynamic changes of individual bacteria or cells can also be observed. Mayet et al.⁹⁹ carried out an AFM-IR test of water-immersed bacteria. Kaltenecker et al.⁹⁴ used a SiN film as the cell cover to explore E. coli and live cancer cells. The s-SNOM images of cancer cells at different times were recorded, showing how the cancer cells changed over time.

For biomedical samples, such as proteins, cell membranes, and cells, the infrared near-field detection techniques based on liquid environments can achieve nanospectral results and have the nanoscale resolution and the single molecule sensitivity similar to that based on dry environments. Biomedical samples retain some activity in their natural environment, so their dynamic processes can be detected. With the further optimization of infrared near-field detection techniques in liquid environments, it will have great development prospects to study pathological process of related tissues in vivo samples while improving the sensitivity and signal-to-noise ratio of the system.