Tip in–light on: Advantages, challenges, and applications of combining AFM and Raman microscopy on biological samples

Abstract Scanning probe microscopies and spectroscopies, especially AFM and Confocal Raman microscopy are powerful tools to characterize biological materials. They are both non‐destructive methods and reveal mechanical and chemical properties on the micro and nano‐scale. In the last years the interest for increasing the lateral resolution of optical and spectral images has driven the development of new technologies that overcome the diffraction limit of light. The combination of AFM and Raman reaches resolutions of about 50–150 nm in near‐field Raman and 1.7–50 nm in tip enhanced Raman spectroscopy (TERS) and both give a molecular information of the sample and the topography of the scanned surface. In this review, the mentioned approaches are introduced, the main advantages and problems for application on biological samples discussed and some examples for successful experiments given. Finally the potential of colocated AFM and Raman measurements is shown on a case study of cellulose‐lignin films: the topography structures revealed by AFM can be related to a certain chemistry by the colocated Raman scan and additionally the mechanical properties be revealed by using the digital pulsed force mode. Microsc. Res. Tech. 80:30–40, 2017. © 2016 Wiley Periodicals, Inc.

divergent sample preparation requirements, the destructive nature of many wet chemistry approaches are several of the inconveniences when trying a multidisciplinary approach, which is essential nowadays in numerous disciplines (Andersen et al., 2011;Cloarec et al., 2008;Drent, 2003;Fowler et al., 2002;Rodríguez-Vilchis et al., 2011;Tharad et al., 2015). The (colocated) combination of different nondestructive techniques, by which the same spot of the sample can be measured by two or more approaches is therefore the "new" trend (Moreno-Flores and Toca-Herrera, 2012). In the past years spectroscopic approaches have especially gained attention as main element for the combination with other modus operandi. Spectroscopy studies the interaction between light (of different frequencies) and matter, from which different properties and characteristics of the material can be derived (Harris and Bertolucci, 1989). Concretely Raman spectroscopy has a wide spectrum of applications, due to its non-destructive nature (if correctly applied) and its suitability for combining with other methods such as scanning electron microscopy (SEM) (Cardell and Guerra, 2016;Timmermans et al., 2016), flow cytometry (Biris et al., 2009), or atomic force microscopy (AFM) (Apetri et al., 2006;Biggs et al., 2012;. In the next paragraphs, the state of the art Raman microscopy in combination with atomic force microscopy will be described as nondestructive approaches giving complementary data about on the one hand surface structure (topography) and other properties (e.g., adhesion, stiffness,. . ..) and on the other hand the molecular structure (chemistry) of the sample.

| C ONFOCAL RAMAN M ICROSC OPY ( CRM )
The underlying principle of Confocal Raman microscopy is based on light scattering. Well known is the "Rayleigh scattering" explaining our perception of the blue sky by elastic (with the same energy in comparison to the excitation wavelength) scattering of the sunlight on the small air particles. Because the human eye is more sensitive to blue than violet light and the blue light has shorter wavelength than the rest of the VIS spectrum, it is scattered with more intensity (according to I a 1/k 4 ) (Milonni and Eberly, 2010). However, a small amount of light (10 210 ) is inelastically scattered (Raman scattering) with more (anti-Stokes) or less energy (Stokes) due to interaction with matter. The population of molecules being in the ground energetic level is favoured (Boltzmann factor as function of temperature) and these molecules will be scattered with less energy (Stokes). The intensity of Stokes scattering is therefore much higher than the anti-Stokes and thus this part of the spectrum is detected in conventional Raman spectroscopy (Parson, 2009). The energy difference corresponds to molecular vibrations (stretching, bending, torsion, etc.) and is responsible for the recorded shift 1 in the spectrum, usually plotted in wavenumbers (ṽ[cm 21 ] 5 10 7 / k) (Griffiths, 2006). How easy a molecular vibration takes place is inherent to the polarizability -"the ability to the electron cloud of a molecule to be disturbed." Therefore nonpolar and symmetric molecules undergo a change in their electron distribution when interacting with monochromatic laser light and are indeed Raman active. The shift in the energy gives rise to a spectrum with different Raman bands that are characteristic for vibrations of specific bonding, functional groups and/or molecules. Depending on the type of vibration, the spring constant of the bond (or bond strength) and the weight of the atoms involved in the movement (as function of Hook es law), 2 the energy for the motion and thus also the location of a specific band in the spectrum will be specific for the involved functional group and/or molecule.
Stronger bonds (higher k) will cause the correspondent band (e.g., C@C compared to CAC) to appear at higher wavenumbers (more energy is lost from the incident light) whereas heavier atoms (higher m) (O-D, being D deuterium oxide, compared to OAH) will "vibrate" at lower wavenumbers. Stretching of atoms is less favourable than bending and thus will generate a higher energy loss: the scattered light will be placed at higher wavenumbers in the spectrum (big shift) (Smith and Dent, 2005). A Raman spectroscopy system consists of an irradiation source (normally a coherent, collimated and polarized laser), a sample mount, a spectrometer and a detection system, usually a charge-coupled device (CCD camera). The coupling of this system with a confocal microscope allows the generation of spatially resolved spectra (Raman imaging) and has improved the lateral resolution of the technique up to 250 nm (diffraction limited spatial resolution given by the Rayleigh criterion 3 ) (Griffith, 2009;Hollricher, 2010). The z-resolution can reach about 800 nm or poorer, but also depending on the instrumental set up (numerical aperture (NA) of the magnification lens, pinhole or fiber diameter) and the transparency or nature of the sample (i.e., refractive index) (Everall, 2004). The high spatial resolution, the possibility to measure in wet environment like water or buffer and easier sample preparation are some advantages compared to infrared microscopy, which also probes molecular vibration. In the VIS and IR range, in contrast to X-rays, the sample does not suffer from direct damage. Different excitation wavelengths (from the ultraviolet to the near infrared) can be used for Raman microscopy. A green laser with 532 nm is common in routine measurements due to the easy use with conventional microscope objectives and accessories (e.g., glass slides and cover slips) together 1 Dṽ 5 1 k0 2 1 k1 , where Dṽ is the Raman shift measured (in wavenumbers) in the spectrum and k 0 and k 1 the wavelength of the incident and scattered photons. where v is the frequency of vibration, k is the force constant of the bond between the atoms, c is he speed of light and m the reduced mass of the atoms involved being m5 M1M2 M1þM2 3 Rayleigh criterion: the lateral resolution is given by r5 0:61k NA where k is the excitation wavelength of the incident laser and NA the numerical aperture of the objective through the signal is focused and collected. with high Raman scattering intensity. However, with this wavelength biological samples often yield high fluorescence and longer excitation wavelength for example, 785 nm, gives better results. Nevertheless toward the higher wavelength (infrared) the Raman intensity is decreasing with the fourth power and the spatial resolution decreases 3 (Griffith, 2009;Hollricher, 2010).
Most modern equipment include a precise scan stage (often piezo driven) that allows nm-wise movements with great accuracy for chemical imaging, that is, recording a Raman spectrum each 250 nm in x and y directions in a matrix-like way. Each spectrum at each position represents a molecular fingerprint and by simply integrating over the specific band of a components characteristic vibration (spectral position), a false colour image of the distribution of that component can be generated (Dieing and Ibach, 2010;Geladi et al., 2010). However, for Raman images of biological systems it is often necessary to apply multivariate data analysis due to many overlapping bands (Felten et al., 2015;Gierlinger, 2014;Mujica Ascencio et al., 2016;Piqueras et al., 2015;Taleb et al., 2006). Stack scans (3D Raman imaging) can be also performed thanks to the use of zmotorized tables when using a confocal microscope. The different focal planes can be resolved by using very narrow fibers as pinholes (Hollricher, 2010).
Every high-speed imaging tool has some drawbacks. Signal to noise ratio of the single spectra is often reduced since a compromise has to be taken with integration time (measurement speed) and the laser intensity not to damage and/or burn biological samples during the scan. Furthermore, large amount of data are generated (1 spectrum each 0.3 mm) which needs computing capacity, external servers and data analysis and management software. Furthermore, "Raman imaging" with high spatial resolution (high numerical apertures) is very sensitive to the focal plane of the measured area which means that the investigated surface should be as flat as possible.
Although Raman microscopy offers chemical (together with structural and conformational) information with high resolution, it is not enough when dealing with phenomena, molecules and bonds from a few angstroms (small peptides) to few nanometres (DNA helix diameter). Only electron microscopy, scanning probe microscopies or superresolved fluorescence microscopy (Huang et al., 2009) achieve the resolution below the diffraction limit of light (Verma et al., 2010).
Nevertheless not all of them deliver chemical information. Not to mention, the implementation of a method should be at least generally applicable to the majority of samples in native conditions and free of time consuming sample preparation steps and specific labelling or staining.
The last condition makes scanning probe techniques more favourable. Especially Atomic Force Microscopy (AFM) (Binnig et al., 1986) Bhushan, 2002). The AFM principle is based on Hooke's Law, 4 which states that the restoring force of a spring is proportional to the displacement applied, or vice versa (Bhushan et al., 2004;Moreno-Flores and Toca-Herrera, 2012). The spring is a cantilever made of silicon (and other materials such as silicon nitride) that scans the sample's surface maintaining the force between cantilever and sample constant (contact mode) (Binnig et al., 1986). One of the most used methods works under the assumption of keeping the oscillation amplitude constant by exiting at a frequency near the resonance frequency of the cantilever. The tip stays in a non-contact or intermittent contact regime (tapping or AC mode) (Hansma et al., 1994). Additional dynamic techniques have also been developed, including frequency modulation of the cantilever (Albrecht et al., 1991) or jumping mode (De Pablo et al., 1998;Moreno-Herrero et al., 2000). In the relative new sub-resonance method digital pulsed force modulation (DPFM) ( Figure 1) the cantilever oscillates far below its resonance frequency with a sinusoidal modulation that allows higher repeat rates (up to 20 kHz). The tip snaps in and out of the sample homologous to what happens by a triangular modulation in force-distance curves: after snip in, further approach causes the bending of the cantilever that depends on the mechanical properties of the sample, whereas in retraction the pull off force reflects the adhesive forces between tip and sample (Gigler and Marti, 2008;Gigler et al., 2007). The resolution achieved in AFM is determined by the sharpness and length of the tip, FIG URE 1 Working principle of atomic force microscopy (AFM) in digital pulsed force mode (DPFM). The cantilever oscillates free with a sinusoidal modulation at lower frequencies than its resonance frequency. At very short distances from the sample surface, the tip snaps into the sample. The repulsive forces increase as the tip pushes towards the sample and they reach a maximum (F max ). From the slope of the "indentation" in the repulsive regime decreases and attractive forces originate between tip and sample, which in turn corresponds to the adhesive local forces between them. When the spring constant of the tip overcomes the adhesion forces, the tip snaps out and a new cycle begins. The baseline corresponds to long range forces and must be set to zero before any read out of absolute values. and its shape (pyramidal, conical, or bead-like, e.g.), but also the radius of the sharp end of the tip (a few nanometres, typically 10 nm) and its softness -spring constant (k). Tip materials or functionalization can also improve the resolution and give further information about electrical, thermal, magnetic or chemical properties (Ebner et al., 2007;Friedsam et al., 2004;Hapala et al., 2014;Wildling et al., 2011). 2.1 | What AFM brings to Raman microscopy and vice versa AFM reveals structural (topography) and mechanical information (e. g., stiffness or Young's modulus, viscosity, local adhesion forces) on the nano-scale complementary to the chemical information delivered by Raman microscopy on the sub-micron scale. Therefore using both methods gives quite a complete picture and better understanding of biological systems. Both combined give additional the chance to reveal chemistry on the nano-scale as summarized in Figure 2 and Table I.
Integrated AFM and Raman (co-located) was first possible in the early 2000s. Several set ups in the market are available for co-located AFM-Raman; some of them allow both methods to work simultaneously. While the AFM scans the top, the Raman scattering is acquired from the bottom part of the sample (bottom illumination) with a high NA objective, which limits the measurement to transparent samples only. In top illumination, another not so optimal version of the combined set up is to use special shaped tips for the AFM which enable the laser light to excite the sample from the top at the free-tip area.
However, illumination of the cantilever with near-IR light (e.g., 785 nm) can also affect in certain degree the AFM scan (due to cantilever heating). Shadowing effect from the tip can also take place in the top illumination configuration, which affects the far field Raman signal. To overcome shadowing effects, the signal can be also collected from the side while the AFM scan is running (side-illumination). In general, the signal when collected from the side is lower in intensity and skewed (only from one side at one specific angle) since the optimal excitation and collection angle have to be optimized (Lucas and Riedo, 2012). Top and side illumination arrangements allow only relatively low magnification (203 or 503) objectives with high working distance (with normally NA not higher than 0.5) (Berweger and Raschke, 2010) since AFM holder and cantilever need a certain space to operate. This reduces dramatically the spatial resolution of the Raman image.
However, most of the co-located AFM-Raman instruments work in a one-after-the-other manner. Polymer science was one of the first disciplines to benefit from the colocated measurements and a proof of principle of the methodology can be found in (Schmidt and others, 2005). After 20 years only few examples of its application in biological science/materials have come up. The "sensitive" nature of biological materials and their heterogeneity add difficulties when performing both methods at the same spot. Many biological materials need a dedicated sample preparation or fixation to be ready for successful AFM measurements and Raman imaging.

| AFM combined with Raman in biology: From medicine to new energy generation
Combined structural and chemical data revealed conformational changes during the a-synuclein aggregation (fibril formation) in the course of Parkinson's disease (Apetri et al., 2006). A closer look at the Raman amide I and III bands together with topographical information led to the conclusion that oligomers with a-helical secondary structures accumulate and then undergo a conformational change to b-sheets in protofilaments, which could be a step to control the kinetics of the fibrillization process. One of the advantages of Raman microscopy is that there are no restrictions on size or form (adsorbed, in solution) of the specimen to analyse. Extracellular polymeric substances produced in biofilms of Acidithiobacillus ferrooxidans grown on uranium have also been studied by micro-Raman and AFM although no differences in the Raman spectra (composition) of the FIGUR E 2 Operation principles when combining surface probe microscopies (SPM) and Raman spectroscopy. Best interpretable Chemical information is achieved applying conventional Confocal Raman microscopy, but spatial resolution is limited by the diffraction of light. Near-Field Raman microscopy and tip enhanced Raman spectroscopy (TERS) overcome this limitation, but are more difficult to be operated on complex biological systems. Atomic force microscopy (AFM) gives structural and mechanical information on the nanoscale and is therefore in combination with Raman microscopy an important tool to reveal structure-function relationships.
The method is not only restricted to bacterial cells but also adherent eukaryotic cells can be imaged, for example, mechanical (changes in cytoskeleton) and chemical profiles (weaker bands) of human breast carcinoma cells at ectopic sites changed after expression of BRMS1 (metastasis suppressor), whereas the tumour site stayed intact . Quantification of RNA, DNA, and proteins has been done for different cells by recording Raman maps and calibration models for each component and correcting focal effects with the topography image provided by AFM (Boitor et al., 2015). Also the assessment of chemical modifications (substitutions) or mechanical processes (milling, flattening) can be studied by combining both techniques. Very often the penetration depth of certain applied modifications is not evenly distributed since it depends on diameter or the thickness of the granule or surface to be modified. This phenomenon was approached by Wetzel et al. (Wetzel, 2010) who exchanged the hydroxyl groups of waxy maize starch granules by hydrophobic octenyl-succcinate to proportionate emulsifying functionality. The spatial distribution of the modification was monitored by Raman microscopy in combination with cluster analysis whereas AFM-phase and -topography images depicted the location and functional charac-teristics of the modified granule. The alkaline treatment on cellulose was also studied in the same way by (Eronen et al., 2008). In medicine, this combination of methods has also been applied to inquire into drug-drug interactions and their susceptibility to agglomerate in the carrier. Taking into account adhesive properties from forcedistance curves between propellant medium and the drug, the formulation can be optimised (Rogueda et al., 2011). Likewise the formation of a pore network during in vitro drug release from a carrier polymer matrix has as well been observed by co-located AFM-Raman (Biggs et al., 2012). Moreover catalytic reactions of metal nanoparticles have also been proofed by this technology (Harvey et al., 2012). In this case, the metal nanoparticles enhanced the Raman sig-   (Lebedev et al., 2014). The measurements showed that the abundance of c-type cytochromes is associated with the transformation of the growing pattern (lone cells to 2D and then 3D associated cells) during lag and exponential phase of their growth generating low and high current, respectively.
2.3 | The curse of light being a wave and the way from micro-to nano-Raman As stated above, Raman microscopy resolution is limited by the wave nature of radiation (diffraction limited) (see Table I (Synge, 1928). SNOM is an optical scanning probe technique that takes advantage of the developed AFM technology (piezo scan tables, miniaturization of tips and beam deflection setup as feedback) and uses AFM-like hollow probes to focus the light through a sub-wavelength aperture and bring very close photons and sample ( Figure 2). As molecules can be defined as an ensemble of dipoles, we might assume that their charge oscillates and thus they are attached to an electrical field. The whole "field" has several boundaries including a near (few nanometers from the object) and a far field (what is seen in optical microscopy). The part of the field approached in SNOM is the near-zone due to the close vicinity between the far end of the tip and sample. The resolution of the optical image reached is then in the range from k/10 to k/50 nm (depending on the diameter of the aperture of the tip). The implementation of AFM probes provides also simultaneously topographical information of the sample's surface (Pohl et al., 1988).
After fulfilling the scope of getting a nanometer resolved optical image and its topographic profile, the following aim is to obtain some more information about the nature of the sample, that is, chemical composition. Near-Field Raman Imaging is achieved by combining SNOM technology together with a very sensitive, high-throughput spectroscopic detection system (Anderson et al., 2005). After focusing the laser light through the SNOM tip and keeping it very close to the surface, an evanescent field is formed at the end of the tip which is able to "excite" only a few nanometers on the sample (the area corresponding to the near field) (Marocchi and Cricenti, 2001). In the same manner as common Raman spectroscopy, only a very small amount of the light is inelastically scattered and the energy corresponds to characteristic vibrations of the molecules in the sample. The scattered light derives only from this small area or volume in the sample and depends on the aperture radius of the tip (Hartschuh et al., 2003). The shift in the energy of the transmitted light is collected by a high-numerical aperture lens from the bottom. 5 By this the Raman spectrum of very thin layers or very small sample volumes can be recorded. While the lateral resolution is highly increased, the spectral signal becomes quite low in comparison to conventional upright Raman systems.
In 2000, tip-enhanced Raman spectroscopy (TERS) was developed as an apertureless SNOM variant (St€ ockle and others, 2000). The method follows the principle of enhancing the Raman signal by approaching a metallic tip very close to the sample surface (<10 nm), similar to the effect of depositing the sample on a specially structured (rather rough) metal layer or particle like in Surface Enhance Raman Scattering (SERS) (Kerker et al., 1980;Nie and Emory, 1997). The excitation of the rough metallic surface with laser illumination originates surface plasmons that lead to strong local electromagnetic fields at very narrow positions. In TERS, the tip acts as an optical antenna that enhances the electrical field at its far end (Stadler et al., 2010;Verma et al., 2010). Tips used in TERS are metallic (normally gold or silver) and can be prepared by different methods such as vacuum evaporation, lithography or electrochemical etching (deposition of Ag on silicon cantilevers). Tip manufacturing is a limiting factor for the field enhancement, since the production is expensive and the reproducibility low.

| A case study: Colocated AFM-Raman on cellulose-lignin films
To not only describe biological samples, but to reveal important Artificial lignin polymer (dehydrogenation polymer, DHP) has been also paid attention for its adhesive properties (Hoareau and others, 2006) and the enormous potential as renewable source of biomass. stiffness and the penetration depth (ranging from 1 to 8 nm in this case) at each point of the image are required to calculate the local stiffness of the sample. Topography of the sample is generated from the feedback mechanism (F max ) of the cantilever.
Raman spectra from the film were acquired using a confocal Raman microscope (alpha300RA, WITec GmbH, Germany) with a 1003 objective (numerical aperture (NA) 5 0.9) (Carl Zeiss, Germany). The sample was excited with a linear polarized (08) coherent compass sapphire green laser kex 5 785 nm (WITec, Germany). The scattered Raman signal was detected with an optic multifiber (100 mm diameter) directed to a spectrometer (600 g mm 21 grating) and finally to the CCD camera (Andor DU401A BR DD, Belfast, North Ireland). The laser power was set at 36 mW and a short integration time of 0.5 s was chosen in order to ensure fast mapping.
One spectrum was taken every 0.3 lm. Images of DHP and CNCs were generated by band integration over the main lignin band at 1,600 cm 21 and the cellulose band at 380 cm 21 (CCC ring, respectively). The Control Four (WITec, Germany) acquisition software was used for both acquisitions and the ProjectPlus (WITec, Germany) software for spectral and image analysis.
Prior to the Raman measurements, AFM was performed in order to avoid possible changes due to laser overheating. In Figure 3A the Topography, based on DPFM measurement is presented of the same area given by the Raman images in Figure 3B. Topography images indicate small agglomerates on the top (from 80 to 209 nm above the bottom) of the film. The Raman images and spectra ( Figure 3B-3C) reveal that these agglomerates have different chemical composition. Based on the spectra it is clear that these globules have higher amount of DHP (in red) and are clearly distinguished from CNCs richer regions (in blue).
DHP shows higher intensity of the aryl C@C stretching at 1,600 cm 21 typical of the phenolic rings present in lignin (Agarwal and Ralph, 1997;Larsen and Barsberg, 2010). Although blue areas show this band too, the intensity is clearly decreased and cellulose bands at 1,096 cm 21 (antisymmetric CAOAC stretching of glycosidic bond) and 380 cm 21 (CCC bending in ring) (Schenzel and Fischer, 2001;Wiley and Atalla, 1987) increased. A smaller area of 1.5 3 1.5 mm 2 comprising a DHP nodule was selected for a more detailed AFM scan ( Figure 4A-4D).
The diameter of the little spheres in the nodule range from 0.1 to 0.5 mm. They are connected by an interface that presents higher adhesive values ( Figure 4C) and thus the more hydrophilic CNCs. The stiffness image ( Figure 4D) does not resolve the nanocrystals as the scan was done on the globules revealed by Raman to agglomerate the DHP on top and a tip radius of 10nm. Yet slightly higher intensities confirm more cellulose in the bridging regions of the nodules. So the combination of AFM and Raman gives a comprehensive picture of the investigated films on surface structure, chemistry and mechanics from the submicron to the nano-scale.

| C O NC LU S I O N S
Raman spectroscopy has shown its feasibility for combination with other techniques over the past decades. The chemical information obtained is complementary to the mechanical/topographical images given by AFM. Their combination in TERS or near-field Raman has high potential to monitor chemical changes at the nanoscale. However, TERS has certain drawbacks that need to be solved: tip manufacturing and reproducibility are main concerns when doing these approaches as well as sample preparation (ideally thin, homogenous in depth and transparent) and low signal Raman spectra with different bands (tip enhancement effect) that are often difficult to interpret. The fact that only a few TERS specialized groups in the world are performing successful measurements is a sign that it is not yet a straight forward method to be applied on every biological sample of interest. Colocated Raman and AFM, despite lower Raman lateral resolution than TERS, has high potential as it offers the possibility to also monitor mechanical properties of the sample and reveal structure-function relationships. In case diffraction limited spatial resolution is not enough for chemical information, also near-filed Raman is an option. Nevertheless again one has to struggle with less Raman signal and more restrictions on sampling.