Contents lists available at ScienceDirect

Trends in Food Science & Technology

journal homepage: www.elsevier.com/locate/tifs

Single-molecule detection of proteins and toxins in food using atomic forcemicroscopy

R. Alexander Reesea, Bingqian Xub,∗

a Department of Chemistry, University of Georgia, Athens, GA, 30602, USAb Single Molecule Study Laboratory, College of Engineering and Nanoscale Science and Engineering Center, University of Georgia, Athens, GA, 30602, USA

A R T I C L E I N F O

Keywords:Atomic force microscopySingle-moleculeBiosensorDynamic force spectroscopyFood safety

A B S T R A C T

Background: Food safety is vital in everyday life. Food toxins are small peptides or proteins that can causedisease by disrupting biological macromolecules such as enzymes or cellular receptors. Toxins are introducedeither externally or from internal, spoilage-related pathogens. To keep these unsafe foods off the market, de-tection techniques for various toxins have been devised. Although many techniques serve this purpose well,challenges remain regarding sensitivity, specificity, and cost, especially for trace amounts of highly lethalproteinaceous toxins.Scope and approach: Atomic force microscopy (AFM) is a widely used nanotechnology for imaging and forcemeasurement in biophysics and biology. Its high sensitivity in probing specific, single-molecule binding betweenmolecules is applicable to food toxin detection. In this review, after a summary of the development of AFM,different detection operation modes are introduced along with examples of their applications.Key findings and conclusions: Cantilever sensing and recognition imaging are found to be the appropriate AFMtechniques for detection. Their shared functionalization approaches are outlined for two categories of surfaces:silicon and gold. Recent progress in AFM biosensors and their applications to food toxin detection are discussed.Single-molecule sensitivity and ease of designing sensing schemes make these AFM techniques excellent can-didates for real-world application. Existing challenges in designing sensing molecules and preventing the foodmatrix from confounding signals are not only applicable to AFM techniques but to most current biosensors.Through the collaboration among materials science, chemistry, and molecular biology, solving these issues willpromote significant advancement in AFM-based food toxin detection.

1. Introduction

Food safety is critically important to human health. The Centers forDisease Control and Prevention calculates that there are 9.4 millionepisodes of foodborne illness in the U.S. alone each year, resulting intens of thousands of hospitalizations and over a thousand deaths(Scallan et al., 2011).

Food safety is compromised by either the introduction of externalpathogens during cultivation, processing, or preparation; or by patho-gens resulting from spoilage. Of key interest here is the contaminationby proteinaceous toxins that are a byproduct of these pathogens andcause sickness at very low concentrations. These toxins include sta-phylococcal enterotoxin B (SEB), a commonly found cause of gastro-enteritis producing 185,000 cases annually (Mead et al., 1999); Clos-tridium botulinum neurotoxins, a deadly toxin found in improperlycanned foods and which is difficult but critical to quickly identify aftersymptom onset; saxitoxin, responsible for shellfish poisoning resulting

from eating shellfish harvested from areas with algal blooms; and ricin,a byproduct of castor bean oil (Buehler, 2005). Rapid, cheap, and ac-curate testing of food for these toxins is an important element to en-suring safety of the food supply.

2. Current toxin detection methods

Plating and culture enrichment can be used to identify pathogens,but these methods are costly in terms of time and labor. Polymerasechain reaction (PCR) emerged to take a major role in testing, but in-hibition of amplification by components of the food matrix can interferewith this method and thus necessitate several controls to ensure suffi-cient amplification of the correct targets (Hoorfar et al., 2003; Scheu,Berghof, & Stahl, 1998). Additionally, these methods are ill-suited torecognizing certain proteinaceous contaminants, which are able tocause illness at extremely low levels (Rasooly & Herold, 2006). Severaladvancements in biosensor research have resulted in a number of

https://doi.org/10.1016/j.tifs.2018.01.005Received 1 September 2017; Received in revised form 10 January 2018; Accepted 16 January 2018

∗ Corresponding author.E-mail address: bxu@engr.uga.edu (B. Xu).

Trends in Food Science & Technology xxx (xxxx) xxx–xxx

0924-2244/ © 2018 Elsevier Ltd. All rights reserved.

Please cite this article as: Reese, R.A., Trends in Food Science & Technology (2018), https://doi.org/10.1016/j.tifs.2018.01.005

methods for recognizing these, typically relying on an antibody or ap-tamer to specifically recognize the toxin of interest and then couplingthat recognition element to a transduction scheme (Turner, 2013).Methods based on piezoelectric crystal sensors, electrochemicalschemes (Patel, 2002), surface plasmon resonance (Homola et al.,2002), and ELISA have all been reported (Freed, Evenson, Reiser, &Bergdoll, 1982). ELISA is widely considered the industry standard withregards to toxin detection, making up for relatively costly sample pre-paration with flexibility and a sub-nanomolar detection limit.

A significant challenge for all techniques remains dealing with thenon-toxic food matrix, which is highly heterogeneous, frequently vis-cous and unamenable to instruments, and can nonspecifically interactwith the recognition element to generate false positives. Advancementsin sample preparation can largely be shared among the various bio-sensing schemes (Myszka, 1999), however, ultimately promoting thosewith lower detection limits and greater cost-effectiveness.

In a manner similar to the other biosensors, a recognition elementcan be coupled to an AFM to produce an instrument that is capable ofproviding extraordinarily high sensitivity, high selectivity, and a cost-effective path toward recognizing these contaminants and securing thesafety of the food supply.

3. Atomic force microscopy

Developed in 1986, an AFM consists of a flexible, sharp (tens ofnanometers) probe mounted on the end of a cantilever. A laser is re-flected off the backside of the cantilever and reflected onto a positionsensitive photodetector, allowing for the deflection of the cantilever tobe measured. The tip can be raster-scanned across a surface to generatea three-dimensional image of the surface morphology (Binnig, Quate, &Gerber, 1986), with the laser deflection used as the input of a feedbacksystem that allows the probe to follow the surface topography (Meyer &Amer, 1988) (Fig. 1).

The high lateral forces exerted on the substrate as the tip is scannedacross can damage of softer biological samples. Amplitude-modulatedAFM (AM-AFM), commonly known as tapping mode, was developed tocircumvent this problem. In AM-AFM, the cantilever is oscillated by anexternal driving force, where it acts as a harmonic oscillator. Thus, atsome specific frequency, it will oscillate with a set free amplitude andcan then be lowered into contact with the surface, reducing the oscil-lation amplitude until it reaches a set fraction of the free amplitude, atwhich the set point of the feedback system is established and main-tained for the duration of the scan (Fig. 1B). Because the tip only in-termittently contacts the surface and only applies a predominantly

normal force, lateral forces are effectively eliminated (Zhong, Inniss,Kjoller, & Elings, 1993). This technique was coupled to AFM's amen-ability in liquid environments (Drake et al., 1989), establishing AFM asa capable tool for probing a broad range of biological samples in phy-siologically-relevant conditions (Müller & Dufrene, 2008).

Another mode resulting from the amplitude modulation is phaseimaging. Dissipative interaction from inelastic soft or viscous samplescauses a phase shift to develop between the sinusoidal signal driving thecantilever oscillation and the recorded output signal. This phase lag,when mapped to each point scanned during the image, shows changesin the mechanical properties of the surface that are otherwise topo-graphically homogeneous (Magonov, Elings, & Whangbo, 1997).

4. Achieving chemical specificity with AFM

4.1. Probe functionalization

An AFM observes only the sample-tip forces as a deflection or am-plitude signal and that all other information must be interpreted fromthis information. Collecting chemical information in addition to thestandard morphological information requires that a chemically func-tional molecule be attached to the probe tip and used to scan a surfacemodified with the cognate molecule of the tip-bound recognition ele-ment.

A variety of chemical functionalization schemes have been devel-oped for this purpose with the two most common methods being sila-nization and exploitation of simple thiol-gold bonding on gold-coatedsurfaces.

4.1.1. Silanization of silicon and silicon nitride probe tipsAFM tips are typically made from silicon or silicon nitride, which

can be modified by organosilanes, forming self-assembled monolayerswith pendant functional groups. (3-Aminopropyl)triethoxysilane(APTES) is a common organosilane used for this and exposes a readily-modified amine group (Fig. 2A). The main challenge with this methodis the hydrolytic instability of APTES and similar compounds. Morerecently, 1-(3-aminopropyl)silatrane (APS) has been demonstrated as amore hydrolytically stable chemical for silanization with otherwise si-milar performance (Shlyakhtenko et al., 2003).

The density of the pendant functional can be controlled by com-bining the reaction solution with another organosilane without anavailable functional group. This is the strategy used by Li and cow-orkers, wherein triethoxy(ethyl)silane (TEES) is added to APTES in aratio of 4:1 (v/v) to reduce the density of the pendant amino group (Li,Michaelis, Wei, & Colombi Ciacchi, 2015).

A bifunctional PEG linker can be coupled to the pendant amine,permitting further functionalization. PEG linkers provide additionaladvantages: preventing non-specific tip-sample interactions and pro-viding conformational flexibility to the bound recognition element,improving its binding (Hinterdorfer et al., 2000).

4.1.2. Gold-coated probe tipsGold surfaces are biologically inert and easily modified with simple

thiol chemistries. It is thus the preferred choice for biological detectionwith AFM. A thin layer of gold (on the order of 10s of nanometers) canbe sputtered onto a standard silicon AFM probe. The sensing elementcan then be directly bound if it features an exposed, suitable thiol ordisulfide. If this is not feasible, the tip can then be incubated with bi-functional polyethylene glycol (PEG) linker with at least one thiol end.Again, a long PEG linker reduces non-specific tip-sample interactionsand allows the bound recognition element the conformational flex-ibility needed bind its cognate ligand (Hinterdorfer et al., 2000). Thethiolated end will react with the gold tip surface and expose the otherfunctional group, allowing further modification in a variety of direc-tions (Barattin & Voyer, 2008; Ebner, Hinterdorfer, & Gruber, 2007)(Fig. 2B). Commonly, carboxylate groups are used, which can be

Fig. 1. An overview of an AFM.(A) Schematic of a generic AFM. (B) Zoomed in view of the tip's motion over a surfacefeature (red) in contact mode (top) and tapping mode (bottom). Reprinted with permis-sion from Dr. Helen Hansma's lab webpage (http://web.physics.ucsb.edu/∼hhansma/biomolecules.htm). (For interpretation of the references to color in this figure legend, thereader is referred to the Web version of this article.)

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

2

activated to an N-hydroxysuccinimidyl (NHS) ester and then be non-specifically linked to a protein's or peptide's primary amines.

These methods largely bind elements together with uncontrollablegeometry. When this is a problem, it is also possible to bind sensingelements using copper-free click chemistry methods (Chen, Ning, Park,Boons, & Xu, 2009), wherein an alkyne- or azide-modified silane orthiol is first reacted to the silicon or gold (respectively) tip, followed byreaction with an azide- or alkyne-modified recognition element (re-spectively) (Fig. 2C). These functional groups are not naturally featuredin proteins and thus provide a singular point for binding, controlling theorientation of the recognition element with respect to the probe tip.

Overall, there is considerable flexibility in functionalizationschemes, lending a high degree of flexibility to the method dependenton the nature of the sensing element—for example, a protein exhibitinga binding site rich in lysine residues may not be suitable for NHScoupling and should instead be engineered with a specific free thiol orazide.

4.2. Substrate modification

In addition to functionalization of the AFM probe, the analyte understudy is usually fixed to a substrate (this is not always the case, as willbe discussed in 5. Cantilever sensors). To achieve high resolution andhigh sensitivity of detection, the surface must exhibit sub-nanometerultraflatness of sufficiently large areas so that any recorded informationrelates to the surface-bound molecules of study and is not masked bythe changes in the underlying substrate height.

Modified mica surfaces are commonly used for AFM studies.Muscovite mica exhibts excellent basal cleavage, producing sheets ofatomic flatness, which are hydrophilic and possess a negative surfacecharge at neutral pH. Silanization is again one of the more commonroutes. Mica is capable of being silanized with APTES, but suffers asignificant decrease in flatness, especially if exposed to aqueous en-vironments over several hours, which is common in AFM studies ofbiological systems. APS is more hydrolytically stable, and provides an

excellent alternative, but it must be synthesized as it is not yet widelycommercially available. Both expose a pendant amine that is readilyconverted to an amide bond with molecules through NHS coupling.

Gold(111) surfaces are another common choice, exhibiting ultra-flatness and readily produced by thermally evaporating gold ontocleaved muscovite mica. As with tip functionalization, gold substratesare ideal for biological AFM study due to negligible toxicity to biolo-gical samples and ready modification with thiolated or disulfide-con-taining biomolecules or with linkers for subsequent cross-linking (Bain& Whitesides, 1989).

Like mica, highly-oriented pyrolytic graphite (HOPG) easily cleaves,exposing an ultraflat surface. Unlike mica, however, it is hydrophobicand conductive, and while popular for imaging applications, the hy-drophobicity has been reported to denature proteins (Muller & Engel,2008), reducing its biological uses. Both HOPG and mica have provento be suitable supports for lipid bilayers (Egawa & Furusawa, 1999;Muller & Engel, 2008; Tanaka & Sackmann, 2005), permitting mem-branes, their embedded proteins, and other associated biological mac-romolecules to be probed with AFM in highly relevant biological con-ditions, while still maintaining tight control over surface morphology.

5. Cantilever sensors

Cantilever sensors, or micromechanosensors, constitute one direc-tion for applying AFM to food toxin detection without the need for amodified substrate. In it, a cantilever bearing a recognition element isoscillated at its resonant frequency and amplitude, establishing abaseline, and is then exposed to a solution bearing the sensing mole-cule's cognate analyte. The mass of bound analyte affects the resonanceconditions, with greater mass proportionally decreasing the resonancefrequency (Xu, 2012). This change can be calibrated to provide quan-tification, with dynamic range limited by the number of recognitionsites able to fit on the cantilever. This method has already been appliedto the detection of SEBs (Maraldo & Mutharasan, 2007), achieving adetection limit of 2.5 fg/mL in apple juice and 25 fg/mL in milk within

Fig. 2. Common methods for functionalizing AFM tips.(A) silanization of a silicon tip (B) thiolation of gold-coated tipReprinted with permission from (Senapati & Lindsay, 2016).Copyright 2016 American Chemical Society. (C) A click chemistrytip functionalization scheme for anti-ricin aptamer. Reprinted withpermission from (Chen et al., 2009). Copyright 2009 AmericanChemical Society. (For interpretation of the references to color inthis figure legend, the reader is referred to the Web version of thisarticle.)

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

3

the short span of 20min by measuring the impedance of the cantileverto detect a frequency decrease. The difference in detection limits be-tween milk and apple juice is attributed to the more complex nature ofmilk as a food matrix (it contains fats and proteins in addition to car-bohydrates), demonstrating the large confounding effects of the foodmatrix in these applications.

Liu and coworkers used a cantilever sensor with a slightly differentsensing scheme to detect botulinum toxins (Liu, Montana, Chapman,Mohideen, & Parpura, 2003). In their work, a cantilever is constructedwith an synaptobrevin 2-modified agarose bead bound to the tip by thecognate protein syntaxin 1A. When botulinum neurotoxin is introducedto the system, it interferes with the binding between the tip and bead,causing the bead to detach and the resonant frequency to increase,signaling detection (Fig. 3). The reported detection limit is 8 nM withina timeframe of minutes. The authors note a tradeoff between themethod's speed and its relatively high detection limit relative to ELISA,but note the inherent amplification in signal by relying on the de-tachment of a large bead could allow for improved sensitivity withfurther work.

Finally, an antibody-functionalized cantilever sensor was used todetect the antigens on the cell membrane of the food- and waterborne

pathogen cholera in a dilute buffer solution, capable of a detection limitof ∼1000 CFU/mL, low enough to preclude the need for enrichment(Sungkanak, Sappat, Wisitsoraat, Promptmas, & Tuantranont, 2010).

6. Recognition AFM: imaging and dynamic force spectroscopy

6.1. Recognition imaging

In topography and recognition imaging (TREC), a magnetic canti-lever is oscillated in liquid by an external alternating magnetic field(Han, Lindsay, & Jing, 1996). Because of the hydrodynamic drag on thecantilever, the quality factor of the oscillation is significantly reduced,as energy is lost from the downswing to the upswing and back. In thisoverdamped state, the energy for oscillation is largely derived from theforcing of the magnetic field on the cantilever. Consequently, themaximum amplitude and the minimum amplitude are decoupled (Strohet al., 2004). TREC requires that a recognition element—usually anantibody or aptamer—is tethered to the tip with a bifunctional PEGlinker. The PEG linker must be long enough both to ensure the tipavoids nonspecific interaction with the surface and to lend flexibility tothe recognition element such that its active binding site is capable ofinteracting with its ligand, but the linker must also be kept as short asfeasible to maintain high spatial resolution (Ebner et al., 2005). Inpractice, linkers anywhere from 5 to 20 nm are commonly used. Thefunctionalized tip's interaction with raised features on the surface re-duces the amplitude on the downswing but this is not carried over tothe upswing as the magnetic actuation restores the amplitude. Con-versely, recognition events reduce the upswing but not the downswing.The upper and lower extrema of the resulting signal versus time plotcan be separated and each converted into an image, the bottom partshowing topography and the upper part showing recognition sites asdark features that correspond in the xy-dimension to the topographicalimage (Fig. 4).

Validation of the specificity of recognition occurs via a series ofcontrols (Stroh et al., 2004): 1) where the tip's recognition element isblocked by a free cognate ligand leading to an image with normal to-pography but no recognition, 2) one in which the surface ligand isblocked by free recognition elements, leading to possible topographical

Fig. 3. Depiction of the cantilever sensing scheme for detection of botulinum toxin.(A) The syntaxin 1A (green) on the functionalized tip interacts with the synaptobrevin 2(red), attaching the bead to the tip as it oscillates. Exosure of botulinum nerurotoxin B(BoNT-B) cleaves the bead-bound synaptobrevin 2, releasing the bead. (B) An overview ofthe components of the cantilever sensor. The tip oscillation is driven by a function gen-erator coupled to a piezoelectric element mounted to the tip. (C–D) An increase in theresonance frequency of the cantilever is detected upon the release of the bead from (C) to(D). Reprinted with permission from (Liu et al., 2003). Copyright 2003 National Academyof Sciences. (For interpretation of the references to color in this figure legend, the readeris referred to the Web version of this article.)

Fig. 4. Cartoon depiction of TREC imaging.A) The Pico TREC box splits the top and bottom halves of the amplitude signal andconverts them to an image. A structural feature on the surface will reduce the bottomhalf's amplitude (seen as a bump in the heavy blue line after the signal is split) whereas arecognition event will reduce the top half's amplitude (seen as a dip in the heavy red lineafter the signal is split). (B) A representative TREC image. The topographical image isshown on the left shows bright spots where avidin molecules are bound to the surface.The corresponding recognition image (right) shows that the tip, biotinylated tip interactswith the bound avidin molecules, generating a recognition signal. Reprinted with per-mission from (Ebner et al., 2005). Copyright 2005 John Wiley & Sons, Inc.

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

4

identification of the now surface-bound recognition elements but againno recognition image, and 3) one in which a blank tip is scanned toensure that stickier domains on the surface are not responsible for anyapparent recognition.

As previously indicated, the PEG linker can reduce spatial resolu-tion, especially for the recognition image, but the problem is limitedbecause the binding event leading to recognition is a probabilisticevent, and as the tip is scanned over and around a recognition site, thetethered recognition element will display some tolerance in its position,but that will be normally distributed over the recognition site, preser-ving high resolution (Senapati & Lindsay, 2016).

6.2. Dynamic force spectroscopy

Additional quantitative information can be achieved from thesefunctionalized AFM probes through the use of single-molecule dynamicforce spectroscopy (DFS). DFS interrogates a ligand-receptor pair byrepeatedly lowering the functionalized tip's ligand into contact with thesurface-bound receptor, allowing a binding event to occur. The tip isthen retracted, the and the force of the interaction causes the cantileverto bend until the bond breaks and the cantilever snaps back to its

equilibrium position. This entire cycle is recorded by the PSD as a force-distance curve (Florin, Moy, & Gaub, 1994) (Fig. 5). The maximumstretching distance and the force of the bond (on the order of pN) ismeasured. A collection of unbinding forces from these curves can becompiled into a histogram, to which a Gaussian function may be fit toestablish the most probable unbinding force, F*. The loading rate (pN/s) of the AFM cantilever can be varied, influencing the binding energylandscape, as higher loading rates study a far-from-equilibrium regime,which causes energy barriers to diminish and some binding states tobecome inaccessible, while lower rates can approach a near-equilibriumregime. F* collected over a wide range of loading rates (a few orders ofmagnitude) allows the plotting of F* as a function of log(loading rate).This analysis, developed by Evans, permits the derivation of the dis-sociation rates (Evans, 2001) and through further theoretical develop-ment by Jarzynski on the relationship between non-equilibrium workand free energy (Jarzynski, 1997), has led to the development offormalisms for extracting free energy profiles from DFS spectra(Hummer & Szabo, 2001; Liphardt, Dumont, Smith, Tinoco, &Bustamante, 2002).

Bioanalytical systems designed with DFS do not necessarily need todirectly measure the analyte. Systems have been designed that record

Fig. 5. An overview of dynamic force spectro-scopy.Top pane: The tip is lowered into contact with thesurface (B), where the ligand-receptor pair binds(between B and C). The tip is then raised until theligand-receptor bond breaks at (E), yielding theunbinding force. Reprinted with permission from(Bizzarri & Cannistraro, 2010). Copyright 2010The Royal Society of Chemistry.Bottom pane: (A) Force-distance curves collectedon the aptamer-ricin system at varying loadingrates. For each loading rate, three curves areoverlaid and shown. (B) A 3D histogram showingthe frequencies of unbinding forces at eachloading rate. (C) F* vs. log(loading rate), as dis-cussed in 6.2. Linear fitting is for both aptamer(red) and antibody (blue). Reprinted with per-mission from (Wang et al., 2011). Copyright 2011The Royal Society of Chemistry. (For interpreta-tion of the references to color in this figure legend,the reader is referred to the Web version of thisarticle.)

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

5

the change in interaction force upon addition of the analyte, such assystems relying on the modulation of aptamer-surface interactions by aprotein analyte (Wei, Steckbeck, Köppen, & Colombi Ciacchi, 2013).DFS's flexibility in experimental designs has led to its use in measuring avariety of analytes, as reviewed by Li and coworkers (Li et al., 2016).

7. AFM applications toward food toxin detection

Recognition AFM has proven to be a capable tool for interrogatingvarious biological systems, owing to its non-destructive nature and itshigh-resolution in biologically-relevant conditions. It has been suc-cessfully applied to the detection of DNA, individual proteins, andproteins on living cell membranes (Lin et al., 2009; Müller & Dufrêne,2011; Stroh et al., 2004; Wang, Park, Xu, & Kwon, 2017). In 2009, Chenet al. used the method to detect the toxin ricin bound to a gold surfacewith an antibody-functionalized tip (Chen, Zhou, Park, & Xu, 2009),achieving single molecule resolution (Fig. 6), corresponding to a sub-femtomolar detection limit. The bonding between AFM probe tip andanti-ricin antibody was achieved using a click chemistry scheme(Fig. 2C), ensuring a specific attachment geometry between the tip andrecognition element (Chen et al., 2009). DFS was used to confirm thesingle-molecule nature of the antibody-ricin interaction, determining itsbinding strength to be 64.89 ± 1.67 pN. Additional work carried outby Wang et al. found that an DNA aptamer worked slightly better (asmeasured by binding interaction strength and koff) than the antibody asa sensing molecule and was even able to distinguish between ricinmolecules bound to the surface in different orientations, further verifiedby the use of molecular simulations (Wang, Guo, Chen, Park, & Xu,2011; Wang, Guo, Zhang, Park, & Xu, 2012) (Fig. 6)). The use of ap-tamers instead of antibodies is a promising direction for biosensors, asthey possess similar or better specificity and sensitivity, while beingmore robust and more easily tailorable to the application (Stadtherr,Wolf, & Lindner, 2005).

Key to the advantage of AFM-based techniques is its ability tomeasure interaction with single-molecule specificity. In a comparisonbetween AFM and SPR, while both technologies were able to detectricin at the minimum of a picomolar concentration, the koff for SPR was∼25 times higher, the result of SPR's measuring the ensemble average

of the interactions in bulk solution, including both specific interactionsand interfering non-specific interactions of the aptamer and ricin mo-lecules (Wang et al., 2011). AFM grants greater control over the mea-surement, even allowing for measurement of specific binding con-formations.

For all biosensors, including AFM-based ones, non-specific interac-tions on the sensing molecule can produce false positives, requiringfollow-up testing leading to a decrease in cost-effectiveness. Onemethod for reducing this chance is to determine the specific interac-tions as determined by their extension lengths. Both the PEG linker andsensing aptamer have elastic properties and extend—spring-like—whenpulled during DFS experiments, where that information is captured inthe force-distance curve. Determining the extension length for a givenunbinding pathway allows for non-conforming force-distance curves tobe removed from further analysis, reducing the chances that a false-positive signal resulting from non-specific adsorption will occur (Wang,Park, Kwon, & Xu, 2014). Visualization of this phenomenon can beachieved through 4D histograms (Fig. 7A), relating unbinding forces,extension length, loading rate, and frequency. Specific interactionpeaks from the force-distance curves collected by DFS are representedby larger data points, allowing more facile segregation of specific in-teractions from non-specific ones. This method has been demonstratedin work on the foodborne pathogen Salmonella typhimurium (Fig. 7B–C)(Wang et al., 2017).

Additionally, applying antifouling coatings to the substrates prior totheir modification with molecules has been explored as a method forreducing the non-specific adsorption of proteins (Sharma, Johnson, &Desai, 2004) and the adhesion of common bacteria (Zeng, Ogaki, &Meyer, 2015). Applying it to prevent fouling by food matrix has beendemonstrated in solid-phase microextraction for gas chromatography(Souza Silva & Pawliszyn, 2012).

8. Conclusion

Because of its perpetual importance, ensuring that foods are safeand free of proteinaceous toxins will remain important. AFM will be acapable tool toward securing food safety. Its biosensing capabilitieshave proven to be highly sensitive and specific, it can operate in all

Fig. 6. Detection of ricin on a gold surface withantibody and aptamer.Panel A. The modification of gold with ricin. (A)A lipoic acid–N-hydroxysuccinimide ester (LA-NHS) binds to a gold surface via its disulfide. (B)LA-NHS reacts to form an amide bond to ricin.Source: (Chen et al., 2009).Panel B. (a) Topographic image of ricin-modifiedgold with antibody-functionalized tip and (b) itscorresponding recognition image. Free ricin wasthen used to block the antibody on the tip,leading to a control (c) topography image and (d)recognition image. Note that the topographychanges little but that recognition is significantlydecreased. Source: (Chen et al., 2009).Panel C. (A) Topographic and (B) recognitionimages of ricin with an aptamer-modified tip. (C)Ricin molecules I-IV from the topographicalimage are selected and zoomed in to identify thespecific conformation ricin has bound to thesurface. These are each compared to a simulatedstructure of ricin, with good agreement. Rep-rinted with permission from (Wang et al., 2011).Copyright 2011 The Royal Society of Chemistry.(For interpretation of the references to color inthis figure legend, the reader is referred to theWeb version of this article.)

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

6

ambient environments, and it exhibits a wide and flexible variety offunctionalization schemes enabling it to be applied to virtually any foodtoxin or pathogenic protein (see Table 1). Work remains to be done indeveloping methods for sample preparation to better remove the con-founding effects of the non-toxic food matrix, but the high degree ofcustomizability in aptamers as recognition elements and work on anti-fouling coatings provide a path forward.

Funding

This review work did not receive any specific grant from fundingagencies in the public, commercial, or not-for-profit sectors.

References

Bain, C. D., & Whitesides, G. M. (1989). Formation of monolayers by the coadsorption of

thiols on gold: Variation in the head group, tail group, and solvent. Journal of TheAmerican Chemical Society, 111(18), 7155–7164. https://doi.org/10.1021/ja00200a039.

Barattin, R., & Voyer, N. (2008). Chemical modifications of AFM tips for the study ofmolecular recognition events. Chemical Communications, 0(13), 1513–1532. https://doi.org/10.1039/b614328h.

Binnig, G., Quate, C. F., & Gerber, C. (1986). Atomic force microscope. Physical ReviewLetters, 56(9), 930. https://doi.org/10.1103/physrevlett.56.930.

Bizzarri, A. R., & Cannistraro, S. (2010). The application of atomic force spectroscopy tothe study of biological complexes undergoing a biorecognition process. ChemicalSociety Reviews, 39(2), 734–749. https://doi.org/10.1039/b811426a.

Buehler, J. W. (2005). Public health response to biological and chemical weapons: WHOguidance. Emerging Infectious Diseases, 11(1), 186b–187 https://doi.org/10.3201/eid1101.040927.

Chen, G., Ning, X., Park, B., Boons, G.-J., & Xu, B. (2009a). Simple, clickable protocol foratomic force microscopy tip modification and its application for trace ricin detectionby recognition imaging. Langmuir, 25(5), 2860–2864. https://doi.org/10.1021/la803523b.

Chen, G., Zhou, J., Park, B., & Xu, B. (2009b). Single ricin detection by atomic forcemicroscopy chemomechanical mapping. Applied Physics Letters, 95(4), 43103. https://doi.org/10.1063/1.3190197.

Fig. 7. Analysis of elastic properties of the unbinding pathway.Panel A. Cartoon of DFS on S. typhimurium and a corresponding force-distance curve, highlighting the extension caused by unbinding.Panel B. (a) A topographical image of S. typhimurium showing the whole cell in liquid. The red frame highlights (b) the resulting zoomed image, with individual outer membrane proteins(OMPs) circled in red. (c–d) The corresponding recognition images for (a–b), showing recognition of the S. typhimurium OMPs. In (e–f), a control experiment is performed on E. coli,demonstrating that the tip-bound aptamer is specific for OMPs of S. typhimurium.Panel C. 4D histograms were constructed to better visualize the relationship between the extension values, loading rates, unbinding force, and frequency of occurrence of the collectedforce-distance curves. The data points' color and size are both representative of frequency, with the color bar showing normalized frequency. Under this visualization, it specific binding“peaks” are readily observed by the few, large data points, interspersed among the smaller ones, which are more likely to be caused by non-specific adsorption.Reprinted from (Wang et al., 2017) under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/). (For interpretation ofthe references to color in this figure legend, the reader is referred to the Web version of this article.)

Table 1A summary of the reviewed methods.

Method Advantages Disadvantages

Cantilever sensor Simple; quick; possible to amplify signal by varying the detection scheme Relatively high LOD compared to ELISA; no imagingcapability

Recognition AFM imaging High spatial resolution; three-dimensional imaging coupled with chemicalspecificity; aids in site selection for DFS

Data not inherently quantitative; relatively slow

Dynamic forcespectroscopy

Rich quantitative data on binding thermodynamics and kinetics; capable of imagingwhen coupled with recognition AFM

No native imaging capability, which can reduce efficiency infinding binding pairs

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

7

Drake, B., Prater, C. B., Weisenhorn, A. L., Gould, S. A., Albrecht, T. R., Quate, C. F., et al.(1989). Imaging crystals, polymers, and processes in water with the atomic forcemicroscope. Science (New York, N.Y.), 243(4898), 1586–1589. https://doi.org/10.1126/science.2928794.

Ebner, A., Hinterdorfer, P., & Gruber, H. J. (2007). Comparison of different amino-functionalization strategies for attachment of single antibodies to AFM cantilevers.Ultramicroscopy, 107(10–11), 922–927. https://doi.org/10.1016/j.ultramic.2007.02.035.

Ebner, A., Kienberger, F., Kada, G., Stroh, C. M., Geretschläger, M., Kamruzzahan, A. S.M., et al. (2005). Localization of single Avidin-Biotin interactions using simultaneoustopography and molecular recognition imaging. ChemPhysChem, 6(5), 897–900.https://doi.org/10.1002/cphc.200400545.

Egawa, H., & Furusawa, K. (1999). Liposome adhesion on mica surface studied by atomicforce microscopy. Langmuir, 15(5), 1660–1666. https://doi.org/10.1021/la980923w.

Evans, E. (2001). Probing the relation between force—lifetime—and chemistry in singlemolecular bonds. Annual Review of Biophysics and Biomolecular Structure, 30(1),105–128. https://doi.org/10.1146/annurev.biophys.30.1.105.

Florin, E. L., Moy, V. T., & Gaub, H. E. (1994). Adhesion forces between individual ligand-receptor pairs. Science, 264(5157), 415–417. https://doi.org/10.1126/science.8153628.

Freed, R. C., Evenson, M. L., Reiser, R. F., & Bergdoll, M. S. (1982). Enzyme-linked im-munosorbent assay for detection of staphylococcal enterotoxins in foods. Applied andEnvironmental Microbiology, 44(6), 1349–1355. Retrieved from http://www.ncbi.nlm.nih.gov/pubmed/7159082.

Han, W., Lindsay, S. M., & Jing, T. (1996). A magnetically driven oscillating probe mi-croscope for operation in liquids. Applied Physics Letters, 69(26), 4111–4113. https://doi.org/10.1063/1.117835.

Hinterdorfer, P., Kienberger, F., Raab, A., Gruber, H. J., Baumgartner, W., Kada, G., et al.(2000). Poly(Ethylene Glycol): An ideal spacer for molecular recognition force mi-croscopy/spectroscopy. Single Molecules, 1(2), 99–103. http://dx.doi.org/10.1002/1438-5171(200006)1:2<99::AID-SIMO99>3.0.CO;2-K.

Homola, J., Dostálek, J., Chen, S., Rasooly, A., Jiang, S., & Yee, S. S. (2002). Spectralsurface plasmon resonance biosensor for detection of staphylococcal enterotoxin B inmilk. International Journal of Food Microbiology, 75(1–2), 61–69. https://doi.org/10.1016/S0168-1605(02)00010-7.

Hoorfar, J., Cook, N., Malorny, B., Wagner, M., De Medici, D., Abdulmawjood, A., et al.(2003). Making internal amplification control mandatory for diagnostic PCR. Journalof Clinical Microbiology, 41(12), 5835. https://doi.org/10.1128/JCM.41.12.5835.2003.

Hummer, G., & Szabo, A. (2001). Free energy reconstruction from nonequilibrium single-molecule pulling experiments. Proceedings of the National Academy of Sciences, 98(7),3658–3661. https://doi.org/10.1073/pnas.071034098.

Jarzynski, C. (1997). Nonequilibrium equality for free energy differences. Physical ReviewLetters, 78(14), 2690–2693. https://doi.org/10.1103/PhysRevLett.78.2690.

Li, Q., Michaelis, M., Wei, G., & Colombi Ciacchi, L. (2015). A novel aptasensor based onsingle-molecule force spectroscopy for highly sensitive detection of mercury ions. TheAnalyst, 140(15), 5243–5250. https://doi.org/10.1039/C5AN00708A.

Lin, L., Fu, Q., Williams, B. A. R., Azzaz, A. M., Shogren-Knaak, M. A., Chaput, J. C., et al.(2009). Recognition imaging of acetylated chromatin using a DNA aptamer.Biophysical Journal, 97(6), 1804–1807. https://doi.org/10.1016/j.bpj.2009.06.045.

Liphardt, J., Dumont, S., Smith, S. B., Tinoco, I. J., & Bustamante, C. (2002). Equilibriuminformation from nonequilibrium measurements in an experimental test ofJarzynski's equality. Science, 296(5574), 1832–1835. https://doi.org/10.1126/science.1071152.

Liu, W., Montana, V., Chapman, E. R., Mohideen, U., & Parpura, V. (2003). Botulinumtoxin type B micromechanosensor. Proceedings of the National Academy of Sciences ofthe United States of America, 100(23), 13621–13625. https://doi.org/10.1073/pnas.2233819100.

Li, Q., Zhang, T., Pan, Y., Ciacchi, L. C., Xu, B., & Wei, G. (2016). AFM-based forcespectroscopy for bioimaging and biosensing. RSC Advances, 6(16), 12893–12912.https://doi.org/10.1039/C5RA22841G.

Magonov, S. N., Elings, V., & Whangbo, M.-H. (1997). Phase imaging and stiffness intapping-mode atomic force microscopy. Surface Science, 375(2), L385–L391. https://doi.org/10.1016/S0039-6028(96)01591-9.

Maraldo, D., & Mutharasan, R. (2007). Detection and confirmation of staphylococcalenterotoxin B in apple juice and milk using piezoelectric-excited millimeter-sizedcantilever sensors at 2.5 fg/mL. Analytical Chemistry, 79(20), 7636–7643. https://doi.org/10.1021/ac070589l.

Mead, P. S., Slutsker, L., Dietz, V., McCaig, L. F., Bresee, J. S., Shapiro, C., et al. (1999).Food-related illness and death in the United States. Emerging Infectious Diseases, 5(5),607–625. https://doi.org/10.3201/eid0505.990502.

Meyer, G., & Amer, N. M. (1988). Novel optical approach to atomic force microscopy.Applied Physics Letters, 53(12), 1045–1047. https://doi.org/10.1063/1.100061.

Müller, D. J., & Dufrene, Y. F. (2008). Atomic force microscopy as a multifunctional

molecular toolbox in nanobiotechnology. 3(5), 261–269 https://doi.org/10.1038/nnano.2008.100.

Müller, D. J., & Dufrêne, Y. F. (2011). Atomic force microscopy: A nanoscopic window onthe cell surface. Trends in Cell Biology. https://doi.org/10.1016/j.tcb.2011.04.008.

Muller, D. J., & Engel, A. (2008). Strategies to prepare and characterize native membraneproteins and protein membranes by AFM. Current Opinion in Colloid & InterfaceScience. https://doi.org/10.1016/j.cocis.2007.09.002.

Myszka, D. G. (1999). Improving biosensor analysis. Journal of Molecular Recognition,12(5), 279–284. http://dx.doi.org/10.1002/(SICI)1099-1352(199909/10)12:5<279::AID-JMR473>3.0.CO;2–3.

Patel, P. D. (2002). (Bio)sensors for measurement of analytes implicated in food safety: Areview. TRAC Trends in Analytical Chemistry, 21(2), 96–115. https://doi.org/10.1016/S0165-9936(01)00136-4.

Rasooly, A., & Herold, K. E. (2006). Biosensors for the Analysis of food-and waterbornePathogens and their toxins. Journal of AOAC International, 89(3), 873–883. Retrievedfrom http://www.ingentaconnect.com.proxy-remote.galib.uga.edu/content/aoac/jaoac/2006/00000089/00000003/art00034.

Scallan, E., Hoekstra, R. M., Angulo, F. J., Tauxe, R. V., Widdowson, M.-A., Roy, S. L.,et al. (2011). Foodborne illness acquired in the United States—major pathogens.Emerging Infectious Diseases, 17(1), 7–15. https://doi.org/10.3201/eid1701.P11101.

Scheu, P. M., Berghof, K., & Stahl, U. (1998). Detection of pathogenic and spoilage micro-organisms in food with the polymerase chain reaction. Food Microbiology, 15(1),13–31. https://doi.org/10.1006/fmic.1997.0134.

Senapati, S., & Lindsay, S. (2016). Recent progress in molecular recognition imagingusing atomic force microscopy. Accounts of Chemical Research, 49(3), 503–510.https://doi.org/10.1021/acs.accounts.5b00533.

Sharma, S., Johnson, R. W., & Desai, T. A. (2004). XPS and AFM analysis of antifoulingPEG interfaces for microfabricated silicon biosensors. Biosensors and Bioelectronics,20(2), 227–239. https://doi.org/10.1016/j.bios.2004.01.034.

Shlyakhtenko, L. S., Gall, A. A., Filonov, A., Cerovac, Z., Lushnikov, A., & Lyubchenko, Y.L. (2003). Silatrane-based surface chemistry for immobilization of DNA, protein-DNAcomplexes and other biological materials. Ultramicroscopy, 97(1–4), 279–287.https://doi.org/10.1016/S0304-3991(03)00053-6.

Souza Silva, É. A., & Pawliszyn, J. (2012). Optimization of fiber coating structure enablesdirect immersion solid phase microextraction and high-throughput determination ofcomplex samples. Analytical Chemistry, 84(16), 6933–6938. https://doi.org/10.1021/ac301305u.

Stadtherr, K., Wolf, H., & Lindner, P. (2005). An aptamer-based protein biochip.Analytical Chemistry, 77(11), 3437–3443. https://doi.org/10.1021/ac0483421.

Stroh, C., Wang, H., Bash, R., Ashcroft, B., Nelson, J., Gruber, H., et al. (2004). Single-molecule recognition imaging microscopy. Proceedings of the National Academy ofSciences, 101(34), 12503–12507. https://doi.org/10.1073/pnas.0403538101.

Sungkanak, U., Sappat, A., Wisitsoraat, A., Promptmas, C., & Tuantranont, A. (2010).Ultrasensitive detection of Vibrio cholerae O1 using microcantilever-based biosensorwith dynamic force microscopy. Biosensors and Bioelectronics, 26(2), 784–789.https://doi.org/10.1016/j.bios.2010.06.024.

Tanaka, M., & Sackmann, E. (2005). Polymer-supported membranes as models of the cellsurface. Nature. https://doi.org/10.1038/nature04164.

Turner, A. P. F. (2013). Biosensors: Sense and sensibility. Chemical Society Reviews, 42(8),3184. https://doi.org/10.1039/c3cs35528d.

Wang, B., Guo, C., Chen, G., Park, B., & Xu, B. (2011). Following aptamer–ricin specificbinding by single molecule recognition and force spectroscopy measurements. Chem.Commun. 48(11), 1644–1646. https://doi.org/10.1039/C1CC15644F.

Wang, B., Guo, C., Zhang, M., Park, B., & Xu, B. (2012). High-resolution single-moleculerecognition imaging of the molecular details of ricin–aptamer interaction. The Journalof Physical Chemistry B, 116(17), 5316–5322. https://doi.org/10.1021/jp301765n.

Wang, B., Park, B., Kwon, Y., & Xu, B. (2014). Determining the elastic properties of ap-tamer-ricin single molecule multiple pathway interactions. Applied Physics Letters,104(19), 193702. https://doi.org/10.1063/1.4876603.

Wang, B., Park, B., Xu, B., & Kwon, Y. (2017). Label-free biosensing of Salmonella en-terica serovars at single-cell level. Journal of Nanobiotechnology, 15(1), 40. https://doi.org/10.1186/s12951-017-0273-6.

Wei, G., Steckbeck, S., Köppen, S., & Colombi Ciacchi, L. (2013). Label-free biosensingwith single-molecule force spectroscopy. Chemical Communications, 49(31), 3239.https://doi.org/10.1039/c3cc40506k.

Xu, S. (2012). Electromechanical biosensors for pathogen detection. Microscopica Acta,178(3–4), 245–260. https://doi.org/10.1007/s00604-012-0831-4.

Zeng, G., Ogaki, R., & Meyer, R. L. (2015). Non-proteinaceous bacterial adhesins chal-lenge the antifouling properties of polymer brush coatings. Acta Biomaterialia, 24,64–73. https://doi.org/10.1016/j.actbio.2015.05.037.

Zhong, Q., Inniss, D., Kjoller, K., & Elings, V. B. (1993). Fractured polymer/silica fibersurface studied by tapping mode atomic force microscopy. Surface Science Letters,290(1–2), L688–L692. https://doi.org/10.1016/0167-2584(93)90906-Y.

R. Alexander Reese, B. Xu Trends in Food Science & Technology xxx (xxxx) xxx–xxx

8