**Spectroscopy-Based Biosensors**

Editors

**Annalisa De Girolamo Vincenzo Lippolis Chris Maragos**

MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade • Manchester • Tokyo • Cluj • Tianjin

*Editors* Annalisa De Girolamo National Research Council of Italy (CNR) Italy

Vincenzo Lippolis National Research Council of Italy (CNR) Italy

Chris Maragos United States Department of Agriculture USA

*Editorial Office* MDPI St. Alban-Anlage 66 4052 Basel, Switzerland

This is a reprint of articles from the Special Issue published online in the open access journal *Biosensors* (ISSN 2079-6374) (available at: https://www.mdpi.com/journal/biosensors/special issues/spectroscopy biosensors).

For citation purposes, cite each article independently as indicated on the article page online and as indicated below:

LastName, A.A.; LastName, B.B.; LastName, C.C. Article Title. *Journal Name* **Year**, *Volume Number*, Page Range.

**ISBN 978-3-0365-1002-6 (Hbk) ISBN 978-3-0365-1003-3 (PDF)**

© 2021 by the authors. Articles in this book are Open Access and distributed under the Creative Commons Attribution (CC BY) license, which allows users to download, copy and build upon published articles, as long as the author and publisher are properly credited, which ensures maximum dissemination and a wider impact of our publications.

The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons license CC BY-NC-ND.

## **Contents**


## **About the Editors**

**Annalisa De Girolamo** is a Senior Researcher at the Institute of Sciences of Food Production, National Research Council of Italy (CNR) (Bari, Italy). She has more than twenty-years of experience in the field of food safety and food authenticity with special interests in development and validation of conventional methods as well as rapid methods based on infrared spectroscopy combined with chemometrics. She has organized several collaborative studies for the validation of analytical methods, proficiency testing for the quality control of the laboratories, and related statistical data processing according to the AOAC/IUPAC guidelines and the ISO/IUPAC standards. She has also specific competence in selection of synthetic mediators (aptamers, molecular imprinted polymers) for several target molecules. Annalisa De Girolamo is involved in several national and EU-funded projects. She is author of 57 peer reviewed publications (h-index of 25, Scopus), 2 book chapters, 2 European reports. Tutor of several PhD and degree thesis on Food Safety. Guest editor of several ISI journals. Teacher at post-graduate masters. Referee for several International Scientific Journals. Speaker at several national and international Conferences/Symposiums.

**Vincenzo Lippolis** is a Senior Researcher at the Institute of Sciences of Food Production, National Research Council of Italy (Bari, Italy). He has more than fifteen-years of experience in the field of food safety and food authenticity with specific focus on the development and validation of targeted and non-targeted analytical methods based on fluorescence polarization immunoassays, lateral flow immunoassays, surface plasmon resonance, electronic noses and mass spectrometry. Vincenzo Lippolis is involved in several national and EU-funded projects. Vincenzo Lippolis is author of more than 40 peer reviewed publications (h-index of 14, Scopus), 1 patent and 2 book chapters. Supervisor of PhD thesis on Food Safety. Tutor of degree thesis in Analytical Chemistry. Teacher at post-graduate masters. Guest editor of several ISI journals. Referee for several International Scientific Journals. Speaker at several national and international Conferences/Symposiums.

**Chris Maragos** is a Lead Scientist in the Mycotoxin Prevention and Applied Microbiology Research Unit at the National Center for Agricultural Utilization Research (Peoria, Illinois, USA), a part of the U.S. Department of Agriculture - Agricultural Research Service. He has more than thirty years of experience in the field of food safety, with particular emphasis on natural toxins. His research interests involve the development of new technologies and materials for the detection of toxic secondary metabolites of fungi (mycotoxins) in commodities and foods. He is the author of 128 peer reviewed publications and book chapters (h-index of 38, Scopus). He is a member of the Editorial Boards of the journal Food Additives and Contaminants, Mycotoxin Research, and the World Mycotoxin Journal. He has served in various capacities within the AOAC International and is a Fellow of the AOAC International.

## **Preface to "Spectroscopy-Based Biosensors"**

Biosensors are analytical devices capable of providing quantitative or semi-quantitative information by using a biological recognition element and a transducer. Depending upon the nature of the recognition element, different surface sensitive techniques can be applied to monitor these molecular interactions. In order to increase sensitivities and to lower detection limits down to even individual molecules, nanomaterials are promising candidates. This is possible due to the potential to immobilize more bioreceptor units at reduced volumes and their ability to act as transduction elements by themselves. Among such nanomaterials, gold nanoparticles, quantum dots, polymer nanoparticles, carbon nanotubes, nanodiamonds, and graphene are intensively studied. Biosensors provide rapid, real-time, accurate, and reliable information about the analyte under investigation and have been envisioned in a wide range of analytical applications, including medicine, food safety, bioprocessing, environmental/industrial monitoring, and electronics. A variety of biosensors, such as optical, spectroscopic, molecular, thermal, and piezoelectric, have been studied and applied in countless fields.

In this book, examples of spectroscopic and optical biosensors and immunoassays are presented. Furthermore, two comprehensive reviews on optical biosensors are included.

> **Annalisa De Girolamo, Vincenzo Lippolis, Chris Maragos** *Editors*

### *Review* **Silver-Based Plasmonic Nanoparticles for and Their Use in Biosensing**

**Alexis Loiseau 1, Victoire Asila 2,**†**, Gabriel Boitel-Aullen 2,**†**, Mylan Lam 2,**†**, Michèle Salmain <sup>3</sup> and Souhir Boujday 1,\***


Received: 3 May 2019; Accepted: 31 May 2019; Published: 10 June 2019

**Abstract:** The localized surface plasmon resonance (LSPR) property of metallic nanoparticles is widely exploited for chemical and biological sensing. Selective biosensing of molecules using functionalized nanoparticles has become a major research interdisciplinary area between chemistry, biology and material science. Noble metals, especially gold (Au) and silver (Ag) nanoparticles, exhibit unique and tunable plasmonic properties; the control over these metal nanostructures size and shape allows manipulating their LSPR and their response to the local environment. In this review, we will focus on Ag-based nanoparticles, a metal that has probably played the most important role in the development of the latest plasmonic applications, owing to its unique properties. We will first browse the methods for AgNPs synthesis allowing for controlled size, uniformity and shape. Ag-based biosensing is often performed with coated particles; therefore, in a second part, we will explore various coating strategies (organics, polymers, and inorganics) and their influence on coated-AgNPs properties. The third part will be devoted to the combination of gold and silver for plasmonic biosensing, in particular the use of mixed Ag and AuNPs, i.e., AgAu alloys or Ag-Au core@shell nanoparticles will be outlined. In the last part, selected examples of Ag and AgAu-based plasmonic biosensors will be presented.

**Keywords:** silver nanoparticles; synthesis; coating; alloy; core@shell; LSPR; biosensors

#### **1. Introduction**

The first use of silver (Ag) as an antimicrobial and antibacterial agent goes back to the ancient Greek and Roman Empire [1,2]. At that time, the medicinal and preservative properties of silver were mainly used to protect vessels from bacterial attacks and to make water and other liquids potable [1,3–5]. Globally, it was already known to be an efficient weapon against the growth of pathogen factors [6]. The antimicrobial effect of silver arises from the interaction of silver ions with thiol groups of vital bacterial enzymes and proteins that lead to cell death [4,5,7]. Over the past decades, silver has been engineered into nanoparticles (NPs) (at least one dimension is smaller than 100 nm) [8]. Although metallic NPs have been present in artefacts for a very long time [9,10], as attested by medieval stained glasses, and even earlier as for the Lycurgus Cup of the British Museum in London, dated from the 5th century, scientific knowledge about NPs is quite recent [11]. Synthesis of metallic NPs can be achieved according to two distinct nanofabrication methods. On the one hand, top-down approaches involve physical processes such as lithography or chemical processes controlled by external experimental parameters

to create nanoscale structures starting from larger dimensions to the nanometer range [12]. This can be achieved by milling or high pressure homogenization [13]. On the other hand, the bottom-up approaches use atoms or small molecules as the building blocks of multi-level structures to build up more complex nanoscale assemblies or directed self-assemblies that perform various operations [14]. This method is extremely valuable since it is free of waste or unused materials [15]. This can be achieved by controlled precipitation (or crystallization) and evaporation from a precursor [13,16]. Generally, top-down techniques produce NPs that are mostly crystalline but high energy or pressure is required to achieve nanometer range comminution, which may also lead to contamination if a milling medium is used. In contrast, bottom-up processes involve dissolution, followed by precipitation or drying. The mechanical energy input is thus minimal, and the resulting NPs can be crystalline or amorphous, depending on the synthesis conditions. Metallic NPs feature unique physical properties such as high ratio-to-surface area and volume. Moreover, the confinement effect confers reactivity as well as mechanical, electromagnetic, chemical and optical properties that differ from those of the bulk metals [17–19]. Indeed, particles properties change drastically at the nanometer scale [11]. Metallic NPs find applications in various fields, from catalysis to the detection of biological molecules in solution [8,20]. In the biomedical field, they can be used in drug delivery, photothermal therapy, or imaging [21–25]. In what follows, we will focus the use of NPs for biosensing applications.

Among all metallic NPs, gold (Au) and silver (Ag) nanoparticles exhibit the most interesting physical properties for biosensing [26,27]. Even if gold nanoparticles (AuNPs) remain the most studied for this application area because of their good chemical stability and biocompatibility [20,28,29], silver nanoparticles (AgNPs) offer better results in terms of sensitivity [30]. One of the most characteristic physical properties of metallic NPs is the localized surface plasmon resonance (LSPR), which is responsible for the bright color of the nanoparticle colloidal suspensions [26,27,31–34]. Indeed, AuNPs and AgNPs are specifically investigated for their optical properties thanks to their strong interactions with light [35–40]. The electrons at the surface of the metallic NPs undergo a collective oscillation when irradiated at a specific wavelength, called surface plasmon resonance (SPR), resulting in the appearance of the electromagnetic field localized on the NPs [30,31,41]. When the oscillations of the electromagnetic field of an incident electromagnetic wave are in resonance with those of the local electromagnetic field of the NPs, the LSPR phenomenon occurs, which is characterized by the resonance oscillation frequencies. Thus, LSPR is the consequence of the confinement of the electric field within a small metallic sphere whose radius is much smaller than the wavelength [10]. This property can be tuned by controlling parameters such as shape, size, uniformity and surface coating [27,31] and is often used for biosensing applications in the field of biology, biomedicine and biochemistry [42]. For this purpose, AgNPs of different shapes and sizes, from the simplest to the most sophisticated, can be readily obtained thanks to the large range of techniques now available that will be presented later in this review for a conceptual opportunity of biosensing. Owing to their plasmonic properties, metallic nanoparticles are also responsible for enhancing Raman scattering of molecules adsorbed at their surface, giving rise to the so-called surface enhanced Raman spectroscopy (SERS) [32,43], a powerful vibrational spectroscopy with impressive enhancement factors of up to 14–15 orders of magnitude [44]. It is worthy to note that this phenomenon is very different from propagative SPR or surface plasmon polariton (SPP) that occurs at the plane surface of large metallic structures, or on metallic nanowires, on which one direction is regarded as infinite. Colloidal suspensions of small spherical AuNPs (10 nm diameter) are red-colored and display an absorption band at 520 nm, while similar AgNPs are yellow and absorb around 380 nm [45]. As these two absorption bands are in the visible part of the electromagnetic spectrum, they allow a colorimetric detection of biomolecules by inducing changes in the position (and possibly in the intensity) of the LSPR band.

As stated above, the position of the LSPR band of AuNPs and AgNPs depends on their size, uniformity, shape, dispersion, composition (ratio Au:Ag), and also on the dielectric constant of the surrounding medium [9,26,30,32,43,46,47]. Therefore, modifying any of these parameters induces wavelength shifts. For example, size increase shifts the LSPR band to higher wavelength, i.e., red-shifted [47,48]; NPs aggregation also induces a red-shift. While isotropic (spherical) NPs have a unique absorption band, anisotropic ones can display several absorption bands [47,48]. Two formats are typically encountered for colorimetric and plasmonic biosensors, i.e., aggregation-based assays [49–51] and LSPR-based ones [38]. The refractive index sensitivity (RIS), expressed in nm/refractive index unit (RIU) (nanometer per refractive index unit) is a measure of the shift in wavelength of the LSPR peak: the more the peak is shifted for small variations of refractive index (RI), the more sensitive the biosensor is (i.e., the highest the sensitivity is). AgNPs are described as more sensitive than AuNPs for the second biosensing strategy [52]. Indeed, a study showed that the RIS for AgNPs and AuNPs increased from 153 to 265 nm/RIU and 128 to 233 nm/RIU, respectively, for sizes 5 to 50 nm [53]. However, combining the two metals is very attractive and offers a wide range of possibilities.

In what follows we discuss multiple strategies to produce AgNPs of various sizes and shapes that are both key factors contributing to the modulation of major optical properties. Then, we will explore different coatings (organic, polymer, inorganic) on silver nanoparticles and highlight the influence of AgNPs coating process on their fate, stability, and toxicity in a given environment as well as their properties regarding plasmonic biosensing applications. We will also cover the use of mixed Ag and Au NPs (i.e., AgAu alloy and Ag-Au core@shell structures) for optical biosensing. Finally, we will present selected examples of Ag and AgAu-based plasmonic biosensors and highlight the merit of silver-containing nanoparticles in this area.

#### **2. Engineering Silver Nanoparticles for Biosensing: Shape-Properties Correlation**

Much less effort has been put on the use of AgNPs for biosensors compared to AuNPs. Indeed, 85,570 papers are referenced in the SciFinder™ database with the keyword "gold nanoparticles" compared to 63,770 for "silver nanoparticles". Combined with the keyword "plasmonic biosensors", these numbers are reduced to 424 and 112 papers, for AuNPs and AgNPs, respectively.

The synthesis of AgNPs is achieved through chemical, physical, or biological strategies. The review written by B. Khodashenas and H. R. Ghorbani in 2015, summarizes the wide range of synthetic methods reported to date as a function of the desired nanoparticle's shape [54]. We summarize in Table 1 the main strategies employed for AgNPs engineering and the related size, shape, and applications.



#### *2.1. AgNPs Synthesis by Chemical Reduction Using Citrate and*/*or Ascorbate*

Nowadays, among the wide range of synthetic methods, the chemical reduction by the bottom-up approach is the most common method to prepare AgNPs. The reaction is performed in either an aqueous or an organic solvent. The commonly used method is inspired from the Turkevich synthesis for AuNPs [67], and was first introduced by Lee's group in 1982 [68]. It involves a silver precursor, usually an inorganic compound such as silver nitrate (AgNO3) that liberates silver ions (Ag<sup>+</sup>) in solution, then, trisodium citrate (Na3C6H5O7) reduces silver and further stabilizes the resulting AgNPs [69]. However, this method often leads to polydisperse nanoparticles and several works attempted to reduce the dispersity [69,70]. Indeed, many factors have been demonstrated to play an important role on the size, shape and color of NPs such as temperature and acidity of the solution [71]. Actually, the acidity of the solution has a strong influence on the AgNPs' shape. In 2009, Dong et al. showed that the shape of AgNPs was significantly influenced by the pH using citrate as a reducing as well as stabilizing agent (Figure 1A) [72]. They found out that the shape of AgNPs at a high pH was a mixture of spherical and rod-like particles while the predominant shapes were triangular and polygonal particles at low pH (Figure 1B) [71,72].

**Figure 1.** AgNPs synthesis using citrate: (**A**) Experimental conditions affecting the silver nanoparticle AgNPs shape and (**B**) TEM images of the AgNPs synthesized at different pH values: (i) 11.1, (ii) 8.3, (iii) 6.1 and (iv) 5.7. Adapted from [72]. Copyright (2009), American Chemical Society.

This observation was confirmed by Qin et al. in 2010 as more spherical particles were obtained when the pH was increased by using ascorbate as a reductant and citrate as a stabilizer [73]. Furthermore, they demonstrated that NP size varied as a function of pH. Indeed, AgNPs were smaller due to the increased reducing activity of citrate or ascorbic acid (AA) when the pH was increased that in turn decreased the number of nuclei [71,73]. These studies also showed that the type of reductant affects the shape of the AgNPs due to the pH-dependent redox potential and thus the pH-dependent reduction rate of the precursor (Ag+). A more precise adjustment of the equilibrium between nucleation and growth allows better control over the shape of AgNPs. The use of ascorbic acid as reducing agent tends to afford spherical particles over the pH range between 6 and 11 conversely to citrate (Figure 2).

**Figure 2.** (**A**) TEM images and (**B**) UV-vis spectra of the AgNPs prepared at pH 6.0, 7.0, 8.0, 9.0, 10.0 and 10.5 by using ascorbate as reductant. Adapted from [73]. Copyright (2010), Elsevier B.V. All rights reserved.

Another study showed that the position of the LSPR band shifted as a function of the acidity [71]. Indeed, the LSPR band intensity increased and the band blue-shifted as the pH increased. The bands became sharper, and the nanoparticles size decreased accordingly as demonstrated by the TEM images in Figure 2. Finally, citrate was also used as reducing agent to form Ag nanoshell (AgNS) on a silica sphere [74–76]. The thickness of the silver shell could be tailored by varying the number of deposition cycles. Such a weak reducing agent was a prerequisite to the growth of a silver shell, so that the Ag seeds grew only in size during the Ag reduction whereas no new nucleation centers were introduced, which ensured the minimal amount of Ag colloids in the suspension accompanying the AgNS growth.

#### *2.2. AgNPs Synthesis: Anisotropic Shapes*

Investigations on anisotropic shapes and morphologies of NPs have increased during the last decade, most often relying on the development of seed-mediated synthetic methods [77]. Such syntheses of anisotropic NPs have attracted interest because the structural, optical, electronic, magnetic, and catalytic properties are different from, and most often superior to, those of spherical NPs for a comparable size [78]. In particular, the most striking properties of anisotropic and hollow NPs lie in the appearance of a plasmon band at a longer wavelength (near-infrared region) than that of spherical NPs [79,80]. Different shapes and sizes of AgNPs can be synthesized thanks to the large range of techniques now available [72,77,81–85]. Inspired by gold nanorods synthesis, silver nanorods (AgNRs) were synthesized following a process involving the reduction of AgNO3 with sodium borohydride (NaBH4) in the presence of citrate followed by growth of seeds into NRs in the presence of AA and cetyltrimethylammonium bromide (CTAB) [86]. In 2011, Zaheer and Rafiuddin achieved the synthesis of flower-like silver NPs at room temperature by a wet chemical reduction strategy [87]. It involves the use of AA as reducing agent and CTAB. Such a shape is the result of the aggregation of small NPls and NRs. Flower-like AgNPs were used as SERS substrates and showed high sensitivity to rhodamine 6G.

Several other studies showed that it is possible to make the transition from spherical to nanowire particles through nanorods by modulating the experimental conditions such as temperature, Ag precursor concentration, pH of the solution, reducing agent (citric acid, AA, and NaBH4), and the method (chemical, physical or biological) (Figure 3) [54,56,88].

**Figure 3.** TEM images of silver nanoparticles with different shapes: (**A**) nanospheres, (**B**) nanoprisms, (**C**) nanobars and (**D**) nanowires. SEM images of (**E**) nanocubes, (**F**) pyramids, (**G**) nanorice and (**H**) nanoflowers. Adapted from [54,56]. Copyright (2009), Springer Science Business Media, LLC.

A particularly interesting morphology for the development of LSPR biosensors is that of silver nanoplates (AgNPls), in which the lateral dimensions are larger than their height, so that they possess an extremely high degree of anisotropy (Figure 4A). Although the systems used for LSPR biosensing have been mainly ordered arrays of triangular NPls (TNPls) prepared by nanosphere lithography (NSL), wet chemistry techniques are one of the most widely used methods with a tight control of size and shape [30,89,90]. The wet chemistry approaches to synthesize AgNPls are light-mediated methods that relate to the use of visible light as a driving force for the oriented attachment of preformed nanoparticles. Some chemical reduction methods are based on the reduction of Ag<sup>+</sup> on Ag seeds using a weakly reducing agent (citrate or ascorbate), in the presence of CTAB, by analogy to the well-known growth of Au nanorods [90,91]. AgNPls optical resonance can be tuned from 500 to 1100 nm by precisely controlling the plate diameter and thickness [92]. AgNPls have extremely large absorbing and scattering cross-sections across the visible and near-infrared (NIR) regions (Figure 4B) [93]; they have applications in SERS, photovoltaics, molecular detection, and photothermal-based therapies [30,89,94].

**Figure 4.** (**A**) TEM image of AgNPls. (**B**) Dispersions of Ag (a) sphere and (b–h) nanoplate colloids with different colors and corresponding UV-vis-NIR extinction spectra that reflect the ability to tune the plasmon resonance of the nanoplates across the visible near-infrared (NIR) portion of the spectrum (500–1100 nm). The nanoplate optical resonance and size are tuned according to different rounds of Ag seed growth. Adapted from [90,93]. Copyright (2008, 2013), Royal Society of Chemistry.

Nanoprisms (NPrs) seem to present a classic triangular shape, but a closer observation showed that the triangles apexes were flat. Compared to spherical nanoparticles, NPrs with flat apexes and (111) crystal planes, exhibited greater antibacterial property [95]. Ag triangular nanoparticles may be also fabricated by NSL. In fact, Haes and Duyne demonstrated a very good RIS by tuning the shape and the size, and a short-range, sensing length scale determined by the characteristic decay length of the local electromagnetic field [57]. NSL is widely used to get mono-disperse, surface-confined Ag triangular NPs. It is based on the creation of a single layer crystal nanosphere mask with a suspension containing monodisperse spherical colloids (polystyrene) onto a glass substrate [96,97]. Then, a drying step is required, and the mask is formed. After that, a film of silver material is deposited over the support and the mask is then removed by a step of sonication in an adequate solvent, as shown in Figure 5. The size of the nanotriangles is controlled by the diameter of the nanospheres deposited [97].

**Figure 5.** Diagram of the nanosphere lithography (NSL) mechanism. SEM image of topography of the triangular Ag nanoparticles fabricated by NSL. Adapted from [96]. Copyright (2011), publisher and licensee Dove Medical Press Ltd.

#### *2.3. AgNPs Synthesis: Chemical Reduction Using Unconventional Ligands*

In 2002, it was reported that silver nanocubes were synthesized with a polyol process (Figure 6A), which involves the reduction of silver thanks to the hydroxyl groups of ethylene glycol [82]. The alcohol acts both as solvent and reducing agent [54]. A capping agent is then added; generally polyvinyl pyrrolidone (PVP) whose role is to build the cubic shape of the NPs [98–100]. Later, Tao et al. (2006) found that the ratio between Ag<sup>+</sup> and the number of repeating units of PVP defined the geometry of the NPs [101]. The nanocube formation was favorable when the ratio was high. On the contrary, the nanowire geometry would have been favorable. In the same year, Siekkinen et al. found out that adding a small quantity of sodium sulfide (Na2S) or sodium hydrosulfide (NaHS) speeds up the reaction, from 16–26 min to 3–8 min (Figure 6B,C) [102].

AgNRs were synthesized by using a method called oxidation reduction growth (ORG) [64]. Firstly, a thin silver film is deposited on a silica surface with a relatively constant flow of argon gas. Then, silver oxide seeds are formed during sputtering. In the sputtering process, the temperature increased to reach 200 to 300 ◦C. Hence, silver oxide dissolved and released oxygen. It allowed silver nanorods to grow without oxygen (Figure 7).

**Figure 6.** (**A**) illustration of (I) Ag<sup>+</sup> reduction by polyol process; (II) formation of Ag clusters; (III) seed nucleation; and (IV) seed growth into nanocubes, nanorods or nanowires, and nanospheres. SEM images of Ag nanocubes synthesized by mixing AgNO3 and polyvinyl pyrrolidone (PVP) *via* polyol process: (**B**) without and (**C**) sulfide-assisted synthesis (reaction time: 45 min vs. 7 min, respectively). Adapted from [98,102]. Copyright (2004), © WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim. Copyright (2006), Elsevier B.V. All rights reserved.

**Figure 7.** (**A**) Synthesis of silver nanorods by sputtering process: oxidation reduction growth (ORG). SEM images of (**B**) Ag nuclei and (**C**) AgNRs arrays. Adapted from [64].

#### **3. Coating of Silver Nanoparticles**

There are very few studies on bare AgNPs as plasmonic biosensors. One of the reasons concerns their toxicity, even if most biosensors operate ex vivo. AgNPs toxicity was extensively described in a book published in 2019 [103]. The second reason, and most probably the major limitation for bare AgNPs use in biosensing, is their poor stability and less straightforward surface chemistry [104,105]. To overcome these limitations, AgNPs were coated by a large variety of compounds; the coating process has a marked influence on the fate, stability, and toxicity of AgNPs in a given environment [106–108]. The coating of the NPs provides electrostatic, steric, or electrosteric repulsive forces between particles, which allows them to resist aggregation phenomena [105]. In the literature, various coating methods have been explored to cover AgNPs with an organic or an inorganic shell and highlighted the interest of coating AgNPs for plasmonic biosensing applications (Table 2). Hence, both the nature of the coating reactant and the thickness of the coating layer have a decisive influence on the optical properties of the NPs. In what follows, we will present examples of AgNPs coatings and discuss their influence regarding plasmonic biosensing. Interest will be first brought to organic-coated (excluding polymers) AgNPs, then to polymer-coated ones in order to improve electrostatic, steric and electrosteric stabilization of AgNPs. Finally, a brief overview of silica coating on NPs will be made.


**Table 2.** Correlation between the coating nature and AgNPs stabilization mode.

#### *3.1. Organic Coatings*

AgNPs synthesis typically uses organic compounds to promote stabilization and prevent aggregation of the particles by adsorption or covalent attachment to the particles surface. In the literature, they are often referred to as "capping agents" when they are applied during synthesis. It was proven that they have an effect on the size and shape control of the AgNPs [98]. Therefore, the function of organic coating in the stabilization and the growth of AgNPs is clearly essential for their further properties [107]. There are different possible shapes of AgNPs including quasi-spheres, nanotubes, rods, or triangular nanoplates (TNPls) which also means different coating methods with capping agents of various chemical nature (Figure 8) [114].

**Figure 8.** Silver nanoparticles core@shell structure. Adapted from [114]. Copyright (2012), American Chemical Society.

Natural organic matter (NOM) is a quite interesting example of organic coating of AgNPs [115,116]; NOM significantly influenced the stability and the surface properties of NPs, and had in turn a direct effect on the transport and the AgNPs toxicity in aqueous systems. In 2015, Gunsolus and his co-workers used NOM to stabilize citrate- and PVP-capped AgNPs against aggregation [117]. AgNPs incubated with NOM showed higher primary extinction peak intensity, which means a larger population of monodisperse particles, and slower aggregate formation by observing the secondary extinction peak (Figure 9). However, we could find no example of use of NOM-coated AgNPs as LSPR biosensors, possibly because the NOM shell is too large (up to 150 nm in Reference [117]) and, as LSPR is a short distance effect, the molecular phenomena occurring at the NOM shell no longer affect the LSPR signal. Besides, another green method was used to synthesize organic-coated AgNPs from extracts of soap-root plant as stabilizers and manna of hedysarum plant as reducer [118].

**Figure 9.** (**A**) Illustration of natural organic matter (NOM) interactions with the surface of silver nanoparticles according to the NOM's chemical composition and the affinity of the capping agent for the AgNP surface. Colloidal stability of (**B**) citrate- and (**C**) PVP-capped AgNPs in the absence or the presence of NOM from various origins. Adapted from [117]. Copyright (2015), American Chemical Society.

Many studies investigated the use of thiol-capping agents as anchoring groups for stabilizing and protecting AgNPs [108,119,120]. The thiol-capping agents are grafted to the AgNPs surface through Ag-S chemical bonds to form the external layer suggesting a core@shell morphology with an Ag core surrounded by Ag2S-like shell. A study showed that stabilized AgNPs, by the organic thiol, allylmercaptane (AM), were synthesized with different Ag/S molar ratios in the presence of tetraoctylammonium bromide (TOAB) and NaBH4 [119] *via* modified Brust−Schriffin method [121]. It has been shown that the increase in Ag/AM ratio led to an increase of the Ag2S layer thickness, and thus larger AgNPs were obtained, while the external AM layer remained unchanged (Figure 10A) [119]. Desireddy et al. prepared ultrastable AgNPs with a uniform size from the reduction of soluble precursor, which uses a protecting shell of *p*-mercaptobenzoic acid in semi-aqueous solution in the presence of NaBH4 and a coordinating solvent. This approach showed better results regarding the stability, purity and yield in very large quantities compared to those for AuNPs, due to an efficient stabilization mechanism [108]. Another approach was used by Cheng et al. using thiol-modified metal-organic framework (MOF) [120]. Herein, MOF was used as a host matrix to obtain AgNPs by using the stabilization ability of the thiol group to prevent further aggregation (Figure 10B). By controlling the initial loading amount of silver ions in the cages of thiol-MOF, monodispersed AgNPs were encapsulated in frameworks by reducing Ag<sup>+</sup> with NaBH4, while adjusting sizes of particles from 5.3 nm to 3.9 nm, which is difficult to achieve for AgNPs because of their strong tendency to aggregate.

**Figure 10.** (**A**) Allylmercaptane-stabilized AgNPs: (i) core@shell morphology for allylmercaptane- (AM)-functionalized AgNPs through Ag-S chemical bonds to form the external layer, (ii) XPS spectra of Ag@AM with four different Ag/thiol ratios, and (iii) TEM images of AgNPs with Ag/AM molar ratio equal to 2/1 (AgNPs dimensions are 9 ± 3 nm and a population of NPs aggregated of 18 ± 6 nm. (**B**) Illustration of the encapsulation of AgNPs in thiol-modified metal-organic framework (MOF) as a host matrix. Adapted from [119,120]. Copyright (2012), American Chemical Society. Copyright (2015), Royal Society of Chemistry.

#### *3.2. Polymer Coatings*

Polymers are molecules that can adopt various conformations in solution. The chain swelling can be modulated by the temperature [122]. This aspect of polymers properties has been investigated because the main interest of polymer coating comes from steric interactions. Indeed, polymers, either grafted or adsorbed on NPs, promote dispersity in the NPs solution [123]. It has long been established that polymers with a large molecular weight and a high grafting density tend to increase the colloidal stability [124,125].

Poly(ethylene) glycol (PEG) is one of the most studied polymers as stabilizing or coating agent for NPs [123,126,127]. This neutral, hydrophilic and biocompatible polymer has been approved by the Food and Drug Administration (FDA) for biomedical and pharmaceutical applications [128]. PEG improves the AgNPs dispersity in physiological conditions by steric hindrance and prevents nanoparticles aggregation [107,115,129]. Figure 11 represents only one of many ways of PEG coating by a green method [130]. Colloidal stabilization for PEG-coated AgNPs probably occurs thanks to the presence of VdW interactions:

$$\text{Ag}^+\text{(aq)} + \text{PEG}\_{\text{(aq)}} \rightarrow \text{[Ag(PEG)]}^+\text{(aq)}$$

2 [Ag(PEG)]<sup>+</sup> (aq) + CH2OH(CHOH)4CHO → 2 [Ag(PEG)](s) + CH2OH(CHOH)4COOH

**Figure 11.** Poly(ethylene) glycol (PEG) coated method of silver nanoparticles. Inspired from [130].

Another polymer, called chitosan, is widely used to coat NPs because of its good biocompatibility [131,132]. It shows a good affinity for the Ag surface and confers a high stability and dispersibility to the AgNPs [91,133]. It also shows shape-directing properties by influencing the shape of the particles from spherical to triangular.

#### *3.3. Silica Coating*

Among all coating materials used for plasmonic nanoparticles capping, silica occupies a pro-eminent position for multiple reasons [134–136]; first, silica provides a biocompatible protective shell, tunable in thickness, preventing aggregation due to electrostatic repulsion and stable in numerous solvents; second, silica synthesis is largely mastered especially through sol-gel and/or Stöber process to achieve a nanometric control of the thickness, the porosity, and the homogeneity; lastly, the presence of silanol groups on silica surface simplify the further chemical modification to introduce various surface functionalities (e.g., COOH, CHO, NH2, or NCO) with readily available coupling agents [137,138]. The Mie theory already predicts effects of silica shell thickness on NPs optical properties in various solvents [139]. There are several ways to coat NPs with silica, among which the modified Stöber process that enables to control the shell growth over a short time period (Figure 12) [135,140,141].

**Figure 12.** Modified Stöber method for coating of AgNPs with silica.

Kobayashi, Liz-Marzán and their co-workers synthesized SiO2-coated AgNPs by sol-gel reaction of tetraethyl orthosilicate (TEOS) [140]. They observed that the shell thickness was controlled through TEOS concentration and observed an increasing red-shift of the LSPR band for thicknesses in the range 28 to 48 nm. Larger silica shell thicknesses, 57–76 nm, induced a blue-shift of the plasmon band as well as a decrease of its intensity, which means that larger silica shells promote significant scattering at shorter wavelengths. Their findings were consistent with the theoretical spectra calculated by the Mie theory. Coating of anisotropic AgNPs, e.g., triangular nanoplates (AgTNPls) is more challenging as methods for silica coating of spherical AuNPs were found to be unsuitable for triangular nanoplates [113]. Silica coating of AgTNPls was achieved through a modified Stöber approach using TEOS as the alkoxide precursor and various primers: diaminopropane priming followed by reaction with TEOS (Figure 13A) that allowed tuning the thickness of the silica shell in the range 7 to 20 nm, or mercaptopropyltrialkoxysilane (either ethoxy or methoxy, MPTES or MPTMS, respectively) priming followed by silica deposition from sodium silicate (Figure 13B). This latter method using MPTES conveyed the highest stability towards salt, while retaining RI sensitivity comparable to that of the original uncoated particles (Figure 13C).

**Figure 13.** Silica coating of Ag triangular nanoplates by (**A**) diaminopropane priming and (**B**) (3-Mercaptopropyl)triethoxysilane (MPTES) priming followed by deposition from Na2Si3O7 solution. (**C**) The silica shell using MPTES on triangular nanoplates (AgTNPls) allows withstanding salts without adversely affecting refractive index (RI) sensitivity in relation to original uncoated particles. Adapted from [113]. Copyright (2013), Elsevier Inc. All rights reserved.

The major interest of these particles is the wide tunability of the plasmonic energies which could have great attention in the development of biosensors [134]. The interest of silica coating was highlighted by Pratsinis and his group [142], who confirmed that silica coating prevented AgNPs agglomeration or flocculation then investigated of the plasmonic Ag@SiO2 NPs the toxicity against a model biological system (*Escherichia coli*) and concluded that it was blocked by coating nanosilver with a silica shell about 2 nm thick. The method used for silica coating is different from those described above as they used a flame aerosol method using hexamethyldisiloxane as silica precursor (Figure 14). To predict the LSPR biosensing performances, they measured the lambda shift upon the adsorption of bovine serum albumin (BSA). The response was improved but this improvement seems to arise from a better dispersion and therefore a higher amount of protein adsorbed, even if no experimental data was provided to confirm this hypothesis [142].

**Figure 14.** (**A**) Effect of a hermetic SiO2 coating on the flocculation and toxicity of nanosilver particles. (**B**) Illustration of the nanosilver encapsulation with a hermetic SiO2 coating using hexamethyldisiloxane as silica precursor in a flame aerosol reactor. (**C**) TEM images of the (a,b) 1.4 wt.% and (c,d) 7.8 wt.% SiO2-coated nanosilver. Adapted from [142]. Copyright (2010), WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.

To conclude this part, the benefit from silica coating is well-established, in terms of chemical and colloidal stability and reduced toxicity. The protective silica shell has a limited effect on biosensing ability as long as the thickness of the layer is limited to few nanometers. Some aspects of silica coating, for instance the porosity, were not discussed herein, but are mentioned in the relevant review cited above. Beside these inputs from silica shells, there are no amplifying or synergetic effects in the plasmonic response of the Ag@SiO2 NPs, conversely to coating or mixing with gold, discussed in what follows.

#### **4. Plasmonic Nanoparticles Based on Silver and Gold: Alloy, Core@Shell, Nanocages and Nanoshells**

Compared to the two other plasmonic nanometals, i.e., gold and copper, silver nanoparticles have a higher theoretical refractive index sensitivity (RIS) (Figure 15) [53,143]. It has been shown that RIS increased from 153 to 265 nm/refractive index unit (RIU), 128 to 233 nm/RIU, and 117 to 212 nm/RIU, respectively, for AgNPs, AuNPs, and CuNPs with sizes from 5 to 50 nm. Spherical AgNPs exhibit a stronger LSPR absorption with a peak at 400 nm, which is from five to 10 times more intense than the gold one at 520 nm [53,143]. Despite this better sensitivity, observed both experimentally and theoretically, AgNPs have several drawbacks for biosensing; in addition to their poor stability and biocompatibility, they display less sensing reversibility due to light alteration, which makes their use in repeated cycles less reliable than that of AuNPs [80].

**Figure 15.** (**A**) Calculated extinction spectra of Ag, Au and Cu spherical NPs (20 nm) in different media. (**B**) Experimental extinction spectra of Ag and Au spherical NPs (10 nm) and (**C**) experimental response expressed as LSPR band shift of biocytin-coated Ag and Au spherical NPs (10 nm) in the presence of avidin. Adapted from [53,143]. Copyright (2014), Springer Science Business Media New York. Copyright (2012), Elsevier B.V. All rights reserved.

Many improvements have come about when AgNPs were combined to other metals and particularly to gold. Of course, the benefit from the previously discussed coating was effective with gold, but in addition, a synergy between these two plasmonic metals allowed for a better efficiency. This combination was mainly done by forming AgAu alloys or Ag@Au core@shell structures (Ag@AuNPs or Au@AgNPs). The main synthesis techniques and the resulting shapes and sizes are summarized in Table 3. In what follows, we will successively discuss AgAu alloys and Ag-Au core@shell structures synthesis. We will also cover the use of AgNPs as sacrificial templates to build gold nanocages (AuNC) or gold nanoshells (AuNS) for improvements of gold plasmonic biosensors.


**Table 3.** Summary of the main Ag-Au bimetallic NPs structures: synthesis techniques, size and shape.

#### *4.1. Silver-Gold Alloy Nanoparticles*

Silver-gold alloy nanoparticles (AgAuNPs) are defined as a mixture of Ag and Au atoms, with no spatial distinction between the gold and silver parts. The chemical synthesis methods mainly consist in the co-reduction of AgNO3 and HAuCl4 with sodium citrate, which give spherical AgAuNPs [144,158]. The mole fraction of each metal in the alloy depends on the concentration of AgNO3 and HAuCl4 introduced in solution. In these conditions, small AgAuNPs (roughly 20 nm) can be synthesized [144]. Besides, simultaneous laser ablation of Ag and Au in colloidal suspension allows the synthesis of AgAuNPs, in the same range of size (Figure 16A) [145]. Another study demonstrated the AgAu alloy interdiffusion at the NPs interface, resulting in an intermediate alloy shell [148]. Indeed, a hydrothermal treatment is necessary during Ag<sup>+</sup> reduction at the surface of AuNPs for Ag diffusion in Au in order to obtain AgAuNPs. This phenomenon is dependent on the temperature, such as the growth of the Ag shell layer until the final structure: core/alloy/shell.

**Figure 16.** (**A**) Pulsed laser ablation in liquid: simultaneous ablation of Ag and Au to synthesize AgAuNPs and ablation of Au in AgNPs colloid to form core@shell structure (Au@AgNPs). Reproduced with permission from [145]. Copyright (2014,) Springer-Verlag Berlin Heidelberg. (**B**) Schematic illustration to form an intermediate AgAu alloy shell by interdiffusion at the NPs interface during hydrothermal treatment. Adapted from [148]. Copyright (2011), American Chemical Society.

AgAuNPs are also prepared by physical techniques, such as metal evaporation (electron beam), followed by thermal treatment which affords supported AgAuNPs on glass support [146,149]. Spherical or quasi-spherical AgAuNPs are obtained due to evaporation of Ag and Au layers in a vacuum chamber. Then, Au and Ag metallic atoms can be deposited on the glass substrate, following an island formation of AgAuNPs, because of the stronger interactions between Ag and Au atoms, than with the glass substrate [149]. The annealing post-treatment increases the crystallinity of AgAuNPs, but destroys completely pure AgNPs, previously synthesized in the same way [146]. In this technique, the AgAuNPs formation seems to be independent of the deposition order of the initial metallic layers on the glass substrate. Besides, in the case of the AuAgNPs elliptical formation with metal evaporation, the LSPR band shift is dependent on the shape, i.e., on the degree of sphericity of the AgAuNPs. AgAuNPs were also synthesized by UV laser radiation in the near-surface region of silicate glass [147]. Finally, nanosphere lithography allows the formation of very ordered arrays of silver-gold alloy nanoprisms (AgAuNPrs) on glass support. It has been shown that AgAuNPs are about four times more sensitive in RI than the equivalent supported spherical AgAuNPs with similar sizes and conditions. Moreover, the RIS of the AgAuNPs with xAu = 0.5 is closer to pure AgNPs, i.e., very superior to pure AuNPs RIS [150].

AgAu alloys are more stable than gold-silver core@shell nanoparticles for the same size and shape [159]. The composition and the molar ratio between the two metals are important factors to be considered regarding their properties. Indeed, the plasmon peak for spherical AgNPs and AuNPs is around 400 and 520 nm, respectively, while the absorption of AgAuNPs can be tuned continuously from 400 to 520 nm *via* changing the alloy composition [34,160,161]. Qi et al. showed that the alloy NPs became less stable when Ag molar ratio increased conversely to core@shell NPs [159]. Hence, optical properties of alloys mainly depend on the ratio of one metal compared to the other, their size and shape. Indeed, AgAuNPs present only one peak, whose intensity and position in wavelength depend on the molar fraction of Au, xAu [144,161]. In the case of spherical NPs of 18 nm average size, Link et al. demonstrated that the single LSPR peak of the AgAuNPs shifted from 400 nm (pure AgNPs) to 520 nm (pure AuNPs) according to the increasing gold molar fraction [144]. Indeed, it has theoretically predicted and experimentally observed that the LSPR band shift from 400 nm to 520 nm is linear, and proportional to the mole fraction of gold xAu. These results are shown in Figure 17. Moreover, theoretical simulations showed that the intensity of the peak decreases when the gold mole fraction increases [53,145].

**Figure 17.** (**A**) TEM image and (**B**) size histogram of spherical AgAuNPs with Au mole fraction xAu = 0.8: the average size is 18 nm. (**C**) Experimental and (**D**) calculated spectra regarding the LSPR shift of 18 nm diameter spherical AgAuNPs with varying Au molar fraction. (**E**) Colloidal suspensions of AuAgNPs with increasing Au concentration. Adapted from [92,144]. Copyright (1999), American Chemical Society. Copyright (2004), Elsevier B.V. All rights reserved.

The application of AgAuNPs to plasmonic biosensing is therefore based on the higher RIS of alloy NPs, compared to pure nanoparticles with equivalent size and shape. Indeed, BSA can be detected with AgAuNPs because a linearity between the LSPR red-shift and the concentration of BSA is observed for concentration between 0.1 and 100 ng/mL, which is better than pure AuNPs [146]. The sensitivity (i.e., the LSPR band shift) and the linearity can even be improved, with use of dopamine coated AgAuNPs.

#### *4.2. Silver and Gold Core@Shell Nanoparticles*

Silver and gold core@shell NPs are made of two spatially distinct layers, each containing a different element: a core, made of the first metal (Ag or Au), and a shell made of the second (Au or Ag, respectively). The core@shell notation places core first, thus, Ag@AuNPs refers to silver core coated by a gold shell and vice versa. The shell formation keeps the initial shape of the NPs, imposed by the metallic core, whatever isotropic or anisotropic [162]. The gold shell has essentially a protecting role on Ag core (Ag@AuNPs), to ensure the chemical stability of the previously synthesized AgNPs, and thus is often very thin. In this case, Au electrodeposition on Ag core is the main technique used with the possibility of successive voltammetric cycles to increase the thickness of the Au shell [151,152,157]. Other techniques exist, such as laser ablation of Au in a solution of Ag colloids, to generate the Ag core, on which the Au layer grows (Figure 16A) [53,145]. Here, the growth of the Au shell thickness is followed by UV-Vis spectroscopy, at different ablation times, according to the Mie theory, which makes the link between shell thickness and LSPR peak position. Chemical reduction of HAuCl4 at the AgNPls surface can also lead to very thin Au shells, by adding very slowly the gold solution [153]. The main problem in these techniques is to avoid galvanic replacement of Ag by Au, which would destroy at least partially the Ag core. Indeed, Ag is a more reductive metal than Au. Recently, our group introduced an original pathway to form Ag@AuNPs from hollow gold nanoshell (AuNS) [163]. Porous AuNS were prepared by galvanic replacement starting from AgNPs generating Ag<sup>+</sup> ions in the

process (Section 4.3 see below). Increase of pH in the presence of these AuNS triggers the reduction of Ag<sup>+</sup> that preferentially occurs at the inner walls of AuNS. The reaction initially relies on the presence of residual Ag<sup>+</sup> inside the AuNS as well as in the surrounding solution, and it proceeds upon external addition of Ag<sup>+</sup> until a solid Ag core is formed inside the AuNS to form Ag@AuNPs (Figure 18). Then, subsequent reduction of Ag<sup>+</sup> occurs on the external surface of the solidified AuNS (Ag@Au@AgNPs). Controlling the Ag content in AuNS allows tuning the LSPR band position at the desired wavelength for biosensing applications.

**Figure 18.** (**A**) Illustration of the reduction and growth process of Ag on the inner and outer surfaces of porous gold nanoshell (AuNS) with increasing amounts of Ag<sup>+</sup> in the surrounding medium. STEM elemental mapping (Ag, Au, and overlay) of AuNS obtained after adding Ag+: (**B**) [Ag+] = 0.16 mM and the corresponding elemental profile along the white hatched line and (**C**) [Ag+] = 0.32 mM and the corresponding elemental profile along the white hatched line. The black ellipse in (**C**) highlights the reduction and growth of Ag at the external surface once the inner volume is completely filled. Adapted from [163]. Copyright (2019), American Chemical Society.

Regarding the synthesis of Au@AgNPs, the chemical reduction of AgNO3 at the AuNPs surface is the main technique used. This requires a reducing agent, which is very often citrate/ascorbate, to form spherical and rod-shaped core@shell structures by Ag chemical deposition on Au core [154–156,161,162]. Besides, a very thin Au layer can be electrodeposited after the Au@AgNPs synthesis, making a peculiar structure, called Au@Ag@AuNPs [157]. Liz-Marzán et al. also realized successive reduction of AgNO3 and HAuCl4 in the presence of AA and CTAB on preformed Au seeds to obtain multilayer bimetallic nanoparticles (Au@Ag, Au@Ag@Au, and Au@Ag@Au@Ag NPs) [164]. According to the Mie theory, for isotropic nanoparticles due to the hybridization between two different plasmonic nanoparticles, the LSPR spectrum should display two peaks, one coming from the core@shell interface between the two metals, and the other one coming from the surface of the shell [53,165]. The position of the former peak mainly depends on the core metal, while the position of the latter one mainly depends on the shell metal, but also on the thickness of the shell layer. As described previously, pure AgNPs have an LSPR extinction peak around 400 nm, and more intense than pure AuNPs. It can be expected that the optimal LSPR properties should occur for very thin Au shells regarding Ag@AuNPs, whereas the Ag shell can be thicker for Au@Ag NPs. Indeed, Zhu et al. demonstrated that the peak of Ag@Au nanowires red-shifted and the intensity decreased when the Au shell thickness increased. In addition, the shell peak is almost inexistent for low Au shell thicknesses. While the peak blue-shifted when the thickness of Ag shell was increased, and the shell peak was more intense for Au@Ag nanowires [166].

Besides, the anisotropic core@shell NPs such as NRs, for which the synthesis was widely described [161,162], the peaks coming from the core@shell interface and the shell surface were enhanced because of the presence of various favored directions, but not all the resulting peaks were always observable. Indeed, four peaks should be observed for Au@AgNRs but only three were actually observed because the two initial peaks were split due to the presence of two favored directions [155]. Moreover, only two remained observable when Ag shell thickness increased. These two peaks corresponded to the longitudinal resonances of the Ag external shell for the shorter wavelength, and Au-Ag interface for the higher wavelength, as in the case of spherical Au@AgNPs [155,167]. The most intense peak observed corresponding to the longitudinal resonance of the Au-Ag interface blue-shifted when the Ag shell thickness increased [156,168]. However, no linearity was observed for Au@AgNRs between the LSPR shift and the Ag shell thickness from 740 nm (Ag shell: 0 nm) to 507 nm (Ag shell: 6 nm), based on the calculated spectra [154]. Figure 19 shows the results in terms of LSPR band position for different Ag shell thickness.

**Figure 19.** (**A**) TEM image of Au@Ag nanorods with 60 × 20 nm dimensions for the Au core and 4 nm thickness for the Ag shell and (**B**) variation of the extinction spectra of Au@Ag nanorods with varying Ag shell thickness (0–8 nm) on the 60 × 20 nm Au core. (**C**) The variations in the calculated LSPR spectra of Au@AgNRs with varying Ag shell thickness (0–6 nm) as well as the zoom on the spectrum allowed seeing the peak corresponding to the Au-Ag interface transversal resonance. Adapted from [154,156]. Copyright (2014), American Scientific Publishers. Copyright (2014), Springer-Verlag Wien.

Several studies pointed out the interesting properties of the core@shell structures based on Ag and Au for LSPR biosensing, because of their high RIS [155]. The LSPR band red-shift is observed when the surrounding RI increases for both Ag@Au and Au@Ag structures [152,156]. Moreover, considering one LSPR peak, the shift in wavelength is proportional to the RI, which is useful for detection of biomolecules in solution [157]. The RIS of core@shell NPs is also dependent on the size and the shape: anisotropic core@shell nanoparticles (NRs, TNPls, etc.) are more sensitive [38] than alloy NPs [169] or anisotropic and isotropic pure NPs [156]. As it has already been described, in the extinction spectrum, there is both an influence of the core/shell interface (depending on the two metals, as in alloys) and shell thickness (depending on the shell metal, as in pure NPs). Additionally, it is possible to deposit a dielectric layer at the surface of Au@Ag, to improve the whole stability as Ag is less stable than Au [170]. In this case, it has been shown that the LSPR sensitivity is not lost, but it can even be raised. Indeed, the LSPR band position increases when the permittivity of the layer is higher than the permittivity of the surrounding medium. The interesting LSPR properties of the core are kept if the protecting layer, which improves stability, is very thin, compared to the core size [151]. Moreover, Dong and his co-workers showed that a certain number of cycles is required for the best efficiency, both for the homogeneity in size and shape and for the RIS in the case of successive Au deposition on Ag core with voltammetric cycles [152,157]. A similar observation was realized for the chemical reduction of Ag on Au, for which a certain quantity of AgNO3 is required for the best sensitivity in RI [156]. Figure 20 shows the linearity between LSPR peak shift and RI, as well as the higher sensitivity of Au@AgNRs compared to Au@AgNPls. A study has also showed Au@Ag core–shell nanorods have better SERS responses, compared AuNRs [154]. Indeed, the SERS intensities increased with the increase of the Ag shell thickness, which demonstrates that the composition and morphology of NPs play key roles on the SERS signals.

**Figure 20.** (**A**) Localized surface plasmon resonance (LSPR) band shifts for Ag@Au hemispherical nanoplates (40 nm radius) supported on ITO glass with increasing RI media (a: air, b: water, c: ethanol, d: cyclohexane, e: carbon tetrachloride), and the linear relation between shift and RI (inset). (**B**) Evolution of the RIS with the number of Au shell electrodeposition cycles on the Ag core for the previous Ag@Au nanoplates. (**C**) The evolution of the RIS with the concentration of AgNO3 for the deposition of the Ag shell on the 20 × 60 nm Au core in the case of Au@AgNRs (TEM image was previously shown in Figure 19A). Adapted from [152,156]. Copyright (2013), American Chemical Society. Copyright (2014), Springer-Verlag Wien.

#### *4.3. Destructive Use of Silver Nanoparticles with Gold*

AgNPs can be used as sacrificial template in the destructive way for the synthesis of gold nanobowls [171], gold nanocages (AuNC) [172,173] or nanoshells (AuNS) (Figure 21A) [79,163,174,175]. Indeed, these structures are synthesized by a galvanic replacement reaction, where the metallic salt with higher reduction potential is added to a suspension containing a metal nanoparticle with lower reduction potential, as the following reaction [174,176]:

$$\text{Au}^{3+} \text{(aq)} + 3\text{ Ag} \text{(N} \text{s}) \rightarrow 3\text{ Ag}^+ + \text{Au} \text{(NC} \text{/} \text{N} \text{s})$$

Specifically, the standard reduction potential of Au<sup>3</sup>+/Au redox pair is 0.99 V vs. the standard hydrogen electrode (SHE), whereas the standard reduction potential of Ag+/Ag is 0.80 V vs. SHE [163]. The difference in reduction potential causes Au to be deposited on the Ag template upon release of Ag<sup>+</sup> into the solution. This method allows designing the shape of the NC or NS as the complementary shape of AgNPs. The nanoshell geometry is ideal for tuning and optimizing the near-field response for SERS on substrates and optical resonance properties of biosensors (Figure 21B) [76,79,174,175]. Tuning the LSPR band of nanoshells into the NIR spectral range leads to a variety of bioapplications.

Regarding the AuNC, the synthesis is done in two steps. The first one is the AgNPs synthesis by electrodeposition on a glass support. Then, the second step consists in the galvanic replacement of Ag by Au, at the AgNPs surface, which can be followed by UV-Vis, using the variation of intensity of the LSPR Au (increasing) and Ag (decreasing) bands, and by cyclic voltammetry. For complete removal of Ag, i.e., to complete gold-silver dealloying, it is necessary to use oxidizing agents such as nitric acid (HNO3) or hydrogen peroxide (H2O2) [173]. Another way to synthesize AuNC is to start from Ag disks deposited on a glass with colloidal lithography, followed by galvanic replacement of Ag by Au to create peculiar AuNC, which are AuAg nanobowls [171].

**Figure 21.** (**A**) The structural evolution of AuAg nanostructures during the galvanic replacement reaction upon addition of HAuCl4 and (**B**) absorption spectra evolution as a function of time of AgNPs titrated with increasing volumes of HAuCl4 to form AuNS: the LSPR band gradually shifts through the whole visible spectrum toward NIR wavelengths. Adapted from [175]. Copyright (2018), American Chemical Society.

#### **5. Selected Applications of Ag and AgAu-Based Plasmonic Nanoparticles in Optical Biosensing**

Few studies have reported the use of Ag and AgAu-based plasmonic nanoparticles for biosensor applications. In this last part, selected examples of the use of Ag and Ag-Au nanoparticles in the development of plasmonic biosensors based on biomolecules recognition will be detailed.

#### *5.1. RI-Based LSPR Biosensors*

Although AgNPs have been used less extensively than AuNPs in the development of biosensors, very interesting works have been published in LSPR optical biosensing. Indeed, a study showed the use of AgNPs exhibited better results for RI-based LSPR biosensing compared to AuNPs, as discussed previously in the manuscript. The LSPR band shift resulting from the addition of biocytin-coated metallic nanospheres by addition of avidin was approx. 5 times higher for AgNPs than for AuNPs, 1.78 nm/nM vs. 10.18 nm/nM for AuNPs and AgNPs, respectively (Figure 15C) [143]. Another study investigated the development of Ag triangular plasmonic NPs on glass substrate, fabricated by NSL, to lead to sensitive and selective nanoscale affinity biosensors for the streptavidin-biotin couple [57]; the limit of detection (LOD) for these LSPR biosensors was found to be in the low-picomolar to high-femtomolar region (Figure 22A–C). A method to amplify the wavelength shift, previously observed, from LSPR bioassays was optimized using Au nanoparticle-labeled antibiotin antibodies. After binding an antigen to the antibody-conjugated Ag nanotriangles, a secondary antibody attached to AuNPs was added. The resulting plasmonic coupling between the Ag nanotriangles and the Au colloids reduced the LOD by three orders of magnitude for more sensitive detection (Figure 22D–F) [177].

**Figure 22.** Silver triangular nanoparticles fabricated by NSL on a glass substrate. (**A**) Tapping mode AFM image of the Ag triangular NPs. (**B**) Surface chemistry of the Ag nanobiosensor. A mixed monolayer of (1) 11-MUA and (2) 1-OT is formed on the exposed surfaces of the AgNPs followed by the covalent linking of (3) biotin to the carboxyl groups of (1) 11-MUA. Schematic illustration of (**C**) streptavidin binding to a biotinylated Ag nanobiosensor and (**D**) biotin covalently linked to the Ag nanobiosensor surface while antibiotin-labeled AuNPs are subsequently exposed to the surface. LSPR spectra (**E**) before (solid black) and after (dashed blue) binding of native antibiotin and (**F**) before (solid black) and after (dashed red) binding of antibiotin-labeled NPs. Adapted from [57,177]. Copyright (2002, 2011), American Chemical Society.

To enhance the sensitivity of the LSPR optical sensor, a new and recent approach used by depositing NPs on an optical fiber. The principle of LSPR optical fiber sensors is also based on the plasmon resonance of metal NPs, but coated on optical fiber surfaces, that are more sensitive to changes in the surrounding medium [178,179]. The label-free and real time detection proposed by this technology is a valuable asset compared to classical techniques. However, there are few studies regarding the development of LSPR optical fiber sensors with AgNPs, although nanosilver films have been proven to be much more sensitive to surrounding medium changes than other metal films [180]. Among these studies, Chen et al. proposed a stable and sensitive reflective LSPR optical fiber sensor based on AgNPs to optimize the fabrication process, including two key parameters (the sensing length and the coating time) [179]. The surface of AgNPs deposited on the optical fibers was then functionalized with an antibody and the antigen-antibody binding process was optically monitored by measuring the wavelength shift in real time (Figure 23). This technique gave a RIS of 387 nm/RIU, which is much higher than that reported for colloidal suspension of AgNPs. Another study found a RIS of 67.6 nm/RIU by photodepositing of AgNPs on the optical fiber end [178]. The sensor response is such that the LSPR peak wavelength is linearly shifted to longer wavelength as the RI is increased.

**Figure 23.** (**A**) Schematic illustration of the experimental set-up used for the LSPR optical fiber sensor. (**B**) SEM image of immobilized AgNPs on optical fiber surface. (**C**) Illustration of the employed strategy for the development of LSPR optical fiber biosensors based on AgNPs. Adapted from [179].

Besides, Patora and Astilean developed LSPR biosensors based on chitosan-coated AgNPs to devise a multi responsive plasmonic sensor [91]. They exploited the anisotropic AgNPs as LSPR chemosensors and *p*-ATP as the target. They showed a gear of plasmon resonance peak, which allows a greater shift toward higher wavelengths. In this same study, chitosan-coated NPs were also used as LSPR sensors for monitoring trace amount of adenine by shifts of LSPR bands proving the binding between the particles and adenine, showing that, the chitosan coated AgNPs make sensitive LSPR sensors and good SERS substrates.

The applications of core@shell NPs combining Ag and Au for RI-based LSPR biosensors are also presented. A study showed that the optical properties of the Au@Ag core@shell NPs were similar to those of pure AgNPs for a given sizes, which was confirmed by means of Mie extinction calculations, while the SERS properties of Au@AgNPs exhibited a higher efficiency than AgNPs under near-infrared excitation [181]. Moreover, the results of three studies focusing on the detection of streptavidin (SA) in solution are briefly discussed below, as a model of optical biosensor, based on the RIS of glass-supported core@shell NPs, and using the receptor-analyte recognition (Figure 24A,B). Biotin, which interacts strongly with the SA target molecule, is fixed on the external shell of the core@shell NPs, which was previously amine-functionalized with 3-aminopropyltrimethoxysilane (APTMS). Two studies used Au@Ag structures, and one Ag@Au structure. Indeed, the SA detection in solution was proven with the use of Au@AgNRs [156], Au@Ag triangular nanoprisms (TNPrs) and Au@Ag@AuTNPrs [157] as well as Ag@Au hemispherical NPls [152]. The results are very comparable: successive shifts in the LSPR peak are observed upon successive additions of APTMS, biotin and SA, in correlation with the induced changes in RI. In addition, linearity is always observed between the LSPR peak shift and SA concentration, as a result of change of local RI. Figure 24C–H shows that the results for SA detection are very similar for Ag@Au and Au@Ag structures, except that the position of the main LSPR peak is red-shifted with Au@Ag core@shell. The Au@AgNRs being more sensitive to RI change, are also more sensitive with respect to SA detection [156] than NPls. Besides, the Au@AgTNPrs have been further coated with a very thin layer of Au. The resulting (Au@Ag@AuTNPrs) keep the initial sensitivity properties of the Au@AgTNPrs, and the linearity between LSPR peak shift and SA concentration [157]. Regarding Ag@Au hemispherical NPls, a complementary study on the biodetection of immunoglobulin G with anti-immunoglobulin G bound Ag@Au NPs, showed similar results to those with SA [152].

**Figure 24.** (**A**,**B**) Schematic illustration showing the preparation of glass-supported core@shell NPs for SA biosensing. (**C**) SEM image of Ag@Au hemispherical nanoplates supported on ITO glass. (**D**) LSPR peak (500 nm) was shifted upon successive treatments with APTMS, biotin and SA. (**E**) Relationship between the LSPR band shift and SA concentration for Ag@Au NPls. (**F**) SEM image of Au@AgTNPls supported on ITO glass. (**G**) LSPR peak (700 nm) shifted upon successive treatment with APTMS, biotin and SA. (**H**) Linear relationship between the LSPR band shift and SA concentration. Adapted from [152,157]. Copyright (2013), American Chemical Society. Copyright (2013), Springer Science Business Media New York.

Finally, the fabrication of substrate-bound AuAg nanobowls arrays synthesized through the galvanic replacement of silver disk arrays is used for size-selective LSPR biosensors. This sensor should prove useful in both size determination and differentiation of large analytes in biological solutions, such as viruses, fungi, and bacterial cells. In these devices, both the LSPR and the SERS signals are enhanced, and the LSPR peak is red-shifted, when the target analyte is small enough to penetrate inside AuAg nanobowls. Otherwise, the previous described effects on LSPR and SERS are not observed [171]. Therefore, this concept was applied towards the detection of a 95 nm H1N1 virus, where the larger diameter nanobowls showed an increased plasmonic response upon addition of the virus.

#### *5.2. Ag and Mixed AgAu Nanoparticle-Based Colorimetric Biosensors*

This approach received considerable attention in the analytical field for naked-eye detection due to its simplicity and low cost, it does not require any expensive or complex instrumentation. Due to these inherent optical properties, colloidal suspensions of AgNPs and mixed Ag and Au NPs have high extinction coefficients and different colors in the visible region of the spectrum when they are dispersed in comparison with when they are aggregated.

#### 5.2.1. Ag Nanoparticles Aggregation-Based Colorimetric Assays

In literature, the development of nanoparticle aggregation based-colorimetric assays has been reported. The optical plasmon properties of AgNPs depend strongly on the interparticle distance between pairs of NPs, small or large aggregates of AgNPs as compared to individual and well-spaced NPs. A decrease in the interparticle distance leads to a strong overlap between the plasmon fields of the nearby particles, causing a redshift in the LSPR band with an increase in intensity and an easily observable change in color solution. Indeed, the analytical performance with high sensitivities because of the strong LSPR and excellent selectivity driven by the interaction between analyte-NPs and its surroundings involving mainly electrostatic and hydrogen bond interactions as well as donor–acceptor chemical reactions. Therefore, a well-designed chemical interaction could lead to a change of color for naked-eye detection of the target analyte [182]. AgNPs-based colorimetric assays have been investigated for melamine detection [183,184]. Han et al. used *p*-nitroaniline (*p*-NA)-modified AgNPs, as a sensitive, selective and simple colorimetric assay, resulting in a color change from yellow to blue in the presence of melamine [183]. This optical method was highly reproducible and concentrations as low as 0.1 ppm of melamine in infant formula can be visualized by the naked-eye. The same strategy was proposed by Ma et al. with dopamine-stabilized AgNPs to detect visually the melamine [184]. Indeed, the color change of the suspension turned from yellow to brown (Figure 25). The results showed the concentrations of detectable melamine were in the range of 10 ppb to 1.26 ppm. AgNPs functionalization and the analysis of melamine can occur in one-step since *p*-NA or dopamine acts as a reducer and stabilizer of AgNPs and as a linker of melamine molecules. Other biomolecules has also been detected by a colorimetric sensor highly selective such as tryptophan [185] or histidine [186] thanks to organic coating on AgNPs: 4.4-bipyridine (4-DPD)- and *p*-sulfonatocalixarene (*p*-SC4)-modified AgNPs, respectively. In the first case, tryptophan interacts with the pyridine ring of 4-DPD *via* π-π interactions, and meanwhile carboxyl acid of tryptophan can also form hydrogen bonds with pyridine, which results in 4-DPD-functionalized AgNPs aggregation and the color change from yellow to red. The LOD was 20 μM for the tryptophan colorimetric detection [185]. In the second case, the aggregation process is due to *p*-SC4, which possesses an electron-rich cyclic cavity, being able to fit imidazole and the side chain of histidine *via* host–guest electrostatic and cation-π interactions. The color change turned from yellow to red and the LOD of 5 μM was obtained [186]. Besides, a derivative of calixarene has also been employed for colorimetric detection of pesticides in water [187].

**Figure 25.** (**A**) Principle of nanoparticle aggregation based-colorimetric assay for the melamine detection with dopamine-modified AgNPs. (**B**) UV-Vis spectra of dopamine-stabilized AgNPs suspensions with different melamine concentrations: (1) 0 mM, (2) 0.08 mM, (3) 0.4 mM, (4) 2 mM, (5) 8 mM and (6) 10 mM. Adapted from [184]. Copyright (2011), Royal Society of Chemistry.

#### 5.2.2. Mixed AgAu Nanoparticles-Based Colorimetric Assays

A colorimetric biosensor with naked-eye detection was designed in which target DNA was indirectly detected through reduction of Ag<sup>+</sup> to AgNPs at the surface of gold nanostars coated by capture probe (Figure 26) [188]. In the presence of target DNA, biotin-labeled complementary oligonucleotide H1, oligonucleotide H2 and avidin-alkaline phosphatase conjugate, ascorbic acid 2-phosphate is converted into AA, which acts as reducer. The resulting observation is a blue-shift of the LSPR peak spectrum, due to the formation of Ag shell on gold nanostars. This technique showed a detection range from 10 fM to 50 pM DNA with a detection limit of 2.6 fM.

**Figure 26.** DNA detection by naked-eye readout with the silver reduction on gold nanostars [188]. Copyright (2015), Elsevier B.V. All rights reserved.

A more recent technique based on non-aggregated Au@Ag core@shell NPs was developed by Mao et al. to detect drugs, such as cocaine, using the coloration of the solution containing NPs [189]. For this purpose, they have used Au@Ag nanoparticles coated with a DNA aptamer specific of cocaine and magnetic beads coated with a DNA sequence, partially complementary to the aptamer sequence allowing cross-linking of Au@Ag nanoparticles and magnetic beads. In the presence of a magnetic field, the nanoparticles leave the suspension with the magnetic beads, and the solution is slightly colored. When cocaine is added to the solution, it interacts with the aptamer, destroying the link between nanoparticles and magnetic beads, allowing the coloration of the solution by the nanoparticles (Figure 27). This study shows the new major improvements in the silver plasmonic biosensing.

**Figure 27.** Schematic illustration of the preparation of the (**A**) reporter probe and (**B**) capture probe as well as the principle of the colorimetric detection of illicit drug based on non-aggregation Au@Ag core@shell NPs. Reproduced with permission from [189]. Copyright (2017), Published by Elsevier B.V.

#### *5.3. Metal-Enhanced Fluorescence (MEF)-Based Biosensors*

Although fluorescence is a highly sensitive technique, where single molecules can readily be recognized, the detection of a fluorophore is usually limited by its quantum yield, auto-fluorescence of the samples and/or the photo-stability of the fluorophores. However, the use of metallic nanostructures such as silver allows modifying favorably the spectral properties of fluorophores and reducing some of these fluorophore disadvantages for metal-enhanced fluorescence (MEF) [190,191]. An interesting study has reported the use of a relatively facile deposition of AgNPs onto glass slides (i.e., silver island films, SIFs). This sensor was used for the development of an enhanced detection limit sandwich-format immunoassay for the cardiac marker myoglobin (Figure 28A) [191]. Indeed, the SIFs and glass surfaces were coated with anti-myoglobin antibodies, and then incubated with fluorophore-labeled anti-myoglobin antibodies. This approach of metal-enhanced planar immunoassay showed a 10–15-fold increase in fluorescence emission observed on the SIFs compared to that naked substrate (Figure 28B) and the results demonstrated the myoglobin concentrations were detected in the 10–1000 ng/mL range (Figure 28C).

**Figure 28.** (**A**) Illustration of a metal-enhanced sandwich immunoassay on silver island films (SIFs). Fluorescence emission of the Rhodamine Red-X-labeled anti-myoglobin antibody attached to the surface-immobilized myoglobin (**B**) for a given myoglobin concentration (100 ng/mL) on SIFs and on glass, and (**C**) at different myoglobin concentrations on SIFs. Adapted from [191]. Copyright (2005), Elsevier Ltd. All rights reserved.

AgNPs functionalized with various thiolates have been also explored for MEF-based biosensors [191]. Indeed, DNA hybridization assays using metal-enhanced fluorescence (MEF) were investigated with thiolated oligonucleotides, which were bound to AgNPs on a glass substrate. This approach suggested the use of AgNPs improved the sensitivity of DNA detection with an increase in the number of detected photons per fluorophore molecule by a factor of 10-fold or more [192]. In addition, another study has also used thiol-organic monolayer-protected AgNPs, which were displaced by oligonucleotides through ligand exchanges, and a fluorophore-labeled complementary oligonucleotide were employed for DNA hybridization. The results showed a possible approach to DNA detection with a surface-enhanced emission after hybridization in the presence of AgNPs based on the aggregation of AgNPs bound by fluorophore-labeled oligonucleotides [193]. Finally, Ag@SiO2 NPs have been also exploited as transducers of DNA hybridization [141], or to achieve MEF-based biosensor [194,195]. In the latter case, 3- to 5-fold enhanced fluorescence signals can be obtained from SiO2-coated AgNPs colloids labeled with cyanine and by their aggregation in suspension. This inert coating reduces the close proximity quenching by noble metals, as well as provides for a wide variety of chemistries for biomolecule attachment. AgNPs@SiO2 can thus become a solution-based enhanced-fluorescence sensing platform [190,191,194].

#### *5.4. Optical Biosensors Based on the Oxidation of Ag*

The oxidation of silver from AgNPs and mixed AgAu NPs, even in destructive way, can be exploited for optical biosensors. Indeed, a work has used the highly reactive properties of H2O2 to modify the nanoparticle shape for improved detection by colorimetric visualization. Xia et al. used the enzyme glucose oxidase mixed in solution with Ag nanoprisms to catalyze the reaction between glucose and oxygen to form H2O2 and gluconic acid. As the reactive H2O2 etched the tips of the nanoprisms, drastic shape and color changes were observed in the LSPR spectrum, resulting in a detection range from 0.2 μM to 100 μM in diluted serum (Figure 29) [196].

**Figure 29.** (**A**) Illustration of the strategy to modify the AgNPs shape from nanoprisms to nanodiscs through Ag oxidation for colorimetric sensing of glucose. SEM images of the Ag nanoprisms before and after incubation with glucose oxidase and glucose (100 μM) for 60 min are also showed. (**B**) Absorption spectra of the Ag nanoprisms after glucose incubation in various concentrations for 40 min with photographs of the corresponding suspensions. Adapted from [196]. Copyright (2013), American Chemical Society.

Furthermore, the destructive use of AgNPs is described by Ag oxidation for the synthesis of gold nanocages (AuNC) as well as hollow gold nanoshells (AuNS). First, the dissolution of the Ag part of the NPs is followed by the changes in the initial UV-Vis spectrum, either the shift of the peaks, or the changes in absorbance allow quantifying the presence of the biomolecule [197]. Several studies also demonstrated that glucose can be detected by this technique, using gold-silver nanoshells (AuAgNS) or Au@AgNPs [198,199]. In both cases, Ag is oxidized to Ag<sup>+</sup> because of H2O2 produced (Figure 30A) from glucose oxidation to gluconic acid in the presence of dioxygen (O2) and glucose oxidase. This strategy is based on the quantification of the amount of metallic Ag oxidized, then the quantity of H2O2 is determined as well as the quantity of glucose. In 2012, a study on Ag oxidation in presence of glucose with dioxygen and glucose oxidase in AuAgNS showed that the LSPR peak is red-shifted with increasing glucose concentration (Figure 30B) [198]. There is linearity between the wavelength shift and the glucose concentration at very low concentrations (down to 20 μM). In another study, the same oxidation process of Ag in presence of glucose with O2 was observed in Au@Ag core@shell structures. The LSPR band intensity, at fixed wavelength was correlated to glucose concentration. In this case, a linearity was observed between absorbance and glucose concentration for higher glucose concentrations (down to 400 μM), and a similar observation has been made for cholesterol detection, using the same method of Ag oxidation [199].

**Figure 30.** (**A**) Illustration showing the oxidation of Ag shell on Au@Ag core@shell NPs to Ag<sup>+</sup> in the presence of H2O2 produced by an enzymatic reaction. (**B**) Schematic illustration showing the glucose sensing mechanism with Ag/Au nanoshells. Adapted from [198]. Copyright (2012) WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.

#### **6. Conclusions**

The interest of silver nanoparticles as highly sensitive materials for plasmonic biosensors design is well-established. Indeed, although AgNPs are less chemically stable and less biocompatible compared to AuNPs, they provide more sensitive plasmonic biosensors owing to their LSPR features. The AgNPs synthesis is now well mastered and well described allowing the fabrication of differently shaped particles from the simplest to special uncommon shapes thanks to the large range of synthesis techniques now available and described in this manuscript for a conceptual opportunity in biosensing. This is a real advantage to explore many more properties. The coating, either organic or inorganic, overcomes the issues of stability and toxicity raised above and allows for the use of the resulting core@shell nanoparticles in plasmonic biosensing. Finally, the use of gold with silver nanoparticles, in alloy and core@shell structures, also provides a protective shell but in addition, enhances the plasmonic response of the resulting colloids.

Most of these findings are recent, and this may explain the few biosensing applications based on AgNPs compared to AuNPs to date. We expect growing interest to the application of these nano-objects in biosensing field, thanks to their higher RIS that allows for a better sensitivity when the strategies are based on the shift of the LSPR band. They are also very promising in naked-eye detection strategies, where multiple scenarios can be envisioned including aggregation, visible lambda shift, or even destruction of a silver shell on a gold core. In this review, the majority of selected applications of Ag and mixed AgAu nanoparticles-based plasmonic biosensors represents only trivial biosensing schemes to emphasize the merit of Ag-related NPs and provide the future prospects silver-based plasmonic nanoparticles in biosensing. In such an application, the expectations for an on-site biosensor, i.e., sensitive, reliable, fast, and user-friendly, are completely fulfilled.

**Author Contributions:** Conceptualization, S.B. initiated and directed the project—V.A., G.B.-A., and M.L. equally contributed to the original draft preparation, A.L. and S.B. reviewed and edited the original draft. A.L. wrote the manuscript starting from the original draft. S.B. and M.S. supervised and corrected the review. All authors contributed to the final manuscript.

**Funding:** This research was funded ANR-FWF grant "NanoBioSensor", ANR-15-CE29-0026-02 and by Idex Sorbonne Universities, Form@Innov.

**Acknowledgments:** The authors thank Sorbonne University and the Master of Chemistry for supporting this initiative.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**


#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Review* **Bio-Recognition in Spectroscopy-Based Biosensors for** *\*Heavy Metals***-Water and Waterborne Contamination Analysis**

#### **Alessandra Aloisi 1,**†**, Antonio Della Torre 1,**†**, Angelantonio De Benedetto <sup>2</sup> and Rosaria Rinaldi 1,2,3,\***


Received: 12 June 2019; Accepted: 25 July 2019; Published: 30 July 2019

**Abstract:** Microsystems and biomolecules integration as well multiplexing determinations are key aspects of sensing devices in the field of heavy metal contamination monitoring. The present review collects the most relevant information about optical biosensors development in the last decade. Focus is put on analytical characteristics and applications that are dependent on: (i) Signal transduction method (luminescence, colorimetry, evanescent wave (EW), surface-enhanced Raman spectroscopy (SERS), Förster resonance energy transfer (FRET), surface plasmon resonance (SPR); (ii) biorecognition molecules employed (proteins, nucleic acids, aptamers, and enzymes). The biosensing systems applied (or applicable) to water and milk samples will be considered for a comparative analysis, with an emphasis on water as the primary source of possible contamination along the food chain.

**Keywords:** water pollution; environmental water; drinking water; milk; heavy metal ions; biosensor; detection limits; optical spectroscopy; proteins; functional nucleic acids

#### **1. Introduction**

Biosensors are currently valid tools, other than laboratory analytical instrumentation, for monitoring the quality of natural water (e.g., in the food production chain) [1]. Biosensors are not meant to take over standard analytical methods, but, when optimal features of a sensing device are met, they offer remarkable advantages over conventional techniques. Overall, in certain conditions, their promptness and low-cost manufacturing make them useful tools to analyze many samples for primary warnings. As defined by the International Union of Pure and Applied Chemistry (IUPAC), "a biosensor is an integrated receptor ± transducer device, capable of providing selective analytical information using a biological recognition element" [2]. Optical biosensors are a group of sensors in which (i) the transducer senses optical fluctuations in the input light resultant from bioreceptor—target interaction, and (ii) the amplitude of these changes hinge on the concentration of the analyte [1].

Even in very small amounts, several metal ions may have important effects on health state, as they are hardly degradable but easily accumulated in the body through the diet [3]. Metal ions are generally not essential nutrients; conversely, they could be damaging to all living species [4].

\* Widely indicated as "heavy metals" (HMs), in a technical report of 2002, the author concluded: "The term *heavy metal* has never been defined by any authoritative body such as IUPAC. No relationship can be found between density and any of the various physicochemical concepts that have been used to define *heavy metals* and the toxicity attributed to *heavy metals* ... Understanding bioavailability is the key to assessment of the potential toxicity... It depends on biological parameters and on the physicochemical properties of metallic elements, their ions, and their compounds. These in turn depend upon the atomic structure of the metallic elements, systematically described by the periodic table" [5].

In the last twenty years, with the aim to quantify trace amounts of such possible contaminants, environmental monitoring has generated a need for innovative and improved approaches that have ever-increasing sensitivity and selectivity, as described in a recent review paper on various analytical techniques-based biosensors [6]. The introduction of biosensors has brought in new and promising approaches, but with still limited application in the environmental field if compared with the biomedical one, where most efforts have converged in the past years.

Much research is still needed before biosensors consolidate as a recognized analytical strategy with respect to environmental and food trace contaminant detection.

In this direction, the integration of nanomaterials and functional biological molecules is part of a new era in the optical biosensor area. Actually, nano-structured materials unveil distinctive sizeand shape-dependent physicochemical properties, showing a number of possible interactions [7] with the biorecognition component, which may act as a reaction catalyst, or may be in equilibrium with macromolecules present in their natural biological settings or isolated and engineered [2]. Essentially, while the sensor sensitivity is influenced by the selected transducer component, the bioreceptor is responsible for the specificity [8]. Many biosensing elements that can be coupled to different transducers are now available for HM detection (Figure 1).

**Figure 1.** Optical biosensor scheme strategies for heavy metal (HM) ion detection in a water/milk "drop". Transduction methods and bioreceptor classes synergistically employed for the development of recently published devices.

In general, depending on the specific mechanism of the bioreceptor component, five groups can be identified: (i) DNA-based metal biosensor, (ii) antibodies, (iii) proteins, (iv) cellular structures or whole cells, (v) biomimetic receptors (gene-engineered molecules, molecularly imprinted polymers [9], and molecularly imprinted membranes [10], mimicking a natural bioreceptor. Most of them, natural and synthetic, are exhaustively described in recent review papers (concerning the interaction of metal ions with DNAs, peptides and enzymes, whole cells, as well as ionophores and small molecules) [8,11–15].

From a functional point of view, optical biosensors can be further categorized as: (i) probing biosensors: Entailing sensors based on target and recognition element affinity interaction; (ii) reacting biosensors: Where the optical responses relies on chemical processes [16]. Concerning the biomolecular probes, the most widely exploited can be collected into two macro groups: proteins and nucleic acids. The specific affinities of these two families of molecules for HM ions are briefly introduced below, before entering the focal topic of this paper.

With regard to metal binding proteins, phytochelatins or metallothionein, metal ligands found in plants, are usually exploited on the surface of the transducer, where protein–metal interactions occur through the formation of a complex [17,18]. Functional proteins with enzymatic activity (purified or directly in a microorganism) catalyze specific chemical reactions also in the presence of metal ions. The mechanisms of action of these elements embrace: (a) Transformation of the analyte into a sensor-detectable product, (b) detection of an analyte behaving as negative or positive enzyme activity modulator, or (iii) appraisal of enzyme properties deviations upon interaction with the analyte [19,20].

Metal ions affinity for amino acid side chains (with sulfur, nitrogen, and oxygen atoms) and the occurrence of such amino acids in antibody-determining regions are expected to influence the ability of antibodies to strongly bind to metal–chelate complexes [21–24].

On a parallel route, functional nucleic acids (FNAs) represent molecules whose usefulness is further than that of encoding genetic information [25], and whose chemical structure is suitable for metal recognition. Two active structures have been developed for this purpose, working as either direct metal binding or metal-assisted deoxyribonucleic/ribonucleic acid catalyst. Definitely, aptamers, metal ion-specific DNA, guanine (G)-rich oligonucleotides, and DNA-based enzymes (DNAzymes) are the most widely reported [26]. In brief, aptamers are able to effectively bind basically any molecule of choice; they consist of artificial short single-stranded (ss) nucleic acid sequences or peptide molecules identified by combinatorial selection, through the Systematic evolution of ligands by exponential enrichment (SELEX) methodology [27,28] that, upon binding to targets, can fold into specific secondary and tertiary structures [29,30].

Basically, DNA and metal ions may interact in three different ways: (i) By HM ions-based exchanging of hydrogen atoms of the Watson–Crick base pairs; (ii) by reversible binding of HM ions with DNA; (iii) forming kinetically inert complexes by persistent crosslinking of DNA with HM ions [31]. The metal ion-specific DNAs are those sequences most commonly rich in thymine (T) or cytosine (C), with great selectivity for metal ions, which promote robust metal-base complex formation—specifically forming T–Hg2<sup>+</sup>–T [32] and C–Ag+–C mismatch [33]. G-rich DNAs are G-rich strands with a tendency to self-associate into non-canonical secondary structures named G-quadruplexes (G4) [34]. On these cations coordination-induced structure/property changes, a number of strategies have been proposed for the detection of Pb2+, Hg2<sup>+</sup>, Ba2<sup>+</sup>, Ag2+, K<sup>+</sup> [26].

A different class is then represented by nucleic acid enzymes (Ribozymes and DNAzymes). These are molecules found in nature like catalytic RNA or in vitro selected DNA sequences, displaying specific strong metal-dependent activity and structure recognition capability, bypassing the need for metal immobilization [25,26].

Remarkably, the choice of a suitable biological element and transduction module makes the biosensor sensitive and analyte specific, thus efficient for toxicological studies. Portable biosensors also make in-situ analysis possible, facilitating real-time monitoring [35].

In this context, the present work aims to review the sensitivity of HM-dedicated optical biosensor systems published in the last decade. Several biosensors relevant for water sample or liquid food monitoring are here described, although only those showing HM ion detection in real and complex matrices are compared, as reported in Table 1.

All methods are listed in order of the prevalence of published biosensors for HM sensing in water or milk matrix, be it a real or laboratory-built aqueous sample. Analytical techniques here presented include luminescence, colorimetry, evanescent wave, surface-enhanced Raman spectroscopy, Förster resonance energy transfer, and surface plasmon resonance. As the core purpose of this review is to recognize which method displays the maximum stated sensitivity—for the selected HM ion, focusing on the biosensing element employed—additional focused tables (Tables 2 and 3) have been worked out and introduced later in the text.

#### **2. Biosensing Methods**

In order to introduce a brief summary of what the reader will encounter during this paragraph, in Figure 2, the HM ion optical biosensor distribution is plotted with respect to the recognition element used, in the frame of the same transduction method, as already classified in Table 1.

**Figure 2.** Distribution of biorecognition elements exploited in recently reported sensors for HM detection in real samples, as classified in Table 1.

With regards to the already mentioned classes of molecules, some considerations have emerged: (1) FNAs are the most employed; (2) direct metal binding DNA sequences (DMB-DNA) subclass, comprising aptamer, metal ion-specific DNA, and G-rich oligonucleotide, occupies a wide portion in the described FNAs-operating sensors; (3) proteins are the least employed, and (4) catalytic active protein-based sensors have been proposed more than those exploiting a non-catalytic protein, or a specific antibody.

In the next subparagraphs, the newly developed biosensors based on these recognition elements will be described, and with regard to the exploited biosensing mechanism, the more representative strategies will be showed in summary figures (Figures 3–10).

#### *2.1. Luminescence*

Luminescence concerns the emission of light from an excited electronic state of an atom or molecular species. A luminescence phenomenon that occurs when a chemical reaction triggers the excitation of an electronic state in a molecular species, that decays emitting light, is named chemiluminescence (CL) [36]; luminescence caused by electrogenerated chemical excitation is named electrochemiluminescence (ECL) [37]. Another luminescence phenomenon is photoluminescence (PL), where a molecule absorbs light, and then decays to a lower energy excited electronic state emitting light with a wavelength different than that of the absorbed light. Depending on the average lifetime of the excited state, the luminescence band can either be fluorescence or phosphorescence [36].

A number of biosensors exploit these phenomena and are here reported. A CL aptasensor for Hg2<sup>+</sup> detection, with a limit of detection (LOD) of 16 pM, was designed by Qi et al. [38]. The sensor is based on positively-charged gold nanoparticles (AuNPs) effect, that show catalytic properties for CL reaction of luminol and H2O2, and on aptamer conformation change induced by Hg<sup>2</sup>+. In the absence of Hg2+, the aptamer causes a weak CL signal because it wraps on positive AuNPs reducing their catalytic properties. Whereas the presence of Hg2<sup>+</sup> leads to a T–Hg2<sup>+</sup>–T complex formation preventing the interaction between aptamer and positive AuNPs, allowing the catalytic reaction to occur (Figure 3).

**Figure 3.** Metal ion-induced T–T complex mechanism for CL-based HM detection. Hg2<sup>+</sup> induces T–Hg2<sup>+</sup>–T complex formation preventing the interaction between aptamer and positive AuNPs, allowing the catalytic reaction occurrence and a stronger CL signal emission. From ref. [38], with the permission of the Publisher.

A different Hg2<sup>+</sup> biosensor, based on two label-free DNA probes and the molecular light switch complex [*Ru*(*phen*)2(*dppz*)] <sup>2</sup>+, was developed by X. Zhang et al. [39]. If Hg2<sup>+</sup> is present, the two label-free DNA probes, with eight T–T mismatches, form stable DNA duplexes which allow the intercalation of [*Ru*(*phen*)2(*dppz*)] <sup>2</sup><sup>+</sup>, leading to a significant Hg2<sup>+</sup>-dependent enhancement of the luminescence intensity. A LOD of 3.5 <sup>×</sup> <sup>10</sup>−<sup>10</sup> M was reached.

A portable multianalyte device, based on a different recognition strategy was designed by R. K. Mishra et al. for Hg2+, Pb2<sup>+</sup>, and Cd2<sup>+</sup> [40], obtaining a LOD of 1, 0.7, and 0.02 μg/L respectively. The device exploits a luminol–H2O2 mixture as a chemiluminescent system and horseradish peroxidase (HRP). The enzymatic inhibition results in a CL suppression that is analyte concentration dependent. Though, in a previous work, Deshpande et al. [41], exploiting a two enzyme based (i.e., alcohol oxidase (AlOx) and HRP) inhibition assay for single HM ion determination, showed a lower LOD (1 pg/mL) for Hg2<sup>+</sup> ions.

Recently, semiconductor sensors have received significant consideration. Electrochemically-etched nano-porous silicon (PS) is considered as a promising material for luminescent chemical sensors [42,43]. Interestingly, PS layers were exploited to develop novel enzyme-based biosensor systems for determination of glucose and urea (direct) as well as HM ions (inhibitory) [44]. In particular, changes in the quantum yield of PS photoluminescence at variations in medium pH. In particular, changes in the quantum yield of PS photoluminescence at variation in medium pH is proposed for the biosensor system. The authors show that the presence of Cu2+, Pb2+, or Cd2<sup>+</sup> ions causes an inhibition of the enzymatic reactions, resulting in a restoration of the PL quantum yield of PS. The LOD of the biosensor was approximately 10 nM. In order to develop handheld enzymatic luminescent biosensors for HMs detection, the integration of luciferase-based microfluidic chip with a portable luminometer has been also realized [45]. The LOD reached for Cu2<sup>+</sup> sulfate was 2.5 mg/L.

#### 2.1.1. Fluorescence

Transducing the molecular recognition events with the fluorescence signals is very attractive and is one of the most widely adopted methods [46]. Simultaneous measurements of multi-elements were arranged by an array-based biosensor exploiting enzymatic activity [47]. Acetylcholinesterase and urease were exploited as model enzymes and combined with a sensing probe (FITC–dextran), for the assessment of pH, urea, acetylcholine, and HMs. A LOD lowered to 10 nM was achieved for Hg2<sup>+</sup> and a LOD of 50 μM was reported for Cd2<sup>+</sup>.

A different kind of fluorescent transducer successfully constructed for determination of Cu2<sup>+</sup> in surface water, exploits the combination of semiconductor quantum dots (QDs) and enzymatic inhibition [48]. AlOx catalyzes methanol oxidation to produce H2O2, inducing the quenching of QDs fluorescence. Copper ions inhibit the enzyme action and, consequently, the quenching of QDs fluorescence decreases (Figure 4b). This hybrid sensor showed a LOD of 2.75 nM.

Useful as new fluorescent sensors, carbon-based QDs (CQDs, namely biodots) have attracted growing interests thanks to their biocompatibility, chemical inertness, and water solubility. In this direction, an application of DNA-derived CQDs in metal ion sensing was demonstrated [49]. Hg2<sup>+</sup> and Ag<sup>+</sup> are predisposed to be captured by the DNA biodots due to the existence of T and C groups (leading to T–Hg2+–T or C–Ag+–C complex), resulting in a quenched fluorescence, with the largest efficiency obtained at pH 7 and a LOD of 48 nM for Hg2<sup>+</sup> and 0.31 μM. for Ag+.

A turn-on aptasensor for Hg2<sup>+</sup> detection based on graphene oxide (GO) and DNA aptamers was proposed, where GO plays a role as nano quencher (Q) to reduce the fluorescence of acridine orange (AO). The recognition process results in the simultaneous formation of T–Hg2<sup>+</sup>–T and G4 structures; the formed G4 can capture AO from the GO surface, leading to fluorescence retrieval. A LOD of 0.17 nM was achieved [50].

Similarly, based on the T–Hg2+–T coordination between two neighboring poly–T strands, two ready-to-use chip-based sensors match well with microarray technology for Hg2<sup>+</sup> detection in the turn-on and turn-off modality [51]. The induced dislocation of the complementary poly-adenine (poly–A) strand, labeled with either a fluorophore (F) or a (Q), allows the turn-off and turn-on detection of Hg2<sup>+</sup>, respectively (Figure 4c). A lower LOD was achieved in the turn-off mode (3.6 vs. 8.6 nM).

Remarkably, with the aim to remove the HM-fluorescence quenching effect, a magnetic separation was integrated for Hg2<sup>+</sup> sensing based on the formation of the T–Hg2+–T structure [52], allowing a LOD value of 0.2 nM.

Another multi-analyte biosensor based on parallel analysis of microarray technology was developed exploiting DNAzymes [53]. In particular, copper and lead ion-dependent DNAzymes are first associated with their corresponding DNA substrates on the surface of aldehyde-modified slides. Then, in the presence of the specific ions, the DNA cleavage of the substrate takes place, inducing a strong variation in fluorescence signal. The sensor showed a LOD value of 0.6 ppb for Cu2<sup>+</sup> and 2 ppb for Pb2<sup>+</sup>. A higher sensitivity for Pb2<sup>+</sup>, with a LOD of 1 nM, was achieved by a similar approach, exploiting a Cy5-labeled DNA/RNA chimera (Figure 4f) as substrate [54].

Working on complex real samples, enzymatic degradation represents a threat to the structural integrity of D-DNAzymes. In this context, L-DNAzymes show similar recognition capability and catalytic capacity with respect to their enantiomer. A promising biosensor for Pb2<sup>+</sup> ion detection was realized by building a Pb2+-specific L-DNAzyme, allowing to obtain a LOD of 3 nM [55]. DNAzymes have also been exploited for Ag<sup>+</sup> detection [56]. As known, the most studied interaction between DNA and Ag<sup>+</sup> is the specific binding with C residues [33,57,58]. This interaction was used to develop Ag<sup>+</sup> biosensors [56,59,60] and for the assembling of fluorescent Ag nanoclusters [61,62]. Saran et al. [63] described the first Ag+-specific RNA-cleaving DNAzyme, successfully integrated in the specific biosensor. A catalytic beacon biosensor is obtained by labeling the 3 end of the DNAzyme strand with a black hole, which, upon hybridization, quenches the signal of the fluorophore located on the 5 end of the substrate. The Ag+-induced substrate cleavage enables fluorescence retrieval. A LOD of 24.9 nM was shown.

Even though DNAzyme-based lead sensors generally demonstrate good sensitivity, the high synthesis cost of these molecule limited their extensive application. A DNA sensor based on Pb2<sup>+</sup>-stabilized G4 formation was proposed with a LOD of 3.79 ppb [64]. In the absence of Pb2+, a fluorescent tracer intercalates with the single-stranded coil and strongly emits. While, in the presence of Pb2+, the random-coil folds into a G4 structure leading to signal reduction (Figure 4a).

Commonly, a DNA-based biosensor for Pb2<sup>+</sup> detection is frequently inclined to interference from Hg2<sup>+</sup>, due to the T–Hg2<sup>+</sup>–T interaction between Hg2<sup>+</sup> and T residues. A label-free system with a LOD in the nanomolar range was optimized (also in the presence of Hg2<sup>+</sup>) based on the Pb2<sup>+</sup>-induced G4 formation with cationic polythiophene water-soluble conjugated polymer (PMNT), as described in the colorimetric transduction method section of this review [65].

In another arrangement, Y.F. Zhu et al. proposed a singly-labeled bifunctional probe consisting of a Cd2+-specific aptamer (CAP), capable to act as the recognition element for Cd2<sup>+</sup> and the signal reporter [66]. The Cd2<sup>+</sup> presence induces the switching of the CAP coil conformation to a stem-loop structure, which brings the four guanosine bases at the 5 end close to 6-Fam at the 3 end, resulting in fluorescence quenching. The biosensor showed a LOD of ~2 nM.

Interestingly, G4 structures have been also exploited to develop a duplex functional fluorescent biosensor for distinct detection of Pb2<sup>+</sup> and Hg2<sup>+</sup> [67]. A K+-induced fluorescent G4 probe was assembled by a G-rich strand and a porphyrin. The sequence presents many T residues in addition to G residues, allowing to bind Pb2<sup>+</sup> or Hg2<sup>+</sup> selectively, changing into a more stably non-fluorescent G4 and a hairpin-like structure, respectively, resulting in PL reduction. LODs of 5.0 nM for Pb2<sup>+</sup> and 18.6 nM for Hg2<sup>+</sup> was reported.

As favorable as fluorescent nanomaterials, DNA-scaffolded silver nanoclusters (DNA–AgNCs) were successfully applied to a novel turn-on fluorescent biosensor [68]. When Pb2<sup>+</sup> is present, the aptamer forms a G4 structure and the two darkish DNA/AgNCs positioned at the 3 and 5 terminus come closer, thus the fluorescence intensity increases [69]. A LOD as low as 3.0 nM was reported.

Light-up biosensors based on the target-induced release of fluorescence-labeled aptamer, from a complex with a Q-labeled short complementary sequence, were developed for Cd2<sup>+</sup> and Pb2<sup>+</sup> [70,71], with a LOD of 40 and 60.7 nM, respectively.

A label-free aptasensor approach for Cd (II) detection was independently exploited by Y. Luan et al. [72] and B. Zhou et al. [73], combining an aptamer with unmodified dsDNA-specific dye. Based on the principle that hybridization of two aptamers boosts the fluorescence engendered during the reaction, B. Zhou et al. showed that, in the absence of Cd2<sup>+</sup>, SYBR green-I binds to the small groove of dsDNA (aptamer-complementary strand) establishing the dsDNA–dye complex and generating high fluorescence signal. The specific recognition and binding of aptamers with Cd2<sup>+</sup> induce the release of the complementary strand from dsDNA and the aptamer conformational switching to a stem-loop structure, causing fluorescence decay (Figure 4d). A LOD of 0.34 ng/mL was reached.

Likewise, Y. Luan et al. reported that, induced by Cd2<sup>+</sup> ions, the aptamer configuration changes from a random coil structure to an aptamer–Cd2<sup>+</sup> complex. After the introduction of complementary strands and Pico Green dye (PG), a hybrid with the residual free aptamers that did not bind with Cd2<sup>+</sup> is formed. This results in a higher PL signal (Figure 4e), allowing a higher sensitivity (LOD of 0.038 ng/mL) [72].

A comparable strategy was proposed for Pb2<sup>+</sup> detection [74]. This biosensor is based on the principle that Pb2<sup>+</sup> induces a structural change of G-rich thrombin aptamer from random coil to G4. This prevents its binding to the complementary sequences to form dsDNA and causes a fluorescence intensity decrease. The results showed a LOD of 1 ng/mL.

A label-free fluorescence sensing system was also developed for As3<sup>+</sup> detection by the exonuclease III (Exo III)-assisted cascade target recycling amplification process [75], exhibiting a LOD of 5 ng/L. As signal indicator and sensing element, the 2-amino-5,6,7-trimethyl-1,8-naphthyridine and the triple-helix molecular switch were used, respectively. This sensor could detect other HM ions with newly-designed triple-helix molecular switch by using aptamer sequences.

**Figure 4.** Various biosensing element constructs for fluorescence-based HM detection. (**a**) Pb2<sup>+</sup>-stabilized G4 formation for turn off detection. In the presence of Pb2+, the T30695 oligonucleotide folds into a G4 structure, leading to a PL signal reduction [64]. (**b**) Cu2<sup>+</sup>-determined enzymatic inhibition for turn on detection. AO catalyzes the oxidation of methanol to hydrogen peroxide, inducing the quenching of QDs fluorescence. Cu2<sup>+</sup> ions inhibits the enzymatic activity decreasing the quenching of QDs fluorescence [48]. (**c**) Metallophilic attraction of the Hg atom in the T–Hg2+–T base pair mismatch. The Hg2<sup>+</sup>-induced dislocation of the complementary labeled poly–A strand allows the turn-off detection mechanism. [51]. (**d**) Cd2+-induced hairpin formation. The release of the complementary strand from dsDNA and the sequence conformational switching to a stem-loop structure lead to a fluorescence decay of the signal reporter [73]. (**e**) Random coil structure to aptamer–Cd2<sup>+</sup> complex. After the addition of complementary strands and PG, the residual free aptamer that did not bind with Cd2<sup>+</sup> forms a hybrid with complementary strands and PG dye which results in a big fluorescent enhancement [72]. (**f**) Pb2+-induced hydrolytic cleavage signal-off. The catalytic strand carries out catalytic reactions for hydrolytic scission of the substrate sequence at the rA site (red arrow). Once the substrate is broken into two pieces, it dissociates from the catalytic strand with a decrease of the surface PL intensities [54]. Adapted with the permission of the Publishers.

In the frame of functional device miniaturization, combining a microfluidic sample pre-treatment module (cation exchange resins) with a DNA aptamer immobilized photoluminescent graphene oxide QD (GOQD), a novel Pb2<sup>+</sup> detection platform sensor was proposed [76], exhibiting a LOD of 0.64 nM. The DNA aptamer on the GOQD specifically captures the target (forming a G4 complex) which can trigger electron transfer from GOQD to Pb2<sup>+</sup> upon UV irradiation, leading the GOQD PL quenching.

#### 2.1.2. Electrochemiluminescence

ECL is the process through which those intermediates generated at the electrodes undergo high-energy electron transfer reactions to produce an excited state that emits light, after relaxation to a lower level [77]; the process is initiated and modulated by switching an electrode voltage [78]. ECL allows small analyte detection at sub-picomolar concentration and wide dynamic range [79].

Various strategies were recently developed, such as biosensors that rely on the formation of the T–Hg2<sup>+</sup>–T and Ru(phen) 2+ <sup>3</sup> or Ru-dppz, which permitted a LOD of 20 or 5.1 pM to be achieved, respectively [80,81].

In their study, X. Zhou et al. [82] reported that Bst DNA polymerase exhibits specific behaviors on the T–Hg2<sup>+</sup>–T biomimetic structure. The sensor exploits the MBs-labeled primer, planned to match the region of the circular padlock probe but with two T–T mismatches at the 3 terminus. If Hg2<sup>+</sup> is introduced, the DNA polymerase reaction with rolling circular amplification (RCA) mechanism is induced. Then, the resulting RCA products hybridize with the tris (bipyridine) ruthenium (TBR)-marked probes and sensed by ECL, once they are attracted to the magnet under the electrode. A LOD of 100 pM was shown.

One more method was designed by Meng Li et al., exploiting a Pb2<sup>+</sup>-specific DNAzyme, achieving a LOD of 9.6 <sup>×</sup> <sup>10</sup>−<sup>13</sup> M [83]. In this sensor, CdS QDs and DNAzyme with Ag/ZnO coupled structures were immobilized on agold nanodendrites-modified ITO electrode. Pb2<sup>+</sup>-activated DNAzyme moves the Ag/ZnO coupled structures near the surface to catalyze the reduction of part of the H2O2, inducing a signal intensity reduction.

Rather than utilizing DNAzyme, L. Lu et al. [84] proposed a sensor to detect Pb2<sup>+</sup> using a graphene/AuNPs-modified electrode and ssDNA labeled with CdSe QDs. When Pb2<sup>+</sup> is present, the G-rich ssDNA adopts the G4 conformation, leading to a shortening of the distance between the CdSe QDs and the graphene–AuNPs nanocomposite (Figure 5). This decreases the ECL intensity, allowing for the detection with a limit of 10−<sup>10</sup> mol/L.

**Figure 5.** Metal ion-induced quadruplex construct for ECL-based HM detection The Pb2<sup>+</sup> causes the G4 structure formation of the G-rich ssDNA, leading to a shortening of the distance between the CdSe QDs and the graphene–AuNPs nanocomposite, thus inducing a reduction of the ECL signal. From [84], with the permission of the Publisher.

A novel ECL sensor to detect Pb2<sup>+</sup> exploiting hemin/G4-based DNAzyme on the core-shell CdSe@CdS QDs, was proposed by X.-L. Du et al. [85]. Pb2<sup>+</sup>-induced G4 combines with hemin to form DNAzyme, which can catalyze H2O2 and oxidize 4-chloro-1-naphthol (4-CN) to form an insoluble precipitate. In the presence of Pb2<sup>+</sup>, more DNAzymes are produced and, thus, more 4-CN molecules are oxidized catalytically, leading to an output signal reduction. A LOD of 0.98 fM was achieved.

Furthermore, a microfluidic paper-based device was successfully applied for concurrent detection of Pb2<sup>+</sup> and Hg2<sup>+</sup> based on the formation of G4 and T–Hg2<sup>+</sup>–T complexes, respectively [86]. Due to the different operational potentials of the two exploited labels (Si@CNCs and Ru@AuNPs), Pb2<sup>+</sup> and Hg2<sup>+</sup> can be quantified with a LOD of 10 pM and 0.2 nM, correspondingly.

#### *2.2. Colorimetric Method*

In colorimetric sensors, the analyte detection occurs by means of a color change of the sensing element. Current technology based on colorimetry focuses on cost reduction, miniaturization, and in-situ detection. Generally, the recognition mechanism is based on molecular interaction on the substrate surface modified with NPs and functional groups [87].

For instance, DNA adsorption by citrate-capped AuNPs could be a function of DNA conformation. DNAs without stable secondary structures allow higher colloidal stability of AuNPs against salt-induced aggregation, because they are more efficiently adsorbed. A sensor exploiting Tl+-induced DNA folding and AuNPs was described by Hoang et al. [88]. The presence of Tl<sup>+</sup> inhibits the DNA adsorption by AuNPs due to G4 sequence folding. Then, adding NaCl solution, a red-to-blue color change is observable because of NPs aggregation. A LOD of 4.6 μM was achieved.

Similarly, a specific Pb2+-induced G4 oligonucleotide (TBAA) probe and the cationic polythiophene (PMNT) readily form an electrostatic PMNT–TBAA red colored complex [65]. This sensor can detect Pb2<sup>+</sup> traces at the micromolar level with the naked eye. Moreover, the authors report that, in the presence of Hg2+, the TBAA sequence (having adenine base) has a higher selectivity for Pb2<sup>+</sup> than TBA (without adenine base in the sequence). As already reported, the same biosensor exhibits a lower LOD, when working in fluorometric mode.

In order to detect Hg2+, Zhu et al. [89] designed a sensor established on ssDNA, phthalic diglycol diacrylate (PDDA) and AuNPs. The T–Hg2+–T structure is much stronger than the interchain contact between ssDNA and PDDA. When the ssDNA recognizes Hg2<sup>+</sup>, a random coil-to-hairpin structure change occurs, avoiding ssDNA interaction with PDDA. Therefore, the free PDDA induces AuNP aggregation (Figure 6b), displaying a color change as a function of Hg2<sup>+</sup> concentration. The LOD was as low as 5 nM.

A multianalyte responsive sensor, able to identify Ag<sup>+</sup>, Hg2<sup>+</sup>, Cr3<sup>+</sup>, Sn4<sup>+</sup>, Cd2<sup>+</sup>, Pb2<sup>+</sup>, Zn2<sup>+</sup>, and Mn2<sup>+</sup> was designed by Tan et al. [90]. It is based on differential colorimetric and fluorescent response of FAM-DNA-AuNP once conjugated to a specific metal ion. A LOD of 50 nM was achieved.

A different approach for visual detection of Hg2<sup>+</sup>, Ag+, Cu2+, Cd2+, Pb2+, Cr6<sup>+</sup>, and Ni2<sup>+</sup> was reported by Hossain and Brennan [91]. An enzymatic reaction is optimized on a sol gel matrix-spotted bioactive paper device; β-galactosidase-substrate catalysis produces a colorimetric signal intensity, which is inversely proportional to the metal ion amount. The sensitivity was different for the diverse ions, as reported in Table 1.

Another paper device was designed by J. Xu et al. [92], for the detection of Pb2<sup>+</sup> via colorimetric and ECL techniques, exploiting a metal-specific DNAzyme and rGO–PdAu–GOx labeled oligonucleotide hybrid. The dual mode sensor showed a lower LOD in the ECL readout (0.14 nM) than in the colorimetric one (LOD: 1.6 nM).

A sensor based on mushroom apo-tyrosinase, entrapped in polyacrylamide gel, was developed by Kaur and Verma [93] in order to detect Cu2<sup>+</sup>, which act as enzyme cofactor for levodopa (L-DOPA) conversion, with a corresponding color change. The shown LOD was 0.01 ug/L.

A colorimetric Hg2<sup>+</sup> detection was also optimized on a test strip, exploiting biotin-labeled and thiolated DNA-modified AuNPs and a T-rich DNA immobilized on the nitrocellulose membrane. Under optimized conditions, the LOD achieved for Hg2<sup>+</sup> was 3 nM [94] or 5 nM [95].

Another colorimetric paper-based platform, involving T–Hg2+–T coordination chemistry and AuNPs aggregation, showed a LOD of 50 nM [96]. In a different way, a linker T-rich DNA and sequences complementary to the AuNPs DNA was designed to induce particle aggregation [97]. Remarkably, Hg2<sup>+</sup> ions induce the linker DNA folding, allowing AuNPs to quickly disassemble and return to red color. A lower LOD (5.4 nM) was shown with respect to the AuNPs aggregation strategy described.

In another work concerning disposable lateral flow strips, the authors examined hairpin probe-modified AuNPs and T–Hg2+–T structure-based strategy. An additional T-rich, digoxin-labeled DNA strand was considered in order to hybridize with T–Hg2+–T coordination. Then, digoxin dsDNA–AuNPs complexes are captured by immunoreaction with the anti-digoxin Ab immobilized (Figure 6a) on the strip and revealed by a red band [98]. Interestingly, a lower LOD for Hg2<sup>+</sup> was shown (0.1 nM), when compared with non-immunochromatographic approaches.

On parallel route, an Exo III-catalyzed target recycling approach was employed to improve the sensitivity of a similar disposable strip on the basis of Hg2+-triggered toehold binding. Using AuNPs as the tracer enables the detection of Hg2<sup>+</sup> with a LOD of ~1 pM [99]. Moreover, in order to sense Cu2+, a lateral flow device based on specific ion-dependent DNA-cleaving DNAzyme and AuNPs was developed, achieving a LOD of 10 nM [100].

**Figure 6.** T–Hg2+–T structure and hairpin probe-modified AuNPs-based strategy for colorimetrybased HM detection. (**a**) The digoxin dsDNA–AuNPs complexes are captured by immunoreaction with the anti-digoxin Ab\* immobilized on the strip and revealed by a red band [98]. (**b**) With the formation of the T–Hg2<sup>+</sup>–T, a random coil-to-hairpin structure change occurs, avoiding ssDNA interaction with PDDA. A color change is observed due to the AuNP aggregation by free PDDA [89]. Adapted with the permission of the Publishers.

#### *2.3. Evanescent Wave*

This method employs the evanescent field of an optical fiber to excite the biological recognition molecule, producing a fluorescence signal. An optical fiber is essentially a cylindrical dielectric waveguide with an inner core having a refractive index greater than that of the cladding. EW exploits the phenomenon of total internal reflection (i.e., the transition of light in the optical fiber by continually reflecting off the cladding–core interface without data loss. If the cladding is removed, the evanescent

field can interact with the fiber surroundings [101]. In order to immobilize biological recognition elements on the optical fiber surface, various methods have been reported [102], such as direct or mediated covalent immobilization, adsorption, or entrapment in polymer matrices.

Based on this transduction, a DNAzyme-based sensor for Pb2<sup>+</sup> detection was developed by N. Yildirim et al. [103], using GR-5 Pb2<sup>+</sup>-dependent DNAzyme. In the presence on Pb2+, the active molecule can catalyze the cleavage of an RNA base embedded in the fluorescent-labeled (Cy5.5) DNA substrate. After that, the released Cy5.5-labeled fragments hybridize with the complementary strands immobilized on the optical fiber, and Pb2<sup>+</sup> detection is revealed by PL signal. Restored over 100 times, this sensor showed no important performance decay and a LOD of 1.03 nM.

Another sensor based on DNAzyme for Pb2<sup>+</sup> detection was realized by R. Wang et al. [104] The whole sensing procedure requires three steps: (i) Pb2<sup>+</sup> ion determines the cleavage of the DNA substrate at the single RNA site, by the DNAzyme, causing the release of a short ssDNA arm which will be used in the second step; (ii) the released ssDNA hybridizes with a complementary DNA strand immobilized on MBs in solution, causing the competitive detachment of the originally hybridized probes (streptavidin–ssDNA–Cy5.5) and the remaining dsDNA–MBs complex is removed by magnetic separation; (iii) the released signal probe is pumped into the flow cell of the biosensing platform, where it can be captured by the desthiobiotin-modified fiber. A LOD of 1 nM was achieved, with the possibility to be reused at least 250 times.

To avoid the use of MBs and to keep the same performance, a similar system was optimized for Hg2<sup>+</sup> sensing [105], by introducing a quencher. Fluorescein-labeled DNA strands with streptavidin (DNA–SA) were designed to hybridize with Q-labeled cDNA strands (Q–DNA). Hg2<sup>+</sup> induces an enhancement in the PL signal because of the Q-DNA release, once the DNA-SA is bound to form a hairpin structure stabilized by the T–T mismatch (Figure 7a). A LOD of 1.06 nM was shown.

With a similar LOD, F. Long et al. [106] developed another sensor based on T–T mismatch pairs and fluorescently (Cy5.5)-labeled cDNA. The DNA probe has two functional elements: a T–T mismatch pair that can bind with Hg2<sup>+</sup> ions, and a short sequence that can hybridize with the fluorescently-labeled cDNA. Via a structure competitive mode, Hg2<sup>+</sup> ions lead to a decrease of the signal. The authors stated a LOD of 2.1 nM, with a reproducibility over 100 times.

In more recent works, the same author proposed two structure-switching DNA optical biosensors for detection of HM ions [107,108]. The developed approaches for Hg2<sup>+</sup> or Pb2<sup>+</sup> detection, respectively, differ by the FNA-based strategy exploited (i.e., T–T mismatch or G4 aptamer (Figure 7b)).

Once introduced in the modified optofluidic cell, the specific metal ion-induced aptamer conformation change reduces the binding of a fluorescently-labeled free DNA with the immobilized DNA probe, causing a decrease of fluorescence signal. A LODs of 1.2 nM for Hg2<sup>+</sup> and 0.22 nM for Pb2<sup>+</sup> were reported. One more sensor for Hg2<sup>+</sup> and Pb2<sup>+</sup> detection, based on T–T mismatch-containing DNA or DNAzyme, respectively, was developed by S. Han et al. [109]. In this system, the detection of HM contaminants is carried out exploiting two complementary DNA sequences, one labeled with a Cy3.3 and one with a Q. The metal ion induces structural modification, causing paired-strand dehybridization and, consequently, the binding of the Cy3.3-labeled segment to the cDNA probe on the fiber surface. By excitation via EW, a detectable fluorescence signal is generated, with a LOD of 22 pM for Hg2<sup>+</sup> and 20 nM for Pb2+.

**Figure 7.** FNAs constructs for evanescent wave-based HM detection (**a**) (1) T–T mismatch-driven biosensing by triple functional DNA–protein conjugates for facile detection of mercury ions; (2) Once the DNA–SA is bound to form a hairpin structure stabilized by the T–T mismatch, an enhancement in the signal is observed [105]. (**b**) G4-driven lead ions biosensing. A decrease of fluorescence signal is recorded by the Pb2+-induced aptamer conformation change (G4) that reduces the binding of the fluorescently-labeled free DNA with the immobilized complementary strand [108]. Adapted with the permission of the Publishers.

#### *2.4. Surface-Enhanced Raman Spectroscopy*

In SERS, definite metallic surfaces are used to intensify Raman scattering of the specific element, by benefitting from localized surface plasmon resonances. Noteworthy: (i) SERS spectra can provide information about the chemical structure of the target, (ii) it permits rapid detection; (iii) weak Raman scattering of water makes its background signal negligible [110].

A highly sensitive DNAzyme-centered SERS quadratic amplification method, based on a bio-barcode and hybridization chain reaction, was developed for Pb2<sup>+</sup> detection [111]. The system includes a DNAzyme-MB complex, a SERS active bio-barcode (composed of the capture probe matching the stem of hairpin DNA, and Raman dye-labeled DNA) on top of AuNPs, to produce a strong SERS signal. Adding Pb2<sup>+</sup>, once a DNA–Pb2<sup>+</sup> complex is formed, a catalytic cleavage of the substrate sequence takes place, giving rise to a series of reaction steps, finally leading to quantitative Pb2<sup>+</sup> detection with a LOD of 70 fM. The method can be further applied to different elements by substituting the lead-responsive DNAzyme with the specific functional DNA.

Combining a specific As3<sup>+</sup> aptamer, a reporter molecule and Raman-labeled Au@Ag core–shell NPs, a novel SERS strategy was proposed [112]. In the absence of As3<sup>+</sup> ions, the aptamer and the reporter are absorbed on Au@Ag; while when they are present, the As3<sup>+</sup> ions compete with NPs for binding to the aptamer, inducing its release from the NP surface, which then aggregate generating SERS "hot spots". This amplification strategy allowed to obtain a LOD of 0.1 ppb.

Likewise, a label-free SERS device was developed for sensing of Hg2<sup>+</sup> [113], exploiting aptamer-derivatized SiO2@Au core–shell NPs. The DNA aptamer consists of two segments, one containing guanine (G) and adenine (A) bases as signal reporters and the other segment, with consecutive T, as the Hg2<sup>+</sup> recognition element. The single strand poly–T shows a flexible structure; when present, Hg2<sup>+</sup> ions cause the formation of T–Hg2+–T pairs via N–Hg2+–N J-coupling bonding. As a result, the DNA molecule adopts vertical alignment (Figure 8a), changing respective Raman intensities of A and G bases in the sequence. In this system, Hg2<sup>+</sup> detection showed a limit of 10 nM.

The system formerly suggested by L. Zhang et al., requiring a fluorescent label, resulted as more sensitive [114]. It was based on nanoporous gold as the plasmonic surface and a Cy5-labeled aptamer as the optical tag for Hg2<sup>+</sup> detection. The coordination of a pair of poly–T oligos with the metal ion induces the molecule perpendicular arrangement, as above described. Consequently, an amplification of the fluorophore surface-enhanced resonance Raman scattering signal (SERRS) variation is observed. A LOD of 1 pM was reported.

Exploiting the same Hg2<sup>+</sup> biorecognition element, W. Ma et al. obtained SERS-enhanced Hg2<sup>+</sup> detection, thanks to the T–Hg2<sup>+</sup>–T-induced assembly of gold nanostar dimers [115]. A great number of "hotspots" were formed, inducing an increase of electromagnetic field over an extensive connecting region. A LOD of 0.8 pg/mL was reached, showing a higher sensitivity if compared with the similar strategy exploited for As3<sup>+</sup> detection (Figure 8b) described in [112].

A selective single nanowire-on-film (SNOF) sensor for Hg2<sup>+</sup> was realized exploiting the hybridization between T-rich ssDNAs and complementary Cy5-labeled DNAs [116]. In the presence of Hg2+, T-rich DNAs fold into hairpin structures to form T–Hg2+–T pairs, leading to an easy release of Cy5-tagged DNAs. The free-labeled ssDNAs are then caught by the SNOF derivatized with cDNAs, turning on the SERRS signal. A LOD of 100 pM was achieved.

**Figure 8.** FNAs constructs for SERS-based HM detection. (**a**) Hg2<sup>+</sup> causes a vertical alignment of DNA molecules due to the formation of T–Hg2<sup>+</sup>–T pairs via N–Hg2+–N J-coupling bonding, changing respective Raman intensities [113]. (**b**) As3<sup>+</sup> induces the aptamer release from NP surface, inducing NP aggregation and the generation of SERS "hot spots" [112]. Adapted with permission of the Publishers.

#### *2.5. Förster Resonance Energy Transfer*

FRET is a physical process where a non-radiative energy transfer from an excited state molecule (donor) to another molecule (acceptor) occurs, by means of intermolecular long-range dipole–dipole coupling. When the energy transfer takes place from donor to acceptor, the fluorescence intensity of the donor decreases. An essential requirement for effective transfer is that an overlap exists between the fluorescence spectrum of the donor and the absorbance spectrum of the acceptor. The rate and the efficiency of the energy transfer depends on the sixth power of the distance between donor and acceptor [117]. Various combinations of donor–acceptor pairs have been used, such as two fluorophores, fluorophore with AuNP, fluorophore with an intercalator or with a dark absorber [118].

For instance, T–Hg2<sup>+</sup>–T complex-induced conformational change of ssDNA allows one-step sensing of Hg2<sup>+</sup> in a AuNPs-based sensor developed by G. Wang et al. [119]. The AuNPs were used as acceptor and FAM as donor. The DNA probes tagged with a FAM on 3 and thiol on 5 end were bound to AuNPs. In order to enable an enhanced FRET process, FAM and AuNPs need to be close to each other, as occurs when the conformation of the DNA probe changes into a hairpin structure leading to fluorescence signal quenching. A LOD of 8 nM was achieved with this approach.

Using the catalyzed hairpin assembly technique, a different aptasensor for Hg2<sup>+</sup> was developed by K. Chu-mong et al. [120]. The suggested strategy exploits a Hg2<sup>+</sup> aptamer–catalyst complex and two hairpin DNA: H1—fluorescein (donor) and H2—tetramethylrhodamine (acceptor). The formation of the T–Hg2+–T complex releases the catalyst strand, triggering the signal amplification step: Hairpin assembly is catalyzed turning H1 and H2 into a duplex. Consequently, FRET efficiency increases and the Hg2<sup>+</sup> concentration can be measured with nanomolar LOD.

An opposite functional scheme was described for Ag<sup>+</sup> sensing by Y.-J. Chen et al. [121]. Fusing the cyan fluorescent protein (donor) and the yellow fluorescent protein through a truncated CupR protein. CupR contains a dimerization helix and a metal binding domain. The presence of Ag<sup>+</sup> ions causes the decrease in FRET efficiency by inducing conformational change of the biorecognition element (Figure 9b).

**Figure 9.** FNAs and protein constructs for FRET-based HM detection. (**a**) The Tl<sup>+</sup> causes the G4 structure formation, leading to a shortening of the distance between the donor and acceptor, thus inducing an enhancement in FRET efficiency [88]. (**b**) The decrease in FRET efficiency is induced by conformational change of the CupR protein, in the presence of Ag<sup>+</sup> [121]. Adapted with permission of the Publishers.

More complex systems were also designed to simultaneously detect several HMs. Using the establishment of C–Ag<sup>+</sup>–C and T–Hg2<sup>+</sup>–T complexes, Cy5 and TAMRA as acceptors and CdTe QDs as donors, C. Hao et al. [122] successfully detected Ag<sup>+</sup> and Hg2<sup>+</sup> with a LOD of 2.5 and 1.8 nM, respectively. When a specific ion is present, if donor and acceptor are in close proximity, a fluorescence intensity increase will take place.

Interestingly, J. Xia et al. engineered specific DNA sequences for Hg2<sup>+</sup>, Pb2<sup>+</sup>, and Ag+, integrating them in two DNA strands and labeling these strands with multicolor fluorophores, in order to realize a cascade FRET [123]. In this way, only one excitation wavelength is needed to obtain a fingerprint-like spectrum in multianalyte monitoring. The sensor works in a dynamic range from 100 nM to 2 μM for Ag<sup>+</sup> and Hg2<sup>+</sup> and can detect as low as 20 nM Pb2+.

As already described, M. Hoang et al. [88] demonstrated that a sensor based on G4 DNAs, FAM (donor), and TMR (acceptor) can be used for Tl<sup>+</sup> detection (Figure 9a) with a LOD of 59 μM, unusually lower than that showed by the colorimetric transduction method.

#### *2.6. Surface Plasmon Resonance*

When light incides on a metal surface, plasmons are generated, whose propagation is very sensitive to the variations in the material refractive index. This alteration can be caused by biomolecular interaction (probe–target) or by a structural modification of the molecules linked to the sensor surface [124,125].

For instance, the detection of Cu2<sup>+</sup> was achieved by associating a SPR biosensor with the competitive adsorption of proteins [126]. The interaction between bio-receptors (native proteins (albumin)) and Cu2<sup>+</sup> ions leads to protein denaturation, inducing a lower affinity between protein–gold surface, thus initiating the competitive displacement by the native one (Figure 10a), which is monitored by SPR measurement with a LOD down to 0.1 mg/L.

**Figure 10.** (**a**) Protein and DNA structure conformational changes, for SPR-based HM detection. The interaction between native proteins and Cu2<sup>+</sup> leads to the protein structure denaturation and weakens its attraction on the sensing surface. The competitive displacement by the native one causes variations in the SPR angle profile [126]. (**b**) The rhodamine-labeled ssDNA folds into the T–Hg2+–T-mediated hairpin loop; this structural change approaches the rhodamine fraction near to the Au surface causing the increase in the SPR signal and the PL quenching [127]. Adapted with permission of the Publishers.

A mercury (II) sensor, based on the dissociation rate of the trans-acting factor MerR from the cis-element, was investigated by SPR [128]. The sensor, modified with dsDNA including the cis-element (Pmer), can monitor the dissociation stage of MerR or protein-tagged MerR from the cis-element, enabling measurement of Hg2<sup>+</sup> with a LOD of 5 μg/L.

Non-specific adsorption can influence the SPR accuracy. In this direction, a laser scanning confocal imaging and SPR were combined to realize a system for Hg2<sup>+</sup> detection [127]. By adding Hg2+, the rhodamine-labeled ssDNA folds into the T–Hg2+–T-mediated hairpin structure and this structural change attracts the rhodamine fraction in proximity to the Au surface (Figure 10b). A double effect is observed: SPR signal heightening and PL quenching. From the PL quenching status, the strand folding is monitored in real time, and the Hg2<sup>+</sup> detection is recorded by the SPR signal, as a function of refractive index and thickness variations of the Au surface, achieving a LOD of 0.01 ng/mL.

In this rich context, the summary of the recently described biosensors is schematized in Table 1. Here, biosensors are classified by their sensitivity (from lower to higher LOD) with respect to a specific ion, within the same transduction method on real samples, but with diverse bio-signaling strategies.

Moreover, in order to clearly illustrate the most sensitive recent methods as well as the bio-recognition elements giving the lowest detection limits, two comparative tables (Tables 2 and 3, respectively) are proposed and shown below. Then, a representative drawing (Figure 11) aims to show the most sensitive detection strategies, with respect to a specific analyte, applied in real samples.






 96


 **Sample**

 T.W.

 M.W.

 T.W. E.W.

 E.W.

 M.W., T.W.

 T.W.

S.W. T.W., R.W.

U.W.

 T.W.

 S.W. T.W., L.W.

S.W.

 T.W. T.W., S.W.

 T.W., S.W.

S.W.

 T.W. M.W.

 [127] [128]

—

 (0.01–100) ng/mL

(101–104) μg/L

—

 [119] [122]

 [121]

[88]

 [115] [113]

 [122] [120]

 [106] [112]

 [111] [114]

 [109] [105]

 [103] [109]

 [108] [104]

 **Reference**

*Biosensors* **2019**, *9*, 96

**SPR** Cu2+Protein ~0.1 mg/L T.W., M.W. [126] Tap water (T.W.); mineral water (M.W.); surface water (S.W.); underground water (U.W.); environmental water (E.W.). Direct metal binding DNA sequence (DMB-DNA); Antibody (Ab\*) (\*indirectly exploited).

**LSCI-SPR**

**SPR**

Hg2+ Hg2+

DMB-DNA (T–Hg2+–T)

DMB-DNA

0.01 ng/mL

5 μg/L

**Figure 11.** Most sensitive biorecognition strategy/specific HM ion, in real samples. (**a**) Disposable strip biosensor based on Hg2<sup>+</sup>-induced toehold binding and Exo III-assisted signal amplification [99]; (**b**) ECL Pb2<sup>+</sup> sensor based on hemin/G4-based DNAzyme biocatalysis [85]; (**c**) Cu2<sup>+</sup> triggered conversion of apo-tyrosinase disc into holo-tyrosinase one, and consequent L-DOPA to dopachrome transformation [93]; (**d**) As3<sup>+</sup> detection by Exo III-assisted cascade target-recycling amplification scheme [75]; (**e**) Possible on-site analysis of HMs by means of the HRP-based bioassay [40]; (**f**) Aptamer-modified NPG-based SERRS sensing of Hg2<sup>+</sup> [114]; (**g**) Ag+ and Hg2<sup>+</sup> detection by FRET between QD and organic dyes [122]. Adapted with permission of the Publishers.


**Table 2.** Most sensitive transduction methods, with respect to the specific analyte (n. of published works ≥ 2), applied in real samples.

**Table 3.** Bio-recognition elements giving the lowest detection limits, within the same transduction method, with respect to a specific analyte (n. of published works ≥ 3).


#### **3. Conclusions**

Nucleic acids, biocatalysts, antibodies, receptors, etc., are natural or biomimetic elements with distinctive features such that they have been engaged as recognition probes since the first public biosensor description in a paper, over 55 years ago, in which Dr. L. C. Clark termed his device as an "enzyme electrode" [129]. In the fields of environmental and food analysis, water and milk exemplify the matrices involved in potential HM ion contamination. In this context, although most of the developed systems were tested only on buffered solutions, plenty of optical biosensors appropriate for real samples showed up in the last decade for possible environmental and food quality monitoring applications. Continuous advances are presented, exploiting nano-microtechnology and biotechnology, such as for miniaturization of integrated systems, genetic engineering of receptors, enzymes, and microorganisms, as well upgrading of bioelement immobilization methods.

Thus far, a number of metals can be selectively sensed by DNA sequences down to the low ppb level [11]. Accordingly, Table 3 shows that direct metal binding DNA sequences allow obtainment of the highest sensitivity. In detail, the biorecognition mechanisms more frequently adopted are those based on T–T mismatch and G-quadruplex, respectively, for Hg2<sup>+</sup> and Pb2+; nonetheless, to the same extent, functional nucleic acids (DNAzyme) are exploited for Pb2+. Among the optical biosensors here reviewed, those applied in real samples, namely milk and water (specifically tap water, mineral water, surface water, underground water), have been assessed by spike test, largely for Hg2<sup>+</sup>, Pb2<sup>+</sup>, and Cd2<sup>+</sup> ions, in descending order, and, in small part, also for Cu2<sup>+</sup>, Ag<sup>+</sup>, Cr3<sup>+</sup>, As3<sup>+</sup>, Tl<sup>+</sup>, and Sn4<sup>+</sup> ions, as summarized in Table 1. Remarkably, ten multianalyte optical devices (able to sense up to eight HM ions) were shown out of a total of more than seventy biosensors here considered, with a large part of them designed to quantitatively discriminate between two ions.

**Author Contributions:** A.A. and R.R. collaborated to design the manuscript. A.A., A.D.T., A.D.B. and R.R. wrote the manuscript. A.A., A.D.T. and R.R. revised the manuscript.

**Funding:** This research was funded by [Regione Puglia, Italy] INNOLABS HYDRO RISK LAB Project (grant number NRNABW5).

**Acknowledgments:** A.D.B., A.A. and R.R. are grateful to MIUR for a PhD grant financed by PON project "Dottorati innovativi con caratterizzazione industriale" (Code: project n.1 DOT1312457).

**Conflicts of Interest:** The authors declare no conflicts of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

## **A Critical Comparison between Flow-Through and Lateral Flow Immunoassay Formats for Visual and Smartphone-Based Multiplex Allergen Detection**

#### **Georgina M. S. Ross 1,\*, Gert IJ. Salentijn 1,2 and Michel W. F. Nielen 1,2**


Received: 3 November 2019; Accepted: 11 December 2019; Published: 12 December 2019

**Abstract:** (1) Background: The lack of globally standardized allergen labeling legislation necessitates consumer-focused multiplexed testing devices. These should be easy to operate, fast, sensitive and robust. (2) Methods: Herein, we describe the development of three different formats for multiplexed food allergen detection, namely active and passive flow-through assays, and lateral flow immunoassays with different test line configurations. (3) Results: The fastest assay time was 1 min, whereas even the slowest assay was within 10 min. With the passive flow approach, the limits of detection (LOD) of 0.1 and 0.5 ppm for total hazelnut protein (THP) and total peanut protein (TPP) in spiked buffer were reached, or 1 and 5 ppm of THP and TPP spiked into matrix. In comparison, the active flow approach reached LODs of 0.05 ppm for both analytes in buffer and 0.5 and 1 ppm of THP and TPP spiked into matrix. The optimized LFIA configuration reached LODs of 0.1 and 0.5 ppm of THP and TPP spiked into buffer or 0.5 ppm for both analytes spiked into matrix. The optimized LFIA was validated by testing in 20 different blank and spiked matrices. Using device-independent color space for smartphone analysis, two different smartphone models were used for the analysis of optimized assays.

**Keywords:** flow-through immunoassay; lateral flow immunoassay; food allergen; multiplex; smartphone analysis; carbon nanoparticle labeling

#### **1. Introduction**

Food allergens are naturally occurring proteins present in a multitude of foods. Individuals with a food allergy are sensitized towards these proteins, and exposure to them can lead to adverse, sometimes life-threatening, health effects [1]. The majority of food allergen-related anaphylaxis in Europe can be attributed to peanut and tree nut allergens [2]. Allergies towards peanuts and tree nuts commonly co-exist, making the simultaneous detection of these problematic allergens desirable [3,4].

The only way for allergic individuals to avoid an allergic reaction is for them to stick to an avoidance diet. Such diets are largely reliant upon proper allergen labeling of food products. However, currently in the European Union (EU), only ingredients which have been intentionally incorporated into a food require labeling [5,6]. This means that allergens that are unintentionally present in food, such as via cross contamination, do not need to be declared, with all associated risks for allergic consumers. As a result, many food manufacturers use voluntary precautionary allergen labeling (PAL) (e.g., 'may contain' statements) in order to safeguard consumers [7].

In theory, PAL statements protect the consumer from potential allergic reactions; in reality the over-use of unregulated PAL has resulted in consumers choosing to ignore these warning statements [8]. Therefore, it is imperative to engage the public with their own food allergen analysis by developing consumer-friendly detection methods [9,10]. The cornerstones to consumer-friendly allergen detection are speed, sensitivity, ease-of-use, affordability, portability, multiplexing capability and a simple read-out system. Although some specifically consumer-oriented allergen sensors are available, such as the portable gluten and peanut sensors from NIMA, more often these biosensors are still proof-of-concept assays rather than commercial tests designed for consumers [11–14], and generally they lack multiplexing and proper validation as screening methods. A shared characteristic of novel allergen detection is the increasing trend to utilize a smartphone as an interface and readout system [9,14–17]. Using a smartphone readout improves the overall ease of result interpretation by introducing an interface that the consumer is already familiar with, alongside providing a means to wirelessly transmit results to relevant stakeholders, such as food manufacturers and restaurant personnel [18]. The Lateral Flow Immunoassay (LFIA) is widely considered the gold standard for easy-to-use, low-cost, sensitive and quick screening for food safety issues. Despite their widespread application, allergen LFIAs are often based on the analysis of a single analyte, owing to the difficulties associated with multiplexing an LFIA, including the need for careful design of test line configuration to prevent upstream detection areas from affecting downstream detection areas [19,20]. Most multiplex LFIAs for food safety focus upon the detection of low-molecular weight compounds, such as antibiotics and mycotoxins [21,22]. However, this past year has seen an increase in the development of multiplex food allergen detection LFIAs, with the development of an assay for the detection of hazelnut, ovalbumin and casein in bakery products within 10 min [23]. A further example is the multiplex, low-ppm detection of both β-lactoglobulin and β-casein, two major allergenic milk proteins, within 10 min [24].

A major drawback typically associated with LFIAs is the assay duration, which usually is 10–20 min, and is affected by mass transport limitations (MTL) and binding kinetics [25]. MTLs are caused by the fact that the target analytes need to be carried across a porous membrane, such as nitrocellulose (NC) by passive, capillary flow, and thus affect the detection speed of the assay [26]. The NC capillary flow rate is measured in the time in seconds it takes the sample front to travel 4 cm. Selection of NC based on this capillary flow rate is a compromise between assay sensitivity and assay speed with mid-speed membranes (120–150 s/4 cm) offering advantages in both areas [27]. When detection speed is not a constraint, a membrane with a slower flow rate and smaller pore size increases the available binding time between the labeled antibody–analyte and the test line antibody which can result in increased assay sensitivity [27–29]. In order to speed up LFIAs, in combination with NC with a good flow rate, antibodies with fast association rates towards their target should be used. Antibodies can be selected for their binding kinetics by in depth surface plasmon resonance (SPR)-based antibody screening and characterization. In this way a carbon nanoparticle-based hazelnut allergen LFIA has been developed, with a 30 s assay time, which as far as we know is a world record for allergen assay speed [30].

In order to overcome restrictions typically associated with LFIAs, a flow-through immunoassay format can be used instead [31,32]. Flow-through immunoassays are reported to offer the benefits of increased assay speeds, better sensitivities—owing to the use of larger sample volumes, excellent multiplexing capabilities and the absence of the 'hook-effect' [27,33,34]. The hook-effect is a phenomenon that is commonly encountered in one-step, sandwich format LFIAs. It occurs where the free analyte and the analyte which is bound to a labeled antibody compete for the limited number of binding sites available on immobilized capture antibodies, leading to a reduction in colorimetric signal and sometimes false negative results [24,35,36]. Therefore, if the correct assay working range is not determined, it could lead to consumers erroneously believing a food with a high allergen content is safe.

Flow-through assays can be prepared in different ways. Passive flow-through assays consist of LFIA materials, but in a stacked arrangement, with the membrane biofunctionalized with capture antibodies on top, and the conjugate and absorbent pads layered underneath or as flow-through enzyme-linked immunosorbent assays (ELISAs) [37–39]. An alternative flow-through approach is to insert a biofunctionalized membrane into a syringe filter holder, applying manual or mechanical pressure to the syringe to actively control the vertical flow of the reagents and the sample [40,41]. Although flow-through formats generally allow greater freedom in geometric assay design, they are prone to inter/intra-user variability [42].

The lack of agreed regulatory allergen thresholds has stalled the development of certified reference materials, preventing true comparisons to be made between various detection methods by different kit manufacturers and researchers [43]. Therefore, in this study, we use the same bioreagents to compare different geometrically designed, paper-based, flow-through and lateral flow immunoassay configurations for the simultaneous detection of hazelnut and peanut allergens with a smartphone readout system.

#### **2. Materials and Methods**

#### *2.1. Reagents and Consumables*

Washing buffer (WB) was composed of 5 mM borate buffer (BB) (pH 8.8) diluted from a mixture of 100 mM sodium tetraborate (VWR, Leuven, Belgium) and 100 mM boric acid (Merck, Darmstadt, Germany) and bovine serum albumin (BSA; Sigma-Aldrich, Zwijndrecht, the Netherlands) was added to a final concentration of 1% (*w*/*v*). Storage buffer (SB) consisted of 100 mM BB containing BSA to a final concentration of 1% (*w*/*v*). Running buffer (RB) was prepared by adding 1% BSA (*w*/*v*) and 0.05% Tween-20 (*v*/*v*) (Merck, Darmstadt, Germany) to 100 mM BB. Phosphate buffered saline (PBS; 0.01 M; pH 7.4) was purchased from Sigma-Aldrich (Sigma-Aldrich, Zwijndrecht, the Netherlands). All solutions were prepared with water from a MilliQ-system (MQ) (>18.2 MΩ/cm) purchased from Millipore (Burlington, MA, USA). 'Spezial Schwartz 4' carbon nanoparticles were purchased from Degussa AG (Frankfurt, Germany). Goat anti-mouse IgG in PBS (pH 7.6) (1.2 mg/mL; AffiniPure F(ab')2 Fragment GAM IgG Fcγ) used for spraying control lines/spots was purchased from Jackson Immunoresearch Laboratories Inc. (Sanbio, Uden, the Netherlands). The hazelnut (50-6B12) and peanut (51-2A12 and 51-12D2) antibodies were developed by Wageningen Food Safety Research (WFSR), Wageningen University and Research (Wageningen, the Netherlands) according to the procedure described by Bremer et al. [44]. All antibodies were buffer exchanged from PBS (pH 7.4) into 5 mM BB (pH 8.8) using Zeba™ Spin Trap columns (Thermo Scientific; Landsmeer, the Netherlands) prior to use. Passive flow-through assays were developed from a Miriad Rapid Vertical Flow toolkit (MedMira, Halifax, NS, Canada). All active flow-through assays were developed on unbacked Whatman 0.45 μm nylon (GE Healthcare, Eindhoven, the Netherlands) 0.45 μm NC or 0.2 μm NC membranes and inserted into 13 mm Swinny syringe filter holders (Merck, Darmstadt, Germany). The assembled filter holder was attached to a 10 mL syringe (Becton-Dickinson, Utrecht, the Netherlands). Lateral flow immunoassays (LFIAs) were developed on 140 CN nitrocellulose membranes (Unisart, Sartorius, Gottinghem, Germany) secured on a plastic backing (G and L, San Jose, CA, USA) overlaid with an absorbent pad (Whatman, GE Healthcare, Eindhoven, the Netherlands). All LFIAs were heat-sealed in foil packets with silica beads and stored at room temperature until use.

#### *2.2. Allergen Extraction*

Currently, a drawback in allergen detection is that no certified, standardized reference materials are commercially available, and antigen standards and blank matrices need to be prepared in-house [45]. The influence of food processing on the protein conformation of allergens can affect their detectability [46], but this was not explicitly investigated in this study, as the focus was comparing the performance of the same antibodies applied in different immunoassay formats.

Extracts were made from hazelnuts, peanuts, blank flour, peanut-spiked flour (8 ppm) and 20 truly different biscuits (i.e., 20 different brands and varieties; see Supplementary Information, Table S1) free from peanuts/tree-nuts, which were supplied by project partners or purchased from local supermarkets. Raw hazelnuts and unsalted peanuts were frozen whole at −80 ◦C for 1 h. The frozen foods were

homogenized using a commercial hand blender (Braun Turbo 600 W Food Processor, Braun, Oss, the Netherlands). A total protein extract was made by adding 10 mL PBS (pH 7.4) per gram of ground sample and incubating at room temperature for 1 h. Following incubation, extracts were centrifuged at 3220× *g* for 20 min. The extracts were then filtered through a series of low protein-binding syringe filters (5 μm > 1.2 μm > 0.45 μm), and the filtrate was aliquoted and stored at −20 ◦C until use. To ensure sample stability, fresh aliquots were defrosted daily for experiments, and protein concentrations were determined using the NanoDrop ND 3300 (Isogen Life Sciences, De Meern, the Netherlands) prior to use. Blank biscuits were homogenized by agitating 0.5 g in a 50 mL tube with ball bearings to a fine powder. Next, 5 mL of 100 mM borate buffer was added to the tubes and agitated for 1 min with the powdered biscuit or flour. The suspension was left at room temperature for 25 min. Afterwards, extracts were filtered through a series of low protein-binding syringe filters (5 μm > 1.2 μm > 0.45 μm), aliquoted and stored at −20 ◦C until use. All experiments, except for matrix experiments, were performed using total hazelnut protein (THP) and total peanut protein (TPP) spiked into running buffer. For matrix experiments, 1 μL of 1000 ppm THP and TPP extract was spiked into 999 μL (*v*/*v*) of the 20 different blank biscuit extracts.

#### *2.3. Carbon Black Nanoparticle Conjugation*

A 1% suspension of carbon nanoparticles (CNPs) was prepared by adding 1 mL of MQ Water to 10 mg carbon and sonicating for 10 min. The resulting 1% carbon suspension was diluted five times in 5 mM BB (pH 8.8) to obtain a 0.2% suspension, which was then sonicated for 5 min. Next, 350 μL purified hazelnut or peanut antibody solution (1 mg/mL in 5 mM BB) was added to 1 mL (to make a total volume of 1.35 mL) of 0.2% carbon suspension and stirred overnight at 4 ◦C. The suspension was split into approximately two equal aliquots (670 μL), and 500 μL of WB was added to each before centrifuging them for 15 min at 13,636× *g* at 4 ◦C. Following this, the supernatants were removed, and the pellets re-suspended in WB. This process was repeated three times. After the final wash, the supernatants were discarded, and the pellets were pooled together with 1 mL storage buffer and stored at 4 ◦C until use.

#### *2.4. Multiplex Passive Flow-through*

The plastic cartridge, biofunctionalized membrane and absorbent pad (absorption volume of 200 μL) from a Miriad Rapid Vertical Flow technology toolkit was used to create the passive flow-through assays. A schematic representation of the passive flow-through assay is shown in Figure 1A.

The membranes were biofunctionalized by manually depositing 0.5 μL of the peanut, hazelnut and control antibody solutions (1 mg/mL) in three distinct regions using a pipette. The tip of the pipette was touched very lightly against the membrane to dispense a consistent antibody spot. The membranes were dried for 45 min. Once dried, three drops of RB were added via a dropper bottle and allowed to saturate the membrane. Immediately after, 50 μL of the mixed allergen extract (diluted in RB; 1000 ppm, 100 ppm, 10 ppm, 1 ppm, 0.1 or 0 ppm) was pipetted dropwise onto the membrane and allowed to absorb fully. Next, a 10 μL suspension of 10 × diluted carbon labeled-monoclonal antibodies (CNP-mAbs) was pipetted onto the membrane and allowed to absorb fully. Finally, three drops of RB were applied to wash the membranes. The assays were read immediately with the naked eye and an image was acquired with a smartphone camera. LOD values for visual inspection were established at the lowest concentration that reproducibly yielded a signal that could be observed and distinguished from the background by the naked eye.

**Figure 1.** Schematic representation (not to scale) of the three flow assay formats developed. Arrows depict the flow direction and C is the control antibody (goat anti-mouse), H is the anti-hazelnut antibody and P is the anti-peanut antibody. Total hazelnut protein (THP) is indicated by the hazelnut graphic and total peanut protein (TPP) is indicated by the peanut graphic. (**A**) The passive flow assay in top-view and side-view. (**B**) The active format flow-through assay, where the syringe filter holder is enlarged, and the membrane is further enlarged to show the biofunctionalized area. (**C**) Both lateral flow immunoassay geometries as defined by the order in which sample will encounter the test and control lines: Peanut, hazelnut, control (PHC) and hazelnut, peanut, control (HPC).

#### *2.5. Multiplex Active Flow-through*

A schematic representation of the active flow-through assay is shown in Figure 1B. First, the most appropriate assays parameters were established including membrane type, pore size, antibody concentration for dispensing and assay conditions.

#### 2.5.1. Simplified Multiplex Flow-through

Allergen-specific antibody solutions (0.5 μL of 1 mg/mL mAb solution) and control antibody solution were manually dispensed by lightly touching the tip of the pipette to the membrane onto 0.2 or 0.45 μm pore size unbacked NC or 0.45 μm unbacked nylon membranes. The membranes were dried for 45 min and then the membranes were placed in 13 mm syringe filter holders and attached to the 10 mL syringe. The assays were performed by manually and sequentially injecting 500 μL sample (concentration series 100–0.1 ppm total protein extract diluted in RB), 1 μL of each CNP-mAb and another 300 μL of RB as a washing step. In this context, sequentially refers to the sequential loading of the syringe with sample with the CNP-mAbs on top of the sample; these were then pushed through by moving the plunger downwards in a single movement, followed by a final washing step with RB. The membranes were then removed from the filter holder, dried for 5 min, read with the naked eye and an image was acquired with a smartphone camera.

#### 2.5.2. Multiplex Flow-through Iterative Optimization

To establish the optimum active flow-through conditions, a number of alternative assay steps were explored. The experiments aimed to reduce background staining, to improve the signal-to-noise ratio and to improve the assay sensitivity.

#### 2.5.3. Volume Optimization

Different sample and reagent volumes were tested to determine the optimum conditions for flow-through operation. Flow-through assays require larger sample volumes compared with LFIA due to reduced contact time between analyte and capture antibodies [42].

When using sample volumes of less than 500 μL, it was necessary to first 'pre-wet' the membrane with running buffer to ensure that the entire surface would be wetted. Initially, membranes were tested using 500 μL RB, followed by a 300 or 500 μL sample and 0.5 μL of each of the CNP-mAbs solutions followed by 500 μL RB as a washing step. In subsequent experiments, the volume of the CNP-mAb solution was increased to 1 μL for each CNP-mAb to maximize the signal intensity. Finally, experiments were performed using 1 mL of sample, with 1 μL of each CNP-mAb solution dispensed on top of the sample, followed by 500 μL RB.

#### 2.5.4. Pre-Mix Method

The assays were tested by pre-mixing the running buffer and CNP-labeled secondary mAbs with sample and injecting the mixture simultaneously. In this approach, 1 mL of sample, 1 mL of RB and 1 μL of each CNP-mAb were injected across the membrane, effectively causing an additional 50% dilution to the sample, when compared to the sequential method described above. The holder was then dismantled, and the membrane dried for 5 min before visual inspection.

#### 2.5.5. Filter Approach

To improve the uniform wetting of the membrane and reduce the background staining caused by the CNPs, a filter approach was tested. In this method, a 0.45 μm NC filter was placed on top of the functionalized membrane before carrying out the assay sequentially. Following the final wash step, the device was dissembled, the 0.45 μm filter carefully removed and disposed of and the membrane dried for 5 min before visual inspection.

#### 2.5.6. Aspiration Approach

To ensure sufficient wetting of the membrane, and to increase the contact time of the sample and the capture antibodies, an iterative aspiration approach was applied. In this way, when sequentially injecting the sample and CNP-mAbs, the plunger of the syringe was pumped up and down, 1, 5 or 10 times. With the increasing number of aspirations, the flux of the analyte past the membrane, and thus past the immobilized antibodies, was increased. After the final aspiration, the RB was flowed through as a washing step, the device was disassembled, and the membrane dried for 5 min before visual inspection and photographing with a smartphone camera.

#### 2.5.7. Multiplex Array Layout

The flow-through array was spotted using the XYZ 3060 BioDot Dispense Platform (Irving, CA, USA). The array was composed of 14 (2 × 7 array) control spots (0.25 mg/mL) and with each analyte having 12 (2 × 6 array) spots (0.25 mg/mL), with a drop size of 100 nL and an offset of 1 mm between each dot (see Figure 1B). The membranes were left to dry overnight prior to testing.

#### 2.5.8. Optimized Active Flow-through Operation Protocol

A 0.45 μm NC filter, acting as a vertical flow diffuser, was placed on top of the biofunctionalized membrane. The filter and membrane were then placed, biofunctionalized side up, into the syringe filter holder. A polytetrafluorothylene (PTFE) gasket was placed on top of the membrane to seal the fluid pathway, giving the assay an actual flow path of 10 mm. The syringe holder was then attached to a 10 mL Luer-Lock™ syringe. The assay was performed sequentially as described in Section 2.5.1. First, 1 mL of sample topped with 1 μL of each CNP-mAb solution was aspirated 10 times across the membrane (only THP or only TPP or mixture of both diluted in RB at 100, 10, 1, 0.1 and 0 ppm). Following this, 500 μL RB, as a washing buffer, was flowed through the membrane. Finally, the syringe filter holder was disassembled, and the membrane removed and placed on an absorbent pad for drying. To determine whether the immobilized test antibodies suffered from non-specific binding towards the other target, the assays were tested using just THP or just TPP extract spiked into RB.

Blank buffer measurements were performed 10 times to test for false positives. The membranes were visually inspected and photographed with a smartphone camera after 5 min. LOD values for visual inspection were established at the lowest concentration that reproducibly yielded a signal that could be observed and distinguished from the background by the naked eye.

#### *2.6. Multiplex Lateral Flow Immunoassay*

Lateral flow immunoassays were manufactured using NC (flow rate of 140 s/4 cm) cut to approximately 4 cm length. The NC membrane was secured on a plastic backing, with 4.5 cm of absorbent pad overlapping one end of the NC. Two different test line configurations (as depicted in Figure 1C) were designed and produced using the XYZ BioDot dispensing platform. The first configuration had the control line (0.25 mg/mL) dispensed at 10 mm from the absorbent pad, the hazelnut line (0.25 mg/mL) at 5 mm from the control line and the peanut line at 5 mm from the hazelnut line, with 10 mm of blank membrane at the bottom of the strip, hereafter referred to as PHC. The second arrangement had the control line at 10 mm from the absorbent pad, the peanut line at 5 mm from the control line and the hazelnut line at 7 mm from the peanut line with 8 mm of blank membrane at the bottom of the strip, hereafter referred to as HPC.

#### Multiplex LFIA Operation Protocol

Firstly, the multiplex LFIAs were tested for non-specific binding by testing 10 × each of the LFIAs in blank running buffer (RB). The LFIAs were inserted into individual microwells of a 96-well plate containing 1 μL of each of the CNP-mAbs and 100 μL of RB (blank). The strips were left to run for 5 min. Next, the LFIAs were tested for specificity by testing in either just THP or TPP extract spiked into RB. LFIAs were placed into the individual microwells of a 96-well plate containing either just THP or TPP (1 μL) spiked into RB, in decreasing concentration with RB (99 μL) and 1 μL of each carbon-labeled mAb. The strips were left to run for 5 min before photographing with a smartphone camera. Finally, the assays were tested using the same conditions in decreasing concentrations (100, 10, 1, 0.5, 0.1 ppm) of both THP and TPP spiked in RB (in triplicate). Calibration series were tested with both formats of the LFIA using (i) 1 μL of sample (diluted in RB) and 99 μL of RB (hereafter, 1:99, sample: RB), (ii) 25 μL of sample (diluted in RB) and 75 μL of RB (hereafter, 25:75, sample: RB), and (iii) 75 μL sample (diluted in RB) and 25 μL of RB (hereafter, 75:25, sample: RB). The 75:25 sample: RB experiments were specifically designed to trigger the hook-effect to determine when the sample volume becomes the limiting factor.

The membranes were visually inspected and photographed with a smartphone camera after running for 5 min. LOD values for visual inspection were established at the lowest concentration that reproducibly yielded a signal that could be observed and distinguished from the background by the naked eye.

#### *2.7. Smartphone Readout and Data Analysis*

Smartphone photographs were acquired using Open Camera (version 4.0.3) and analyzed using a Huawei P20 smartphone (Huawei Technologies, Shenzen, China) according to the method developed by Ross et al. [27] using two freely downloadable apps from the Google Play Store. The red, green, blue (RGB) values were obtained for test regions of assays using the RGB Color Detector (version 1.0.58). Using the crosshair function in the app, test dots on the flow-through membrane or three distinct regions on the test line of the LFIA were selected and the color values were averaged and recorded. Background measurements were also made above and below the test areas to determine an overall background level for subtraction from results. Alternatively, results were normalized by dividing the value of each test region by the corresponding control region, as has been performed in literature [35,47,48]. Using 'Nix Pro Color' (version 1.31), the RGB values were converted to luminosity, A, B (LAB) values; a device-independent color space that more accurately represents how humans interpret color intensity.

Additionally, to show the device-independent nature of LAB measurements, the optimized assays were also analyzed using a Google Pixel 2 XL smartphone (Google, Mountain View, CA, USA). The obtained values were used to plot calibration curves for L (luminosity) of the LAB values as a function of allergen concentrations spiked into RB, using Microsoft Excel. LOD values were obtained from these calibration curves by visual evaluation.

#### *2.8. Matrix Experiments and Validation*

To validate the assays, they were also tested in spiked food matrices. All assays were tested in a decreasing concentration of THP and TPP, spiked directly into a blank biscuit matrix extract to determine the matrix effects. Additionally, the optimized LFIA (PHC) was more extensively validated by testing in 20 truly different blank matrix extracts. In this way, LFIAs were placed in individual microwells containing 25 μL blank matrix extract (*n* = 20) and 75 μL RB and left to run for 10 min to determine whether any false positives occurred. Additionally, 1 ppm of THP and TPP was spiked into the 20 different blank matrix extracts (1 μL of 1000 ppm THP and TPP sample into 999 μL (*v*/*v*) blank matrix extract) and the LFIAs were tested using both 25 μL spiked matrix plus 75 μL RB and 1 μL spiked matrix extract plus 99 μL RB. Assays were left to develop for 10 min. Finally, the optimized LFIAs were also tested in blank flour matrix extract and spiked peanut flour matrix extract in both 25:75 and 1:99 dilutions in RB.

#### **3. Results and Discussion**

#### *3.1. Multiplex Passive Flow-through Assay*

An overview of conditions, quantitative and qualitative results for spiked buffer experiments for the passive flow-through assay, can be found in Table 1. The visual limit of detection (LOD) for the passive flow-through was established by testing in decreasing concentrations of THP and TPP extracts spiked in RB. The visual LODs were determined as 0.1 ppm and 1 ppm and smartphone LODs 1 and 10 ppm for hazelnut and peanut, respectively (*n* = 3), whereas no visible spot was obtained for blanks (see Table 1 and Supplementary Information, Figure S1A). Following the addition of the CNP-mAbs to the passive flow-through assay, the positive spots appeared within 5 s, a detection speed which is unparalleled by LFIA. Even when using the high-speed LFIA described in [30] the appearance of the positive result took 30 s, due to MTL limitations of the solution that needs to wick through the membrane before reaching test lines. Three drops of RB were added to the flow-through assay to wash the unbound CNPs from the membrane. Using dropper bottles with pre-defined drop volumes for the delivery of RB makes the assay easy to perform and means that pipettes are unnecessary. A further benefit is that the result can be directly read through the window of the cassette by the naked eye without having to disassemble the device. However, when recording a smartphone image of the membranes, these do need to be removed from the plastic cassette to avoid shadowing. Despite the washing step, the membranes had variable background staining, which made it impossible to obtain calibration curves from the images acquired with a smartphone. The reason for the appearance of background staining probably lies with the polydispersity of the CNP, which can form aggregates of several hundred nm, which are too large to be flowed through the pores. A drawback of this specific passive flow assay format is the lack of freedom in geometric assay design as bio-reagents required manual spotting by pipette. However, such a limitation could be easily overcome by biofunctionalization of the membranes before having them cut to the factory-made circular size.


ComparisonofoptimizedFlow-throughandLateralFlowparameters(RB

 measurements were made using total hazelnut protein and total peanut protein (THP and TPP) spiked into running buffer (RB). Where the peanut, hazelnut, control geometrydefinedbyPHCandthehazelnut,peanut,controlgeometryisdefinedbyHPC.

#### *Biosensors* **2019** , *9*, 143

#### *3.2. Multiplex Active Flow-through*

An overview of conditions, quantitative and qualitative results for spiked buffer experiments for the active flow-through assay can be found in Table 1. The assays using the 0.45 μm pore size nylon and NC membranes were ineffective, and no spots (including control spots) appeared on these membranes. This can be attributed to 0.45 μm being too large a pore size and the majority of the analyte and labeled antibodies passing through the membrane, which is confirmed by the dark coloration of the waste liquid when using this assay membrane. Therefore, the 0.2 μm pore size NC membrane was determined to be the most suitable for this application.

During the optimization steps, active flow-through assays were tested using 0.5 μL of each CNP-mAb solution, but this only yielded faint detection spots. In subsequent experiments the volume of the CNP-mAb solution was increased to 1 μL of each CNP-mAb which improved the readability. Additionally, volumes of 500 μL and 1 mL of sample were tested, with the sensitivity improving with the increased sample volume, without the appearance of a hook-effect, even at high concentrations. Although in this manually spotted initial format, LODs of 0.5 and 0.1 ppm could be reached for peanut and hazelnut (see Supplementary Information, Figure S1B), respectively, false positives were also detected when testing the assays in a blank sample (1 in 5 false positives). Using a pre-mix approach did improve the overall user-friendliness of the assay, as the operator only needed to pass the liquid containing the sample, CNP-labeled mAbs and RB through once without the necessity of removing and reinserting the plunger, but this method consistently resulted in false positives in the blank samples. Contrastingly, using the sequential method increased the difficulty of the assay, but prevented false positives owing to the washing step at the end. The addition of a 0.45 μm NC filter on top of the biofunctionalized membrane increased the (smartphone) readability of the assay. Besides filtering the larger sized CNPs, reducing the level of background staining, the filter also acted as a flow diffuser. In this way, uniform wettability of the membrane was achieved, resulting in better reproducibility compared to when it was performed without the filter. Although the filter improved the readability of the membranes, it also further complicated the user-friendliness of the method, as it needed to be carefully removed from the biofunctionalized membrane before the results could be read.

The sensitivity of the assay was improved by increasing the number of sample aspirations across the membrane (see Supplementary Information, Figure S2). Flow-through assays are subject to unidirectional flow and require capture antibodies with rapid association rates in order to achieve binding or require extended sample/reagent incubation times [48]. By increasing the number of sample aspirations, the flux of the CNP-mAb-analyte complex past the immobilized antibodies, and the potential of binding, is increased. Of all the tested parameters the most appropriate assay conditions were determined to be a 0.45 μm filter on top a 0.22 μm NC membrane biofunctionalized with 0.25 mg/mL control and test spots and aspirating 1 mL of sample with 1 μL of CNP-mAb solution 10 times back and forth through the membrane. Subsequently, 500 μL of RB was injected as a washing step. Although these conditions allowed for the assay to reach very low LODs, they also meant that this method generated a high volume of chemical waste (1.5 mL), which needs to be safely disposed of.

When testing active flow-through membranes in decreasing concentrations of THP and TPP spiked into RB, visual LODs of 0.05 ppm (*n* = 3) could be reached for both targets, an LOD which is so far un-met by commercially available allergen assays [8]. This LOD is less obvious from the smartphone image (LODs of 0.5 ppm for both THP and TTP) compared with reading by naked eye (see Figure 2). Therefore, eye symbols are inserted in Figure 2 to designate the lowest concentration that could still be read visually. Despite the active flow-through approach reaching lower LODs than the passive flow-through assay, the assay was more complicated to perform and used a far greater sample volume.

**Figure 2.** Active flow-through assay calibration range. Assays were tested in decreasing concentrations (100–0.05 ppm) of Total Hazelnut Protein (THP), Total Peanut Protein (TPP) spiked into Running Buffer (RB) and in blank RB. The control region is indicated by C and outlined in red, the hazelnut region by H and outlined in dark blue and the peanut region by P and outlined in light blue. There is an evident decrease in test dot intensity as the concentration of total protein in the sample decreases. The eye icon is used to indicate test regions that are visible to the naked eye but more difficult to read in the smartphone image. The visual limit of detection is established at 0.05 ppm for both analytes.

#### *3.3. Multiplex Lateral Flow Immunoassay*

An overview of conditions, quantitative and qualitative results for spiked buffer experiments for the LFIAs can be found in Table 1. The LFIAs were both able to achieve single analyte detection and a true blank result every time (0% false positives at 0 ppm; *n* = 10). When testing PHC with 1 μL of sample, 1 μL of each CNP-mAb and 99 μL of RB, visual LODs of 1 and 5 ppm were achieved by the naked eye (see Figure 3A) for hazelnut and peanut, respectively, with a clear decrease in intensity in the test line with decreasing concentration of the sample. When the LFIAs have a low signal intensity, the naked eye is still superior at distinguishing between a positive or negative signal, and the lower visual LODs are indicated by the eye icon in Figure 3. However, these visual readings are performed by a trained person, and the distinction between signal and no signal at the lowest concentrations is not trivial. In comparison, when the same anti-hazelnut antibody was applied in a single-plex LFIA, an LOD of 0.1 ppm in spiked buffer was reached, which suggests that having an additional test line on the LFIA can compromise the overall sensitivity [30]. Still, the multiplex LODs are in accordance with commercially available allergen single-plex LFIAs, which report LODs within this range. However, lack of standardized, certified reference materials in the allergen industry means that each reported assay is developed using antibodies specific to different allergenic components (total soluble protein vs. allergen-specific proteins) and tested and validated using different analytes [9,45], thus underlining that true comparisons can only be made when bioreagents and samples are kept constant, as in this research. To optimize the multiplex LFIA and improve the LOD, the sample volume was increased to 25 μL (diluted in RB) in 75 μL RB. By increasing the sample volume to 25 μL (thus concentrating the sample 25 × compared with the 1 μL sample volume) LODs of 0.1 and 0.5 ppm for hazelnut and peanut were reached respectively (see Figure 3B).

**Figure 3.** Calibration range (100–0.05 ppm) of Total Hazelnut Protein (THP), Total Peanut Protein (TPP) spiked into Running Buffer (RB) and blank RB, where the control line is indicated by C, the hazelnut test line by an H and the peanut test line by a P. A positive result can be still read with the naked eye, but is difficult to see in the smartphone image, thus an eye icon has been used to indicate the visual LOD. (**A**) Peanut, Hazelnut, Control (PHC) line configuration using 1 μL of spiked sample and 99 μL RB. (**B**) PHC using 25 μL of spiked sample and 75 μL RB. (**C**) PHC using 75 μL of spiked sample and 25 μL RB.

Despite the assay sensitivity improving with the increased sample volume, with these conditions at concentrations of 100 ppm and higher, a reduction of the intensity of the upper line (hazelnut) could be observed, as has been witnessed by Galan-Malo et al. [24]. Although this was not considered a false negative, as three distinct lines were still clearly visible, it did warrant further exploration into the extent of the hook-effect in more concentrated samples.

To further investigate the extent of the hook-effect and its potential to limit the upper dynamic range of the LFIA assay, the PHC format was also tested in 75 μL of sample extract diluted with 25 μL RB (see Figure 3C). These conditions resulted in a more pronounced hook-effect with LFIAs tested at 1000 ppm appearing to be false negatives, and at 100–50 ppm exhibiting decreased test line signals. As well as just testing high analyte concentrations, it is important to test different sample-to-RB ratios, as increasing sample volume has a noteworthy influence on the appearance of the hook-effect. In order to avoid the hook-effect it is imperative to use the correct volume of diluted sample. Despite this, PHC in the 75:25 conditions did achieve a lower LOD of 0.05 ppm for both analytes in RB. Therefore, PHC could still be used with 75:25 conditions for testing trace allergen levels, so long as the sample is also tested in the 1:99 and 25:75 conditions to ensure no false negatives arise at high concentrations. The optimum conditions from PHC were determined to be 25:75. When testing HPC in the 1:99 conditions, LODs of 5 and 1 ppm (see Supplementary Information, Figure S3) were reached for peanut and hazelnut, respectively, with the LODs decreasing to 1 and 0.1 with the 25:75 arrangement. But for HPC, the hook-effect was greater in 25:75 compared with PHC with concentrations of 100 and 50 ppm experiencing reduced intensity on both the control and the peanut lines, complicating quantitative analysis. The larger hook-effect in this configuration could be because the upstream (hazelnut) test line comes into contact with the sample first, and this mAb has a rapid association rate and high affinity for THP, and so it becomes quickly saturated [30].

So, the optimum condition for HPC was the 1:99 protocol, although this was significantly less sensitive compared with the optimized PHC assay. For this reason, PHC was determined

to be the optimum test line configuration with the best working conditions being 25:75 in the working range of 100–0.1 ppm. Therefore, PHC was used for further smartphone quantification and validation experiments.

#### *3.4. Smartphone Readout and Analysis*

Smartphones are ever-increasing in popularity for analyzing colorimetric assays. Most often, smartphone analysis is based on specific apps which relate a particular color intensity to a certain concentration of analyte. In the absence of a specific app, it has been shown by Ross et al. [30] that it is possible to use freely downloadable apps from the Google Play Store to analyze endpoint, smartphone image color intensity values. By converting RGB values to LAB values, luminosity or intensity can be plotted as a function of concentration in a calibration curve. In sandwich immunoassay formats with CNP labels, a higher L value corresponds to a lower analyte concentration. As LAB color space is device-independent, the same results can be potentially achieved using different smartphone models. For analysis of PHC and HPC (in triplicate) the normalization of the (L)LAB values was carried out by dividing the L values of the test lines by the L values of the control lines. The method of dividing the test line response by the control line response (T/C ratio) is a technique commonly used for the quantification of sandwich LFIAs [35,47–49]. The results for PHC can be found in Figure 4, and the HPC smartphone calibration curve can be found in Figure S4 in Supplementary Information.

Two smartphone models were used for the device independent LAB analysis of PHC assays (in RB in triplicate), as can be seen in Figure 4 where A, C and E show the curves for THP in 1:99, 25:75 and 75:25 (sample: RB) and B, D and F show the curves for TPP in 1:99, 25:75 and 75:25 (sample: RB). A higher normalized L value was obtained for hazelnut at 25–100 ppm using the 25:75 conditions, as can be seen in Figure 4C. Comparatively, peanut did not appear to be subject to the hook-effect under 25:75. Using 75:25 conditions (see Figure 4E), concentrations of 50 and 100 ppm resulted in a higher normalized L value for hazelnut (i.e., weaker signal). Furthermore, under these conditions the hazelnut T/C ratio for 10 ppm and 25 ppm gave the same normalized L value, highlighting that the hook-effect was still evident, even at these lower concentrations. Comparatively, peanut in 75:25 (see Figure 4F) gave higher normalized L values at concentrations of 25–100 ppm, again indicating with increasing sample volume and concentration the likelihood of the hook-effect being increased. The only crucial variation between the two smartphone measurements using the different models was obtained for the peanut line using 75:25 at 0.1 ppm (see Figure 4F). However, this is the smartphone LOD, and detection spots were already more difficult to read. As well as this, the current method relies on manually selecting regions of interest on the control and test lines, rather than being able to read the values across the whole line. Therefore, please note that the results also include any errors due to not selecting the exact same regions, and this can also cause variations in the obtained color values.

Additionally, to compare different smartphone quantification methods, all smartphone readable assays were also analyzed by making a background subtraction as can be seen in Supplementary Information (see Figures S5 and S6). However, when analyzing the LFIAs in this way the differences in background readings, due to discrepancies in lighting conditions caused by recording an image of the entire calibration range simultaneously under ambient lighting conditions, meant that a simple background subtraction was insufficient. However, for active flow-through assays the background subtraction was found to be the most effective analysis method (see Supplementary Information, Figure S6A), whereas the T/C method resulted in larger standard deviations (see Supplementary Information, Figure S6B). This could be attributed to the membranes being photographed independently, so the small membranes were subject to the same ambient lighting conditions and did not have such variable background readings. By using two data processing methods it is evident that the selected data processing method plays a crucial role for the quality of the semi-quantitative information that can be obtained from raw results.

**Figure 4.** Smartphone calibration curves for the normalized (L) LAB values of the test lines of a Peanut Hazelnut Control (PHC) assay as a function of the concentration of Total Hazelnut Protein (THP), and Total Peanut Protein (TPP) (100–0.1 ppm) tested using two different smartphone models. All calibration ranges were performed in triplicate in spiked Running Buffer (RB). All L(LAB) values have been normalized by dividing the test line values by the control line values. (**A**) Hazelnut tested in 1 μL of sample in 99 μL of running buffer (RB) (**B**) Peanut tested in 1 μL of sample in 99 μL of RB. (**C**) Hazelnut tested in 25 μL sample in 75 μL of RB. (**D**) Peanut tested in 25 μL sample in 75 μL of RB. (**E**) Hazelnut tested in 75 μL sample in 25 μL of RB. (**F**) Peanut tested in 75 μL of sample in 25 μL of RB. Error bars show standard deviation (SD) from triplicate measurements.

#### *3.5. Matrix Experiments and Validation*

To determine their applicability to real life samples, the assays were tested using THP and TPP spiked into blank biscuit matrix extracts. The passive flow-through format was able to achieve visual LODs of 5 and 1 ppm for peanut and hazelnut. These LODs are higher than previously observed in spiked buffer experiments, showing that the matrix extract did have some influence on the detection of the analytes. When testing in this way, the passive flow membranes had greater background staining compared with in spiked buffer experiments. This can be attributed to the overall reduction of reagents, BSA and tween-20 in the assay buffer, as the sample was spiked into a matrix extract rather than into the RB.

In comparison, the active flow-through membranes did not suffer with increased background staining due to the use of the additional filter on top of the membrane and subsequent washing step. The active-flow assay reached visual LODs of 0.5 and 1 ppm for THP and TPP in spiked matrix extract, however the intensity of the detection spots was fainter compared with spiked buffer samples because of the reduction of buffer reagents responsible for good flow.

Therefore, whilst visual readout was possible, the construction of calibration curves based on smartphone images could not be achieved.

PHC was tested in both 25:75 and 1:99 of spiked matrix in RB to determine the visual LOD in matrix extract, as can be seen in Supplementary Information Figure S7. When using 25 μL sample (THP and TPP spiked into matrix extract) and 75 μL RB a LOD of 0.5 ppm could be reached for both analytes (see Supplementary Information, Figure S7A). At higher concentrations (100 ppm +) there was decreased intensity for the hazelnut line. This can be attributed to the hook-effect. For the spiked matrix extract experiments, the PHC assays were run for 10 min, due to the reduction of reagents BSA and tween-20 from spiking sample into matrix extract rather than RB, affecting the flow of the sample. Additionally, PHC was tested in 1 μL of spiked matrix extract:99 μL of RB (see Supplementary Information, Figure S7B). Visual LODs of 10 and 5 ppm were reached for peanut and hazelnut, respectively. The PHC assay was fully validated using 25:75 conditions by evaluating 20 truly different blank matrices and determining that no false positives occurred. Additionally, the 20 blank matrices were spiked with 1 ppm THP and TPP. In the absence of agreed regulatory levels for food allergens, a screening target concentration (STC), based on VITAL 2.0 levels of 1 ppm, was selected [8,50]. The LFIAs were able to detect the allergens with both visual and smartphone readout at 1 ppm in all 20 samples, as can be seen in Figure 5 and as is summarized in Table 2. The excellent reproducibility at the STC level clearly suggests that a simple device-independent smartphone readout may provide semi-quantitative data.


**Table 2.** Matrix experiments for the optimized PHC assay, all measurements made in spiked matrix extract.

\* PHC = Peanut, hazelnut, control geometry lateral flow immunoassay. \*\* Reproducibility defined as Relative Standard Deviation (RSD) × 100% 1 ppm of Total Hazelnut Protein (THP), Total Peanut Protein (TPP) spiked into blank biscuit matrix extract (*n* = 20). Data based on normalized L (LAB) values.

Finally, to confirm the capability of the optimized LFIA in detecting allergens in raw ingredients, blank flour and peanut-spiked flour samples were briefly tested. The LFIAs correctly did not detect either of the allergens in the blank flour (*n* = 4). Furthermore, PHC specifically detected only peanut

in the peanut-spiked flour (*n* = 4) with no false hazelnut positives being observed. The detection of peanut was not adversely affected by using the accelerated 30 min extraction procedure for the spiked flour. Further developments should include simplified and faster extraction methods.

**Figure 5.** Smartphone validation of Peanut Hazelnut Control (PHC) assay using 20 truly different blank biscuit samples (square markers) and 20 truly different biscuit samples spiked at the screening target concentration of 1 ppm Total Hazelnut Protein (THP), and total peanut protein (TPP). Normalized L (LAB) values were obtained by dividing the test line response by the corresponding control line response.

#### **4. Conclusions**

Quick and accurate detection of food allergens is of critical importance for food safety; it is particularly relevant if such testing procedures can be easily performed by the consumer, and therefore, there is an evident requirement for simple and robust testing procedures. Two formats of multiplex flow-through immunoassays have been developed and compared with two test line configurations of LFIA, all developed using the same bioreagents and against the same targets in order to allow a true comparison.

Two recent review papers have extensively outlined commercially-available and proof-of-concept single-plex and multiplex allergen immunoassays and biosensors, and the assays reported in this study have matched or surpassed these previously-reported LODs [9,51]. All the developed multiplex assays were able to detect both analytes in the low ppm range within minutes. It is important to note here that our screening concentrations always related to total protein extracts from either peanuts or hazelnut, and therefore, the concentration of specific allergenic proteins is expected to be even lower than the reported values. This in turn means that the reported LODs are underestimating the true sensitivity of the immunoassays in this work. The passive flow-through format offered a way to rapidly develop a fast flow-through assay. However, this specific format was limited by the need to manually biofunctionalize the membranes, limiting their reproducibility. The active flow-through assay could achieve very low limits of detection with no false negatives when following the optimization steps. However, it is these optimization steps that made the assay more complicated to perform for a non-expert user such as a consumer. In future versions, the use of a mechanical pump could improve the user-friendliness, although this would introduce an additional and costly element into the procedure, limiting the portability of the assay. It should be reiterated that the assays within this study were performed by a trained scientist, and the active flow-through method is not recommended for untrained users. In comparison, the LFIAs, when using the optimized assay conditions for each configuration, resulted in

no false positives. However, outside of working conditions, both configurations of LFIA did experience a hook-effect at high concentrations, a phenomenon commonly encountered in sandwich LFIA, where a falsely low signal occurs at high analyte concentrations. As the hook-effect is concentration-dependent, it can be avoided/limited by assay optimization.

To demonstrate their applicability to real life bakery products and raw ingredients, all assays were tested in decreasing concentrations of analyte spiked into the matrix extract. Additionally, the PHC assay was validated as a screening method in spiked matrix extract, blank matrix extract (*n* = 20) and incurred spiked flour, proving its capability of detecting the target even in complex matrices. The majority of commercially-available allergen detection LFIA test kits can detect a single analyte at 1–10 ppm [9,51]. Comparatively, PHC was able to detect both analytes at 0.5 ppm of THP and TPP spiked into a blank biscuit matrix extract, affirming its place as one of the most sensitive allergen LFIAs. This LOD was in agreement with the LOD using the same hazelnut antibody in a previously-reported single-plex assay [30]. Finally, all assays were (semi-)quantified by smartphone readout. At this stage no additional external equipment was used for the image recording, so the LFIA membranes were subject to ambient lighting conditions. To compensate for the lighting conditions a normalization factor (T/C ratio) was applied. By using device-independent (L)LAB values, it was possible to obtain comparable results using two distinct smartphone models. The ability to use different smartphone models for reading the same assays is a characteristic that is highly desirable, but not often reported, within smartphone analysis. In future developments, researchers should focus on improving the ease of use of these assays by integrating sample preparation, limiting the user interaction with the assay, as well as by developing a consumer-friendly app as a user interface which can directly analyze data with minimal user input.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2079-6374/9/4/143/s1, Figure S1: Calibration range for multiplex flow-through assays using passive and active flow. Figure S2: Calibration range for multiplex flow-through assay optimization: Sample aspirations. Figure S3: Calibration range for lateral flow immunoassay with test line configuration: Hazelnut, peanut, control (HPC). Figure S4: Calibration curve for smartphone analysis of hazelnut, peanut, control (HPC) lateral flow immunoassay. Figure S5: Calibration curve for smartphone analysis of lateral flow immunoassays using background subtraction. Figure S6: Calibration curve for smartphone analysis of active flow-through immunoassay. Figure S7: Optimized lateral flow immunoassay calibration curve in spiked matrix extract. Table S1: Ingredient and allergen information for the 20 varieties of biscuit used for matrix experiments.

**Author Contributions:** Conceptualization, G.M.S.R. and G.I.S.; methodology, G.M.S.R.; software, G.M.S.R.; validation, G.M.S.R. formal analysis, G.M.S.R.; investigation, G.M.S.R.; resources, G.M.S.R., G.I.S. & M.W.F.N.; data curation, G.M.S.R.; Writing—Original Draft preparation, G.M.S.R.; Writing—Review and Editing, G.M.S.R., G.I.S. & M.W.F.N.; visualization, G.M.S.R.; supervision, G.I.S. & M.W.F.N.; project administration, M.W.F.N.; funding acquisition, M.W.F.N.

**Funding:** This project has received funding from the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No. 720325.

**Acknowledgments:** The authors would like to thank project partner Barilla for providing test materials for blank and spiked flour samples.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*
