key: cord-0716912-ekc3jo5p
authors: Anh, Nguyen Ha; Doan, Mai Quan; Dinh, Ngo Xuan; Huy, Tran Quang; Tri, Doan Quang; Ngoc Loan, Le Thi; Van Hao, Bui; Le, Anh-Tuan
title: Gold nanoparticle-based optical nanosensors for food and health safety monitoring: recent advances and future perspectives
date: 2022-04-07
journal: RSC advances
DOI: 10.1039/d1ra08311b
sha: dcb4ab28f10fd4856196a17a74fb8bcb7a32aa41
doc_id: 716912
cord_uid: ekc3jo5p

Modern society has been facing serious health-related problems including food safety, diseases and illness. Hence, it is urgent to develop analysis methods for the detection and control of food contaminants, disease biomarkers and pathogens. As the traditional instrumental methods have several disadvantages, including being time consuming, and having high cost and laborious procedures, optical nanosensors have emerged as promising alternative or complementary approaches to those traditional ones. With the advantages of simple preparation, high surface-to-volume ratio, excellent biocompatibility, and especially, unique optical properties, gold nanoparticles (AuNPs) have been demonstrated as excellent transducers for optical sensing systems. Herein, we provide an overview of the synthesis of AuNPs and their excellent optical properties that are ideal for the development of optical nanosensors based on local surface plasmon resonance (LSPR), colorimetry, fluorescence resonance energy transfer (FRET), and surface-enhanced Raman scattering (SERS) phenomena. We also review the sensing strategies and their mechanisms, as well as summarizing the recent advances in the monitoring of food contaminants, disease biomarkers and pathogens using developed AuNP-based optical nanosensors in the past seven years (2015–now). Furthermore, trends and challenges in the application of these nanosensors in the determination of those analytes are discussed to suggest possible directions for future developments.

Never in the history of humanity have food and health not been major global concerns. In the Christian Bible, the Four Horsemen of the Apocalypse are gures representing four heartstruck fears of human: Pestilence, War, Famine and Death. Nowadays, by virtue of the rapid developments of agriculture and food processing industry, the worries of famine are less serious in most areas of the world, instead, people have to deal with the threat of unsafe food and drinks. There have been incidents of heavy metal contaminants in wheat, 1 melamine in milk, 2 pronil in chicken eggs and poultry products, 3 Cholera outbreaks, 4 etc. Food contamination may lead to serious gastrointestinal infections, 5 malnutrition, 6 and even cancers. 6, 7 Moreover, humanity has been ghting against diseases such as cancer and diabetes for decades. Besides, the continuous appearance of new pathogenic viruses such as Severe Acute Respiratory Syndrome coronaviruses (SARS-CoV) and Middle East respiratory syndrome coronavirus (MERS-CoV) keeps challenging every medical system. Recently, the COVID-19 outbreak has proved that human beings have never escaped from the threat of fatal diseases and pathogens. All of these problems have been putting human health in serious dangers. Therefore, it is essential to develop rapid and accurate tools and techniques to detect food contaminants, as well as diseases and pathogens.

The traditional methods such as high performance liquid chromatography (HPLC), 8 mass spectrometry (MS), 9 liquid chromatography-MS, 10 gas chromatography-MS, 11 and polymerase chain reactions (PCR) 12 are powerful techniques to detect different kinds of chemical compounds and biological elements. Despite their high sensitivity, accuracy and stability, they are time consuming, laborious and require expensive instruments. 13, 14 Therefore, sensing systems have been investigated to develop rapid, convenient, effective and less expensive detecting tools. In general, a sensor requires a target recognition element to ensure the selectivity and a transducer element to signal the binding events and manage the sensitivity of the detection. 15 Antibodies, 16 enzymes 17 and aptamers 18 are obviously effective for specic recognitions. A useful transducer element has to possess the ability to convert physical and electrical, electrochemical, catalytic and especially, optical properties are remarkable. 21, 23, 24 Every modication of size, shape, analyte binding, aggregation, etc. leads to change(s) in optical behaviors of AuNPs such as wavelength shi, color change, and enhancement of Raman scattering. 25, 26 They are detectable response signals that researchers are always looking for in an appropriate transducer. Therefore, it is not surprising that AuNPs-based optical sensors have attracted their interest for several years. Fig. 1 demonstrates the statistics of publications about AuNPs-based optical sensors over the last seven years (2015 -February 2022) according to Scopus data. Even aer many years of investigation, it is still a well-concerned topic of researchers all around the world without any sign of stop. Possessing the properties of local surface plasmon resonance (LSPR), uorescence quenching and surface-enhanced Raman scattering (SERS), AuNPs have been employed to develop different kinds of optical sensing systems including LSPR, colorimetric, uorescence resonance energy transfer (FRET)-based and SERS-based sensors (Fig. 2) . In this review, we focus on the strategies to develop those sensors, especially for the applications in food safety monitoring, and in vitro disease diagnosis. Recent advances in optical nanosensors for those analytes will be highlighted, focusing on the use of monometallic AuNPs, however, potential of multi-metallic goldbased optical nanosensors will be also introduced. Trends and challenges will be discussed to consider the future perspectives.

Recently, several reviews relating AuNPs-based sensors for food safety 14, 27 and disease diagnosis [28] [29] [30] have been established. However, most of them only focused on colorimetric sensors. 14, [27] [28] [29] [30] Other types of AuNPs-based optical sensors were mentioned in a few reviews but only as an introduction without full discussion. [28] [29] [30] This review exhibits a systematic look of AuNPs-based optical sensors for food and health safety monitoring with three kinds of sensors including colorimetric, FRET and SERS sensors. Further discussion about the sensing strategies developed for each kind of those optical sensors would provide readers more choices when considering designing a AuNPs-based optical sensor for their desired analyte. On the other hand, instead of focusing on the sensing mechanism for each kind of analytes as in the literature, [28] [29] [30] this review shows several comparisons of sensing strategies for different selection of analyte for one identical target, for example, nucleic acid or protein biomarker for cancer detection; nucleic acid, capsid protein or antibody for virus determination; etc., which is useful for researchers when selecting a suitable analyte for their study. Therefore, this review would give researchers who are working on AuNPs-based sensors some suggestions to develop their sensing strategies to their designed targets in the future.

Analogously to other nanomaterials, AuNPs can be prepared by both "top-down" and "bottom-up" methods. In general, the "top-down" fabrication of AuNPs involves breaking down the bulk Au into nanoparticles, while the "bottom-up" method is based on the assembly of precursor molecules or atoms. For the "top-down" approach, various techniques have been developed, such as chemical etching, lithography, pyrolysis, laser ablation, sputtering, etc. 20, 21, 31 Despite their advantages of fast and simple fabrication procedures, these "top-down" methods commonly encounter the limitations in controlling the particle size and shape. 32 In addition, some of these synthesis procedures require extreme conditions, including high vacuum, high temperature, and large amount of energy as well as expensive instruments. 21 In contrast, the "bottom-up" methods commonly involve the reduction of Au ions into Au atoms, followed by the assembly of the atoms to form NPs. In this case, the formation of AuNPs normally follows two steps: the nucleation and the subsequent growth. 32, 33 In the rst step, under certain circumstance, Au atoms are assembled to form small clusters, which function as nuclei. The addition of the incident atoms to the surface of the nuclei leads to the formation of AuNPs in the second step. An isotropic growth normally results in spherical nanoparticles. In order to tailor the particle morphology, surfactants such as cetyltrimethylammonium bromide (CTAB), 34 sodium dodecylsulfonate (SDS), 35 and poly(vinylpyrrolidone) (PVP) 36 are usually added. This is due to the fact that surfactants may suppress the growth in certain directions and promote the growth in the other directions, resulting in the anisotropic growth of the NPs. 37, 38 Therefore, by choosing an appropriate surfactant, the particle size and shape can be controlled. This is a great advantage of the "bottom-up" methods over the "topdown". Nevertheless, the use of toxic chemicals and solvents is a major drawback of this approach. 39 The era of size-and shape-controlled synthesis of AuNPs started with the method developed by Turkevich in 1951, in which AuNPs were synthesized by reducing hydrogen tetrachloroaurate (HAuCl 4 ) in the presence of sodium citrate. 40 This method was further improved by Frens et al. in 1957 , who demonstrated the size control of AuNPs by varying the ratio between HAuCl 4 and sodium citrate. 41 Although this protocol requires a high dilution of the Au salt that leads to the low AuNP concentration, its simplicity is a great advantage that retains it as the rst choice of researchers for the synthesis of spherical and monodispersed AuNPs with diameters in the range of 10-20 nm. In 1994, Brust and co-workers developed a novel approach for the synthesis of ultra-small AuNPs, namely twophase liquid-liquid reduction of AuCl 4 À by sodium borohydride in the presence of alkanethiol, which is nowadays known as the Schiffrin-Brust method. 42 In this method, HAuCl 4 was transferred from the aqueous solution to toluene using tetraoctylammonium bromide as the phase-transfer reagent. The addition of dodecanethiol and aqueous sodium borohydride facilitated the reduction of AuCl 4 À . In this process, the growth of Au clusters was accompanied by the simultaneous attachment of self-assembled thiol monolayers on Au surface. This provided an excellent control of the particle size, which allowed the synthesis of AuNPs with diameters in the range of 1-3 nm. 42 These are the pioneering works that paved the way for the tremendous development in AuNPs synthesis in the last several decades. [42] [43] [44] At the same time, the shape-controlled growth has also received enormous attention due to the shape-dependent properties of AuNPs. 45, 46 In this regard, the most common approach is utilizing surfactants to promote the anisotropic growth of AuNPs. A diversity of AuNPs such as nanorods, nanostars, nanoowers, and many others has been fabricated by adding CTAB as a surfactant. 37, 38 PVP has been reported to be useful to control the formation of star-like Au nanoplates. 47 Recently, surfactant-free techniques have been developed to control the shape of AuNPs. Wall et al. demonstrated the fabrication of AuNPs with different morphologies, including nanostars, nanospheres, nanorods, and nanoplates without using any surfactant. Instead, the morphology was controlled by varying the ratio between the reactants, pH, temperature and reaction time. 48 As the environmental issues raise the concerns, chemical solvents are gradually replaced by other environmentally benign reducing and capping agents such as biomolecules like lactic acids and chitosan, 49 or other biological agents such as extracts from Citrus maxima fruit or alfalfa plants. 50 Alternative approaches used physical agents to assist the reduction of Au ions, such as UV irradiation or electrical current in the electrochemical and photochemical methods. 33, 51, 52 Laser ablation is another popular physical method of synthesis as it can be employed to synthesize AuNPs in both aqueous and organic solvents. 53, 54 Moreover, this process does not require the removal of excess reagents aer the synthesis. It also helps avoid the contamination on the surface of the nanoparticles and reduce the toxicity of the prepared nanoparticles. Vinod et al. demonstrated a successful synthesis of AuNPs by pulsed laser ablation. 55 The obtained AuNPs exhibited excellent SERS and photothermal properties. 55 Electron-beam-induced reduction method has also been employed for the synthesis of AuNPs with a diameter of less than 20 nm that can be used as bioimaging probes in mammalian cells. 56 However, in comparison with the chemical synthesis methods, the physical approaches are less convenient and provide a lower yield.

For technological sensing applications, AuNPs are usually modied by appropriate chemical or biological elements for target recognition. Binding a self-assembled monolayer (SAM) to the AuNPs surface is a common technique to functionalize AuNPs. In general, SAM involves the adsorption of organic molecules onto the solid surface. 57, 58 Thiol, amides and carboxylic acids are ideal candidates for SAM due to their affinities to the AuNPs surface. Among them, thiols are most used because the Au-S bond is the strongest ($40 kcal mol À1 ) in comparison with Au-N ($8 kcal mol À1 ) and Au-COO ($2 kcal mol À1 ). 59 Thanks to these interactions, a wide range of molecules can be anchored to AuNPs. For example, peptides or proteins can be attached. 60 Kumar and coworkers used cysteinylated peptide to modied AuNPs and developed a sensor for Newcastle disease virus. 61 Similarly, thiolated aptamers were bound to AuNPs in a colorimetric sensor for interleukin-6, a biomarker for acute inammation. 62 Thanks to this great biocompatibility, AuNPs can be the scaffold for many kinds of recognition elements, leading to the ability to sense various targets, from inorganic to organic and from molecules to cells.

2.2. Optical properties of gold nanoparticles for sensing applications 2.2.1. Localized surface plasmon resonance: LSPR-based and colorimetric sensors & sensing mechanism. Localized surface plasmon resonance (LSPR) is one of the most important optical properties of AuNPs. LSPR is a phenomenon in which a collective coherent oscillation of the surface conduction electrons of the AuNPs is induced by the electric eld of the incident light. The interaction between the electric eld and the free electrons induces a charge separation, leading to the accumulation of charges on the surface of NPs. 25, 26, 63, 64 LSPR is highly dependent on shape and size of AuNPs. Sharp corners of the NPs lead to the rise of charge polarization, resulting in the red-shis of LSPR spectra while highly symmetric structures increase the intensity of dipole resonances, followed by the increase in LSPR intensities. 65 In addition, AuNPs may exhibit multiple resonance absorption peaks depending on the number of modes that they are polarized. 25, 66 Therefore, nonspherical NPs usually exhibit multiple, red-shied peaks in comparison to spherical particles. 67 For example, with two axis, transverse and longitudinal, gold nanorods can be polarized along them, leading to the appearance of two absorption peaks. 65 In contrast, spherical AuNPs exhibit only one peak of absorption. 65 Moreover, the longitudinal LSPR peak of gold nanorods is usually found at the longer wavelength region in comparison to the absorption band of spherical AuNPs. 25 The wavelength-shi of LSPR peak is also observed with the change in size of AuNP. It was reported that for spherical AuNPs, changing the size from 10 to 100 nm results in a red-shi of 47 nm. 68 In addition, the interaction among NPs is also an inuencing factor to LSPR. Interacting particles, either in a dimer or in an aggregation, would result in red shis of the LSPR spectra as well as the appearance of an absorption band at the region of lower energy (larger wavelength). 25, 26 Furthermore, wavelength shis in LSPR spectra can also be the results of changes in refractive index at the surface of NPs. The relationship between the LSPR wavelength shi, Dl (in nm), and the change in refractive index Dn (in refractive index unit, RIU) is described by the equation: 25, 63 Dl ¼ mDn

in which m is the bulk refractive index response of the nanoparticles (in nm/RIU), d is the thickness of adsorbed layer (in nm) and l d is the characteristic electromagnetic eld decay length (in nm). Eqn (1) expresses the inuence of the changes in the surrounding medium as well as the attachment of other molecules onto the surface of the NPs through Dn on their LSPR spectra. Therefore, every change in size, shape, interparticle interaction, surrounding environment and surface modication may result in LSPR wavelength shis. This is the underlying mechanism of optical sensors based on inherent LSPR property of AuNPs. Due to their strong LSPR absorption in the green region, AuNP solutions usually exhibit a characteristic ruby red color. In most cases, an LSPR wavelength-shi can lead to a change in color of the solution. For example, an aggregation of AuNPs that caused a red shi of the LSPR spectra was also observed as a red-to-purple/blue change in the solution color. 69 Therefore, the changes in LSPR spectra can be detected by either using an UV-visible spectrophotometer or bare-eyes. Hence, from the LSPR characteristics of AuNPs, two typical types of optical sensors have been developed: LSPR-based and colorimetric sensors. Importantly, the shi of LSPR spectra is strongly correlated with the aggregation of AuNPs. This correlation has been utilized in many sensing systems that are related to the aggregation of AuNPs (Fig. 3) .

The aggregation state of AuNPs can be triggered through either crosslinking or non-crosslinking strategies. In the crosslinking method, ligand-functionalized AuNPs are assembled in the addition of a target element. Herein, the ligand acts as recognition probes for the target compound, which binds to this compound, decreases the interparticle distance and triggers aggregation of the AuNPs (Fig. 2) . For instance, in the study of Megarajan and Veerapan, the aggregation of N-lauryltyramine (NLTA) capped AuNPs was induced in the addition of Al 3+ , which led to a pink-to-purple change in the solution color as well as a red-shi from 538 nm to 670 nm of the LSPR peak. The mechanism was suggested to be the chelation effect between NLTA and Al 3+ . 70 In contrast, the non-crosslinking approach, which is also known as de-protection method, is based on the removal of stabilizing agent on the surface of AuNPs due to the presence of the target element. In this strategy, AuNPs are well-protected with the stabilizing agent to avoid aggregation. However, in the addition of the target element, which has higher affinity to the stabilizing agent, the AuNPs lose their coating agent (i.e. they are "de-protected"). This loss of electrosteric stabilizations results in AuNP aggregation. Aptamer-anchored AuNPs are widely used to develop colorimetric sensors based on this mechanism. In the addition of salts and the target elements, the aptamers are disassociated from AuNPs to "capture" the targets while high-salt condition induced the aggregation of the AuNPs. Wu et al. have applied this strategy and designed aptasensors for detection of Escherichia coli O157:H7 and Salmonella typhimurium with the speci-city of 100%. 71 On the contrary, a change of color from blue to red is observed in another strategy of colorimetric sensing called antiaggregation. In this method, a linking molecule (or linker) induces aggregation of NPs by crosslinking mechanism. However, as the target element has a higher affinity to the linker, their interaction inhibits the binding of the linker to the ligand on AuNP surface and consequently prevents the aggregation of those AuNPs. For example, Gao's group introduced a sensing system for Ag + , in which thiamazole played the role of a linking molecule to induce aggregation of AuNPs. 72 The presence of Ag + ions triggered the formation of preferential combination between the pyrimidic nitrogen of thiamazole and Ag + ions. Moreover, thiol group of thiamazole was also oxidized by Ag + . Therefore, the aggregation of AuNPs was inhibited, which resulted in a blue-to-red change in the solution color 72 Fig. 3 AuNPs-based optical sensors and their sensing strategies. (Fig. 4) . The same strategy has been utilized to detect other heavy metals, food additives, etc. 14, 73, 74 Mainly concerning the aggregation of AuNPs (either formation or destruction), all strategies mentioned above do not require any specic shape of the AuNPs. Therefore, any kind of AuNPs can be employed to develop these colorimetric sensors, however, spherical AuNPs are the most selected in practice. 75 The choice of isotropic AuNPs helps simplify the fabrication procedures of the sensors, consequently, it is more time-and cost-effective to develop these sensors. Brief information about AuNPs-based colorimetric sensors and their particle shapes can be found in Table 1 and 2, and further discussion about those sensors will be presented in the following sections.

Based on the aggregation of AuNPs, the colorimetric sensing strategies mentioned above are fast and simple with only a few steps of preparation and testing. Moreover, the obvious color change can be observed with bare-eye without any complicated instrument. However, the main limitation of these approaches is that AuNPs can be aggregated due to numerous factors, including pH, temperature and ionic forces that can lead to false positive/negative results. Therefore, researchers have investigated AuNPs-based sensors applying non-aggregation mechanisms, including etching, growth and nanozyme. 76 Etching techniques focus on the effect of enchants such as H 2 O 2 and I À ions to change the sizes and shapes of AuNPs, which leads to a shi in LSPR peak and a subsequent color change of reaction solution. Thus, etching is the only strategy, in which anisotropic NPs, such as Au nanorods and multibranched AuNPs, are preferable when designing colorimetric sensors. 66, 77 An example of this strategy is a urine glucose sensor established by Zhang et al., in which glucose oxidase enzymatic reaction led to the release of H 2 O 2 that oxidized I À to I 2 rapidly. 66 Au nanorods were etched by I 2 , especially at their tips, thus, a longitudinal LSPR blue-shi and a blue-to-red color change were observed for glucose detection. 66 Besides, tetracycline, one of the over-used antibiotics for farm animals, could reduce Au 3+ ions to trigger the growth of AuNPs and introduce the characteristic red color to the solutions. Hence, it has become the principle to develop tetracycline sensing systems. 78 However, it is worth mentioning that the selectivity of this method is low among other antibiotics in the same family (with tetracycline) due to their similar phenolic structures. 78 In another approach, the peroxidase-like activity of AuNPs allows them to oxidize TMB in the presence of H 2 O 2 to achieve the colorless-to-blue color change in solutions. Hence, target elements that enhance the catalytic activity such as Hg 2+ (ref. 79) and Fe 2+ , 80 or induce the formation of H 2 O 2 such as glucose 81 (in the interaction with glucose oxidase) can be detected. In another study, the color change was reversed upon the addition of glutathione, which suggested another principle to sense this disease marker. 82 Furthermore, AuNPs can mimic the catalytic activity of several other enzymes such as glucose oxidase and peroxidase. 76, 83 With the highest surface area-to-volume ratio, there is no surprise that spherical AuNPs are the most selected material for this approach (see Tables 1 and 2) .

AuNPs-based sensors using non-aggregation approaches are more complicated than those using aggregation or antiaggregation methods. For instance, to detect zearalenone, a mycotoxin, the sensors fabricated based on the catalytic activity of AuNPs to convert 4-nitrophenol (yellow) into 4-aminophenol (colorless) needed to be functionalized with two kinds of single stranded DNA (ssDNA), including one aptamer and its complementary strand. Moreover, the detecting procedure required two steps of adding exonuclease III and 4-nitrophenol. 84 Nevertheless, these novel methods have helped researchers to escape from the "red-to-blue" trap as with every kind of target element, the aggregation methods only exhibit the colors in that range. The novel indirect colorimetric approaches introduce new color ranges which would be useful to develop multi-target detecting systems.

In several immunoassays, AuNPs play the role of a chromatic substrate, in which the characteristic red color of AuNPs would be a reporting signal for the detection. It is convenient to generate immune testing strips. [85] [86] [87] 2.2.2. Fluorescence modulation: FRET-based sensors and sensing mechanism. Fluorescence is an optical phenomenon in which a photon is absorbed by a uorophore and consequently emits another photon at a lower frequency. The emission of a uorophore can be modulated by AuNPs as it can interact with the LSPR of AuNPs. 88 It has been reported that AuNPs, especially spherical particles, exhibit high uorescence quenching ability. 89, 90 Therefore, AuNPs are ideal candidate for acceptors in a uorescence resonance energy transfer (FRET)-based system. FRET is a non-radioactive phenomenon involving the energy transfer from an excited donor uorophore to an acceptor uorophore. FRET only occurs when the distance between donors and acceptors is sufficiently short, in general, less than 10 nm. 89, 91, 92 Thanks to the high extinction coefficient and broad absorption spectrum that overlaps the emission spectra of many organic dyes and quantum dots, AuNPs are great quenchers. Without any dened dipole moment, spherical AuNPs are ready for energy transfer from any orientation of the nearby donors. 93 Moreover, the high surface area-to-volume ratio of spherical AuNPs provides a large interacting area for multiple donors at once, increasing the efficiency of the FRET system. 89 More importantly, using spherical AuNPs as acceptors, the distance between the donors and the AuNPs can be extended up to 70-100 nm instead of less than 10 nm. 89, 91, 94 This allows them to be functionalized with recognition elements such as aptamers or peptides to ensure specic binding to donors without reducing the FRET effect between them. A FRET-based sensor for detection of aatoxin B1 (AFB1), a mycotoxin that might be found in food and drinks, was fabricated using uorescein (FAM) molecules as donors and AuNPs as acceptors. FAM molecules were labeled by an aptamer specic for AFB1 while AuNPs were functionalized with complementary DNA strands. In the absence of AFB1, two DNA strands hybridized, resulting in the immobilization of FAM molecules on the surface of AuNPs. FRET phenomenon occurred as AuNPs quenched the uorescence of FAM. In the presence of AFB1, the aptamers were bound to AFB1, and FAM molecules were released from AuNPs, "turning-on" the uorescent signal. 95 Similarly, different kinds of organic dyes, uorescent proteins, quantum dots, and unconventional NPs have been utilized as donors, from which excited energy is transferred to AuNPs within a short distance. 89, 96 AuNPs quench the uorescence signal, so the system is in "turn-off" state. Hence, the key strategy to "turn-on" a FRET sensor is to elongate the distance between the donor and the acceptor, leading to the recovery of uorescent signal. Therefore, the target element should bind to either the donor or the acceptor in a competitive mechanism. For example, Dong et al. reported a "turn-on" uorescent sensor to detect glutathione (GSH) based of FRET effect between N,S dual-doped carbon dots (N,S-CDs) and AuNPs. Positively Table 2 Several recently reported AuNPs-based optical nanosensors for charged N,S-CDs bound to negatively charged AuNPs via electrostatic interaction. 97 Fluorescence intensity of CDs was quenched by AuNPs due to FRET effect. In the addition of GSH, the strong Au-S bond between GSH and the NPs leads to the release of N,S-CDs from the NPs surface, so uorescence recovery occurs (Fig. 5 ). The limit of detection (i.e., LOD) of this method was as low as 0.21 mM. As the concentration of GSH in human serum is up to mM levels, the sensing system was expected to be able to determine GSH in diluted real samples. 97 This example demonstrates the high sensitivity and accuracy of AuNPs-based FRET sensors that take advantage of the uorescence efficiency of the donors and the excellent quenching ability of AuNPs. However, further studies are required to achieve the optimal conditions for FRET sensing. Each of donors possess its own advantages and disadvantages. For example, organic dyes are small, simply-prepared and cost-effective, but they have high toxicity, high photobleaching rate and low stability; uorescent proteins are nontoxic, but they display wide emission spectrum as well as large sizes that can elongate the distance to the acceptors; QDs exhibit high quantum yield, narrow emission spectrum and high photostability, but they are highly toxic, etc. 89 Therefore, researchers need to balance out these advantages and disadvantages to choose the suitable donors for FRET sensors based on the desired target elements and the sensing environment. Moreover, other NPs have been investigated as the promising candidates for the role of donors. For instance, unconventional nanoparticles (UCNPs) can emit uorescence under the excitation wavelengths in the near infrared region, which is less harmful for biological samples. 89, 98 Recently, graphene NPs have been regarded as a potential choice as they can be fabricated easily at large scale of production. Moreover, their cost-efficiency, high solubility and biocompatible are advantages. However, the structure of this nanomaterial should be further studied to improve their wide emission spectra. 89, 99 In general, AuNPs are non-luminescent materials, however, ultrasmall AuNPs (<3 nm), which are known as gold nanoclusters (AuNCs), have molecular-like properties, including photoluminescence. 100 Therefore, they have been studied to design uorescence sensors for many different targets. However, in this review, we only focus on AuNPs. The uorescent sensing mechanisms using AuNCs have been reviewed in detail by Halawa et al. 100 2.2.3. Surface enhanced Raman scattering: SERS-based sensors and sensing mechanism. Surface enhanced Raman scattering (SERS) is a sensitive spectroscopic technique involving inelastic scattering of incident light and a surface molecule, leading to the vibrational spectral peaks of the molecule. When an analyte is adsorbed onto a rough plasmonic metal, its Raman signal is enhanced. SERS signal provides information about the structure, the interaction and environment of the analyte, even at trace levels, as well as a platform via a rapid detection. AuNPs are widely used as substrates for SERSbased sensors. 25, 101 In 1977, the signal enhancement was explained in both chemical and electromagnetic mechanisms (CM and EM) by lbrecht & Creighton 102 and Jeanmaire & Van Duyne, 103 respectively. These two mechanisms are now regarded as the underlying fundamentals of SERS effect. The intensity of Raman scattering is proportional to the square of the induced dipole moment P, which is expressed by the equation: 25,104,105

in which a is the polarizability of the molecule (chemical), and E is the electric eld (electromagnetic). CM requires the analyte to be chemisorbed to the substrate, allowing a charge transfer between them. On the other hand, EM is the result of coupling of the incident electromagnetic eld and the surface plasmon of the SERS substrate. This is a local phenomenon that only inuences the molecule at or near the metal surface. 105 Therefore, both CM and EM concern the distance between the SERS substance and the analyte/reporter. Minimizing this distance can enhance the SERS signal. In addition, it has been reported that SERS enhancements could originate from the plasmon formed at the small gaps between NPs (usually <10 nm) or sharp tips and corners of NPs, known as "hot spots", with an enhancement factor up to 10 12 . 106 Therefore, analyte binding and "hot spots" are the keys to design a SERS-based sensor (Fig. 6) . Besides, the distance between Raman reporter and the substrate (AuNPs) is also important. In this part, we summarize several strategies to design AuNPs-based SERS sensors and their detecting mechanism (Fig. 3) .

The analyte or target element of a SERS-based sensor might be adsorbed directly to the surface of AuNPs or binds indirectly through a recognition element. This element can be an antibody, an aptamer, a small molecule or a polymer. Using AuNPs as SERS substrates, molecules containing thiol or cyanide group can be introduced onto the NP surface directly due to the high affinity between their functional groups and Au. 108, 109 Employing unfunctionalized AuNPs is simple to prepare. However, the lack of specicity may lead to the low selectivity of the sensors, especially for samples of complex matrices. More importantly, there are only few functional groups and ions that can attach to the surface of AuNPs. Thus, functionalization of AuNPs is a solution to avoid these two disadvantages because it allows specic binding of target elements on AuNPs. However, the limitation of this approach is that the functional molecule (or receptor) can be extremely large, for example, antibody, which makes the distance between the target element and the substrate too far to experience the signal enhancement. 110 Thus, it has been a demand of investigating smaller molecules to functionalize AuNPs. Van Duyne's group has designed bisboronic acid receptors with two arms of boronic acid forming covalent bonds with glucose and one arm of thiol being immobilized on AuNP surface. In addition, the highly polarizable aromatic structure of the molecule acted as a Raman reporter to enhance SERS signals. The detection of glucose in physically relevant condition range allowed the diagnosis of both hypoglycemia and hyperglycemia. 111 Besides, Raman signal of a target element can be amplied when it is adsorbed on hot-spot regions. 107 Hence, in addition to directing the binding of analyte onto NPs, researchers also pay attention in creating hot-spots via material fabrication. Anisotropic AuNPs, such as nanostars, nanorods, have been synthesized and employed as SERS materials, leading to signicant enhancement in SERS signal. 112, 113 Besides, hot-spots could be also created between isotropic AuNPs by self-assembly 114 or polymer assisted assembly 115 on the substrates, or even the targets themselves. 116 In a strategy of hot-spots modulation, hot-spots are created or destroyed by the target elements themselves, consequently, SERS-based sensors are generated via either "turn-on" or "turnoff" mechanisms. "Turn-on" approach focuses on decreasing the distance between two adjacent AuNPs to create hot-spots to increase SERS signals. In contrast, in "turn-off" approach, the target elements disrupt the connections between AuNPs, elongate their interparticle distance to destroy the hot-spots, thus, decrease SERS signals. In fact, these mechanisms are similar to aggregation and anti-aggregation mechanisms in colorimetric sensing. An example for the "turn-on" method is the detection of Hg 2+ ions based on the assembly of AuNPs functionalized with single stranded DNA into chains reported by Thanks to the T-Hg 2+ -T bond between two thymine base pairs, double stranded DNA was form between NPs to decrease the interparticle distance. As a result, SERS signals are signicantly enhanced, even at low concentration of Hg 2+ . Hence, the method was extremely sensitive with an LOD of 0.45 pg mL À1 . 117 On the other hand, heparin, an anticoagulant, was detected via "turn-off" method. Zhang et al. induced aggregation state of AuNPs using the interaction between 4-mercaptobenzoic on the surface of AuNPs and protamine. However, the higher affinity to protamine helped heparin to bind to this protein, resulting in anti-aggregation of the NPs and a decrease in SERS signals. 118 To achieve high signals for SERS sensing, it is important to maintain the close distance between the substrates and the reporter. A similar strategy of "turn-on" and "turn-off" using the modulation of this distance can be also generated using the mechanisms mentioned above. 119 However, applying either hotspots between AuNPs or distance between AuNPs and reporter, "turn-on" approach is commonly preferred. The reason is that at low concentration levels of target elements, the decrease in SERS signals obtained by "turn-off" method is not as noticeable as the presence or not of the signal in "turn-on" approach.

Since SERS is a vibrational spectroscopy, once the structure or conformation of the recognition elements on the surface is changed, the spectral variations in some bands of SERS responds responses can be detected. In a 2019 study, Popp and coworkers reported the detection of Cu 2+ in white wine using this principle. The interaction of Cu 2+ and the dipicolylaminebased ligand bound to the surface of AuNPs caused the changes in SERS spectra associated with pyridine breathing mode. 120 This chemo-optical sensing method is a new approach that takes advantage of the chemical sensitivity and specicity of SERS signals. However, it is not all of the chemical and conformational changes that can be detected in SERS responses. In general, changes of the groups linked directly to aromatic ring tend to have greater inuence on the polarizability of the probe, therefore the spectral variations are also more noticeable. 104 In addition to the active metal ions that can change the chemical/conformation of the ligands bound to the surface of NPs, large elements have also been detected using this mechanism. For example, the interaction between inuenza-H1 antigens and their antibodies, which were linked to 4-ATP, led to the relaxation of the stretched aromatic linker on the surface of AuNPs. This conformational change resulted in signicant up-shis in the peaks corresponding to the C-S stretching and C-C breathing modes. 121 This sensing mechanism has overcome the disadvantages of using large antibodies in SERS sensors as the long distance between target elements and AuNPs did not limit the SERS responses. This mechanism has paved the way to the development of novel methods to detect biological elements using SERS-based sensors. 122,123

Taking advantage of the optical, chemical and electromagnetic properties of the AuNPs, combining with the knowledge about desired analytes, researchers can design and develop optical nanosensors with various of advantages. However, there are still several disadvantages with which researchers have to cope when working on these types of sensors. Although some advantages and disadvantages have been mentioned previously, in this section, we would like to summarize their pros and cons, in comparison to several other AuNPs-based nanosensors.

Optical AuNPs-based sensors are simple and rapid. 14,124 Especially, the results obtained for colorimetric sensors can be observed by bare-eye. 14,125 However, they have to deal with a risk of false positive/negative results due to aggregation of AuNPs, especially in complex sample matrix. 126, 127 Moreover, the inadvertent aggregation of AuNPs may also reduce the sensibility and accuracy of SERS and FRET sensors. 128, 129 These risks in precision and accuracy can be avoided by functionalizing AuNPs with a recognition element specic for the desired analyte. Functionalization denitely improves the selectivity of optical AuNPs-based sensors, however, sometimes, it scaries the stability of AuNPs. 130 Nevertheless, the introduction of netuning techniques resulted in more stable modied-AuNPs. 131 Another drawback of functionalization of AuNPs in optical nanosensors is that large recognition element, such as polymer, protein, antibody, etc., elongate the distance between the surface of the particle and the analyte (in a SERS sensor) or the acceptor (in a FRET sensor), which prevents the analyte from experiencing SERS effect or prevents the uorescence of the acceptor from being quenched. However, thanks to great quenching ability of AuNPs, the distance between them and the acceptor this distance to be extended up to 70-100 nm, 89, 91, 94 thus, it is the problem of only the SERS sensors, and the solution is avoiding the large recognition elements despite their high specicity. Besides, the manufacture of multi-target optical sensors requires functionalizing AuNPs with multiple recognition probes. It can increase the complexity and the cost of sensor fabrication procedures, howbeit, to the best of our knowledge, researchers have to accept this drawback to design these optical sensors. 132 In term of instrument, expect for colorimetric sensors, LSPR, FRET and SERS sensors all require specic spectroscopies for measurements. 133 However, being developed for a long period of time, the spectroscopies for these techniques have become common and popular, 134,135 which can be found in many laboratories all around the world, thus, it is convenient for researchers to generate these optical sensors.

Also using AuNPs as the main substrate for analyte detection, some other nanosensors can help researchers overcome the disadvantages of the optical sensors. Electrochemical sensors are promising for multi-target sensing as they can determine different analytes at the same time. 136 However, they require noticeable electroactivity of the analytes, which limits the range of target molecules. 137 Furthermore, electrode fouling may be a problem for sensing repeatability. 138 Another novel technique based on plasmonic property of AuNPs is chaotic sensor, in which the evolution of absorption spectra of AuNPs was recorded under the excitation of a laser source and analyzed by electronic signals using a chaotic circuit. 139 In another word, it is an opto-electric sensor, which senses the changes in conductivity arising from the absorption of the samples, thus, its sensibility is impressive. This sensor does not require any expensive spectroscopy. However, it is necessary to design a complicated electric modulator. 139 Thus, it needs specialized human resources to design and operate the sensing system. In another approach, AuNPs can also act as an assisting material in terahertz (THz) sensors. THz waves are potential for detecting analytes at trace levels, especially biological elements whose absorption peaks fall in THz region. For example, L-histidine shows a strong absorption band at $0.78 THz. 140 Unfortunately, AuNPs lack plasmonic property within this region. 140 However, it has been reported that the assist of AuNPs could enhance the performance of metamaterials for THz sensors, leading to impressive LODs of the sensors. 140, 141 Avidin, Staphylococcus aureus (S. aureus), and microRNAs were detected using this kind of sensors at the concentration down to fM, pM, and even aM level, respectively. [140] [141] [142] In spite of that excellent sensibility, THz sensors still exhibit several drawbacks. Compared to optical sensors, material fabrication for THz sensors is more complicated and expensive. [140] [141] [142] [143] Furthermore, being a novel technique, THz spectroscopy is not as popular. 143 Moreover, due to several reported limitations, THz need to be further studied and modied before being employed for real samples. 144 Therefore, the application of THz sensors is still limited.

Each of the AuNPs-based sensors has its advantages and disadvantages that researchers can consider developing the appropriate sensors based on their desired analytes and the conditions of their laboratories. With the convenience and simplicity of material fabrication and sensing system operation, as well as the popularity of the spectroscopies, AuNPs-based optical sensors still attract the interest of researchers with a huge number of sensors developed for food and health safety monitoring in recent years which will be presented in Tables 1 &  2 in the following sections. 3. AuNPs-based optical nanosensors for food and health safety monitoring

The rising attention about food quality has triggered many studies on monitoring food safety. In general, they focus on the detection of the contaminants (i.e., heavy metals, pesticides, herbicides, farm animal drugs, etc.) and food additives (i.e., preservatives, food dyes). This part will present some examples of recent advances in AuNPs-based sensing to detect these targets (Fig. 7) . Detailed information about limits of detection (LOD), linear ranges and real sample matrices is presented in Table 1 .

3.1.1. Heavy metals. Heavy metals, such as mercury (Hg), lead (Pb), copper (Cu), iron (Fe), nickel (Ni), cobalt (Co), cadmium (Cd), etc., are omnipresent naturally in the environment. Most of them are even essential for life. However, at high concentrations, they can cause damage to many tissues and organs in human body, including kidney, liver, bones, and central nervous system. 146 Due to industrial and agricultural developments, their contents have increased and accumulated through the food chains. To protect the consumers from food and drinks containing high concentrations of heavy metals, it is important to develop sensing system to monitoring their contents.

In general, most of sensors using AuNPs as probes for detecting heavy metals operate based on the interaction between the ligands on the NP surface. There are several widelyused ligands that can be chosen to be suitable for the desired target. For example, the high specicity of thymine-Hg 2+thymine (T-Hg 2+ -T) affinity have been applied for many Hg 2+ sensing systems as in the presence of Hg 2+ , T-T mismatches are formed. For example, Amiri et al. have introduced a FRET aptasensor for Hg 2+ using carbon dots as donors and AuNPs as acceptors. 145 The carbon dots were labeled with T-rich single stranded DNA, while the AuNPs were functionalized with a complementary DNA sequence (cDNA). When being mixed together, the AuNPs quenched the uorescence of carbon dots. However, in the presence of Hg 2+ , the cDNA was replaced due to the formation of T-Hg 2+ -T on the T-rich single stranded DNA. As a result, uorescence recovery occurred. The FRET sensor achieved an impressive LOD of 7.5 Â 10 À13 M of Hg 2+ . 145 A similar principle has been employed to design an indirect colorimetric sensor, in which T-rich DNA sequence were used as a capturing probe for Hg 2+ . In the presence of Hg 2+ ions, it bound to T-rich oligonucleotide to form a T-Hg 2+ -T complex. The formation of this complex prevented the adsorption of oligonucleotide on the surface of AuNPs. The AuNPs then were available as a nanozyme to oxidize TMB, resulting in a much deeper blue color of the solution compared to the previously coated AuNPs. 147 Besides, the interaction between DNAzyme and Pb 2+ also attract researchers' concern for anti-aggregation approach. 148 In the presence of Pb 2+ , DNAzyme would cleavage the target sequence. Therefore, this phenomenon was employed to separate AuNPs and graphene quantum dots in a 2018 study, resulting in uorescence recovery (Fig. 8A ). 149 The sensor had an LOD of 16.7 nM. However, compared to the utilization of T-rich oligonucleotide, using DNAzyme is more complicated and less convenient. Therefore, this approach is not widely used.

In addition to the interaction with oligonucleotide, heavy metals can also be captured by peptides that are rich in reactive groups. Glutathione (GSH) is an example. Each GSH possesses 8 binding sites, including 1 sulydryl, 1 amino group, 2 carboxyls and 4 carbonyl and amide donors. 150 Therefore, heavy metal ions exhibit high affinity to this peptide. The thiol group allows GSH to be easily introduced onto the surface of AuNPs while other reactive groups are responsible for capturing heavy metal ions, causing the aggregation of AuNPs. Pb 2+ and As 3+ have been detected via this sensing principle. [151] [152] [153] Incubated with GSH before the addition of AuNPs, Cd 2+ exhibited such a high affinity to the peptide that AuNPs were prevented from binding to GSH, which led to the NP aggregation. The LOD of the method was calculated to be 4.3 pM. 150 Similarly, other molecules containing reactive sites, either natural or synthesized ones, can become promising ligands to detect heavy metals in AuNPs-based assays. 154 3.1.2. Pesticides and herbicides. Due to the development of agriculture and high demand of grains, vegetables and fruits all around the world, agro-products oen contain more pesticides that are harmful to human health. For example, organophosphorus, a group of pesticides, is most used due to its broad spectrum of insecticidal activities. However, it acts as an inhibitor for acetylcholinesterase (AChE), an essential enzyme to decompose acetylcholine for neurotransmission. Thus, the detection of organophosphorus pesticides has attracted a great concern of researchers, resulting in the development of interesting approaches. In addition to the studies using the widely-used concept of AuNP aggregation via de-protection of functionalized AuNPs, 155, 156 Wu et al. took advantage of the characteristics of the pesticides to realize a sensor with high selectivity. 157 This method was based on combination of AChE hydrolysis reaction and dissolution of AuNPs in Au 3+ -cetyltrimethylammonium bromide (Au 3+ -CTAB) solution. In the absence of organophosphorus pesticides (OPs), AChE converted acetylthiocholine to thiocholine that reduced Au 3+ in the solution. Due to the consumption of AChE, there was no Au 3+ to dissolve AuNPs. Moreover, the reduction of Au 3+ triggered the growth of AuNPs, which turned the solution into dark red. In contrast, the presence of OPs resulted in the inhibition of AChE. Without reducing thiocholine, Au 3+ ions remained in the solution and dissolved the AuNPs with the aid of CTAB. Consequently, the solution changed to light pink or colorless. The sensing system achieved a low LOD of 0.7 ppb, and well performed in practical samples including tap water, apple washing solution and sea water. 157 However, this method could not distinguish different kinds of pesticides in the organophosphate group. In a more recent study, Zhang et al. designed a multi-analyte sensing system allowing the detection of three organophosphate pesticides (triazophos, parathion, and chlorpyrifos). 158 Three kinds of AuNP probes were fabricated with responding antibodies for each pesticide and three different kinds of uorophores (i.e., FAM, Cys 3 and Texas red), respectively. The detection was based on a competitive reaction between the pesticides and antigens in the plate to bind to AuNP probes. The probes binding to pesticides were washed away while the others were immobilized in the plate. Subsequently, dithiothreitol (DTT) was added and the affinity between DTT and AuNPs led to ligand exchange. As a result, uorophores were released from the uorescence quenching AuNPs. The uorescence signal was recovered and then analyzed to calculate the concentration of each pesticide in the samples. The work proposed a novel method for multi-residue analysis with appropriate performance in various food samples. 158 Another approach for the detection of pesticides is SERSbased sensors. Different kinds of organophosphorus pesticides and fungicides, such as thiram and carbendazim, have been detected using this method. 112, 159, 160 In general, these residues can adsorb directly on the surface without any modi-cations. Therefore, to enhance the SERS signal, researchers have been focusing on the fabrication of the AuNPs. For example, Au nanostars (Fig. 8C) , snowake-like AuNPs, and spherical AuNPs deposited on a "sticky" tape have been employed as substrates for the detection of thiram and several organophosphorus pesticides. 112, [159] [160] [161] 3.1.3. Farm animal drugs. To meet the growing demand for milk, meat and other livestock products, the number of global livestock has kept increasing for years. As a result, the consumption of farm animal drugs is also in a rising trend. Different from human, for which medicine consumption is carefully calculated and evaluated, veterinary drug consumption is much more massive and less controlled. These drugs can be accumulated in the meat and eggs of the animals or secreted into their milk. In addition, a large part is excreted into the ecosystem via urine and feces, which nally ends up in water sources. Therefore, human have to cope with the risk of consuming food and drinks containing high contents of these drugs. Farm animal drugs include b-agonists (i.e., ractopamine, salbutamol, clenbuterol, etc.), which promote leanness in livestock, and different kinds of antibiotics.

In general, there has been only few studies mentioning the detection of b-agonists using optical sensing systems. Most of them were based on the crossing-linking principle to cause aggregation of AuNPs. However, using different linking elements including melamine and aptamer has improved the selectivity of the detection, 162,163 allowing them to perform in real complex samples. Recently, Han et al. reported on a ractopamine sensing system based on the growth of AuNPs in combination with a competitive immunoassay. 164 In this system, ractopamine in the samples had to compete with ractopamine immobilized in the ELISA to bind to the antibodies. Then, biotinylated second antibodies were added, followed by the introduction of streptavidin label catalase, which could convert H 2 O 2 to H 2 O and O 2 . In the presence of ractopamine, there was less catalase immobilized in the ELISA well, so Au 3+ ions in the well were reduced by H 2 O 2 rapidly, resulting the formation of AuNPs with the red color. In contrast, more catalase consumed a large amount of H 2 O 2 . The lack of H 2 O 2 slowed down the kinetics of AuNP fabrication and induced the formation of aggregation with blue color. 164 Antibiotics are also widely used in farming in a massive amount. The risk of antibiotic resistance caused by consuming agro-products containing antibiotics cannot be ignored. Therefore, for the last 10 years, many studies have been reported on the detection of antibiotics in food and animal samples. Most of them are colorimetric and uorescent sensors, however, the detecting approaches are quite diverse. Using aptamers as capturing probes, ampicillin and kanamycin were detected via aggregation and anti-aggregation, respectively. 165, 166 In another approach, kanamycin, neomycin, streptomycin, and members of tetracycline family were employed to trigger the growth of AuNPs. The color of AuNPs conrmed their presence and concentrations in samples. Although these methods did not exhibit signicant difference among different antibiotics, 78, 167 it is worth noting that most of those systems could be employed to detect antibiotics in milk and animal tissues. Recently, researchers have expressed a great interest in developing multi-residue sensors. In a 2020 study, Wu et al. described a colorimetric aptasensor for multiplex antibiotics, including chloramphenicol (CAP) and tetracycline (TET). 132 A multifunctional aptamer was designed and adsorbed onto the AuNP surface. In the presence of one antibiotic, the specically recognized fragment of the aptamer bound to that antibiotic and dissociated from the AuNP surface, while the non-specic one still attached and controlled the aggregation of AuNPs. Different antibiotics caused different states of AuNPs aggregation that could be distinguished by the colors of the solutions (Fig. 8B) . Therefore, one sensing system was successful for the separate detection of two kinds of antibiotics with low LODs of 32.9 nM and 7.0 nM for TET and CAP, respectively. 132 Similarly, Xiang's group reported an even more complicated system including two kinds of AuNPs labeled with 2 distinct aptamers to detect 5 types of aminoglycoside antibiotics. 168 Due to different binding sites of those types of antibiotics, the obtained signal was distinguished. 168 Therefore, 5 types of antibiotics could be detected separately using one sensing system. However, those two sensing systems were not possible to detect two or more antibiotics at once.

3.1.4. Food additives. Food additives are substances utilized during food processing to improve quality and durability of the product, as well as to control the food color and avor. However, illegal food additives are toxic substances that are banned in food products. Nevertheless, the use of the illegal ones is not completely stopped that put consumers in health hazards. Hence, it is urgent to develop sensors to detect these poisonous food additives. In this review, we summarize several examples of advanced AuNPs-based sensing systems for melamine, some preservatives and food dyes. Examples of other additives are presented in Table 1 .

Melamine is a trimer of cyanamide which has been widely used in manufacture of polymer resins, fertilizers, kitchenware, etc. However, due to its high nitrogen contents, it has been added illegally into animal feeds and dairy products to inate the protein contents. 170 Consuming melamine containing food and drinks may cause serious damages to human kidneys and reproductive systems. 170, 171 Thus, it is important to develop reliable sensing methods to detect melamine. Due to the structure containing 3 amino groups, it is easily bound to 2-3 AuNPs at once, through the interactions with their coating agents such as citrate, 171 thymine, 172 and 1,4-dithiothreitol (DTT). 173 Therefore, the detection of melamine is usually based on the crosslinking aggregation approach. 171-175 Gao et al. even combine crosslinking and de-protection strategies to develop a colorimetric assay for the detection of melanine (Fig. 9A) . 174 To improve the sensitivity of the method, several novel strategies have been proposed. Zhang et al. took advantage of AuNP aggregation to generate a "turn-off" FRET sensor using conjugated polymer NPs as donors. 176 In the absence of melamine, uorescence signal was quenched by AuNPs. The introduction of melamine led to aggregation of the AuNPs, resulting in the uorescence recovery. The method achieved a low LOD of 1.7 nM. 176 Chen et al. focused on widening the dynamic detection range. 170 The authors found that aer introducing melamine to AuNP solution, AuNPs quickly aggregated and formed large aggregations. The precipitation of these aggregations led to the rapid evanishment of solution color, which reduced the accuracy of the colorimetric signals. Therefore, they introduced polyethylene glycol onto selected area of the AuNP surface, thereby stabilizing the AuNPs. The binding of melamine still occurred at the citrate-site of the AuNPs and induced a purpleto-blue color change. The linear range of the method covered from 1.05 nM to 1 mM. 170 Preservatives have been playing an important role in the food industry by improving the duration of food products and allowing them to be stored and transported to consumers. However, illegal food preservatives are extremely toxic to human. For example, high intake nitrite and formaldehyde might lead to the risk of gastric cancer and leukemia. 177, 178 In general, there are only few studies on optical sensors to detect food preservatives, in comparison to electrochemical ones. Martínes-Aquino et al. employed resorcinol functionalized AuNPs for colorimetric detection of formaldehyde with an LOD of 0.05 ppm, but the method has not been applied in real samples. 179 Nie et al. detected formaldehyde using nonfunctionalized AuNPs in a SERS sensor with an LOD of 1 Â 10 À4 mg mL À1 , which exhibited satisfying performance in real samples of duck blood and rice our. 180 To detect thiabendazole, an antifungal agent and also a food preservative that is not approved in the EU, Alsammarraie et al. developed a SERS sensor, in which the utilization of Au nanorods as substrates has enhanced SERS signal. 181 The sensor achieved LODs of 149, 216 and 179 mg L À1 in lemon, carrot and mango juice, respectively. 181 Food dyes provide attractive colors to food and drinking products. However, due to their toxicity, many kinds of food dyes have been prohibited, including Sudan dyes, Malachite Green, Allura Red, Sunset Yellow, etc. Malachite Green (MG) is not only a dye but also an antimicrobial in aquaculture. Therefore, MG can be detected in sh and other seafood products. Due to its affinity to RNA aptamer, Jia et al. employed aptamer labeled AuNPs to detect MG using an anti-aggregation strategy. 182 In the presence of MG, RNA aptamer would bind to MG instead of protecting AuNPs. As a result, AuNPs were aggregated at a high concentration of NaCl and the solution color turned from red to blue. The LOD of the sensor was as low as 15.95 nM. 182 Also taking advantage of that affinity, Zhao et al. developed an indirect colorimetric sensor using CTAB as inhibitors for peroxidase-like activity of AuNPs. 183 The system had 3 states of color. In the absence of RNA aptamer and MG, CTAB inhibited the enzymatic activity of AuNPs, so TMB could not be oxidized and the solution was nearly colorless (the concentration of AuNPs was low). In the addition of RNA aptamer, AuNPs were stabilized and protected, which consequently reduced the adsorption of CTAB on AuNPs, leading to the oxidization of TMB and a blue color in the solution. In the presence of MG, some of the aptamers bound to MG instead of AuNPs, reducing the level of protection, resulting in a light blue color in the solution. The LOD of the method was 1.8 nM (ref. 183) (Fig. 9B) .

3.1.5. Other contaminants. Food and drinks can be contaminated by bacteria, which cause food poisoning or diarrheal disease to human. To control food safety, these food/ water-borne pathogens are also the detecting targets. Several AuNPs-based sensors have been designed and developed for the detection of those bacteria. In general, researchers usually choose antibodies [184] [185] [186] or aptamers [187] [188] [189] [190] [191] to recognize and capture the bacteria. Feng et al. reported the development of a simple sensor for Shigella exneri using aptamer labeled AuNPs. 191 Due to the high affinity of the bacteria to the aptamer, they de-protected the AuNPs. The lack of stabilizing agents made AuNPs easily get aggregated under a high-salt condition. In contrast, in the absence of the bacteria, AuNPs were wellprotected by the aptamer coating and would not be aggregated even at high concentration of NaCl. The sensor exhibited a low LOD of 80 CFU mL À1 and a linear range of 10 2 -10 6 CFU mL À1 . 191 Concerning the specicity of the detection, it is necessary to mention antibody as an effective recognition ligand. Fu et al. proposed a colorimetric immunoassay to sense Vibrio parahaemolyticus, in which AuNPs played the role of a chromogenic substrate, and antibody labeled MnO 2 NPs acted as the recognition element. 186 Unbound MnO 2 NPs were eliminated by external magnetic separating. Once being immobilized on the surface of the bacteria, the MnO 2 NPs generated Mn 2+ due to the etching effect of ascorbic acid. A high concentration of Mn 2+ induced the aggregation of AuNPs, leading to a red-toblue color change. The sensor had an LOD as low as 10 CFU mL À1 and a detecting range covering from 10 to 10 6 CFU mL À1 . 186 Another choice for the capturing element is 4-mercaptophenylboronic acid (4-MPBA). This molecule contains a thiol group to bind to AuNP surface and a boronic acid group that can form covalent bonds with the peptidoglycan cell wall of bacteria. Huang et al. reported a colorimetric sensor using AuNPs functionalized with 4-MPBA via anti-aggregation strategy. 192 In the absence of bacteria, a high concentration of NaCl induced the aggregation of AuNPs. In contrast, in the presence of bacteria, AuNPs are bound to them through the boronic acid group of 4-MPBA, so the solution remained red instead of turning blue, even at high-salt conditions. The obtained LOD was appropriate at 1.02 Â 10 3 CFU mL À1 for Escherichia coli (E. coli). 192 The utilization of 4-MPBA is more simple, easily-prepared and cost-effective, comparing to using antibody or aptamer. Unfortunately, because of the presence of peptidoglycan in every bacterium, it lacks specicity. Therefore, other molecules have been investigated to nd the alternatives for expensive antibodies, which usually requires special experimental conditions. As vancomycin has been conrmed to be able to bind specically to Gram-positive bacteria, 193, 194 Song and coworkers have employed this antibiotic as recognition element in a study of a FRET sensor for S. aureus in 2017, 195 followed by the use of another antibiotic in the same family, teicoplanin, in 2021. 190 The newest system consisted of aptamer labeled quantum dots and teicoplanin factionalized AuNPs. In the presence S. aureus, both of them specically bound to the bacteria, and the distance was close enough for AuNPs to quench uorescent signal of the quantum dots. The method achieved an LOD of 2 CFU mL À1 . 190 In addition, chimeric bacteriophage is another potential candidate. Thanks to the developments on phage display technique, Chen's group have introduced a set of engineered M13 phages displaying the receptor-binding protein from a phage naturally targeting several bacteria, including Pseudomonas aeruginosa (P. aeruginosa), Vibrio cholerae (V. cholerae), and E. coli. 169, 196 The structure of those chimeric M13 phage has been described in a 2019 study (Fig. 8D) , 169 followed by a report on the detection of P. aeruginosa using a thiolated chimeric phage in a crosslinking aggregation approach in 2020. 196 Chimeric phage is a promising candidate to replace antibody as the recognition element for detecting bacteria due to its specicity and robustness. Moreover, it is non-toxic to human.

Biotoxins occur naturally in the environment, which are secreted by organisms to protect themselves from being eaten by others. For instance, mycotoxins are produced by the ones from fungus kingdom while saxitoxin is the most harmful shellsh toxin. To avoid food poisoning, it is important to determine these toxins in potential food before consuming. Therefore, several studies have been carried out to sense biotoxins using AuNPsbased optical sensors. Mycotoxins, such as fumonisin B1 and T-2, have been detected using colorimetric sensors via crosslinking and non-crosslinking approaches, respectively. 197, 198 Cao et al. employed cysteine-modied AuNPs as detecting probes to sense saxitoxin as the toxin could bind to cysteine via electrostatic interactions and hydrogen bonds. As a result, saxitoxin was detected with an LOD as low as 1 Â 10 À7 M. 199 

For disease diagnostics, AuNPs-based sensors aim to detect the target molecules related to a specic disease. They can be molecules and peptides (e.g., glucose for diabetes, dopamine for schizophrenia disorder, glutathione for reperfusion injury, etc.), protein and nucleic acid biomarkers for non-infectious diseases (e.g., cancers, liver diseases, heart diseases, etc.), or infectious pathogens (e.g., bacteria, viruses, etc.) and their antibodies for infectious diseases in human blood, serum, saliva, urine, etc. Although it has been reported that appropriate nanomaterials for biomedical sensing should have the absorbance adjusted to the near-infrared region by changing their sizes or morphology. Thus, the ideal materials for disease diagnosis are large nanospheres, nanorods, nanocage, etc. 75 However, it is the criteria for the substrates for in vivo sensing, i.e., bioimaging. 75, 223 In this review, we focus on in vitro disease diagnosis. Thus, there is no large difference in the nature of the samples in food and health safety monitoring sensors. Therefore, similar to the sensors described in the previous section, in this section, the selection of particle shape in each kind of AuNPs-based optical sensors for in vitro disease diagnosis is still obeyed what we mentioned in Section 2.2. Herein, we will introduce some AuNPs-based sensing systems for both non-infectious and infectious disease diagnostics, focusing on diabetes, cancers and pathogenic bacteria and viruses, especially SARS-CoV-2 (Fig. 10) . The information about other sensors for other diseases is presented in Table 2 .

3.2.1. Non-infectious diseases 3.2.1.1. Diabetes. The glucose level in human blood, serum, saliva as well as urine is an indication of health status. Both high and low glycemic levels are harmful for human health. According to World Health Organization, a level below 3.0 mM (54 mg dL À1 ) of plasma glucose aer a period of fasting can be considered as hypoglycemia. 226 In contrast, plasma glucose maintaining at a high concentration of above 7.0 mM (126 mg dL À1 ) may present hyperglycemia. 227 Persistently high glucose level can be the result of the inability to produce insulin of the pancreas or the inefficiency of utilizing insulin for uptaking glucose into the cells. 228 This is one of the characterized symptoms for diabetes. Therefore, monitoring the glucose level is important to diagnose as well as manage diabetes. In this review, we focus on glucose as a typical example of the molecules that have been detected using many different sensing principles.

Glucose sensors using AuNPs as the detecting probes can be enzymatic or non-enzymatic. Glucose oxidase (GOx) is an enzyme for converting glucose to gluconolactone that releases H 2 O 2 as a by-product. 229 Thus, the level of glucose can be determined by the presence of this by-product. In 2020, Yang et al. designed a tapered ber structure decorated with AuNPs and GOx to sense glucose. 230 AuNPs integrated on surface of the tapered ber were functionalized with GOx. In the addition of glucose solution, GOx decomposed the glucose and produced H 2 O 2 that changed the refractive index (RI) of the medium surrounding the AuNPs. As mentioned in the previous part, the RI change Dn always leads to the shi of LSPR band. The absorbance peak shied toward the longer wavelength region along with the increase in glucose level. This sensor achieved an LOD as low as 322 mM. 230 This sensor also introduced the idea of designing a reusable sensor, which would reduce the cost of the system. However, the limitation of this sensor is that the RI can be affected by various factors. Intrinsic variabilities in complex biological samples can inuence the accuracy of the method. This might be the reason why the sensing system was not employed to detect glucose in real samples.

Another strategy is taking the advantage of the reducing activity of H 2 O 2 to accelerate AuNP etching reaction in the presence of halogenic ions. In a recent study, Donati et al. use Br À as the halogenic ion in a sensing system for glucose. 77 Due to the enzymatic reaction between glucose and GOx in the AuNP solutions, H 2 O 2 was produce. This by-product converted Br À ions into Br 2 molecules, which quickly etched the multibranched AuNPs at their tips and reshaped them into spherical AuNPs. Interestingly, in real samples of saliva, the salivary thiols (i.e., cysteine, glutathione and others) acted as an organic shield that protected and stabilized the NP surface. Therefore, the AuNPs were not aggregated, and the blue-to-red color change of the solution was the result of the transformation of AuNP shapes. At a higher concentration of glucose, a pink color was observed in the reaction solution, indicating a hyperglycemia event (Fig. 11A) . The LOD of the sensor calculated aer analyzing saliva samples from different donors was 1.4 mg mL À1 . Although this LOD is not in line with the best performing glucose sensors mentioned in this review, the system is promising to intrinsic biological variabilities or real clinic samples. 77 The by-product H 2 O 2 can also acts as the analyte in an indirect sensor for glucose. Candem and coworkers designed a sensing system based on 3-mercaptophenylboronic acid (3-MPBA) modied (AuNPs). 231 In the presence of H 2 O 2 , 3-MPBA is oxidized into 3-hydroxythiophenol (3-HTP). The change in the structure of boronate probe was presented by the change among two SERS spectra of the nanoprobe before and aer the H 2 O 2 treatment. For the real samples, GOx and nanoprobes were added into glucose-containing urine solutions, resulting in a change in SERS spectra indicating conversion of the boronate into the phenol as expected. A similar result was obtained in serum samples. 231 Due to the specicity of the enzyme to its substrate, the enzymatic approach provides high selectivity to the AuNPsbased sensors. However, the presence of enzyme requires specic storage and reaction conditions. Thus, we also concern non-enzymatic strategies. Based on the formation of covalent bonds between glucose and boronic acids, Na et al. have employed 3-aminobenzeneboronic acid as the recognition element in a sensing assay using AuNPs and graphene oxide quantum dots (GOQDs). 232 GOQDs displayed orange uorescence, however, this uorescence was quenched by AuNPs. In the presence of glucose and 3-aminobenzeneboronic acid, they bound together to form a cationic species that could cause aggregation of anionic AuNPs. As a result, the uoresce signal of GOQD was recovered. This method resulted in an LOD of 0.65 mM, 232 lower than that of a similar sensing system using phenylboronic acid, a boronic acid derivative in colorimetric approach. 233 Besides, the complex of boronic acid and glucose is also useful to reduce the distance between AuNPs and create "hot spots" for SERS sensing. 111 In another approach, glucose can act as a reducing and capping agent for AuNP synthesis. Hence, Pinheiro et al. have designed a colorimetric sensor with the direct detection of glucose without any intermediate product. 234 The authors stated that low concentrations of glucose resulted in large AuNPs while high concentrations of glucose led to the formation of smaller AuNPs. The difference in sizes of AuNPs was claimed based on the LSPR shis and the color changes of the reaction solutions. The method gave an LOD of 0.65 mM, 234 which is not as low as some others that we have discussed above, but it is still appropriate in physiological range. More importantly, this study has introduced a fast and easily-prepared AuNPs-based sensor without NP modication.

In addition to glucose, glycated hemoglobin is another glycemic indicator, especially to measure the average glucose concentration in the blood of diabetic patients. In comparison to the glucose level, glycated hemoglobin level is more stable due to the 4 months lifespan of the hemoglobin. 235 Therefore, Kwak et al. developed a sensor for detection of glycated hemoglobin in an enzymatic approach. 236 Glycated hemoglobin consists of a glucose molecule binding to N-terminal valine residue of the hemoglobin b-chains. H 2 O 2 was generated due to oxidative cleavage of fructosyl valine in glycated hemoglobin in the presence of fructosyl amino acid oxidase. Subsequently, H 2 O 2 reduced Au 3+ ions into Au atoms and formed AuNPs. The characteristic red color of AuNP solutions and their absorption peak at 520 nm represented for the presence of glycated hemoglobin. Based on the intensity of absorption peak, they calculated the ratio of glycated hemoglobin to total hemoglobin. The sensor exhibited an LOD of 0.124% of total hemoglobin and a linear range of 4.6% to 13.5% of total hemoglobin. Because normal level of glycated hemoglobin levels is within the range of 4-6%, this method was promising in vitro diagnosis for diabetes. 236 3.2.1.2. Cancers. Cancers can be detected through the presence of biomarkers in blood, tissues and other body uids. 237 Cancer biomarkers can be proteins (either secreted or cell surface proteins), carbohydrates or nucleic acids (i.e., DNA, microRNA, etc.). 238 Besides, in some studies, extracellular vesicles and even cancer cells were chosen to be biomarkers for cancer diagnosis. 239, 240 Most of AuNPs-based cancer sensors focus on detecting protein and nucleic acid biomarkers due to the well-understood knowledge of the specic interactions such as antigen-antibody, complementary nucleotide sequences, etc. The discovery of numerous cancer markers has enabled the detection of cancers, for example, p53 (tumor suppressor gene/ protein), 241,242 BRCA, 243 HER2 (ref. 244) and Era (breast cancer), 245 AFP (liver cancer), 246 PSA (prostate cancer), 247 etc.

Nucleic acid biomarkers. To sense nucleic acid cancer biomarkers, the main strategy is based on the interaction between complementary nucleotide sequences. However, because the number of each gene copies in one cell is limited to 2, the concentration of DNA in the samples is consequently low, thus these sensors usually require a step of amplication. 248, 249 It can be either the amplication of target, which is carried out before the detection step, or the amplication of signal, which takes place aer the detection step. As the most widely-used method to amplify DNA molecules is polymerase chain reaction (PCR), a colorimetric sensor for PSA3 has been designed by combining PCR technique and AuNP functionalization. In this 2019 study, Htoo et al. designed primers for PSA3 amplifying using PCR technique, in which the forward primer was thiolated. 247 In the presence of the target gene (PSA3), PRC products were produced as thiolated double-stranded PSA3 sequences, which could bind to AuNPs due to the affinity between their thiol groups and the NP metallic surface. Thanks to the stabilization of the DNA ligands, AuNPs did not aggregate at highsalt condition (red color). In contrast, without the presence of the target gene, no PCR product was formed to stabilize AuNPs, thus, AuNPs aggregated at a high concentration of NaCl (blue color). The red or blue color of AuNP solution represents the presence or absence of PSA3 biomarker in the urine samples, respectively. 247 Instead of thiolated primers, in another study, Trau and coworkers utilized biotin-contained primers to amplify 3 DNA sequences containing 3 important melanoma mutations. 250 These amplicons could be conjugated onto streptavidin coated magnetic bead via biotin-streptavidin binding. Meanwhile, AuNPs were functionalized with 3 kinds of DNA tags complementary to 3 different barcodes on the forward primers for recognition. Moreover, each kind of AuNPs was modied with one Raman reporter. As a result, the presence of the targeted mutation was detected through the characteristic peaks of the appropriate reporter. Thanks to DNA ampli-cation step, this multi-target SERS system achieved an impressive LOD of 0.1% mutation (10 copies). 250 Despite of its advantages, DNA amplication still have several drawbacks including the risk of erroneously amplifying contamination and time-consuming, thus, researchers have developed procedures in which the response signal was ampli-ed instead. For example, instead of amplifying the number of target DNA molecules, Yu and coworkers employed a strategy of "reusing" them. 241 They introduced a sensor for p53 cancer gene using strand displacement amplication on functionalized AuNPs. FAM-labeled hairpin probes were bound on the surfaced of AuNPs, therefore, their uorescence was quenched by AuNPs via FRET principle. In the presence of p53 gene, it hybridized with the hairpin probe due to the complementary sequence. The hybridization opened the hairpin, which not only led to uorescence recovery, but also allowed the primer to bind to the opened hairpin and trigger the polymerization/displacement reactions. Once being replaced, p53 gene was release, then bound to another hairpin to start another round of reaction (Fig. 11B) . The sequences of reactions have amplied the uorescence signal, which allowed p53 gene to be quantied at a concentration as low as 1.6 pM. 241 Another approach of signal amplifìcation is focusing on material fabrication. In a 2020 study, Bai et al. took advantage of the catalytic activity of AuNPs as they can trigger the reduction of 4-nitrophenol, resulting in a color change from yellow to colorless. 243 They designed a capture probe and a signal probe, each of which contains a DNA sequence that is complementary to the target BRCA1 mutated gene. Capture probes were bound to the surface of the wafer while the signal ones were attached to 0D AuNPs. In the presence of the targets, three DNA sequences form a sandwich structure, and the AuNPs with signal probes were immobilized on the surface of the wafer. The colorimetric reaction catalyzed by these AuNPs provided signals for detection of BRCA1 mutation. Then AuNPs were graed on the surface of 2D materials including GO and Bi 2 Se 3 nanosheets, resulting in 4-and 10-fold amplication in signal, respectively. Thus, by improving the material property, the method reached an impressive LOD of 10 À18 M and a linear range of 10 À18 to 10 À12 M. 243 To simplify sensing procedures, researchers are also interested in fabrication of amplication-free sensors for DNA biomarkers. Yu et al. reported on a SERS sensor, in which the target sequence was sandwiched between 2 probe DNAimmobilized on the surface of 2 kinds of NPs: AuNPs and magnetic beads. 251 AuNPs were modied with a Raman reporter. In fact, the structure of this sensing system is relatively similar to the one developed by Trau and coworkers that we mentioned above. 250 However, Yu et al. did not amplify the target gene. Instead, their strategy was enhancing the SERS signal by utilizing hollow AuNPs, which could localize electromagnetic elds through their pinholes. Additionally, the targeted sequence was short (45 nucleotide), so the gaps between two particles were also small enough to create "hot-spots". As a result, the method achieved an LOD of 2.7 fM. 251 Despite the impressive result, it was only an experimental model, in which the target sequence was a PCA3 mimic gene. In real samples, it is difficult to isolate a gene sequence from genome without PCR. This technique might be more appropriate for other kinds of nucleic acid, which are short and single stranded. MicroRNA is a promising candidate.

A dual (colorimetric and FRET) sensor for microRNA-155 was developed by Borghei et al. 252 This microRNA was reported to be over-expressed in many human cancers, 253 so amplication was not necessary. They utilized DNA-modied silver nanoclusters (DNA-AgNCs) as donors and AuNPs as acceptors. In the absence of microRNA, the single-stranded DNA sequences were adsorbed onto AuNPs. Consequently, AuNPs quenched the uorescence emitted from DNA-AgNCs. At the same time, this adsorption also prevented salt-induced aggregation of the AuNPs. In contrast, upon addition of microRNA, the hybridization between microRNA and the DNA immobilized on the AgNCs separated the DNA-AgNCs from the AuNPs. As a result, uorescence recovery occurred. Also, AuNPs easily aggregated at high concentration of salt, reected by a red to purple color change. Although this method exhibited an acceptable LODs of 0.6 nM and 0.4 pM for colorimetric and FRET sensor, respectively, it was reported in the article that those LODs were still higher than amplication step-containing methods. 252 Hence, amplication is denitely important for the sensitivity of sensors for nucleic acid biomarkers.

Due to the highly specic interaction between complementary sequences, the selectivity of cancer sensors using nucleic acid targets is undeniable. In all of the studies mentioned above, an obvious difference was observed in the signal of the target genes in comparison with their mutant sequences, even with 1 or 2 mis-matched bases. 241, 243, 247 However, the necessity of amplication can be a limitation to generate a fast and simple procedure of detecting cancer.

Protein biomarkers. Compared to nucleic acid, protein is a much richer source of biomarkers because many kinds of protein are over-expressed in cancer tissues. 254 However, researchers may encounter the problems of selectivity as the mutations of a few nucleotides only cause a little change in the protein conformation. 255 It requires full understandings of the interactions between these biomarkers and other biological molecules to generate sensitive and selective sensing systems for cancers.

The well-understood immune-interaction of antigen-antibody might be a simple and specic principle to design sensors for protein cancer biomarkers. In a 2015 study, Chen et al. reported on a gold nanorods-based LSPR sensor for cancer antigen 15-3 (CA15-3) using gold nanorods modied with CA15-3 antibody. The presence of CA15-3 led to increase of intensity of absorption spectra, red-shis of the absorption peaks and the broadening of the peaks, corresponding to the assembly of the gold nanorods. 256 Similarly, the mucin protein, MUC4, which is overexpressed in pancreatic cancer, was also recognized thanks to antibody-antigen binding in a SERS sensor. 257 However, due to the large size of the antibody immobilized on AuNPs, the hotspots for SERS signal enhancement were not created due to the assembly of AuNPs through binding with the same antigen. Instead, they designed a template stripped gold (TSG) substrate to immobilized antibodies. Thus, MUC4 antigens were captured between the substrate and the AuNPs in a sandwich model. The AuNPs were modied with a Raman reporter. Therefore, the signal of the reporter represented for the presence or absence of MUC4. In addition, the concentration of antibody immobilized on the substrate could be controlled to ensure the distance among those antibodies as well as interparticle distance among the AuNPs aer the recognition step were close enough to create SERS hot-spots. 257 Another example of the SERS sensor using this interaction is a sensor for AFP developed by Zhang et al. in 2018. 258 This system consisted of both AuNPs and AgNPs, in which AgNPs were functionalized with AFP antibodies while AuNPs acted as the capturing probe for the attachment of AFP antigens in the sample. Due to the antibody-antigen interaction, the distance between the nanoparticles was reduced, leading to formation of "hot spots" that enhanced the SERS signal. As expected, the method exhibited high specicity. However, in complex real samples, it expressed poor sensitivity compared to other methods. 258 This can be explained by the high biocompatibility of AuNPs, which allows the attachment of many other molecules rather than AFP antigens.

Recently, based on the interactions between protein molecules and DNA sequences, AuNPs have been functionalized with aptamers and employed as sensing probes for colorimetric sensors. Protein biomarkers such as PSA and p53 was detected using AuNPs-based aptasensors via cross-linking strategy. 242, 259 Meanwhile, de-protection strategy was employed for detection of others such as HER2, 244 EAa 245 and carcinoma embryonic antigen (CEA). 260 In cross-linking models, the aptamers had to be designed as thiolated DNA 242 or containing poly A sequences 259 so that they could be attached to AuNPs. In the presence of target biomarker, aptamers-modied AuNPs aggregated due to the specic interaction between their coating agents and the target protein. As a result, a red to purple color change could be observed with bare eye. 242, 259 On the other hand, aptamer acted as a protecting agent to prevent saltinduced aggregation of AuNPs by absorbing onto the metallic surface in de-protection models. In the addition of target protein, the aptamer would bind to the target due to a higher affinity, and the AuNPs aggregated easily at high concentration of salt, which also resulted in a red to purple color change. This method is simple and fast with the incubation time of only 5-10 minutes. 244, 245, 260 Moreover, it can avoid the use of antibodies, which are required strict conditions of storage and utilization. More importantly, the selectivity of aptamer is undeniable. For example, in the study on p53, the mutant p53 could be distinguished from the wild-type. 242 The lack of a functional tetramerization domain in mutated p53 prevented it from inducing aggregation of aptamer-functionalized AuNPs. In contrast, the AuNPs aggregated in the presence of wild-type p53 as the aptamer contained recognition element for the binding of p53.

In another attempt to detect AFP, Zhou et al. combined both antibody-antigen and protein-DNA interactions in a FRET sensing system, where the aptamer labeled luminescent CdTe quantum dots (QDs) were the donors and AFP antibody graed AuNPs were the acceptors. 246 In the presence of AFP antigen, its affinity to antibody and aptamer reduced the distance between the two kinds of NPs. As a result, AuNPs quenched the uorescence of QDs. This "turn-off" sensor exhibited a linear range of 0.5-45 ng mL À1 , with an LOD of 400 pg mL À1 . Moreover, it also showed a satisfying performance in human serum samples. 246 It is worth noting that in this study, the author used 2 sites of recognition instead of one, which increased the specicity of the sensing method.

The combination of different strategies may result in interesting results. In a 2021 study, Shen et al. were successful to design and develop a FRET sensor for cancer antigen 125 (CA-125) at trace amounts by combining protein-aptamer interaction and DNA amplication. 261 They designed 2 partly complementary sequences, named Oligo 1 and Oligo 2. Oligo 1 contained an aptamer for CA-125. In the absence of CA-125, the partially hybridized sequences were extended with Klenow fragment (exo) polymerase to obtain a new double stranded DNA molecule. This new DNA was employed as the template for PCR-amplication. Both forward and reverse primers contained a spacer, resulting in the formation of single-stranded DNA at two sides of the PCR product, which was complementary with Oligo 3 and Oligo 4 immobilized on the surface of AuNPs. Moreover, each forward primer also contained a uorescein (FAM). Aer hybridization, the uorescence of FAM was quenched by gold nanoparticles due to FRET effect, resulting in a decrease in uorescence signal. In contrast, in the addition of CA-125, the antigen bound to Oligo 1 and prevented the hybridization between Oligo 1 and Oligo 2. It also terminated all the following process. Thanks to its conformation (Fig. 11C) , the forward primer did not hybridize to the DNA probes on the AuNPs directly, thus, their uorescence intensity remained.

Despite its complexity, the method achieved an impressive LOD of 1.5 fg mL À1 . 261 Moreover, it has revealed the unlimited potential of combining different strategies and techniques in fabrication of novel systems.

Other biomarkers. In addition to nucleic acid and protein biomarkers, other tumor-related elements have been studied as new markers to detect cancers, including extracellular vesicles (EVs) secreted from cancer cells 262 and cancer cells themselves. 128 Among them, extracellular vesicles are more widelyused because they are the richer source. The advantages of using these large biomarker is that they contain many different membrane proteins, which can be employed as targets for detection. 263 In 2020, Wang and coworkers introduced a sandwich-type SERS sensor for EVs tumor. 264 The system included two kinds of aptamer targeting two different sites of the EVs that were used to label AuNPs and agarose beads, respectively. Even in complex samples, EVs could selectively bind to agarose beads, followed by the attachment of AuNPs to form sandwich complexes. The enhancement in SERS signal was observed due to this formation. The LOD of the sensor was as low as 2.44 pg mL À1 . 264 Similarly, Oliveira-Rodríguez et al. also employed two tetraspanins on the membrane of exosome as detecting target in their colorimetric sensor. 263 The utilization of two binding sites on one target can increase the selectivity of the sensing systems.

On the other hand, cancer cells were the biomarkers in several studies. Butler et al. developed a near infrared SERS sensor based on 150 nm AuNPs to detect MCF-7 breast cancer cells. 128 A signicant enhancement of Raman signal was observed in the presence of MCF-7 cells. However, without specic functionalization, this sensing system were not effective in protein-rich samples, such as serum, with a low signal-tonoise ratio. 128 Also in an effort of sensing MCF-7 cells, Li et al. introduced a colorimetric sensor, in which the cancer cells were recognized by Mucin 1 protein (MUC-1) aptamer-modied gold nanorods. 265 In the presence of the cancer cells, the gold nanorods bound onto their membrane, which can be observed by UV-vis spectroscopy. Thanks to the specic binding of the aptamer and MUC-1 protein, the sensors could distinguish the target MCF-7 and untumorigenic cells HepG-2. 265 3.2.2. Infectious pathogens control. Infectious diseases are caused by different kinds of pathogens such as bacteria and viruses. Therefore, there are several approaches for diagnosis of one specic infectious disease, including detecting the whole or a part of the pathogen particle, the whole or a part of its genome or the antibody produced by the body in the presence of pathogen. 266 As mentioned in the part about food safety, one whole bacterium can be captured and determined by a sensing system. 169, 196 However, considering the high infectivity of pathogenic bacteria in human clinic samples, it is safer to inactivate them and use their genome for detection. For example, Mycobacterium tuberculosis (TB) collected from oral swabs should be heat-treated before being further studied. 267, 268 Tsai et al. proposed a colorimetric method to sense a TB sequence containing IS6110, using the non-crosslinking strategy. 268 Similar to other studies on detecting DNA sequences mentioned above, 247 the target sequence had to be amplied by PCR before the detection. The authors designed a single stranded DNA probe that was complementary to IS6110. The DNA probes were adsorbed on the surface of AuNPs. Once the sample containing IS6110 was added to the solution, the DNA probe would bind to the target sequence, thus de-protected the AuNPs, and AuNPs were easily aggregated (Fig. 12A) . As a result, the absorption spectra of the solution shied towards the longer wavelength region while its color became more blueish. The method had an LOD of 1.95 ng mL À1 for TB DNA. 268 In another study, Owens , an TB antigenic marker, as the target to detect TB. 269 Two kinds of anti-ManLAM antibodies were employed to functionalize the AuNPs and the capture substrate, respectively. Hence, ManLAM antigen would be captured between them. The SERS signal was measured to determine the presence of ManLAM with an LOD of 10 ng mL À1 . 269 Similarly, for virus detection, it is possible to detect the whole virus or its genome. Many kinds of inuenza viruses were detected as whole virus particles, [270] [271] [272] in which AuNPs were functionalized with recognition elements to sense the viruses. The most frequently-used capturing probe is antibody so that AuNPs would assemble on the surface of the virus. Thus, the presence of viruses could be determined due to a red to purple color change caused by that assembly. 271, 272 In another approach, Ahmed et al. reported that the organization of AuNPs on the surface of virus had enhanced the peroxidase-mimic enzymatic activity of positively-charged AuNPs. 270 Compared to AuNPs only, AuNPs-decorated inuenza viruses exhibited stronger catalytic activity when catalyzing the oxidation of TMB by H 2 O 2 , resulting in a deeper blue color in solution, reecting in higher absorption intensity at 665 nm. It was the main principle to determine the presence of inuenza virus H1N1 and H3N2. 270 Instead of using antibody, Li and coworkers employed AuNPs modied with glycan containing terminal sialic acid (SA). 273 The structure of this terminal SA played a decisive role for the binding of the functionalized AuNPs to the viruses. The binding of SA receptors to the viruses resulted in the aggregation of the AuNPs on the virus surface, and subsequently the blue color of the solution. In contrast, in case the SA receptor was not specic to the virus, the AuNPs would not bind to the viruses and the solution would remain the red color. Using a set of AuNPs with 7 different structures of SA receptors, the authors differentiated 14 inuenza virus strains, including the most major subtypes in human and avian populations. 273 Besides, an example for virus detection through its genome is a colorimetric assay to detect MERS-CoV developed by Kim et al. 274 They designed a pair of thiolated probes at 5 0 or 3 0 end, which were complementary to MERS-CoV sequence. In the presence of the target DNA sequence, they reorganized into a double stranded DNA, which was assembled on the surface of AuNPs and prevented them from aggregation at high-salt condition. In contrast, the absence of target sequence led to the aggregation of AuNPs. The difference in color could be observed using bare eye or UV-Vis spectroscopy. The assay had a low LOD of 1 pmol mL À1 . 274 3.2.3. Emerging COVID-19 disease control. Since December 2019, the whole world has been under the threat of a new Severe Acute Respiratory Syndrome Coronaviridae virus, known as SARS-CoV-2. Many detecting approaches for this kind of virus have been investigated, including the use of AuNPs. However, short research time and limited understanding about the structure of this virus only allowed researchers to provide simple reports of SARS-CoV-2 sensing. For fast detection, most of them described the sensors to detect the whole virus particle. 224, 275, 276 For example, Alfassam et al. proposed taking advantage of hemagglutinin (HA) protein on the surface of the virus as target for detection. 275 Therefore, they fabricated sialic acid (SA) coated AuNPs, so the binding of HA of the virus to SA on AuNPs would lead to the aggregation of AuNPs, resulting in the spectral shi of LSPR. However, this sensor lacked of selectivity because it exhibited the same positive result for all three types of respiratory virus: SARS-CoV-2, inuenza B, and MERS-CoV. 275 Ventura et al. described another sensor using 3 kinds of antibodies targeting 3 kinds of protein on SARS-CoV-2 surface, including spike (S), membrane (M) and envelop (E). 224 Obvious changes in solution color and absorbance have been observed in the presence of the virus (Fig. 12B) . 224 To increase sensitivity and selectivity of the sensor, Huang et al. have immobilized the virus on a sensing chip via antibodies coated on the chip surface in an immunoassay. 276 AuNPs were functionalized with another kind of antibodies to bind to a different epitope of the virus. Using the whole virus directly as a target element, this method is time effective (<15 min) with a low LOD of 30 particles. 276 Jiao and coworkers have avoided the threat of being infected while working with viruses by choosing a different target, IgM antibody against SARS-CoV produced in the human body in the presence of the virus. 86 They fabricated a testing strip, where SARS-CoV-2 nucleoprotein (SARS-CoV-2 NP) was coated on a membrane to capture the target antibody. Antihuman IgM was attached to AuNPs. In the presence of the antibody, it was immobilized between the nucleoprotein and AuNPs, which introduced a line in characteristic red color on the strip. 86 Through this study, the authors have reported a fast testing of SARS-CoV-2 using serum samples rather than oral swabs. The limitation of this approach was that it required a relatively high concentration of antibody in human serum to form a visible red line on the testing strip. Therefore, SARS-CoV-2 could not be detected in early stage of infection using this method. In an effort to fabricate a more sensitive sensor for SARS-CoV-2 antibody, Lew et al. designed a colorimetric sensor to detect SARS-CoV-2 IgGs in patients' plasma. 277 The IgG binding affinity of four immunodominant linear B-cell epitopes, located on the spike (S) and nucleocapsid (N) proteins, were characterized to nd the two most sensitive ones, which were subsequently biotinylated and conjugated to streptavidin-coated AuNPs. In the presence of SARS-CoV-2 IgGs, these epitopes played the role of recognizing target antibodies, triggering to the aggregation of AuNPs (Fig. 12C) . Moreover, co-immobilization of two epitopes led to an improvement in the sensitivity of the method, in comparison to single-epitope modied AuNPs. The assay achieved an LOD of 3.2 nM. Hence, it could be potentially used to detect IgG for convalescent COVID-19 patients. Using for 35 clinic samples, including Covid-19 patients' samples, the novel method successfully identied SARS-CoV-2 infection with 100% specicity and 83% sensitivity. 277 In addition to the method using antigen-antibody interaction, recently, detecting SARS-CoV-2 using aptamers has attracted the interest of researchers. 278, 279 In a recent study, Aithal et al. employed aptamer-functionalized AuNPs to target S protein of the virus in a colorimetric sensor. 279 In an antiaggregation approach, the binding of S protein to the aptamer prevented to the aggregation of the AuNPs at high-salt condition (Fig. 12D) . The sensor was well-performed in the presence of isolated S protein as well as heat-inactivated SARS-CoV-2 viruses. The nanoprobes could detect 3540 genome copies/mL inactivated SARS-CoV-2 virus. However, it should be further validated with real patients' samples. 279 

Recently, the introduction of multi-metallic nanomaterials, in general, and multi-metallic gold-based nanoparticles, in particular, has opened a potential of developing a new generation of optical sensors. To achieve the best plasmonic effect, noble metals such as silver (Ag), platinum (Pt), copper (Cu), etc. have been selected to combine with Au in the manufacture of multi-metallic nanostructures. 289, 290 It has been reported that these nanomaterials exhibited stronger LSPR bands than monometallic ones. 289, 291 Moreover, the position of LSPR band in UV-Vis spectrum of a multi-metallic material can be controlled by adjusting the ratio and the arrangement of its components. [289] [290] [291] Besides, in another study, Bai et al. calculated electromagnetic eld distributions of AuNPs and several multimetallic NPs using Lorenz-Mie theory. 292 The results revealed signicant electromagnetic enhancement of alloy and coreshell multi-metallic structures (i.e., Ag-Au NPs and Au@Ag NP, respectively), and especially alloy core-shell structure (i.e., Au@Ag-Au NPs), in comparison with mono-metallic AuNPs. 292 Therefore, multi-metallic gold-based NPs are promising to develop colorimetric (or LSPR) and SERS nanosensors. However, there has been no report on the property of quenching uorescence of these NPs, especially in comparison to AuNPs. As a results, studies on multi-metallic gold-based NPs as substrates for optical sensors for food and health safety monitoring have focused on colorimetric and SERS sensors. For instance, the strong LSPR band of bimetallic Ag-Au NPs allowed George et al. to develop ultrasensitive colorimetric sensors for Mn 2+ and ciprooxacin, an antibiotic banned for animal food. 293 Mn 2+ and ciprooxacin were detected at low concentrations down to 18.40 and 10.26 nM, respectively. 293 In another approach, core-shell Au@Ag nanorods were employed to detect AFP, a protein marker of hepatocellular cancer. 294 By controlling the thickness of Ag shell and the aspect ratio of Au core, the authors adjusted the color of nanorod solution to achieve the appropriate color range for bare-eyed observation. 294 Besides, the noticeable electromagnetic enhancement of core-shell structures, especially at the internal gap between the core and the shell, has attracted the attention of researchers in the effort of designing and developing effective SERS substrates. 292, 295 For example, Au@Ag NPs was designed and optimized to detect pesticides, including tricyclazone and thiram in pear extracts, resulting in impressive LODs of 0.005 and 0.003 ppm, respectively. 296 In addition, Cu@Au NPs exhibited a great potential to be appropriate SERS substrates for various biomolecules, including acetaminophen, dopamine, cholesterol, etc. 290 Furthermore, the complex core-shell alloy structure of Au@Ag-Au NPs showed the best performance compared with core-shell (Au@Ag NPs), alloy (Ag-Au NPs) and mono-metallic structures (AuNPs) in SERS sensing system for cardiac troponin I, a biomarker for cardiac injury. 292 Despite the great potential, the application of multi-metallic gold-based NPs for development of optical nanosensors for food and health safety is still limited with only a few studies, and very few of them were carried out in real samples. 296 Compared to mono-metallic AuNPs, complicated procedures of multimetallic NP manufacture 297 might have been prevented them from being widely-employed in sensing systems. Readers who are interested in the manufacture of multi-metallic NPs may nd useful information in the established review by Srinoi et al. 297 Nevertheless, the superior optical properties of multimetallic gold-based NPs are undeniable, therefore, they still have space to grow as substrates for optical sensors for food and health safety monitoring in the future. The huge amount of information in the tables has provided a picture of recent developments of AuNPs-based sensing systems for food safety control and in vitro disease diagnosis. Fig. 13 exhibits the most concerned detecting targets in both elds with the presence of long-term and emerging ones. In general, the development of AuNPs-based optical sensors have always caught up with urgent needs of the society. For example, virus outbreaks have triggered the investigation of their detecting sensors. High toxicity, even at low concentrations, of pesticides 298 and melamine have led to the demand of monitoring these food contaminant/additive, especially when melamine contamination is usually related to enfant milk. 299 Diabetes and cancers are also concerning as the number of patients suffering and dying of these disease keeps rising every year. 300, 301 Herein, we present several trends of developing AuNPs-based optical sensors in the recent 7 years.

With a large number of reported studies, obviously, AuNPsbased optical nanosensors are widely studied to monitor food safety. Since 2015, many food contaminants and additives have been detected using colorimetric, FRET-and SERS-based sensors. Starting with analytes possessing simple chemical structure such as heavy metals (e.g., Hg 2+ , Pb 2+ ) and melamine, researchers have design various sensing systems based on their well-understood interactions with DNA, 201 DNAzyme 205 and peptides 207 (for heavy metals) as well as citrate, 219 methanobactin 217 and DTT, 173 etc. (for melamine). Thus, heavy metals and melamine were the targets for many reported optical nanosensors. Moreover, recent studies about melamine detection also concern the improvements in technologies, including newly-investigated ligands 218 and the combination of known ligands 170 to evaluate the performance of the optical sensors. They not only introduced effective sensing tools for the most popular contaminants and additive in food but also provided the concepts for the following studies. The enlarged knowledge about other food contaminants and their specic interactions to other compounds allow the fabrication of other sensing systems in the later period. We can clearly observe an upward trend in determination of toxins as many analytes in this category have been detected using different kinds of optical sensors and in various forms of real samples in recent 3 years. 95, 199, 214 This trend may continue to grow in the following years. Detection of pesticides and herbicides is also a long-term concern of researchers. In fact, before 2016, pesticides and herbicides were determined using AuNPs-based optical sensors. 156, 302 However, it was not until 2017 that we observed a wave of research on pesticide and herbicide detection as many kinds of targets have been detected. The developments in nanotechnology have led to the signicant improvements in the sensitivity of the sensors. For instance, SERS sensors for pesticide detection designed by Huang et al. in 2020 159 resulted in impressive LODs which were 10 times lower than those of the studies reported in 2018 (ref. 112 ) and 2016, 161 thanks to the fabrication of snowake-like AuNPs and the utilization of specic ligands. Recently, many states and countries has participated in numerous Free Trade Agreements, which allows the trading of goods, including various kinds of food, all over the world. Therefore, cost-and time-effective sensing tools to monitor food safety and quality have been in high demand. It opens an opportunity for AuNPsbased optical sensors to be further investigated and applied in this large area in the near future.

Besides, the AuNPs have maintained their position as one of the most appropriate substrates to develop optical nanosensors for in vitro disease diagnosis during the observed period (i.e., 2015-now) thanks to their excellent biocompatibility. 22, 39, 88 The non-infectious diseases such as diabetes and cancers have been always long-term concerns of researchers as many kinds of optical sensors have been generated to monitor the glucose level or determine varied tumor markers. Because diabetes and cancers are still unsolved problems of modern mankind, 303, 304 these studies may be continued in several years. On the other hand, the trends of investigating optical sensors for detection of infectious diseases usually follow their outbreaks. For examples, inuenza viruses, which have been the perpetrators for the outbreaks of u, including bird u, 273 swine u, 270 and seasonal u, 271 etc., have attract the interest of researchers for several years. The turning point came in December 2019 with the outbreak of a new corona virus (i.e., SARS-CoV-2), which then developed into COVID-19 pandemic, one of the most serious health crisis in the human history. 305, 306 Consequently, we are witnessing an explosion of researches on detecting COVID-19, including AuNPs-based optical sensors. Until now (February 2022), the world is still under the threat of the pandemic with the appearance of the new variants, and the virus shows no sign of stop. Therefore, the studies on SARS-CoV-2 will be continued, and the AuNPs-based optical sensing systems for rapid COVID-19 detection is still on the road to optimization.

Following the recent advances, we can see that antibody is losing their place as the most ideal recognition element. Although antibody provides selective targeting, more and more groups are choosing aptamer, which is more stable and smaller in size compared to antibodies. However, it cannot be denied that biological molecules are not as stable as chemical ones. Their affinity to analytes would reduce aer a long time of storage, especially in hot weather. Researchers have started solving this challenge by nding alternative such as synthetic organic compounds, 111 antibiotics 190,195 and chimeric phages. 169, 196 Another trend is the utilization of smart phone in recording and analyzing signals, especially with colorimetric sensors, in RGB color model. 132, 158, 221 Because almost everyone today possesses a smart phone, it is a step closer to the application of sensor in real life.

Along with the developments of knowledge and new technology, researchers have focused on various AuNPs-based sensing systems for food and health safety monitoring. However, we are still facing several challenges to improve and optimize the performance of these sensing systems, including their sensibility, reliability and practicality.

4.2.1. Sensibility. An effective sensor requires a low limit of detection and a large linear detecting range. In general, the reported AuNPs-based optical sensors listed in the tables exhibits appropriate LODs. For example, all the LODs of those melamine sensors are lower than the maximum acceptable concentration in food set by US FDA (50 ppb $0.3 mM). 307 However, in colorimetric sensors, the linear range is usually limited due to the aggregation of AuNPs. The introduction of analytes at high concentration leads to the rapid formation of large aggregates and precipitates of AuNPs, causing a fast reduction in the intensity of absorption spectrum, which is usually out of the linear range. Chen et al. have suggested a solution for this challenge by improving the stability of AuNPs against aggregation by partially functionalizing the AuNPs with PEG. 170 In this study, they attached citrate-stabilized AuNPs onto a APTES-activated glass substrate via electrostatic attraction. Subsequently, thiol-PEG-acid chains were introduced onto the "untouched" surface of AuNPs. Denitely, this approach has enlarged the linear range of the AuNPs-based melamine sensor toward the higher-concentration side. However, the preparation of AuNPs was complicated. Moreover, it is difficult to apply this concept to other sensing systems using any ligands other than citrate. Therefore, improving linear ranges for colorimetric sensors is still challenging.

4.2.2. Reliability. In order to develop appropriate AuNPsbased optical sensors, the parameters of reliability such as selectivity, stability, reproducibility, etc. cannot be ignored. Following the recent reports presented in the tables, it is not difficult to notice that in many sensing systems, citrate has been chosen as the recognition element for the detection of varied analytes such as melamine, 170 dopamine, 225 paraquat, 212 etc. This choice has simplied the fabrication of AuNPs-based sensors. Moreover, those sensors have exhibited appropriate LODs. However, it is obvious that those sensors are lacked of selectivity. To improve the selectivity of the methods, it requires the utilization of more specic ligands on the AuNPs surface. Nevertheless, in SERS sensors, there is a conict between the selectivity and the sensibility of the method. SERS sensors require an intimate contact between the AuNPs surface and the analyte for signal enhancement while the specic recognition element may enlarge the distance between them. The trend of investigating alternative ligands mentioned in the previous part of this Section may be the solution to balance this conict. In addition, being performed on dried substrates, SERS sensors usually exhibit lower reproducibility compared to the ones performed in solutions. Neither drop-casting 159 nor pastingand-peeling 161 technique is able to ensure the homogeneity and reproducibility of SERS signal on the substrate. 4.2.3. Practicability. Many studies have described rapid, sensitive, selective sensors. However, they are still laboratorybased, and practicality is still a major challenge. Researchers are in the struggle between generating simple and user-friendly sensors versus selective and sensitive ones. We are impressed by the LODs and linear ranges of many sensors, but their complicated preparation limits the potential for wider application. 241, 243, 246 Among all sensing systems mentioned above, many of them were claimed to have been employed in real samples (Table 1 & 2) . However, most of these "real samples" need to be managed in the laboratory with many steps of pretreatment. 170, 222, 308 Even though there are a few sensors fabricated using simply-prepared samples, 170, 222, 308 they are all clinical samples, which have been managed strictly through a standard procedure. It would be much more difficult for a real sample of food. Real food samples are extremely complicated and unpredictable. They might contain a large amount of salts that induces the aggregation of AuNPs. In addition, they can include oil and oil-soluble elements. Therefore, even when a sensor is designed for one specic food product, the pH of the sample can change due to the inuence of unexpected contaminations. Similarly, saliva samples for rapid sensing of drugs or biomarkers are also uncontrollable with many variables. 258 Besides, due to the complexity of real samples, each of which might contain more than one desired target, optical sensors have been designed and developed towards multi-target directions. However, among numerous examples that we have mentioned above, there is only one sensor that can detect more than one molecule at once. It is the sensing systems for 3 kinds of pesticides by Zhang et al. in 2020, 158 in which 3 different uorescence signals from 3 kinds of uorophores were recorded, analyzed in RGB model to calculate the contents of each kind of pesticides. However, a uorescence-based sensor always requires some instruments and tools to record and analyze the signals. Colorimetric sensor is a more user-friendly approach. Therefore, several groups have investigated multi-target colorimetric sensor. Wu et al. employed multi-functional aptamer labeled AuNPs 132 while Yan et al. proposed a pattern recognition methods using two kinds of AuNPs functionalized with two different aptamers. 168 However, due to the "red-to-blue" trap, these sensors are not accurate in complex samples contained two or more targets.

Gold nanoparticles are excellent transducers for optical nanosensors including LSPR, colorimetric, FRET-based and SERSbased sensors for detection of food contaminants and diseases. Examples on advanced sensing systems have been summarized to draw a general picture of the recent developments of AuNPs-based optical nanosensors. In-laboratory works have shown impressive achievements in detecting many kinds of analytes. However, on the road of optimizing their sensing performance, there are several challenges. In the future, by solving all those challenges and taking advantage of the developments of every eld of science and technology, sensordevelopers may fabricate a new generation of AuNPs-based optical sensors, which is smarter, achieves signicant improvements in sensing performance and accomplishes the goal of practical applications. To generate smart sensors, researchers need to combine the latest technologies including Internet of Things (IoT), Articial Intelligence (AI) and machine learning to their sensing systems. These tools are able to collect, manage and analyze a large amount of data. Therefore, the data recorded by the sensors may be rapidly collated to the data bank, followed by the fast determination of the sensing system. Besides, the sensing performance of the sensors will be improved in every aspect. Knowledge about the interactions between analytes and recognition elements will be fullled, so more highly-specic elements (either natural or synthetic ones) at appropriate sizes will be employed for the detection of desired targets. Obviously, stable and durable ligands are highly desirable to enhance to stability of the systems. In addition, further studies on multi-metallic gold-based nanomaterials and their applications of optical sensing may be a direction to elevate the sensibility of the sensors. Finally, the new sensors will be more practicable. Standard but irreducible preparation procedures will be investigated in order to minimize the inuence of environmental factors on the sensing performance, but still be user-friendly. Besides, the success of glucose meter has established the idea of developing hand-held spectrometers for LSPR, uorescence and SERS measurements. The manufacture of these devices in combination with the development of professional analysis tools in smartphones might make AuNPsbased optical nanosensors become accessible for everyone. Furthermore, multi-target sensors will be the new direction to monitor complex samples, especially food products. Beside the red-to-blue range, sensor-developer may add more color ranges by combining indirect colorimetric sensing to the systems. In another approach, taking advantage of the different colors of uorophores may be the key for the detection of several targets at once using uorescence sensors. In term of material fabrication, AuNPs should be synthesized using robotic systems. It will ensure the homogeneity among different synthesis batches. Moreover, it will not only reduce the requirement of specialized human resources, but also allow highly-yielded manufacture. As a result, AuNP fabrication will be more cost-and time-effective, which is essential for the goal of high practicability. Aer several decades of research, the practical applications of optical sensors based on AuNPs are still limited. Hence, greater efforts should be made towards their practical utilization in the future. 

The authors declare that they have no known competing nancial interests or personal relationships that could have appeared to inuence the work reported in this paper.

Body Sensor Networks

Raman Spectroscopy in the Undergraduate Curriculum

Metal Nanoparticles for Drug Delivery and Diagnostic Applications

Division of Child Health and Development and Maternal Health and Safe Motherhood Program, Hypoglycaemia of the newborn: review of the literature, World Health Organization

Denition and diagnosis of diabetes mellitus and intermediate hyperglycemia

Advanced Techniques in Diagnostic Microbiology

Global Climate Change

This research was funded by Vietnam National Foundation for Science and Technology Development (NAFOSTED) under grant number 103.02-2018. 48 . The authors would like to thank application-driven research program on Smart Sensing Technologies between ITIMS-HUST, IMS-VAST and MSE-NANO at the Phenikaa University.