Open Access book publisher

Milk and dairy products are important nutrients for all age groups. However, the use of antibiotics for the treatment of food-producing animals generates the risk to human health, as these compounds and their metabolites can be transferred into milk. Rapid testing of the presence of antibiotics in raw milk to grant its quality has become a major task for farmers and dairy industry. The conventional analytical methods are either too slow or do not enable quantitative detection of antibiotic residues, so alternative methods that are rapid, cost effective, and easy to perform should be considered. The present chapter gives an overview of the recent developments and issues of the construction of different biosensors for the detection of antibiotic residues in milk.


Introduction
Milk and dairy products, generally considered to be healthy and nutritionally balanced natural food, comprise essential nutrients for all age groups and are an important part of our everyday diet. However, the use of antimicrobial drugs for the treatment of food-producing animals generates the risk to human health due to the transmission of the residues and metabolites of these compounds into the food chain. Hypersensitive consumers may be subject to allergic reactions or even more severe health problems. At present, up to 10% of people have already been diagnosed to be hypersensitive or allergic to antibiotics [1,2]. In addition to direct health problems, the presence of antibiotic residues in milk has an adverse effect on milk fermentation processes in dairy industry, as many starter cultures used for the production of fermented food products (cheese, yoghurt, etc.) may be inhibited by antimicrobial substances and the product quality will be impaired [1,2]. Scientists and health experts also fear that wide application of antimicrobial agents is contributing to the rise and spread of antibiotic-resistant bacterial infections [3].
The overall sales of veterinary antimicrobial agents, used in food-producing animals in EU (24 countries, excluding Croatia, Greece, Malta, and Romania), were 7974.2 tons of pure ingredients in 2012. The sales of the most commonly used antibiotic classes are shown in Table 1 [4]. Based on the sales data, the classes of antibiotics shown in Table 1 are of main concern regarding the need for practical analyses of antibiotic residues in food, including milk.  Table 1. Sales of veterinary antibiotics used in food-producing animals in EU (24 countries, excluding Croatia, Greece, Malta, and Romania) in 2012 [4] To protect consumers, strict legislative regulations have been imposed for the treatment of animals with antibiotics and maximum residue limits (MRLs) in foods of animal origin, which are not to be exceeded, established for residues of antibiotics or their metabolites. The residue levels in milk and other body fluids are dependent on the physicochemical properties of a particular drug, which determines the pharmacokinetics and time course of drug concentrations in the body [5]. Table 2 lists the MRL values in milk in EU for the most common antibiotics used for the treatment of dairy cows [6].
Since the introduction of EU regulations on MRLs of pharmacologically active substances in foodstuffs of animal origin ( *Sum of the parent compound and its 4-epimer **Sum of all sulphonamides should not exceed 100 µg/l Table 2. The MRL values for the most commonly used antibiotics in milk [6]. Routine testing of milk for the detection of residues of different antibiotics to grant the quality and safety of milk has become a major task for farmers and dairy industry. At present, two major technologies are commonly used for milk analyses: qualitative milk screening tests [7][8][9] and various chromatography-based techniques [7,10]. Qualitative microbial inhibition tests have been proven to be very suitable for milk screening purposes [7][8][9]. These tests comprise spores of specific bacteria, sensitive to particular antibiotics on agar gel including nutrients for bacterial growth and a pH indicator. After milk is added to the test, it is incubated at the appropriate temperature to germinate and grow the spores. In the absence of antibiotic residues, the growth of bacteria can be detected visually either by the change of opacity of the agar medium or by the color change of the pH indicator, resulting from acid production and change of pH. In the presence of antibiotic residues (or any other inhibitors), the growth of bacteria is suppressed and there are no observable changes of the color. The main advantages of these tests are their low cost, simple performance, and broad selection toward different antibiotics. Although these tests are considered to be rapid, they take 3-24 hours to perform in an incubator. The bacterial strains used in tests should be constantly monitored to ensure that they have not become resistant to the antibacterials. The interpretation of test results is quite subjective and may lead to false negative or positive results. The presence of natural inhibitors in abnormal milk (e.g., milk of mastitic cows or colostrum) can be the cause of false positive results [7][8][9].
In addition to microbial inhibition tests; there are different rapid tests, based either on immunoassay or enzymatic operation, available for the screening of a number of antibiotic residues in milk. These tests provide results usually within 30 min [11]. As already said, milk tests are specific to the particular class of antibiotics. As a rule, the detection limits of milk tests are in the range of the established MRL values [7,8,[11][12][13].
Chromatography is the most reliable technique for quantitative detection of antibiotic residues [7,10]. Nowadays more than 80% of the analytical techniques for the determination of veterinary drugs use high-performance liquid chromatography in combination with mass spectrometry (HPLC/MS) [10]. However, chromatography-based methods require expensive equipment and trained personnel with high experience. In addition, HPLC techniques demand laborious pre-treatment of samples for the extraction of the compound analyzed, from the sample matrix.
As the milk matrix is one of the most complex ones, the application of biosensors, enabling a selective detection of particular compounds in natural or only minimally pre-treated samples is a good option for the on-site assessment of milk quality [7,[14][15][16]. Biosensors are compact devices transferring the selective biochemical recognition into a measurable physical signal, which can be translated into an indication of the safety or quality of milk. Physically biosensors comprise of bio-recognition and signal transduction elements. Biosensors offer an opportunity for the development of quick and portable devices for real-time analysis in complex matrixes, operating fully automatically or manually, so the user does not require special skills. At present, most of the biosensing methods focus on the detection of single antibiotic groups, but there are also studies dealing with the simultaneous determination of different groups of antibiotics.
The present review gives an overview of the developments and construction issues of antibiotic biosensors, applicable for the analysis of bovine milk quality, during the last 15 years. Due to the large number of different technologies used, these biosensors are classified into five separate groups according to the bio-recognition employed. The basic parameters of performance of different developments are collected into concise tables.

Receptor and enzyme-based biosensors
In these biosensors, specific receptors or enzymes are utilized to generate a bio-recognition reaction, whose signal is then detected with a suitable transducer. Receptor/enzyme-based biosensors usually employ optical or electrochemical signal detection principles [7,15]. For optical detection, surface plasmon resonance (SPR) has been used most commonly. The application of SPR technology secures low detection limits, even below the established MRL values. The main drawbacks of SPR biosensors are their high cost; nonspecific binding of compounds of sample matrix to the sensor surface; and assay time (including chip preparation, incubation of receptors, detection, and system regeneration), which could take even a couple of days [17][18][19][20][21][22][23].
A receptor-based SPR biosensor, where a conjugate of cephalosporin C and a H1 monoclonal antibody (mAb) was attached to the sensor surface before injecting the milk sample mixed with DD-carboxypeptidase (EC 3.4.16.4), was developed for β-lactam antibiotics (β-Ls) by Gustavsson et al. [18]. In the presence of β-L residues in the sample, this receptor (DDcarboxypeptidase) did not bind to the complex on the sensor surface, making the sensor response inversely proportional to the β-L concentration in the sample. The limit of detection for the studied β-Ls was below or near the corresponding MRL. The same workgroup also developed an alternative SPR-based biosensor assay for β-Ls, using the catalytic properties of carboxypeptidase from Streptomyces R39 [19,20,23]. The enzyme catalyzes the hydrolysis of a tri-peptide (acetyl-L-Lys-D-Ala-D-Ala) into a di-peptide (acetyl-L-Lys-D-Ala). In the presence of β-Ls, the enzymatic activity is inhibited and less di-peptide will be formed. The assay described measured the amount of remaining enzymatic substrate (tri-peptide). The detection limits of this biosensor system for most of the studied β-Ls were below or equal to the established MRL values, except for cloxacillin (CLOX) and ceftiofur (CEFT).
A frequently used receptor for the detection of β-L residues is penicillin-binding protein (PBP), which covalently binds to penicillin G (PEN) and other β-Ls. High molecular mass PBPs exhibit DD-carboxypeptidase activity and catalyze the final steps of peptidoglycan cross-linking [17]. PBPs are used as binding reagents in receptor and enzyme-based assays, specific to β-Ls, and the biosensor system measures the inhibition of the enzymatic activity of DD-carboxypeptidase. A SPR-based biosensor to measure the inhibition of the binding of digoxigenin-labelled ampicillin (DIG-AMP) to a soluble PBP 2x* from Streptococcus pneumonia was proposed by Cacciatore et al. [17]. The nonspecific binding was minimized and kept on a constant level by applying heat-treatment and centrifugation steps and the addition of carboxymethylated dextran. Sample pre-preparation enabled to detect some of the studied β-Ls at MRL levels. The assay, however, does not distinguish between different β-Ls and is intended to be used as a screening method prior to identification and quantification of the individual analytes by other methods, such as by HPLC/MS [17]. To detect β-Ls, Lamar and Petz [21] immobilized PBP 2x* from S. pneumonia to a micro-plate. DIG-AMP complex was added to the samples. The amount of DIG-AMP bound via its AMP part to the PBP was decreasing along with the increase of β-L concentrations in the sample. In the detection step, anti-DIG Fab fragments marked with horseradish peroxidase (HRP) were added. The more DIG-AMP complex was bound to the receptor protein, the more antibody fragments were bound via the DIG part of the complex. A maximum color development with a chromogen agent was achieved, when no β-L residues were present and the amount of bound Fab fragments with peroxidase was highest. With this system, it was possible to detect the studied β-L residues at levels corresponding to 50% of their respective MRL values in milk. To eliminate the matrix interferences and increase sensitivity, it was necessary to remove fat from the milk samples. Setford et al. [22] integrated PBP with screen-printed electrodes to measure PEN. Although the system was simple to perform and easy to use, it was only possible to distinguish between PEN concentrations below or above 1.3 MRL levels.
A group of enzyme-based biosensors for the detection of β-Ls are called penicillinase (EC 3.5.2.6, PCNase) biosensors. PCNase is produced by bacteria providing resistance toward β-L antibiotics [24]. All β-L antibiotics have a common element in their molecular structure: a fouratom ring known as a β-Lactam ring. PCNase catalyzes the opening of the β-Lactam ring turning PEN into penicilloic acid and so deactivating the molecule's antibacterial properties [24]. The measurement of the hydrolysis of β-L ring by detecting the pH change has been used in biosensors for β-L antibiotics [25][26][27][28]. Chen et al. [25] proposed a PEN sensor, where hematein (pH indicator) was co-immobilized with multi-walled carbon nanotubes (MWCNTs) and PCNase onto glassy carbon electrode. MWCNTs were used to enhance electron transfer.
In case PEN was present in the sample, the pH value decreased due to PEN hydrolysis into penicilloic acid catalyzed by PCNase. Once [H] + was accepted, hematein as a pH-sensitive redox probe was reduced to hematoxylin, what induced the increase of electrochemical signal. Severe interferences from the matrix of raw milk, caused by the adsorption of milk proteins and fat onto the electrode surface, probably forming a barrier between the enzyme and analyte molecules, were observed. To eliminate these effects, proteins and fat were separated from milk samples using salting-out and centrifugation processes. However, the interferences from the milk matrix were still observable and the detection limit remained quite high-9 mg/l. Hence, this sensor is not applicable for the detection of PEN in milk but only in water samples. Wu et al. [26] reported a similar PEN biosensor using single-graphene nanosheets (SGCs) instead of MWCNTs. Hematein was attached directly to graphene by adsorption, then ionic liquid was added due to its good biocompatibility, favoring further immobilization of PCNase. They found that at higher concentrations of PEN, the PCNase activity was lowered due to accumulation of the acidic products, decreasing the system sensitivity. The PEN detection limit was declared to be at 0.04 pg/l. The main drawback of pH-dependent biosensors is the fact that they exhibit only a limited range of applicability. For example, in complex fermentation media these biosensors are useless, as there can be dramatic pH changes caused by the fermentation itself.
Concalves et al. [29] proposed a PEN sensor using a cysteine-based self-assembled monolayer to immobilize PCNase onto a gold electrode. Ferrocene was added to establish the effectiveness of the biosensor development, although cysteine and PCNase inhibit the electron transfer to ferrocene. The PCNase used was a metallo-β-lactamase. The reaction could be monitored cronoamperometrically without any redox pH probe (e.g., hematein). The obtained detection limit was 1.5 µg/l. Ismail and Adeloju [30] developed a potentiometric biosensor for PEN by exploiting the nonconducting polytyramine-PCNase film, which was placed on Pt electrode. In potentiometric setup, the main nonspecific factors influencing the detection sensitivity are buffer concentration and pH. As PCNase was used in this sensor system, pH was fixed at 7.0 to ensure the optimal enzymatic activity. The detection limit of the sensor in buffer solutions was 0.3 µM (100 µg/l). In milk samples, the PEN recovery was not reproducible, being 110±80% at 1 mg/l, 92±32% at 5 mg/l, and 78±16% at 10 mg/l PEN. The same workgroup [27] proposed also a bilayer potentiometric biosensor for PEN. The benefit of using a bilayer configuration, where the enzyme is immobilized in both layers, is a considerable enhancement of the sensitivity through the efficiency of electron transfer between the enzyme and the electrode. In the bilayer configuration, polypyrrole film was used for PCNase immobilization. The PEN detection limit was not improved in the buffer solutions, being 0.3 µM. In milk, the detection limit was found to be 5 mg/l; unfortunately, at higher PEN concentrations the recoveries in milk were very low (30-60%).
A biosensor array, based on the analyses of the patterns of change of oxidation kinetics of lactose and its metabolites, has been proposed for a rapid multiplex detection of the most common veterinary antibiotics in raw milk [31]. In this biosensor array, different oxidoreductases were used to catalyze the oxidation of lactose and its hydrolysis products galactose and glucose in separated sample flow channels. The combination of different reaction parameters of different biosensors forms a pattern of milk sample and in the presence of antibiotics this combination forms the fingerprints of particular antibiotics.

Microbial biosensors
There are a few biosensors for detecting antibiotic residues in milk based on the application of enzymatic activity of microorganisms [35][36][37]. Systems for the monitoring of β-Ls are based on similar principles as microbiological inhibition tests [8,10], with the difference that the biorecognition reaction signal is detected quantitatively or semi-quantitatively. The microbial biosensors are based on the measurement of the inhibition of bacterial growth due to the presence of antibiotics [7][8][9][10].
Ferrini et al. [35] presented a hybrid biosensor combining classical microbiological screening of antibacterials with electrochemical detection and reading. In this system, Bacillus stearothermophilus var. calidolactis was used as a test microorganism and its growth was followed electrochemically measuring the quantity of CO 2 produced. The presence of microbial inhibitors (e.g., antibiotics) in the milk sample prohibits the growth of test strain and thus decreases the CO 2 production rate. This variation in CO 2 production was recorded during the initial 120 min in comparison to a control milk sample. The detection limits were at MRL levels.
A Bacillus cereus 66 assay based on its β-lactamatic activity and using iodine as reaction indicator has been proposed by Das et al. [36]. The system was examined for different β-Ls and other antibiotics to study its selectivity. In case antibiotics were not present, cultures did not show any color change in the test ampoules meaning that the basic enzyme production by microorganisms was not sufficient to reduce the starch iodine mixture. In the presence of antibiotics, a color change was observed within 15-25 min. The growth of B. cereus 66 was inhibited by β-Ls at ≥ 100 mg/l, which is much higher than the allowed MRL values for any of the studied β-Ls. Other antibiotics studied showed inhibition at very high concentrations ranging from 2.5 to 1000 mg/l, indicating the low sensitivity of the system.
For the detection of quinolones (Qs) and tetracyclines (TCs), an electrochemical microbial biosensor was proposed by Pellegrini et al. [37]. The detection was based on the measurement of CO 2 production rate in relation to the inhibition of microbial growth by antibiotics. The microorganism used in this study was Escherichia coli (ATCC 11303). E. coli was chosen for its good sensitivity to Qs and TCs. The inhibition degrees were evaluated after 120 min. Qs and TCs residues were detectable at 25 µg/l. The biosensor was not sensitive toward other studied antibiotics (macrolides, β-Ls, aminoglycosides, and sulfonamides).

Immunosensors
The largest group of biosensors, used for the detection of antibiotic residues in milk, is based on the exploitation of immunochemical biorecognition reactions. The most frequently applied immunosensors are the electrochemical and optical ones, the latter most often being an SPR biosensor. Although immunosensors are very selective, the speed of analysis depends on the incubation time required to form antigen/antibody complex. In addition, the full regeneration of the sensor can also be quite time consuming.
For the detection of β-L residues in milk, a SPR biosensor based on a commercial anti-ampicillin (AMP) antibody, which had much higher affinity toward open β-lactam rings than the closed ones, was constructed by Gaudin et al. [38].
In order to open the β-lactam rings and increase the assay sensitivity, two different methods for sample pre-treatment-enzymatic (with the help of PCNase) and chemical (basic hydrolysis)-were tested. The application of pretreatment enabled to achieve AMP detection limits at 33 µg/l and 12.5 µg/l after enzymatic and chemical pre-treatment, respectively. Another SPR immunosensor for the analysis of AMP was described by Zhang et al. [39]. This biosensor was a competitive binding assay between AMP covalently immobilized on the sensor surface and AMP-containing sample, mixed with monoclonal anti-AMP antibody. The sensor detected the amount of free antibody bound to the sensor surface after the injection of milk sample. The limit of detection of AMP with this sensor was 2.5 µg/l.
A portable SPR immunosensor for the determination of fluoroquinolone (FQ) antibiotics (enrofloxacin (ENRO), ciprofloxacin (CIP), and norfloxacin (NOR)) in milk was proposed by Fernandez et al. [40]. The assay worked in indirect inhibition format based on binding of the polyclonal anti-FQ-haptenized protein (FQ-BSA) antibody to the SPR sensing surface, activated with FQ-BSA, while the presence of FQs in sample inhibited the binding. The limit of detection was 2.0 µg/l. An earlier version of this portable SPR biosensor was constructed for the simultaneous detection of three different antibiotic classes (FQs, sulfonamides, and phenicols) [41]. This sensor was based on a similar competitive assay format. The limits of detection were 1.7 µg/l for ENRO, 2.1 µg/l for sulfapyridine (SPY), and 1.1 µg/l for chloramphenicol (CAP). Rebe Raz et al. [42] developed a microarray biosensor, based on an imaging SPR platform for the simultaneous detection of aminoglycosides in milk. The detection of antibiotics was carried out by combining seven different specific immunoassays on one sensor chip and was based on the competitive inhibition of antibody binding. The immunosensor showed ppb (µg/l)-level sensitivity toward the target compounds if 10 times diluted milk samples were used. A SPR immunosensor for CAP residues in milk, designed also as a binding inhibition assay, was proposed by Ferguson et al. [43]. The detection limit of CAP in milk with this assay was quite low-0.05 µg/l. For the screening of streptomycin (STR) residues in milk, Haasnoot et al. [44] tested both direct and competitive binding SPR immunoassays, based on monoclonal anti-dihydro STR antibodies. The limit of detection for STR was 20 µg/l, both for the direct and competitive binding assays. One more competitive STR immunosensor, exploiting commercial Qflex TM antibodies, was reported by Ferguson et al. [45]. This assay enabled the analysis of STR in whole bovine milk (fat content 3.5%) at concentration level 30 µg/l.
A parallel affinity immunosensor array (PASA) for the analysis of 10 different antibiotics in milk (see Table 5), using multi-analyte immunoassays with an indirect competitive ELISA format, was presented by Knecht et al. [46]. Hapten conjugated with different antibiotics was attached to modified microscope glass slides to prepare disposable microarrays. Specific monoclonal antibodies against each antibiotic allowed the simultaneous detection of each individual analyte. Antibody binding was detected by a secondary antibody, labelled with horseradish peroxidase (HRP) generating enhanced chemiluminescence. The detection limits ranged from 0.12 µg/l up to 32 µg/l. Kloth et al. [47] proposed an improved PASA system, enabling the multiplexed analysis of even 13 antibiotic residues in milk (see Table 5). In these re-generable microarray chips, hapten-antibiotic conjugates were coupled onto epoxyactivated polyethylene glycol (PEG) chip surfaces. The simultaneous detection of the 13 antibiotics in raw milk samples close to the corresponding MRL values was possible within 6 min.
A disposable amperometric magneto-immunosensor, using a polyclonal sheep anti-tetracycline (TC) antibody immobilized on the surface of protein G-functionalized magnetic beads (ProtG-MBs) and screen-printed carbon electrodes (SPCEs) for the detection of TCs in milk, was described by Conzuelo et al. [48]. TC detection was performed through competitive binding between TC in the sample and a HRP-labelled specific tracer (TC-HRP) for binding sites of the capture antibodies. The detection limits were 8.9 µg/l for TC, 1.2 µg/l for oxytetracycline (OTC), 66.8 µg/l for chlortetracycline, and 0.7 µg/l for doxycycline. Conzuelo et al. [49] proposed a similar immunosensor for the specific detection and quantification of sulfonamide (SAs) residues in milk. They used polyclonal rabbit antibodies immobilized onto the electrode surface, modified with 4-aminobenzoic acid. The limit of detection was 0.15 µg/l. Another direct competitive immunoassay for the determination of SPY was based on antibody immobilized onto the surface of protein G-modified glassy carbon plates [50]. The limit of detection of this immunosensor for SPY was on a similar level-0.13 µg/l.
A wavelength interrogated optical sensor (WIOS) technology has been employed for the development of biosensors for SAs, FQs, β-Ls, and TCs. A competitive immunosensor for the simultaneous detection of three antibiotics-SPY, CIP, and OTC-in raw milk has been described by Suarez et al. [51]. These three assays were performed in indirect formats with three specific haptens. Raw milk samples were mixed with different antiserum receptors that specifically reacted with particular antibiotics and further spiked with three antibiotics at corresponding MRLs (100 µg/ml). In contact with the sensing surface, excess antibodies not bound to antibiotics were attached to the hapten-coated sensing region. The attached antibodies were revealed with a secondary antibody. Adrian et al. developed WIOS immunosensors for the detection of SAs [52] and for the simultaneous screening of SAs and other most frequently occurring antibiotic classes: FQs, β-Ls, and TCs [53]. These sensors relied on a competitive immunoassay format, where haptenized proteins for SAs, FQs, β-Ls, and TCs were immobilized on the chip surface, where they formed independent sensing zones. Milk samples were mixed with the specific antibodies and bio-receptors were added. The detection limits of these sensors were 0.5 µg/l for SPY, 1.3 µg/l for CIP, 3.1 µg/l for AMP, and 34.2 µg/l for OTC.
An impedimetric immunosensor for CIP that comprised polyclonal anti-CIP antibody immobilized on an electrogenerated N-hydroxysuccinimide-functionalized polypyrrole film was developed by Ionescu et al. [54]. The antibody-antigen affinity reaction resulted in an extremely sensitive and specific impedance response, even with CIP concentrations as low as 10 pg/ml. A flow injection impedimetric immunosensor for the direct detection of PEN in milk samples, based on immobilized monoclonal anti-PEN on self-assembled thioctic acid monolayer on gold electrode, was developed by Thavarungkul et al. [55]. Binding of PEN to anti-PEN on the electrode surface causes the impedance to increase. The limit of detection of this immunosensor was 1 pg/l, much lower than the corresponding MRL in milk, but the sensor preparation took as long as 2 days.
A competitive amperometric immunoassay for PEN in milk was developed by Merola et al. [56]. This immunosensor was based on the competitive binding of free PEN and BSA-PEN conjugate immobilized on the sensor membrane, to anti-PEN-biotin-avidin-peroxidase complex. The limit of detection of this immunosensor was as low as 5 ng/l. Another amperometric immunosensor for PEN in milk has been described by Wu et al. [57]. This biosensor was based on the interaction of PEN with the covalently bound new methylene blue (NMB) and HRP-labelled PEN polyclonal antibody (HRP-PEN-Ab) on a glassy carbon electrode. Cyclic voltammetry and impedance spectroscopy enabled to gain the detection limit 0.6 µg/l. An electrochemical magneto immunosensor for CIP has been proposed by Pinacho et al. [58]. Magnetic beads for the attachment of CIP, which are modified with antibody (Ab171) and HRP-BSA, are collected after incubation with the sample onto magnetic electrode and detected electrochemically following the H 2 O 2 oxidation, catalyzed by HRP. The detection limit for CIP 9 ng/l was almost as low as gained in Ref. [56].
A nanogold resonance-scattering (RS) spectral assay for the determination of PEN was developed by Jiang et al. [59]. The binding of PEN to anti-PEN, immobilized on the surface of gold nanoparticles, leads to the cleavage of nanoparticles. Uncovered nanoparticles aggregated in the presence of PEN and generated the resonance scattering (RS) effect, measured at 560 nm. The detection limit of the assay was 0.78 µg/l. Another immunosensor, based on the application of the aggregation of nanoparticles, was developed by Chen et al. [60] for the detection of kanamycin (KAN) in milk. This sensor, using superparamagnetic iron oxide (SPIO) nanoparticles, acted as a magnetic relaxation switch. The target analyte KAN competed with KAN immobilized on the surface of the SPIO nanoparticles and hence affected the formation of SPIO aggregates. The dispersed and aggregated states of the SPIO modulate the spin-spin relaxation time (T 2 ) of the neighboring water molecules, which change due to the effect of the target analyte. The limit of detection with this biosensor for KAN was 0.1 µg/l.
Piezoelectric immunosensors for PEN, AMP, and the total content of penicillin antibiotics have been developed by Karaseva et al. [61]. The receptor coating of the sensors was prepared by the immobilization of PEN-or AMP-hapten-protein conjugates on the polypyrrole film via glutaraldehyde. The limits of detection obtained were 0.8 µg/l for PEN, 3.9 µg/l for AMP, and 1.7 µg/l for total penicillins.
Finally, a surface acoustic wave (SAW) biosensor has been introduced by Gruhl et al. [62] for the rapid detection of PEN in milk. This was a binding inhibition assay, using the interaction of PEN in the sample and PEN epitopes immobilized on the sensor surface with monoclonal anti-PEN. The binding of antibodies onto the sensor surface was followed by means of acoustic (gravimetric) detection. Low PEN concentrations led to a high binding load of free antibodies to the surface and hence to high signals. The detection limit for PEN in low-fat milk was 2.2 µg/l.

Aptasensors
For the detection of antibiotic residues in milk, aptamer-based biosensors (aptasensors) have been developed during last 5 years [63][64][65][66][67][68][69][70][71][72][73][74][75][76][77]. Aptamers can be considered as chemical or "synthetic" antibodies because of their in vitro production based on the systematic evolution of ligands by exponential enrichment (SELEX) [78]. The SELEX process enables the fabrication of aptamers also for non-immunogenic and toxic targets that cannot be produced by natural immune systems [79,80]. The selection provides specific aptamers fold into well-defined threedimensional shapes, which can recognize their target molecules with high affinity [78,81]. In most studied biosensing systems, the dissociation constants of the aptamer-target molecule complexes are in the nanomolar range. In addition, aptamers are quite stable and are not affected by reasonable temperature or pH shifts; at optimal conditions they can restore their original conformation. Aptamers are smaller in size compared to antibodies enabling them to reach previously blocked or intracellular targets [80].
The immobilization of aptamers is carried out with the help of two major techniques [82]: (i) direct attachment of aptamers on a bio-coated sensor surface via suitable linkers; and (ii) noncovalent conjugation of aptamers to functionally activated surfaces. To facilitate direct immobilization, the aptamer must be functionalized by adding a terminal functional group such as biotin or amine [82]. Sometimes an extra linker is used to create flexibility between the aptamer and terminal functional group. To minimize the nonspecific binding caused by the linker, the linker usually consists of a string of thymidine [82,83]. A serious disadvantage using aptasensors for milk analyses is the presence of milk proteins and fat, and the non-transparency of the samples, which hamper the application of optical detection methods [64,68,71]. Commonly pretreatment of milk samples is required [64, 68-71, 73, 74, 77].
For the detection of β-L antibiotics, several aptasensors have been proposed [64,67,69,71]. Dapra et al. [67] developed polymer biosensor chips comprising a micro-fluidic system and immobilized aptamers, integrated with the measurement of electrochemical impedance. Polymers exhibit excellent properties to master the task of transducing a binding event between an analyte and a biological probe into a measurable signal. Polymers have been used in biosensing as alternatives to traditional electrode materials due to their inexpensive fabrication and simple functionalization [84]. In aptasensors proposed by Dapra et al., singlestranded DNA (ssDNA) aptamers functionalized with fluorescein amidite (FAM) were used [67]. With this sensor it was possible to detect ampicillin (AMP) concentrations below the established MRL (the detection range was 35 ng/l to 350 µg/l). Song et al. [69] proposed an aptasensor for sulfadimethoxine (SDMX), which consisted of FAM-modified ssDNA aptamers attached to coordinated polymer nanobelts (CPNBs). The detection limit for this aptasensor for SDMX in milk was 10 µg/l, which is 10 times below the allowed MRL in EU. However, the aptamer-antibiotic interaction was interfered by other components of raw milk (proteins, fat, [K] + , [Na] + ), reported also by other authors [63,70,71,85]. It has been proposed that changes of [K] + and [Na] + inactivate aptamers by preventing them to fold into correct structures, and accordingly at least 10-fold diluted matrix should be used for milk analyses [63,85].
There are numerous aptasensors for the detection of chloramphenicol (CAP) [65,68,70], whose use is actually banned in food-producing animals and whose presence in milk is strictly prohibited. The regulations set the MRL for CAP in milk at 0.3 µg/l, which is the LOD with HPLC-MS [6]. Wu et al. [68] proposed a fluorescence-based aptasensor, using aptamerconjugated magnetic nanoparticles (MNPs) for both CAP recognition and concentration. In the absence of target molecules (antibiotics), MNP-aptamer complex hybridizes to its complementary DNA (cDNA) modified with upconversion nanoparticles (UCNPs) to form the duplex structure giving a maximum fluorescent signal. Upon CAP addition, the aptamer preferentially binds with CAP and causes the dissociation of some cDNA, liberating some UCNPs-cDNA complexes, and leads to a decreased fluorescence signal on the surface of MNPs. Under optimal conditions, a linear CAP detection range from 0.01 to 1 µg/l was achieved. Alibolandi et al. [70] constructed an aptasensor for the detection of CAP, where aptamers were conjugated with Cd-Te quantum dots (QDs), which exhibit high resistance to photo-bleaching, stable fluorescence, high quantum yield, narrow and symmetric emission band, and broad adsorption spectra [86]. The detection limit of CAP with this aptasensor was 0.2 µg/l [70].
Various aptasensors for the detection of tetracyclines (TCs) in milk have been proposed [73][74][75][76][77]85]. Jeong and Paeng [85] used two different enzyme-linked aptasensors (ELAA) based either on a ssDNA-aptamer or on an RNA-aptamer. A competitive assay with sequential mode detection was adapted for both systems. A similar system was developed by Kim et al. [73], who used ssDNA aptamers for the detection of oxytetracycline (OTC). In terms of specificity, detection limit, and dynamic range, the results obtained with ELAA in both the abovementioned studies were not an improvement in comparison with the commonly used immunoassays (ELISA) [73]. For the detection of TC, colorimetric aptasensors have been proposed by He et al. [74,75]. In these systems, aptamers were adsorbed onto the gold nanoparticles (AuNPs) by electrostatic interactions. To generate and stabilize the AuNPs, poly(diallyldimethylammonium) (PDDA) [75] and hexadecyltrimethylammonium bromide (CTAB) [74] were used. The detection of TCs was based on the aggregation phenomenon of functionalized AuNPs, which can be detected by the change in color of the solution from blue to wine red. According to the information presented, if AuNPs were stabilized with PDDA, nanoparticles aggregated in the presence of TC [75]. If CTAB was used, the aggregation of nanoparticles was not favored in the presence of TC and an opposite effect was detected [74]. These assays also required pretreatment of milk samples with acetic acid in order to remove Ca 2+ . In both systems, the average recoveries of TC were in the range 81-112%. Compared to the PDDA assay, the CTAB assay had 2.7 times higher detection limit (20 µg/l and 54 µg/l, respectively). A label-free electrochemical method based on modified glassy carbon (GC) electrode and ssDNA-aptamers has been proposed by Zhang et al., also for the detection of TCs [76]. The advantage of this electrochemical aptasensor in comparison with the optical ones is the fact that the proposed system did not need any sample pretreatment. The linear detection range for TC was 0.1-100 µg/l. An ultrasensitive resonance scattering method for the detection of TC in milk was described by Luo et al. [77]. This biosensor was based on the competition of aptamers between nanogold and TC. SsDNA-aptamer coats the surface of nanogold particles through van der Waals and intermolecular forces and prevents the aggregation of nanogold particles. But in the presence of TC, the nanogold surface becomes naked, because of the high affinity between TC and aptamer, and the nanoparticles aggregate. The constructed biosensor could be used to detect trace levels of TC in milk samples with good sensitivity (the limit of detection was 22 µg/l).
The overview of aptasensors used to determine antibiotic residues in milk is given in Table 6. Assay time does not include aptamer and electrode preparation. Table 6. Aptasensors for the detection of antibiotic residues in milk.

Molecularly Imprinted Polymer (MIP) sensors
A recent development in the biosensing of antibiotic residues in milk is the application of molecularly imprinted polymer sensors [87][88][89][90][91][92][93]. Molecular imprinting is a technique for the creation of synthetic materials containing specific receptor sites having high affinity toward the target molecule. Molecularly imprinted polymers (MIPs) are cross-linked organic structures containing pre-designed molecular recognition sites complementary in shape, size, and functional groups to the template molecule [94]. MIPs are effective alternatives for the natural receptors in biosensor assays.
For the detection of TCs, a photonic MIP sensor was developed by combining colloid crystal templating and molecular imprinting techniques [93]. Three structurally similar TCs (tetracycline, oxytetracycline, and chlortetracycline) were used as template molecules for MIPs synthesis. Target antibiotic molecules generated optical changes in the Bragg diffraction peak, detected directly using fiber optic spectroscopy. No labelling or sample pretreatment was necessary. The MIPs prepared enabled to discriminate between the very similar TCs, indicating that the cooperative effect of shape, size, and interaction sites of the formed binding areas plays a critical role in the selective molecular recognition process of MIPs. Hydrogen bonding is suggested to be the main interaction responsible for the retention of TCs in MIPs. Because hydroxyl groups form hydrogen bonds with carboxylic groups of the polymer, the difference in the amount of OHgroups along with the molecular structure of particular TCs result detectable variations in retention [93].
A voltammetric MIP sensor for the detection of sulfadiazine (SDZ) based on carbon paste electrode modified with SDZ-MIP and measuring the potential change due to the binding of antibiotics was proposed by Sadeghi and Motaharian [92]. However, the selectivity of the MIP sensor was rather poor, as sulfonamides including two aromatic cycles generated a difference in the measured currents less than 20% in the case of SDZ and sulfapyridine, and around 30% in the case of benzenesulfonamide and 4-methyl-benzenesulfonamide. The milk matrix did not interfere with the measurements.
An electrochemical MIP sensor was fabricated for the detection of erythromycin (ERY) by decorating a gold electrode with chitosan-platinum nanoparticles/graphene-gold nanoparti-cles double-layer nanocomposite [91]. The MIPs were prepared using HAuCl 4 and 2-mercaptonicotinic acid, and ERY as a template. The selectivity of the sensor was evaluated toward four different antibiotics (ERY, kanamycin, neomycin, and spiramycin). The response of the MIP electrode to ERY was approximately 80% higher than toward the other studied compounds, revealing the good selectivity of the MIP sensor. The limit of ERY detection in milk was around 75 µg/l [91].
MIP sensors have also been developed for the detection of different aminoglycosides. For tobramycin (TOB), an MIP-based pyrrole glassy carbon electrode [89] and a quartz crystal microbalance nanosensor [88] have been proposed. These MIP sensors were very sensitive, with the detection limits in milk under 1 µg/l. The selectivity and stability of these sensors were also excellent. Liu et al. [87] proposed a method for the detection of streptomycin (STR) residues in milk, based on nanogold-encapsulated poly(o-phenylenediamine) shell on magnetic iron oxide core. The assay was carried out in a competitive-type mode between the target molecule and glucose oxidase-labelled STR. The sensor response was based on the catalytic oxidation of glucose by glucose oxidase, amplifying the signal and improving the sensitivity. The application of magnetic beads facilitates the construction of the sensors and enables the concentration of samples. In milk, the detection limit for STR of this system was 1 µg/l [87]. For the detection of neomycin (NEO), a multilayer sensor system of gold electrode, composite material (chitosan-silver nanoparticles/graphene-multiwalled carbon nanotubes), and poly(pyrrole)-based MIP were used [90]. The selectivity of this sensor was good toward NEO, and in milk the detection limit was 6 µg/l.
The MIP sensors, which have been developed, are very sensitive with low detection limits. However, the linear ranges of detection of these sensors tend to be below the allowed MRL values, so complicating their practical applications. The condensed overview of MIP sensors used to determine antibiotic residues in milk is given in Table 7. n.a. 0.05-20 µg/l [93] TOB-tobramycin; NEO-neomycin; SDZ-sulfadiazine; STR-streptomycin; ERY-erythromycin; TC-tetracycline; OTC-oxytetracycline; CTC-chlortetracycline; LOD-limit of detection; n.a.-data not available; MIP-molecularly imprinted polymers Table 7. Molecularly imprinted polymer sensors for the detection of antibiotic residues in milk.

Conclusions
Due to the urgent need for on-site analysis of milk quality and safety, the research activity for the development of biosensors for the detection of antibiotic residues in milk has been very high during the last years. Most studies are focusing on the detection of antibiotics, which are most commonly used for the treatment of food-producing animals: β-lactams, tetracyclines, sulphonamides, and aminoglycosides. Biosensors proposed for these antibiotics often exhibit detection limits below or equal to the allowed maximum residue levels in milk. There are almost a hundred studies within the last decade reporting about biosensors for the detection of different antibiotic residues in milk; unfortunately, we could not find any studies dealing with biosensing of colistin or trimethoprim.
In most studies, milk samples spiked with a selected antibiotic or antibiotics have been used for the "proof of concept" and validation of the proposed technology. Analyses of the milk of animals, who are undergoing antibiotic treatment, are scarce, although these real samples can be a key factor to indicate the applicability of biosensor technology for practical analyses, as these "natural" samples can contain in addition different metabolites of antibiotic compounds. Some complex biosensing systems require additional pretreatment of milk samples to remove fat and proteins-the implication of these technologies for on-site analyses also seems problematic. The average detection time (excluding pretreatment) is 30-40 min or even up to 2 hours in case a longer incubation period for bio-recognition is required.
The biggest problem in biosensing is the stability of compounds, used for bio-recognition and the life-time of sensors. In terms of stability, the most prospective are MIP biosensors, where the biological material is used only as a template, and aptasensors. Concerning the regeneration of biosensor systems, there is very little data available. Usually, the full dissociation of complexes, formed during the bio-recognition, takes 30-40 min. In terms of rapid analyses, single use biosensors have a great advantage.
In analytical developments, reliability and robustness are the keywords of future trends. We hope that along with automation and independent action, these features enable biosensors to become an essential tool for on-site analysis of milk and other substances.