**3. Development of a highly sensitive LFIA for measuring AFM1 in milk**

With the aim of producingasystem sensitive enough to reach the limits imposed by Europe‐ an regulations, we developed a competitive lateral flow immunoassay which exploited rab‐ bit polyclonal antibodies towards AFM1that had been previously employed in the development of a sensitive ELISA [19]. A classic device, including a NC membrane (onto which the two lines of reagents had been immobilized), cellulose sample and adsorbent pads, and a glass fibre release pad (on which GNP-labelled antibodies are pre-adsorbed) was conceived.

**3.1. Materials and methods**

and stored at 4°C until use.

*3.1.2. Lateral flow immunoassay procedure*

lateral flow assay.

distributed near the lower edge of the pads and left to dry.

Gold colloids with mean diameter of about 40 nm were prepared using the sodium citrate method as previously described [46]. The saturation concentration of the anti-AFM1 anti‐ body for conjugation with gold nanoparticles was determined according to Horisbergand Rosset [69]. GNP-antibody conjugation was carried out using an amount of antibodies which is the half the saturation concentration and was carried out as follows: 100 µl of a 0.5 mg ml-1 anti-AFM1antibodies in borate buffer was added to 10 mL of pH-adjusted colloidal gold solution. After 30' incubation at room temperature, 1 ml of borate buffer containing 1% of BSA was added. The mixture was centrifuged and the pellet was washed twice by re-sus‐ pension in borate buffer with 0.1% BSA added. Finally, the pellet was re-suspended in bo‐ rate buffer supplied with 1% BSA, 0.25% Tween 20, 2% sucrose, and 0.02% sodium azide

Lateral Flow Immunoassays for Aflatoxins B and G and for Aflatoxin M1

http://dx.doi.org/10.5772/51777

325

Release pads were previously treated with borate buffer supplied with 1% BSA, 0.25% Tween 20, 2% sucrose, and 0.02% sodium azide. After drying, gold-labelled antibodies were

Test and Control lines were spotted upon a NC membrane as follows: the AFM1-protein conjugate (SR 4) at 0.3 mg/ml was the capture reagent, and the goat anti-rabbit IgG antibod‐ ies (2 mg/ml) formed the C-line. Then, the membrane was dried. Strips were composed as follows: from the top; the adsorbent pad, the NC membrane, the release pad and the sample pad were pasted, in sequence, with 1-2 mm overlap. Release pad was positioned so that the line of GNP-labelled antibodies was on the opposite site from the edge of the membrane. The prepared membrane was cut into strips of 5 mm, which were inserted into rigid plastic cassettes.

Cassettes were stored in plastic bags containing silica at room temperature until use.

Pasteurized milk samples were purchased in large stores, and raw milk samples were ob‐ tained from farms. Whole and semi-skimmed milk (1 ml) were centrifuged for 2 min at 6000 rpm. The upper fat layer was discharged, 500 µl of the underlying serum was transferred into a tube and 25 µl of 10% Tween 20 was added. The mixture was immediately used in the

The test was carried out by placing 100 µl of sample into the sample well. After 15 minutes of incubation at 37°C, the cassette was placed above a mobile scanner connected to a laptop. The Skannex 3.0 software (SkannexAS,Hoenefoss, Norway) was used to acquire and process images. Calibration curves were obtained by plotting the ratio between the intensity of the test (T) and the control line (C) [46] against the log of AFM1 concentration. For each experi‐ ment, a calibration curve was determined by a nonlinear regression analysis of the data us‐ ing the four-parameter logistic equation [70]. For the construction of the standard curve and for recovery experiments blank milk samples that did not show any detectable residues of

*3.1.1. LFD preparation*

The method was designed to be a competitive LFIA, in which the Test line comprised an AFM1 conjugate (competitor) and the Control line was composed of anti-rabbit IgG antibod‐ ies. GNP-labelled anti-AFM1 antibodies were furnished as pre-adsorbed in a release pad. When re-suspended by the sample, flowed across the membrane where first encountered the T-line and bound to the immobilized AFM1 conjugate. A red colour became visible at the T-line, due to the focusing of nanoparticles. If some AFM1 was present in the sample, it com‐ peted with the immobilized AFM1-BSA for binding to the GNP-labelled antibodies, resulting in a reduction of the T-line intensity. The anti-rabbit IgG antibodies on the Control line cap‐ tured any excess GNP-labelled antibodies to produce a C-line as a visible confirmation of particle flow. Signal intensities of the two lines were read by a portable scanner connected to a laptop and processed by dedicated software, which acquires images, determines colour in‐ tensity, interpolates values on a memorized standard curve and returns the concentration of the analyte in the sample.

Since the methodin development was a competitive immunoassay, its sensitivity was influ‐ enced by several well-known factors, such as antibody dilution and competitor concentra‐ tion, provided that a definite antiserum was used. Additional factors that could be considered were: the chemical structure of the hapten (actually, the use of heterologous competitors had been shown to improve sensitivity [67]), the structure of the antigen used as the competitor in the assay (as far as the nature of the carrier-protein and the degree of conjugation between the hapten and the carrier-protein itself were considered); the specific response of the reporter used to label the antibody; the extent of antibody labelling (moles of reporter per mole of antibody). In effect, the work of Byzovaet al [68] firstly reported the ef‐ fect of varying some of the described factors on LFIA performances and, in particular, showed that the diminishing of the molar substitution ratio (SR) between the hapten and the carrier-protein in the preparation of the competitor significantly improved as say sensitivity. The same authors also studied the binding capacity of different anti-species antibodies (which were used to trace the C-line) concluding, in this case, that no evident differences could be observed.

The need of developing a very high sensitive assay for determining AFM1 in milk at lev‐ els of regulatory concern according to EU regulation [2], forced us to investigate further in these directions and to question other established practices, such as the assumption that the labelling of antibodies should be conducted in such a way to obtain a complete coat‐ ing of GNP surfaces.

Therefore, the effects of varying: the competitor (use of homologous or heterologous hapten; nature of the carrier-protein and hapten-to-protein molar ratio) and the reporter (extent of antibody labelling)on sensitivity were studied and optimized.

#### **3.1. Materials and methods**

## *3.1.1. LFD preparation*

pads, and a glass fibre release pad (on which GNP-labelled antibodies are pre-adsorbed)

The method was designed to be a competitive LFIA, in which the Test line comprised an AFM1 conjugate (competitor) and the Control line was composed of anti-rabbit IgG antibod‐ ies. GNP-labelled anti-AFM1 antibodies were furnished as pre-adsorbed in a release pad. When re-suspended by the sample, flowed across the membrane where first encountered the T-line and bound to the immobilized AFM1 conjugate. A red colour became visible at the T-line, due to the focusing of nanoparticles. If some AFM1 was present in the sample, it com‐ peted with the immobilized AFM1-BSA for binding to the GNP-labelled antibodies, resulting in a reduction of the T-line intensity. The anti-rabbit IgG antibodies on the Control line cap‐ tured any excess GNP-labelled antibodies to produce a C-line as a visible confirmation of particle flow. Signal intensities of the two lines were read by a portable scanner connected to a laptop and processed by dedicated software, which acquires images, determines colour in‐ tensity, interpolates values on a memorized standard curve and returns the concentration of

Since the methodin development was a competitive immunoassay, its sensitivity was influ‐ enced by several well-known factors, such as antibody dilution and competitor concentra‐ tion, provided that a definite antiserum was used. Additional factors that could be considered were: the chemical structure of the hapten (actually, the use of heterologous competitors had been shown to improve sensitivity [67]), the structure of the antigen used as the competitor in the assay (as far as the nature of the carrier-protein and the degree of conjugation between the hapten and the carrier-protein itself were considered); the specific response of the reporter used to label the antibody; the extent of antibody labelling (moles of reporter per mole of antibody). In effect, the work of Byzovaet al [68] firstly reported the ef‐ fect of varying some of the described factors on LFIA performances and, in particular, showed that the diminishing of the molar substitution ratio (SR) between the hapten and the carrier-protein in the preparation of the competitor significantly improved as say sensitivity. The same authors also studied the binding capacity of different anti-species antibodies (which were used to trace the C-line) concluding, in this case, that no evident differences

The need of developing a very high sensitive assay for determining AFM1 in milk at lev‐ els of regulatory concern according to EU regulation [2], forced us to investigate further in these directions and to question other established practices, such as the assumption that the labelling of antibodies should be conducted in such a way to obtain a complete coat‐

Therefore, the effects of varying: the competitor (use of homologous or heterologous hapten; nature of the carrier-protein and hapten-to-protein molar ratio) and the reporter (extent of

antibody labelling)on sensitivity were studied and optimized.

was conceived.

324 Aflatoxins - Recent Advances and Future Prospects

the analyte in the sample.

could be observed.

ing of GNP surfaces.

Gold colloids with mean diameter of about 40 nm were prepared using the sodium citrate method as previously described [46]. The saturation concentration of the anti-AFM1 anti‐ body for conjugation with gold nanoparticles was determined according to Horisbergand Rosset [69]. GNP-antibody conjugation was carried out using an amount of antibodies which is the half the saturation concentration and was carried out as follows: 100 µl of a 0.5 mg ml-1 anti-AFM1antibodies in borate buffer was added to 10 mL of pH-adjusted colloidal gold solution. After 30' incubation at room temperature, 1 ml of borate buffer containing 1% of BSA was added. The mixture was centrifuged and the pellet was washed twice by re-sus‐ pension in borate buffer with 0.1% BSA added. Finally, the pellet was re-suspended in bo‐ rate buffer supplied with 1% BSA, 0.25% Tween 20, 2% sucrose, and 0.02% sodium azide and stored at 4°C until use.

Release pads were previously treated with borate buffer supplied with 1% BSA, 0.25% Tween 20, 2% sucrose, and 0.02% sodium azide. After drying, gold-labelled antibodies were distributed near the lower edge of the pads and left to dry.

Test and Control lines were spotted upon a NC membrane as follows: the AFM1-protein conjugate (SR 4) at 0.3 mg/ml was the capture reagent, and the goat anti-rabbit IgG antibod‐ ies (2 mg/ml) formed the C-line. Then, the membrane was dried. Strips were composed as follows: from the top; the adsorbent pad, the NC membrane, the release pad and the sample pad were pasted, in sequence, with 1-2 mm overlap. Release pad was positioned so that the line of GNP-labelled antibodies was on the opposite site from the edge of the membrane. The prepared membrane was cut into strips of 5 mm, which were inserted into rigid plastic cassettes. Cassettes were stored in plastic bags containing silica at room temperature until use.

#### *3.1.2. Lateral flow immunoassay procedure*

Pasteurized milk samples were purchased in large stores, and raw milk samples were ob‐ tained from farms. Whole and semi-skimmed milk (1 ml) were centrifuged for 2 min at 6000 rpm. The upper fat layer was discharged, 500 µl of the underlying serum was transferred into a tube and 25 µl of 10% Tween 20 was added. The mixture was immediately used in the lateral flow assay.

The test was carried out by placing 100 µl of sample into the sample well. After 15 minutes of incubation at 37°C, the cassette was placed above a mobile scanner connected to a laptop. The Skannex 3.0 software (SkannexAS,Hoenefoss, Norway) was used to acquire and process images. Calibration curves were obtained by plotting the ratio between the intensity of the test (T) and the control line (C) [46] against the log of AFM1 concentration. For each experi‐ ment, a calibration curve was determined by a nonlinear regression analysis of the data us‐ ing the four-parameter logistic equation [70]. For the construction of the standard curve and for recovery experiments blank milk samples that did not show any detectable residues of the target when analysed by a reference ELISA (LOD 5 ng l-1) [19] were fortified with appro‐ priate amounts of an AFM1 standard solution.

On the contrary, the substitution of the bovine serum albumin (which had been used to pre‐ pare the immunogen) with ovalbumin as the carrier-protein seemed irrelevant. In fact, anti‐ bodies binding the BSA used as the immunogenic carrier-protein are saturated in the preparation of the gold labelled- antibody. This preparation involves the GNP overcoating with exceeding amount of BSA to prevent aggregation; however, the inhibition of further

Lateral Flow Immunoassays for Aflatoxins B and G and for Aflatoxin M1

http://dx.doi.org/10.5772/51777

327

Optimization of LFIA usually involves checkerboard titrations where the amounts of anti‐ bodies and of the competitor are varied to achieve the lower limit of detection and the maximum slope of the calibration curve. Varying the amount of antibodies is exclusively intended as diluting the colloid of GNPs coated with antibodies themselves. The parame‐ ter used to measure this dilution is the optical density (OD) of the gold colloid, assuming that GNPs surface had been saturated with antibodies; a typical protocol prescribes that the saturation amount of antibodies, intended as the amount that prevent GNP aggregation, has to be determined firstly and this stabilizing amount or, more usually, a small excess of antibodies, has to be conjugated to GNPs to prepare the signal reporter. Nevertheless, con‐ trarily to this generally accepted assumption, Laycock et al reported a huge increase in sensitivity due to the reduction of antibodies coated onto GNPs in comparison to the stabi‐

Therefore, besides studying the effect of varying GNP-labelled antibody (intended as vary‐ ing the OD under saturating conditions); we considered that dilution of antibodies to favour competitive conditions would also be achieved by reducing the number of molecules of anti‐ body bound per GNP at a fixed OD value. Consequent risk of GNP aggregation, due to in‐ complete shielding of the superficial GNP charges, could be efficiently prevented by the further addition of exceeding amount of other proteins, such as for example BSA, which is particularly effective in this purpose.The variation of the amount of GNP-labelled antibod‐ ies dispensed at different ODs (3 and 6) under saturating conditions, apparently did not di‐ rectly influence the sensitivity of the LFIA (data not shown) Nevertheless, the increasing of the OD allows the development of more intense absolute signals, which in turn means that the amount of competitor could be decreased in the T-line therefore improving detectability.

To study antibody dilution intended as the reduction of antibody amount per GNP, differ‐ ent amount of antibodies were reacted with portions of the same GNP colloid as follows: saturation amount (AbSAT), excess of antibody (Ab/ AbSAT = 1.5), defect of antibody (Ab/ AbSAT = 0.7), and half the saturation amount (Ab/ AbSAT = 0.5). The four GNP-antibody prep‐ arations were dispensed onto release pads at OD 3 and applied to strips where the AFM1- BSA with SR of 22 and a concentration of 0.2 mg/ml had been traced upon to form the T-line. AFM1 calibrators prepared in milk were run onto these strips in triplicate. Resulting curves are show in in Figure 4. Besides a significant signal reduction, a certain improvement in sen‐ sitivity was observed when the amount of antibody was lowered from saturating conditions (IC50 = 1.71 ± 0.01) to its half (IC50 = 0.99 ± 0.01); however detectability was influenced in a

non-specific binding to BSA of antigens could also be attained.

*3.2.2. Labelling of antibodies with gold nanoparticles*

lizing amount [47].

#### **3.2. Optimization of the LFIAs**

#### *3.2.1. Effect of varying the hapten, the AFM1-protein substitution ratio and the carrier protein in the T line*

The polyclonal antiserum used in this work had showncertain cross-reactivity towards aflatoxin B1 (about 35% when measured by means of the ELISA); therefore a competitor synthesized by using a hapten derived from this toxin was considered as a "heterolo‐ gous" competitor respect to AFM1 protein conjugates (which were homologous to the im‐ munogen). Therefore, three conjugates of AFM1 with Bovine Serum Albumin (AFM1- BSA) conjugates which varied in the hapten-to-protein ratio, one conjugate of AFM1with ovalbumin (AFM1 –OVA) and one conjugate between AFB1and BSA (AFB1-BSA) were eval‐ uated as potential competitors to be immobilized in the Test line (Table 1). Each conju‐ gate was dispensed on the membrane at the same rate and volume (1µl/cm), however the concentration was varied to obtain an absolute signal of about 20-25 arbitrary units in the T-line when the strip were read by means of the software. AFM1 standard solutions (0-10-100-1000 ng l-1) prepared in a blank pasteurized whole milk were measured in tripli‐ cate and IC50 values were compared (Table 1). The AFB1 conjugate qualitatively behaved as the AFM1 conjugate with a similar SR, except for the absolute signal, which was less intense at the same concentration of dispensing. Interestingly, the decrease of the amount of AFM1 per mole of protein strongly influenced the sensitivity of the assay. Indeed, the reducing of the substitution ratio (SR) from about 22 to about 4 allowed an improvement of nearly 10-folds in the IC50 to be obtained. This result is in good agreement with the observation of Byzova and co-workers [68] and with expectations based on the experi‐ ence with competitive immunoassays in other formats (such as for example in ELISA). In parallel, the absolute signal decreased and forced to increase the amount of antigen to be dispensed. Nevertheless, the advantage of reducing the hapten density strongly predomi‐ nated over the increase of the absolute antigen concentration.


**Table 1.** Effect of varying the competitor to be used in the Test line of the LFD. Protein conjugates were dispensed onto the membrane at different concentrations to reach an absolute signal comprises between 20 and 25 arbitrary units on the T-line. SR represents the molar substitution ratio between the toxin and the protein which had been estimated by spectrophotometric measurements.

On the contrary, the substitution of the bovine serum albumin (which had been used to pre‐ pare the immunogen) with ovalbumin as the carrier-protein seemed irrelevant. In fact, anti‐ bodies binding the BSA used as the immunogenic carrier-protein are saturated in the preparation of the gold labelled- antibody. This preparation involves the GNP overcoating with exceeding amount of BSA to prevent aggregation; however, the inhibition of further non-specific binding to BSA of antigens could also be attained.

#### *3.2.2. Labelling of antibodies with gold nanoparticles*

the target when analysed by a reference ELISA (LOD 5 ng l-1) [19] were fortified with appro‐

*3.2.1. Effect of varying the hapten, the AFM1-protein substitution ratio and the carrier protein in the*

The polyclonal antiserum used in this work had showncertain cross-reactivity towards aflatoxin B1 (about 35% when measured by means of the ELISA); therefore a competitor synthesized by using a hapten derived from this toxin was considered as a "heterolo‐ gous" competitor respect to AFM1 protein conjugates (which were homologous to the im‐ munogen). Therefore, three conjugates of AFM1 with Bovine Serum Albumin (AFM1- BSA) conjugates which varied in the hapten-to-protein ratio, one conjugate of AFM1with ovalbumin (AFM1 –OVA) and one conjugate between AFB1and BSA (AFB1-BSA) were eval‐ uated as potential competitors to be immobilized in the Test line (Table 1). Each conju‐ gate was dispensed on the membrane at the same rate and volume (1µl/cm), however the concentration was varied to obtain an absolute signal of about 20-25 arbitrary units in the T-line when the strip were read by means of the software. AFM1 standard solutions (0-10-100-1000 ng l-1) prepared in a blank pasteurized whole milk were measured in tripli‐ cate and IC50 values were compared (Table 1). The AFB1 conjugate qualitatively behaved as the AFM1 conjugate with a similar SR, except for the absolute signal, which was less intense at the same concentration of dispensing. Interestingly, the decrease of the amount of AFM1 per mole of protein strongly influenced the sensitivity of the assay. Indeed, the reducing of the substitution ratio (SR) from about 22 to about 4 allowed an improvement of nearly 10-folds in the IC50 to be obtained. This result is in good agreement with the observation of Byzova and co-workers [68] and with expectations based on the experi‐ ence with competitive immunoassays in other formats (such as for example in ELISA). In parallel, the absolute signal decreased and forced to increase the amount of antigen to be dispensed. Nevertheless, the advantage of reducing the hapten density strongly predomi‐

priate amounts of an AFM1 standard solution.

nated over the increase of the absolute antigen concentration.

estimated by spectrophotometric measurements.

**Conjugate SR Dispensing concentration (mg ml-1) IC50 (µg l-1)**

**Table 1.** Effect of varying the competitor to be used in the Test line of the LFD. Protein conjugates were dispensed onto the membrane at different concentrations to reach an absolute signal comprises between 20 and 25 arbitrary units on the T-line. SR represents the molar substitution ratio between the toxin and the protein which had been

AFM1-BSA 4 0.8 0.2 AFM1-BSA 15 0.4 1.1 AFM1-BSA 22 0.2 1.7 AFM1-OVA 10 0.8 0.6 AFB1-BSA 24 0.4 1.6

**3.2. Optimization of the LFIAs**

326 Aflatoxins - Recent Advances and Future Prospects

*T line*

Optimization of LFIA usually involves checkerboard titrations where the amounts of anti‐ bodies and of the competitor are varied to achieve the lower limit of detection and the maximum slope of the calibration curve. Varying the amount of antibodies is exclusively intended as diluting the colloid of GNPs coated with antibodies themselves. The parame‐ ter used to measure this dilution is the optical density (OD) of the gold colloid, assuming that GNPs surface had been saturated with antibodies; a typical protocol prescribes that the saturation amount of antibodies, intended as the amount that prevent GNP aggregation, has to be determined firstly and this stabilizing amount or, more usually, a small excess of antibodies, has to be conjugated to GNPs to prepare the signal reporter. Nevertheless, con‐ trarily to this generally accepted assumption, Laycock et al reported a huge increase in sensitivity due to the reduction of antibodies coated onto GNPs in comparison to the stabi‐ lizing amount [47].

Therefore, besides studying the effect of varying GNP-labelled antibody (intended as vary‐ ing the OD under saturating conditions); we considered that dilution of antibodies to favour competitive conditions would also be achieved by reducing the number of molecules of anti‐ body bound per GNP at a fixed OD value. Consequent risk of GNP aggregation, due to in‐ complete shielding of the superficial GNP charges, could be efficiently prevented by the further addition of exceeding amount of other proteins, such as for example BSA, which is particularly effective in this purpose.The variation of the amount of GNP-labelled antibod‐ ies dispensed at different ODs (3 and 6) under saturating conditions, apparently did not di‐ rectly influence the sensitivity of the LFIA (data not shown) Nevertheless, the increasing of the OD allows the development of more intense absolute signals, which in turn means that the amount of competitor could be decreased in the T-line therefore improving detectability.

To study antibody dilution intended as the reduction of antibody amount per GNP, differ‐ ent amount of antibodies were reacted with portions of the same GNP colloid as follows: saturation amount (AbSAT), excess of antibody (Ab/ AbSAT = 1.5), defect of antibody (Ab/ AbSAT = 0.7), and half the saturation amount (Ab/ AbSAT = 0.5). The four GNP-antibody prep‐ arations were dispensed onto release pads at OD 3 and applied to strips where the AFM1- BSA with SR of 22 and a concentration of 0.2 mg/ml had been traced upon to form the T-line. AFM1 calibrators prepared in milk were run onto these strips in triplicate. Resulting curves are show in in Figure 4. Besides a significant signal reduction, a certain improvement in sen‐ sitivity was observed when the amount of antibody was lowered from saturating conditions (IC50 = 1.71 ± 0.01) to its half (IC50 = 0.99 ± 0.01); however detectability was influenced in a considerably lesser extent respect than when modifying the nature of the competitor (i.e.: the SR of the conjugate used to obtain the T-line), as discussed above.

LFIA for other mycotoxin [56, 61-63], matrix-matched calibration should be carried out to fit experimental results obtained on milk samples. Therefore, a pasteurized whole milk sample in which AFM1 was found out undetectable when analysed by the reference ELISA kit was used to prepare diluted calibrators. A typical calibration curve is depicted in Figure 5. A LOD (calculated as the average of the blank minus 3 standard deviations from the average)

Lateral Flow Immunoassays for Aflatoxins B and G and for Aflatoxin M1

http://dx.doi.org/10.5772/51777

329

**Figure 6.** A typical calibration curve for AFM1 measurement in milk by the developed LFIA. Error bars represent SD of 3

Accuracy of the developed LFIA was evaluated on different kind of milk samples (Table 2). Milk samples were purchased on the market and were found undetectable according to the developed LFIA. Therefore, accuracy was evaluated on samples fortified at two levels (50 and 100 ng l-1). Acceptable results were obtained, although a slight overestimation or under‐ estimation were observed for the raw and the UHT samples, respectively, which can be at‐

The intra- and inter-day precision was evaluated at 3 levels of fortification (0-50-100 ng l-1). RSD values were generally high (above 30%) which makes reliability of quantification

tributed to the fact that calibration was carried out in pasteurized milk.

replicates.

questionable.

and IC50 of 20 ng l-1 and 102 ± 19 ng l-1 were estimated, respectively.

**Figure 5.** Effect of the amount of antibodies coated onto the GNPs (Ab) compared to the amount needed for saturat‐ ing GNP surface (AbSAT)for varying Ab/ AbSAT: 0.5 (), 0.7 (▲), 1 (), 1.5 ().GNP-antibodies dispensed at OD 3; T-line: AFM1-BSA conjugate (0.2 mgml-1, SR=22).

#### **3.3. AFM1 detection in milk by the developed LFIA**

Protein and fat contents of milk may influence test results in various ways: the sample flow can be altered (for example fat content strongly affectsviscosity) and any of the milk compo‐ nents can give specific or non-specific interactions with immunoreagents involved in the as‐ say. In fact, we observed that casein determined a strong signal depression of both the Test and Control lines. With the aim of developing a unique system that could be used on milk samples undergone to different thermal treatments, i.e.: with different levels of protein de‐ naturation (raw, pasteurized, UHT milk) and with variable fat content (whole, semi-skim‐ med, skimmed milk), samples were standardized by a rapid centrifugation stepto allow the removal of the fat layer and by adding Tween 20 to control protein interferences.

After development (15' at 37°C), strips were scanned. Dedicated software acquires and proc‐ essed images and the signal, intended as the T/C ratio, was plotted against the logarithm of AFM1 concentration to carry out calibration. As previously observed in the development of LFIA for other mycotoxin [56, 61-63], matrix-matched calibration should be carried out to fit experimental results obtained on milk samples. Therefore, a pasteurized whole milk sample in which AFM1 was found out undetectable when analysed by the reference ELISA kit was used to prepare diluted calibrators. A typical calibration curve is depicted in Figure 5. A LOD (calculated as the average of the blank minus 3 standard deviations from the average) and IC50 of 20 ng l-1 and 102 ± 19 ng l-1 were estimated, respectively.

considerably lesser extent respect than when modifying the nature of the competitor (i.e.:

**Figure 5.** Effect of the amount of antibodies coated onto the GNPs (Ab) compared to the amount needed for saturat‐ ing GNP surface (AbSAT)for varying Ab/ AbSAT: 0.5 (), 0.7 (▲), 1 (), 1.5 ().GNP-antibodies dispensed at OD 3; T-line:

Protein and fat contents of milk may influence test results in various ways: the sample flow can be altered (for example fat content strongly affectsviscosity) and any of the milk compo‐ nents can give specific or non-specific interactions with immunoreagents involved in the as‐ say. In fact, we observed that casein determined a strong signal depression of both the Test and Control lines. With the aim of developing a unique system that could be used on milk samples undergone to different thermal treatments, i.e.: with different levels of protein de‐ naturation (raw, pasteurized, UHT milk) and with variable fat content (whole, semi-skim‐ med, skimmed milk), samples were standardized by a rapid centrifugation stepto allow the

After development (15' at 37°C), strips were scanned. Dedicated software acquires and proc‐ essed images and the signal, intended as the T/C ratio, was plotted against the logarithm of AFM1 concentration to carry out calibration. As previously observed in the development of

removal of the fat layer and by adding Tween 20 to control protein interferences.

AFM1-BSA conjugate (0.2 mgml-1, SR=22).

328 Aflatoxins - Recent Advances and Future Prospects

**3.3. AFM1 detection in milk by the developed LFIA**

the SR of the conjugate used to obtain the T-line), as discussed above.

**Figure 6.** A typical calibration curve for AFM1 measurement in milk by the developed LFIA. Error bars represent SD of 3 replicates.

Accuracy of the developed LFIA was evaluated on different kind of milk samples (Table 2). Milk samples were purchased on the market and were found undetectable according to the developed LFIA. Therefore, accuracy was evaluated on samples fortified at two levels (50 and 100 ng l-1). Acceptable results were obtained, although a slight overestimation or under‐ estimation were observed for the raw and the UHT samples, respectively, which can be at‐ tributed to the fact that calibration was carried out in pasteurized milk.

The intra- and inter-day precision was evaluated at 3 levels of fortification (0-50-100 ng l-1). RSD values were generally high (above 30%) which makes reliability of quantification questionable.


To achieve the useful ability to discriminate compliant from noncompliant samples, a prop‐ er cut-off value should be established. The eligible EU MRL value (i.e.: 50 ng l-1) would be expected to be at tain able given the high sensitivity of the developed LFIA. Nevertheless, the definition of a cut-off level should consider precision and technical limitations of the meth‐ od, besides sensitivity. Moreover, the calibration curve being a continuously descending curve characterized by a finite slope, the definition of a single-point cut-off value is less appropri‐ ate than the identification of an indicator range of analyte concentrations within which uncer‐

Lateral Flow Immunoassays for Aflatoxins B and G and for Aflatoxin M1

http://dx.doi.org/10.5772/51777

331

tain or "non-attributable" results (neither "compliant" nor "noncompliant") fall [44].

precision did not allow to reliably attributing them to one or another group.

sults (samples certainly attributed as compliant or noncompliant).

tion of sensitivity, selectivity and efficiency, are reported in Table 3.

ered acceptable.

As regard precision, European legislation for screening methods of analysis defines as appro‐ priate a relative uncertainty of 47% of the MRL and as acceptable even 94% for AFM1 based on the application of Horwitz equation [71]. Accepting the more restrictive criterion, this means that any screening methods should be able to discriminate between AFM1 content less than 26.5 ng l-1 (negative sample) and AFM1 content over 73.5 ng l-1 (positive sample). Samples that have AFM1 content close to the thres hold limit should thus be defined as uncertain because

In spite of this, it should be noted that a "non-attributable" judgement would determine re‐ jection of the sample with a considerable economic damage, as discussed above. Therefore, the minimum number of non-attributable results would be expected for a worth while meth‐ od and this number obviously depends on the combination of accuracy and precision of the method itself. To indicate the capability of a qualitative/semi-quantitative method to pro‐ duce the lowest score of non-attributable results, for a defined uncertainty interval, we intro‐ duced a new parameter indicated as "efficiency" of the method, defined as the ability of the method itself to detect truly non-attributable as non-attributable. Efficiency was thus calcu‐ lated as the number of truly non-attributable tests divided by the sum of known non-attrib‐ utable samples, in strict analogy with "sensitivity" and "selectivity" of qualitative and semiquantitative as says, which are defined as the rate of truly positive e and truly negative test results, respectively [50, 60]. The more efficient the assay, the highest the score of useful re‐

The ability of the developed LFIA to correctly attribute to each of the groups milk samples found on the market was thus assessed; in particular, negative (compliant) samples were de‐ fined as those in which AFM1 content was below 30 ng l-1, positive (noncompliant) samples those in which AFM1 content was above 70 ng l-1 and uncertain (non-attributable) those hav‐ ing an AFM1 content between 30 and 70 ng l-1. Since all tested samples were always contami‐ nated below 30 ng l-1 as established by the reference ELISA, positive samples were generated through fortification at 50 and 100 ng l-1. Results of this evaluation, together with the defini‐

It can be observed from data that the definition of an indicator range instead of a cut-off lev‐ el allowed us to avoid occurrence of false compliant and false noncompliant. Incorrect attri‐ bution occurred in 15% of samples (6/40), though 3 of them would represent a minor issue being assigned as non-attributable rather than noncompliant, which anyhow mean that sam‐ ples would be discarded. The efficiency is relatively low, however it could still be consid‐

**Table 2.** Recovery of AFM1 determination from artificially contaminated milk samples undergone to various thermal treatments and with different fat content as determined by the developed LFIA. Recovery was calculated as follows: (estimated AFM1 for the fortified sample – estimated AFM1 for the non fortified sample) / fortification level \*100

#### **3.4. Intra-laboratory validation of the semi-quantitative LFIA**

The objective of analytical methods such as those based on the LFIA technology is the parting between samples surely complying with legislation in force and samples which do not comply. However, a further category of samples should be considered and is represent‐ ed by those samples in which the toxin content is close to the legal limit which because of measure uncertainty cannot be classified as compliant or noncompliant (Figure 4). These "uncertain samples" should be submitted to further controls before entering the transfor‐ mation chain. In the case of milk, rejection is more often the fate of such uncertain sam‐ ples (as for noncompliant samples), because the perishable nature of milk discourages timeconsuming investigations. Therefore, the purpose of the work could become the development of a very rapid screening method which allowed the semi-quantitation of AFM1 in milk in such a way to permit the discrimination between compliant and noncompliant samples. The instrumental quantification of coloured lines and their correlation with a calibration curve, in this context, could be regarded as a way to limit subjectivity in the interpreta‐ tion of results and to improve detectability [52, 44] rather than going into the direction of factual quantitative measurements.

To achieve the useful ability to discriminate compliant from noncompliant samples, a prop‐ er cut-off value should be established. The eligible EU MRL value (i.e.: 50 ng l-1) would be expected to be at tain able given the high sensitivity of the developed LFIA. Nevertheless, the definition of a cut-off level should consider precision and technical limitations of the meth‐ od, besides sensitivity. Moreover, the calibration curve being a continuously descending curve characterized by a finite slope, the definition of a single-point cut-off value is less appropri‐ ate than the identification of an indicator range of analyte concentrations within which uncer‐ tain or "non-attributable" results (neither "compliant" nor "noncompliant") fall [44].

**Milk sample**

**AFM1 measured by ELISA (ng l-1)**

330 Aflatoxins - Recent Advances and Future Prospects

**Fortification level (ng**

**Estimated AFM1± SD**

50 78.4 ± 6.2 121 100 153.2 ± 14.1 135

50 40.0 ± 2.0 80 100 121.5 ± 9.8 122

50 79.0 ± 8.6 126 100 125.5 ± 11.0 126

50 74.4± 4.0 117 100 113.0 ± 20.5 97

50 46.8 ± 5.3 94 100 87.5 ± 10.8 88

**Table 2.** Recovery of AFM1 determination from artificially contaminated milk samples undergone to various thermal treatments and with different fat content as determined by the developed LFIA. Recovery was calculated as follows: (estimated AFM1 for the fortified sample – estimated AFM1 for the non fortified sample) / fortification level \*100

The objective of analytical methods such as those based on the LFIA technology is the parting between samples surely complying with legislation in force and samples which do not comply. However, a further category of samples should be considered and is represent‐ ed by those samples in which the toxin content is close to the legal limit which because of measure uncertainty cannot be classified as compliant or noncompliant (Figure 4). These "uncertain samples" should be submitted to further controls before entering the transfor‐ mation chain. In the case of milk, rejection is more often the fate of such uncertain sam‐ ples (as for noncompliant samples), because the perishable nature of milk discourages timeconsuming investigations. Therefore, the purpose of the work could become the development of a very rapid screening method which allowed the semi-quantitation of AFM1 in milk in such a way to permit the discrimination between compliant and noncompliant samples. The instrumental quantification of coloured lines and their correlation with a calibration curve, in this context, could be regarded as a way to limit subjectivity in the interpreta‐ tion of results and to improve detectability [52, 44] rather than going into the direction of

**(ng l-1) Recovery (%)**

**l -1)**

raw 17.8 0 <LOD

whole 1 < LOD 0 <LOD

whole 2 16.0 0 <LOD

skimmed 15.7 0 34.6 ± 1.2

UHT <LOD 0 <LOD

**3.4. Intra-laboratory validation of the semi-quantitative LFIA**

factual quantitative measurements.

As regard precision, European legislation for screening methods of analysis defines as appro‐ priate a relative uncertainty of 47% of the MRL and as acceptable even 94% for AFM1 based on the application of Horwitz equation [71]. Accepting the more restrictive criterion, this means that any screening methods should be able to discriminate between AFM1 content less than 26.5 ng l-1 (negative sample) and AFM1 content over 73.5 ng l-1 (positive sample). Samples that have AFM1 content close to the thres hold limit should thus be defined as uncertain because precision did not allow to reliably attributing them to one or another group.

In spite of this, it should be noted that a "non-attributable" judgement would determine re‐ jection of the sample with a considerable economic damage, as discussed above. Therefore, the minimum number of non-attributable results would be expected for a worth while meth‐ od and this number obviously depends on the combination of accuracy and precision of the method itself. To indicate the capability of a qualitative/semi-quantitative method to pro‐ duce the lowest score of non-attributable results, for a defined uncertainty interval, we intro‐ duced a new parameter indicated as "efficiency" of the method, defined as the ability of the method itself to detect truly non-attributable as non-attributable. Efficiency was thus calcu‐ lated as the number of truly non-attributable tests divided by the sum of known non-attrib‐ utable samples, in strict analogy with "sensitivity" and "selectivity" of qualitative and semiquantitative as says, which are defined as the rate of truly positive e and truly negative test results, respectively [50, 60]. The more efficient the assay, the highest the score of useful re‐ sults (samples certainly attributed as compliant or noncompliant).

The ability of the developed LFIA to correctly attribute to each of the groups milk samples found on the market was thus assessed; in particular, negative (compliant) samples were de‐ fined as those in which AFM1 content was below 30 ng l-1, positive (noncompliant) samples those in which AFM1 content was above 70 ng l-1 and uncertain (non-attributable) those hav‐ ing an AFM1 content between 30 and 70 ng l-1. Since all tested samples were always contami‐ nated below 30 ng l-1 as established by the reference ELISA, positive samples were generated through fortification at 50 and 100 ng l-1. Results of this evaluation, together with the defini‐ tion of sensitivity, selectivity and efficiency, are reported in Table 3.

It can be observed from data that the definition of an indicator range instead of a cut-off lev‐ el allowed us to avoid occurrence of false compliant and false noncompliant. Incorrect attri‐ bution occurred in 15% of samples (6/40), though 3 of them would represent a minor issue being assigned as non-attributable rather than noncompliant, which anyhow mean that sam‐ ples would be discarded. The efficiency is relatively low, however it could still be consid‐ ered acceptable.


the prevalent economic impact of cereals in comparison to other commodities potentially af‐

Lateral Flow Immunoassays for Aflatoxins B and G and for Aflatoxin M1

http://dx.doi.org/10.5772/51777

333

The development of reliable devices for AFM1 detection, conversely, suffers the extreme sen‐ sitivity required to analytical methods aimed at measuring such a contaminant. Very few papers have been published which describe LFIAs for AFM1and none actually meet those requirements, despite the high interest in obtaining adequate systems for the rapid and on

In this paper, we demonstrated that modifying the format of the classic lateral flow assay (such as tailoring the toxin conjugate, used as the competitor in the T-line, and the anti‐ body labelling procedure)a greatdetect ability improvement could be obtained. The estimat‐ ed LOD of the developed semi-quantitative LFIA was one order of magnitude lower than previously published LFIAs for AFM1, therefore allowed us to effectively discriminate be‐ tween compliant and noncompliant samples at a level required by the most severe legisla‐ tion in force. Matrix-matched calibration was necessary to level results obtained on milk samples, however, various matrices (undergone to different thermal treatment and with differing fat contents) could be analysed after a very rapid and easy sample treatment, which involves 2' centrifugation followed by the addition of a small volume of a concentrated

, Claudio Baggiani, Cristina Giovannoli and Gianfranco Giraudi

[1] International Agency for Research on Cancer (IARC). (2002). Evaluation of carcino‐

[2] European Commission. (2006). Commission Regulation (EC) No 1881/2006 of 19 De‐ cember 2006 setting maximum levels for certain contaminants in foodstuffs. *Official*

[3] European Commission. (2010). Commission Regulation (EU) No 165/2010 of 26 Feb‐ ruary 2010 amending Regulation (EC) No 1881/2006 setting maximum levels for cer‐ tain contaminants in foodstuffs as regards aflatoxins. *Official Journal of the European*

[4] European Commission. (2003). Commission Directive 2003/100/EC of 31 October 2003 amending Annex I to Directive 2002/32/EC of the European Parliament and of

fected by aflatoxin contamination.

site monitoring of this toxin.

solution of a surfactant.

\*Address all correspondence to: laura.anfossi@unito.it

Department of Chemistry, University of Turin, Giraudi

genic risks in humans. 82, *Lyon (France):IARC*, 171-274.

*Journal of the European Community*, L364, 5-24.

*Community*, L050, 8-12.

**Author details**

Laura Anfossi\*

**References**

**Table 3.** Evaluation of LFIAs performances on 40 milk samples: 16 negatives, 16 positives and 8 uncertain. The AFM1 reference content was determined by an ELISA kit [19]. Abbreviations used: tp, truly positive (AFM1 below 30 ng l-1); tn, truly negative (AFM1 above 70 ng l-1); tu, truly uncertain (AFM1 between 30 and 70 ng l-1); fn, false negative; fp, false positive; fun, false uncertain and known to be negative; fup, false uncertain and known to be positive.

Finally, the stability of the overall device at room temperature was evaluated as the possibil‐ ity of correctly measuring samples contaminated at low (<30 ng l-1) and high levels (> 70 ng l -1) and by using calibration curves carried out with freshly prepared strips; nevertheless, it could not be confirmed for periods longer than a month.
