Cyclodextrin-Promoted Fluorescence Detection of Aromatic Toxicants and Toxicant Metabolites in Commercial Milk Products

Abstract

The detection of polycyclic aromatic hydrocarbons (PAHs) and their metabolites in food and in agricultural sources is an important research objective due to the PAHs’ known persistence, carcinogenicity, and toxicity. PAHs have been found in the milk of lactating cows, and in the leaves and stems of plants grown in PAH-contaminated areas, thereby making their way into both cow milk and plant milk alternatives. Reported herein is the rapid, sensitive, and selective detection of 10 PAHs and PAH metabolites in a variety of cow milks and plant milk alternatives using fluorescence energy transfer from the PAH to a high quantum yield fluorophore, combined with subsequent array-based statistical analyses of the fluorescence emission signals. This system operates with high sensitivity (low micromolar detection limits), selectivity (100% differentiation even between structurally similar analytes), and general applicability (for both unmodified lipophilic PAHs and highly polar oxidized PAH metabolites, as well as for different cow and plant milk samples). These promising results show significant potential to be translated into solid-state devices for the rapid, sensitive, and selective detection of PAHs and their metabolites in complex, commercial food products.

INTRODUCTION

Polycyclic aromatic hydrocarbons (PAHs) are a class of ubiquitous organic pollutants known for their long-term environmental persistence (Manzetti 2013; Marni et al. 2013), suspected carcinogenicity (Jarvis et al. 2014), and well-documented toxicity to humans and animals (Balcioglu 2016; Hayakawa 2016). PAHs are released into the environment via the incomplete combustion of petroleum products through natural and anthropogenic events, including forest fires (Wang et al. 2017), oil spills (Allan 2012; Forth et al. 2017), vehicular emissions (Ma et al. 2017), and tobacco products (Vu et al. 2015). Once released, PAHs spread widely throughout the environment, persisting in the water (Zgheib 2008), air (Brown et al. 2013), soil (Ruby et al. 2012), and sediment (Wang et al. 2015). PAHs are known to bioaccumulate and biomagnify in the food chain, and can cause substantial toxicity to humans (Wang et al. 2017) and other organisms (Mearns et al. 2015).

Human exposure to PAHs through the aforementioned events has been well-documented, and individuals who have been exposed to PAHs often contain oxidized PAH metabolites in their breast milk (Madhavan et al. 1995), blood (Qin et al. 2011), and urine (Jongeneelen et al. 1987). To understand the extent of human exposure to PAHs, sensitive, selective, and reliable detection methods are necessary. Currently used methods for PAH detection rely on the use of separation techniques such as liquid chromatography (Nieva-Cano et al. 2001) or gas chromatography (Khalili et al. 1995) followed by detection via mass spectrometry (Aiken et al. 2007) or fluorescence spectroscopy (Delgado et al. 2004). Although these methods are highly sensitive, they are time-consuming and expensive (Frysinger et al. 2003), and thus limit the practical applicability of these methods as well as the ability to conduct rapid, high throughput assays of large populations to measure widespread toxicant exposure.

Previous work in our laboratory has focused on the use of cyclodextrin-promoted energy transfer for the rapid, sensitive, and selective fluorescence detection of aromatic toxicants (Mako et al. 2012; Serio et al. 2013b; Serio et al. 2014b; Serio et al. 2015c). We have shown that such detection operates successfully in human urine (DiScenza et al. 2016a; Serio et al. 2014a) and breast milk (DiScenza et al. 2017a), that it can be used for the detection of non-polar aromatic toxicants and highly polar toxicant metabolites (Serio et al. 2015b), and that it can form part of oil spill remediation strategies using cyclodextrin-promoted PAH extraction followed by fluorescence detection (Serio et al. 2013a; Serio et al. 2015a; Serio and Levine 2015; Serio and Levine 2016). Non-photophysically active analytes have also been detected using a modified cyclodextrin-promoted fluorescence modulation (DiScenza and Levine 2016a,b; DiScenza et al. 2016b; DiScenza et al. 2017b).

One fluid that is particularly important as a target for toxicant detection is cow milk, because of its well-documented prevalence in the human diet (Haug et al. 2007; Rozenberg et al. 2016). Cow milk has a high fat content, which enables non-polar lipophilic PAHs to readily accumulate (Abou-Arab 2014; Girello et al. 2014). In fact, researchers have found that lactating cows in PAH-contaminated regions excrete PAHs and PAH metabolites in their milk (Srogi 2007). In addition to widespread cow milk consumption in the human diet, plant milk consumption (i.e. almond milk, cashew milk, and soy milk) has become increasingly popular (Amini 2010; Massey 2003), and these plants can also contain PAH contamination. Such contamination is exacerbated by the presence of waxy substances on leaves that promote the accumulation of hydrophobic compounds, including PAHs (Howsam et al. 2001; Srogi 2007). The presence of PAHs and PAH metabolites in cow milk and plant milk alternatives and subsequent consumption by humans can lead to exposure-related diseases, although the correlation between toxicant exposure and the development of exposure-related disease remains poorly understood. Obtaining better methods for detecting such exposure is a crucial first step in understanding the relationship between PAH exposure and the development of such disease (Srogi 2007).

Reported herein is the rapid, sensitive, and selective detection of 10 PAHs and oxidized PAH metabolites in cow milk and plant milk alternatives using cyclodextrin-promoted energy transfer to high quantum yield fluorophores. This system has been proven to operate with high sensitivity (micromolar detection limits), selectivity (100% differentiation even between structurally similar analytes), and general applicability (for both unmodified lipophilic PAHs and highly polar oxidized PAH metabolites, as well as for different cow and plant milk samples).

MATERIALS AND METHODS

¹H NMR spectra were obtained using a Bruker 300 MHz spectrometer. UV-visible spectra were obtained using a Shimadzu UV-3600 Plus UV-Vis-NIR spectrophotometer. Fluorescence spectra were obtained using a Shimadzu 5301PC spectrophotofluorimeter. GC-MS measurements were obtained using a Shimadzu GC-MS-QP2020 gas chromatograph-mass spectrometer. All toxicants and toxicant metabolites (compounds 1–10, Figure 1) were purchased from Sigma Aldrich and used as received. Fluorophore 11 was synthesized following literature-reported procedures (Shepherd et al. 2004). Fluorophores 12 and 13 were purchased from Sigma Aldrich and used as received. Cow milk (whole and organic whole) and plant milk alternative products (almond milk, cashew milk, soy milk) were purchased from a local grocery store. Raw milk was purchased from a local farm. All milk samples were frozen and stored in a freezer until needed.

Figure 1.

Structures of analytes 1–10 and fluorophores 11–13

General Sample Preparation Procedure for Fluorescence Experiments

Frozen cow milk was thawed overnight in the refrigerator. The milk was then allowed to come to room temperature on the lab bench and shaken vigorously to ensure a uniform suspension. 1875 μL of this suspension and 20 μL of an analyte solution (1 mg/mL analyte in tetrahydrofuran (THF), compounds 1–10, Figure 1) were added to a centrifuge tube. The solution was shaken and allowed to stand for 10 minutes. 938 μL of absolute ethanol and 938 μL of a 10 mM γ-cyclodextrin solution in phosphate buffered saline (PBS) were added to each tube, and the tubes were shaken and allowed to sit overnight in the refrigerator. The solution was centrifuged at 3000 rpm for 10 minutes, after which time it separated into three layers.

2.5 mL of the middle layer was transferred to a quartz cuvette and the fluorescence of the solution was measured as a result of excitation at the analyte’s excitation wavelength and a result of excitation at the fluorophore’s excitation wavelength. Then, 100 μL of a fluorophore solution (0.1 mg/mL in THF, compounds 11–13, Figure 1) was added to the cuvette, and the solution was again excited at the analyte’s excitation wavelength and at the fluorophore’s excitation wavelength.

The same procedure was followed but without the addition of an analyte to obtain data in undoped samples.

General Procedure for Energy Transfer Experiments

The efficiency of the energy transfer was quantified as the ratio of the integrated emission of the fluorophore via analyte excitation to the integrated emission of the fluorophore via direct excitation:

$$\text{Energy}\text{transfer}\text{efficiency}=\frac{{\mathrm{I}}_{\text{DA}}}{{\mathrm{I}}_{\mathrm{A}}}\times 100\%$$ (Eq. 1)

Where I_DA is the integration of the fluorophore from analyte excitation in the presence of analyte and I_A is the integrated fluorophore emission from direct excitation.

Analyte comparison ratios were calculated for each experiment according to Equation 2:

$$\text{Analyte}\text{comparison}=\frac{{\mathrm{I}}_{\text{analyte}}}{{\mathrm{I}}_{\text{blank}}}$$ (Eq. 2)

where I_analyte is the integrated emission of the fluorophore from excitation at the analyte’s excitation wavelength in the presence of the analyte, and I_blank is the integrated emission of the fluorophore from excitation at the analyte’s excitation wavelength in the absence of the analyte. Analyte comparisons greater than 1 indicate that the fluorophore emission via excitation at the analyte wavelength is higher in the presence of the analyte, indicating legitimate energy transfer is occurring. Analyte comparison ratios less than 1 indicate that the fluorophore emission via excitation at the analyte wavelength is lower in the presence of the analyte, indicating analyte-driven fluorescence quenching. Analyte comparisons near 1 indicate little to no interaction of the analyte with the fluorophore, resulting in a fluorescence emission signal that is essentially unchanged with the addition of the analyte (the putative energy donor).

General Procedure for Limit of Detection Calculations

The limit of detection (LOD) is defined as the lowest concentration of analyte at which a signal can be detected. These experiments were conducted following literature-reported procedures (Cheng et al. 2016). To determine this value, the following steps were performed for each analyte-fluorophore combination:

1. Undoped, analyte-free milk samples were prepared according to the general sample preparation procedure for fluorescence experiments.

2. 2.5 mL of the middle layer was transferred to a cuvette. 100 μL of a 0.1 mg/mL solution of fluorophore 11 in THF was added to the cuvette, and the solution was excited at the fluorophore’s excitation wavelength and analyte’s excitation wavelength and fluorescence spectra were recorded. Six repeat measurements were taken.

3. 5 μL of a 1.0 mg/mL analyte solution in THF was added to the cuvette and the solution was again excited at the fluorophore’s excitation wavelength as well as at the analyte’s excitation wavelength. Six repeat measurements were taken.

4. Step 2 was repeated, adding analyte in 5 μL increments until a final volume of 40 μL was added. In each case, the solution was excited at the fluorophore’s excitation wavelength and analyte’s excitation wavelength and the fluorescence emission spectrum was recorded six times.

5. All fluorescence emission spectra were integrated vs. wavenumber, and calibration curves were generated with the analyte concentration on the X-axis (in μM) and the energy transfer ratio (I_DA/I_A) on the Y-axis. The curve was then fitted to a straight line and an equation for the line was determined.

6. For each case, the solution before any analyte was added was also excited at the excitation wavelength for the fluorophore and excitation wavelength for the analyte and the fluorescence emission spectra were recorded (as per step 1). These measurements are referred to as the “blank.”

7. The limit of detection is defined according to Equation 4:

   $$\text{LOD}=\frac{3({\text{SD}}_{\text{blank}})}{\mathrm{m}}$$ (Eq. 4)

   Where SD_blank is the standard deviation of the blank and m is the slope of the calibration curve. In cases where the slope of the trend line was negative, the absolute value of the slope was used to calculate LOD. In all cases, the LOD was calculated in μM.

General Procedure for Array Generation Experiments

Array-based analysis was performed using SYSTAT 13 statistical computing software with the following settings:

1. Classical Discriminant Analysis

2. Grouping variable: Analytes

3. Predictors: γ-cyclodextrin/BODIPY, γ-cyclodextrin/Rhodamine 6G, γ-cyclodextrin/Coumarin 6

4. Long-range statistics: Mahal

General Procedures for Undoped Characterization Experiments

GC-MS Experiments

Sample preparation experiments for GC-MS were conducted following literature-reported procedures (Baduel et al. 2015). In brief, 3 mL of each milk sample were added to 8 mL centrifuge tubes. Each sample was vortexed for one minute and allowed to settle for 15 minutes. 3 mL of acetonitrile were added, and each tube was shaken vigorously by hand. Then 1.2 grams of MgSO₄ and 0.3 grams of NaCl were added to each centrifuge tube and shaken for 1 minute. Samples were placed in the freezer for 10 minutes, followed by centrifugation at 3700 rpm for 10 minutes. The supernatant was removed from each sample and analyzed by GC-MS.

All GC-MS measurements were obtained on a Shimadzu-QP2020 gas chromatograph-mass spectrometer following literature-reported procedures (Pfannkoch et al. 2010). The GC-MS operating conditions were as follows: column, Shimadzu SH-Rxi-5SilMS (30 m × 0.25 mm × 0.25 μm); carrier gas, helium at 2.0 ml/min; oven temperature, 60 °C (1 min) →15 °C/min 325 °C (3 min); injection temperature, 250 °C; splitting ratio, splitless; electron impact (EI) ionization mode; MS ion source temperature, 230 °C; interface temperature, 150 °C; total run time, 22 minutes.

Fluorescence Experiments

Defrosted milk samples (3 replicates from each milk type) were prepared following the general sample preparation procedures. 100 μL of fluorophores 11–13 were added, and each sample was excited at various wavelengths characteristic of photophysically active organic contaminants. Energy transfer efficiencies were calculated using Equation 1, and arrays were generated for each sample following experimental procedures for array generation experiments.

RESULTS AND DISCUSSION

Undoped Milk Characterization

Sample characterization experiments using gas chromatography-mass spectrometry (GC-MS) and fluorescence spectroscopy were performed on undoped (analyte-free) milk samples to identify key differences in the composition of the cow milk and plant milk samples.

Fluorescence Experiments

Analyte-free samples to which small amounts of fluorophores 11–13 were added demonstrated unique, sample-specific fluorescence response patterns from excitation at a wide variety of wavelengths. These specific wavelengths were chosen to represent excitation wavelengths characteristic of aromatic toxicants. Samples were excited at various excitation wavelengths to measure how the signals change in the presence of inherent aromatic toxicants. Linear discriminant analyses of these fluorescence responses generated distinct pattern identifiers for all wavelengths in all milk samples (Figure 2). The goal of linear discriminant analysis with jackknifed reclassification is to maximize the distance between different samples (i.e. 360 nm vs. 365 nm) while minimizing the distance between the four trials of each specific sample to result in 100% differentiation between samples (Bajaj et al. 2010). Array-based analysis has been used by our research group (Serio et al. 2015b) and a number of other research groups to measure selectivity of detection systems (Creran et al. 2015; Hizir et al. 2017; Miranda et al. 2010; Tao et al. 2017).

Figure 2.

Array response patterns using linear discriminant analysis from undoped samples of: (A) whole milk, (B) organic whole milk, (C) raw milk, (D) almond milk, (E) cashew milk, and (F) soy milk using cyclodextrin-fluorophore predictors.

Each milk sample demonstrated a unique visual response pattern, likely due to the differences in the fatty acid composition of each sample (see GC-MS section for more details) (Leroux et al. 2016), which lead to different local micro-environments for fluorophores 11–13 that are detectable spectroscopically. Analogous sensitivities to local environments for BODIPY (Jiang et al. 2013; Marfin et al. 2014; Zhang et al. 2013), Rhodamine (Ji et al. 2017; Ren et al. 2016), and Coumarin (Phatangare et al. 2014; Sarkar et al. 2015) fluorophores have been reported previously.

Some visual similarities between the group of plant milks (soy, almond, cashew) and the group of cow milks (organic, whole, and raw) were also observed. In particular, the arrays generated for whole milk and organic whole milk appear to be mirror images of each other, as do the arrays generated from almond and cashew milk. Moreover, the location of the signals from 380 nm excitation and 385 nm excitation are remarkably similar in the whole milk and raw milk arrays, as well as in comparing the arrays generated from almond and soy milk. Interestingly, even as little as a 5 nm difference in excitation wavelength (360 vs. 365 or 380 vs. 385 nm) resulted in 100% differentiation between response signals using array-based analysis. This shows the ability of our system to differentiate between extremely small variations.

Energy transfer efficiencies in the analyte-free milk solutions (to which small amounts of fluorophores 11–13 were added) were calculated according to Equation 1, and the results are summarized in Table 1. These results show that the energy transfer efficiencies were highest in the presence of fluorophore acceptor 11, which is a consequence of the high quantum yield of that fluorophore (0.96 in tetrahydrofuran) (Zhang et al. 2016; Zhao et al. 2017), its electrostatic potential surface which is strongly complementary to that of the aromatic analytes (see ESI for more details), and its demonstrated ability to bind in or near the γ-cyclodextrin cavity and facilitate proximity-induced energy transfer (Milles et al. 2013). In contrast, the structure of fluorophore 12 makes it less able to penetrate the cyclodextrin cavity (both because of the large steric size as well as its central twisted biphenyl axis) (Zhao et al. 2010), and the quantum yield of fluorophore 13 is lower than that of fluorophore 11 (Rodionova et al. 2008). The energy transfer efficiencies observed in the soy milk samples were markedly lower than those observed in the other milk samples, likely as a result of soy protein aggregation (vide infra).

Table 1.

Calculated energy transfer efficiencies in undoped milk samples^a

Fluorophore	Organic	Whole	Raw	Almond	Cashew	Soy
11	21.8 ± 0.1	21.3 ± 0.4	31.8 ± 2.0	21.7 ± 0.2	23.1 ± 0.2	5.1 ± 0.0
12	8.5 ± 0.0	4.3 ± 0.0	7.4 ± 0.0	7.4 ± 0.0	7.2 ± 0.0	2.6 ± 0.0
13	15.5 ± 0.0	14.8 ± 0.0	14.6 ± 0.0	7.1 ± 0.1	6.7 ± 0.0	3.3 ± 0.0

^aEnergy transfer efficiencies were calculated using Equation 1; λ_ex (analytes) = 360 nm; λ_ex (11) = 460 nm; λ_ex (12) = 490 nm; λ_ex (13) = 420 nm. Results presented in table represent an average of four trials.

GC-MS

Differences between classes of cow milk and plant milk samples were further confirmed by GC-MS analysis (see ESI for full details). The cow milk samples (whole, organic, raw) showed peaks corresponding to fatty acid compounds typically found in cow milk, with raw milk showing a higher number and higher intensity of peaks (Figure 3A). The GC-MS spectra from plant milk samples showed peaks that indicated the presence of fatty acids, nut shell or plant protein components, flavorings, and other additives commonly found in plant milk (Figure 3B) (Fagbemi 2009; Scilewski da Costa Zanatta et al. 2017; Valdes et al. 2015). Notably, soy milk showed the highest number of peaks, which can be attributed to the hydrolysis of soy proteins into amides that occurs during soy milk preparation procedures (Jaiswal et al. 2015). Other peaks in the soy milk GC-MS spectra correspond to furfurals, which are known agricultural byproducts, and a variety of alcohols, glycols, and stearates which are commercially used de-foaming agents (Kamath et al. 2011).

Figure 3.

GC-MS overlay of: (A) cow milk samples where the black line represents organic whole milk, the pink line represents whole milk, and the blue line represents raw milk, and (B) plant milk samples where the black like represents soy milk, the pink line represents cashew milk, and the blue line represents almond milk.

Analyte Selection

The analytes selected include PAHs that have been found in dairy products (compounds 2, 6, 8) (Santonicola et al. 2017), known PAH metabolites (compounds 1, 3, 4, 7, and 9) (Lee et al. 2015), an aromatic amine that is known to cause bladder cancer (compound 5) (Zhao et al. 2015), and a partially hydrogenated aromatic amine that has been investigated for pesticide activity (compound 10) (Choudhury and Das 1983; Das et al. 1998).

Analyte-Doped Energy Transfer

To enable the efficient detection of specific analytes in the complex milk environments, the milk samples were doped with micromolar concentrations of analytes 1–10 (in addition to micromolar concentrations of the fluorophores) prior to sample preparation and centrifugation procedures. Doping analytes and fluorophores at this early stage provides a mechanism to assess the extent to which preparation procedures for all milk samples retain the innate toxicants of interest. The amount of analyte and fluorophore that remained after sample preparation was quantified using UV-visible spectroscopic measurements, and the results indicated that significant amounts of all analytes and fluorophores remained in the analyzed layer (see ESI for more details).

Table 2 shows energy transfer efficiencies from analytes 1–10 to fluorophore 11 in analyte-doped milk samples (data obtained using the other fluorophore acceptors are shown in the ESI). The efficiency of the energy transfer was quantified as the ratio of the integrated emission of the fluorophore from analyte excitation to the integrated emission of the fluorophore from direct excitation, and is shown as a percentage value (Equation 1). The extent to which the observed fluorescence signals represent legitimate energy transfer was determined from the analyte comparison ratios (Equation 2, values shown in parentheses in Table 2), with ratios less than 1 indicating the presence of “legitimate” energy transfer rather than a result of exciting the toxicant donor at a wavelength where it has a notable absorption cross-section.

Table 2.

Calculated energy transfer efficiencies and analyte comparisons in analyte-doped milk samples with fluorophore 11^a

Analyte	Whole	Organic	Raw	Almond	Cashew	Soy
1	19.1 ± 0.4 (0.87)	19.2 ± 0.3 (0.75)	20.7 ± 0.6 (1.04)	24.2 ± 0.0 (1.12)	23.7 ± 0.1 (1.79)	13.7 ± 0.5 (0.44)
2	23.3 ± 0.2 (1.14)	25.4 ± 0.1 (1.11)	19.8 ± 0.2 (0.99)	21.3 ± 0.2 (2.08)	18.3 ± 0.2 (1.25)	20.6 ± 1.5 (0.57)
3	22.4 ± 0.1 (1.11)	23.2 ± 0.3 (1.09)	20.0 ± 0.3 (0.99)	21.4 ± 0.2 (1.08)	17.4 ± 0.1 (1.14)	13.2 ± 1.1 (0.59)
4	22.8 ± 0.2 (1.04)	22.9 ± 0.4 (0.86)	20.6 ± 0.2 (1.06)	21.0 ± 0.3 (1.27)	16.7 ± 0.1 (1.10)	17.1 ± 0.7 (0.39)
5	23.5 ± 0.2 (1.12)	23.7 ± 0.5 (0.88)	17.6 ± 0.2 (0.76)	20.0 ± 0.1 (1.04)	16.8 ± 0.2 (1.14)	10.4 ± 0.1 (1.03)
6	18.3 ± 0.2 (0.89)	21.0 ± 0.2 (0.74)	20.9 ± 0.0 (0.92)	22.2 ± 0.1 (1.19)	16.9 ± 0.0 (1.12)	10.6 ± 0.3 (0.99)
7	19.8 ± 0.1 (0.95)	22.6 ± 0.3 (1.02)	17.6 ± 0.3 (0.91)	23.3 ± 0.1 (1.25)	25.3 ± 0.2 (1.61)	8.9 ± 0.6 (0.82)
8	23.3 ± 0.2 (1.15)	25.5 ± 0.3 (1.01)	20.0 ± 0.3 (0.99)	20.7 ± 0.3 (1.11)	17.0 ± 0.2 (1.25)	13.4 ± 1.4 (1.59)
9	31.0 ± 0.3 (0.52)	35.9 ± 0.6 (0.41)	23.0 ± 0.3 (0.49)	20.7 ± 0.1 (0.96)	15.9 ± 0.2 (0.88)	14.1 ± 0.9 (0.46)
10	13.8 ± 0.2 (0.65)	15.5 ± 0.2 (0.60)	13.7 ± 0.0 (0.62)	17.6 ± 0.1 (0.87)	13.1 ± 0.1 (0.98)	4.5 ± 0.4 (0.45)

^aEnergy transfer efficiencies were calculated using Equation 1; Analyte comparisons in parentheses were calculated using Equation 2; λ_ex (analytes) = 360 nm; λ_ex (11) = 460 nm; Results presented in table represent an average of four trials.

For the whole and organic cow milks, legitimate energy transfer (as indicated by analyte comparison ratios greater than 1) was observed for analytes 2, 3, and 8. In the case of the other analytes, although no energy transfer is occurring, it is noteworthy that the introduction of the analyte led to noticeable changes in the fluorophore emission signal (as indicated by analyte comparison ratios that are substantially different from 1). Previous work in our group has demonstrated the ability of several aromatic and non-aromatic toxicants to modulate the fluorescence emission of fluorophores 11–13 (DiScenza and Levine 2016a, b; DiScenza et al. 2016b; DiScenza et al. 2017b). Analyte 8 is a particularly important target for detection due to its known toxicity (Chepelev et al. 2015), carcinogenicity (Ceccaroli et al. 2015), and teratogenicity (Wells et al. 1997). Moreover, its strong binding in the hydrophobic γ-cyclodextrin cavity has been well-documented in the literature (Ghosh et al. 2012; Mousset et al. 2014; Munoz de la Pena et al. 1991). In the case of raw milk, only analytes 1 and 4 act as legitimate energy donors likely due to the complex nature of raw milk (see GC-MS section above).

More cases of legitimate energy transfer were observed for the nut-based almond and cashew milks, with all analytes except compounds 9 and 10 acting as legitimate energy donors in this context. This differential behavior is likely a result of the increased non-covalent interactions due to the composition of these milks (see GC-MS section above). In the case of soy milk, low energy transfer values were observed for all analytes investigated. These lower values compared to the other milk samples are likely a result of the aggregation of large soy proteins upon centrifugation and subsequent binding of PAHs in the aggregated proteins, a phenomenon that has been well-documented in the literature (Nik et al. 2008).

Both the fluorescence modulation or legitimate energy transfer occurring from doped analytes to fluorophore 11 in organic whole milk, whole milk, raw milk, almond milk, and cashew milk can be attributed to amphiphilic fatty acids in milk that are known to form micelles with hydrophobic interiors (Bahri et al. 2017). These micelles work synergistically with the cyclodextrin to promote PAH inclusion (Dai et al. 2017), leading to significant energy transfer efficiencies. The energy transfer efficiencies observed for all cow milks were similar in magnitude, which highlights the robust nature of this cyclodextrin-promoted detection system to operate successfully in multiple environments. These results represent improvements over our previous results obtained in purified buffer solution where many of the presumed cases of cyclodextrin-promoted energy transfer were determined to be a result of directly exciting the analyte at a wavelength where it has noticeable absorption rather than a result of legitimate energy transfer. These improved results can be attributed to the differences in media composition between the aqueous buffer and lipophilic milk, as previously described above.

Practical Considerations

Sensitivity

The sensitivity of the system was determined by calculating limits of detection for all analytes in all milk samples using literature-reported procedures (Table 3). Both aromatic toxicants as well as toxicant metabolites can be detected at micromolar concentrations, which are comparable (after mathematical conversion of the micromolar values to parts-per-million) to the reported permissible exposure limits of 0.2 ppm for compounds 2, 6, and 8 (Occupational Safety and Health Administration 2012). The only notable exceptions to such low LODs were the values obtained in raw milk, which is likely due to the complex nature of that environment. Current efforts in our laboratory are focused on lowering all detection limits to obtain optimal detection sensitivities. To that end, we are exploring the possibility of incorporating a pre-concentration step to obtain higher effective concentrations and enable practical detection systems.

Table 3.

Calculated limits of detection for analytes 1–10 with fluorophore 11in cow milk and plant milk alternatives.^a

Analyte	Whole (μM)	Organic (μM)	Raw (μM)	Almond (μM)	Cashew (μM)	Soy (μM)
1	5.85 ± 0.00	3.43 ± 0.46	3.21 ± 0.00	0.09 ± 0.00	5.87 ± 0.06	2.30 ± 0.05
2	5.40 ± 0.00	1.33 ± 0.09	35.40 ± 0.00	0.38 ± 0.01	2.70 ± 0.00	1.85 ± 0.00
3	2.22 ± 0.37	2.30 ± 0.00	21.19 ± 0.00	3.58 ± 0.00	20.60 ± 3.46	31.00 ± 0.00
4	4.01 ± 0.00	6.54 ± 0.00	6.24 ± 0.00	6.50 ± 0.00	27.08 ± 10.26	4.30 ± 0.66
5	87.54 ± 11.67	10.97 ± 0.00	44.86 ± 0.00	3.38 ± 0.00	27.33 ± 8.24	45.11 ± 0.00
6	4.17 ± 0.13	17.26 ± 0.00	2.42 ± 0.00	0.24 ± 0.00	0.13 ± 0.00	0.07 ± 0.00
7	0.39 ± 0.00	1.76 ± 0.00	8.42 ± 0.00	1.53 ± 0.02	3.54 ± 0.13	6.79 ± 0.00
8	23.18 ± 0.00	8.13 ± 0.00	4.63 ± 0.00	2.24 ± 0.00	2.77 ± 0.07	4.70 ± 0.00
9	0.83 ± 0.07	0.33 ± 0.02	5.38 ± 0.00	3.39 ± 0.40	11.22 ± 1.39	3.10 ± 0.00
10	5.92 ± 0.00	4.34 ± 0.00	10.27 ± 0.00	2.74 ± 0.26	8.04 ± 0.00	33.44 ± 0.00

^aLimits of detection were calculated using the procedures in Chen 2016; see Electronic Supporting Information for more information.

Selectivity

The selectivity of the system in correctly identifying analytes was determined using array-based statistical analysis of the integrated fluorophore emission signals. Linear discriminant analysis (LDA) of this data resulted in 100% success in the differentiation of all 10 analytes (Figure 4), even among the structurally similar compounds (i.e. compounds 1 vs. 6, and compounds 3 vs. 4). Such high selectivity is particularly relevant for consumer safety detection applications, because structurally similar analytes often have widely disparate toxicities. One way to understand this high level of selectivity is to look at the fluorescence spectra of each analyte-fluorophore-milk combination, which shows clear differentiation and leads to a unique ‘fingerprint’ for each combination (see ESI for more details). These results highlight the uniqueness of this approach for achieving a high level of selectivity and differentiation between structurally similar analytes.

Figure 4.

Array response patterns using linear discriminant analysis from analyte-doped samples of: (A) whole milk, (B) organic whole milk, (C) raw milk, (D) almond milk, (E) cashew milk, and (F) soy milk using cyclodextrin-fluorophore predictors.

Generality

The general applicability of this detection method was demonstrated by comparing the fluorescence response signals of the toxicant analyte donors in the six different milk samples, all of which have significant differences in their chemical compositions, including notable differences in their lipid contents. For these experiments, we selected a benzo[a]pyrene (compound 6) as a high-impact target for detection due to its significant toxicity. The integrated fluorescence responses from analyte 6 in the presence of the six different milk samples yielded six unique response signals (Figure 5), with the statistical analysis providing 100% success in identifying the milk environment in which the analyte was found. The distinct arrays seen for each milk sample highlight the potential for profiling samples based on such identification, and creating milk-specific patterns for accurate sample identification.

Figure 5.

Array response patterns using linear discriminant analysis of response patterns of analyte 6 in (A) cow milk, (B) plant milk, and (C) all milk samples using cyclodextrin-fluorophore predictors.

In conclusion, cyclodextrin-promoted, proximity-induced fluorescence modulation and energy transfer from 10 PAHs and PAH metabolites to high quantum yield fluorophores 11–13 enabled successful detection in multiple types of cow milk and plant milk alternatives. This fluorescence energy transfer operates with high sensitivity (micromolar detection limits), selectivity (100% differentiation between analytes and 100% selectivity in accurate identification of the milk environment), and general applicability (for both unmodified lipophilic PAHs and highly polar oxidized PAH metabolites, as well as for samples with varying chemical composition). These results show significant potential in the development of practical detection systems for consumer safety applications. Current efforts in our laboratory are focused on the development and optimization of such systems, and results of these and other investigations will be reported in due course.