Development and evaluation of a real-time PCR assay for the detection of bovine Milk in foods

Abstract

The Food Drug & Cosmetic Act designates milk as a major food allergen and requires its labeling on food packages. As such, robust, sensitive detection methods are required to ensure accurate labeling. Immunoassays such as ELISAs detect the allergenic proteins but may be limited by cross-reactivity and protein degradation that occurs during food processing. DNA-based detection methods are often less affected by food processing and offer a robust alternative. In this work, a real-time PCR assay targeting the COI mitochondrial gene for Bos taurus was developed and evaluated for the detection of bovine DNA, specifically in the context of bovine milk in foods. Bovine milk was spiked at known concentrations into four complex food matrices –dark chocolate, vegan cookies, dairy-free muffins, and orange juice—chosen for their relevance to allergen recalls and robustness testing. Because milk is considerably lower in DNA concentration than other bovine tissues, beef was used as a control in a parallel set of spiking experiments to evaluate assay performance. The assay detected milk at concentrations as low as 100 ppm across food matrices, except in dark chocolate where internal controls indicate the presence of PCR inhibitors. Comparison between beef and milk spike data suggests that the lower sensitivity of the assay for milk detection is attributable to inherently low bovine DNA concentration in milk. This method should be used as a screening tool or in conjunction with immunoassays when food processing has compromised allergenic proteins.

Graphical abstract

Highlights

• •Real-time PCR method targeting COI gene of Bos taurus for detection of milk.
• •Detection of milk in food matrices representing major allergen recall categories.
• •Minimal cross-reactivity from other ruminant species.
• •Assay sensitivity limitations are attributable to low DNA concentration in milk.

1. Introduction

The prevalence of food allergies in the United States has been estimated as high as 10%, with several reports suggesting an increase in the number of affected individuals over the past two decades. Among specific allergens, the prevalence of milk allergy is estimated at approximately 3% (Boyce et al., 2010; Gupta et al., 2019; Seth et al., 2020). The Food, Drug, & Cosmetic Act designates milk as one of nine major regulated allergens in the United States. Recent guidance further defines “milk” as milk from domesticated cows, goats, sheep, and other ruminants, and mandates that food labels include the species or common name (United States Food and Drug Administration, 2025). Despite these regulations, milk was the most frequently cited major food allergen in US recalls from 2007 to 2019 — a finding consistent with both global food allergen recall data and, more recently, 2024 U.S. recall data. Notably, milk was the only major allergen cited in recalls across all 25 food product categories with the top three categories being bakery products, dough, bakery mixes, icings; chocolate and cocoa products; and snack food items (flour-, meal-, or vegetable- based) (Sharma et al., 2023; United States Food and Drug Administration, 2026). Because avoidance remains the only prevention strategy for allergic individuals, the prevalence of milk allergy and its pervasiveness across the food supply underscore the need for accurate and reliable detection methods to ensure proper labeling.

Immunoassays are the predominant methods for detecting allergenic proteins. However, DNA-based detection methods can serve as complementary tools in cases where compromised protein integrity or immunoassay specificity limit accurate detection and quantification of allergenic proteins. It is well established that commercial food processing practices, matrix effects, cross-reactivity, and variance among commercial kits can affect the accurate detection and quantification of proteins by immunoassays (Cordeiro et al., 2021; Gomaa & Boye, 2013; Panda et al., 2026; Prado et al., 2016; Zhang et al., 2025). DNA-based methods are generally less susceptible to these same factors and can offer species-specific identification and sensitive detection across a variety of complex food matrices, performing favorably in side-by-side comparisons with ELISA (Costa et al., 2014; Eischeid et al., 2021; Hird et al., 2006; Palle-Reisch et al., 2015; Pedersen et al., 2008; Poms et al., 2004; Stephan & Vieths, 2004). Nevertheless, an important limitation of DNA-based methods is that they indirectly detect the allergenic ingredient and cannot differentiate between animal tissues (i.e., meat vs. milk). Therefore, the advantages and limitations of each method should be considered in the context of the target allergen and the food matrix. In the case of milk detection, protein-based assays may offer greater sensitivity, while DNA-based methods may be particularly useful for species verification when cross-reactivity between homologous proteins is a concern.

It is well documented that milk is inherently low in DNA concentration compared to other bovine tissues (Baumert, 2014; Gantner et al., 2015; Holzhauser, 2018; Holzhauser et al., 2020; Liu et al., 2014; Roy et al., 2020). Genomic DNA in milk is derived solely from somatic cells — white blood cells and epithelial cells sloughed from the mammary gland — which are present in significantly lower quantities than blood or tissue (Alhussien & Dang, 2018; Liu et al., 2014; Pokorska et al., 2016). Milk also harbors a complex microbiome with bacterial load estimates comparable to somatic cell counts in healthy cows (Oikonomou et al., 2020; Quigley et al., 2011; Quigley et al., 2013a; Quigley et al., 2013b). Consequently, the bovine gene target is present both in low abundance and constitutes a smaller proportion of total DNA than would be expected in other bovine tissues.

Despite the challenges posed by low DNA concentration, several real-time PCR assays for the detection of cow's milk have been developed to identify fraud and adulteration in cheese and other dairy products (Kalogianni, 2018). These studies aimed at detecting bovine milk as a percentage of other dairy products, primarily for authentication purposes, and not necessarily at the low detection levels required for allergen detection. Few works have focused on detecting cow's milk as an undeclared allergen in non-dairy foods (Villa et al., 2019; Villa et al., 2020). Villa et al. used milk protein concentrate (MPC) to incur meat samples, emulating the addition of dry milk to meat products. However, DNA yields from MPC are considerably higher than from liquid milk and pose less of a challenge for assay sensitivity (Liao et al., 2017; Villa et al., 2020). As such, these studies do not fully address the challenge of detecting liquid milk as an ingredient in complex food matrices — a gap that is particularly relevant given that bakery products, snack food items, and chocolate are among the food categories most frequently recalled for undeclared milk. Therefore, this work focuses on the development of a real-time PCR assay targeting the COI mitochondrial gene of Bos taurus, or domestic cow, for detection of bovine DNA — specifically liquid milk —in complex food matrices.

The COI mitochondrial gene target was selected for two primary reasons: its well-documented ability to distinguish organisms at the species level and its high copy number relative to nuclear gene targets (Hebert et al., 2003; Kalogianni, 2018; Maudet & Taberlet, 2001). High-copy gene targets have been shown to improve assay sensitivity which is particularly advantageous when DNA concentration is low, as is the case with milk and egg (Holzhauser, 2018; Holzhauser et al., 2020; Sanchiz et al., 2021; Villa et al., 2023). A comparison of several mitochondrial gene targets (COI, Cyt b, 12S) for real-time PCR detection of bovine DNA revealed comparable sensitivities and similar cross-reactivities across all three targets (Villa et al., 2020). Taken together, these findings supported the selection of COI for assay development and evaluation.

The matrices for assay evaluation were selected based on three criteria: representation of foods in which milk commonly appears (baked goods), significance to undeclared allergen concerns (i.e., milk in dark chocolate), and robustness testing under conditions known to degrade DNA, such as low pH (orange juice) (Bauer et al., 2003; Bedford et al., 2017; Blom et al., 2018). These matrices also align with the top product categories implicated in FDA food allergen recalls from 2013 to 2019 with “bakery products, dough, bakery mixes and icings” ranked first (367 recalls) and “chocolate and cocoa products” ranked second (120 recalls) (Sharma et al., 2023).

Given the challenges of low bovine DNA concentration in milk, beef was included as an alternative tissue to provide a more concentrated and homogeneous source of bovine DNA. Bovine skeletal muscle is mitochondria-rich relative to the epithelial or immune cells that contribute DNA to milk, resulting in a higher copy number of the COI mitochondrial gene target in beef (Cole, 2016; Rath et al., 2024). By evaluating assay performance using beef as a high-DNA-yield source of bovine DNA, limitations attributable to assay design can be distinguished from those arising from the low bovine DNA concentration inherent to milk. Strong performance with beef but reduced sensitivity with milk would indicate that assay limitations are driven by DNA concentration rather than assay design, whereas poor performance with both tissues would suggest a fundamental limitation of the assay itself.

2. Materials and methods

2.1. Primer/probe design and selection

COI mitochondrial gene sequences for Bos taurus (GenBank accession numbers FJ958332, FJ958333, FJ958334, FJ958335, FJ958336, GU130589, GU130590) were aligned using Geneious software (Geneious Prime® 2022.1.1 (https://www.geneious.com); National Center for Biotechnology Information, n.d). Five primer/probe sets ranging from 100 to 300 bp were designed within a 700 bp segment of the gene using the “primers” tool. Primer and probe sets were first tested with respect to linear range and efficiency using target DNA in buffer. Primer sets with acceptable efficiencies were screened for cross-reactivity with milk and meat samples from other ruminant species. The primer/probe set with the best efficiency and least amount of cross reactivity was chosen for further optimization studies. The primer and probe sequences are as follows: forward primer 5’-GTAACCGCACACGCATTTGT-3′, reverse primer 5′- GGAATGAGGGAGGGAGGAGT-3′, probe 5′- 56-FAM-AAATGCTAT-ZEN-ATCGGGAGCAC-3IABkFQ-3′. All primers, probes, and template DNA for assay testing and internal controls (Deer et al., 2010) were obtained from Integrated DNA Technologies (Coralville, IA, USA) as custom qPCR probes and custom DNA oligos. Internal control sequences are as follows: forward primer 5’-CTAACCTTCGTGATGAGCAATCG-3′, reverse primer 5’-GATCAGCTACGTGAGGTCCTAC-3′, probe 5’-HEX-AGCTAGTCG-ZEN-ATGCACTCCAGTCCTCCT-3’-IowaBlack FQ, IAC template TACAGCACCCTAGCTTGGTAGAATCGATCAGCTACGTGAGGTCCTACGACGATCGCCAAGCATGCCCTAGCTAAGATGCATCGATTGCTCATCACGTACGTTAGGTCGACTAGGAGGACTGGAGTGCATCGACTAGCTAAGATGGTTCGATTGCTCATCACGAAGGTTAGGTCGACTACGAACGAGTCGTACGTTAGGTT (Deer et al., 2010).

2.2. Efficiency, limit of detection (LOD), and cross-reactivity tests

Ten-fold serial dilutions of DNA extracted from beef and whole milk (1 ng/μL- 1 × 10⁻⁶ ng/μL) were analyzed over twelve replicates to calculate assay efficiency, R² coefficient, limit of detection (LOD), and limit of quantification (LOQ) according to validation guidelines published by the U.S. Food and Drug Administration (United States Food and Drug Administration, 2023).Beef was evaluated in parallel with milk throughout assay validation as a source of more concentrated bovine DNA target. Cross-reactivity was assessed in three independent experiments with samples analyzed in triplicate. Samples included milk and meat from other ruminant species, as well as commercial genomic DNA standards (Zyagen, San Diego, CA, USA) to control for possible cross-contamination with bovine DNA in market samples (Table 1).

Table 1.

Cross-Reactivity Data. DNA was extracted from each sample, diluted to 1 ng/μl, and tested for cross-reactivity in triplicate over three separate experiments. Samples with no amplification for at least 5 of the 9 replicates were considered negative.

	Experiment 1			Experiment 2			Experiment 3
Sample	CT 1	CT 2	CT 3	CT 4	CT 5	CT 6	CT 7	CT 8	CT 9	Average CT	SD	CV (%)
Bovine Standard	19.72	19.52	19.53	18.19	18.23	18.16	16.74	16.71	16.85	18.18	1.15	6.3
Beef (meat)	15.07	15.03	14.99	23.42	23.44	23.43	18.02	18.21	18.10	18.86	3.47	18.4
Whole Milk	21.00	20.75	20.91	23.76	23.80	23.75	21.35	21.43	21.40	22.02	1.26	5.7
Pork Standard	29.91	29.51	29.70	28.02	27.86	28.05	27.10	27.49	27.39	28.34	1.01	3.6
Water Buffalo Milk	26.09	25.29	25.37	29.21	28.92	28.56	31.56	29.29	31.37	28.41	2.23	7.8
Goat Meat	27.39	27.27	27.28	30.57	30.75	30.71	31.11	31.12	31.11	29.70	1.70	5.7
Lamb	27.10	27.02	27.04	30.71	31.12	30.55	31.45	31.38	31.40	29.75	1.93	6.5
Camel Standard	36.29	N/A	36.34	34.36	34.20	33.59	35.07	34.01	32.95	34.60	1.14	3.3
Pork	31.25	31.18	31.97	35.02	34.70	35.54	36.44	43.30	37.25	35.18	3.56	10.1
Goat Milk	37.93	N/A	35.98	N/A	N/A	37.47	N/A	38.46	N/A	–
Chicken	35.61	34.84	40.30	N/A	37.08	N/A	N/A	N/A	N/A	–
Sheep Milk	36.15	39.34	N/A	N/A	N/A	N/A	N/A	N/A	N/A	–
Sheep Standard	N/A	N/A	N/A	36.52	37.58	N/A	N/A	N/A	N/A	–
Chicken Standard	N/A	N/A	N/A	N/A	N/A	N/A	35.92	N/A	N/A	–
Camel Milk	N/A	N/A	37.26	N/A	N/A	N/A	N/A	N/A	N/A	–
Bison	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	–
Bison Standard	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	–
Goat Standard	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	–

2.3. Sample preparation and DNA extraction

Four food matrices were obtained for testing: dark chocolate bars, dairy-free muffins, vegan cookies, and orange juice. Ingredient lists are provided for each in supplementary materials (Table S1). Prior to spiking, dark chocolate, dairy-free muffins, and vegan cookies were homogenized separately in large batches to achieve uniform consistency. Whole milk and beef were spiked into each matrix to final concentrations of 0.01, 0.1, 1, 10, 100, 1000, 10⁴, and 10⁵ ppm (Table S2). All spikes were homogenized with the food matrix to a total 5 g in 18 mL of 1% SDS in IKA tube mills at 10,000 rpm for 40 s or until fully homogeneous. A matrix blank consisting of 5 g of the matrix (5 mL for orange juice) and 18 mL of 1% SDS buffer was included for each matrix. Homogenization blanks containing 1% SDS alone were run between each matrix to confirm the absence of cross-contamination during the homogenization process or of the Tube mill itself. All samples were frozen at −20 °C prior to DNA extraction. Spiking experiments were conducted in duplicate across all matrices, using independent experiments performed on different days with separate batches or lot numbers of each matrix.

DNA extraction was carried out per manufacturer's instructions with the QIAGEN DNeasy mini plant kit (QIAGEN, Venlo, The Netherlands) with minor modifications: the mechanical tissue lysis and rupture step was omitted, 300 μL of homogenate was used directly in the extraction procedure, and DNA was eluted in 60 μL of Buffer AE. The DNeasy mini plant kit was selected based on the plant-based composition of the food matrices. Extracted DNA was quantified with a DeNovix spectrophotometer (DeNovix, Wilmington, DE, USA) and normalized to 1 ng/μL prior to PCR. Cross-reactivity experiments in matrices were conducted in the same manner with pork as the spike in cookies and orange juice.

2.4. PCR

PCR conditions were optimized with respect to primer, probe, magnesium, and dNTP concentration, as well as annealing temperature (data not shown). DNA extracted from beef was used in a separate, parallel set of optimization experiments as a control source of concentrated target DNA. Real-time PCR was performed using a Bio-Rad CFX-96 Real-Time PCR Detection System (Bio-Rad Laboratories, Hercules, CA, USA). An internal amplification control (IAC) with primers, probe and synthetic DNA template was included in all reactions (section 2.1; Deer et al., 2010). Each 12.5 μL reaction contained 5 μL analyte DNA and 7.5 μL PCR master mix with final concentrations of: 1× PCR Buffer (Invitrogen, Carlsbad, CA, USA), 5 mM MgCl₂ (Invitrogen, Carlsbad, CA, USA), 500 nM target primer, 200 nM target probe, 1.2 mM dNTP mix (Thermo Fisher Scientific, Waltham, MA, USA), 0.05 U/μL Platinum Taq Polymerase (Invitrogen, Carlsbad, CA, USA), 200 nM IAC primer, 200 nM IAC probe, and 5 × 10⁵ copies IAC template DNA per reaction. Thermal cycling consisted of an initial denaturation step at 95 °C for 5 min followed by 45 cycles of denaturation at 95 °C for 15 s and annealing and extension at 65 °C for 2 min 30 s. Fluorescence was measured at the end of 2 min in the annealing/extension step. All analyte DNA was analyzed in triplicate per standard qPCR practice, with PCR performed three times per matrix for each independent spiking experiment. CT values at each spike concentration are reported as the mean of the six PCR replicates (Table S3).

2.5. Data analysis

Data analysis was performed using CFX Manager software (Bio-Rad, Hercules, CA, USA). C_T values were obtained from the single threshold value calculated by the software and the amplification curves for each reaction. Standard curves were created by plotting the C_T values as a function of the log of the starting quantity of DNA in buffer or the log of ppm spiking level for matrix evaluation. Linear standard curves were analyzed with respect to statistical R² values, range of linearity, sensitivity, and assay efficiency.

2.6. Sanger sequencing for amplicon verification

PCR reactions were performed as described above, with water substituted for the target probe and the internal control primers, probe, and template. Template DNA from milk, beef, and all cross-reactivity samples was sequenced to verify the intended amplicon or identify unintended secondary amplification products. PCR cleanup, cycle sequencing reactions, sequencing reaction cleanup, and Sanger sequencing were conducted according to FDA-validated sequencing protocols (Eischeid et al., 2016; Handy et al., 2011; United States Food and Drug Administration, 2024) with the exception that milk assay primers were substituted for M13 primers in the cycle sequencing reaction.

3. Results and discussion

3.1. PCR performance

To evaluate the assay under optimized PCR conditions, ten-fold serial dilutions of target DNA (both beef and milk) in TE buffer were analyzed across twelve replicates with respect to dynamic range, LOD, LOQ, R² values, standard deviation, and reaction efficiency (Fig. 1, Table 2). Per guidelines for real -time PCR methods, standard curves should be linear over at least three orders of magnitude, ideally extend over 5–6 orders of magnitude within the detection range, have R² values ≥0.98, and have reaction efficiencies between 90 and 110% The limit of detection (LOD) is defined as the lowest concentration at which detectable amplification is observed across all twelve replicates. The limit of quantitation (LOQ) is defined as the lowest concentration at which detectable amplification is observed across all twelve replicates with a standard deviation of no more than 0.5 C_T values (Broeders et al., 2014; Bustin et al., 2009; United States Food and Drug Administration, 2023).

Fig. 1.

Average C_T values of twelve replicates for milk and beef ten-fold serial dilutions (1 ng/μL- 1 × 10⁻⁶ ng/μL (1 fg/μL)) in TE buffer. Error bars show 1 standard deviation above and 1 standard deviation below the mean. Reactions contain 5-5 × 10⁶ fg total DNA. Efficiency is calculated from the equation E = 10^–1/m-1, where E is efficiency and m is the slope of the dilution curve.

Table 2.

C_T values across twelve replicates of milk and beef ten-fold serial dilutions (1 ng/μL- 1 × 10⁻⁶ ng/μL (1 fg/μL)) in TE buffer.

MILK
Starting Quantity (fg)	CT Rep 1	CT Rep 2	CT Rep 3	CT Rep 4	CT Rep 5	CT Rep 6	CT Rep 7	CT Rep 8	CT Rep 9	CT Rep 10	CT Rep 11	CT Rep 12	Average CT	Std Dev
5,000,000	23.07	22.91	23.15	23.06	23.16	22.87	22.89	22.98	23.03	23.02	23.06	23.11	23.03	0.09
500,000	26.62	26.46	26.49	26.46	26.42	26.48	26.56	26.32	26.55	26.60	26.42	26.59	26.50	0.08
50,000	30.27	30.24	30.31	30.04	30.39	30.69	30.26	30.33	30.21	30.35	30.44	30.21	30.31	0.15
5000	34.98	33.55	33.99	34.16	34.88	33.48	33.95	33.04	33.47	33.46	32.91	35.00	33.91	0.70
500	(−)	35.90	(−)	(−)	(−)	(−)	35.70	(−)	34.86	35.80	36.30	35.78	35.72	0.43
BEEF
Starting Quantity (fg)	CT Rep 1	CT Rep 2	CT Rep 3	CT Rep 4	CT Rep 5	CT Rep 6	CT Rep 7	CT Rep 8	CT Rep 9	CT Rep 10	CT Rep 11	CT Rep 12	Average CT	Std Dev
5,000,000	19.05	19.07	19.13	19.11	19.04	19.07	19.07	19.07	19.07	18.97	18.89	19.02	19.05	0.06
500,000	22.75	22.71	22.66	22.62	22.74	22.76	22.71	22.66	22.60	22.60	22.63	22.68	22.68	0.05
50,000	26.36	26.07	26.04	26.06	26.03	26.14	26.14	26.11	26.11	26.14	26.07	26.22	26.12	0.09
5000	29.64	29.98	29.94	29.94	29.92	29.66	29.49	30.02	30.04	29.66	29.72	29.66	29.80	0.18
500	33.50	34.36	33.25	33.04	33.55	32.96	33.64	33.01	33.10	33.61	33.21	34.32	33.46	0.45
50	(−)	38.74	36.16	36.22	36.12	(−)	(−)	36.20	35.60	36.07	(−)	42.79	37.24	2.28

With respect to linearity, the assay performed within acceptable parameters for both beef and milk DNA. The standard curves generated from DNA extracted directly from both beef and milk demonstrate the PCR is linear five orders of magnitude for milk and six orders of magnitude for beef with R² values of ≥0.99 with efficiencies of 102% and 89%, respectively. The beef efficiency of 89% falls marginally below the recommended 90% lower threshold. The LOD of the assay is 1000 fg/μL (0.001 ng/μL) for milk and 100 fg/μL (0.0001 ng/μL) for beef. The LOQ for milk is 10,000 fg/μL (0.01 ng/μL) and 100 fg/μL (0.0001 ng/μL) for beef (Fig. 1, Table 2). These LODs are consistent with, or more sensitive than, other previously published bovine PCR assays that reported absolute LOD values (Bottero et al., 2002; Giglioti et al., 2022; Rentsch et al., 2013; Villa et al., 2018).

3.2. Performance in food matrices

3.2.1. Dark chocolate

Dark chocolate was selected as one of the evaluation matrices not only for its relevance to undeclared milk recalls (Bedford et al., 2017; Sharma et al., 2023), but for the well-documented challenges it presents for DNA purification and amplification (Gryson et al., 2007; Rosman et al., 2016), making it an ideal candidate for robustness testing. Chocolate is known to contain PCR inhibitors including polyphenols, lipids, and polysaccharides that can interfere with PCR amplification (de Oliveira et al., 2022; Gryson et al., 2007; Pirondini et al., 2010; Sajali et al., 2018). Consistent with this, initial PCR testing of dark chocolate samples spiked with milk yielded no amplification of either the target DNA or the internal control, indicating the presence of inhibitors. Dilution of extracted samples to 1 ng/μL total DNA prior to PCR enabled amplification, but detection was limited to the 10⁵ and 10⁴ ppm spikes. In comparison, the dark chocolate samples spiked with beef, also normalized to 1 ng/μL, demonstrated better sensitivity by two orders of magnitude with detection at 100 ppm and a PCR efficiency of 95% (Fig. 2a).

Fig. 2.

Whole milk and beef standard curves spiked into a.) dark chocolate, b.) vegan cookies, c.) dairy-free muffins, and d.) orange juice in concentrations 0.01 to 10⁵ ppm. Average C_T values over six PCR experiments: three each for the first and second spiking experiment with standard deviation error bars.

While dilution of the extracted samples allowed for amplification, sensitivity for both milk and beef in dark chocolate remained lower comparable to the other matrices, suggesting that inhibitors can still decrease assay sensitivity and efficiency even if amplification is achieved (Wilson, 1997). To further address inhibition, bovine serum albumin (BSA) was evaluated as an adjuvant as several studies have reported its ability to improve sensitivity in the presence of multiple inhibitors (Bessière et al., 2023; Farell & Alexandre, 2012; Kreader, 1996). BSA improved beef detection by an order of magnitude, with detection down to 10 ppm in dark chocolate, but had no effect on milk detection sensitivity (Fig. S1). The contrasting response to BSA- improved detection for beef and no effect on milk- may suggest that PCR inhibition is not the only factor contributing to the limited sensitivity observed for milk. The direct comparison with beef suggests that low copy number of target DNA in milk may also be a factor.

3.2.2. Baked goods (vegan cookies and muffins)

Milk spikes in baked goods were detected in the range of 10³ to 10⁵ ppm in both cookies and muffins, with PCR efficiencies of 99% and 122%, respectively (Fig. 2b and c). Consistent with findings from dark chocolate spike experiments, the detection of beef in baked goods was approximately three orders of magnitude more sensitive than the detection of milk. PCR efficiencies for beef spike detection were 107% and 105% for cookies and muffins, respectively, well within the accepted range for real-time PCR assays (Fig. 2b and c).

This comparison offers further support that limited sensitivity and elevated PCR efficiency observed in milk detection in muffins could be attributed to low copy number of target DNA in milk. Low target copy number can both reduce sensitivity and inflate apparent PCR efficiency, a pattern consistent with the improved assay performance observed when a more concentrated source of bovine DNA is used in spiking experiments. It is established that low copy number reactions are more susceptible to high C_T variance among replicates due to PCR stochasticity and the higher impact of Poisson distribution sampling error at low target concentrations (Bustin et al., 2009; Hassibi et al., 2004; Ruiz-Villalba et al., 2021). This variance can reduce the linearity of mean C_T values (R² < 0.99), broaden their distribution, decrease the slope of the standard curve, and therefore produce a spurious increase in calculated reaction efficiency. The absence of this trend in beef spike detection, where bovine DNA is more concentrated, might suggest the stochastic effects of lower copy number as a possible explanation for high assay efficiency in milk detection. Given its relatively lower sensitivity and higher efficiency, this method may be better suited as a qualitative screening tool in certain matrices, such as the muffin matrix demonstrated here.

3.2.3. Orange juice

Milk spikes in orange juice were detected in the range of 10² to 10⁵ ppm, with a PCR efficiency of 107%. Beef spikes in the same matrix were detected across a broader range of 10⁻¹ to 10⁴ ppm, with a PCR efficiency of 104% (Fig. 2d). Both values fall within the accepted range for real-time PCR assays, indicating the absence of any significant PCR inhibition. Although the reduced sensitivity observed for milk detection could suggest DNA degradation attributable to the acidic nature of orange juice, the sensitivity of beef detection in the same matrix was unaffected and does not support this explanation. Instead, the data agree with a consistent pattern in which beef detection is approximately three orders of magnitude more sensitive than milk detection. Given similar reaction efficiencies for both targets, differences in copy number between the two types of tissue offer a plausible explanation for the observed disparity in detection sensitivity.

3.2.4. Comparison of assay performance across all matrices

This study evaluated the detection of liquid milk as an allergenic ingredient in complex food matrices. Standard curves for beef and milk spiked into food matrices (Fig. 2) demonstrate reliable detection of bovine milk in all evaluated matrices except dark chocolate, where internal controls indicated the presence of PCR inhibitors, precluding comparable data analysis for this matrix. Assay performance parameters for each matrix are summarized in Table 3. Excluding dark chocolate, the lowest detected quantity of milk was 100 ppm in orange juice and 1000 ppm in vegan cookies and muffins. Although assay sensitivity for the detection of milk is relatively low compared to previously FDA-validated animal tissue qPCR assays, the sensitivity for the detection of beef is three orders of magnitude greater than for milk in all evaluated matrices and is comparable to those validated assays (Eischeid, 2016; Eischeid, 2019; Eischeid, 2022; Eischeid & Kasko, 2015; Eischeid & Stadig, 2018; Puente-Lelievre & Eischeid, 2021). The consistent difference in detection sensitivity between beef and milk across all matrices suggests that a lower concentration of target DNA in milk relative to beef is a plausible explanation for the reduced sensitivity observed in milk detection.

Table 3.

Summarizes slope, efficiency, R², the lowest ppm spike detected, and whether the presence of inhibitors was indicated for each matrix with both milk and beef spikes.

	Milk					Beef
Matrix	Slope	Efficiency	R2	lowest ppm detected	Inhibitors indicated (Y/N)	Slope	Efficiency	R2	lowest ppm detected	Inhibitors indicated (Y/N)
Dark Chocolate	−3.45	N/A	N/A	1.00E+04	Y	−3.45	95%	0.9879	100	Y
Vegan cookies	−3.34	99%	0.9904	1000	N	−3.17	107%	0.9925	10	N
Muffins	−2.88	122%	0.9827	1000	N	−3.21	105%	0.9971	1	N
Orange Juice	−3.18	107%	0.9991	100	N	−3.23	104%	0.9991	1	N

3.3. Cross-reactivity

Cross-reactivity was evaluated with milk and meat samples from other relevant species (Table 1). Cross-reactivity was observed with water buffalo milk six C_T values later than bovine milk at the same concentration and with weaker amplification (Table 1, Fig. 3). This is likely attributable to sequence similarity with Bos taurus at the primer and probe binding sites, differing by only three base pairs in the probe and one base pair in the reverse primer (Fig. 4). This cross-reactivity is not considered problematic as buffalo milk is significantly more expensive than bovine milk and is used almost exclusively for high-end dairy products such as artisanal cheeses, yogurt, and ghee, making it unlikely to appear as an undeclared ingredient in foods (Sindhu & Arora, 2011). Additionally, low-level cross-reactivity was observed for select meat samples — namely lamb, goat, and pork — with average C_T values of approximately 29 or later (Table 1, Fig. 3). This is consistent with findings from other studies using mitochondrial gene targets, which similarly report some degree of cross-reactivity with pork, goat, or sheep (Kesmen et al., 2009; López-Andreo et al., 2005; Villa et al., 2020; Zhang et al., 2007).

Fig. 3.

Average C_T vs Average End RFU (average relative fluorescence unit (RFU) value reached at the plateau phase of an amplification curve) for cross-reactivity samples. In this plot, a strong positive signal is indicated by higher end RFU and lower C_T values. The end RFU provides a second metric to measure positive signal than C_T values alone. Bovine samples include beef, beef genomic standard, and whole milk.

Fig. 4.

Alignment of Bos taurus (cow), Bubalus bubalis (water buffalo), and probe/reverse primer binding sites. The three base pair mismatches are marked by red arrows. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

4. Conclusions

This study describes the development and evaluation of a real-time PCR assay targeting the COI mitochondrial gene of Bos taurus for the detection of liquid bovine milk as an allergenic ingredient in complex food matrices. The assay demonstrates reliable detection of bovine milk in three of four evaluated matrices with the lowest detected concentration of 100 ppm (0.01%) orange juice and 1000 ppm (0.1%) in vegan cookies and muffins. PCR inhibition limited detection in dark chocolate. Extensive steps to mitigate the effects of inhibitors were not explored within the scope of this study.

A key finding of this study is that beef detection is consistently three orders of magnitude more sensitive than milk detection regardless of matrix type or the presence of PCR inhibitors. This pattern suggests that the low concentration of bovine DNA in liquid milk — rather than a fundamental limitation of the assay design — is a plausible explanation for the reduced sensitivity observed in milk detection.

Cross-reactivity with water buffalo milk was observed but is not considered problematic given the significant price difference between bovine and buffalo milk and the limited use of buffalo milk as a food ingredient. Low-level cross-reactivity with lamb, goat, and pork is consistent with findings from other studies using mitochondrial gene targets and should be considered when interpreting results in the context of complex matrices containing meat.

Despite sensitivity limitations, this assay consistently detects bovine milk down to 0.1% in food matrices and may serve as a suitable screening tool for milk present as a more abundant ingredient, rather than for the detection of trace allergen contamination. It may also serve as a complementary species identification tool in cases where immunoassay cross-reactivity between milk proteins of different ruminant species limits accurate allergen species identification. Future work to improve milk detection sensitivity would further support the utility of this method for regulatory applications.

CRediT authorship contribution statement

Sarah R. Stadig: Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Anne C. Eischeid: Writing – review & editing, Supervision, Methodology, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material

Data availability

Data will be made available on request.