Transfer of cannabinoids into the milk of dairy cows fed with industrial hemp could lead to Δ⁹-THC exposure that exceeds acute reference dose

Hemp (Cannabis sativa) production has recently expanded with regulatory changes, introducing many hemp-derived products to the market. Hemp contains cannabinoids that interact with endocannabinoid systems in animals and humans, including psychoactive Δ⁹-THC and pharmacologically active cannabidiol (CBD). Although EU regulations limit Δ⁹-THC in industrial hemp varieties, consumer safety concerns remain when hemp by-products or whole plants are used as animal feed. Transfer of cannabinoids into foods of animal origin is conceivable; however, experimental data on Δ⁹-THC transfer into cow’s milk are scarce, and analytical methods have often failed to differentiate psychoactive Δ⁹-THC from its non-psychoactive precursor Δ⁹-THCA, which can be substantially more abundant. This study investigates the effects of feeding industrial hemp silage to lactating dairy cows, quantifies cannabinoid transfer into milk, evaluates potential animal health effects, and assesses consumer risk. The approach includes LC–MS/MS analyses distinguishing Δ⁹-THC from Δ⁹-THCA across matrices and toxicokinetic modeling to predict transfer of multiple cannabinoids (Δ⁹-THC, Δ⁹-THCV, CBD, CBN, CBDV and Δ⁹-THC metabolites) from feed into milk.

Prior work on cannabinoid transfer into ruminant milk is limited; earlier studies primarily reported Δ⁹-THC and its metabolite THC-COOH in milk after Δ⁹-THC exposure, with limited differentiation from Δ⁹-THCA. Reports on cannabinoid effects on feed intake in animals are mixed, with some studies observing reduced intake after Δ⁹-THC or CBD administration while others showed no change or increases. In humans, Δ⁹-THC and CBD transfer into breast milk after marijuana use has been documented, while 11-OH-THC and THC-COOH are inconsistently detected. The Community method used in the EU for determining Δ⁹-THC in hemp has analytical limitations (incomplete extraction, poor Δ⁹-THC/Δ⁹-THCA separation, potential Δ⁹-THCA decarboxylation), potentially underestimating total THC compared to LC–MS/MS methods. Overall, literature indicates the need for accurate analytics distinguishing Δ⁹-THC from Δ⁹-THCA and more data on cannabinoid transfer and effects in dairy systems.

Design: A controlled feeding experiment with lactating Holstein Friesian cows (n=10 initially; 8 included in final statistics) at the German Federal Institute for Risk Assessment. Two groups: L (low hemp) and H (high hemp). Four sequential periods: control (7 days, hemp-free), adaptation (7 days, low-cannabinoid hemp silage A partially replacing corn silage at 0.31–0.92 kg dry matter (DM) per cow per day), exposure (6 days, cannabinoid-rich silage E at 0.84 kg DM per cow per day for group L and 1.68 kg DM per cow per day for group H), and depuration (8 days, hemp-free). Diets were partially mixed rations with individual concentrate allocation based on milk yield; water ad libitum. Hemp silages: Silage A (whole-plant, variety Ivory) had low cannabinoid levels; Silage E (leaves/flowers/seeds, variety Finola) had higher cannabinoid levels. Measured concentrations (mg/kg DM): Silage A vs E respectively: Δ⁹-THC 58.3 vs 1,254.7; Δ⁹-THCA 7.4 vs 70.1; Δ⁹-THCV 0.2 vs 12.5; CBD 804.7 vs 8,304.1; CBN 9.4 vs 38.9; CBDV 5.1 vs 450.1; others < LOD. Measurements: Daily recording of feed intake, milk yield, respiratory rate, heart rate, body temperature, and behavioral observations. Milk composition (fat, protein, lactose, DM, urea, somatic cell count) at period ends. Milk for cannabinoids collected on days 7, 14–24, 26, 28; evening and following morning milks pooled proportionally and stored at −20 °C. Plasma and faeces collected from two cows per group on days 7, 14, 20, 22. Analytics: LC–MS/MS (HPLC–MS/MS) with isotope-labelled internal standards quantified native cannabinoids: Δ⁹-THC, Δ⁹-THCA, Δ⁸-THC, Δ⁹-THCV, CBD, CBN, CBDV and Δ⁹-THC metabolites (11-OH-Δ⁹-THC, THC-COOH). Matrix-specific extraction and clean-up (milk: acetone extraction, hexane LLE with saturated NaCl, GPC clean-up; plasma: extraction with acetone/acetic acid; β-glucuronidase trials showed losses so non-cleavage data used; feed/faeces: Soxhlet acetone extraction and GPC). LODs: milk 0.01–0.4 ng/ml, plasma 0.1–0.5 ng/ml, feed/faeces 0.2–2.0 ng/g DM. Recoveries >85%, CV <15%. Statistics: Non-parametric analyses due to non-normality. Kruskal–Wallis tests across six groups (combined periods and treatments), with post hoc Tukey-type multiple comparisons via multcomp (glht). Visualization with violin plots. Modeling: Two-compartment toxicokinetic models. Model A (for Δ⁹-THC and CBD) includes GI tract and plasma, with outputs to milk, faeces, and elimination; parameters include absorbed fraction (F_a), GI passage rate k_p fixed at 1.3 d⁻¹ (~5% h⁻¹), milk transfer rate constant k_m, and elimination rate k_e. Model B (for Δ⁹-THCA, Δ⁹-THCV, CBN, CBDV) simplifies to GI tract and central compartment with outputs to milk and total elimination, using absorbed-bioavailable fraction F_b. Inputs I(t) were calculated from feed intake and silage cannabinoid content; milk amounts computed from concentrations and yields; faecal DM estimated by mass balance and digestibility equations. Model performance assessed by Pearson r or r² against observed milk, plasma, and faeces data.

Animal health and performance: Feeding low-cannabinoid silage A during adaptation showed no significant effects on physiological parameters. Feeding cannabinoid-rich silage E caused significant decreases in feed intake (Kruskal–Wallis χ²=28.9, P<0.001, d.f.=5) and milk yield (χ²=44.5, P<0.001, d.f.=5) from day 2 of exposure in both groups. Respiratory rate (χ²=50.2, P<0.001) and heart rate (χ²=77.4, P<0.001) decreased within hours; some animals exhibited bradypnea and bradycardia. Behavioral changes included pronounced tongue play, increased yawning and salivation, nasal secretions, reddened nictitating membrane, somnolence, and unsteady gait in some high-dose animals; effects resolved within two days after stopping hemp feeding. Milk constituents, body temperature, and body weight were unaffected. Intakes: Average ingested doses (mg/kg BW) during exposure: Group L Δ⁹-THC 1.6±0.3, CBD 10.7±1.9; Group H Δ⁹-THC 3.1±0.7, CBD 20.4±4.4; other cannabinoids ≤1.1. Milk cannabinoid levels: Detected Δ⁹-THC, Δ⁹-THCA, Δ⁹-THCV, CBD, CBN, CBDV in milk during adaptation and exposure. Max concentrations up to 316 µg/kg milk (Δ⁹-THC) and 1,174 µg/kg (CBD); others max: Δ⁹-THCA 1.9 µg/kg, Δ⁹-THCV 8.0 µg/kg, CBN 2.5 µg/kg, CBDV 10.1 µg/kg. On last depuration day, Δ⁹-THC remained detectable (group L 1.4±0.4 µg/kg; group H 5.0±0.6 µg/kg) and CBD (group L 7.0±1.9 µg/kg; group H 16.2±2.6 µg/kg). Δ⁹-THC, Δ⁹-THCV, and CBD showed milk/plasma ratios indicating accumulation in milk (Δ⁹-THC 6–26×, Δ⁹-THCV 3–5×, CBD 11–32× over plasma at exposure end); Δ⁹-THCA did not accumulate; CBDV trend unclear. Modeling: Models reproduced milk data well (r²=0.92). Approximations for plasma and faeces achieved Pearson r of 0.93 and 0.90, respectively, despite limited data. Predictive milk models underestimated exposure-period levels by ~12–26% on average but aligned with observed ranges (Pearson r 0.59–0.88). Modeled fate: Δ⁹-THC output fractions ~77% eliminated, 23% faeces, 0.20% milk; CBD ~64% eliminated, 36% faeces, 0.11% milk. Biphasic elimination observed with rapid decline after cessation followed by slower decline. Transfer rates at steady state (milk/feed µg/µg): Δ⁹-THC 0.20% (about 33–100% higher than prior estimates), Δ⁹-THCA 0.015%, Δ⁹-THCV 0.56%, CBD 0.11%, CBN 0.043%, CBDV 0.0080%. Human exposure and risk: Using EFSA RACE and an ARfD of 1 µg Δ⁹-THC/kg BW, mean and maximum milk Δ⁹-THC concentrations from group L led to ARfD exceedances up to 14-fold for consumers <18 years on average consumption and up to 57-fold at the 95th percentile across groups. For group H exposure levels, ARfD exceedances reached up to 120-fold across all population groups. Even adaptation-period milk led to 1.5-fold ARfD exceedance for infant high consumers (95th percentile). Potential conversion of Δ⁹-THCA to Δ⁹-THC during processing could further increase exposure.

The study demonstrates that cannabinoids from industrial hemp silage are transferred into cow’s milk and can adversely affect dairy cow physiology and behavior, addressing the central question of transfer magnitude and potential health implications. Distinguishing Δ⁹-THC from Δ⁹-THCA was essential, revealing that Δ⁹-THC, despite low transfer percentages, can reach appreciable milk levels due to high feed intake and lipid partitioning into milk fat. Toxicokinetic modeling supports substantial bioavailability with predominant elimination and modest faecal excretion, and indicates biphasic decline in milk after feeding stops. The estimated milk transfer rates for multiple cannabinoids provide a basis for exposure assessment; among them, Δ⁹-THC and CBD are most relevant due to higher silage concentrations and milk accumulation. Consumer exposure scenarios suggest that Δ⁹-THC intake from milk and dairy derived from hemp-fed cows could exceed the ARfD in several groups, raising public health concerns. While Δ⁹-THC likely drives observed cow health effects, contributions from CBD, other cannabinoids, and non-cannabinoid phytochemicals or dietary composition changes cannot be excluded. Further mechanistic understanding of gastrointestinal transformations, metabolism, and interconversion among cannabinoids would refine risk assessments and modeling.

Feeding dairy cows with cannabinoid-rich industrial hemp silage (leaves, flowers, seeds) decreased feed intake and milk yield and adversely affected heart and respiratory rates and behavior, while low-cannabinoid whole-plant silage showed no such effects. LC–MS/MS analytics differentiating Δ⁹-THC from Δ⁹-THCA enabled accurate quantification across matrices. Toxicokinetic modeling estimated milk transfer rates below 1% for all cannabinoids, yet Δ⁹-THC and CBD reached substantial milk levels, with predicted consumer Δ⁹-THC exposures potentially exceeding the ARfD in several populations. The findings highlight the need for cannabinoid analysis of hemp feed materials before use and caution in feeding cannabinoid-rich fractions to dairy cows. Future research should elucidate cannabinoid fate and interconversion in the ruminant GI tract, refine unified mechanistic models including metabolism, expand datasets for cannabinoids such as CBD to enable risk assessment, and explore how plant variety, plant parts, harvest timing, and processing influence transfer and animal health outcomes.

Potential confounding from partial replacement of corn silage with hemp silage altered nutrient profiles and introduced other phytochemicals (terpenes, flavonoids), which may have contributed to observed effects. Behavioral observations were not standardized by an ethogram. Urine cannabinoid data were not usable due to analytical issues after glucuronidase treatment. Sample size for full statistical analysis was limited (eight animals), and plasma/faeces sampling was limited to two cows per group and selected days. The modeling framework lumped multiple elimination processes and lacked explicit interconversion pathways; limited knowledge of cannabinoid transformations in the ruminant gastrointestinal tract constrains mechanistic interpretation. The exposure assessment assumes milk Δ⁹-THC concentrations translate to dairy products without processing losses or gains, and potential decarboxylation of Δ⁹-THCA during processing could change consumer exposure.