Four electronic databases were used to retrieve papers on the use of feed additives for dairy cattle with an impact on CH₄ emissions: PubMed, CAB Abstracts, Web of Science, and Scopus (accessed on 30^th July 2024). The search string applied was as follows: (feed additives OR supplementation OR animal feed OR essential oil OR PUFAs OR polyunsaturated fatty acids OR 3-NOP OR 3-nitrooxypropanol OR monensin) AND (animal husbandry OR farm OR farming) AND (dairy cattle OR cow) AND (Methane OR emission OR GHG OR greenhouse gas OR CH4).

The bibliographic search was restricted to studies published between 2001 and 2024. Studies published before 2001 were excluded because diet composition and formulation were described using different parameters (NRC, 2001). The statements reported in the Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 were followed throughout the study (Page et al., 2021).

Articles meeting the following inclusion criteria were selected for the meta-analysis:

Although a limited number of studies have attempted to predict enteric methane emissions using in vitro models (Hindrichsen et al., 2004), only in vivo studies were included in this meta-analysis. For studies employing a Latin square or cross-over design, the total sample size for each treatment was considered in the methane emission calculations, and study duration was reported as the sum of the adaptation period and the experimental period.

The data extraction process addressed enteric methane emissions expressed as g/day, g/kg dry matter intake, g/kg milk produced, and g/kg energy-corrected milk in treated and control groups. Details of the data extraction and management process are summarized below.

When selected studies provided only the standard error of the mean (SEM) for enteric methane emission values, the standard deviation (SD) was calculated using the following Equation 1:

where n=sample size.

When a 95% confidence interval (CI) was available for an absolute effect measure (e.g., standardized mean difference, risk difference, or rate difference), or when standard error differences (SED) were provided and the number of observations in the control and treatment groups was similar, SEM was calculated using the following Equations 2, 3:

When standard error was reported on the log-transformed scale and non-log-transformed (back-transformed) means were provided, the original standard error was calculated using the following Equation 4:

In addition, when methane emissions were reported as L/d or L/kg, data were converted using the following Equation 5:

where 0.000668 g/cm³ represents methane density at 20 °C.

When data were not provided in the retrieved papers or were depicted only graphically, the corresponding authors of each study were contacted by email; however, no responses were received.

Data were analyzed using the log response ratio (lnRR) as the effect size, the I² statistic for heterogeneity, and three-level random-effects models. The lnRR is based on the transformation of the mean difference into a percentage measure, as relative differences between control and treated groups allow comparability across studies reported in different units. This approach is robust across measurement scales (g/day, g/kg DMI, g/kg milk, and g/kg ECM) and reduces dependence on absolute methane production levels, thereby improving comparability between studies. The lnRR approach is widely recommended in ecological and nutritional meta-analyses to address bias arising from baseline differences (Laujeunesse, 2011, Laujeunesse, 2015, Livingstone et al, 2015). The lnRR calculation and its percentage transformation are reported in the following Equations 6, 7:

In lnRR variance calculation for paired studies, the Pearson correlation coefficient (r) was set to 0.5. lnRR variance calculation for unpaired and paired studies are reported in the following Equations 8–10:

where SD = standard deviation, n = sample size.

where SD = standard deviation, n = sample size.

Heterogeneity (I²) was calculated using the following equation (Higgins et al., 2003):

where Q= Cochran’s Q statistic and df= degrees of freedom.

A three-level random-effects model was applied to estimate variability between studies (τ²) and within studies (multiple effect sizes from the same study), providing more reliable heterogeneity estimates (Konstantpoulos, 2011; Van Den Noortgate et al., 2013). Standard errors and p-values for model covariates were corrected using cluster-robust variance estimation (CR2) with the Satterthwaite method to account for the small number of studies and dependent effect sizes (Tripton, 2015; Satterthwaite et al., 1946). Leave-one-out diagnostics were performed to assess whether pooled estimates were driven by individual studies (Viechtbauer and Cheaung, 2010).

When the number of independent studies for a specific feed additive was fewer than five (e.g., monensin, n = 4), multilevel models were considered statistically unreliable (Tripton, 2015; Hedges et al., 2010). In these cases, only descriptive forest plots of lnRR (%) were reported to illustrate the direction and magnitude of effects. Additionally, when covariate-adjusted model p-values were not statistically significant, forest plots of lnRR (%) with and without covariate correction were reported for descriptive purposes.

Covariates included in the model formulation to explain heterogeneity were study duration (days), methane measurement technique (SF₆, respiration chambers, or GreenFeed), experimental design (randomized or cross-over), crude protein (CP), and neutral detergent fiber (NDF) content of the basal diet. Although dosage is recognized as a key factor in feed additive efficacy, it was not included as a covariate because it was reported using heterogeneous units across studies (e.g., mg/kg DMI, g/day per cow, percentage of diet), precluding standardization (Beauchemin et al., 2020).

Statistical significance was set at p< 0.05. All statistical analyses were performed using R 4.5.0 (R Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org/ accessed on September 17^th, 2025) with the packages “dplyr”, “readr”, “writexl”, “metafor”, “readxl”, “ggplot2”, “stringr”, “cowplot”, “tibble”, “clubSandwich”, and “purr”.

A total of 4,828 studies were identified in the primary search across the four selected databases: 871 from PubMed, 122 from CAB Abstracts, 1,179 from Web of Science, and 2,654 from Scopus. Duplicates, reviews, and meta-analyses were removed using Zotero, resulting in 3,416 studies for title screening. Of these, 868 articles were selected for abstract screening, and 796 were excluded, leaving 72 articles to be assessed for eligibility. Based on the inclusion and exclusion criteria, 38 additional papers were excluded for the following reasons: 13 contained a control diet with feed additives or did not report the composition of the control diet; five used supplements not addressed in this meta-analysis; three did not report a method for measuring methane; and 17 did not meet other inclusion criteria. Ultimately, 34 studies met the inclusion criteria, from which 84 methane measurements were extracted: 45 associated with PUFAs supplementation, 23 with 3-NOP, 12 with essential oils (EO), and four with monensin. The PRISMA flow chart detailing the search and selection process is presented in Figure 1, based on Page et al. (2021).

PUFAs supplementation consistently reduced enteric methane emissions across all four metrics evaluated in this meta-analysis: absolute emissions (g/day), emissions normalized to dry matter intake (g/kg DMI), emissions per unit of milk yield (g/kg milk), and emissions per unit of energy-corrected milk (g/kg ECM). Uncorrected random-effects meta-analytic estimates based on lnRR revealed statistically significant reductions for each metric (g/day: overall REML = -18.46, 95% CI: -23.30 to -13.62; g/kg DMI: overall REML = -10.77, 95% CI: -15.37 to -6.18; g/kg milk: overall REML = -18.57, 95% CI: -24.95 to -12.19; g/kg ECM: overall REML = -12.58, 95% CI: -17.95 to -7.21), as illustrated in Figures 2–5. These results confirm the methane-mitigating potential of PUFAs in dairy cattle, with effect magnitudes varying by metric but remaining consistently negative and statistically significant.

High heterogeneity characterized all datasets (I² ranging from 85.78% to 94.54%), indicating substantial variability in effect sizes both within and between studies. For methane emissions expressed in g/day, variance was predominantly within studies; for emissions expressed in g/kg DMI, variance was primarily between studies; and for emissions expressed in g/kg milk and g/kg ECM, variance components were comparable across within- and between-study levels. Heterogeneity tests were highly significant across all metrics (all p< 0.0001).

Multilevel models incorporating CR2 correction were applied to address potential sources of heterogeneity; however, no covariates exerted statistically significant effects on PUFAs effect sizes (covariate tests: p = 0.3829 for g/day; p = 0.966 for g/kg DMI; p = 0.6465 for g/kg milk; p = 0.4448 for g/kg ECM). These results indicate that the study-level and dietary factors examined did not substantially explain the observed variability.

Leave-one-out (LOO) validation analyses further underscored the robustness of the models while identifying influential studies. For methane emissions expressed in g/day, studies 10, 15, and 24 most strongly influenced fit indices (AIC, BIC, and log-likelihood), whereas studies 22, 30, and 34 had negligible effects. Similar patterns were observed for other metrics: studies 10 and 24 for g/kg DMI; studies 2, 9, 14, 23, and 24 for g/kg milk; and studies 1, 3, 4, 9, 11, 14, 23, and 24 for g/kg ECM. Importantly, exclusion of these influential studies did not alter the direction or statistical significance of pooled effects, confirming the stability of the meta-analytic conclusions.

Complete model specifications, variance components, LOO diagnostics, and corrected forest plots are reported in the Supplementary Material (Supplementary Tables S2–S9; Supplementary Figures S1-S4). Overall, these results highlight PUFAs as a reliable nutritional strategy for mitigating enteric methane emissions, although the persistent heterogeneity observed across studies underscores the need for standardized reporting and longer-term trials.

3-nitrooxypropanol (3-NOP) supplementation reduced enteric methane emissions across all four outcome metrics evaluated in this meta-analysis: absolute emissions (g/day: overall REML = −22.93, 95% CI: −29.04 to −16.82), emissions normalized to dry matter intake (g/kg DMI: overall REML = −16.19, 95% CI: −23.06 to −9.31), emissions per unit of milk (g/kg milk: overall REML = −10.86, 95% CI: −23.92 to 2.19), and emissions per unit of energy-corrected milk (g/kg ECM: overall REML = −16.54, 95% CI: −22.92 to −10.16). Uncorrected random-effects models based on lnRR indicated statistically significant reductions for g/day, g/kg DMI, and g/kg ECM, whereas the reduction for g/kg milk was not statistically significant, primarily due to the limited number of available studies. After CR2 correction, confidence intervals widened and statistical significance was not consistently retained across metrics. Nevertheless, pooled lnRR estimates remained negative for all outcomes, confirming a consistent methane-mitigating effect of 3-NOP (Figures 6–9; Tables 2, 3; Supplementary Tables S10–S13).

All datasets exhibited high heterogeneity (I² ranging from 83.28% to 97.99%), reflecting substantial variability in methane responses within and between studies. For methane emissions expressed in g/day and g/kg DMI, variability was predominantly between studies, suggesting that differences in experimental conditions, diets, and management practices were primary drivers of effect size dispersion. For methane emissions expressed as g/kg milk and g/kg ECM, heterogeneity was more evenly distributed across within- and between-study levels, consistent with the smaller number of comparisons and the greater sensitivity of intensity-based metrics to production-related variation. Heterogeneity tests were statistically significant for g/day and g/kg DMI models (all p< 0.0001). Covariate tests (g/day: p = 0.0057; g/kg DMI: p< 0.0001) indicated that included moderators only partially explained the observed variability. For g/kg ECM, the heterogeneity test was not significant (p = 0.1169), whereas the covariate test remained significant (p< 0.0001), suggesting a meaningful contribution of moderators.

Multilevel models with CR2 correction identified crude protein concentration in the basal diet as the most consistent covariate associated with methane reduction magnitude for g/day, and both crude protein concentration and study design (randomized vs. non-randomized) for g/kg DMI. These findings indicate that dietary composition exerts a stronger influence on the methane-mitigating response to 3-NOP than animal-related factors or other design features. Leave-one-out analyses showed that only a limited number of studies influenced model fit indices, and their exclusion did not change the direction or magnitude of pooled effects, supporting the robustness of the findings.

Taken together, these results demonstrate that 3-NOP consistently reduces enteric methane emissions across all expression metrics. However, substantial residual heterogeneity indicates that diet- and design-related covariates only partially explain response variability. Notably, the limited data available for g/kg milk precluded multilevel modeling and LOO analyses; therefore, results for this metric are presented for descriptive purposes only.

Essential oil (EO) supplementation showed no statistically significant reduction in enteric methane emissions across all four outcome metrics evaluated in this meta-analysis: absolute emissions (g/day: overall REML = −1.58, 95% CI: −4.71 to 1.55), emissions normalized to dry matter intake (g/kg DMI: overall REML = −1.13, 95% CI: −3.70 to 1.43), emissions per unit of milk (g/kg milk: overall REML = −1.17, 95% CI: −2.96 to 0.62), and emissions per unit of energy-corrected milk (g/kg ECM: overall REML = −2.72, 95% CI: −6.79 to 1.34).

Heterogeneity varied across datasets, ranging from moderate (I² = 54.07%–60.13% for g/day, g/kg DMI, and g/kg ECM) to absent (I² = 0% for g/kg milk). For methane emissions expressed in g/day and g/kg DMI, variability was predominantly within studies, indicating that differences in experimental units or measurement protocols within individual experiments contributed more to effect size dispersion than between-study differences. For methane emissions expressed in g/kg milk (n = 4 studies) and g/kg ECM (n = 5 studies), the limited number of studies precluded reliable multilevel modeling, CR2 application, and leave-one-out (LOO) diagnostics, consistent with established guidelines for meta-analytic robustness. The g/kg milk metric exhibited complete consistency across the limited number of studies (n = 4). Heterogeneity tests were not statistically significant for g/day (p = 0.0683) and were significant for g/kg DMI (p = 0.0144).

Multilevel models with CR2 correction applied to the g/day and g/kg DMI datasets identified no statistically significant covariates influencing effect size (covariate tests: p = 0.8476 and p = 0.7691, respectively). Tested moderators, including dietary composition, animal characteristics, and study design, did not meaningfully account for observed variability, suggesting that unmeasured factors or inherent biological variability may contribute to the lack of EO efficacy.

LOO validation analyses for the g/day and g/kg DMI models revealed that a small number of studies exerted measurable influence on fit indices, particularly study 6 across both metrics and study 16 for g/kg DMI. Exclusion of these influential studies improved model fit but did not alter the direction, magnitude, or non-significance of pooled lnRR estimates. Other studies showed negligible impact, confirming overall model stability despite moderate within-study variability. Heterogeneity patterns remained consistent following sequential exclusions.

Taken together, these results demonstrate that EO supplementation does not yield statistically significant reductions in enteric methane emissions across any expression metric in dairy cattle. The consistent lack of effect across datasets, combined with predominantly within-study variability and the absence of influential covariates, suggests that EO has limited methane-mitigating potential under the experimental conditions represented in the literature. The small number of studies for milk-based intensity metrics (g/kg milk and g/kg ECM) further limits inferential power; therefore, results for these metrics are presented for descriptive purposes only. Complete model specifications, LOO diagnostics, and forest plots are provided in the Supplementary Material (Supplementary Figures S8–S13; Supplementary Tables S14–S17).

Only three studies on monensin met the inclusion and exclusion criteria, with one study providing data at two time points, resulting in a total of four data points. While these outcomes are informative for descriptive purposes, they are insufficient to support reliable statistical analysis. Notably, none of the studies measured enteric methane emissions in relation to the productive performance of dairy cows. Consequently, data were available only for enteric methane emissions expressed as g/day (n = 4) and g/kg DMI (n = 2). Heterogeneity was moderate for g/day (I² = 57.47%) but absent for g/kg DMI (I² = 0%).

Due to the very small sample size, model formulation, CR2 correction, and LOO validation were not feasible. However, forest plots of lnRR (%) are provided in the Supplementary Material as Supplementary Figures S14, S15 for g/day (overall REML = 0.14, 95% CI: −4.71 to 5.00) and g/kg DMI (overall REML = −4.21, 95% CI: −8.55 to 0.14), respectively. These results should be interpreted with caution, as they are primarily descriptive in nature.

This meta-analysis reveals distinct patterns in the methane-mitigating efficacy of nutritional additives in dairy cattle. PUFAs and 3-NOP consistently reduced methane emissions across all emission metrics, whereas essential oils (EO) and monensin exhibited limited and inconsistent effects. PUFAs and 3-NOP operate through different mechanisms. PUFAs primarily modify rumen microbial ecology and fermentation pathways due to their toxicity toward cellulolytic bacteria responsible for acetic acid production. Acetic acid is, in turn, one of the primary substrates utilized by methanogens as a hydrogen donor. In contrast, 3-NOP directly inhibits methyl coenzyme M reductase, a key enzyme in the methanogenesis pathway. Both additives produced biologically meaningful reductions in methane emissions, whether expressed as absolute output or relative to production performance parameters. In contrast, the effects of EO and monensin varied widely. For EO, this variability is likely attributable to differences in EO blend composition, as several oils are known to possess antibacterial activity. For monensin, variability likely reflects the influence of rumen microbial populations and dietary substrates. These findings highlight the importance of additive-specific characteristics and contextual factors in determining methane mitigation potential.

The high heterogeneity observed across analyses underscores the influence of unmeasured experimental variables. Diet composition emerged as a key moderator, particularly crude protein concentration in relation to 3-NOP efficacy. This aligns with established principles of rumen biochemistry. Indeed, higher protein levels may increase hydrogen availability for alternative metabolic pathways (e.g., the propionate pathway), thereby enhancing the inhibitory effect of 3-NOP on methanogenesis.

PUFAs responses similarly reflect interactions driven by lipid profiles. The findings of this meta-analysis align with previous meta-analyses and experimental studies reporting 10%–25% reductions in enteric methane emissions, depending on PUFAs dose and type (Shingfield et al., 2013; Hristov et al., 2013a; Martin et al., 2010). In contrast, EO and monensin lacked significant covariates, reflecting their broader mechanistic variability. In particular, EO bioactive compounds (e.g., cinnamaldehyde and eugenol) exhibit compound-specific potency against methanogens, while monensin primarily shifts propionate production in high-forage diets.

These findings carry important practical implications for dairy management and emission reduction policies. PUFAs supplementation can be flexibly integrated into existing rations and offers established production benefits. Similarly, 3-NOP provides targeted, dose-dependent methane reduction compatible with precision-feeding strategies. However, the temporal decline in 3-NOP efficacy observed in longer-duration trials—likely due to microbial adaptation—highlights the need for strategies such as intermittent dosing or combined mitigation approaches.

EO and monensin, despite lower overall reliability, may serve niche applications when specific formulations align with particular dietary conditions. The variability in EO effects can be attributed to distinct combinations of bioactive compounds present in EO formulations, which differ in their mechanisms of action on rumen methanogenesis, microbial composition, and fermentation pathways (Patra and Yu, 2012; Benchaar and Hassanat, 2024). Certain EO compounds, such as garlic derivatives and cinnamaldehyde, can inhibit methanogenic archaea or modulate volatile fatty acid production, but their efficacy is highly context-dependent (Patra and Yu, 2012). The two studies showing statistically significant effects of EO on enteric methane emission reduction were Khurana et al. (2023); study ID: 18), which evaluated an EO mixture containing garlic–citrus extract, and Tondini et al. (2024); study ID: 31), which used a mixture of capsicum oleoresin, eugenol, and cinnamaldehyde. Monensin can reduce enteric methane through its ionophore activity, which alters rumen fermentation and inhibits gram-positive bacteria, thereby increasing propionate production at the expense of acetate and hydrogen availability and reducing methane formation. However, the magnitude of monensin’s effect depends strongly on diet composition, animal type, and rumen microbial adaptation (Beauchemin et al., 2008; Min et al., 2022).

Several key knowledge gaps remain, including the lack of standardized additive dosing protocols, harmonized methane measurement methodologies, and evaluations of long-term treatment effects (>90 days) that account for rumen microbiome and physiological adaptation dynamics. Trial duration is particularly critical, as short-term studies may overestimate sustained mitigation potential. Diet formulation—particularly protein-to-neutral detergent fiber (NDF) ratios—should inform additive selection, as both starch and NDF contents have been identified as influential factors (Kebreab et al., 2023). In the present study, NDF was included in the model formulation, whereas starch was excluded due to the limited number of studies reporting starch values and its multicollinearity with other dietary covariates. The systematic organization of experimental trials addressing these identified gaps would improve the refinement and implementation of feed additive strategies, thereby supporting verifiable methane emission reductions in dairy cattle and enhancing the sustainability of the dairy food system.

Meta-analyses are subject to several limitations, ranging from the literature search process to the conclusions drawn from study outcomes (Walker et al., 2008). These limitations include potential gaps in paper retrieval, the selection of inclusion and exclusion criteria, and issues related to the quality and heterogeneity of included studies (Allen, 2020). This meta-analysis exhibited a high degree of heterogeneity, making model development essential to explore and account for this feature. Heterogeneity was addressed by applying a three-level model that accounted for multiple sources of variability influencing study outcomes. The models identified a significant effect only for 3-NOP, which appeared to be driven primarily by basal diet composition. Previous research has indicated that forage-to-concentrate ratio and study duration can also modulate the effectiveness of 3-NOP (Shilde et al., 2021). Another important consideration concerns study design. Direct comparisons between cross-over and randomized trials are challenging due to the lack of a clearly defined and consistent control group; however, no statistically significant effect of study type was observed in this analysis. Study duration also warrants careful evaluation. Most trials were relatively short, lasting at most 1 month (nine studies for PUFAs, three for 3-NOP, six for EO, and three for monensin, including follow-ups). Only seven studies lasted between 1 and 2 months (four for PUFAs, two for 3-NOP, and one for monensin, including follow-ups), and just eight studies extended beyond 2 months (one for PUFAs, five for 3-NOP, and two for EO, including follow-ups). These limitations were considered in the model formulation. Regarding study duration, the literature clearly highlights that potential long-term side effects of additive supplementation on animal welfare, as well as the economic costs associated with feed additive production and use, are often insufficiently addressed and typically fall outside the scope of studies included in this meta-analysis.

While meta-analyses aim to provide evidence in areas where the literature presents conflicting results, they cannot compensate for a lack of research in a specific field. Consequently, findings should be interpreted cautiously and within the appropriate context. Regular updates and periodic revisions of meta-analyses are essential to incorporate newly published research and maintain relevance (Walker et al., 2008). This is particularly relevant for the use of feed additives to improve sustainability and reduce methane emissions in dairy cattle.

Despite these limitations, meta-analysis plays a valuable role in guiding future research and improving experimental design. One of the primary challenges in traditional reviews is the difficulty of simultaneously evaluating results from multiple studies and drawing conclusions beneficial to the field. Meta-analysis addresses this challenge by complementing qualitative reviews with quantitative data analysis, achieving increased statistical power through aggregation of sample sizes across studies. To the best of the authors’ knowledge, the present work extends previous research by incorporating a large number of in vivo studies and providing an updated assessment of feed additive efficacy in reducing methane emissions from dairy cattle. In addition, this meta-analysis incorporated an in-depth investigation of heterogeneity through model formulation to identify factors significantly influencing effect size and study outcomes. As a result, meta-analysis can reveal modest yet meaningful effects that might otherwise remain undetected, offering insights critical for the design and interpretation of future research trials. In conclusion, meta-analysis applies advanced statistical techniques to synthesize findings and enhance the quality of information derived from multiple bibliographic sources and is therefore regarded as the gold standard in evidence-based scientific research (Seidler et al., 2020).

The results of this study showed that PUFAs supplementation significantly reduces methane emissions across all expression metrics (i.e., g/day, g/kg DMI, g/kg milk, and g/kg ECM) without adversely affecting animal productive performance. Regarding 3-NOP, this additive generally achieved the largest average reduction in methane emissions, although the statistical significance of the effect varied by metric. Crude protein content of the diet emerged as a significant covariate influencing 3-NOP efficacy. Additionally, some evidence suggests a potential attenuation of its mitigating effect over time, potentially due to rumen microbiome adaptation.

Essential oils (EO) produced inconsistent, formulation-specific results and were not statistically significant across methane measurement metrics, whereas evidence for the efficacy of monensin was too scarce to support reliable inference.

Despite these findings, further research employing more detailed and robust study designs is required to fully assess the long-term efficacy of feed additives in mitigating enteric methane emissions from dairy cattle. There is a clear need to establish globally accepted methods for measuring methane emissions, including comparative evaluations of the three approaches considered in this meta-analysis, as well as emerging methods such as sniffer-based technologies. In addition, studies investigating feed additives in animal nutrition should, where feasible, extend beyond a few weeks or months. Finally, dosage reporting should be standardized to facilitate identification of thresholds that optimize methane mitigation without compromising production performance. Addressing these issues would enable more robust conclusions regarding the efficacy of feed additives in reducing methane emissions in dairy cattle and support meaningful comparisons among different mitigation strategies.