1. Introduction
Aflatoxin M1 (AFM1) is the primary hydroxylated metabolite of aflatoxin B1 (AFB1), secreted into the milk of lactating mammals within 12–24 hours of dietary AFB1 exposure
[1][2][7][3][4][5][6]
The thermal stability of AFM1 under conventional pasteurization and ultra-high temperature (UHT) processing underscores the inadequacy of standard milk safety protocols and has stimulated a diverse body of research into post-contamination mitigation strategies
[8][12][13]
The first is a
[16][17][18]
The second is an Analytical Gap: a substantial fraction of published studies fail to report foundational analytical validation parameters, including LOD, LOQ, recovery rates, and matrix effect assessments, that are prerequisite for industrial application and regulatory acceptance [19], [20], [21], [22]. The objective of this systematic review is to critically assess the efficacy, mechanistic classification, and analytical quality of microorganism-based interventions for AFM1 mitigation in dairy matrices, and to provide a structured evidence base for the design of next-generation decontamination technologies.
2. Methodology
This systematic review was designed and reported in accordance with the PRISMA 2020 statement
[25][26]
Inclusion criteria required studies to be original, peer-reviewed, primary research articles that:
1) quantitatively measured AFM1 removal efficiency;
2) employed biological agents as the primary or co-primary decontamination strategy;
3) used a recognized dairy matrix. Studies were excluded if they focused exclusively on physicochemical or thermal treatments without a biological component, or lacked quantitative efficacy data
[27][28]
Review articles, letters, conference abstracts, and book chapters were excluded.
A standardized data extraction template was applied to all included studies, capturing: decontamination agent identity (species, strain, viability status), dairy matrix, intervention methodology, peak AFM1 removal efficiency (%), proposed mechanism of action, experimental conditions (temperature, pH, incubation time), analytical method and instrument, and reporting of critical validation parameters (LOD, LOQ, recovery rate, matrix effect correction). Data were managed in a structured spreadsheet.
Analytical and methodological quality was evaluated using a modified RoB framework adapted from SYRCLE and ARRIVE 2.0 guidelines
[29][30][31]
1. Analytical Validation: whether LOD, LOQ, and recovery rates were explicitly reported.
2. Mechanistic Clarity: whether the study tested for toxin desorption or identified degradation metabolites.
3. Matrix Specificity: whether efficacy data were generated in the target dairy matrix.
4. Procedural Transparency: whether experimental conditions were reported with sufficient detail for independent replication.
Each domain received a risk rating of Low, High, or Unclear.
Studies were categorized as adsorption-based or degradation-based according to explicit mechanistic claims supported by experimental evidence. A study was classified as demonstrating degradation only if it provided mass spectrometric or chromatographic identification of AFM1 transformation products, or demonstrated irreversible loss of AFM1 under desorption-challenge conditions with confirmed non-recovery. Studies claiming degradation without metabolite identification were flagged under the Mechanism Ambiguity Bias category.
3. Results
The systematic search retrieved 499 records (PubMed n=153; Scopus n=143; Web of Science n=109; Google Scholar n=63; Semantic Scholar n=23; OpenAlex n=8). Following deduplication (n=210 removed), title screening (n=136 excluded), abstract screening (n=91 excluded), and full-text review (n=28 excluded — 19 for lacking quantitative AFM1 efficacy data, 9 for absence of original experimental data), 34 studies met all inclusion criteria, as shown in Figure 1
[32][33]
Flowchart illustrating the systematic process of identifying, screening, and determining the eligibility of studies for inclusion in the final review, resulting in 34 included sources for the final analysis, following the PRISMA 2020 guidelines
LAB were the predominant decontamination agents, appearing in 33 of 34 studies, followed by yeasts (n=9),
[34][22][20]
Reported removal efficiencies ranged from 13% to >99%, representing substantial heterogeneity attributable to differences in microbial species and strain, biomass concentration, dairy matrix composition, initial AFM1 concentration, temperature, and pH. Twelve studies reported peak removal exceeding 90%. However, 30 of 34 studies (88.2%) attributed AFM1 removal to physicochemical adsorption to microbial cell-wall components rather than enzymatic degradation of the toxin molecule (see Table 1). Only four studies provided evidence consistent with degradation [23], [46], [47], [62], and of these, only Martínez et al. [23] claimed identification of less-toxic degradation metabolites, though without full structural characterization by tandem mass spectrometry.
AFM1 Removal Efficacy, Proposed Mechanisms, and Key Methodological Limitations Across 34 Included Studies
| Study | Microorganism(s) | Dairy Matrix | Peak Removal (%) | Proposed Mechanism | Key Methodological Limitation |
| [37] | LAB starter cultures | Fermented milk | 13–31 | Adsorption | Low efficacy; toxin dissociation observed during refrigerated storage |
| [38] | L. casei | Reconstituted milk | 75–88 | Binding | Efficacy inversely correlated with initial AFM1 concentration |
| [39] | Native LAB | Raw milk | >50 | Binding/physical | Required centrifugation and filtration; industrially cost-prohibitive |
| [40] | Probiotics | Various milk types | 40–70 | Adsorption | High variability with milk fat content; mechanism undetermined |
| [35] | L. rhamnosus, S. cerevisiae | Milk | >90 | Adsorption | Efficacy contingent on Box–Behnken process optimization |
| [16] | Cottage cheese LAB isolates | Milk | 35–65 | Binding | Strain-specific; binding destabilized under high-acidity conditions |
| [20] | Probiotics | Doogh | 50–65 | Adsorption | Significant alterations to sensory and fermentation profiles |
| [23] | Mixed bacteria and yeasts | Milk | 80–100 | Degradation (claimed) | One of four degradation-claiming studies; metabolites not structurally identified by MS/MS |
| [41] | + LAB | Milk | 60–80 | Biophysical | Multi-step biophysical process; scalability undemonstrated |
| [42] | Probiotic cocktails | Milk | >90 | Adsorption | Non-specific binding of milk macronutrients documented |
| [17] | Non-viable LAB | Frescal cheese | 40–60 | Adsorption | Efficacy dependent on high biomass concentrations (>10⁹ CFU/mL) |
| [43] | Indian Lactobacilli | Simulated GI model | 30–55 | Binding | Significant toxin desorption during simulated gastric phase |
| [18] | High-concentration microbes | Milk | 45–95 | Binding | Requires excessive microbial loads (>10⁹ CFU/mL) for peak efficacy |
| [44] | Cell fractions | Milk | 30–50 | Surface binding | Fractionated cells showed lower efficacy versus intact cells |
| [45] | Inactivated LAB | Milk | 60–85 | Adsorption | Heat-inactivation compromised binding site structural integrity |
| [46] | Synbiotic + ZnO nanoparticles | Milk | >95 | Complexation | Potential cytotoxicity of ZnO nanoparticle residues not assessed |
| [47] | Nanoencapsulation cocktail | Milk | 85–98 | Adsorption | High technical complexity; residual toxin risk unquantified |
| [22] | L. reuteri | Sarshir | 70–90 | Adsorption | Confirmed high desorption rates under acidic and ionic conditions |
| [48] | LAB biofilms | Milk | 60–80 | Bio-sequestration | Industrial biofilm maintenance not feasible at scale |
| [49] | Kefir starter culture | Milk | 40–55 | Adsorption | Fermentation time constraints limit practical applicability |
| [33] | Kefir microbiota | Milk | 30–60 | Binding | High inter-grain variability; standardization unaddressed |
| [50] | Kefir-derived polysaccharides | Milk | 40–55 | Adsorption | Low binding capacity compared to living biomass |
| [51] | Multiple LAB strains | In vitro (milk) | 20–80 | Multi-toxin binding | Non-specific; co-sequestration of fat-soluble vitamins documented |
| [52] | Probiotics | Dairy products | 45–70 | Binding | Post-storage toxin recovery confirmed in multiple product formats |
| [36] | B. bifidum | Skim milk | 55–80 | Adsorption | Efficacy markedly reduced in full-fat matrix |
| [53] | + inulin | Milk | 65–85 | Adsorption | Prebiotic–toxin interactions complicate safety interpretation |
| [19] | L. rhamnosus biofilm | Milk | 80–100 | Adsorption | High removal rate, but biofilm sloughing risk identified |
| [34] | Mixed methods | Cheese | 40–60 | Binding | Incomplete removal in complex solid matrix |
| [54] | Probiotics vs. natamycin | Milk | 50–75 | Adsorption | Microbial binding less stable than chemical natamycin benchmark |
| [55] | L. brevis | Raw milk | 35–55 | Binding | Low efficacy against raw milk contaminants; mechanism uncharacterized |
| [56] | Probiotics | Milk | 50–70 | Adsorption | pH fluctuations triggered measurable toxin release |
| [57] | NRRL B4496 | Milk | 40–65 | Adsorption | Reversibility confirmed by wash-step desorption experiments |
| [21] | Novel biological agents | Milk | 70–95 | Adsorption | AFM1 bioaccessibility remained elevated following treatment |
| [58] | W. confusa, L. plantarum | Milk/aqueous | 60–90 | Adsorption | High efficacy but mechanism confirmed as purely surface-based |
The RoB assessment revealed that over 65% of included studies did not report LOD, LOQ, or percent analyte recovery (Table 2). Furthermore, fewer than 25% of studies rigorously tested for toxin reversibility through desorption-challenge experiments, meaning that the permanence of AFM1 removal was empirically unverified in the large majority of studies
[59][62][63]
Risk of Bias (RoB) Assessment of Analytical Quality Across Study Categories
| Study Category | n | LOD/LOQ Reported (%) | Recovery Rate Reported (%) | Matrix Effects Assessed (%) | Desorption Tested (%) | Overall RoB |
| Standard LAB studies | 18 | 35 | 40 | 15 | 20 | High |
| Yeast/mixed culture studies | 8 | 45 | 50 | 25 | 15 | Medium–High |
| Nano/synbiotic studies | 4 | 75 | 80 | 60 | 10 | Medium |
| Recent studies (2024–2026) | 4 | 90 | 90 | 85 | 75 | Low |
4. Discussion
The overwhelming prevalence of adsorption-based removal (88% of studies) highlights a widespread scientific preference for rapid, surface-level results over structural detoxification. Surface binding is an equilibrium-driven process rather than a terminal one. Studies such as Hashemi and Amiri
[22][43][66][24]
It is important to acknowledge that high removal efficiencies reported in adsorption-based studies, particularly those exceeding 90%, remain technically significant, especially in contexts where initial concentrations far exceed regulatory thresholds. However, the permanence of this reduction under physiologically relevant conditions has not been established for the majority of these systems. The regulatory frameworks of EFSA and FDA would require precise mechanistic characterization and stability data before any biological decontamination agent could be approved for commercial dairy application
[13]
Over 65% of existing literature fails to report fundamental validation parameters (LOD, LOQ, and recovery rates), suggesting a systemic lack of rigor in the field's current analytical culture
[59][60]
A positive trend is evident in studies published in 2024–2026, with 90% reporting LOD/LOQ, 90% reporting recovery rates, and 75% conducting desorption-challenge experiments. This suggests the field is maturing toward standards necessary for industrial application. Future work must align with AOAC or ISO validation frameworks to enable meaningful cross-study comparison and regulatory submission.
A recurring limitation is the predominant use of simplified liquid milk models. The limited studies conducted in fat-rich matrices (e.g., sarshir, full-fat milk) or solid matrices (e.g., Frescal cheese) consistently report reduced decontamination efficacy compared to skim or reconstituted milk models
[17][36][21][18][46]
The collective limitations of microbial binding point toward a necessary evolution into small-molecule biotransformation. A promising avenue involves the use of natural polyphenolic compounds that possess innate degradative capabilities
[67][71][69][68][75]
5. Conclusion
This systematic review of 34 original research studies demonstrates that while microorganism-based biological interventions can achieve substantial AFM1 removal from dairy matrices, with peak efficiencies exceeding 90% in 12 studies, the dominant mechanism (reversible cell-wall adsorption in 88% of studies) renders current strategies fundamentally insufficient for guaranteed consumer protection. The equilibrium-driven nature of adsorption creates an inherent risk of toxin desorption under the physiological conditions of the gastrointestinal tract, a risk that is empirically quantified in only a minority of the reviewed studies. Compounding this mechanistic limitation is a pervasive analytical gap, with over 65% of studies failing to report LOD, LOQ, or recovery rates.
Two interconnected research imperatives follow from these findings. First, a mechanistic pivot is required, from passive adsorption toward enzymatic biodegradation strategies, with full structural identification of degradation metabolites by high-resolution MS techniques (e.g., Q-TOF or Orbitrap-based MS/MS) [64]. Second, analytical standardization is essential: all future decontamination studies must report AOAC- or ISO-compliant validation parameters, including matrix-matched calibration, recovery rates, and desorption-challenge experiments [74], [75]. Only through this dual commitment, to irreversible molecular transformation and to rigorous analytical verification, can biological AFM1 decontamination achieve the scientific and regulatory credibility required for adoption within integrated dairy safety management systems.
6. AI Declaration
During manuscript preparation, Elicit AI was used for automated literature identification across multiple databases and preliminary extraction of study metadata. All extracted data were subsequently verified against original full-text articles by the author, who takes sole and complete responsibility for the scientific content herein.