Bacteriocins for Safety of Animal-Derived Foods: Systematic Mapping, Multilevel MIC Analysis, Food-Matrix Applications, and Emerging Antiparasitic Evidence

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript addresses an important topic and is methodologically rigorous in its search strategy and PRISMA compliance. However, the incomplete bacteriocin coverage, superficial mechanistic and translational analysis, and overinterpretation of both the meta-analysis and antiparasitic evidence prevent acceptance in its current form.

1. The review omits several well-established bacteriocins highly relevant to animal-derived foods, including sakacins (e.g., sakacin P, sakacin K), pediocin PA-1/AcH, leucocins, curvaticins, lacticin 3147, bavaricin, and carnobacteriocins. The strict inclusion criterion of "purified bacteriocins" has excluded a substantial body of credible evidence where the compound is well characterized but not purified to homogeneity in every study. This significantly limits the generalizability of the conclusions and undermines the claim of providing a "systematic mapping."
2. The Discussion describes general modes of action in a textbook-like manner (membrane permeabilization, lipid II binding, etc.) but fails to link MIC values to mechanistic differences that explain potency variations across families and target species. No attempt is made to correlate observed activity with receptor-binding affinities, membrane insertion kinetics, or peptide structural features. For Gram-negative bacteria, the review ignores the extensive literature on outer-membrane permeabilization strategies (e.g., EDTA, lactoferrin, chitosan) that are essential for practical applications.
3. Table 2 and Figure 4 provide a count of applications but do not address whether the bacteriocin concentrations used in foods correspond to the MIC values from the meta-analysis. The review also fails to systematically discuss matrix interference effects (fat content, protein binding, salt, pH, proteolytic degradation) that profoundly affect bacteriocin performance in real foods. Moreover, there is no discussion of sensory impact, consumer acceptance, producer-strain safety (GRAS/QPS), or regulatory pathways, which are critical for industrial adoption.
4. Only three studies are available, all in murine models, with only one testing purified enterocins. The other two used producer strains without isolating the bacteriocins, so the observed effects cannot be specifically attributed to bacteriocin production. The abstract and title imply a broader scope than the data support. This component should be reframed as a preliminary, hypothesis-generating observation rather than a core finding.
5. MIC values are pooled across different studies, target strains, and assay conditions (broth media, inoculum size, incubation time) without reporting heterogeneity statistics (I², τ²) or performing sensitivity/subgroup analyses. The wide confidence intervals (e.g., lactocin: 5.5 ± 5.0 mg/mL) render the pooled estimates statistically fragile. The authors should report heterogeneity metrics, perform sensitivity analyses, and clearly state that the pooled MICs are descriptive summaries rather than definitive comparative rankings.
6. The brief mention of bacteriocins combined with essential oils and surface sanitation (lines 520–530) is interesting but lacks depth. This content should either be expanded into a separate subsection or removed.
7. The terms "plantaricians" and "plantaricins" are used interchangeably throughout the text and should be standardized.
8. Several citations in Table 2 appear incomplete (e.g., [58–77] lack full titles and page numbers). All DOIs and reference details should be verified.
9. Figure 2 (pie charts) adds little beyond Table 1. Consider replacing it with a heatmap showing bacteriocin family × target species coverage to better illustrate gaps and distribution patterns.

Author Response

Comment 1

The review omits several well-established bacteriocins relevant to animal-derived foods and the former emphasis on purified bacteriocins substantially limits the claimed systematic mapping.

Response

We agree and substantially expanded the nomenclature and search strategy. The high-sensitivity search now includes specific bacteriocin names/classes such as nisin, enterocins, plantaricins, pediocins, sakacins, leucocins, curvacins/curvaticins, lacticin 3147, bavaricins, carnobacteriocins, subtilosin, microcins and related terms. We also removed purification/MIC wording as automatic title/abstract exclusion criteria. In the Discussion, we now explicitly include translational examples for sakacin K, pediocin AcH, lacticin 3147 and leucocin A while keeping these examples analytically distinct from the primary exact-MIC tier.

Location in revised manuscript

Lines 139-159 (expanded high-sensitivity search); lines 181-192 (purification/MIC no longer automatic exclusion criteria); lines 525-533 and 616-624 (broadened bacteriocin coverage and translational examples).

Comment 2

The mechanistic discussion was superficial and did not sufficiently link MIC variation to receptor binding, membrane insertion, structural features, or Gram-negative outer-membrane permeabilization strategies.

Response

The Discussion was rewritten to connect observed MIC heterogeneity with biologically relevant mechanisms. We now discuss receptor-mediated recognition, lipid II binding, thiopeptide translation inhibition, membrane composition, peptide charge/hydrophobicity, insertion kinetics and growth state. For Gram-negative organisms, a dedicated passage discusses the outer-membrane barrier and sensitization strategies including EDTA, lactoferrin, organic acids, essential-oil components, heat, freezing and high pressure.

Location in revised manuscript

Lines 513-524 (mechanistic determinants of MIC heterogeneity); lines 534-544 (Gram-negative outer-membrane barrier and sensitization strategies).

Comment 3

The manuscript did not relate food-use concentrations to MIC values and insufficiently addressed matrix interference, sensory effects, producer-strain safety, GRAS/QPS status, and regulatory pathways.

Response

We addressed this point in both Methods and Discussion. A dedicated dose-versus-MIC audit was introduced using strict criteria for bacteriocin identity, target, concentration basis, matrix exposure and unit compatibility. The revised manuscript explicitly explains why broth MIC cannot generally be converted into an in-food dose and discusses protein/fat binding, pH, salt, proteolysis, diffusion, delivery systems and hurdle technologies. We also added a specific safety/regulatory discussion covering virulence determinants, antimicrobial resistance, transferable resistance, toxigenic potential, genome stability, QPS/GRAS considerations, sensory neutrality and consumer acceptance.

Location in revised manuscript

Lines 266-273 (dose-versus-MIC methodology); lines 430-441 (dose-versus-MIC audit results); lines 545-555 and 575-606 (matrix effects, delivery and hurdle strategies); lines 607-639 (producer-strain safety, regulatory and technological considerations).

Comment 4

The antiparasitic component was overinterpreted because the evidence was sparse and largely based on producer strains or murine models rather than purified bacteriocins.

Response

We fully agree and reframed the antiparasitic component. The title now uses the wording “Emerging Antiparasitic Evidence,” the abstract explicitly states that the evidence is preliminary and hypothesis-generating, and the Methods separate direct peptide exposure, producer-strain evidence, indirect probiotic evidence and sensitivity-tier evidence. The Results and Discussion explicitly state that no included antiparasitic study is a food-matrix challenge test and that therapeutic/animal-model effects must not be interpreted as demonstrated food biopreservation.

Location in revised manuscript

Lines 3-5 (revised title); lines 34-39 (abstract framing); lines 274-292 (evidence categories and interpretive separation); lines 442-479 (study-by-study antiparasitic evidence); lines 641-673 (explicit separation of therapeutic models from food biopreservation); lines 718-727 (conclusions and future validation needs).

Comment 5

The pooled MIC analysis lacked heterogeneity statistics and sensitivity analyses, and the estimates should not be interpreted as definitive family rankings.

Response

The quantitative synthesis was completely redesigned. Formal pooling is now target-specific and restricted to study-target cells where within-study variance can be estimated empirically. The model uses inverse-variance random effects, REML estimation, modified Hartung-Knapp confidence intervals, Cochran Q and I². Leave-one-study-out analyses are provided for targets with at least three contributing studies, and a one-stage mixed model using all 312 primary observations is included as a secondary sensitivity analysis. The revised manuscript explicitly avoids interpreting descriptive family distributions or the mixed-model family term as a potency ranking, and the substantial heterogeneity is emphasized in the Results, Discussion and Limitations.

Location in revised manuscript

Lines 209-233 (revised statistical framework); lines 370-398 (pooled estimates, tau², I², Q, mixed-model results and forest plots); lines 486-512 and 556-573 (interpretation without simple family ranking); lines 676-690 (limitations related to heterogeneity and sparse/unbalanced evidence).

Comment 6

The section on essential oils and surface sanitation should either be expanded or removed.

Response

The material was reorganized into the dedicated food-translation subsection rather than left as isolated examples. Combined-hurdle strategies are now interpreted as a distinct intervention category, and essential oils are discussed together with EDTA, lactoferrin, organic acids, chitosan, heat, freezing and high pressure. The revised text also notes that co-treatments may independently drive microbial reductions and may introduce sensory constraints, so factorial controls are required.

Location in revised manuscript

Lines 575-606, especially lines 597-606.

Comment 7

The terminology for plantaricins should be standardized.

Response

The terminology was standardized throughout the revised manuscript to “plantaricin/plantaricins,” with nomenclature harmonized across the search strategy, evidence mapping and Discussion.

Location in revised manuscript

Examples at lines 141-145, 525-533 and 616-624; terminology was standardized globally.

Comment 8

Incomplete citation ranges and reference details should be verified, including DOIs and full bibliographic information.

Response

Reference reporting was revised. Article-level references and DOI/source links for food-matrix studies are now provided in the supplementary tables rather than relying on incomplete citation ranges in the main text. The Supplementary Materials statement also specifies the datasets and documentation accompanying the article.

Location in revised manuscript

Lines 252-254 (article-level DOI/source links in Supplementary Table S5); lines 422-424 (references, DOI links, doses and outcomes in Supplementary Tables S5-S7); lines 731-735 (complete Supplementary Materials statement); lines 748-751 (Data Availability Statement).

Comment 9

Replace the former pie-chart presentation with a heatmap showing bacteriocin family × target coverage.

Response

We implemented this suggestion. In the current numbering, Figure 3 is a bacteriocin-family × target-pathogen heatmap based on the primary exact-MIC tier. It reports the number of distinct bacteriocins represented in each cell and is designed to show coverage gaps without treating repeated observations or the same bacteriocin across multiple pathogens as independent unique compounds. Figure 2 provides the complementary observation-level distribution of primary MIC values.

Location in revised manuscript

Lines 336-369 (Figures 2 and 3, including the bacteriocin-family × target-pathogen heatmap and captions).

Reviewer 2 Report

Comments and Suggestions for Authors

The topic is interesting and fits the scope of the journal. The review may be valuable; however, in its current form I have serious concerns regarding the search and study-selection methodology, data selection and pooling, the validity of the meta-analysis, and the consistency of some results with the supplementary dataset. The criteria used to select target pathogens, as well as the completeness and consistency of the reference list, also require clarification. Detailed comments are provided in the attached review.

Comments for author File: Comments.pdf

Author Response

The original Referee 2 report was narrative rather than numbered. For clarity, the principal concerns are organized below as thematic points R2.1-R2.5.

Comment R2.1

The term “meta-analysis” and the former title were questioned because the quantitative synthesis was not sufficiently demonstrated as a formal meta-analysis and the antiparasitic evidence was much smaller than the antibacterial evidence.

Response

The title and scope were revised. The quantitative synthesis is now explicitly a target-specific MIC meta-analysis and is performed only where empirical within-study variance is available. The antiparasitic component is described as “emerging” rather than as an equivalent evidence stream, and the abstract explicitly reports its limited and heterogeneous nature.

Location in revised manuscript

Lines 3-5 (revised title); lines 15-39 (revised abstract); lines 209-233 (formal meta-analytic framework); lines 641-673 (antiparasitic interpretation).

 

Comment R2.2

The selection of target pathogens required clearer justification, including strain/pathotype/serovar-level relevance, toxin-mediated hazards, and additional parasite targets such as Taenia and Sarcocystis.

Response

The Methods now define the hazard panel and explicitly state that Regulation (EC) No 2073/2005 is used as one regulatory reference rather than as a comprehensive list of pathogens. Epidemiological relevance is assessed at the most specific level available: STEC/EHEC and E. coli pathotypes, non-typhoidal Salmonella enterica with serovar information, C. jejuni/C. coli, pathogenic Y. enterocolitica bioserotypes/virulence, and toxigenicity/toxin relevance for S. aureus, B. cereus, C. perfringens and C. botulinum. Taenia spp. and Sarcocystis spp. were added to the complementary parasite search.

Location in revised manuscript

Lines 124-138.

 

Comment R2.3

The search strategy was too narrow, particularly because it relied on the generic word “bacteriocin,” omitted specific bacteriocin names and synonyms, and failed to recover the Giardia study by Amer et al.

Response

The search was expanded substantially through a complementary high-sensitivity component incorporating specific bacteriocin names/classes, pathogen and parasite synonyms, species-level terms, food-matrix terminology and known producer-strain terms. The parasite search now explicitly includes Giardia lamblia, G. intestinalis and G. duodenalis. The Amer et al. Giardia study was recovered and is now described in the Results and Discussion.

Location in revised manuscript

Lines 139-159 (expanded search); lines 274-292 (parasite-specific search and evidence categories); lines 443-454 (recovered Giardia evidence and explicit zero-evidence parasite targets); lines 642-647 (Giardia interpretation).

 

Comment R2.4

The previous manuscript did not sufficiently distinguish pathogenic foodborne strains from laboratory/reference strains or toxin-associated hazards.

Response

We now retain strain, serovar and pathotype information wherever available and explicitly separate pathogen-specific food-safety evidence from broader antimicrobial-spectrum evidence. Clinical MRSA and laboratory reference strains are not interpreted as proof of enterotoxin control, and MICs against toxigenic species are not interpreted as inhibition of toxin synthesis unless toxin production was directly measured.

Location in revised manuscript

Lines 131-138 (eligibility framework); lines 328-335 (interpretive safeguards in Results); lines 560-565 (Discussion).

 

Comment R2.5

The overall reliability of the results was limited by search incompleteness, unclear eligibility, MIC selection/pooling, and limited reproducibility.

Response

These concerns prompted a full reconstruction of the evidence base. The revised review includes 25,499 source rows, 12,700 unique bibliographic records, 577 curated MIC observations from 146 studies, and 312 exact conventional MIC observations from 94 studies in the primary quantitative tier. The reconstructed evidence base also contains 204 direct food-matrix studies and 11 antiparasitic publications. All eligible MIC observations are retained at observation level; artificial standard errors are not assigned to study-target cells lacking empirically estimable variance; formal pooling is restricted to eligible study-target cells; sensitivity analyses are reported; and reproducible inputs/code are supplied in Supplementary Tables S9 and S21-S23 and Supplementary Code S1.

Location in revised manuscript

Lines 160-176 and 295-310 (search/reconciliation arithmetic and evidence-set counts); lines 194-233 (MIC extraction and statistical analysis); lines 370-398 (quantitative results); lines 731-735 and 748-751 (supplementary data, reproducible code and Data Availability Statement).

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript addresses an interesting and timely topic and has the potential to provide a useful overview of bacteriocins relevant to animal-derived food safety. However, substantial methodological concerns currently limit the reliability of the quantitative conclusions. In particular, the search/filtering strategy may omit relevant literature, no formal quality/risk-of-bias assessment is reported, selection of the lowest MIC from each study introduces systematic bias, and the statistical basis for the random-effects meta-analysis of individual MIC values requires clarification or reconsideration. The food-matrix and antiparasitic components also require clearer methodological separation and more cautious interpretation. I therefore recommend major revision.

Comments to the authors

1. Major — The literature-search strategy appears too restrictive for a systematic review. The search is essentially followed by keyword filtering for “purified/purification” and then “MIC/minimum inhibitory concentration”. This approach can easily exclude relevant studies in which the word bacteriocin, purified, or MIC does not appear in the title/abstract even though these data are reported in the full text. It may also miss articles referring directly to specific bacteriocins such as nisin, pediocin, enterocin, plantaricin, etc. The authors should provide the complete database-specific search strategies and substantially justify or revise the keyword-based exclusion procedure. Search terms should preferably include synonyms, specific major bacteriocin classes/names, plural variants, and pathogen synonyms. Any language, publication-date, or document-type restrictions should also be clearly stated.
2. Major — The study-selection process is insufficiently documented for PRISMA compliance. The manuscript states that two independent reviewers extracted data, but it is not equally clear whether title/abstract and full-text screening were independently performed by two reviewers. In addition, the Results introduce a “journal article” publication-type filter that removed 53 records, although this filter is not adequately described in the search methodology. The PRISMA diagram reports 46 full-text exclusions but does not provide the specific exclusion reasons. Please fully report the screening process, reviewers, disagreement resolution, exclusion reasons, and whether a review protocol was prospectively registered. If no protocol/registration exists, this should simply be stated.
3. Major — No methodological-quality/risk-of-bias assessment is presented. This is especially important because the manuscript performs a quantitative synthesis and compares bacteriocin families. I could not identify a formal assessment of study quality or risk of bias. The included studies presumably differ considerably in MIC methodology, microorganism strains, inoculum, growth medium, endpoint determination, peptide purity, and experimental conditions. Without assessing these aspects, highly heterogeneous studies may contribute equally to apparently quantitative conclusions. The authors should introduce an appropriate quality-assessment framework, or an adapted predefined checklist, and use the findings when interpreting the evidence.
4. Major — The statistical basis of the MIC “meta-analysis” requires substantial clarification and possibly reconsideration. The authors convert MICs to mg/mL, log-transform them, group bacteriocins by family and Gram status, and state that random-effects meta-analytic models were applied. However, individual MIC measurements are not conventional study effect estimates accompanied by sampling variances. It is therefore unclear how study-level standard errors/variances and inverse-variance weights were obtained. The manuscript should specify the exact random-effects model, estimator for between-study variance, weighting procedure, and source of within-study variance. Heterogeneity statistics such as τ², I² and Q should also be reported. If suitable within-study uncertainty is unavailable, a descriptive quantitative synthesis of log-MIC distributions may be statistically more defensible than presenting this as a conventional meta-analysis.
5. Major — Selecting the lowest MIC from each article creates a systematic bias toward greater apparent antimicrobial activity. The manuscript explicitly states that when multiple eligible MIC values were reported, “the lowest MIC value was retained”. This is a major concern. A study evaluating many strains or conditions has a greater chance of producing an unusually low value, and using this best-case value will systematically underestimate representative MICs. The authors should reconsider this rule. Possibilities include analysing all relevant MIC observations using a multilevel model, using a predefined representative value, or summarizing multiple measurements within each study using an appropriate central measure such as a geometric median/mean. At minimum, a sensitivity analysis comparing alternative selection rules is needed.
6. Major — Pooling at the bacteriocin-family level may obscure very large biological heterogeneity. For example, enterocins and plantaricins comprise structurally and mechanistically diverse peptides, while MICs are measured against different species and strains under different experimental conditions. Consequently, a pooled “family MIC” may not represent a biologically meaningful property of that family. The authors should provide a table of the individual MICs entering each pooled analysis—including bacteriocin identity, target species/strain, study, assay conditions and converted value—and consider stratification by individual bacteriocin and/or target microorganism where sufficient data exist. Claims that one bacteriocin family is “more effective” than another should be strongly tempered.
7. Major — Figure 3 needs statistical clarification. The manuscript discusses original-scale mean MICs such as 0.0015 ± 0.0010 mg/mL for micrococcin, whereas Figure 3 reports pooled log10 MIC estimates and 95% CIs. The relationship between these two sets of values is not sufficiently transparent. In addition, some intervals in the forest plot appear unusual—for example, the micrococcin estimate has an essentially zero-width CI, whereas lactocin has an extremely wide interval. The authors should verify these calculations and provide the underlying data, study weights, k, pooled log estimate, 95% CI, τ²/I² and preferably the back-transformed estimate. The x-axis label “Effect Size” should also be replaced with the actual metric, e.g., pooled log10 MIC (mg/mL).
8. Major — There appears to be a numerical inconsistency in Table 1/Figure 2. The text states that Staphylococcus aureus is associated with 29 distinct bacteriocins, but I count 28 entries in the corresponding Table 1 row. Accordingly, the Gram-positive total calculated directly from Table 1 would be 28 + 22 + 4 + 1 = 55, whereas the manuscript reports 56 Gram-positive bacteriocin–pathogen associations. Please verify all counts and Figures 2/Table 1. It would also be clearer to call these bacteriocin–pathogen associations rather than simply “number of active bacteriocins,” because the same bacteriocin can occur under several pathogens.
9. Major — The supplementary food-matrix mapping is not reproducible from the Methods. The Results suddenly state that 204 records were considered, 199 remained after duplicate removal, 46 involved direct food-matrix testing, and 34 were animal-derived applications. However, Section 2.2.3 mainly provides classification criteria and does not explain how these 204 records were identified: databases, search strings, search date, screening procedure and relationship to the primary 5,291-record search are unclear. This component should have a reproducible search methodology and preferably its own flow description.
10. Major — Direct bacteriocin efficacy should be distinguished from combined-hurdle or producer-culture effects. Table 2 combines purified bacteriocin addition, bacteriocinogenic cultures, active packaging/coatings, and interventions containing other hurdles such as high pressure, nitrite, refrigeration, lactic acid, heat or cold plasma. These cannot all be interpreted equivalently as evidence for bacteriocin efficacy. For example, the authors themselves appropriately recognize that lactocin C-M2 was used together with cold plasma, making its independent contribution uncertain. The application evidence should therefore be categorized according to whether the bacteriocin was tested alone, as a producer culture, in a delivery system, or in a combined hurdle treatment.
11. Major — The antiparasitic inclusion criterion is internally inconsistent and the conclusions are somewhat overextended. The Methods allow bacteriocinogenic strains when bacteriocin involvement is explicitly reported. However, the Discussion later states that no bacteriocin was isolated or directly tested for Lactobacillus brevis PQ214320 or Bacillus subtilis PQ198038, and that the effects therefore cannot specifically be attributed to bacteriocin production. I agree with the latter interpretation. These studies should either be excluded from the bacteriocin evidence set or clearly placed in a separate category of potentially bacteriocinogenic/probiotic strain evidence. Otherwise readers may incorrectly infer direct antiparasitic bacteriocin activity.
12. Major — Murine antiparasitic efficacy should not be equated with food biopreservation. All eligible Trichinella investigations were murine infection studies, rather than direct food-matrix challenge studies. Nevertheless, the Conclusion proposes subsequently evaluating the strains as “bacteriocin-producing protective starter cultures”. This is an interesting research hypothesis, but the translational step is presently large. The title, abstract and conclusions should more clearly distinguish in vivo anti-Trichinella/probiotic effects from demonstrated food-biopreservative effects.
13. Major/Minor — Safety considerations deserve substantially greater attention before proposing producer strains for food application. In particular, strain-specific safety evaluation is essential for enterococci before considering them as protective cultures, including virulence determinants, antimicrobial resistance, transferable resistance genes and relevant regulatory/safety status. This issue should be discussed explicitly rather than presenting the identified strains mainly in terms of efficacy.
14. Minor — The predefined pathogen panel and the final results should be connected more transparently. The Methods include Campylobacter, Yersinia, Clostridium botulinum and several other targets, yet the antibacterial Results ultimately discuss six pathogens. The authors should explicitly state which predefined bacterial pathogens produced zero eligible studies, just as they do for the parasite targets. This would strengthen the systematic-mapping aspect of the review and identify genuine research gaps.
15. Minor — A dedicated limitations subsection is needed. Important limitations include the restrictive search approach, small number of studies in several families, heterogeneity of MIC protocols, conversion of different units, use of the lowest MIC, family-level aggregation, incomplete molecular characterization of antiparasitic agents, and scarcity of direct food-matrix evidence. These limitations presently appear only indirectly rather than as a structured assessment.
16. Minor — The Discussion could be shortened and made more focused on evidence generated by the review. Several long mechanistic descriptions of individual bacteriocin classes occupy considerable space, while comparatively less attention is paid to methodological heterogeneity and evidence quality. The sections on food-waste reduction and indirect surface/essential-oil applications also go somewhat beyond the core systematic evidence. The manuscript would be stronger if the Discussion focused more heavily on what the systematic review itself demonstrates and where the evidence remains uncertain.
17. Minor — The manuscript requires careful editorial revision. Examples include missing spaces such as “included[12]” and “Gram-negative(n=26)”, missing punctuation, inconsistent use of “Subtilosin/subtilosin A”, expressions such as “application/trials”, and some awkward phrasing. In addition, the “Supplementary Materials” statement is currently incomplete (“The following supporting information can be downloaded at:”), although the Data Availability Statement refers to supplementary material. The supplementary files should contain, at minimum, full search strategies, excluded full-text studies with reasons, individual extracted MIC data/conversions, and the dataset/code underlying the quantitative synthesis.

 

Comments on the Quality of English Language

The manuscript is generally understandable; however, the English language would benefit from careful editing to improve clarity, grammar, punctuation, and sentence structure. Several sentences are overly long or awkwardly phrased, and there are occasional spacing and formatting inconsistencies. A thorough language revision by a fluent English speaker or professional editing service is recommended to improve readability and precision.

Author Response

Comment 1

The literature-search strategy was too restrictive and should include synonyms, major bacteriocin names/classes, plural variants, pathogen synonyms, and clearly stated restrictions.

Response

We substantially broadened the search strategy using a complementary high-sensitivity search containing specific bacteriocin names/classes, pathogen/parasite synonyms, species-level terms, food-matrix terms and known producer strains. No language or publication-date restriction was applied to the high-sensitivity component, and document type was assessed during screening rather than used as a database filter. Purification and MIC wording were removed as automatic exclusion criteria.

Location in revised manuscript

Lines 139-159 and 181-192.

 

Comment 2

The study-selection process was insufficiently documented for PRISMA compliance, including reviewer roles, disagreement resolution, exclusion reasons, publication-type filtering, and protocol registration.

Response

The revised Methods now state that no prospective protocol was registered and clearly distinguish automated prioritization from scientific eligibility assessment. All deduplicated records, irrespective of priority category, underwent manual title/abstract assessment by two reviewers using the predefined eligibility criteria, and no record was excluded solely on the basis of the model-generated priority assignment. High- and medium-priority records were systematically sought for full-text retrieval; low-priority records were assessed from the available title, abstract, indexing record and bibliographic metadata, with full texts assessed when available but not systematically sought. Full-text eligibility assessment and data extraction were performed independently by two reviewers, disagreements were resolved by consensus, and source/retrieval status was retained separately from scientific eligibility. Complete search arithmetic, access status and exclusion coding are reported in Supplementary Tables S2, S8 and S20. The use of ChatGPT for record prioritization is also disclosed in the Acknowledgments.

Location in revised manuscript

Lines 96-122 (reviewer roles, prioritization/retrieval workflow, consensus and source-status handling); lines 160-176 (search reconciliation, retrieval and access/exclusion coding); lines 728-730 (Acknowledgments disclosure).

 

Comment 3

No methodological-quality/risk-of-bias assessment was presented.

Response

We added an adapted predefined nine-domain methodological/reporting appraisal for the 94 studies contributing primary quantitative MIC observations. The domains include bacteriocin identity, preparation/purity, strain/serovar/pathotype characterization, exact/convertible MIC endpoint, MIC assay reporting, growth medium, inoculum, incubation conditions and source verification. It is explicitly interpreted as a reporting/methodological appraisal rather than a clinical risk-of-bias instrument.

Location in revised manuscript

Lines 255-265 (Methods) and 399-407 (Results).

 

Comment 4

The statistical basis of the MIC meta-analysis required clarification, including the source of within-study variance, random-effects model, estimator, weighting, and heterogeneity statistics.

Response

The statistical model was rebuilt and is now fully specified. Study-target effects are calculated only when at least two primary MIC observations are available; the effect is the mean log10 MIC and the within-study variance is the sample variance divided by the number of observations. Study-target means are pooled using inverse-variance random effects, REML for tau², the modified Hartung-Knapp adjustment for 95% CIs, Cochran Q and I².

Location in revised manuscript

Lines 209-233; results at lines 370-398.

 

Comment 5

Selecting the lowest MIC from each article creates systematic bias and should be reconsidered.

Response

This rule was removed. The revised dataset retains all eligible MIC observations at the bacteriocin-target-MIC observation level. Exact conventional values form the primary tier, while non-equivalent/censored/formulation-dependent endpoints are retained transparently outside the primary pooling or reserved for sensitivity analysis.

Location in revised manuscript

Lines 194-208. The resulting dataset is summarized at lines 370-372.

 

Comment 6

Pooling at bacteriocin-family level may obscure major biological heterogeneity and family rankings should be tempered.

Response

We agree. Formal pooling is now target-specific rather than family-level. Family is used for mapping and mechanistic context, and the secondary mixed model tests family effects only after accounting for target and study clustering. The revised manuscript treats the observation-level family distributions as descriptive rather than as a potency ranking, while the high heterogeneity and sparse/unbalanced representation of several families are explicitly acknowledged in the interpretation and Limitations.

Location in revised manuscript

Lines 209-233; lines 370-388; lines 486-512 and 556-565; lines 676-690.

 

Comment 7

Figure 3 required statistical clarification and should show the actual pooled metric, study counts, CIs, tau²/I² and supporting inputs.

Response

The former quantitative figure was replaced by a target-specific forest plot, now presented as Figure 4, showing study-level mean log10 MIC (mg/mL) and 95% CIs for Staphylococcus aureus, Escherichia coli and Listeria monocytogenes. Table 2 reports k, pooled log10 MIC, back-transformed MIC, 95% CI, tau², I², Q, degrees of freedom and p-values; the Methods and Table 2 specify REML estimation and the modified Hartung-Knapp adjustment. Supplementary Tables S21-S23 provide the study-level inputs, weights and sensitivity analyses.

Location in revised manuscript

Lines 370-398, especially lines 389-398 (Table 2 and Figure 4).

 

Comment 8

There was a numerical inconsistency in bacteriocin counts and the terminology should be “bacteriocin-target associations” rather than “number of active bacteriocins.”

Response

The quantitative evidence was reconstructed and the terminology was corrected. Counts are now explicitly termed bacteriocin-target evidence, and Table 1 reports MIC observations, distinct bacteriocins and studies separately for each target. The same bacteriocin appearing across multiple pathogens is not counted as an independent unique compound in the heatmap.

Location in revised manuscript

Lines 316-369.

 

Comment 9

The supplementary food-matrix mapping was not reproducible from the Methods.

Response

The food-application mapping is now explicitly described as deriving from the same reconciled evidence base plus complementary food-matrix search terms. Inclusion/exclusion rules are stated, animal-derived matrices are defined, intervention categories are prespecified, and article-level matrix, target, intervention, DOI/source link and outcome are reported in Supplementary Table S5.

Location in revised manuscript

Lines 235-254. Application results and counts are at lines 409-441.

 

Comment 10

Direct bacteriocin efficacy should be distinguished from producer cultures, delivery systems and combined-hurdle effects.

Response

We implemented this distinction throughout. Each food application is classified as direct bacteriocin/BLIS addition, producer/protective culture, delivery system/active packaging/coating, or combined-hurdle treatment. The Results explicitly state that these categories are not equivalent evidence of independent bacteriocin efficacy.

Location in revised manuscript

Lines 246-254; lines 409-424; lines 430-441.

 

Comment 11

The antiparasitic inclusion criterion was internally inconsistent; producer-strain effects should not be equated with direct bacteriocin activity.

Response

The antiparasitic evidence is now prospectively separated into direct peptide exposure, bacteriocinogenic/producer-strain evidence, indirect probiotic-strain evidence, and sensitivity-tier postbiotic/preprint evidence. These categories are not pooled as equivalent. The Results explicitly show an example where a live producer strain was active while the isolated AP7121 peptide showed no larvicidal effect.

Location in revised manuscript

Lines 274-292; lines 455-472; lines 648-663.

 

Comment 12

Murine antiparasitic efficacy should not be equated with food biopreservation.

Response

This distinction is now explicit in the Methods, Results, Discussion, title, abstract and Conclusions. No included antiparasitic study is presented as a food-matrix challenge test, and the manuscript states that current evidence supports only mechanistic or therapeutic hypotheses until direct food-matrix validation is performed.

Location in revised manuscript

Lines 289-292; lines 467-479; lines 641-673; lines 718-727.

 

Comment 13

Safety considerations for producer strains, particularly enterococci, require substantially greater attention.

Response

A dedicated safety paragraph was added. It addresses strain-specific virulence determinants, antimicrobial-resistance genes, transferability of resistance, toxigenic potential, genome stability and jurisdiction-specific regulatory status. It also notes that enterococci do not receive an automatic species-level EFSA QPS presumption of safety and require case-by-case assessment.

Location in revised manuscript

Lines 607-615; broader regulatory, technological and sensory considerations at lines 625-639.

 

Comment 14

The predefined pathogen panel and final results should be connected more transparently, including explicit zero-evidence targets.

Response

The revised Results report target-level coverage for the full bacterial evidence set and explicitly identify Clostridium botulinum as having no exact conventional MIC in the primary tier. The parasite section also records zero direct peptide-exposure evidence for Toxoplasma gondii, Anisakis spp., Taenia spp., Sarcocystis spp. and other screened parasite targets.

Location in revised manuscript

Lines 316-335 (bacterial coverage, including zero primary-tier observations for C. botulinum); lines 443-454 and 645-647 (parasite zero-evidence targets).

 

Comment 15

A dedicated limitations subsection is needed.

Response

A dedicated Section 4.4, “Limitations,” was added. It addresses heterogeneity in MIC methods, restricted eligibility for formal pooling, small contributing study numbers, sparse and unbalanced family representation, strain/serovar/pathotype and assay variation, matrix/dose heterogeneity, producer-strain safety and regulatory assessment, limited antiparasitic evidence, and the absence of prospective protocol registration. The prioritization and retrieval workflow is described separately and explicitly in the Methods, including the fact that low-priority full texts were assessed when available but were not systematically sought.

Location in revised manuscript

Lines 676-698 (Limitations); lines 96-122 (prioritization and retrieval workflow).

 

Comment 16

The Discussion should be shortened and focused more directly on evidence generated by the review, methodological heterogeneity and evidence quality.

Response

The Discussion was reorganized into focused subsections: antibacterial interpretation, translation to animal-derived foods, emerging antiparasitic evidence, and limitations. Mechanistic material is now used to explain observed heterogeneity rather than to construct family rankings, while food-translation and evidence-quality issues receive dedicated treatment.

Location in revised manuscript

Lines 486-573 (antibacterial interpretation); lines 575-639 (food translation); lines 641-673 (antiparasitic evidence); lines 676-698 (limitations).

 

Comment 17

The manuscript required editorial revision and the Supplementary Materials statement was incomplete; supporting files should include full search strategies, exclusions, MIC data/conversions, and analysis inputs/code.

Response

The Supplementary Materials statement is now complete and specifies Supplementary Tables S1-S23 and Supplementary Code S1, including database search strategies, screening/deduplication/access-exclusion audits, MIC extraction and reporting-quality datasets, food-matrix and dose-versus-MIC audits, antiparasitic mapping, study-level meta-analysis inputs, sensitivity analyses and reproducible code. The Data Availability Statement also confirms that the extracted datasets and quantitative inputs/outputs are supplied.

Location in revised manuscript

Lines 731-735 (Supplementary Materials) and 748-751 (Data Availability Statement).

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The revised quantitative analysis is now reproducible; however, I remain concerned about the biological meaning of the study-target effect and the variance assigned to it. This issue can be illustrated directly using the supplementary dataset. In the study DOI 10.1007/s12602-012-9095-x, the Listeria monocytogenes study-target cell contains 26 MIC observations representing two different bacteriocins, nisin and lacticin 3147, each tested against 13 strains. When analysed separately, the geometric mean MIC is approximately 0.0193 mg/mL for nisin and 0.00171 mg/mL for lacticin 3147, an approximately 11-fold difference. Combining all 26 observations produces the reported study-target mean of approximately 0.00573 mg/mL. This value is mathematically well defined, but it does not directly represent the activity of either bacteriocin.
More importantly, the variance among these 26 MIC observations reflects at least two biologically distinct sources of variability: differences between bacteriocins and differences between bacterial strains. Treating this variability divided by n as the sampling variance of a single study-level effect assumes that the observations are exchangeable measurements of one underlying effect. In this example they are not simple replicates of one bacteriocin-strain effect. Consequently, biological heterogeneity becomes part of the quantity used as within-study uncertainty and therefore directly affects inverse-variance weighting.
I therefore suggest reconsidering the analytical hierarchy rather than only the numerical calculation. The observation-level dataset is a major strength of the revised manuscript and could be retained in a multilevel framework in which MIC observations remain linked to study, bacteriocin identity and bacterial strain. Study-level clustering should be modelled explicitly, while bacteriocin identity/family, target species or strain characteristics, and relevant methodological variables could be considered as additional levels or moderators, depending on data availability. Robust variance estimation may also be considered when multiple dependent estimates originate from the same study.
A useful applied example is Isola et al. (2026, The Veterinary Journal, DOI 10.1016/j.tvjl.2026.106844), who analysed log10 MIC data using a multilevel random-effects model with strain-level observations nested within studies and robust variance estimation. Vanacker et al. (2023, Frontiers in Microbiology, DOI 10.3389/fmicb.2023.1186920) similarly retained multiple E. coli strain-level observations within studies and explicitly separated within-study from between-study heterogeneity. These examples may provide a useful framework for retaining the rich observation-level structure of the present dataset rather than collapsing biologically different bacteriocin-strain combinations into a single study-target mean.
If such modelling is not feasible because true sampling variances or sufficient replication are unavailable, I would consider a structured descriptive quantitative synthesis preferable to assigning biological between-bacteriocin/between-strain variation the role of within-study sampling variance.

Author Response

We thank the Reviewer for this important methodological observation. We agree that aggregating 
biologically distinct MIC observations within a study-target cell may obscure relevant differences among 
bacteriocins and strains, and that the resulting within-cell dispersion should not be interpreted as a 
conventional sampling variance for inverse-variance weighting.
We therefore substantially revised the quantitative analysis. The primary inferential analysis is now 
conducted at the level of the individual MIC observation, retaining all 312 exact conventional MIC values 
rather than first aggregating them into study-target means. The revised linear mixed-effects model 
includes bacteriocin family and target category as fixed effects, together with random intercepts for 
Study_ID, Study_Strain, and bacteriocin identity. Study_Strain denotes the study-specific target/strain 
grouping as reported, thereby retaining strain-level information whenever available. This structure was 
selected to account explicitly for dependence among observations from the same study, repeated 
observations within the same study-strain grouping, and repeated measurements associated with the 
same bacteriocin.
The final model therefore had the following structure:
log10(MIC) ~ Family + Target category + (1|Study_ID) + (1|Study_Strain) + (1|Bacteriocin)
The model included 312 observations from 94 studies, 212 study-strain clusters, and 117 bacteriocin 
identities. Both bacteriocin family and target category remained globally significant after accounting for 
this hierarchical structure: family, F(11, 52.58) = 3.568, p < 0.001, and target category, F(6, 115.42) = 
4.219, p < 0.001.
Importantly, the revised model also separates the different sources of variability. The estimated variance 
components were 1.098 for Study_ID, 0.015 for Study_Strain, 0.015 for bacteriocin identity, and 0.099 
for the residual component. The relatively large study-level component indicates that experimental 
context remains an important source of MIC variability, whereas the additional bacteriocin and study
strain components explicitly represent biological dependencies that were previously combined within the 
study-target variance.
In accordance with the Reviewer’s suggestion, the previous study-target aggregation is no longer used 
for an inverse-variance-weighted inferential meta-analysis. It is retained exclusively as a secondary 
descriptive summary to illustrate the distribution of MIC patterns at the study level. We no longer interpret 
variability among biologically distinct bacteriocin-strain observations as sampling variance.
We also considered the Reviewer’s suggestion regarding robust variance estimation. Because sampling 
variances were not available at the level of the individual MIC observation, and because the new primary 
analysis directly models dependence through a multilevel random-effects structure, we did not construct 
artificial sampling variances solely to apply a separate robust-variance procedure. Instead, the 
hierarchical dependence highlighted by the Reviewer is addressed directly within the mixed-effects 
model.
The Methods, Results, Discussion, Limitations, Conclusions, and the quantitative figure have been 
updated accordingly. Figure 4 now presents adjusted estimates based on the final multilevel model 
rather than the previous study-target inferential pooling.
We believe that this revision addresses the central methodological issue raised by the Reviewer by 
preserving the biological identity of the individual MIC observations and explicitly separating sources of 
dependence related to study, strain, and bacteriocin, rather than treating them as interchangeable 
sampling variability