Forensic source classes and toxic-equivalent concentrations of polycyclic aromatic hydrocarbons in soils and sediments: a systematic review and meta-analysis

1 Introduction

Polycyclic aromatic hydrocarbons (PAHs) are persistent, largely carcinogenic contaminants of soils and sediments and a recognized burden on human and ocean health (Landrigan et al., 2020; Samarajeewa, 2023). Environmental forensics conventionally partitions their origin into pyrogenic (combustion) and petrogenic (petroleum-derived) classes using molecular diagnostic ratios, such as Fla/(Fla + Pyr), Ant/(Ant + Phe), BaA/(BaA + Chr), and IcdP/(IcdP + BghiP) (Yunker et al., 2002; Tobiszewski and Namieśnik, 2012). Because pyrogenic mixtures are typically enriched in high-molecular-weight congeners carrying the largest toxic-equivalency factors (Nisbet and LaGoy, 1992), it is widely assumed that source class carries information about the toxicity of the contaminated matrix.

This assumption has practical consequences. If source class predicted toxic-equivalent concentration, the diagnostic ratios already computed in thousands of site investigations could serve as a rapid, low-cost triage of hazard, and forensic attribution could inform remediation priority directly. If it does not, attribution and hazard assessment must remain separate exercises. The question is therefore worth evaluating rather than assuming.

The inferential chain must, however, be stated precisely because it is weaker than it first appears. Diagnostic ratios are affected by weathering, mixing, differential degradation, analytical censoring, and disagreement between threshold conventions, and they can misassign sources (Katsoyiannis and Breivik, 2014; Dymova et al., 2021). Benzo[a]pyrene toxic-equivalent concentration (TEQBaP), in turn, weights measured concentrations by a chosen toxic-equivalency-factor scheme; it expresses relative carcinogenic potency as a benzo[a]pyrene equivalent and does not measure bioavailability, exposure, or any organism response. This review therefore evaluates an association between a study-level source classification and a reported toxicity-potential metric. It does not evaluate whether the true emission process causes ecological harm, and we use the terms toxicity potential and toxic-equivalent concentration, rather than ecological hazard or ecotoxicological burden, throughout.

We framed the review using a PECO structure. Population: solid environmental matrices (soil and sediment). Exposure/moderator: the forensic source class assigned by the primary authors from molecular diagnostic ratios (pyrogenic, mixed, and petrogenic). Comparator: studies assigned to a different source class. Outcome: the reported benzo[a]pyrene toxic-equivalent concentration (TEQBaP, ng g^-1 dry weight, or the identical unit μg kg^-1). Unit of analysis: the study, because sample-level TEQBaP paired with sample-level source assignment is almost never published. The primary contrast is pyrogenic versus petrogenic, the classes expected to differ most in intrinsic potency; because petrogenic-dominant solid-matrix studies reporting TEQBaP proved scarce, the pyrogenic-versus-mixed contrast is reported as the only estimable comparison and is identified as such, rather than as a pre-registered primary endpoint. On mechanistic grounds, we expected pyrogenic assemblages to show the higher TEQBaP per unit total PAH, and we regarded a two-fold difference in geometric mean TEQBaP as the smallest difference that would be practically meaningful for screening or triage; this margin was defined for interpretation, not used as a formal equivalence test.

The evidence window was set at 2013–2026. The original corpus covered 2015–2026; during revision, the window was extended backward after a supplementary search identified eligible studies from 2013 to 2014, including the only petrogenic-dominant study in the entire evidence base that reports both TEQBaP and total PAH. The window is bounded below by the period over which combined diagnostic-ratio and TEQ reporting became routine in this literature; we note explicitly that it remains a pragmatic, rather than a principled, boundary and that earlier studies exist which report source profiles and TEQ separately.

2 Materials and methods

2.1 Reporting standard

The review follows PRISMA 2020 (Page et al., 2021) and is reported as a systematic review with meta-analysis. Each methodological decision described below is identified as either pre-specified or data-informed, so that the confirmatory and exploratory elements of the analysis can be distinguished.

2.2 Information sources and search strategy

The primary search was executed on 2 July 2026 across Scopus, PubMed, and Web of Science, restricted to peer-reviewed journal articles reported in English. The string was applied to title, abstract, and keyword fields and combined four concept blocks, namely, analyte, source attribution, toxicity or risk, and environmental compartment as follows:

TITLE-ABS-KEY((“polycyclic aromatic hydrocarbon” OR PAH OR PAHs OR “polyaromatic hydrocarbon“) AND (pyrogenic OR petrogenic OR pyrolytic OR combustion OR “diagnostic ratio” OR “isomer ratio” OR “molecular ratio” OR “source apportion” OR “source identif” OR “source signature” OR “source diagnos”) AND (toxic* OR ecotox* OR bioassay OR biomarker* OR “oxidative stress” OR EROD OR CYP1A OR genotox* OR mortalit* OR LC50 OR EC50 OR “effect concentration” OR bioavailab* OR “ecological risk” OR “sediment quality guideline“) AND (sediment OR soil OR freshwater OR river OR lake OR aquatic OR benthic OR estuar)).

The search returned 2,119 records after duplicates were removed on DOI and title. Title and abstract screening retained 765 records, excluding items that were off-topic, reported no source attribution, reported no toxicity or risk outcome, or were reviews. Screening of the full records retained 234, excluding studies without a solid environmental matrix or without an extractable outcome. Retrieval and assessment of those 234 reports produced the review corpus of 105 full texts, and the remaining 129 were unavailable in full text or report the outcome in units that could not be reconciled. Eligibility assessment of the 105 excluded a further 39 with the reasons and counts itemized in Figure 1, leaving 66 studies from this stream. A supplementary search conducted during revision, using the same eligibility criteria with the window extended back to 2013, identified four additional studies meeting all quantitative criteria; these are reported separately in Figure 1 in accordance with PRISMA 2020 item 16a rather than pooled with the database records. After the primary-source audit described in Section 3.1 removed one study, 67 studies were included in the review.

The supplementary search referred to above was conducted on 12 August 2026, after the first review round, for two stated reasons: to assess whether the 2015 lower bound of the original window had excluded eligible evidence, and to search specifically for petrogenic-dominant solid-matrix studies, the class the first analysis had shown to be unestimable. It used the same three databases and the same four concept blocks as the primary search, with two modifications. The publication window was extended back to 2013, and the source block was additionally run in a targeted form pairing the analyte and compartment blocks with (petrogenic OR petroleum OR “oil spill” OR oilfield) and the outcome terms (“toxic equivalen” OR TEQ OR “BaP equivalen“). This was combined with backward citation checking of the reference lists of all included studies and forward citation checking of the three most recently published of them. After removal of duplicates and of records already held in the corpus, 64 records were screened against the unchanged eligibility criteria described in Section 2.3. Sixty were excluded, the recurring reasons being that no TEQBaP value was reported in extractable form, that no diagnostic-ratio source classification was given, or that the matrix was water rather than soil or sediment. Four studies met every criterion and were added: Nasher et al. (2013), Azimi Yancheshmeh et al. (2014), Raeisi et al. (2016), and El-Maradny et al. (2025). Three of the four predate the original window, and one of them is the only petrogenic-dominant study in the entire evidence base that reports ΣPAH alongside TEQBaP; this is why the window was extended to 2013 rather than left at 2015. All four were extracted and appraised by the same procedures applied to the database records, and each is listed with its extraction provenance in Supplementary Table S1.

Screening and extraction were performed by one reviewer and independently verified by a second against the primary source; disagreements were resolved by discussion, and where a value or a source label could not be agreed, the study was excluded, and the reason recorded. Every extracted number is accompanied in Supplementary Table S1 by its exact location (page, table, or supplementary item) in the primary publication.

2.3 Eligibility criteria

Studies were eligible for the quantitative synthesis if they (i) reported TEQBaP for soil or sediment in ng g^-1 dry weight or μg kg^-1; (ii) reported a source classification derived from molecular diagnostic ratios; and (iii) reported a central value with sufficient information to identify the summary statistic used. Excluded were reviews, non-solid matrices, water-only TEQ, figure-locked values, and values in units that could not be reconciled. A study contributing the same underlying samples as another included report was treated as a duplicate cohort, and only the more complete report retained; this rule replaces the vague criterion “redundant regional surveys adding no distinct evidence” used in the original submission, which we accept was not a defensible exclusion ground.

Studies that met the review question but not the quantitative criteria were retained for a structured narrative synthesis (Section 4.5) under pre-specified rules: they had to address PAH source attribution or toxicity potential in an environmental matrix and report either a source classification or a standardized risk metric. Their role is to characterize the concentration and reporting landscape from which the quantitative subset is drawn, not to corroborate the meta-analytic estimate.

2.4 Source classification: decision algorithm

Because a study-level label is an aggregation of potentially discordant evidence, source class was assigned using an explicit algorithm rather than transcribed. (1) Where the authors stated an overall dominant source in the abstract or conclusions, that statement was taken as the label. (2) Where two classes were reported without a stated dominance, or where different sites or seasons yielded different assignments, the study was labeled mixed. (3) Where individual ratios conflicted (for example, Fla/(Fla + Pyr) indicated combustion, while Ant/(Ant + Phe) indicated petroleum), the authors’ own reconciliation was followed, and the discordance is recorded in Supplementary Table S1. (4) Labels derived only from principal component analysis or positive matrix factorization, without diagnostic ratios, were not accepted. Each assignment was verified independently against the primary tables by a second reviewer. Because collapsing a study to one label can cause exposure misclassification that attenuates a true moderator association, we report this as a principal limitation and treat the misclassification-prone mixed category as an aggregation artefact, rather than a mechanistic class.

2.5 Analytical comparability and study appraisal

Unit equivalence does not establish comparability. For every included study, we therefore recorded the congener panel entering the TEQ sum, the toxic-equivalency-factor scheme, the concentration basis, the summary statistic reported, whether TEQBaP and ΣPAH derived from the same samples, and the meaning and independence of n (Table 1). Studies were appraised in five domains: analytical comparability, source-classification validity, sampling design and independence of the unit of analysis, transparency of the summary statistic, and selective reporting, each rated low, moderate, or high concern against pre-defined rules that are stated in full in Table 2 and applied study by study in Table 3. Appraisal was used to interpret and to structure sensitivity analyses, not to weight studies.

TABLE 1. Characteristics, analytical comparability, and extraction provenance of the 18 included studies.

TABLE 2. Pre-defined criteria used to assign concern ratings in each appraisal domain.

TABLE 3. Study-level appraisal against the criteria provided in Table 2 (L = low; M = moderate; H = high concern).

2.6 Effect size and variance derivation

Because environmental TEQBaP is strongly right-skewed, the effect size was ln(TEQBaP), and pooled estimates are back-transformed to geometric means. Sampling variances on the log scale were derived by one of four routes, each with its estimand stated: (i) from a reported arithmetic mean and SD using the delta-method approximation var ≈ CV²/n; (ii) from a vector of site-level values as the variance of the log values divided by the number of sites, which targets the mean of the log values; (iii) from a reported range, using the sample-size-dependent estimator SD = (ln max − ln min)/ξ(n) with ξ(n) = 2Φ⁻¹((n − 0.375)/(n + 0.25)) as given by Wan et al. (2014), and the geometric midpoint as the central value; and (iv) for studies reporting only a central value, imputation of the cohort-median log-SD. We correct here an error in the original submission, which applied a fixed divisor of four, the Hozo et al. (2005) range rule, while citing Wan et al. (2014). The fixed divisor materially overstates precision at small n; with n = 4, for example, ξ(n) ≈ 2.06 rather than 4. All estimates below use the corrected estimator. Because routes (i) and (ii) target subtly different estimands and route (iv) treats an estimated precision as known, the primary model is restricted to routes (i)–(iii), and the analysis is repeated under equal weighting, which requires no variance assumption at all.

2.7 Statistical synthesis

The primary model is a random-effects meta-analysis with the between-study variance τ² estimated using restricted maximum likelihood and confidence intervals computed using the Hartung–Knapp–Sidik–Jonkman method, which is better behaved than DerSimonian–Laird when k is small and heterogeneity is extreme. A Q-profile confidence interval is reported for τ², and the results are additionally presented under DerSimonian–Laird, Paule–Mandel, and equal weighting so that the sensitivity of every conclusion to the estimator is visible. Heterogeneity is summarized using Q, I², τ², and a 95% prediction interval.

Analyses are organized into a short hierarchy. One estimable contrast pyrogenic versus mixed, expressed as a ratio of geometric means with a confidence interval, carries the review’s inference. Everything else, including matrix, sampling environment, continent, and the meta-regression, is exploratory and hypothesis-generating; no multiplicity adjustment is applied to these, and none of them is interpreted as confirmatory because, with 13 studies and several correlated candidate moderators, the information available cannot support stable adjusted inference. The cross-tabulation of source class against the sampling environment is reported so that empty and sparse cells are visible. Robustness is assessed using leave-one-out re-estimation of the source contrast itself rather than of the pooled mean alone. Formal small-study-effect testing was not retained: with k = 13, near-total heterogeneity, source-class clustering, and standard errors constructed via different routes, funnel asymmetry is not identifiable as publication bias, and the Egger regression reported in the original submission has been removed rather than reinterpreted.

Analyses were performed using Python 3.11 with NumPy 1.26, SciPy 1.11, and pandas 2.1.

3 Results

3.1 Included studies and primary-source audit

A total of 67 studies were included in the review, and 16 met the quantitative criteria (Figure 1). The primary-source audit was then repeated in full, with every extracted TEQBaP and ΣPAH value assessed against the page, table, or figure of the publication it came from; the locations are recorded study by study in Supplementary Table S1. The audit changed the evidence base in four ways. First, three studies were removed. Chen et al. (2024) was excluded because its matrix is sewage sludge rather than soil or sediment, and the original table had recorded it as soil. Nechita et al. (2026) was excluded because the values recorded as its TEQBaP range, 0.36–3.99 ng g^-1, are in fact that study’s limit-of-quantification range: in the two-column layout, the sentence reporting the limits of detection and quantification is interleaved with the TEQ equation, and the study’s actual toxic-equivalency values are presented only in a figure, so they are not extractable, and the study fails the eligibility criterion. Abakumov et al. (2021) was excluded because the value recorded for it could not be verified: the study reports BaP-equivalents of 56.2 μg kg^-1 for its single most contaminated point rather than a study-level central value, which is not a comparable estimand. Second, the summary convention of every ΣPAH entry was assessed against that of its TEQBaP entry, and ΣPAH was retained only where the two derive from the same convention. This withdrew the ΣPAH values for Hassaan et al. (2024), which the reviewers had correctly identified, and on the same principle for Deelaman et al. (2023), Toscanesi et al. (2025), Brandão et al. (2026), Zhao et al. (2022), Nawrot et al. (2023), Nasher et al. (2013), and El-Maradny et al. (2025); the entries previously recorded for Parchami et al. (2025) and Chaber and Gworek (2020) could not be located in those publications at all and have also been withdrawn. Third, three ΣPAH values were corrected to the figures actually reported: 318.37, 515.37, and 61.67 ng g^-1 for Zhang and Zhu et al. (2021), Sun et al. (2021), and Cui et al. (2025), respectively. Fourth, the supplementary search added four studies (Nasher et al., 2013; Azimi Yancheshmeh et al., 2014; Raeisi et al., 2016; El-Maradny et al., 2025), one of which is petrogenic-dominant and reports ΣPAH on a convention matching its TEQBaP. We report these corrections in full because they change the evidence base materially and because the direction of every one of them is to reduce, rather than enlarge, what the synthesis can claim.

Of the 18 studies, 12 yielded a directly derivable sampling variance and form the primary model; 14 are available when imputed variances are admitted; and all 18 enter the equal-weights model, which requires no variance. Two studies (Azimi Yancheshmeh et al., 2014; Raeisi et al., 2016) do not report the number of independent sampling units and are therefore restricted to the equal-weights and descriptive strata. Across the 16 studies, the source classes are as follows: 7 pyrogenic, 7 mixed, and 2 petrogenic. These totals refer to the full quantitative set; the corresponding counts in the primary model are as follows: 6 pyrogenic, 5 mixed, and 1 petrogenic because studies with an imputed or underivable variance are excluded from it. Every subgroup k reported in Figure 2 and Table 4 is therefore a primary-model count and is smaller than the class total in Table 1; the two should not be read as inconsistent.

TABLE 4. Pooled ln(TEQBaP) under alternative heterogeneity estimators and weighting schemes.

3.2 Pooled toxic-equivalent concentration and heterogeneity

The pooled TEQBaP under the primary REML + HKSJ model was 104.2 ng g^-1 (95% CI = 32.4–335.6; k = 12), with τ² = 3.27 (95% CI = 1.58–9.44), I² = 99.9%, and a 95% prediction interval of 1.6–6,942 ng g^-1 (Figure 2). The estimate is essentially invariant to the heterogeneity estimator and the weighting scheme (Table 4; Figure 3a), including under equal weighting, which discards the variance model entirely.

We state plainly what this quantity is and is not. With a prediction interval spanning four orders of magnitude, these studies do not estimate a practically coherent common concentration for any defined environmental population, and inverse-variance weighting returns a precision-weighted mean of the sampled study populations rather than a geographically representative mean for soils and sediments worldwide. The pooled value is therefore reported as a descriptive summary of the assembled evidence and the reference point for the moderator analyses; it is not the finding of the review, and we do not interpret its stability under leave-one-out deletion as support for any scientific conclusion.

3.3 The estimable source contrast

Pyrogenic assemblages pooled to 64.7 ng g^-1 (95% CI = 18.3–229.6; k = 6) and mixed assemblages to 114.2 ng g^-1 (95% CI = 6.0–2,162.2; k = 5). Expressed as the quantity of interest, the ratio of geometric mean TEQBaP for pyrogenic relative to mixed assemblages was 0.57 (95% CI = 0.04–8.0; p = 0.639).

This interval is the central result of the review, and it is uninformative in a specific and reportable way. It includes unity, so no difference was detected; however, it also includes ratios far beyond the two-fold margin we regarded as practically meaningful, in both directions. The data are thus compatible with pyrogenic assemblages carrying substantially less, or roughly twenty times more, toxic-equivalent concentration than mixed assemblages. We therefore make no claim of equivalence, and we have removed the description of this contrast as adequately powered: no power calculation supports it, and none is reported here because a post hoc power calculation conditioned on the observed effect would add nothing.

For the petrogenic class, we make no inference at all. Three petrogenic studies were identified across the whole search, only one of which supplies a derivable variance, and their reported TEQBaP values span more than an order of magnitude: a heavily oil-contaminated oilfield soil (Liu et al., 2022), a natural-source Antarctic soil (Abakumov et al., 2021), and a petrogenic-dominant Caspian sediment (Azimi Yancheshmeh et al., 2014). The three-level between-group test reported in the original submission rested on a single-study petrogenic cell and has been withdrawn; it was not a valid class comparison, and interpreting its result as a petrogenic signal and then dismissing it as one contaminated site conflated an outlier assessment with inference about a class that was never estimable. The correct statement is that the theoretically central pyrogenic-versus-petrogenic contrast could not be evaluated and that this remains true after the search was widened by 3 years and a further database stream.

3.4 Exploratory moderator analyses

Secondary moderators are reported for completeness and are hypothesis-generating only. The cross-tabulation of source class against the sampling environment (Figure 3d) shows why: of fifteen source-by-environment cells, seven are empty, and none contains more than three studies, so source class, land-use context, matrix, and contamination intensity cannot be separated in these data. The sampling environment showed the largest apparent differences, with urban studies at the high end, but each environment stratum contains two to four studies and retains near-total residual heterogeneity, and the test is one of several unadjusted secondary comparisons. We therefore no longer describe environment as a genuine moderator, as the key positive result, or as structuring or governing the burden; the earlier claim that a non-significant continent comparison showed continent to be free of confounding was also incorrect and has been removed. The meta-regression on source class and matrix has been withdrawn from the confirmatory results: with 13 studies, several correlated dummy predictors, and empty cells, it cannot support stable adjusted inference, and reporting it alongside the subgroup analyses invited exactly the over-interpretation the reviewers identified.

3.5 Robustness

Leave-one-out re-estimation was applied to the source contrast itself. Omitting any single study leaves the pyrogenic-versus-mixed ratio between 0.28 and 1.18, with every confidence interval still spanning unity and every p-value above 0.4 (Figure 3b). The conclusion is therefore not driven by any one study, but neither does it become informative under any deletion; the stability is a stability of imprecision.

3.6 Mixture potency factor: exploratory feasibility

The mixture potency factor (MPF = TEQBaP/ΣPAH) is presented as a reporting proposal and evaluated as feasibility work, not as a validated forensic indicator. It is a concentration-weighted average toxic-equivalency factor only when the numerator and denominator derive from the same samples, the same concentration basis, and a stated congener set; a ratio of study-level means is not a sample-mean TEF, and different definitions of total PAH change the ratio even when mixture composition does not. Bounded between zero and one under a TEF scheme whose maximum weight is unity, it removes scale but not confounding by analyte coverage, censoring, weathering, matrix, or bioavailability. Values were derivable for five studies: pyrogenic median of 0.124 (k = 1), mixed median of 0.017 (k = 3), and a single petrogenic value of 0.036 (Figure 3c). Pyrogenic and mixed groups did not differ (Mann–Whitney U = 3.0, p = 0.500), and with six or seven studies per group, this test is not evidence of equivalence.

The addition of a petrogenic study reporting total PAH is nonetheless the single most useful outcome of the widened search because in the original submission, neither petrogenic study reported ΣPAH, and the contrast for which the metric was devised could not be computed at all. Its MPF is the lowest of the three classes, which is the direction the compositional argument predicts, but one study cannot evaluate that prediction, and we draw no conclusion from it. The relationship between the two quantities is shown directly in Figure 4a: reported TEQBaP increases approximately in proportion to ΣPAH across three orders of magnitude, with almost every study falling between mixture potency factors of 0.01 and 0.5 and the source classes overlapping throughout that band. This is the confound described in Section 4.2 made visible, and it is why a normalized metric is required before any compositional signal can be isolated. No variance can be attached to individual MPF values because ΣPAH dispersion is essentially never reported, so no pooling was attempted, and no uncertainty is propagated from the numerator to the denominator. Validating MPF requires congener-level data and sample-level calculation, and we set that out as a research recommendation in Section 5 rather than as a result.

4 Discussion

4.1 What the evidence does and does not show

This review set out to evaluate whether forensic source class predicts the toxic-equivalent concentration of PAH-contaminated soils and sediments. The honest answer is that the available study-level evidence cannot decide the question. In the only estimable comparison, pyrogenic and mixed assemblages differed by a factor whose confidence interval spans more than two orders of magnitude; the theoretically central pyrogenic-versus-petrogenic comparison could not be made at all. This is an absence of evidence, and we are careful not to present it as evidence of absence. The distinction matters practically: a null that reflects sparse, heterogeneous data invites better primary reporting, whereas a demonstrated null would license abandoning the hypothesis.

4.2 Why the evidence base is weak

Three structural features limit what any synthesis of this literature can achieve. First, the outcome is compound: TEQBaP = Σ(Cᵢ × TEFᵢ) scales with both absolute concentration and congener composition, so a petroleum point source delivering a large ΣPAH mass can exceed the TEQBaP of a compositionally more potent but less concentrated pyrogenic assemblage. Second, comparability is not established by unit equivalence: the studies here use congener panels ranging from 15 to 23 compounds, and the number and identity of the PAHs quantified materially affect both TEQBaP and ΣPAH. Third, the moderator is a study-level aggregation of site-level evidence, and the mixed category, in particular, is as much an artefact of aggregation as a mechanistic class. Near-total heterogeneity follows naturally from pooling absolute concentrations across Antarctic soils, tropical mangroves, urban catchments, and oilfields; we report it transparently rather than treating it as a nuisance, but we no longer claim that it validates any particular analytical choice.

4.3 Implications for environmental forensic practice

The practical implication is cautionary rather than affirmative. Diagnostic-ratio attribution and hazard assessment should be treated as separate exercises until evidence exists that the first predicts the second because at present, no such evidence exists in either direction for solid matrices. A pyrogenic or petrogenic label should not be read as a toxicity ranking in casework or screening and equally should not be assumed uninformative. Source apportionment retains its established role in provenance and liability questions (Ilić et al., 2021; Farid et al., 2020), which do not depend on any link to toxicity.

4.4 What primary studies should report

The most useful contribution this review can make is to specify what would make the question answerable. Primary studies reporting TEQBaP should state the congener panel and TEF scheme explicitly; report ΣPAH from the same samples and on the same basis; report a central value with dispersion and the number of independent sampling units; retain site-level or sample-level TEQBaP paired with the site-level source assignment rather than only a study-level summary; and report the concentration-normalized ratio (MPF = TEQBaP/ΣPAH) alongside the raw value. Sample-level pairing is the decisive item: it would convert this question from an underpowered between-study comparison into a within-study one, where source and contamination magnitude can be separated and where multilevel models or quantitative source proportions could replace a three-category label.

4.5 Structured narrative synthesis of the wider evidence

The 51 studies that met the review question but not the quantitative criteria are summarized here under a pre-specified structure source attribution, toxicity potential, measured biological effects, and remediation, and their function is to characterize the reporting landscape, not to corroborate the meta-analysis. On source attribution and concentration, regional surveys across Africa, Asia, Europe, and the Americas (Umeh et al., 2023; Okechukwu et al., 2021; Osman et al., 2024; Zaker, 2022; Haris et al., 2024; Bellanova et al., 2022; Matys Grygar et al., 2025; Velandia-Aquino et al., 2024; Wen et al., 2025) report combustion-dominated fingerprints across a concentration range spanning several orders of magnitude, consistent with the dispersion observed in the quantitative subset. On toxicity potential, source-resolved apportionment coupling positive matrix factorization to TEQ (Zhang et al., 2022) and standardized sediment-quality quotients (Long et al., 1995) address related but distinct questions: they grade hazard by magnitude and apportion it to emission processes, and we no longer present them as direct support for the MPF, which they are not. On measured effects, the corpus contains genuine organism-response studies (Meier et al., 2020; Singh et al., 2023; Kim et al., 2026) whose data were not control-referenced in poolable form; these define the evidence gap between toxicity potential and demonstrated harm rather than closing it. On remediation, the pyrogenic-carbon and treatment literature (Li et al., 2016; Chang et al., 2024; Gajić et al., 2018; Narayana et al., 2025) is reported descriptively; the earlier recommendation that remediation be prioritized by normalized potency has been withdrawn as it did not follow from any evidence set assembled here. Bioaccumulation and biomonitoring studies (Ihunwo and Ibezim-Ezeani, 2021; Kristensen et al., 2025; Sun et al., 2025; Styszko et al., 2023) are reported as context; because MPF normalizes measured concentration and does not isolate bioavailability, they cannot validate that property, and the earlier suggestion that they do has been removed.

4.6 Limitations

The review has substantial limitations, several of which are structural rather than remediable by reanalysis. The quantitative evidence base is small (k = 13 in the primary model), and the petrogenic class is not estimable. The outcome is a carcinogenic-potency metric, not a measured biological effect. Source labels are study-level aggregations subject to misclassification, which would attenuate a true association. TEQ definitions vary across congener panels in ways that unit conversion cannot repair. Two variance routes target subtly different estimands, and two studies supply no usable variance at all. Taken together these means, the review should be read as a structured characterization of an evidence gap.

5 Conclusion

In a systematic review of soil and sediment studies reporting benzo[a]pyrene toxic-equivalent concentrations alongside diagnostic-ratio source attribution, no difference was detected between pyrogenic and mixed-source assemblages (ratio of geometric means = 0.57, 95% CI = 0.04–8.0). However, the interval is wide enough to be compatible with substantial differences in either direction, and the pyrogenic-versus-petrogenic contrast, the comparison of primary theoretical interest, could not be estimated because only three petrogenic studies exist in the assembled evidence and only one supplies a derivable variance. We therefore conclude that the present study-level evidence is insufficient to determine whether forensic source class predicts toxic-equivalent concentration rather than that source class fails to predict it. Secondary moderator analyses, including the apparent gradient across sampling environments, are exploratory. The mixture potency factor is offered as a reporting convention requiring validation with compatible congener-level data, not as an established metric. Closing this gap requires primary studies to publish sample-level toxic-equivalent concentrations paired with sample-level source assignments, together with the congener panel, the TEF scheme, and total PAH on a common basis; until then, forensic source attribution and toxicity assessment should remain distinct steps in environmental practice.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/feart.2026.1948291/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Valentin Romanovski, University of Virginia, United States

Reviewed by: Ronan Adler Tavella, Federal University of São Paulo, Brazil

Kunlong Hui, Chinese Research Academy of Environmental Sciences, China