Spatial‐temporal patterns of perfluoroalkyl substances in the biota of the Laurentian Great Lakes: A meta‐analysis

Abstract

To better understand the dynamics of per‐ and polyfluoroalkyl substances (PFAS) in complex freshwater ecosystems, we performed a systematic meta‐analysis of PFAS distributions and spatiotemporal variance in biota of the Laurentian Great Lakes watersheds. We reviewed 50 publications that contained 2489 records (primarily of fish and birds) spanning 42 years of biological sampling. Using this dataset, we built generalized additive models for six compounds—perfluorooctanesulfonic acid (PFOS) and five perfluoroalkyl carboxylic acids (PFCAs)—routinely detected in biological tissues. Estimated concentrations of PFOS, the dominant compound in biota, increased along a lake gradient from west (Lake Superior) to east (Lake Ontario), and PFCA levels also varied across the lakes. Modeled temporal trends of PFOS in biota were highly significant but non‐linear, and may be correlated with industrial production, the timeline of compound phase‐out, food web shifts, and lake‐specific conditions. In the eastern lakes, biotic PFOS concentrations were highly variable through time, spanning one to two orders of magnitude, and model estimates generally declined following industrial phase‐out of the compound. In the western lakes, PFOS levels did not demonstrate substantial changes from lower baseline concentrations. PFOS, but not PFCA, levels biomagnified from primary producers to apex predators across the Great Lakes. Model output indicated that eggs, blood, and liver samples were consistently the most contaminated tissues. Our review also revealed several data constraints in the literature revolving around lake coverage, taxonomic biases, and methodological inconsistencies. Addressing these data gaps will maximize the inferential ability of future synthetic studies of PFAS.

Plain Language Summary

Per‐ and polyfluoroalkyl substances (PFAS), known as “forever chemicals,” can remain in the environment for decades and pose significant concerns for the Laurentian Great Lakes. By analyzing PFAS data for Great Lakes biota over 42 years, we found that PFAS burdens across multiple taxonomic groups showed a distinct west‐to‐east increase from Lake Superior to Lake Ontario. Concentrations within organisms were driven by their position in the food chain, their lifestyle, and the affinity of PFAS for protein and fatty tissues. Voluntary phase‐out of one compound has led to declines in body burdens, showing promising potential for reducing contamination overall in the Great Lakes. This review also identifies research gaps in what taxa have been sampled, where samples have been collected, and how concentration data have been reported, and we provide recommendations for future research.

Abbreviations

GAM

generalized additive model

PFAS

per‐ and polyfluoroalkyl substances

PFCA

perfluoroalkyl carboxylic acid

PFDA

perfluorodecanoic acid

PFDoA

perfluorododecanoic acid

PFNA

perfluorononanoic acid

PFOS

perfluorooctanesulfonic acid

PFSA

perfluoroalkane sulfonic acid

PFTrDA

perfluorotridecanoic acid

PFUnA

perfluoroundecanoic acid

w.w.

wet weight

1. INTRODUCTION

Per‐ and polyfluoroalkyl substances (PFAS) are a family of synthetic organofluorines with some 15,000 recognized compounds (US Environmental Protection Agency, ²⁰²²). Due to their desirable chemical properties, PFAS have been utilized in a number of commercial products since the 1940s (Blake & Fenton, 2020; Buck et al., ²⁰¹¹). However, intense industrial production has also led to the release of PFAS into the environment, with important consequences for ecological systems. PFAS are extremely mobile within the environment: compounds have varying degrees of solubility in water (Sima & Jaffé, 2021), can sorb to sediment (Bhhatarai & Gramatica, 2011; McGarr et al., ²⁰²³; Xiao et al., ²⁰¹⁹; Zhao et al., ²⁰¹⁴), and can also be transported through the atmosphere (D'Ambro et al., ²⁰²¹; Pfotenhauer et al., ²⁰²²; Xia et al., ²⁰²⁴). In the environment, PFAS can accumulate in the bodies of wildlife, leading to contaminant burdens that may detrimentally affect organismal health and behaviors (Coy et al., 2022; Custer et al., ²⁰¹⁷; Fenton et al., ²⁰²¹; Lau et al., ²⁰⁰⁷). Due to their strong carbon‐fluorine bonds, PFAS are also extremely resistant to degradation (Cousins et al., 2020; Lemal, ²⁰⁰⁴; O'Hagan, ²⁰⁰⁸).

The toxicity, mobility, and persistence of this chemical class make PFAS contamination a particular concern for the watersheds of the Laurentian Great Lakes. Formed at the end of the last Glacial Period, ∼10,000 years ago, the Laurentian Great Lakes today provide drinking water for 40 million people in the United States and Canada. Diverse ecological communities occupy the Great Lakes watersheds and provide numerous ecosystem services, including provisionment of food and water resources, management of nutrient regimes, and protection from flooding (Wichert et al., 2005). In contemporary times, the biodiversity of the Great Lakes has been subjected to anthropogenic stressors such as eutrophication, overfishing, the introduction of exotic species, and habitat destruction. Contaminants from industrial, urban, and agricultural sources also enter the system, with recent pressure posed by PFAS inputs. PFAS can reside in the Great Lakes for decades due to the years‐long water retention times of the five lakes (Quinn, 1992), and both point and nonpoint sources are involved in the spread of PFAS (Hu et al., 2016; Y. Lin et al., ²⁰²¹; Moody et al., ²⁰⁰²). The prevalence of these contaminants poses a substantial risk to the Great Lakes’ biota (Remucal, 2019), as PFAS can bioaccumulate and biomagnify in lake food webs (Kannan et al., 2005; J. W. Martin, Whittle, et al., ²⁰⁰⁴; Miranda et al., ²⁰²³).

Over the past 15 years, a growing body of research has worked to synthesize available information on the dynamics of PFAS in aquatic environments, including in the Great Lakes. Reviews have focused on bioaccumulation and biomagnification potential (and their toxicity to biota) across diverse aquatic systems (Ahrens & Bundschuh, 2014; Ankley et al., ²⁰²⁰; Burkhard, ²⁰²¹; Lewis et al., ²⁰²²; Miranda et al., ²⁰²²; C. Wang et al., ²⁰²⁵) or on specific prominent taxa or environmental compartments in the Great Lakes (Furdui et al., 2007; Gewurtz et al., ²⁰¹³; Klečka et al., ²⁰¹⁰; Point et al., ²⁰²¹; Remucal, ²⁰¹⁹). For example, Remucal (2019) found that the ratio of perfluoroalkane sulfonic acid (PFSA) to perfluoroalkyl carboxylic acid (PFCA) concentrations in Great Lakes water, sediment, lake trout tissue (Salvelinus namaycush), and herring gull eggs (Larus argentatus) increased across a west‐to‐east gradient, mirroring shifts in the dominant sources of PFAS inputs to the lakes. Using PFAS samples of air, water, sediment, fish, and birds from across Canada (including from portions of the Great Lakes), Gewurtz et al. (2013) linked higher PFAS concentrations to proximity to urban areas and known waste sector contaminant sources, and found group‐specific variation in PFAS concentrations through time. To date, however, no study has used available data on biotic PFAS concentrations in the Great Lakes to quantitatively infer how contamination has changed through time and space (Glass, 1976; Page et al., ²⁰²¹). Such large and complex land‐water systems may exhibit location‐specific responses to the presence of contaminants, making it difficult to recognize ecological patterns at the local scale (Brandt et al., 2024; Clements et al., ²⁰¹⁶). Robust quantitative inference, across scales and beyond single trophic levels, is needed for the Great Lakes.

To address this knowledge gap, we conducted a meta‐analysis of PFAS distributions in the biota of the Laurentian Great Lakes. We built statistical models, using 42 years of published data (1979–2021), to investigate predictor‐response relationships for PFAS in Great Lakes biota at spatial and temporal scales that are typically intractable for single studies. Specifically, we addressed three hypotheses. First, we hypothesized that watershed concentrations of PFAS would reflect shifts in the sources and historical intensity of contamination across the Great Lakes. We predicted that modeled PFAS concentrations would increase along a west‐to‐east gradient (i.e., from Lake Superior to Lake Ontario), consistent with patterns previously observed for specific taxonomic groups. Second, we hypothesized that changes through time in the modeled concentrations of specific PFAS compounds would align with their industrial phase‐out. Specifically, we predicted that following phase‐out in 2002, modeled concentrations of perfluorooctanesulfonic acid (PFOS) would decline through time across the Great Lakes and that these declines would be more variable in the eastern lakes (i.e., lakes with shorter hydraulic residence times and more industry) than in the western lakes. Third, we hypothesized that the taxa‐ and tissue‐specific accumulation of the six long‐chain PFAS included in this review (see below) would be driven by the known biochemical affinities of these PFAS and the dietary and life history patterns of the organisms in which these PFAS accumulate. Specifically, we predicted higher levels of PFAS contamination in (1) organisms higher up in the food chain (due to biomagnification), (2) organisms associated with benthic sediments (due to the role of contaminated sediment as a PFAS source), and (3) protein and lipid‐rich tissues (due to the lipo‐ and proteinophilic nature of these PFAS). Finally, our analysis also allowed us to identify sampling biases of surveys and studies to date, and to suggest directions for future research.

2. METHODS

2.1. Literature search and data handling

We performed two keyword searches in Web of Science that targeted research related to the Great Lakes, PFAS, and biota. From these searches, we identified a total of 2489 biological samples across 50 independent studies that met our criteria for inclusion in the meta‐analysis (Supporting Information Text S1 and S2; Table S1; Figure S1). Samples were distributed relatively evenly across the five watersheds of the Great Lakes (Figure 1). The compiled dataset included 76 species/genera from eight taxonomic groups and seven trophic levels. Fish (50 species and 1681 samples) and birds (eight species and 633 samples) were the most frequently surveyed taxonomic groups, and the most prominent trophic levels in the data were quaternary consumers (i.e., large‐bodied piscivorous fish like salmon and pike; 15 species and 1163 samples) and piscivorous/insectivorous birds (e.g., gulls and cormorants; six species and 444 samples). For a complete breakdown of sampling distributions across the levels of relevant variables, please see Supporting Information Text S3. Six perfluoroalkanes with sufficient concentration data (units of ng/g wet weight) were selected for analysis (see Supporting Information Text S4 for the specific selection criteria). The six PFAS were PFOS, perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), perfluoroundecanoic acid (PFUnA), perfluorododecanoic acid (PFDoA), and perfluorotridecanoic acid (PFTrDA). PFOS is a PFSA, whereas PFNA, PFDA, PFUnA, PFDoA, and PFTrDA are all in the PFCA subclass. Whenever possible, data from each publication were screened to include concentration values for only the linear isomers of these six compounds.

FIGURE 1.

Comprehensive map of samples (N = 2489) included in this meta‐analysis. Waterbody classifications were determined using the buffered watershed divisions defined by the US Geological Survey (2010) (in blue, with the inner border corresponding to the original watershed boundary and the outer border the buffer; see Text S2 for further details). Lake Superior (n = 362), Lake Michigan (n = 532), Lake Huron (n = 453), Lake Erie (n = 362), and Lake Ontario (n = 780) all have large sample sizes distributed throughout their respective watersheds.

2.2. Statistical analysis

Quantitative analyses were performed in RStudio with R version 4.4.1 (R Core Team, ²⁰²⁴). Since left‐censored values (i.e., values reported as less than the limit of detection; Supporting Information Text S4) can bias model estimates (Helsel, 2006; Newman, ²⁰¹²), we imputed these values by applying a model‐based ordinary Expectation‐Maximization algorithm and a simulation‐based Data Augmentation algorithm from the zCompositions package (Palarea‐Albaladejo & Martín‐Fernández, 2015) (Supporting Information Text S5; Figure S2).

Following imputation, we modeled biotic concentrations of each of the six PFAS from across the Laurentian Great Lakes using generalized additive models (GAMs) from the mgcv package (Wood, 2017), with restricted maximum likelihood estimation. For each contaminant (i.e., model), a scaled t‐distribution (Wood et al., 2016) and log₁₀‐transformed concentration data were found to optimize model fit and performance. Models were built and validated using 80:20 holdout validation, with a single random partition of the data into training and test sets. Specifics on variable formatting, model selection, and fitting can be found in Supporting Information Text S6. All six model functions were specified as follows:

$$\mathrm{lo}{\mathrm{g}}_{10}\mathrm{PFA}{\mathrm{S}}_{\mathrm{i}}\sim \mathrm{t}.\mathrm{scaled}\left({\mathrm{\mu }}_i,v,\mathrm{\sigma }\right)$$ (1)

(2)

$${\mathrm{\epsilon }}_i\sim \mathrm{Normal}\left(0,{\mathrm{\sigma }}^2\right)$$ (3)

$$i=1,2,\text{…},n$$ (4)

where the scaled t‐distribution is defined by three parameters: the shape parameter ν, the scale parameter σ, and the location parameter µ (Equation 1). The value µ_i (calculated for each data point i from 1 to the analyte‐specific sample size n, Equation 4) is determined by a combination of linear and smooth predictors (Equation 2), where β ₀ is the overall intercept of the model, calculated using effects parameterization; β _{1–4( i )} are the coefficient estimates for the levels of the Waterbody variable (the watershed from which the sample was collected, either Superior, Michigan, Huron, Erie, or Ontario); β _{5–6( i )} for Waterbody_Type (the nearest type of waterbody, either inland waters, connecting channels, or lake basin); β _{7–12( i )} for Trophic_Level (the trophic level of the species surveyed, which was assigned using the reported feeding habits and life histories of each species included in the meta‐analysis; see Froese & Pauly, ²⁰²⁴); β _{13–17( i )} for Revised_Tissue (the type of tissue that was collected); f _{0–4( i )} are the watershed‐specific nonlinear smooth functions for the Sampling.Year and Water_Level_sc variables (the sampling year and the water level, scaled and centered, of the lake basin during that sampling year, respectively); f_i (Longitude _i • Latitude _i ) is the two‐dimensional nonlinear smooth function for the two geographic coordinate variables Longitude and Latitude; s_i is a random effect smooth for the Composite variable (whether that sample was a composite tissue sample or not); and ${\mathrm{\epsilon }}_i$ is a normally‐distributed residual error term (Equation 3) (for more information on the variables collected for this meta‐analysis, see Supporting Information Text S2 and S3 and Tables S2 and S3). Wald tests (which produces a χ ² test statistic and corresponding p‐value) were computed with mgcv's anova() function to assess the significance of all parametric and smooth terms (Table S4).

To estimate average concentrations for each group of each predictor variable of interest, we calculated estimated marginal means with the emmeans package (Lenth, 2024), using Tukey multiple comparison tests for pairwise contrasts. Estimated marginal means are calculated by averaging the model's predictions of all possible parameter combinations, beyond just the combinations present in the dataset. This calculation method thus controls for the values of non‐focal predictors and accounts for the asymmetries of discontinuous datasets (i.e., not all trophic levels, tissue types, or waterbody types were surveyed in each year of our dataset). The values that we report can therefore be interpreted as model‐adjusted, geometric means: they were calculated on log₁₀ scale and then back‐transformed to the original scale of biotic concentrations (ng/g wet weight [w.w.]). Estimated marginal means were calculated for categorical predictor variables and were used in conjunction with functions from the gratia package (Simpson, 2024) to visualize the nonlinear relationships between smooth predictors and PFAS concentrations. For our discussion of modeled temporal trends, we chose to focus our analysis on the period of time from 1990 onward, due to the limited sample sizes in the lakes prior to this date (Figures S3 and S4). Estimated marginal means were calculated only for the years that were represented in the data frame, and the starting point for each watershed's trendline was the first year that that watershed had been sampled. Splines were fit to these geometric means to help visualize the smooth temporal trends.

For the waterbody type, tissue type, and trophic level predictor variables, we performed a scale transformation to the estimated marginal means: for each variable, we summed concentration estimates across all levels and then divided each estimate by the sum, converting concentrations to percentages. This approach allowed us to visualize the relative distribution of the analyte across the modeled compartments (i.e., across the levels of the predictor variable), as well as how the distribution changed with increasing C‐F chain length across the five PFCA models. This scale conversion was critical for comparing patterns across analytes while controlling for compound‐specific variation in concentration estimates.

3. RESULTS AND DISCUSSION

3.1. PFAS analyte trends

Across the variables analyzed in these GAMs (Tables 1 and 2 and Tables S5 and S6), the PFOS model estimated higher average concentrations in biota than any of the models constructed for the PFCA subgroup of compounds (PFNA, PFDA, PFUnA, PFDoA, or PFTrDA), which is consistent with the widely documented prevalence of PFOS in biota worldwide (Chen et al., 2021; Giesy & Kannan, ²⁰⁰¹; Houde, Martin, et al., ²⁰⁰⁶; Sun et al., ²⁰²³). Additionally, we found a chain length‐dependent trend in concentration estimates across the PFCA models. Specifically, modeled concentrations followed a unimodal shape across the five GAMs (i.e., as chain length increased): values peaked around PFUnA (C₁₁), while the models for PFCAs with both shorter and longer chains estimated lower concentrations. As PFAS chain length increases, bioaccumulative potential also increases due to the augmenting hydrophobicity and lipophilicity of the compound (Gagliano et al., 2020). However, PFAS with longer chain lengths also exhibit decreasing bioavailability due to the higher affinity of these compounds for organic matter in the sediment and for suspended particulate matter (Bertin et al., 2014; Higgins & Luthy, ²⁰⁰⁶; Lewis et al., ²⁰²²; Prevedouros et al., ²⁰⁰⁶). Thus, for the PFCAs with a perfluoroalkyl chain longer than 11 carbons, bioaccumulative potential is limited by bioavailability, preventing contaminant burdens from reaching levels comparable to those of PFUnA. This mechanism, which has been observed elsewhere (Lewis et al., 2022; J. W. Martin et al., ²⁰⁰³; J. W. Martin, Whittle, et al., ²⁰⁰⁴), would explain our findings of a chain length‐dependent pattern in mean biotic concentration estimates across the five PFCA models.

TABLE 1.

Mean biotic concentrations of per‐ and polyfluoroalkyl substances (PFAS) (ng/g wet weight) [95% CI] and patterns of significance for the five delineated watersheds of the Laurentian Great Lakes (all biota combined).

Estimated marginal means [95% CI]	PFOS (C8)	PFNA (C9)	PFDA (C10)	PFUnA (C11)	PFDoA (C12)	PFTrDA (C13)
Lake Superior	5.74 [1.54, 21.3]	0.766 [0.182, 3.23]	0.360 [0.112, 1.16]	0.967 [0.284, 3.29]	0.854 [0.265, 2.75]	0.657 [0.178, 2.42]
Lake Michigan	15.2 [5.76, 40.0]	0.306 [0.0921, 1.02]	0.885 [0.351, 2.23]	3.62 [1.09, 12.0]	1.41 [0.480, 4.14]	0.877 [0.175, 4.39]
Lake Huron	14.1 [7.10, 28.0]	0.843 [0.514, 1.39]	2.41 [1.51, 3.83]	4.46 [2.75, 7.25]	2.58 [1.38, 4.81]	1.04 [0.636, 1.69]
Lake Erie	94.0 [10.3, 855]	0.511 [0.0628, 4.17]	0.910 [0.127, 6.53]	1.63 [0.239, 11.1]	1.41 [0.223, 8.88]	0.565 [0.130, 2.46]
Lake Ontario	184 [19.2, 1774]	9.66 [1.03, 90.8]	2.92 [0.356, 24.0]	4.89 [0.654, 36.6]	2.47 [0.381, 16.1]	0.739 [0.165, 3.31]
Pairwise contrasts (p‐values)	PFOS (C8)	PFNA (C9)	PFDA (C10)	PFUnA (C11)	PFDoA (C12)	PFTrDA (C13)
Lake Erie/Lake Huron	0.464	0.989	0.862	0.836	0.969	0.917
Lake Erie/Lake Michigan	0.525	0.991	0.999	0.950	0.999	0.993
Lake Erie/Lake Ontario	0.446	<0.001***	0.164	0.295	0.884	0.960
Lake Erie/Lake Superior	0.175	0.997	0.919	0.990	0.990	0.999
Lake Huron/Lake Michigan	0.999	0.398	0.186	0.997	0.832	0.999
Lake Huron/Lake Ontario	0.188	0.203	0.999	0.999	0.999	0.991
Lake Huron/Lake Superior	0.703	0.999	0.0104*	0.0952°	0.389	0.945
Lake Michigan/Lake Ontario	0.231	0.0365*	0.821	0.999	0.983	0.999
Lake Michigan/Lake Superior	0.496	0.608	0.429	0.275	0.929	0.998
Lake Ontario/Lake Superior	0.0577°	0.294	0.393	0.635	0.863	0.999

Note: Tukey multiple comparison tests, with the corresponding results (p‐values) of significance testing (α = 0.05), are reported below for all six compounds.

Abbreviations: CI, confidence interval; PFDA, perfluorodecanoic acid; PFDoA, perfluorododecanoic acid; PFNA, perfluorononanoic acid; PFOS, perfluorooctanesulfonic acid; PFTrDA, perfluorotridecanoic acid; PFUnA, perfluoroundecanoic acid.

° < 0.1. * < 0.05. *** < 0.001.

TABLE 2.

Mean tissue concentrations (ng/g wet weight) [95% CI], relative amounts (% in parentheses; values estimated through scale transformation of marginal means), and patterns of significance for six per‐ and polyfluoroalkyl substances (PFAS) in the different tissue compartments sampled from organisms of the Laurentian Great Lakes (all biota combined). Note that percentages for specific tissues do not sum to 100%, since not all tissues (i.e., “Misc. Tissue”; see Table S2) used in the percentage calculations are reported herein.

Estimated marginal means (%) [95% CI]	PFOS (C8)	PFNA (C9)	PFDA (C10)	PFUnA (C11)	PFDoA (C12)	PFTrDA (C13)
Eggs	126 (41.3) [43.9, 360]	2.07 (20.6) [0.703, 6.09]	3.16 (24.8) [1.20, 8.33]	10.9 (34.8) [4.19, 28.1]	7.63 (39.2) [3.08, 18.9]	5.49 (52.8) [2.23, 13.5]
Blood	86.7 (28.5) [30.4, 248]	5.50 (54.7) [1.84, 16.4]	5.16 (40.6) [1.93, 13.8]	12.5 (40.2) [4.73, 33.1]	5.30 (27.2) [2.12, 13.3]	3.38 (32.5) [1.35, 8.43]
Liver	54.2 (17.9) [18.9, 156]	0.785 (7.8) [0.262, 2.36]	2.80 (22.0) [1.05, 7.47]	4.83 (15.5) [1.84, 12.7]	3.79 (19.5) [1.51, 9.53]	0.752 (7.2) [0.295, 1.92]
Whole organism homogenate	8.93 (2.9) [3.15, 25.3]	1.02 (10.2) [0.359, 2.90]	0.286 (2.3) [0.111, 0.736]	0.618 (2.0) [0.245, 1.56]	0.288 (1.5) [0.121, 0.690]	0.225 (2.2) [0.0947, 0.536]
Muscle	5.01 (1.7) [1.76, 14.3]	0.252 (2.5) [0.0872, 0.727]	0.155 (1.2) [0.0597, 0.405]	0.417 (1.3) [0.164, 1.06]	0.172 (0.9) [0.0707, 0.418]	0.147 (1.4) [0.0612, 0.353]
Pairwise contrasts (p‐values)	PFOS (C8)	PFNA (C9)	PFDA (C10)	PFUnA (C11)	PFDoA (C12)	PFTrDA (C13)
Blood/eggs	0.00222**	<0.001***	0.0118*	0.944	0.155	0.197
Blood/liver	0.0115*	<0.001***	0.0147*	<0.001***	0.492	<0.001***
Blood/muscle	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***
Blood/whole organism homogenate	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***
Eggs/liver	<0.001***	<0.001***	0.975	<0.001***	<0.001***	<0.001***
Eggs/muscle	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***
Eggs/whole organism homogenate	<0.001***	0.0028**	<0.001***	<0.001***	<0.001***	<0.001***
Liver/muscle	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***
Liver/whole organism homogenate	<0.001***	0.821	<0.001***	<0.001***	<0.001***	<0.001***
Muscle/whole organism homogenate	<0.001***	<0.001***	<0.001***	<0.001***	<0.001***	0.00640**

Note: Tukey multiple comparison tests for pairwise tissue comparisons, with the corresponding results (p‐values) of significance testing (α = 0.05), are reported below for all six compounds.

Abbreviations: CI, confidence interval; PFDA, perfluorodecanoic acid; PFDoA, perfluorododecanoic acid; PFNA, perfluorononanoic acid; PFOS, perfluorooctanesulfonic acid; PFTrDA, perfluorotridecanoic acid; PFUnA, perfluoroundecanoic acid.

*< 0.05. **< 0.01. ***< 0.001.

3.2. Temporal trends of PFOS

Modeled (geometric) mean concentrations of PFOS in all biota of the Great Lakes demonstrated distinct, watershed‐specific temporal trends over the 42‐year sampling period (Figure 2) that are likely influenced by human activity, natural processes, and co‐occurring stressors. For the watersheds of Lake Michigan (χ ² [df] = 0.043 [1.001], p = 0.838, range ∼0.84 ng/g w.w.) and Lake Superior (χ ² [df] = 23.898 [3.166], p < 0.001, range ∼4.9 ng/g w.w.), model‐estimated concentrations showed minor, even if significant, fluctuations through time. In Lakes Erie, Ontario, and Huron, temporal trends were all significantly nonlinear, with more sizable changes in estimated concentrations over the course of the sampling period (Erie: χ ² [df] = 24.162 (5.778), p < 0.001, range ∼104.7 ng/g w.w.; Ontario: χ ² [df] = 118.549 [6.603], p < 0.001, range ∼255.4 ng/g w.w.; Huron: χ ² [df] = 16.794 [5.076], p = 0.0121, range ∼20.8 ng/g w.w.) (Table S4; Figure 2 and Figure S5).

FIGURE 2.

Comparisons of temporal trends in perfluorooctanesulfonic acid (PFOS) concentrations (ng/g wet weight) in the different watersheds of the Great Lakes from 1979 to 2021 (all biota combined). Estimated yearly averages from the generalized additive model for the full dataset are plotted over the sample point distribution (N = 2474 because 15 samples did not quantify PFOS concentrations). Splines were fit to the averages to help visualize the watershed‐specific temporal trends; we focused the analysis on the period of time from 1990 onward due to the limited sample sizes for the lakes prior to that date. The interval during which PFOS was voluntarily phased out of production in the United States (2000–2002) is highlighted in orange. For full plots of the watershed‐specific temporal trends of PFOS, with 95% confidence intervals, see Figure S5.

Because PFAS are very persistent in the environment, modeled declines in PFOS suggest that exports are now exceeding inputs to the system, a trend consistent with lake‐wide PFAS budgets assembled by Xia et al. (2024). Notably, PFOS was voluntarily phased out of production in the United States from 2000 to 2002 by the major US manufacturer (US Environmental Protection Agency, ²⁰⁰⁰). The temporal trends in Lakes Ontario and Erie reflect this phase‐out of the contaminant. Concentrations peaked after the year 2000 (2002 in Ontario and 2006 in Erie), declined through the decade, and then plateaued from 2010 to 2015, declining again in the tissues of wildlife in recent years (2015–2021). Lake Huron showed a subtler decline following its post‐2002 peak, while Lakes Superior and Michigan did not demonstrate any such decline.

These contrasting temporal patterns appear to be linked to the lakes’ hydraulic residence times: that is, the amount of time that a substance (e.g., water and solutes) spends in a waterbody before being flushed out through connecting channels. Lakes Erie and Ontario, the shallowest lakes, have the shortest residence times (2.7–7.5 years; Quinn, ¹⁹⁹²), allowing for relatively rapid flushing of PFAS from these systems after inputs were restricted (Xia et al., 2024). In contrast, Lakes Michigan and Superior have much longer residence times (62–173 years; Quinn, ¹⁹⁹²), and as a result, PFOS inputs can persist for many decades. Lake Huron, with its intermediate residence time (21 years; Quinn, ¹⁹⁹²) and hydrological connections to both pairs of lakes, displays a hybrid pattern (Figure 2 and Figure S5): it has relatively low concentrations through time, like Superior and Michigan, but more perceptible fluctuations and a post‐2002 peak, similar to Ontario and Erie.

In addition to hydrologic residence times, temporal patterns prior to phase‐out (i.e., pre‐2000; Figure 2) may be tied to lake‐wide shifts in the structure of food webs caused by species invasion. During the 1990s, Lakes Ontario and Erie demonstrated a parabolic pattern of biotic concentrations, with declines in estimated average PFOS levels until about 1996—1997. This timeframe aligns with the invasion of dreissenid mussels into the Great Lakes, which shifted food web pathways from nearshore (Diporeia spp.) to offshore (Mysis spp.) pelagic invertebrate food sources (Edlund et al., 2021; O'Gorman et al., ²⁰⁰⁰; Rogers et al., ²⁰¹⁴). If the nearshore environment and associated organisms were more heavily contaminated with PFOS than offshore dietary sources, as others have posited (Furdui et al., 2008; J. W. Martin, Whittle, et al., ²⁰⁰⁴), the declines estimated by our model could be explained by this change to the food web. However, the sample size of our dataset is limited during the early 1990s, and modeled mean concentrations in Lake Ontario fall above any of the reported values from the watershed. While the statistical methods of this study were chosen for their ability to handle discontinuous datasets, and while food web shifts are a plausible biological explanation of the modeled trends, we acknowledge that inherent limitations in data availability may have contributed to the parabolic pattern of our model's estimates.

Overall, the variability in temporal trends across these lakes is likely a product of (1) the legacy of PFOS from heavy industrial production that remains in lake water and sediments, (2) continued inputs of these contaminant legacies into lake watersheds from upstream environmental and proximate, small‐scale industrial sources, (3) removal of PFOS through water outflow and potentially through biota leaving the system, (4) changes in the biotic communities inhabiting these systems, and (5) the industrial phase‐out of PFOS and its regulated precursors by large‐scale manufacturers.

3.3. Watershed comparisons

Model results revealed differing spatial trends between the biotic concentrations of PFOS and the five PFCAs. Watershed was a significant predictor variable in the PFOS model (χ ² [df] = 11.63 [4], p = 0.0203) (Table S4), and while pairwise contrasts of watersheds were generally not significant (with the exception of the marginally significant Lake Ontario/Lake Superior contrast; p = 0.0577), modeled PFOS concentrations increased from west to east: Lake Superior < Lake Michigan ≅ Lake Huron < Lake Erie < Lake Ontario (Table 1; Figure 3 and Figure S6).

FIGURE 3.

Estimated concentrations of six per‐ and polyfluoroalkyl substances (PFAS) compounds (ng/g wet weight) in samples from the five watersheds of the Laurentian Great Lakes (all biota combined). Values are presented as absolute concentrations, and different analytes are color‐coded. The five watersheds are ordered from west (Lake Superior) to east (Lake Ontario). PFDA, perfluorodecanoic acid; PFDoA, perfluorododecanoic acid; PFNA, perfluorononanoic acid; PFOS, perfluorooctanesulfonic acid; PFTrDA, perfluorotridecanoic acid; PFUnA, perfluoroundecanoic acid.

West‐to‐east variation in PFAS signatures across the Great Lakes has been noted by others. Y. Lin et al. (2021) documented increasing PFOS concentrations from west to east in fish collected in 2010 by the EPA National Coastal Condition Assessment, as did Furdui et al. (2007) for individual whole body homogenates of lake trout. Remucal (2019) reported similar spatial patterns for ∑₁₂PFAS in water, surface sediment, and biota samples (lake trout homogenates and herring gull eggs), with these patterns typically driven by PFOS. In general, the west‐to‐east trend is likely attributable to contemporary and historical patterns of contamination in the Great Lakes (Boulanger et al., 2005; Murphy et al., ²⁰¹²; Scott et al., ²⁰¹⁰), spatial variation in the dominant sources of PFAS to the lake basins (Remucal, 2019; Xia et al., ²⁰²⁴), and lake‐specific hydraulic residence times (Quinn, 1992).

A potential advantage of our methodological approach can be shown when we compare our modeled mean watershed concentrations for PFOS in all biota and tissue types with trends in Remucal's (2019) empirical averages for PFOS in lake trout whole‐body homogenates (∼6 vs. 4 ng/g w.w. [Superior]; 15 vs. 14 [Michigan]; 14 vs. 29 [Huron]; 94 vs. 109 [Erie]; and 184 vs. 75 [Ontario]). Despite methodological (modeled vs. empirical estimates) and sample size (50 vs. nine studies) differences, a consistent west‐to‐east gradient is evident in both datasets. Yet one notable difference is that for lake trout, Remucal (2019) reported higher concentrations in Lake Erie than in Lake Ontario, a finding that contrasts with our model's results and with her empirical data for other compartments (water, sediment, and herring gull eggs). She attributed this result to the low number of lake trout samples from Lake Erie. The breadth of our meta‐analysis, which integrates more studies and has a larger number of samples, may have helped correct for this sample‐size‐induced deviation from the broader west‐to‐east spatial pattern.

While PFOS showed a clear west‐to‐east increase across the Great Lakes, modeled estimates of watershed PFCA concentrations in biota did not demonstrate a clear spatial trend. Rather, for all five compounds, concentrations tended to increase from Lake Superior through Huron, declined in Lake Erie, and were typically highest in Lake Ontario (Table 1; Figure 3 and Figure S7). We detected few pairwise differences in PFCA concentrations between lakes, especially as PFCA chain length increased (Table 1). These findings could be due to a number of factors, such as the hydraulic residence times of the lakes (i.e., Lake Erie may “flush” contaminants faster than other lakes due to its shorter hydraulic residence time) or the chain‐length‐dependent carbon‐water partitioning dynamics of these compounds (Higgins & Luthy, 2006; Prevedouros et al., ²⁰⁰⁶; Xiao et al., ²⁰¹⁹). PFCA sorption dynamics drive down the bioavailability of this subgroup as chain length increases, which could produce lower biotic concentrations that preclude statistical differentiation. These dynamics could also restrict peaks in PFCA concentrations to areas in the Great Lakes system where corresponding sediment concentrations are high, such as industrial hotspots with a significant load of proximate point sources. As Y. Lin et al. (2021) suggested, peaks in PFNA contamination can be correlated with nearby industrial manufacturing sites in the Great Lakes. PFCA model results could also be related to specific exposure pathways for this long‐chain subgroup (e.g., atmospheric transport of precursors), which can homogenize contamination profiles across large geographic ranges (D'Ambro et al., ²⁰²¹; Pfotenhauer et al., ²⁰²²; Scott et al., ²⁰¹⁰; Z. Wang et al., ²⁰¹⁴).

3.4. Waterbody type

We contrasted biotic measurements from within lake basins to estimates from channels connecting the lakes and to estimates from inland waters (e.g., tributaries and inland lakes) that ultimately feed the lake basins. Model estimates of biotic PFAS concentrations were similar between the lakes and their connecting channels, while the inland waters had significantly lower concentrations than their lakes (all six contrasts with p < 0.005) and, in some instances, lower concentrations than connecting channels (two of six contrasts with p < 0.05) (Tables S4 and S5).

The consistently lower estimated concentrations of PFAS in the biota of tributaries and inland lakes compared to the biota of the Great Lakes is surprising, given the widespread contamination of such systems from nearby point sources (de Solla et al., ²⁰¹²; Moody et al., ²⁰⁰²; Oakes et al., ²⁰¹⁰). However, our findings suggest that these water bodies act more like rapid transporters of contaminants to the receiving waters of the Great Lakes, rather than as long‐term reservoirs of PFAS themselves. Indeed, Balgooyen and Remucal (2022) implicated tributary discharge as a major contributor of PFAS to the lake basins, noting that while abiotic compartments in these tributaries reflected local point sources, significant desorption of PFAS from tributary sediments could occur when in contact with lake basin water. Thus, even with exposure to PFAS pulses from proximate sources, including wastewater, industrial sites, military bases, and airports, the rapid dispersal of these compounds in inland freshwater systems, through high discharge rates and low retention times, might limit the bioavailability of PFAS.

3.5. Patterns across trophic levels

Trophic position was a significant predictor variable of biotic PFAS concentrations in all six models (p < 0.001) (Table S4). For PFOS, modeled concentrations increased across trophic levels in the Great Lakes, such that primary producers < primary consumers < piscivorous/insectivorous birds < secondary consumers ≅ tertiary consumers < quaternary consumers < apex predators (Table S6; Figure 4). This overall trend in estimated PFOS concentrations indicates biomagnification across the full Great Lakes food web, with two orders of magnitude difference between primary producers (1.4 ng/g w.w.) and apex predators (105.8 ng/g w.w.).

FIGURE 4.

Log₁₀ distribution of biotic tissue samples measured for concentrations of perfluorooctanesulfonic acid (PFOS) (ng/g wet weight) across the seven different trophic levels of a generalized Great Lakes food web. Darker areas of the distribution represent higher densities of samples. Overlaid are the estimated marginal means (red dots), with associated 95% confidence intervals (vertical lines), for each trophic level, predicted using the generalized additive model built for PFOS concentrations. The x‐axis labels signify the following: PP, primary producer; PC, primary consumer; SC, secondary consumer; TC, tertiary consumer; QC, quaternary consumer; PIB, piscivorous/insectivorous bird; AP, apex predator. Created in BioRender P. O. L. Martin et al., in 2026 (https://BioRender.com/agrxn66).

In contrast to patterns for PFOS, trend analysis across trophic levels was variable for the PFCA models (Table S6). At lower trophic levels, we found evidence of biomagnification from primary producers to secondary consumers. However, models showed no clear increase in estimated PFCA concentrations above secondary consumers. Interestingly, scale‐transformed model estimates (i.e., percentages; see Section 2.2) showed that as chain length increased across the five models, the percentage of the analyte detected in apex predators and piscivorous/insectivorous birds (i.e., air‐breathing animals) decreased (Table S6). In contrast, in the aquatic portion of the food web, PFCA percentages in the secondary, tertiary, and quaternary consumers increased as chain length increased, and model‐estimated concentrations peaked in the secondary and quaternary consumers (Table S6). Primary producers and primary consumers did not have a discernible chain‐length‐dependent pattern of accumulation and contained low absolute (ng/g w.w.) and relative (%) estimated concentrations of PFCAs (Table S6). Overall, our models suggest that for these five long‐chain PFCAs, more of the environmental contaminant burden in the Great Lakes is contained in higher aquatic trophic levels that closely associate with the sediment (i.e., secondary consumers such as suckers, redhorse, drum, goby, crayfish, and sculpin) (Table S3).

Although long‐chain PFCAs have been documented to bioaccumulate and biomagnify in certain environments and food webs (Houde, Bujas, et al., 2006; Kelly et al., ²⁰⁰⁹; Loi et al., ²⁰¹¹; J. W. Martin, Smithwick, et al., ²⁰⁰⁴; Ren et al., ²⁰²²), our findings illustrate the complex nature of PFAS movement across trophic levels. Models estimated patterns of PFCA uptake at the top of the food web, across fish, mammals, and birds, that do not necessarily demonstrate increases in contamination from the base of the food chain (Conder et al., 2008; J. W. Martin, Smithwick, et al., ²⁰⁰⁴; Miranda et al., ²⁰²¹; Sinclair et al., ²⁰⁰⁶; Tomy et al., ²⁰⁰⁴). This lack of clear biomagnification of PFCAs could be influenced by three general mechanisms (see also Miranda et al., ²⁰²²).

First, the limited sampling efforts for apex predators have been biased toward birds: of the 202 datapoints compiled in this meta‐analysis for the “Apex Predator” group, only 13 come from reptiles or mammals, and tissues for many predatory species from the Great Lakes ecosystem, including those of river otter, black bear, and coyote, have no measurements for PFAS content to date (Supporting Information Text S3; Figures S3 and S4). Even those species that are represented in the data frame have been sampled only sporadically through time and across the watersheds of the Great Lakes. Second, different sources of contamination may exist between aquatic and terrestrial species. For example, birds have greater mobility than top fish predators, and they integrate PFAS from both aquatic and terrestrial sources, potentially across a larger geographic area. Third, lower trophic levels in the aquatic food web are often more closely tied to the sediment, a notable ambient sink and source of PFAS, especially of long‐chain PFCAs (Prevedouros et al., 2006; Xiao et al., ²⁰¹⁹). The importance of the sediment exposure pathway is supported by our scale‐converted estimates (i.e., percentages): as C‐F chain length increased, our models estimated higher relative amounts of the analyte in the aquatic portion of the food web (Table S6). This pattern is mirrored by the sorption dynamics of the PFCA group—as chain length increases, the affinity of PFCAs for the sediment also increases (Bertin et al., 2014; Higgins & Luthy, ²⁰⁰⁶; Lewis et al., ²⁰²²; Xiao et al., ²⁰¹⁹). These parallel trends suggest that sediment associations may be driving patterns of accumulation, producing concentration peaks in the aquatic secondary and quaternary consumers and higher relative amounts of PFCAs in the sediment‐proximate, aquatic portion of the food web. Thus, the proximity of particular taxa to the sediment, especially in areas of the Great Lakes where sediments are heavily contaminated from nearby point sources, can interact with dietary exposure to influence patterns of PFCA accumulation across food webs (Langberg et al., 2020; J. W. Martin, Whittle, et al., ²⁰⁰⁴; Miranda et al., ²⁰²³; Ren et al., ²⁰²²). These patterns can become even more complicated at finer taxonomic scales: while sculpin and goby are both labeled as secondary consumers in this meta‐analysis, due to their similar life histories and diets (Froese & Pauly, 2024), prior research has shown how disparate PFAS accumulation can be between the two species (Asher et al., 2012; Miranda et al., ²⁰²³; Ren et al., ²⁰²²). Broad‐scale estimates of contamination across trophic levels may therefore be complemented by generic or species‐specific patterns of PFAS accumulation.

3.6. Differential tissue accumulation

Tissue type was a significant predictor of contaminant concentrations in all six models (p < 0.001) (Table S4). In our models, PFOS had the highest estimated concentrations in the different tissue samples, a result similar to other studies (Conard et al., 2022; George et al., ²⁰²³; J. W. Martin, Smithwick, et al., ²⁰⁰⁴; Verreault et al., ²⁰⁰⁵). Across all six models, the highest estimated PFAS concentrations were found in fish and bird eggs, blood, and liver, while the tissue compartments with the lowest concentrations were whole‐organism homogenates and muscle tissue (Table 2). These results agree with the known biochemical/binding affinities of PFAS, compounds that form strong associations with hepatic proteins (e.g., fatty‐acid binding proteins), albumin/albumin‐like proteins in blood plasma and eggs, egg fat stores, and egg yolk proteins (Bangma et al., 2022; Dassuncao et al., ²⁰¹⁹; Fedorenko et al., ²⁰²¹; Jones et al., ²⁰⁰³; F. Wang et al., ²⁰¹⁹; Zhang et al., ²⁰¹³).

In addition to showing preferential accumulation within specific organs, our models also indicate that the partitioning of PFAS within the body may be chain‐length dependent. PFCAs with 9–11 carbons (C_9–11: PFNA, PFDA, and PFUnA) had the highest estimated mean concentrations in blood > fish and bird eggs > liver, while for the longest PFCA carbon chains (C_12–13: PFDoA and PFTrDA), as well as for PFOS (C₈), modeled concentrations were highest in eggs > blood > liver, suggesting that PFAS uptake in these tissues may depend on carbon chain length. When concentrations are normalized to percentages (see Section 2.2), we see that as carbon chain length increased, the percentage of the analyte in eggs substantially increased (from 20.6% [PFNA] to 52.8% [PFTrDA]; Table 2). The binding affinity of these long‐chain compounds to proteins in the liver and the blood, coupled with their lipophilicity, likely enables their transfer to eggs, as has been shown in various study organisms (Berger, 2008; Conard et al., ²⁰²²; Gebbink & Letcher, ²⁰¹²; Lewis et al., ²⁰²²; Ricolfi et al., ²⁰²⁴; Shi et al., ²⁰¹⁸; Zhang et al., ²⁰¹³). Furthermore, maternal exposure to PFAS, as well as the biophysiological complexities of the transfer process (e.g., the variability introduced from diet, maternal body condition, egg traits, clutch size, taxonomic differences, etc.), will also influence concentration patterns in eggs (Conard et al., 2022; Gebbink & Letcher, ²⁰¹²; Ricolfi et al., ²⁰²⁴).

4. LIMITATIONS, RESEARCH GAPS, AND RECOMMENDATIONS

While our methodology allowed us to make a systematic review of the literature and derive predictor‐response relationships for PFAS concentrations in Great Lakes biota, there are several limitations to the current work. GAMs can integrate data over large spatial and temporal scales, representing powerful tools for statistical inference, but they are not mechanistic models. While we can posit potential explanations for the results produced from statistical analysis, these explanations are only working hypotheses. More complex models, built explicitly to incorporate the physical and biochemical dynamics of PFAS, are needed to simulate the behavior of these compounds in multimedia environments and to make definitive conclusions about the underlying mechanisms of temporal, spatial, and biotic concentration patterns (Eichler et al., 2022; Fischer et al., ²⁰²⁵; J. Lin et al., ²⁰²³; Lyu et al., ²⁰²²). Further, our models do not include every variable that should be accounted for when characterizing PFAS contamination within biota of the Great Lakes: not all tissue types had enough samples to be included in the model (e.g., adipose tissue, brain, gastrointestinal tract, etc.), nor could more taxonomically or methodologically explicit comparisons be made (e.g., between species or taxonomic groups within the same trophic level or between different analytical methods with varying degrees of quantitative accuracy). This limitation reflects the absence of and points to the need for larger and more comprehensive datasets.

Our meta‐analysis identified three critical gaps in data coverage and reporting within the existing literature. First, the work revealed a spatial gap in biological sampling for PFAS contamination within the Great Lakes. We noted a lack of published data on PFAS concentrations in biota from the Canadian shores of Lakes Superior, Huron, Erie, and Ontario, as well as the St. Lawrence River (Figure 1). To some extent, at least in Canada, this pattern reflects less populated areas of the lakes, but efforts should be made to increase coverage of these under‐sampled areas.

Second, we found a taxonomic gap with respect to the groups of organisms commonly surveyed. Research on PFAS in the Great Lakes system remains predominantly focused on contamination patterns in fish, likely because of their role as bioindicators, the direct pathway to human exposure, and the ease of sample collection. More comprehensive coverage of lower trophic levels (e.g., aquatic plants and invertebrates) and higher trophic levels (e.g., amphibians, reptiles, mammals, and certain bird species) is needed to characterize the full extent of PFAS contamination in aquatic‐terrestrial food webs (Supporting Information Text S3; Figures S3 and S4). Better resolution of these biological parameters would help to clarify the context‐dependency of contaminant accumulation (Clements et al., 2012) and to characterize the effects of certain migratory species (e.g., salmon, terns, eagles, walleye, etc.) on the movement of PFAS within the watersheds of the Great Lakes (Brandt et al., 2024).

Third, we encountered a methodological gap that is centered on the standardization of methods, data accessibility, and the completeness of metadata. Specifically, we noted inconsistent reporting of sampling locations (e.g., latitude–longitude coordinates), collection dates, and biological information (e.g., weight, length, body condition, sex, and stable isotopes). We recommend reporting all available sampling and methodological information, as well as making any biological data available on an individual sample basis, as opposed to reporting sample pool‐wide averages. We provide additional recommendations for best practices in reporting data for re‐use by other researchers in Supporting Information Text S7.

Together, these issues highlight the need for spatially complete, taxonomically robust, and clearly reported PFAS monitoring efforts. Addressing these three gaps will enhance our understanding of which variables drive PFAS accumulation in biota, how PFAS move through complex food webs across the Great Lakes at the land–water interface (Kraus et al., 2021), and how biotic concentrations may vary through space and time in conjunction with contamination sources. Data generated with these best practices in mind will also help identify priority areas where concerted industrial regulation, source control, and environmental remediation efforts are needed.

5. CONCLUSIONS AND ENVIRONMENTAL IMPLICATIONS

Our review represents the first systematic investigation of the spatiotemporal dynamics of PFAS in all major trophic levels of the Great Lakes. We applied GAMs to the concentration data of six bioaccumulative PFAS (i.e., PFOS and the five PFCAs), using nearly 2500 observations collected over 42 years. Across the Great Lakes, declines in biotic concentrations of PFOS aligned with the chronology of its industrial phase‐out, while watershed‐specific hydrologic parameters, such as hydraulic residence time, modulated these responses. Specifically, the temporal trends of PFOS differed between the western and eastern lakes. Mean estimates in biota of the western lakes (Lakes Superior and Michigan) were relatively stable, while patterns in biota of the shallow, industrially impacted eastern lakes (Lakes Erie and Ontario) showed variable, nonlinear declines. Temporal trends are likely therefore a product of complex interactions within individual watersheds involving food web shifts, industrial activity, PFAS regulation, and waterbody‐specific parameters. These complex and context‐dependent dynamics may influence the long‐term environmental outcomes of the industrial regulation of PFAS.

Spatial patterns of the six focal PFAS were influenced by the waterbody type and history of industrialization, and showed strong evidence of a west‐to‐east gradient in biotic contaminant concentrations across a diverse assemblage of Great Lakes taxa. Variation in PFAS physicochemical properties (e.g., carbon chain length) likely influenced how these compounds move through the food web and accumulate in different tissues. We found consistent evidence in our model estimates that PFAS preferentially accumulated in protein and lipid‐rich tissues (e.g., eggs, blood, and liver). Mean PFOS concentrations showed a clear trend of biomagnification across a generalized Great Lakes food web (i.e., an estimated difference of two orders of magnitude between primary producers and apex predators), whereas concentration estimates of the long‐chain PFCAs reflected life history differences rather than dietary patterns. For these contaminants, our models tended to estimate elevated concentrations in biota closely associated with the sediment.

Our findings also highlight the importance of comprehensive regulation and monitoring, not only of specific terminal PFAS analytes, but also of the precursor compounds that can transform into these analytes. Replacement of known bioaccumulative and toxic compounds by these precursors may dilute or nullify the environmental effects of regulations on the industrial production of PFAS. Finally, we identified sampling gaps for the Great Lakes (and likely other systems) that represent key priorities for future research in the effort to provide more comprehensive, accessible, and reusable datasets. Such datasets will facilitate further synthetic work on the drivers and spatiotemporal patterns of contaminant concentrations in vital ecosystems like the Laurentian Great Lakes.

AUTHOR CONTRIBUTIONS

Peter O. L. Martin: Conceptualization; data curation; formal analysis; investigation; methodology; validation; visualization; writing—original draft; writing—review and editing. Alison M. Zachritz: Conceptualization; methodology; supervision; writing—review and editing. Amaryllis K. Adey: Methodology; validation; writing—review and editing. Sarah M. Klepinger: Data curation; methodology; writing—review and editing. Whitney M. Conard: Conceptualization; writing—review and editing. Gary A. Lamberti: Conceptualization; supervision; writing—review and editing. Daniele A. Miranda: Conceptualization; supervision; writing—review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. This work is not a product of the United States Government or the United States Environmental Protection Agency. The author affiliated with the US EPA did not contribute work in any governmental capacity. The views expressed are their own and do not necessarily represent those of the United States or the US EPA.

Supporting information

Supporting Information

Martin, P. O. L. , Zachritz, A. M. , Adey, A. K. , Klepinger, S. M. , Conard, W. M. , Lamberti, G. A. , & Miranda, D. A. (2026). Spatial‐temporal patterns of perfluoroalkyl substances in the biota of the Laurentian Great Lakes: A meta‐analysis. Journal of Environmental Quality, 55, e70183. 10.1002/jeq2.70183

DATA AVAILABILITY STATEMENT

All R code for this project has been made available on GitHub and Zenodo (P. O. L. Martin et al., ²⁰²⁶; https://doi.org/10.5281/zenodo.18683257) with the publication of this work, as well as the finalized dataset used in modeling, which was archived at the time of submission on Zenodo (P. O. L. Martin et al., ²⁰²⁵; https://doi.org/10.5281/zenodo.16173643). For a list of studies (with full citations), the data from which are included in this release, please reference Supporting Information Text S1 and Table S1. The release contains data that have been heavily formatted and manipulated and are the end product of the workflow for this meta‐analysis. The data frame has not been reviewed or approved by agencies or entities that may have been involved in the individual studies included in the meta‐analysis. For copies of the original raw data files, we recommend contacting the corresponding authors of the appropriate papers and/or accessing the data portals where those files are available. The data sources used for this project are provided in the Supplementary Information (Table S1).