Evaluating toxicity of heavy fuel oil fractions using complementary modeling and biomimetic extraction methods

Predicted dissolved exposures

The overall limits on dissolved exposures for the substances in the present study were reached at a relatively low oil–water loading (∼50 mg/L), consistent with the high-MW compounds typical of the HFO fractions. The magnitudes of the dissolved exposure were very different among the oil fractions (Figure 1). At very low loadings (∼0.02 mg/L), dissolution of the whole oil was approximately 0.003 mg/L and reached a maximum of 2.8 mg/L at very high loadings (>50 mg/L). The F2 fraction was enriched in more water-soluble constituents and reached a maximum dissolved concentration near 10.4 mg/L at high loadings. In contrast, the F3 and F4 fractions were relatively insoluble, with total dissolved concentrations of 0.77 mg/L and 0.21 mg/L, respectively. Although the whole oil was composed of only 7% of the F2, these constituents dissolved and contributed to the higher total dissolved concentrations predicted for HFO. Note that the difference between the predicted dissolved concentration and nominal loading (1:1 line) increased from F2 to F4, reflecting the increasing fraction of oil that was not dissolving and was present in droplet form (Figure 1).

To further illustrate how WAF composition changes depending on oil type and loading, the aqueous concentrations of 2-methylnaphthalene, 2-methylphenanthrene, and chrysene, as predicted by PETROTOX, were plotted in Figure 1. At low loadings of HFO (<1 mg/L), the concentration of each hydrocarbon contributed to the dissolved exposure in a linear manner proportional to both their single-chemical solubility and abundance within the oil (e.g., Raoult's Law). At higher loadings (1–50 mg/L), the predicted solubility limit of chrysene and 2-methylphenanthrene were reached, whereas the aqueous concentration of the more soluble 2-methylnaphthalene continued to increase. The magnitude of the predicted dissolved concentrations varied across the fractions. For example, despite 2-methylnaphthalene having the highest individual chemical solubility of the 3 constituents, at lower loadings of F3, C1-PHE and chrysene concentrations are predicted to be higher because of their greater abundance in this substance than 2-methylnaphthalene.

The range of loadings in Figure 1 spanned the range of oil-in-water loadings used in the toxicity tests performed by Adams et al. 17. In summary, the concentration and composition of dissolved constituents in WAFs were predicted to vary as a function of oil composition and loading.

Comparison with measured WAF chemistry

Measurements of the CEWAF chemistry at the 0.1% v/v loading prepared for the present study were compared with predicted dissolved concentrations of individual constituents (Figure 2). The measured concentrations of individual PAHs (and SHCs) were approximately 2 to 50 times greater than the predicted dissolved exposure, as indicated by data falling below the 1:1 line in the top row of Figure 2. Constituents in the HFO and F4 fractions showed less deviation than the F2 and F3 fractions, probably because of the viscous nature of the former, which were not readily dispersed into aqueous media.

The speciation caused by the partitioning of hydrocarbons between oil and water phases was evaluated using the individual constituent data to minimize potential errors from the experimental and modeling methods. Both CEWAFs and PDWAFs were enriched in 3- to 4-ring PAHs relative to the typically more soluble 2-ring PAHs, as indicated by the increasing ratio of measured aqueous to predicted dissolved concentrations with increasing log K_OW (Supplemental Data, Figure S2). The predicted total concentrations showed better agreement with measured data (lower row, Figure 2). The TPH-normalized concentration, or the effective oil composition (C_oil), were calculated by dividing the measured aqueous concentrations (C_aq) for each constituent, i, by the TPH—for example, C_oil,i (µg/mg) = C_aq,i (µg/L)/TPH (mg/L)—was compared with the neat oil substance compositions to confirm the presence of entrained oil in the test system. The normalized concentrations were consistent with the chemistry of the neat oil (Supplemental Data, Figure S3), confirming the presence of entrained oil in WAF prepared for all fractions.

Similar observations were noted for the PDWAF dosing systems at the 0.1% v/v loading (Figure 2). Although the concentrations of TPH for F2 and F3 fractions were approximately 4 times lower than in the CEWAF tests, measured aqueous concentrations of individual PAHs were still well above the predicted dissolved concentrations, and the CEWAF composition was indicative of entrained oil. The TPH concentrations for whole oil and F4 fractions in the PDWAF system were approximately 10 times lower than the CEWAF but still 2 to 5 times greater than the predicted dissolved concentrations. However, this trend was less consistent across the range of PAHs shown in Supplemental Data, Figure S2, although the presence of SHCs in the CEWAFs and PDWAFs supported the presence of oil, perhaps at trace concentrations. Similar to the CEWAF results, the PDWAFs were enriched in 3- and 4-ring PAHs, and the TPH-normalized composition of the PDWAFs for these 2 substances was similar to that of the whole oil.

The high-energy mixing involved in CEWAF and PDWAF preparation via vortex and sonication appeared sufficient to generate droplets that were relatively stable in solution. The droplets appeared stable because the chemical measurements that indicated droplet oil were performed over a period of hours to days after WAF preparation and sampling. Chemical dispersants are thought to stabilize droplets, preventing re-coalescence 31, but differences in the behavior of droplets between the 2 WAF systems were not measured. However, when comparing the results between CEWAF and PDWAF solutions, the dispersants had the greatest impact on total aqueous hydrocarbon concentrations of the relatively heavier whole oil and F4 fractions. The TPH-normalized oil concentrations for the dosing systems were similar to the measured oil concentrations and are indicative of oil droplets. The variability in the HFO and F4 tests were attributable in part to concentrations that were near the detection limit but also to the poor dispersion of these highly viscous materials, such that a high proportion of hydrocarbons in the test solution (though generally low) were in the dissolved phase rather than in droplets.

The oil-in-water speciation model 28 was used to estimate the concentration of droplets in these systems (Supplemental Data, Table S5; Figure 2), which provided an indication of the dispersion effectiveness of the different WAF preparation systems. This calculation was performed only where concentration data for individual PAHs and SHCs were available in the WAF, because this was primarily a method for estimating hydrocarbon speciation. Consistent with the TPH measurements and WAF composition discussed earlier in this section, the estimated droplet concentrations were highest for F2 and F3 in the CEWAF and PDWAF systems. Comparatively, the droplet concentrations were lower by approximately 10 to 100 times for whole oil and F4 substances in these 2 systems, likely because of their higher viscosity.

These estimates were based on chemical speciation measured for the neat oil test substances and corresponding CEWAFs or PDWAFs. At low oil loadings, model estimates became uncertain based on noise in the WAF analysis data. Correspondingly, the standard deviations were small for the estimates with obvious entrained oil (Supplemental Data, Table S5) and high for the droplet estimates at low concentrations. The results of the present analysis confirmed that droplets were present in the CEWAF and PDWAF systems for some substances (F3, F2) or at trace to negligible concentrations for others (whole oil, F4).

Predicted toxic units and chronic toxicity

Dissolved-phase hydrocarbons are assumed to be the most bioavailable fraction of oil. The composition of dissolved hydrocarbons in WAF is therefore expected to determine the toxicity toward aquatic species. To illustrate, concentration-loading relationships (Figure 1) were converted to TUs using the TLM. Although 2-methylnaphthalene was typically the more abundant hydrocarbon in WAFs of whole oil, F2, and F4 because of its relatively higher water solubility, it was not a major contributor to toxicity based on the TLM framework (Figure 3).

At lower loadings, the more hydrophobic 2-methylphenanthrene and chrysene often contributed more to the predicted toxicity than 2-methylnaphthalene for all oils tested. Only after the solubility limits were reached for 2-methylphenanthrene and chrysene for whole oil, F2, and F4 did 2-methylnaphthalene become a larger contributor. Furthermore, the composition of F3 included many 3- and 4-ring PAHs, which were more hydrophobic than the compounds in F2. Despite the lower predicted dissolved hydrocarbons of F3 WAFs (compared with F2), the maximum chronic TUs were 12 for rainbow trout, which were comparable to the maximum TUs for F2 (28 TUs) and the whole oil (9 TUs). The maximum TUs for the distillate fractions F2, F3, and F4 do not add up to the maximum TUs for the whole oil. After distillation, each of the fractions behaved as a separate substance whose dissolution and toxicity are controlled by the respective loading and composition.

A key goal of the present study was to describe the observed toxicity to trout embryos in terms of toxic units and passively sampled WAFs. Toxicity calculations with PETROTOX were based on measured neat oil chemistry and loadings estimated using measured aqueous TPAH concentrations 17. The underlying principle for toxicity calculations with CEWAF and PDWAF systems is that both dissolved material and oil droplets are transferred from the stock solution to the more dilute loadings. All material (dissolved and droplet) from the initial WAF preparation (e.g., 10% v/v stock) were diluted in the lower loadings and the transferred droplets then re-equilibrated in the new exposure water.

The TPAH concentrations from Adams et al. 17 were used to estimate the loadings from transferred droplets for the TU calculations, because these concentrations were representative of the exposures experienced by trout embryos. The TPAH concentrations were not measured for each of the loadings in the present analysis, because solutions that were toxic were often too dilute for accurate analyses. Hence, nominal loadings were used in TU calculations as scaled by the calculated transfer efficiency where measured TPAHs were estimated from dilutions of known TPAH concentrations in stock WAF solutions.

Differences were found in the TPAH concentrations obtained in CEWAFs prepared in the present study and results reported by Adams et al. 17 (Supplemental Data, Table S4 and Figure S6), likely because of the inter-laboratory variability of the WAF preparation protocol and differences in analytical methods used for TPAH characterization. For example, samples analyzed in the present study were taken directly from WAF preparation volatile organic analysis (VOA) vials, whereas Adams et al. 17 sampled exposure solutions during chronic toxicity tests. Despite the difference in absolute TPAH concentrations, the composition of the TPAH based on the individual PAH analysis (Supplemental Data, Figure S7) remained similar.

Comparison with measured chronic toxicity

The observed mortality of trout exposed to CEWAF spanned approximately 3 orders of magnitude across the different fractions of HFO (Figure 4A) because of the very different nature (e.g., hydrophobicity) of the constituents present. Predicted aqueous concentrations generated using PETROTOX were used to predict toxicity 19. Dissolved-phase TUs were compared with the observed effects data (Figure 4B).

The chronic TUs normalized the different effects from the 4 test oils to within a factor of 3. The modeling approach predicted relatively higher TUs for the lighter F2 and F3 fractions compared with the F4 and whole oil, which is composed of approximately 60% F4 constituents. The differences between these 2 groups of data were within a factor of 2, which is consistent with typical variability in the model framework and experimental toxicity work (factor of 2–5) 20 with petroleum substances. It was, therefore, difficult to identify trends in the model–data comparisons among fractions, given the variability in the comparisons and the small number of fractions analyzed. The exposures of trout embryos to whole oil, F2, and F3 fractions caused mortality that spanned from control levels (<10%) to 100%. The TUs across the exposures ranged from less than 0.001 to 18, with observed dose responses between mortality and TUs that were characteristically steep, typical of nonpolar organics.

Exposures to F4 resulted in 25% mortality at the highest loading, which was associated with only 0.24 TUs. Whether this toxicity was a result of biological variability or exposure to hydrocarbons is not clear. The observation of sublethal endpoints (deformity, blue sac disease) 17 were consistent with exposure to hydrocarbons, but the TU analysis and SPME results (see later discussion) suggested that dissolved-phase hydrocarbons may not have been a large contributor to F4 toxicity or are at least at the lower end of the uncertainties associated in the models and exposure assumptions.

Uncertainty around the TU–mortality comparison was characterized by evaluating the standard errors of the logistic model fit to the data sets for each test oil. This provided a means for establishing confidence intervals around the toxicity predictions and for comparing 1 dataset with another. For example, the chronic TUs associated with 10% mortality (95% confidence intervals in parentheses) were 0.92 (0.32) for whole oil, 1.8 (0.15) for F2, 2.8 (0.97) for F3, and 0.2 (0.25) for F4. The TUs at the observed EC10s were similar, within a factor of 2, with confidence intervals that overlapped or were very similar, except for F4. The similarity in the confidence intervals indicated very subtle differences in the observed exposure–response relationships. Furthermore, the uncertainty associated with TLM predictions (e.g., factor of approximately 3) 20 suggested very little difference among the observed dose responses from the different HFO fractions. Thus, the observed effects were consistently described by this TU approach, which provided a mechanistic means for relating observed toxicity to the chemical composition of the different oil fractions in these complex exposure systems (CEWAFs).

Ideally, the EC10s would be associated with a TU of 1, but there is some uncertainty attributable to assumptions in the modeling or the underlying biological variability in toxicity tests with complex hydrocarbons. In addition, this analysis relied on some estimated exposure concentrations, and losses of material through handling or biodegradation were possible and a likely source of variability. Variability in the oil dissolution model, including the assumption that droplets were in equilibrium with the exposure water, is another source of potential error. However, mortality increased with increasing TUs, and the clear dose responses were generally consistent with theory. Furthermore, the assumption of constant exposures is supported by the reasonably close agreement between the model and observations. However, developing methods that deliver constant exposures of complex substance is needed to minimize variability and to support valid interpretation of dose response data. Hence, variability in toxicity attributable to the estimated exposure concentrations of oil fractions was considered acceptable relative to the other uncertainties discussed previously. The application of PETROTOX to these toxicity data reduced the differences in toxicity (e.g., 1000× on loading basis) among HFO fractions to within a factor of 3.

To further illustrate how the oil dissolution model impacts the interpretation of the toxicity in these exposure systems, the predicted TUs for each loading were plotted at their target dilution from the stock solution (Figure 4C). The curves (as symbols) were shallow, following the general pattern of the dissolved concentration and TU curves in Figures 1 and 3. The calculated TUs based on absolute dilution of the stock solution (e.g., no dissolution) are shown as lines decreasing linearly with increasing dilution of the stock solution. The 2 approaches gave very different profiles and differed by nearly 1000-fold at very dilute loadings. This provides a mechanistic insight for heavy substances that substantial dilutions from a concentration stock may not appreciably alter the dissolved profile because of the solubility characteristics of these substances. This observation is dependent on the composition of the entrained oil as well as the amount of entrained oil, and at present it may be difficult to generalize beyond the results of the present study.

Correlation between toxic units and biomimetic extraction measurements

The use of passive samplers is increasing as a way to provide a more relevant exposure measure for effects assessment and as a potential alternative method to predict toxicity 26. The biomimetic method used in the present study provides an integrated measurement of dissolved hydrocarbons in a complex WAF, which is similar, in principle, to the use of additive TUs. In the present study, the application of this method to the observed effects data was considered qualitative because direct biomimetic extraction measurements were not taken during the toxicity exposures of Adams et al. 17. Instead, the comparisons were made on the basis of predicted TUs calculated from the PETROTOX model.

The biomimetic extraction results for both CEWAFS and PDWAFs at selected loadings for the different oils are provided in Supplemental Data, Table S1. The SPME fiber concentrations of total dissolved, bioavailable hydrocarbons for F2, F3, and whole oil ranged from 0.5 mM to 290 mM PDMS fiber concentrations for F4 CEWAFs, and PDWAFs were low to nondetectable (∼0.5 mmol/L PDMS). Fiber concentrations exhibited a linear trend with predicted TUs over approximately 2 orders of magnitude (Figure 4D). Consistent linear patterns were observed within each dosing system. This suggests that these results were not confounded by fouling of fibers by oil droplets. If fouling were to occur, significant variability in analytical results would be expected. However, replicate measurements of SPME concentrations in CEWAF and PDWAF samples indicated low coefficient of variations of typically less than 20% (Supplemental Data, Table S4).

Furthermore, the SPME chromatograms of CEWAF and PDWAF samples of HFO and the distillate fractions did not resemble the chromatogram for the neat oils but was rather enriched in the more water-soluble components as expected by Raoult's Law. In addition, the pattern of SPME chromatograms changed as a function of loading, with the more water-soluble components enriched at higher loading (Supplemental Data, Figure S8). This is consistent with the expected increased concentration of dissolved-phase components at higher loadings (Figure 1). Based on these observations, the passive sampling method used in the present study appears to provide a viable surrogate for measuring dissolved hydrocarbons in different WAF systems.

Recent guidance suggests that equilibrium sampling with passive samplers should meet both negligible depletion and equilibrium requirements 32. The extent to which a passive sampler depletes the dissolved phase chemical in which it is equilibrated is given by

(1)

where K_PDMS–Water is the PDMS–water partition coefficient, and the second term is the volume ratio of PDMS to water, which in the present study was 0.000132 mL/22 mL = 6 × 10⁻⁶. Smedes et al. 33 provide the following correlation between log K_PDMS–Water and log K_OW:

(2)

Substituting this relationship into the previous equation indicates that a hydrocarbon with log K_OW of 4, 5, and 6 would be depleted by 4%, 28%, and 77%, respectively. However, the actual amount of depletion may be much less than calculated because of replenishment of the dissolved phase by droplet dissolution. This also may explain the presence of the material at high retention time (>7 min) in the fibers, which corresponds to high boiling point chemicals with typically low solubility.

A second concern is if the 100-min extraction time is insufficient for the fiber to reach equilibrium with the dissolved phase hydrocarbon components. Previous work in our laboratory showed that fiber concentrations determined in heavy fuel oil WAFs and determined using the method described in the present study were approximately 75% of the value obtained at equilibrium, which was reached in approximately 400 min (data not shown). Thus, the potential depletion of higher log K_OW hydrocarbon constituents and nonequilibrium sampling conditions may have underestimated the concentrations in target lipid. This is consistent with the fact that the PDMS concentration corresponding to 1 chronic TU was approximately 1 µmol/mL PDMS, which was less than expected based on partitioning theory (Figure 4D). Because the partition coefficient between target lipid and water (K_{Target Lipid–Water}) was estimated using the bioconcentration model from the TLM [20]

(3)

Target lipid is estimated to have a 1.4-fold higher (10^0.17) affinity than PDMS (e.g., the intercept of the K_PDMS–Water regression). The chronic CTLBB for trout derived from the TLM and used in the present study was 7.2 µmol/mL octanol (octanol is assumed to approximate target lipid). Translating this value into PDMS concentrations based on these partition relationships gave a predicted value of 7.2/1.4 = 5.1 µmol/mL PDMS, which was within the range but slightly lower than the empirical value deduced from Figure 4D. Future work should consider using larger water samples and thinner fiber coatings that have faster sampling kinetics for approaching equilibrium so that equilibrium concentrations of dissolved hydrocarbons in WAFs are not underestimated.

One goal of the present study was to support the EDCF efforts by identifying important classes of toxic constituents in the HFO and distillate fractions. This was done by summing the contributions of predicted TUs within major chemical classes for each of the petroleum substances evaluated. The calculated TUs for each substance were calculated at the predicted EC10, which corresponds to a TU of 1 and ideally correlates to the onset of chronic toxicity. In the case of F4, in which a TU of 1 was not reached, the TU distribution at the highest loading tested was used (Figure 5).

The modeling suggested that for whole oil, most (60%) of the dissolved TUs were derived from 3 + ring aromatics between log K_OW 5 and 7, approximately 25% were from 2 + ring aromatics, and the remaining contributions were from branched monoaromatic and aliphatic constituents in the same range of log K_OW. A similar pattern was observed for the F3 fraction, though the distribution reflects a narrow range of log K_OW. The TU distribution for F2 is more evenly divided between the 2 ring and 3 + ring aromatic fractions. Approximately half of the TUs in the F4 fraction come from 3 + ring PAHs with log K_OW near 6.

To summarize, the TU distributions vary with composition of the neat oils. The patterns observed are specific to these substances and would vary for different petroleum substances categories (e.g., gas oil, kerosene, crude oil). The modeling framework used in the present study allowed the systematic identification of important subsets of hydrocarbon classes for additional research to refine our understanding of the process and constituents that contribute most to the toxicity of these petroleum substances.

Putting the results of laboratory-based toxicity testing in the context of real world exposures is important. In open-ocean spills, the concentrations of dispersed oil are expected to drop below 1 ppm in hours to less than a day 34, thereby limiting the exposure to levels of dissolved hydrocarbons that cause adverse effects. Dissolved hydrocarbons in dilute oil concentrations (<1 ppm) are not likely to be influenced substantially from replenishment by entrained oil droplets (Figure 1). However, that ecological and biological and factors can also modify the risks of oil toxicity must be recognized. Freshwater ecosystems can have more limited volume and relatively more shoreline that can lead to stranding of oil in shoreline habitat and shallow sediments, creating more prolonged exposures.