Use of carbon isotopic ratios in nontargeted analysis to screen for anthropogenic compounds in complex environmental matrices

Abstract

Analytical data for ultra-high-performance liquid chromatography (UHPLC), nontargeted, high-resolution, mass-spectrometry (HR/MS) molecular features from a wide array of samples are used to calculate ¹³C₁¹²C_(n-1)/¹²C_n isotopologue ratios. These ratios increase with molecular carbon number roughly following a trend defined by atmospheric carbon. When the effective source reservoir ¹³C/¹²C ratio is calculated from the isotopologue ratio (assuming a fractionation factor of unity), features in biotic samples uniformly are tightly grouped, proximate to atmospheric ¹³C/¹²C ratio. In contrast, features in soil natural organic matter (NOM), dust NOM and anthropogenic compounds range from proximate to relatively divergent from atmospheric ¹³C/¹²C. For the NOM, ¹³C/¹²C ratios are consistent with an expected preferential volatilization of ¹²C, rendering features in soil NOM ¹³C-enriched and some features in dust NOM ¹³C-depleted. Anthropogenic compounds tend to diverge most dramatically from atmospheric ¹³C/¹²C, generally toward ¹³C-depletion, but pesticides we tested tended toward ¹³C-enriched. This pattern is robust and evident in: i) anthropogenic vs natural features in dust; ii) perfluorinated compounds in standards and as soil contaminants; and iii) sunscreen compounds in commercial products and wastewater. Considering the observed wide ¹³C/¹²C range for anthropogenic compounds, we suggest Rayleigh distillation during synthetic processes commonly favors one isotope over the other, rendering a source reservoir that is progressively depleted as synthesis proceeds and, consequently, generates a wide variation in ¹³C/¹²C for man-made products. However, kinetic-isotopic effects and/or synthesis from petroleum/natural gas might contribute to the anthropogenic isotopic signature as well. Regardless of cause, ¹³C/¹²C can be used to cull HR/MS molecular features that are more likely to be of anthropogenic or non-biotic origin.

1. Introduction

The last decade has seen rapid advances in nontargeted mass-spectral analyses of complex samples. For example, using an ion-trap mass spectrometer, in 2009 Evans et al. [1] reported on the nontargeted analysis of biological samples, followed by searches against an in-house library of 339 small biological molecules containing data for retention time, m/z, fragmentation patterns and other conventional mass-spectral data. In 2010, Meng et al. [2] reported using gas chromatograph (GC) mass-spectral deconvolution software developed for the US Department of Defense as well as an early-generation “molecular-feature extractor” that used liquid chromatograph, time-of-flight (LC/TOF) mass-spectral (MS) data including isotopic ratios, adducts and fragments to produce molecular features composed of retention time, accurate mass and abundance, which were searched against an accurate-mass database to identify several known compounds. In 2011, Hayward et al. [3] compared several chromatographic-mass-spectral systems for identifying pesticides spiked into spinach and ginseng extracts using database searches. They reported high identification rates for TOF/MS and Orbitrap/MS instruments, demonstrating the utility of nontargeted analysis with these high-resolution, fast-acquisition instruments. With these and similar successes, nontargeted analysis has found widespread application in metabolomics, proteomics and similar studies of endogenous compounds.

Typically generating ±10,000 nontargeted molecular features per sample, identification of unknowns has remained a prevalent challenge. Probabilistic elucidation of likely formulae for unknown molecular features (accurate m/z, isotopes, fragments, elution time) are based on heuristic filtering according to rules addressing mass, completion of valence shells, isotopic patterns, H/C ratios, heteroatom ratios and similar [4]. Some combination of these rules commonly are incorporated into commercial software to generate possible formulae automatically for detected molecular features, along with a metric of confidence in the formula assignments [5]. Typically, formula candidates are subjected to efforts at confirmation by i) matching spectra and elution times against standards, ii) searching against libraries or within literature, iii) spectrum matching with congeners, or iv) manual interpretation of precursor and fragment spectra [6,7].

Among tools for manual interpretation, mass defect has been particularly useful. Mass defect is the difference between an exact mass and the nearest nominal integer mass [8]. Because halogens have unusually negative mass defects among atoms commonly in organic molecules, mass-defect filtering has been particularly successful for detecting unknown halogenated compounds in complex matrices including natural and anthropogenic Cl- and Br-compounds in dolphins [9,10], anthropogenic Cl- and Br-compounds in freshwater animals [11], anthropogenic F-compounds in water [12] and metabolites of F-compounds spiked into rats [13]. Mass defect has been used with success in nontargeted studies with non-halogenated compounds as well, e.g., Grabenauer et al. [14] identified synthetic cannabinoids in “herbal incenses,” however, non-halogenated nontargeted studies invoking mass-defect filtering appear to be relatively fewer in number.

Given the wide array of marketplace pesticides, herbicides, drugs including antibiotics and, other synthetic products in use, screening for unknown degradation products of anthropogenic compounds in environmental chemistry or unknown metabolites in toxicological studies is a central objective in many ongoing research efforts. Large numbers of anthropogenic compounds intended to be bioactive, and their degradation products, bear no halogens, e.g., many organophosphates, carbamates, pyrethroids, sulfonylureas, among other families of compounds. Consequently, there is a need in nontargeted mass spectrometry for a filter that can help cull, from the thousands of molecular features for a typical sample, those unknown features that likely are of anthropogenic provenance.

Among the components of molecular features in nontargeted analyses is information on carbon isotopes, usually including the “M + 1 peak,” which is a secondary peak having a m/z = 1.003355 Da greater than the primary peak when molecular charge (z) is 1, equating to the difference in mass of ¹³C-¹²C. With respect to carbon, the primary peak generally is composed entirely of ¹²C, whilst the M + 1 peak represents the abundance of molecules containing a single ¹³C with the remaining carbons all being ¹²C. The M + 1 peak is smaller than the primary peak because most carbon is ¹²C; for example, ¹²C represents about 0.98898 of recent stable atmospheric carbon and ¹³C represents the balance, 0.01102 [15]. Most chemical reactions tend to fractionate isotopes by any of a number of elementalmechanisms [16] so that products often have a different isotopic signature than reactants and consequently different global reservoirs commonly have distinct isotopic signatures [17]. Considering differences that commonly are present between biotic and anthropogenic compounds for i) elemental mechanisms of carbon assimilation, ii) isotopic signature of carbon source reservoirs, and iii) the infinite vs finite source reservoir scale, here we investigate whether the carbon isotopic signatures, determined with nontargeted mass spectral analyses, might be used to screen for unknown anthropogenic molecular features in complex environmental matrices.

2. Background

2.1. Carbon isotopes

Carbon isotopic content conventionally is described by delta C 13 (δ¹³C), given by[17]:

$${\delta }^{13}C(\text{‰}))=\frac{\left[{\left(\frac{{13}_C}{{12}_C}\right)}_{spl}-{\left(\frac{{13}_C}{{12}_C}\right)}_{std}\right]}{{\left(\frac{{13}_C}{{12}_C}\right)}_{std}}\times 1000$$ (1)

where units are parts-per-thousand (i.e., per-mil), spl designates a measured sample and std designates a standard. In modern isotopic chemistry the standard often is the Vienna Peedee belemnite which is taken as ¹³C/¹²C = 0.0112372[18].

For carbon and most lighter multi-isotopic elements, in many chemical reactions and physical processes (e.g., diffusion), isotopes partition uniquely between reactant and product giving rise to isotopic fractionation between phases. This effect often is expressed as a fractionation factor (α), which is defined for carbon as[17]:

$$\alpha =\frac{{\left(\frac{{13}_C}{{12}_C}\right)}_R}{{\left(\frac{{13}_C}{{12}_C}\right)}_P}=\frac{{\delta }^{13}{C}_R+1000}{{\delta }^{13}{C}_P+1000}$$ (2)

where R signifies reactants and P is products. Fractionation factors arise from both equilibrium and kinetic effects and in complex multi-step fractionation, sequential elemental steps can enhance or suppress the effect of previous steps [16]. As a consequence of these fractionation effects, isotopic signatures have been shown to vary among plants having differing photosynthetic cycles (i.e., C3, C4, CAM), between specific tissues within plants [16], bacteria and hydrocarbon sources [19,20] (Supporting Information (SI), Table S1). The isotopic signature of plants commonly is a composite result of numerous elemental equilibrium and kinetic factors combined [16].

When a carbon source is very large relative to what is consumed in a reaction (e.g., the atmosphere), the source δ¹³C remains effectively constant, so products of fractionating processes also produce relatively constant δ¹³C values, simply given by Eq. (2). However, when a source reservoir is finite relative to the reactant flux for a reaction, the source reservoir is progressively enriched in the more discriminated isotope. So when various fractions (f) of the reservoir remain, they can vary widely in isotopic composition, becoming progressively enriched in the discriminated isotope relative to initial conditions, a process described by Rayleigh distillation [17]:

$$\frac{{\left(\frac{{13}_C}{{12}_C}\right)}_f}{{\left(\frac{{13}_C}{{12}_C}\right)}_0}=f(\alpha -1)$$ (3)

Where 0 designates initial state of the source reservoir, and f the reservoir state when fraction f remains.

Within context of this background, possible sources of variation in isotopic signature for ‘biotic vs anthropogenic’ compounds include: i) source isotopic composition, ‘atmosphere vs often synthesized from hydrocarbons’ (Table S1); ii) synthesis rate, ‘equilibrium/kinetic composite vs dominantly kinetic?’ (Eq. (2)); and iii) reservoir scale, ‘∼constant atmospheric composition vs variable finite’ (Eq. (3)).

2.2. Carbon isotopes in molecules (isotopologues)

Given an atmosphere in which stable carbon is 0.98898 ¹²C [15], for a fractionation factor of α=1, the probability of a biomolecule, comprised of n atmospheric-source carbons, being comprised entirely of ¹²C (P(¹²C_n)) is given by:

$$P\left({}^{12}{C}_n\right)={\left[\frac{{12}_C}{\left({}^{12}C{+}^{13}C\right)}\right]}_{atm}^n={0.98898}^n$$ (4)

The probability of the isotopologue having one or more ¹³C (P(m≥1 ¹³C_m)) is given by:

$$P\left(m\ge 1^{13}{C}_m\right)=1-P\left({}^{12}{C}_n\right)=1-{\left[\frac{{12}_C}{\left({}^{12}C{+}^{13}C\right)}\right]}_{atm}^n$$ (5)

And the probability of the isotopologue having only one ¹³C (P(¹³C₁)) is given by:

$$P\left({}^{13}{C}_1\right)=n\left\{{\left[\frac{{12}_C}{\left({}^{12}C{+}^{13}C\right)}\right]}_{atm}^{n-1}-{\left[\frac{{12}_C}{\left({}^{12}C{+}^{13}C\right)}\right]}_{atm}^n\right\}$$ (6)

Then the mean ratio [(¹³C₁¹²C_(n-1))/¹²C_n] is given by:

$$\frac{{}^{13}{C}_1{}^{12}{C}_{(n-1)}}{{}^{12}{C}_n}=\frac{P\left({}^{13}{C}_1\right)}{P\left({}^{12}{C}_n\right)}=n\left\{{\left[\frac{{}^{12}C}{({}^{12}C+{}^{13}C)}\right]}_{atm}^{-1}-1\right\}$$ (7)

For MS data, when z=±1, Eq. (7) describes the (M + 1)/M peak-area ratio for a molecular species in equilibrium with the atmosphere and having α = 1. Eqs. (4), (5), (6), (7) are plotted as a function of molecular carbon number in Fig. 1. Fig. 1illustrates that the probabilities of ¹²C_n (Eq. (4)) and ¹³C_m (m ≥ 1; Eq. (5)) reflect the atmospheric ¹²C fraction at n = 1 and curve asymptotically toward the opposing limiting fraction with increasing carbon number, always summing to unity probability. The probability of ¹³C₁ (Eq. (6)) rises to a maximum at n = 90, coincident to where it exceeds P(¹²C_n), then decreases because of the increasing probability of isotopologues having ¹³C_>1. The ratio ¹³C_n¹²C_(n-1)/¹²C_n (Eq. (7)) increases linearly with carbon number, exceeding unity value at n = 90 as well. Rearranging Eq. (7) to solve for the effective source reservoir (src; e.g., atmosphere for plants) ¹³C/¹²C with α = 1:

$${\left(\frac{{13}_C}{{12}_C}\right)}_{src}=\frac{\left(\frac{{13}_{C_1}{}^{12}{C}_{(n-1)}}{{12}_n}\right)}{n}=\frac{\left(\frac{M+1}{M}\right)}{n}$$ (8)

Figure 1:

Carbon isotopic composition for molecules comprised of atmospheric carbon assuming α = 1. Probabilities (P) are plotted on left axis, ¹³C₁¹²C_(n-1)/¹²C_n is plotted on the right.

The form of Equation 8 is particularly useful in that it describes a linear relationship between source reservoir and isotopologue compositions (Figure 1), making calculation of one from the other convenient.

2.3. Mass defects

Mass defects arise from systematic variations among the elements in nuclear binding energy according to Einstein’s mass-energy equivalence, E = mc². Assigning ¹²C a mass of 12.0000 Da, elements lighter than ¹²C have positive mass defects and elements heavier than ¹²C have negative mass defects [21]. Consequently, homologous series or compounds having common moieties, tend to fall in tight ranges of mass defect so that plots of these compounds in mass-defect vs nominal-mass space often form bands. Working with a high-resolution (but not fast acquisition) mass spectrometer, Kendrick [22] recognized that the slopes of these discrete bands can be rotated by choice of mass scale. Kendrick found that, when masses were scaled by setting the methyl unit mass to CH₂ = 14.0000, bands of homologous series of aliphatic compounds plotted horizontally. Similarly, judicious selection of other mass scales can rotate any band to horizontal orientation [23]. Horizontal band orientation offers the advantage of allowing user definition of filters, centered on a chosen mass defect +/− a tolerance, thereby diminishing noise [21].

3. Material and Methods

3.1. Samples

Given the documented ¹³C/¹²C variation among plants of different photosynthetic cycles (Table S1), we selected vegetative tissue from four C3-cycle plants (spinach, potato, cotton, Loblolly pine), two C4-cycle plants (corn, Bermuda grass) and one CAM-cycle plant (pineapple). Specifics on tissues and extraction methodology for these and the following samples are described in the SI.

Representing prokaryotic life, we selected known cyanotoxins from cultured cyanobacteria (SI). We selected liver tissue from laboratory mice to represent animal tissue (SI). We also analyzed natural organic matter (NOM) for an uncontaminated soil from wooded land in Conyers, GA (SI), collected as part of a previous study [24].

To represent anthropogenic compounds in clean matrices, we selected eight perfluorinated standards, 17 pesticide products, and 21 compounds in commercial sunscreen products (SI). Representing anthropogenic compounds in environmental matrices, we analyzed 17 perfluorinated compounds in soil, 16 sunscreen and antimicrobial compounds identified in wastewater. House dust samples were analyzed, with subsequent identification of both natural (e.g., fragrances and nicotine) and anthropogenic features (e.g., sunscreen, antiseptics, insect repellent).

3.2. Analytical

All analyses were performed using Waters Xevo G2-XS QToF MS (Waters, Milford, MA) with Acquity UPLC Class I using electrospray ionization (ESI) in positive and negative sensitivity modes. MS^e function was used in all acquisitions to obtain low- and high-energy spectra of m/z’s simultaneously. A high-energy ramp from 10 to 45 eV was chosen to fragment molecules. To characterize (M+1)/M ratios, we selected the M+H fragments in positive ESI and the M-H fragments in negative ESI (SI). Details regarding the liquid chromatography, fragments and data workflow are presented in the SI.

3.3. Assessment of precision

We assessed analytical uncertainty of our (M+1)/M data by injecting a single spinach extract ten times, integrating primary and (M+1) peaks for five molecular features having iFIT scores ≥90%, calculating effective source ¹³C/¹²C from (M+1)/M using Equation 8, and reporting summary statistics for these data. To assess extraction plus analytical uncertainty by this type of procedure, we injected ten replicate natural organic cotton extracts, examining eight molecular features having iFIT scores ≥93%.

4. Results

Taken altogether, we measured and calculated the (M+1)/M ratios of 168 molecular features including:

1. 67 biotic (Table S2) consisting of 26 features of C3 photosynthetic plants (spinach, potato, cotton, pine), 20 features in C4 photosynthetic plants (corn beard, kernel and inner husk, Bermuda grass), 7 features of CAM photosynthetic plants (pineapple leaves, flesh), 7 known natural toxins in cyanobacteria, and 7 features of mouse liver;

2. 17 NOM features (Table S2), 7 from soil and 10 from dust;

3. 46 anthropogenic compounds prepared from standards or commercial products (Table S3) including 8 perfluorocarboxylates, 21 sunscreen components, and 17 pesticides; and

4. 51 anthropogenic compounds in environmental media (Table S4) including 18 in household dust (e.g., sunscreens, DEET; Table S5), 7 perfluorocarboxylates at background levels in soil[24], 10 perfluorocarboxylates in soil amended with biosolids[25], and 16 antimicrobial and sunscreen compounds in wastewater influent (Table S5).

These data are plotted as a function of molecular carbon number in Figure 2. Collectively, these data clearly follow the trend line calculated for recent atmospheric carbon composition. In general, features in biotic samples fall closely proximate to the atmospheric line, presumably deviating by amounts related to the samples’ fractionation factors. NOM features also follow a roughly atmospheric trend but have a slightly greater variance than do biotic features. And, whilst anthropogenic compounds also follow the atmospheric trend, they appear to span a wider variance than biotic or NOM features, dominantly skewing toward depletion in ¹³C. These observations suggest that there is sufficient resolution to discern differences in isotopic composition among at least some source materials.

Figure 2:

Isotopologue ratio (≡(M+1)/M when z=1) for all data vs molecular carbon number follows the general trend defined for equilibrium with the atmosphere and α=1. Biotic features generally are closely proximate to the atmospheric trend line. NOM features from soil and dust deviate from roughly atmospheric composition over a slightly wider range. Anthropogenic compounds coarsely follow the atmospheric trend, but appear to have larger variance about atmospheric composition than biotic or NOM, especially toward depleted-¹³C composition.

4.1. Precision

Analytical uncertainty was assessed with ten repeated measures of a single spinach extract, for five molecular features (Table S6). Considering that ¹³C/¹²C can vary among tissues within a plant [16], we characterize uncertainty for these data as the mean of the standard deviations of ten repeated measures for each molecular feature, so mean standard deviation is ¹³C/¹²C=1.53×10^-4.

Using a similar approach to characterize extraction plus analytical uncertainty by injecting ten separate extracts of natural cotton (Table S7), for eight molecular features the mean standard deviation is ¹³C/¹²C=1.87×10^-4.

Comparing these uncertainties to typical uncertainties using conventional isotope-ratio-monitoring (irm) GC/MS, Freeman et al.[26] reported ¹³C/¹²C standard deviations among repeated measures ranging from 1.1×10⁻⁶ to 5.6×10⁻⁶ for n-alkanes in oxic seawater. Our analytical uncertainties are roughly 30- to 140-times these typical irm-GC/MS results, and our extraction plus analytical uncertainties are roughly 30- to 170-times the irm-GC/MS data.

Considering typical literature values, the natural reservoirs summarized in Table S1 vary from atmospheric composition by ¹³C/¹²C=7×10⁻⁵ for C4 plants to 3.58×10⁻⁴ for natural gas. The standard deviations we report for our data (i.e., for repeated measures and between replicates) fall toward the high end of this range. Based on these data then, nontargeted TOF/MS is unlikely to discern the small ¹³C/¹²C variations commonly observed among many natural carbon sources.

4.2. Patterns of deviations from atmospheric composition

Using Equation 8 to express ¹³C₁¹²C_(n-1)/¹²C_n data in terms of effective source ¹³C/¹²C accentuates scatter of the data about the atmospheric trend line (Fig. 3). Consistent with Fig. 2, the biotic source samples uniformly plot tightly on and near the atmospheric trend line, with every one of the 67 biotic data points falling in the field bound by the two green dashed lines. Considering that these data represent all three of the photosynthetic cycles of land plants (eukaryotes), as well as the primitive photosynthetic cycle of prokaryotic cyanobacteria and animal tissue, the tight spread of these data evidently reflects the atmospheric carbon source of all photosynthetic carbon-fixers and heterotrophs, along with modest fractionation effects [16].

Figure 3:

Effective source ¹³C/¹²C reservoir, calculated with Equation 8, vs molecular carbon number. All 67 biotic features plot closely proximate to the atmospheric trend line, with the outermost ¹³C/¹²C extremes defined by the green dashed lines. Unlike the biotic samples, NOM features commonly fall outside of the biotic limits, enriched in ¹³C for soil. Anthropogenic compounds also commonly plot outside of both biotic limits, but preferentially to the depleted ¹³C side of the biotic field.

Grouping data in Figure 3 by molecular provenance (Figure 4) reveals that NOM generally deviates more from atmospheric ¹³C/¹²C than do biota. In fact, four of the seven soil NOM features fall outside of the biotic ¹³C/¹²C range, uniformly falling in the “heavy” (¹³C-enriched) field (Figure 3). This ¹³C enrichment is consistent with expectations for soil NOM. Fractionation processes during decomposition of plant material in soil are known to generate CO₂ depleted in ¹³C [27] and, over long periods of decay, the resulting soil NOM can be enriched in ¹³C [28]. In contrast to soil NOM, dust NOM tends to deviate to the “light” (¹³C-depleted) field for both mean and range (Figure 4). These dust NOM features largely are fragrances and nicotine (Table S5). To the extent that the mode of occurrence of these compounds on dust is by volatilization from original substrates (e.g., skin or food), the lighter isotopologues can be expected to volatilize preferentially, finding their way to sorb on dust.

Figure 4:

Effective ¹³C/¹²C reservoir of molecular features (mean±1SD) grouped by provenance. The blue line delineates atmospheric ¹³C/¹²C for α=1. WW designates waste-water influent at a sewage treatment plant.

For the anthropogenic features, 39 out of 97 points plot anomalously, i.e., outside of the entire biotic range (¹³C/¹²C = 0.009876–0.01313), most of them, 33 compounds, to the light side of the biotic field (Figure 3). Comparing to biotic mean ±2 standard deviations, 47 anthropogenic points, 48%, fall outside of this interval.

Taken altogether, when molecular features are grouped by provenance (Fig. 4), overarching patterns emerge: i) all biotic samples (plants, bacteria, animal) fall close to atmospheric composition and have small variances; ii) NOM tends to deviate from atmospheric more so than biota and exhibit larger variances, possibly related to fractionation processes in their mode of occurrence, with soil NOM skewing to the heavy field and dust NOM to the light; iii) most anthropogenic compounds deviate to the light side of atmospheric and exhibit larger variances than biota or the light dust NOM; and iv) the pesticides we tested also deviate from atmospheric composition with wide variance albeit to the heavy field. Considering that variation in ¹³C/¹²C among source reservoirs tends to be modest (Table S1) relative to these anthropogenic variances (Fig. 4), this wide anthropogenic ¹³C/¹²C variability reported here might suggest Rayleigh distillation (Eq. (3)) as a cause for this variation.

Figure 4 also is instructive for confirming that a diagnostic ¹³C/¹²C signal of anthropogenic compounds, deviant from atmosphere and wide variance, is not obfuscated by complex environmental matrices. For example, the ¹³C/¹²C of PFAS compounds are similar for standards and in soil (Figure 4). Likewise, the ¹³C/¹²C of sunscreen compounds are similar for sunscreen products and in wastewater (Figure 4). Addressing a single sample matrix, dust, anthropogenic compounds range to distinctly lighter ¹²C/¹³C values than do the natural dust components (Figure 4). In addition, anthropogenic anti-microbials and preservatives in wastewater exhibit the ¹³C depletion and large variance characteristic of most anthropogenic compounds (Figure 4).

Although the anthropogenic groups of compounds uniformly possess a larger standard deviation than the biotic features, the extent of overlap varies among groups (Figure 4). This variation in overlap among groups suggests variation in probability of detecting anthropogenic compounds in a generally biotic matrix. For example, the cosmetic preservatives in wastewater and sunscreens in dust overlap little with biotic features; a large fraction of these compounds might be detected with ¹³C/¹²C. In contrast, the PFASs in soil we report fully envelope biotic features, consequently some of these compounds might not be discerned in a biotic matrix. Whereas most anthropogenic compounds are depleted in ¹³C, the pesticides we tested are enriched in ¹³C. For this pattern, pesticide-degradation products might be discerned in food matrices but are unlikely to be discerned in soil matrices with this isotopic approach (Figure 4).

Considering that anthropogenic and biotic features generally share some common range in ¹³C/¹²C, using ¹³C/¹²C in combination with mass defect might bolster efficacy for culling anthropogenic from biotic compounds. In Figure 5, all biotic features except for five (1 potato flesh, 4 cyanobacteria) fall in the mass-defect range = 0.0199–0.4929 Da; 33 anthropogenic compounds fall outside of this biotic range. Comparing the ¹³C/¹²C and mass-defect, 27 data points are exclusively anomalous for ¹³C/¹²C and 23 exclusively for mass defect. Taking ¹³C/¹²C and mass defect together, 60 of 97 anthropogenic compounds fall outside the biotic field (62%; Figure 5), an outcome that is more effective than using either filter alone.

Figure 5:

Mass defect vs Reservoir ¹³C/¹²C. The green dashed lines delineate all biotic samples for ¹³C/¹²C and most samples for mass defect, excepting 5 outliers at a defect of roughly −0.4 (4 prokaryotic cyanobacteria and 1 potato flesh).

To our knowledge, prior to our report here, mass defect has been the only variable that has been used widely to identify possible unknown anthropogenic compounds in nontargeted analyses of environmental samples. Largely relying on the characteristically negative mass defect of halogens, the efficacy of mass defect been greatest for identifying halogenated anthropogenic compounds [[9], [10], [11], [12], [13]], potentially leaving a wide array of non-halogenated anthropogenic compounds undetected. For the data we report here, >2/3 s of the anthropogenic features falling outside the ¹³C/¹²C biotic range fell within the mass-defect biotic range, rendering them unlikely to have been culled for further assessment except by this ¹³C/¹²C isotopic screening. As such, ¹³C/¹²C might prove to be an important new tool for culling unknown anthropogenic compounds from vast numbers of environmental features in nontargeted analysis.

4.3. Limitations to use of ¹³C/¹²C in nontargeted analyses

The mass-defect filter for identifying groups of compounds, especially halogens, requires only an accurate m/z and an inference as to whether z = 1 or a larger charge. In contrast, use of ¹³C/¹²C to identify anthropogenic compounds in nontargeted analyses requires an inferred carbon number for a molecular feature. When a molecular feature is unknown, carbon number can be inferred using commercial algorithms based on heuristic rules similar to those described in Kind and Fiehn [4], e.g., iFIT in Waters’ Unifi software, and this filter performs only as well as the estimate of the carbon number. Another fundamental limitation is that the primary (M) and M + 1 peaks must be integrated internally consistently. Some commercial nontargeted software does not allow manual chromatographic peak integration, so when primary and M + 1 peaks integrate differently, algorithms based on isotopic ratios can yield misleading formula estimates.

5. Conclusions

Here we used UPLC coupled to a high-resolution mass spectrometer to acquire ¹³C/¹²C ratio data for nontargeted molecular features in a wide variety of biotic and other environmental matrices as well as anthropogenic commercial products. These features included natural compounds that were components of the natural samples, anthropogenic compounds in natural matrices, and anthropogenic compounds in standards and commercial products.

Biotic molecular features from primitive algal cells, C3, C4 and CAM plants as well as animal tissue, tightly clustered about the line delineating an atmospheric ¹³C/¹²C reservoir, consistent with modest isotopic fractionation processes operating from an effectively infinite atmospheric carbon reservoir. Soil NOM ¹³C/¹²C ranged from the biotic field to ¹³C-enriched, suggesting a Rayleigh-type distillation of preferential ¹²C mineralization to ¹²CO₂ thereby enriching the remaining OM reservoir in ¹³C. The ¹³C/¹²C ratio of anthropogenic compounds ranged from nearly atmospheric composition to far outside the biotic field, whether the compounds be in environmental matrices, or neat or in commercial products. For most anthropogenic compounds, the ¹³C/¹²C ratio skews toward ¹³C-depleted. The 17 pesticides we tested exhibited a tendency opposite that of other anthropogenic compounds, skewing toward ¹³C-enriched; we remain uncertain whether this pattern extends to pesticides generally as well as what might cause this anomalous departure from other anthropogenic compounds.

Because of the varying degrees of overlap in ¹³C/¹²C ranges among environmental matrices and anthropogenic compounds, the fraction of all anthropogenic compounds that might be detected in a sample will vary with anthropogenic compound type and sample matrix. For example, a large fraction of sunscreencompounds, preservatives and PFASs might be detected using ¹³C/¹²C in biological media and soils (Fig. 4). Likewise, a large fraction of pesticides might be detected in biological media (Fig. 4), such as residues in food crops, or trace pesticides and their metabolites in toxicology studies. However, considering that the ¹³C/¹²C ratios of the pesticides we tested were completely enveloped by those of soil NOM (Fig. 4), the ¹³C/¹²C contrast of pesticide residues and pesticide degradation products with soil matrices is likely to be poor to non-existent, rendering this approach ineffective for this application. Despite limitations such as this, that vary among sample matrices and anthropogenic compound groups, few alternative routines exist to this ¹³C/¹²C screening method for identifying anthropogenic compounds, that do not bear halogens, in the large datasets typical of high-resolution MS nontargeted analyses of environmental samples.