Nanoplastic concentrations across the North Atlantic

Sampling

The samples were collected aboard RV Pelagia during cruise 64PE480 in November 2020. Samples were taken at nine stations along a transect through the temperate to subtropical North Atlantic and at three stations positioned on the European continental shelf (Fig. 1). To enable cross-comparison between different stations, three depths (10 m and 1,000 m water depths and 30 m above the seafloor) were sampled at every deep-ocean station (stations 1–9). Consequently, the actual depth below the sea surface of the deepest sampling point varied as a function of the local water depth. A conductivity, temperature and depth (CTD) sensor phalanx with a rosette sampler comprising an array of 24 polytetrafluoroethylene (PTFE)-lined, PVC Niskin bottles with a volume of 12 l was used for profiling water properties and recovering discrete water samples. During the hydrocast, the Niskin bottles were kept open so that they were flushed with local water during descent and ascent until closure at the desired water depth. Once the CTD sensor was placed on deck, the bottle faucet and tubing used for tapping seawater were thoroughly flushed with sample water before sampling. Then, 2-l gl*** bottles (Fisherbrand, FB8002000) with PTFE stoppers were rinsed three times with water from the clean deionized water system of the ship and subsequently pre-rinsed (three times) with sample water from the Niskin bottle. Finally, a 2-l aliquot was tapped from the Niskin bottle into the gl*** bottle and immediately sealed with the stopper. The samples were stored in a dark and cool environment until further ***ysis in our home laboratories. To safeguard against contamination concerns, we performed a series of field blanks (see the ‘Quality ***urance and control’ section).

TD-PTR-MS ***ysis

The water samples were processed in the PTR-MS lab at the Institute for Marine and Atmospheric Research Utrecht. During the time of ***ysis, the lab was thoroughly cleaned and dedusted on a weekly basis. Typically, only one person was present in the lab during ***ysis to minimize potential contamination. Blanks were included with every sample batch to account for the risk of airborne contamination. For future work, processing samples in a cleanroom should be considered, although the effectiveness of clean labs in eliminating plastic contamination at the nanoscale is at present uncertain. The 2-l samples were homogenized by shaking the bottle before subsampling. Immediately afterwards, an aliquot of 10 ml was taken from the 2-l gl*** bottle and stored in a pre-combusted gl*** chromatography vial (VWR). To separate nanoplastics from microplastics, the 10-ml aliquot was filtered through a 1.0-µm PTFE syringe filter. For further ***ysis, subsamples were prepared in triplicate, for which 1.5 ml of sample was pipetted into a new pre-combusted gl*** chromatography vial. The water matrix was removed using an evaporation/sublimation system⁵⁸. The dried samples were introduced to the PTR-MS unit through a thermal desorption system, using a heating protocol defined as follows: starting temperature of 50 °C, followed by a quick increase at 1 °C s⁻¹ to 100 °C, then a temperature increase to 200 °C at a rate of 0.19 °C s⁻¹ and, finally, the temperature was increased to 360 °C at a rate of 0.44 °C s⁻¹. The final dwell time was 1 min at 360 °C. The thermally desorbed compounds were carried by a constant stream of zero air at 50 SCCM to the PTR-ToF-MS instrument (PTR-TOF 8000, Ionicon Analytik). The inlet temperature was set to 180 °C and the drift tube operation parameters were set to 2.90 mbar, 477 V and 120 °C, resulting in an E/N of approximately 120 Td.

Nanoplastic quantification

The software PTRwid was used to extract the m*** spectra⁵⁹. For data reduction, the m*** spectra were averaged over a time period of 5 min once the thermal desorption unit reached a temperature of 200 °C, that is, we only considered the time window from 200 °C to 360 °C, during which most of the plastic thermally desorbs. Hence, much of the organic matter matrix was excluded from ***ysis, as many monomers and most volatile compounds typically desorb at temperatures below 200 °C (refs. ^4,33,58). Data integration for oven temperatures from 200 °C to 360 °C not only excludes volatile compounds but also avoids pyrolysis and extensive thermolysis of the sample matrix. Consequently, our method measures collectively free nanoplastics and nanoplastics that are loosely ***ociated to organic matter or that are aggregated, provided that the aggregates p*** filter pores (≤1 µm) during prefiltration. To account for background contamination, the m***-specific average of the lab blanks from the corresponding sample batch was subtracted from the averaged nanoplastic m***es in the samples. After subtraction, a 3σ limit of detection filter was applied, for which the m***-specific signal was set to zero when it did not exceed three times the standard deviation of the lab blanks. The lab blanks consisting of HPLC water (VWR, filtered with 0.2-μm filter, CAS number 7732-18-5) were subjected to similar preparation and ***ysis as performed for the normal samples. In this manner, we corrected for background noise and possible procedural contamination in the samples. The pre-processed data were subsequently used for nanoplastic fingerprinting against chemically unaltered plastics (the library m*** spectra) as described in detail in previous works^4,33. The fingerprint algorithm compares the spectra against a library comprising the seven most prevalent polymers: PE, PET, PS, PP, PPC, PVC and tyre wear. A matching score of 2σ (z-score = 2, P < 0.02275, one-tail distribution) was considered a positive fingerprint. Algal organic matter may slightly increase false-positive PS detection (see the ‘Quality ***urance and control’ section and Sarg***um experiment in Extended Data Table 1). To minimize this risk of false-positive annotations, we only considered a z-score of 4 or higher as a positive fingerprint match for PS. Matching scores are indicated with * (z-score > 2), ** (z-score > 3) and *** (z-score > 4), for which a higher matching score indicates a better fit with the library m*** spectra. We conducted a Monte Carlo ***ysis to ***ess the potential interference of organic matter with plastic fingerprinting. The ***ysis showed that plastic overestimation did not exceed 31% before the match fails (Extended Data Fig. 7). Ion counts were converted to mole fraction using:

in which k is the reaction rate coefficient, t the residence time of the primary ions in the drift tube, [MH⁺] the protonated ***yte and [H₃O⁺] the proton donor, hydronium. tr(mH₃O⁺) and tr(mMH⁺) represent the transmission functions of the hydronium and protonated ***yte. The mole fractions were then converted to plastic concentrations (mg m⁻³) by correcting for the sample load and dilution factor. Duplicate measurements instead of triplicate are available for station 9 in the mixed layer, stations 5 and 8 at 1,000 m water depth and station 5 in the bottom-water layer owing to file-corruption issues. Presented nanoplastic concentrations are semiquantitative as not all of the plastic material is eventually converted into detectable ions. This is because of (1) thermal desorption not being perfectly efficient and (2) fractions of the ***yte ending up as non-***ysable ions. Hence, the reported concentrations represent the lower limit of nanoplastic concentrations. Spike-and-recovery experiments were carried out for PS. Homogenized suspensions of 100 or 200 ng of PS was loaded into a vial along with 1.5 ml of seawater sample. Fingerprinting these spiked samples consistently yielded positive matches for PS with z-scores of 4 or higher. By contrast, only 29.4% of the unspiked mixed-layer samples with PS showed z-scores of 4 or above. Spiking experiments were performed in triplicate to obtain a reliable recovery rate (Extended Data Table 2). The spiking experiment revealed a recovery/ionization efficiency rate of roughly 7% ± 2.2, which agrees with our previous works^4,33,35. This entails that the actual PS concentrations in the samples might be 14 times higher. Because of the difficulties in loading precise amounts of plastic in the nanogram range, spike-and-recovery experiments have not yet been performed for PVC or PET. In a previous study, a linear correction factor of 5.28 ± 1.48 for PS and a nonlinear correction factor between 15.05 ± 0.9 for 59 ng PET load and 26.06 ± 6.8 for 177 ng PET load have been reported⁴. A cross-library correction was applied for PS and PVC concentrations, as these polymer m*** spectra partially overlap, resulting in artificially higher PS concentrations when PVC is present and vice versa. These cross-library corrections were calculated on the basis of a 1:1 mixture of 1,000 ng PS and 1,000 ng PVC constructed from library m*** spectra which were subsequently fingerprinted.

Moreover, high PS contents were found to lower the PVC matching score, potentially leading to false negatives in PVC detection. This probably affected the surface samples at station 12, at which high amounts of PS but low amounts of PVC were observed. Concentrations of PET were found to be unaffected by the presence of other polymers, owing to its very distinctive m*** spectrum.

Quality ***urance and control

Several field blanks were carried out to monitor potential plastic contamination during sampling. We performed field blanks in triplicate at the beginning, middle and end of the cruise, amounting to nine field blanks in total. The Niskin bottles were flushed twice using Milli-Q water and rinsed once more with HPLC water. Then, 2.5 l of HPLC water was poured into the Niskin bottles and left for 1 h in the Niskin bottle to simulate the time that is needed for the CTD sensor to reach the surface of the ocean after closing a Niskin bottle at depth. The Niskin bottle with HPLC water was then sampled in a similar manner as for the normal seawater samples. Field blanks were ***ysed in the same batches as normal samples. Although we found a low background signal of nanoplastics in the lab blanks (0.90 ± 1.45 mg m⁻³ averaged over all polymers and all lab blanks), the field blanks did not contain substantial further nanoplastic contamination (Extended Data Figs. 3 and 4); hence, we concluded that the low concentrations of background nanoplastics originated from the preparation and procedures in our laboratory and not from the sampling procedure. The average nanoplastic background concentration of 0.90 ± 1.45 mg m⁻³ is low compared with the transect averages of 18.1 ± 2.1 mg m⁻³ for the mixed layer, 10.9 ± 1.6 mg m⁻³ for 1,000 m depth and 5.5 ± 0.6 mg m⁻³ for the bottom layer.

To ***ess potential false positives from organic matter, we ***ysed Sarg***um biom*** samples as a proxy for complex organic material. Sarg***um is abundant in the Sarg***o Sea and disperses to other parts of the Atlantic, including the northeast⁶⁰. Approximately 0.5 mm³ of Sarg***um biom***—collected during our previous campaign and stored frozen—was dried in an oven at 50 °C for 2 h before TD-PTR-MS ***ysis. The Sarg***um biom*** samples (no digestion applied) showed no positive matches for PE, PP, PET, PVC, or tyre wear particles and only a negligible match for PS, characterized by a low final PS quantity and a low algorithm matching score (see Extended Data Table 1). To maintain a conservative approach, we considered this PS match as a potential false positive in our water samples and, accordingly, increased the PS matching threshold to eliminate such false positives across all samples.

The missing PE and PP nanoplastic paradox

We could not detect PE and PP nanoplastics in this study (Extended Data Fig. 8). The only other study investigating nanoplastics in surface waters of the NASG (using pyrolysis–gas chromatography–m*** spectrometry)²⁸ could also not find a clear PE signal matching the pyrolytic fingerprint of their PE standard. Neither PE nor PP nanoplastics were reported along Atlantic or Pacific coastlines⁵. This is surprising considering that PE and PP account for about half of the global plastic production⁶¹ and have been found as the most abundant floating polymer types in the ocean, including the NASG^6,7,46. We cannot fully explain this at present as our method has proved suitable to measure PE and PP—provided the chemical composition remains unaltered—in freshwater, air and marine biota samples^33,35,62,63, in which it was the dominant polymer. Consequently, possible explanations are the following: (1) the nanoplastics are chemically modified in seawater compared with unaltered polymers so that m*** spectrometric fingerprinting cannot detect the modified PE/PP; (2) the concentration of PE and PP nanoplastics were below our detection limit; or (3) the chemical composition of PE or PP is masked by the organic background in ocean water. We cannot rule out any of these explanations. However, through a Monte Carlo ***ysis (Extended Data Fig. 7), we could indeed show that PE identification was most sensitive to the effect of randomly added organic matter. It also seems very likely that photodegradation not only leads to the production of secondary nanoplastics from parent macroplastics/microplastics^3,24 but that the secondary PE and PP nanoplastics have also undergone some chemical alteration^23,28 (for example, photooxidation introduces carbonyl groups³). This might result in a disparity with the diagnostic fingerprint and would explain why the ions typically ***ociated with PE or PP were not detected.

Calculation of the mixed-layer volume

The dynamic height anomaly (DHA) contours of Ψ (m² s⁻²) as defined in Section 3.27 of ref. ⁶⁴ were used to define the NASG:

Here k = (0, 0, 1), f is the Coriolis parameter (s⁻¹), v is the geostrophic velocity (m s⁻¹) with respect to some reference pressure P_ref and v_ref is the reference velocity at P_ref. The gradient of the DHA was taken at constant pressure as ({nabla }_{P}varPsi =left(frac{partial varPsi }{partial x},frac{partial varPsi }{partial y},0right)). For this study we choose P_ref = 1,000 dbar. This was combined with flow velocities derived from Argo floats at parking level⁶⁵. Ψ_ref was defined as the relative DHA, set relative to 1,000 dbar. Ψ_ref was defined as the reference DHA, such that the sum

equals the DHA. Here Ψ_rel can be directly obtained from the thermal wind balance.

To calculate Ψ_rel, we used the annual mean World Ocean Atlas 2018 1° gridded climatology⁶⁶ as input for in situ temperature and practical salinity. This was then converted into conservative temperature (CT) and absolute salinity (SA) using the Gibbs Seawater software toolbox⁶⁷. Both CT and SA were used as input for the gsw_toolbox function ‘gsw_geo_strf_dyn_height’ to calculate Ψ_rel with respect to 1,000 m (Extended Data Fig. 5b). To obtain Ψ_ref, we constructed an inverse estimate (Extended Data Fig. 6) equated as follows:

Here i represent longitudes and j represents latitudes, both limited to the North Atlantic basin. Δx and Δy are the related distances and u and v are the eastward and northward velocities, respectively. Each Ψ_ref can be included in up to four equations, which can be written as Ax = b. Here x are the unknown stream functions, b is the known right-hand side values of equations (4) and (5) and A is a matrix containing −1 or 1 that multiplies the unknown x (Ψ) values. This set of equations is solved using MATLAB least-squares minimization machinery given by x = Ab, giving the reference DHA Ψ_ref (Extended Data Fig. 5a).

To define the NASG, we first considered that the gyre is mostly concentrated in the upper 400 m (Fig. 1 in ref. ⁶⁸). On the basis of the World Ocean Atlas vertical grid sizes, we averaged over the upper 410 m. The resulting streamlines of the DHA (Extended Data Fig. 6) correspond well to model-based Lagrangian trajectories (Figs. 1d and 3 in ref. ⁶⁸) and stream function (Fig. 1 in ref. ⁶⁹). This supports that the observation-based DHA streamlines calculated here are an accurate indication of the flow field.

To further define the gyre, we selected the last streamline (8 m² s⁻²) that loops from the northern part of the NASG to the southern part without crossing the coast (Extended Data Fig. 6). We used a lower bound latitude cut-off of 8.5° N, as this corresponds with the most western extent of the 8 m² s⁻² contour line. The northern bound of our study region was set at 55° N, as that separates the subpolar area from the temperate to subtropical region in which we sampled. The NASG is then bounded by the 8 m² s⁻² contour (black dots in Extended Data Fig. 5c), whereas the residual area bounded landwards by a 200-m isobath is defined as ‘outside gyre’ (red plusses in Extended Data Fig. 5c).

The climatological mixed-layer depth was calculated⁷⁰ using World Ocean Atlas November mean data (Extended Data Fig. 5c). The station mixed-layer depths were calculated from the CTD sensor measurements from this study (Extended Data Fig. 5c). Although the CTD sensor occasionally measured deeper instantaneous mixed-layer depths than the climatological mean, they are within expectations. Therefore, we used the World Ocean Atlas climatological mixed-layer depth values as a first-order estimate to determine the mixed-layer volume both inside and outside the gyre. For the calculation of the macroplastic/microplastic m*** inside and outside the NASG, we extracted the modelled concentration values from ref. ¹ and overlaid these onto the World Ocean Atlas grid points. This allowed us to make a direct comparison with our nanoplastic data.

Sensitivity ***ysis of the fingerprinting algorithm

To evaluate the uncertainty in potential overestimation of our plastic identification approach (for example, owing to the presence of natural organic matter), we performed a Monte Carlo ***essment⁷¹. We simulated the addition of organic matter to the m*** spectra of our plastic library and ***essed identification and quantification performance. We systematically added 50–350% (increment of 50%) of signal randomly spread over up to 5, 10 and 40 ions of our library used for the identification of nanoplastics. Each sequence of the run was done in 1,000 replicas.

Our Monte Carlo ***ysis showed that the identification of PET and PS was least affected by the simulated addition of organic matter. We could add 200% of the organic matter in relation to the polymer signal without compromising identification of these two plastics. PVC plastic identification was affected more strongly; addition of more than 100% progressively reduced the plastic identification of the fingerprinting algorithms. PE identification was mostly affected by organic matter presence, for which the recognition of the polymer was greatly affected already when about 50% organic matter was added.

On the other hand, the Monte Carlo ***ysis showed that the overestimation in all scenarios (different levels of organic matter impurity spread over different numbers of ions) for all plastic polymers did not exceed 31%. For PET, for example, increasing the organic matter background by 100%, 150%, 200% or 250% of the polymer signal, the overestimation was only about 20%, 27%, about 31% (peak) and about 10%, respectively (Extended Data Fig. 7). In other words, if a sample contains a high amount of natural organic matter, the plastic recognition (fingerprint match) is likely to fail before the nanoplastic amount is overestimated by >31%. Thus, we consider our results conservative, with a possible overestimation of roughly 30% owing to the organic matrix effects.