Diverse arsenic‐containing lipids in the surface ocean

Arsenic is present at nanomolar levels throughout the ocean, and microbes assimilate this potentially toxic element due to its similarity to inorganic phosphorus. Although dissolved arsenic has been a focus of several oceanographic studies, the size and chemical character of the particulate arsenic pool is poorly understood. We measured particulate arsenic in five samples from the open ocean and determined the contribution of arsenic‐containing lipids to this pool. Here we show that the accumulation of arsenic into lipids is a widespread phenomenon in the surface ocean. Particulate arsenic concentrations were 15–42 pmol L−1 with 7–20% of the particulate arsenic pool within arsenolipids. We found that arsenosugar phospholipids dominated the arsenolipid pools in our samples with a minor component of arsenohydrocarbons and other unidentified lipids. A significant portion of the arsenosugar phospholipids (up to 35%) was present as previously undescribed mixed acyl ether lipids, suggesting a bacterial source.

Arsenic is present at nanomolar levels throughout the world's oceans. Its nutrient-like profile indicates accumulation within surface ocean marine microbial communities and subsequent export via sinking particles into the deeper ocean (Middelburg et al. 1988;Cutter and Cutter 1995). Within the dissolved pool, simple organic arsenic compounds such as monoarsenic acid and dimethylarsenic acid build up in nutrientdepleted areas of the ocean, particularly in regions with low phosphorus (Andreae 1978;Cutter et al. 2001;Wurl et al. 2013). The distribution of reduced and methylated arsenic species in seawater suggests that arsenic biotransformations occur primarily in low-phosphorus environments to detoxify arsenate Á as a reducing agent in energy-yielding reactions (Mazumder et al. 2020), and these processes have been inferred to exist in modern marine systems from gene expression studies (Saunders et al. 2019)

[Correction added on 22
October, 2021, after first online publication: AsO 3À 3 À Á was changed to AsO 3À 4 À Á ]. In picocyanobacteria, the genetic potential for arsenic methylation (arsenic methyl transferase, asrM) is more common than the genetic potential for the "reduce and efflux" arsenic detoxification mechanism (Saunders and Rocap 2016), and asrM is widespread among other marine phytoplankton clades (Chen et al. 2017). In coastal ecosystems, macro and microalgae transform arsenic into organicarsenic compounds well beyond what one would expect as a detoxification mechanism (Vinogradov 1953;Lund 1973;Sanders 1979). These organic arsenic compounds accumulate in fish (i.e., Taleshi et al. 2014) and are produced by monocultures of green algae, diatoms, and cyanobacteria in laboratory settings (Duncan et al. 2013;Xue et al. 2014;Řezanka et al. 2019). Arsenolipids have also been observed in lake sediments and lacustrine suspended particles (Glabonjat et al. 2019(Glabonjat et al. , 2020, and in suspended particles from the Atlantic Ocean (Glabonjat et al. 2021). Together, these findings suggest that the production of organic arsenic compounds is a widespread process in aquatic settings. In this work, we quantify the stock of arsenic associated with microbial biomass and explore the composition of arsenolipids in mixed microbial communities in the surface ocean.

Overview
Surface samples for particulate arsenic speciation and quantification were collected from five locations noted in  Table S1). These samples include two from oligotrophic subtropical regions: one from the North Pacific (A Long-Term Oligotrophic Habitat Assessment [ALOHA]) and one from the North Atlantic (Bermuda Atlantic Time Series [BATS]). Three additional samples were collected from the equatorial upwelling influenced Eastern Tropical North Pacific (ETNP), including two offshore samples (ETNP-PS1 and ETNP-PS2) and one coastal sample (ETNP-PS3). We took quantitative subsamples for total particulate carbon and total particulate arsenic. We extracted samples for arsenolipids which we analyzed by liquid chromatography-inductively coupled plasma mass spectrometry (LC-ICP-MS, for quantification) and liquid chromatography-high-resolution electrospray ionization mass spectrometry (LC-HR-ESI-MS, for identification of individual lipids). Details for each step are in Supporting Information Supplemental Methods.

Total particulate carbon
Subsamples of known area were taken from filters and dried overnight at 60 C. Dried samples were pelletized in 9 Â 10 mm tin capsules (Elemental Microanalysis and Costech) and run on a vario ISOTOPE select (Elementar) CHNOS Elemental Analyzer to quantify total C. Measurements were calibrated with an inhouse aminocaproic acid standard (ACROS).

Total particulate arsenic
Three quantitative subsamples from each sample were taken and digested for 4 h in 1 mL of 50% HNO 3 at 110 C (Planquette and Sherrell 2012). We reconstituted the dried digests in 1 mL of 2% HNO 3 , 2.5% MeOH, 1 ppb Rh in MQ water. We analyzed concentrations of these extracts on a Thermo quadrupole ICP-MS (iCAP-RQ, Thermo Scientific) paired with a Thermo Dionex 3000 HPLC used as an autosampler (no column). The ICP-MS was set to monitor 75 As, 77 Se (to monitor for 35 Cl 40 Ar interference), 78 Se, and 103 Rh. We performed sample analyses in kinetic energy discrimination mode with He collision gas at 2.0 L min À1 which eliminated 35 Cl 40 Ar interference.

Arsenolipid analyses
We separated water-soluble compounds from lipid-soluble compounds to yield a crude lipid extract which we analyzed for total extractable arsenolipids. We performed a silica gel clean up step as previously described (Glabonjat et al. 2014) that selects for arsenolipids. We used the same LC configuration for both LC-ICP-MS (for quantification) and LC-HR-ESI-MS (for identification) so peaks could be aligned. We tested multiple gradients with our ALOHA sample to ensure that the full 75 As signal eluted within our chromatographic parameters and ran blanks between each sample. In brief, our LC set up was a Dionex Ultimate 3000 bio-inert LC equipped with a C18 column (ZORBAX SB-C18 by Agilent, 0.5 Â 150 mm, 5 μm particle size, 80 Å pore size) with a flow rate of 40 μL min À1 and a gradient of a mixture of water and isopropyl alcohol, with 1% formic acid.
For quantification of arsenolipids, we injected crude lipid extracts and directed the flow from LC to the ICP-MS and monitored the 75 As trace over time. As there are no commercial standards for arsenolipids, to quantify we used a proxy standard that was retained on our C18 column (cyanocobalamin), using an approach similar to what has been used to quantify nickelbound ligands in seawater (Boiteau et al. 2016). We quantified each chromatographically resolved peak from the baseline (as calculated from the baseline; Liland et al. 2010) and also quantified the total 75 As signal retained on the LC column (not including the 75 As signal associated with unretained compounds in the first 5 min).
For identification of arsenolipids, we injected both the crude and the silica gel-cleaned extracts, directing the flow of the LC to the hybrid quadrupole Orbitrap (Q-Exactive HF, Thermo Scientific) for HR-ESI-MS. We collected both MS 1 and MS 2 data using data-dependent acquisition. We searched the resulting MS 1 and MS 2 scans for masses (MS 1 ) or fragments (MS 2 ) associated with arsenolipids from an in silico database of arsenolipids curated in house from general structures previously reported (summarized in Supporting Information Table S2, database supplied in Supporting Information  Tables S3, S4).

Bulk quantification of total arsenic and arsenolipids in marine suspended material
Total particulate arsenic ranged from 15 to 42 pM and the highest arsenic and As : C ratios were in the oligotrophic samples (HOT and BATS, Table 1). Peaks in 75 As in the LC-ICP-MS chromatograms suggested the presence of individual arsenolipids that could be probed for identification (Fig. 1). By integrating the total signal in the LC-ICP-MS traces (excepting small peaks near the start of the run associated with compounds not retained on the LC column), we estimated the concentration of total extracted arsenolipids to be 2.6-5.7 pM, or 7-20% of the total particulate arsenic pool in our five samples (Table 1).

Identification of organic-extractable arsenic compounds
Distinct peaks in the LC-ICP-MS 75 As chromatograms in each of our samples corresponded to individual arsenolipids when aligned with the LC-HR-ESI-MS data. All samples showed a peak early in the chromatogram at approximately 14 min (Fig. 1), which we identified as an arsenohydrocarbon (AsHC) with an m/z of 333.2139 (AsHC332, Fig. 2), confirmation based on the MS 2 fragmentation spectra (Fig. 3A). In all samples, we saw peaks at 23 and 26 min in the LC-ICP-MS chromatograms which we identified as diacyl glycerol arsenosugar phospholipids (AsSugPLs) with m/z's of 955.4893 (AsSugPL954) and 1011.5519 (AsSugPL1010). These masses were also associated with MS 2 spectra showing diagnostic fragments for AsSugPLs including a 409.0245 fragment corresponding to the phosphorylated arsenosugar with a glycerol headgroup (C 10 H 23 AsO 10 P,  (Fig. 4). We observed additional peaks in the 75 As LC-ICP-MS chromatograms that we could not identify by searching for MS 1 masses of known arsenolipids from our in-house database. In the oligotrophic ALOHA sample, we observed two large 75 As LC-ICP-MS peaks at approximately 21 and 22 min and found two masses (1001.485 and 1003.500) with differing m/z and separation in retention time corresponding to one unsaturation at those times (labeled "unknown but m/z known" in Fig. 1 and Supporting Information Fig. S1). Both masses had major MS 2 fragments of 255.021, 237.010, and 195.000 m/z ( Fig. 3E and Supporting Information Fig. S1), which are likely associated with ions of C 7 H 16 O 5 As, C 7 H 14 AsO 4 , C 5 H 12 O 3 As, matching observed fragmentation spectra for other arsenosugar lipids Glabonjat et al. 2018Glabonjat et al. , 2019 but lacking fragments that would support a phosphate moiety. The high masses of these lipids (m/z's > 1000) suggested two chains, but the There are no significant peaks before the earliest time shown (12 min) except some 75 As signal associated with water soluble species that elutes with the solvent front. Note that different amounts of material extracted accounts for much of the difference in intensities, details in Supporting Information Table S1.
retention times suggested that these lipids were more polar than the smallest observed AsSugPL.
In the ETNP-PS2 and ETNP-PS3 samples, we saw late eluting peaks in the LC-ICP-MS 75 As chromatograms that did not correspond to any masses initially in our database (retention time >26 min in Fig. 1). After consideration, we have identified these as arsenosugar phospholipids with mixed acyl ether glycerols (AEGs), which we refer to as arsenosugar phospho acyl ether lipids (AsSugPeLs, Fig. 2). We have four tiers of evidence to support these identifications. In the LC-HR-ESI-MS data, we saw near identical MS 2 fragmentation between these lipids and the earlier eluting AsSugPL's in the <450 m/z range, strongly suggesting that these lipids all have the same headgroup (Fig. 3, Supporting  Information Fig. S2). The masses that were isolated for fragmentation which resulted in these diagnostic fragments (971.55, 973.57, 999.59, and 1001.60) did not match any predicted m/z's of AsSugPLs with either even or odd chain length but matched well to the calculated masses of AsSugPeLs (<2 ppm for all except AsSugPeL972 which was <6 ppm, Supporting Information Table S5). We also saw less abundant fragments in the MS 2 s of each of these lipids that correspond to the loss of the arsenosugar phospho headgroup of the lipid (neutral loss of 408.016) and to the loss of the fatty acid chain (m/z = 719.347, Fig. 2, Supporting Information Fig. S2). Finally, the later retention time relative to the AsSugPLs of similar masses suggests a less polar lipid class, consistent with an AEG. The even numbered nature of the ether-bound chain in the proposed structures suggests that the ether bound chains are not branched, though this could not be confirmed by our analyses.

Bulk particulate arsenic measurements
Dissolved arsenic in open ocean surface waters is typically 10-18 nM (Cutter et al. 2001;Cutter and Cutter 2006). Our measurements show that particulate arsenic is nearly 1000 times less abundant than dissolved arsenic. This high seawater background obscures our view of transformations of arsenic occurring within or on particles and has hindered measurements of suspended particulate arsenic in aquatic systems for decades (Andreae 1978;Glabonjat et al. 2020). This background is important to take into consideration when loading particulate material onto a filter for analysis-we filtered 193-747 L of surface seawater onto 142-mm filter (Supporting Information  Table S2). To account for sorbed dissolved arsenic we analyzed wet blanks which were never more than 10% of the signal observed in our samples. This work provides an important baseline for future particulate arsenic measurements, both in establishing a general range of values of open ocean particulate arsenic and providing methodology to obtain these data.

Arsenolipids as part of the bulk particulate arsenic pool
Within the bulk particulate arsenic pool, we found that 7-20% was present in a lipid-soluble fraction (Table 1). This arsenolipid pool may act as an important conduit for arsenic transfer into the deeper ocean via sinking particles to maintain the nutrient-like profiles observed for this element (Cutter and Cutter 2006). Furthermore, arsenolipids have been hypothesized to be the precursor for arsenobetaine-an arsenic-containing metabolite that accumulates in mesopelagic particles and higher trophic levels (Duncan et al. 2015;Heal et al. 2021). Lipid-bound arsenic was first observed in seaweed in the 1950s (Vinogradov 1953) and has since been seen within microalgae, macroalgae, and many species of fish (i.e., Lund 1973;Amayo et al. 2013;Raab et al. 2013;Taleshi et al. 2014;Xue et al. 2014;Glabonjat et al. 2018). Algal and cyanobacterial monocultures vary widely in the percent of total cellular arsenic present as arsenolipids in large part because of accumulated or sorbed inorganic arsenic (Duncan et al. 2013;Xue et al. 2014;Glabonjat et al. 2018Glabonjat et al. , 2020. Our observations of 7-20% of total particulate arsenic as arsenolipids are on the higher end of these observations. Lund (1973) found that algae produced more arsenolipids under low arsenic conditions, which has also been observed in the cyanobacterium Synechocystis (Xue et al. 2014). These authors hypothesized that when arsenic exposure is low, arsenolipids serve as a biologically useful molecule (presumably membrane lipids) and simultaneously act as a sink for arsenic, keeping the element's cytosolic concentration low. We observed a higher percentage of particulate arsenic as arsenolipids in the ETNP samples (each about 20% of total particulate arsenic) than either oligotrophic samples (ALOHA and BATS). The ETNP samples are influenced by upwelling and likely have a higher dissolved phosphorous to arsenic ratio (P : As; and therefore lower arsenic exposure), hinting that arsenolipids may accumulate to higher levels when low arsenic exposure does not necessitate an immediate need to efflux toxic levels of arsenic.

Composition of arsenolipids
In our samples, AsSugPLs were the most abundant type of arsenolipids. The diacyl varieties of these lipids have been observed in one species of cyanobacteria and several species of brown macroalgae (García-Salgado et al. 2012;Raab et al. 2013;Xue et al. 2014), thus we hypothesize that cyanobacteria or microbes related to brown macroalgae (e.g., diatoms) synthesize these lipids in the open ocean. Recent work has characterized arsenolipids in sediments and suspended matter in relatively arsenic-rich lakes like Mono Lake in California (Glabonjat et al. 2020) and the Great Salt Lake in Utah (Glabonjat et al. 2019). At both locations, arsenolipid composition was dominated by a rich array of isoprenoidal arsenolipids related to phytol, hypothesized to be produced by green algae including Picocystis which dominate the Mono Lake phytoplankton community (Glabonjat et al. 2020). Additional green algae species produce arsenolipids of similar structure (Glabonjat et al. 2018;Řezanka et al. 2019). In contrast to these lake systems, we did not detect any of these isoprenoidal arsenolipids in these marine samples. One recent study characterized the arsenolipid pool in suspended particles from the Atlantic Ocean and found AsSugPLs, AsHCs, and a minor component of isoprenoidal arsenolipids (Glabonjat et al. 2021). Although chemotaxonomy of arsenolipids may be premature at this point, our observations do not support green algae as a major source for arsenolipids in the open ocean.
In the Atlantic Ocean, Glabonjat et al. (2021) identified six arsenolipids, including two which we also saw in this study (AsHC332 and AsSugPL958). It is possible that our unidentified arsenolipids between 16 and 20.5 min correspond to AsHC360 or AsSugPhytol, but we do not have LC-HR-ESI-MS data to support this (possibly due to low concentrations). The apparent mismatch between the AsSugPLs identified in our samples compared to this previous work (Glabonjat et al. 2021) highlights the importance of cross-lab collaboration in oceanographic studies. These observed differences could be rooted in the oceanographic settings but may also arise from differences due to extraction and sample preparation techniques or differences in detection approach. For example, our extractions did not use NH 3 which Glabonjat et al. (2021) found to aid extraction of arsenolipids from samples collected on glass fiber filters; this may have resulted in an underestimation of arsenolipids in our samples. We relied on LC-HR-ESI-MS for detection of our arsenolipids whereas Glabonjat et al. (2021) used low-resolution mass spectrometry for compound detection in singular samples. The LC-HR-ESI-MS requires more costly equipment, but provides more robust evidence for identification, especially when performed in manner that allows the user to obtain high resolution MS 2 data. For future work in highly complex samples like marine particles, we highly suggest the use of LC-HR-ESI-MS (both MS 1 and MS 2 ) for compound identification, as has been adopted by other fields relying on mass spectrometry for compound identification in complex systems (Sumner et al. 2007). Further comparison between labs would also benefit by the deposition of HR-ESI-MS data into public data repositories, reference standards, intra-lab calibration studies, and a robust extraction comparison.

Mixed AEG arsenolipids
Our observations of AEG arsenolipids are especially intriguing. To our knowledge, these are the first observations of etherbound arsenolipids. Other AEGs have been seen in natural marine systems either as intact lipids (Schubotz et al. 2018)   open ocean sites (Hernandez-Sanchez et al. 2014), and intact polar AEGs have been seen in surface waters near our study sites in the ETNP (Schubotz et al. 2018). Therefore, we think it is unlikely that our observed AEGs originate from sulfate-reducing bacteria in our oxic surface samples. Overall, while AEG lipids are unusual in surface waters, they are not unprecedented.
Eukaryotic algae are considered the main source of arsenolipids in natural systems (Glabonjat et al. 2019(Glabonjat et al. , 2020. Our provocative findings of AsSugPeL lipids suggest at least some of the lipids may be of bacterial origin. Ether-bound lipids in the surface ocean are hypothesized to originate from aerobic bacteria (Hernandez-Sanchez et al. 2014). Previous work in this area has led to the hypothesis that cyanobacteria are a possible source of AEGs based on observations of AEGs with a sulfoquinovosyl headgroup that are commonly associated with cyanobacteria (Schubotz et al. 2018). Especially at the ETNP-PS2 (offshore) and PS3 (coastal) stations where 20-30% of the identified arsenolipids were AEG lipids (Table 1, Fig. 1), we hypothesize that bacteria-either cyanobacteria or aerobic heterotrophic bacteria-process arsenic in ways beyond efflux-based detoxification and biosynthesize arsenolipids.

Conclusions
We analyzed marine suspended particles from five ocean locations to establish a baseline for particulate arsenic and its accumulation into lipids. This work supports the hypothesis that arsenolipid biosynthesis is a widespread phenomenon in marine systems. The characterization of the arsenolipid pool suggests that in addition to eukaryotic algae, bacteria are a possible overlooked source of these fascinating compounds in natural systems.