Bacteriohopanetetrol-x: constraining its application as a lipid biomarker for marine anammox using the water column oxygen gradient of the Benguela upwelling system

Interpreting lipid biomarkers in the sediment archive requires a good understanding of their application and 10 limitations in modern systems. Recently it was discovered that marine bacteria performing anaerobic ammonium oxidation (anammox), belonging to the genus Ca. Scalindua, uniquely synthesize a stereoisomer of bacteriohopanetetrol (‘BHT-x’). The ratio of BHT-x over total bacteriohopanetetrol (BHT; ubiquitously synthesized by diverse bacteria) has been suggested as a proxy for water column anoxia. As BHT has been found in sediments over 50 Myr old, BHT-x has the potential to complement and extend the sedimentary biomarker record of marine anammox, conventionally constructed using ladderane lipids. Yet, 15 little is known about the distribution of BHT-x in relation to the distribution of ladderanes and to the genetic evidence of Ca. Scalindua in modern marine systems. Here, we investigate the distribution of BHT-x and the application of the BHT-x ratio in relation to distributions of ladderane intact polar lipids (IPLs), ladderane fatty acids (FAs) and Ca. Scalindua 16S rRNA genes in suspended particulate matter (SPM) from the water column of the Benguela upwelling system (BUS), sampled across a large oxygen gradient. In BUS SPM, high BHT-x abundances were restricted to the oxygen deficient zone on the continental 20 shelf (at [O2] <45 μmol L, in all but one case). High BHT-x abundances co-occurred with high abundances of the Ca. Scalindua 16S rRNA gene (relative to the total number of bacterial 16S rRNA genes) and ladderane IPLs. At shelf stations with [O2] >50 μmol L, the BHT-x ratio was <0.04 (in all but one case). In apparent contradiction, ladderane FAs and low abundances of BHT and BHT-x (resulting in BHT-x ratios >0.04) were also detected in oxygenated offshore waters ([O2] up to 180 μmol L), whereas ladderane IPLs were undetected. The index of ladderane lipids with five cyclobutane rings (NL5) 25 correlates with in situ temperature. NL5 derived temperatures suggested that ladderane FAs in the offshore waters were not synthesized in situ but were transported down-slope from warmer shelf waters. Thus, in sedimentary archives of systems with known lateral organic matter transport, such as the BUS, relative BHT and BHT-x abundances should be carefully considered. In such systems, a higher BHT-x ratio may act as a safer threshold for deoxygenation and/or Ca. Scalindua presence: our results and previous studies indicate that a BHT-x ratio of ≥0.2 is a robust threshold for oxygen-depleted waters ([O2] <50 30 μmol kg). In our data, ratios of ≥0.2 coincided with Ca. Scalindua 16S rRNA genes in all samples (n = 62), except one. Lastly, when investigating in situ anammox, we highlight the importance of using ladderane IPLs over BHT-x and/or ladderane FAs; these latter compounds are more recalcitrant and may derive from transported fossil anammox bacteria remnants.

To constrain past and present N cycle variations, lipid biomarkers can be employed (see Rush and Sinninghe Damsté, 2017 for a review). Subsequently, biomarker information can be applied for predictions of future N cycling variations (e.g. Monteiro et al., 2012). Anammox bacteria uniquely synthesize ladderane fatty acids (FAs) and ladderane glycerol monoethers, which contain three or five linearly concatenated cyclobutane rings, designated respectively as   (Rush et al., 2014), testifying to its potential application as anammox marker. Moreover, since BHT is ubiquitous and the total BHT (BHT-x ratio) to be a proxy for water column deoxygenation ([O2] <50 µmol kg -1 ). These studies show the potential of BHT-x to complement and extend the ladderane biomarker record.
To better interpret BHT-x as a biomarker in the sedimentary record, either as an indicator of the presence of marine anammox bacteria, Ca. Scalindua spp., or as a proxy for water column deoxygenation, it is imperative to establish how  x is distributed in modern marine oxygen-depleted systems. In this study, we combine measurements of BHT-x, ladderane lipids (both as IPLs and FAs) and 16S rRNA marker genes in suspended particulate matter (SPM) across a redox gradient in the water column of the Benguela upwelling system (BUS). The BUS, located along the southwest African continental margin ( Fig. 2), supports one of the most productive regions in the world. The high primary productivity on the broad but shallow continental shelf results in a perennial OMZ off-shelf, between ~200-500 m below sea surface (mbss). Additionally, annual 95 variation in upwelling intensity leads to a seasonal, on-shelf oxygen deficient zone (ODZ; here defined as [O2] <5 µmol L -1 in bottom waters), which develops in late austral summer (Chapman and Shannon, 1987 therefore is an optimal modern marine system to assess the distribution of BHT-x in relation to that of ladderane IPLs, ladderane 100 FAs and Ca. Scalindua 16S rRNA gene sequences, as a function of water column oxygenation.

Hydrographic setting
The BUS is located off the southwest African coast, where the cold and nutrient rich waters of the Benguela current (BC) are 115 upwelled through a combination of wind-driven Ekman transport and collision with the African continental shelf. The studied area is situated in the northern part of the BUS (16-26°S and 10-16°E; Fig. 2). Here, changes in the offshore wind field, which affect upwelling and hence primary production, result in seasonal variations and movements of the oxygen depleted waters (Chapman and Shannon, 1987). The northern border of the BUS is delineated by the dynamic Angolan Benguela Front (ABF; ~16-20°S; Fig. 2), where the warm and oligotrophic waters of the Angola current (AC), transporting the oxygen poor (<45 120 µmol L -1 ) South Atlantic Central Water (SACW) southwards, converge with the cold and nutrient-rich waters of the equatorward BC. Seasonal variations in the intensity of the AC and the BC control dissolved oxygen (DO) concentrations in the BUS (Mohrholz et al., 1999;Brüchert et al., 2006). This results in a near permanently present OMZ located off the continental shelf (~200-500 mbss; [O2] ~20-50 µmol L -1 ) and a seasonally variable ODZ on the continental shelf (~50 mbss to seafloor; [O2] <5 µmol L -1 ), where the most severe oxygen depletion occurs during late austral summer (Chapman and 4 Shannon, 1987;Bailey et al., 1991;Mercier et al., 2003;Ekau and Verheye, 2005;Brüchert et al., 2006;Mohrholz et al., 2008).
In the south, the northern BUS is bordered by the Lüderitz upwelling cell around ~26°S (Boyer et al., 2000).  (Chapman and Shannon, 1987), which, combined with limited cross-shore bottom-water ventilation at the start of the annual upwelling cycle, led to severe oxygen depletion (Mohrholz et al., 2008). Sampling was performed at various water depths at 13 stations (11°22' 36.5''-14°47'34.8''E and 17°16'38.3''-25°12'25.0''S; Fig. 2; Table 1), covering a large range in water column oxygen concentrations ( Fig. 3a; Fig. 4b, f). From here on, the deeper stations (>300 mbss stations 1,2,8,9,59), sampled in the OMZ off the continental shelf will be termed 'offshore stations', whereas the shallower stations (<120 mbss; 140 stations 3-6, 10, 18, 117, 140) sampled on the continental shelf will be termed 'shelf stations' (Table 1). At each station, physical parameters of the water column were recorded with a Sea-Bird SBE911+ conductivity-temperature-depth (CTD) system. The CTD was equipped with an additional SBE 43 oxygen electrode (Sea-Bird Electronics, WA, USA) to measure DO (detection limit of 1-2 µmol L -1 ). A NIOZ-made Rosette sampler of 24 x 12L Niskin bottles with hydraulically controlled butterfly lids was used to collect water for nutrient and DNA analysis. A ~0.5 bar overpressure of N2 gas was applied to the 145 Niskin bottles to collect water without introducing oxygen. Water samples for DNA analysis were collected into pre-cleaned acid-washed Nalgene bottles. Ca. 2 L of water was filtered over 0.2 µm pore diameter Millipore Sterivex filters from the Nalgene bottles kept on ice in a climate-controlled container set at 4°C. 2 mL of RNALater (RNA protect bacteria reagent, Qiagen) was applied to the Sterivex cartridges, which were then sealed with parafilm and stored at -20 °C until further processing. Suspended particulate matter (SPM) for lipid analysis was collected using four McLane Large Volume Water 150 Transfer System Sampler (WTS-LV) in situ pumps, which were deployed for 4h (~40-900 L filtered; McLane Laboratories Inc., Falmouth, MA, USA). Water was filtered over pre-ashed GF75 grade glass fibre filters of 0.3 µm pore size and 142 mm diameter (Advantec MFS, Inc., USA). Filters were wrapped in aluminium foil and stored at -80 °C until further processing.
Water column sampling depths were chosen based on the CTD profiles, focusing around and below the redoxcline (Table S1).

Nutrient analysis 155
Samples for nutrient analysis were sub-sampled from the Niskin bottles with pre-flushed 60 mL high-density polyethylene syringes with a three-way valve. Samples for PO4 3-NO2 -, NO3and NH4 + analysis were filtered over a combined 0.8-0.2 µm Supor Membrane Acrodisc PF syringe filter (PALL Corporation, NY, USA) into pre-rinsed 5 mL polyethylene vials and analysed onboard with an autoanalyzer (QuAAtro39 Gas Segmented Continuous Flow Analyser, Seal Instruments). Detection limits for PO4 3-, NO2 -, NO3and NH4 + were 0.005, 0.003, 0.015 and 0.019 µmol L -1 , respectively. The fixed inorganic nitrogen 160 deficit was calculated as: in which 16 reflects the Redfield ratio of N:P (Redfield et al., 1963). 165 and centrifuged for 2 min at 3000 rpm. The supernatant was then transferred to another tube, while the residue was re-extracted thrice (i.e. total of four extraction rounds), where during the last two extractions, the phosphate buffer was replaced with a trichloroacetic acid solution to enable optimal recovery of IPLs (Sturt et al., 2004). Phase separation between the solvent layer and aqueous layer was induced by adding additional DCM and phosphate buffer to obtain a ratio of 1:1:0.9 (v:v:v). The bottom DCM layer, containing the lipid extract, was collected, while the aqueous layer was washed two more times with DCM. The 175 combined DCM layers were dried under N2 gas. This extraction method was also performed on freeze-dried biomass from a Ca. Scalindua brodae enrichment culture, obtained from an anoxic sequencing batch reactor at Radboud University, Nijmegen,

Modified Bligh and Dyer extraction
The Netherlands (described in Schwartz-Narbonne et al., 2019).  Table S4 for exact masses). Only PC C20[3] monoalkylether (from here on termed 'PC ladderane) and an ether-ester PG C20[3]-C18[5] (from here on termed 'PG ladderane') were detected in the BUS SPM samples (Fig. 1c). ladderane IPLs, abundances are reported as the peak area response (response unit, ru) per litre of filtered water. Although this does not allow for quantification of absolute concentrations, it does allow for quantification of the relative abundances, as the 210 response factor should be identical across the sample set.

Ladderane fatty acids
SPM from stations 2, 6 and 140 were additionally analysed for ladderane FAs. BDE aliquots were saponified by adding 2 ml of KOH (1 M in 96% MeoH) and refluxing for 1 h. Then, 2 mL of bidistilled water was added and the pH was adjusted to 3 with 1:1 HCL:MeOH (v:v). To collect the fatty acid fraction, 2 mL of DCM was added, after which the tube was sonicated for 215 5 min and centrifuged for 2 min at 3000 rpm. The fatty acid fraction (DCM layer) was collected, and the procedure was repeated two more times. Fatty acid fractions were then dried over a sodium sulfate (Na2SO4) column. The fractions were then methylated using diazomethane to convert FAs into their corresponding fatty acid methyl esters (FAMEs). To remove polyunsaturated fatty acids (PUFAs), extracts were dissolved in DCM and eluted over a silver nitrate (AgNO3) impregnated silica column. Lastly, the FAME fractions were filtered through a 0.45 µm PTFE filter (BGB, USA) using acetone. 220

Ladderane fatty acid analysis
Purified FAME fractions were analysed on an Agilent 1290 Infinity I ultra high performance liquid chromatographer (UHPLC) equipped with a thermostatted auto-injector and column oven, coupled to a Q Exactive Plus Orbitrap MS, with an atmospheric pressure chemical ionization (APCI) probe (Thermo Fischer Scientific, Waltham, MA). Separation was realised with a ZORBAX Eclipse XDB C18 column (Agilent, 3.0×250 mm, 5 µm), maintained at 30°C. MeOH was used as an eluant at 0.18 225 ml min -1 with a total run time of 20 min. Optimal APCI source settings were determined using a qualitative standard mixture isolated from an anammox enrichment culture grown in sequencing batch reactors, containing both Ca. Scalindua wagneri and Ca. Kuenenia stuttgartiensis (described in Kartal et al., 2006). The index of ladderane lipids with five cyclobutane rings (NL5) was calculated to quantify the trends in ladderane chain lengths with respect to temperature, using: (3) 240 Following, the relationship between NL5 and temperature is then given by: with temperature (T) in °C (Rattray et al., 2010). 245

DNA extraction and phylogenetic analysis
Millipore® Sterivex™ filters were extracted for DNA using the Qiagen DNeasy Powersoil kit®. PCR reactions of the DNA templates were conducted with the Qiagen® PCR reagents (Taq PCR Master Mix Kit). The universal prokaryotic primer pair, forward 515F-Y and reverse 926R (Parada et al. 2016) was used to target the V4-V5 small subunit ribosomal RNA region, and was modified with 12 different nucleotide barcodes at both the forward and reverse primer. The 515F-Y/926R primer pair 250 has been found to successfully target Ca. Scalindua (e.g. Yang et al., 2020) and is reported to reflect marine community compositions well (Parada et al., 2016). Reagents were mixed with fifty times diluted DNA template (2 µL) by addition of: 11.75 µL Nuclease free water, 5 µL 5x Qiagen Phusion buffer, 2 µL dNTPs (2.5 mM), 3 µL of 515F-Y/926R primer pair (4 µmol L -1 ), 1 µL BSA (20 mg ml -1 ) and 0.25 µL Taq polymerase (5 units µL -1 ). Negative controls were included during extractions and PCR reactions. Amplification was performed using the following PCR program: 30 s at 98°C (1 cycle), 10 s 255 at 98° followed by 20 s at 50°C and 30 s at 72°C (30 cycles), 7 min at 72°C (1 cycle) and ending with 5 min at 4°C. To quantify DNA concentrations of the PCR reagents, PCR products were mixed with a Xylene Ficoll loading dye and loaded on a 2% agarose gel, together with a home-made Escherichia coli quantification standard dilution series (20, 10 and 1 ng µL -1 ). Gel electrophoresis was performed for 1 h at 75 V. After, the gel was stained with Ethidium bromide. Gels were imaged using GeneSys (lightning, TLUM -Mid Wave; filter, UVO32). The 400 bp bands were then quantified using the 'QuickQuant' 260 option. Following quantification, all PCR products were pooled in equimolar amounts (40 ng DNA per sample) and loaded on a 2% gel stained with SYBRsafe®. The 400 bp band was extracted from the gel using the QIAquick® PCR Gel Extraction Kit. Concentration of the pooled PCR product (20 ng uL -1 ) was quantified using Qbit (Thermo Fisher Scientific Inc.). Library preparation was performed with a 16S V4-V5 library preparation kit and sequencing with an Illumina MiSeq 2x300 bp sequencing platform (Illumina, San Diega, CA) at the University of Utrecht Sequencing Facility (USEQ, the Netherlands). 265 The prokaryotic 16S rRNA gene amplicon sequences were analysed using the Cascabel data analysis pipeline (Abdala Asbun et al. 2020). Raw forward and reverse reads were merged using PEAR (Zhang et al. 2014). Barcode reads were demultiplexed using QIIME (Caporaso et al. 2010), allowing a maximum of 2 barcode mismatches, a maximum of 5 consecutive low-quality base calls and a maximum unaccepted Phred quality score of 19. Reads were then filtered based on length using the values of the median distribution, with an offset of 10 bp. Sequences were dereplicated with VSEARCH, and 270 subsequently clustered to operational taxonomic units (OTUs) using the UCLUST algorithm in QIIME, with a 97% threshold.
From each OTU, the most abundant sequence was picked as representative with QIIME (Caporaso et al, 2010). Taxonomy  was assigned based on the SILVA database (SSU 138 Ref NR 99), using the VSEARCH tool. OTUs representative of the order Brocadiales (to which Ca. Scalindua spp. belong) were extracted with QIIME using filter_taxa_from_otu_table.py. The filtered sequences were imported in the SILVA NR99 SSU 138 Ref database using the ARB parsimony tool in the ARB software package (ARB SILVA, Germany) to assess the phylogenetic affiliation of the partial 16S rRNA gene sequences. 285 Affiliated sequences were checked for homology and imported in MEGAX using the BLAST search query (Kumar et al. 2018). The twelve OTUs with the largest number of reads and 27 reference sequences were aligned, based on 422 bp, with the Clustal W alignment tool. A Kimura 2-parameter model with Gamma distributed sites was then used to calculate pairwise distances between sequences and to create a maximum likelihood tree in MEGAX, using a bootstrap with 1,000 replicates and the maximum parsimony method to create the initial tree (Kimura, 1980;Kumar et al. 2018). To estimate the relative 290 abundance of Ca. Scalindua spp. 16S rRNA reads in relation to the total amount of bacterial 16S rRNA reads, relative Ca.
Scalindua spp. reads were calculated for each sample (in % of total bacterial reads). Though this does not allow for absolute quantification, it does allow for a comparison of relative abundances throughout the dataset of this study, as all samples were processed and analysed in the same way.

Statistical analysis 295
A multivariate binomial regression was performed with anammox biomarker lipids and Ca. Scalindua 16S rRNA gene amplicon sequences. 16S rRNA gene amplicon sequences were used dichotomously, defined as either the presence or absence of Ca. Scalindua assigned OTUs in a given sample. Pearson's correlations between anammox lipid biomarkers and the physiochemical parameters were also investigated (Matlab, R2019a).

Ladderane IPLs 340
The presence of all ladderane IPLs reported for the Ca. Scalindua brodae enrichment culture (Table S4) and those previously reported for Ca. Scalindua spp. (Rattray et al., 2008) was evaluated in the BUS SPM samples. However, at the time of sampling, only the PC and PG ladderanes (Fig. 1c) were detected in the BUS water column. Furthermore, these ladderane IPLs were found in SPM from a limited number of shelf stations located south of the ABF (stations 5, 10, 117 and 140), below~30 mbss waters. Concentrations ranged between 1.1×10 4 -6.4×10 5 ru L -1 for the PG ladderane and between 2.1×10 4 -4.2x10 5 ru 345 L -1 for the PC ladderane (Table S3). At station 5, PC and PG ladderanes were both present. At stations 10, 117 and 140 the PG ladderane was less abundant or absent at the water column depths where the PC ladderane was found. From here on, their summed abundances are reported as 'ladderane IPLs' (Fig. 4c, g; Fig. 5b, i).   Table S5). At offshore station 2 (Fig 7a) (Table S5) Scalindua (Arb SILVA 132R database). These sequences were further analysed to assess the distribution and phylogeny of 380 Ca. Scalindua in the BUS. Negative controls did not contain reads taxonomically assigned to Ca. Scalindua spp.

Distribution Ca. Scalindua spp. 16S rRNA sequences in the BUS
The relative abundances of Ca. Scalindua spp. 16S rRNA gene sequences in respect to the total amount of bacterial 16S rRNA gene sequences was calculated to estimate the distribution of Ca. Scalindua in the BUS (Figs. 5g, n). Sequences taxonomically assigned to Ca. Scalindua spp. were detected in 11 out of 13 stations, in SPM collected at the shelf (<120 mbss; stations 3-6, 385 10 and 140) and at offshore stations (>300 mbss; stations 1, 2, 8 and 9) but were not detected at stations located north of the ABF (stations 18 and 59). At shelf stations (Fig. 5g), the relative Ca. Scalindua spp. gene read abundance ranged between 0-2.7‰, with highest abundances found below 50 mbss. In surface waters (<30 mbss), no Ca. Scalindua spp. 16S rRNA gene sequences were detected, except at station 140, where Ca. Scalindua spp. was present throughout the water column (0.03-0.53‰). At offshore stations, the relative abundance of Ca. Scalindua spp. 16S rRNA gene copies ranged between 0-1.5‰, 390 with highest relative abundances found in bottom waters at station 2 at 700 and 710 mbss (Fig. 5n). At offshore stations 1, 8 and 9, Ca. Scalindua spp. was detected between 50-390 mbss.

Ca. Scalindua phylogeny 400
Ca. Scalindua 16S rRNA reads (422 bp) were assigned to 66 operational taxonomic units (OTU 1-66) based on 97% sequence similarity (Table S6). Most of the Ca. Scalindua spp. reads (88%) could be assigned to twelve OTUs (OTU-1 to -12) ranging from the relative most abundant to least abundant OTU (Fig. S1). To estimate the phylogenetic relationship of these 12 OTUs to Ca. Scalindua spp. sequences from other OMZs, a maximum likelihood tree was constructed with reference sequences from various other OMZs and anammox enrichment cultures (Fig. 8). In addition, the pairwise distances between these sequences 405 (based on 422 bp) were calculated (Table S7). The phylogeny of the BUS OTUs revealed a cluster of ten OTUs (OTUs 1-4 and 6-9, 11 and 12; Fig. 8), with an overall sequence identity of 96%. OTU-10 displays a relatively large evolutionary divergence (>12%; Fig. 8) and limited sequence identity (88%) to the other BUS OTUs. Highest sequence identity of the BUS OTU cluster is observed with environmental sequences isolated from the Guaymas deep sea hydrothermal vent sediment (98%) and the Black Sea suboxic waters (53 mbss; also 98%), which in turn both exhibited the highest sequence identity to Ca. 410 Scalindua sorokiini (again 98% in both cases). Sequence identity to Ca. Scalindua brodae and Ca. Scalindua spp. sequences detected previously in the Namibian OMZ (Kuypers et al., 2005) was 96%. Lowest sequence identity is seen with Ca. Scalindua wagneri (93%) and the Arabian Sea (94%). OTU-5 is placed outside of the BUS cluster (Fig. 8) and shows the highest sequence identity to Gulf of Mexico and Indian Ocean sediments (98% and 97% respectively). Sequence identity of OTU-5 in relation to the BUS OTU cluster is 94%. 415

Discussion
In the BUS, seasonal shifts in upwelling intensity create large spatiotemporal variability in oxygen concentrations (Bailey, 1991). Anammox has been reported previously in the low oxygen BUS water column (Kuypers et al. 2005), which therefore presents an ideal location to investigate anammox biomarkers. We assessed the distribution of BHT-x across the redox gradient in the BUS, to provide further insights into the application of BHT-x as a biomarker for Ca. Scalindua spp., and its ratio over 420 total BHT (BHT-x ratio) as a proxy for deoxygenation in dynamic upwelling systems.

Anammox markers are constrained to the ODZ in the southern BUS shelf waters
During expeditions 64PE449 and 64PE450, the seasonal ODZ had developed in the southern BUS shelf waters (20°S-26°S) between ~50 mbss and the seafloor, with [O2] down to ~1.5-5.5 µmol L -1 . Nutrient analyses revealed a large N deficit in the 425 ODZ (N deficit >10 µmol L -1 ; Fig. 4d), suggesting major losses of bioavailable N by anammox and/or denitrification.
Relatively high BHT-x abundances are detected at the southern shelf stations (stations 3-6, 10, 140) below 30 mbss (Fig. 5a).  Table 2) and on-shelf N deficiencies were accompanied with relatively high BHT-x and ladderane IPL abundances (Fig. 4). This suggests anammox was at least in part responsible for loss of bioavailable N. In summary, the co-occurrence of BHT-x with Ca. Scalindua spp. 16S rRNA reads, 435 ladderane IPLs and on-shelf N deficiencies, indicate the presence of living (or recently dead) anammox cells in the BUS shelf waters (below ~30 to 50 mbss), consistent with earlier reports of anammox activity on the Namibian continental shelf waters (Kuypers et al., 2005).  , who showed that Ca. Scalindua spp. colonize microscopic particles in the BUS, which provide suitable anaerobic micro-niches. Nevertheless, this was found to be restricted to ambient [O2] levels below 25 µmol L -1 . Likely, our evidence for the presence of anammox bacteria in the more oxygenated BUS shelf waters reflects material transported upwards from the deeper ODZ. Upwelled waters from the BC were clearly distinguishable at stations south of the ABF (stations 1-6, 10, 117 and 140; 20°S-26°S), as indicated by the relatively low SST and salinity at the halocline (Fig. 3b,  450 c; Table S1).

Absence of anammox biomarkers near and north of the Angolan Benguela Front
At the end of austral summer (i.e. the timing of expeditions 64PE449 and 64PE450), the ABF reaches its most southern point 460 and is generally found around 20°S. At this time, strongest oxygen depletion is known to occur around ~24-26°S, while less severe oxygen depletion is observed near the ABF (Chapman and Shannon, 1987;Boyer et al., 2000). At the time of sampling, large horizontal gradients in SST and salinity existed around ~19.8ºS, fanning out seaward (Fig. 3b, c), indicating that the ABF had developed at this latitude. The ODZ did not extend past the frontal zone, and north of the ABF, oxygen depletion occurred only down to ~20 µmol L -1 (between 150-500 mbss; Fig. 4f At station 117, located just south of ABF, [O2] was below ~20 µmol L -1 at ~50 mbss (down to ~3 µmol L -1 at 85 mbss; Fig. 5c), yet evidence of anammox was sparse. Station 117 was the only BUS station where N was not limited (N deficit < 0 µmol L -1 ; Fig. 4h; Table S2), revealing that loss of bioavailable nitrogen by anammox and/or denitrification was absent or 475 Salinity, temperature or nutrient (NO2and NH4 + ) concentrations were not seen to influence biomarker distributions in the BUS: i.e. no correlation was observed between these physiochemical parameters and BHT-x or ladderane IPLs (Table   2). This agrees with earlier findings.

Lateral transport of anammox biomarkers to oxygenated offshore waters
In the more oxygenated offshore waters (up to ~180 µmol L -1 ), BHT-x was observed at stations 1, 2, 8 and 9, whereas ladderane IPLs were not detected and the relative abundance of the Ca. Scalindua spp. 16S rRNA gene was extremely low (Fig. 5).
Potentially, ladderane IPLs (and hence, fresh anammox bacterial cells) were present, but simply below the detection limit of 495 our method. However, various studies have reported a high sensitivity of IPLs when analysed using HPLC-ESI-MS. Especially IPLs with a PC headgroup were found to have a high response factor, likely due to the charged quaternary amine moiety on the PC headgroup (Sturt et al. 2003; van Mooy & Fredricks, 2010;Wörmer et al., 2015). In light of these results, it is unlikely that BHT-x was detected while the PC ladderane remained below the detection limit. Rather, it seems that a living anammox community was absent in offshore waters. Indeed, the offshore N deficit was limited (<4 µmol L -1 ; Fig. 4d) and earlier reports 500 (Kuypers et al., 2005) did not find anammox bacteria to be active in the BUS at an offshore station where bottom waters exceeded 20 µmol L -1 .
Yet, in apparent contradiction, high concentrations of ladderane FAs were detected at offshore station 2 at 125, 250 and 710 mbss, with peak concentrations at 250 mbss. In addition, BHT-x was observed at 250, 310 and 710 mbss, with the highest abundance found at the lowest depth (Fig. 7a). To determine the provenance of ladderane FAs observed at station 2, 505 the NL5 index was used (Table S5) suggests that ladderane FAs observed offshore likely originated in the warmer shelf waters and were transported down-shelf. Mollenhauer et al. (2007) showed that radiocarbon ages of lipid biomarkers in the BUS increased with distance from shore and water depth, as a consequence of lateral organic matter transport over the Namibian margin. In fact, most of the organic matter deposited offshore was found to derive from the shelf (Mollenhauer et al., 2007)

Seasonality in anammox biomarker distributions
The Lüderitz upwelling cell has been identified as one of the most intense upwelling regions in the BUS. In austral winter, the 525 water column near the cell is relatively oxygenated, due to the upwelling of oxygen-rich South Atlantic Central Water (Bailey et al., 1991). However, low-oxygen conditions and even anoxia prevail during austral summer due to the respiration of sinking organic matter supplied by phytoplankton blooms (Bailey et al., 1991;Brüchert et al., 2006). Consequently, continental shelf waters between 24-26°S display large temporal variations in DO concentrations under the influence of the Lüderitz upwelling cell. At the time of sampling, the Lüderitz upwelling cell was apparent at ~26°S, appearing as a water mass with a low SST, 530 low salinities (Fig. 3b, c) and high chlorophyll α concentrations (Table S1). Here, the water column was sampled once in February (station 6) and once in March (station 140), to explore the occurrence and distribution of anammox lipid biomarkers and 16S rRNA gene sequences, as the ODZ developed on the continental shelf (sediment depth 100 mbss).
In February (Fig. 7b), the nutrient, oxygen and temperature profiles show a highly stratified water column. A strong oxycline is present around ~20 mbss, with near anoxic conditions in the bottom waters (down to ~3 µmol L -1 ). 16S rRNA 535 amplicon sequences of Ca. Scalindua spp. and BHT-x were detected below 40 mbss and 50 mbss, respectively, with (relative) abundances increasing with depth. Ladderane FAs followed a similar distribution, increasing in concentration with water column depth. However, ladderane IPLs were not detected throughout the water column, which may indicate that anammox bacteria were not yet a dominant feature in the water column community. Possibly, BHT-x and ladderane FAs at this station were laterally transported from more southern shelf sites (Mollenhauer et al., 2007;Blumenberg et al., 2010). The accumulation 540 of ammonium in the bottom waters (Fig. 7b), corresponding to a very high N deficit of 38 µmol L -1 ( Fig. 4h; Table S2), would suggest that denitrification was more active than anammox (Richards et al., 1965).
In March (Fig. 7c), the same sampling location showed distinct differences in physiochemical properties. This is consistent with previously reported seasonality: lower temperatures and increased upwelling commence in austral autumn,

Application and constraints on the use of BHT-x as a biomarker for Ca. Scalindua
In the BUS, sequences taxonomically assigned to Ca. Scalindua spp. were detected at 11 out of 13 stations (Fig. 5g, n). A 560 phylogenetically closely related cluster of Ca. Scalindua OTUs could be identified (i.e. the BUS OTU cluster indicated in Fig.   8 Fig. 5g). Low abundance of marine anammox bacteria in comparison to other phylogenetic groups in marine ecosystems has been reported previously  and is likely caused by slow cell division rates (Strous et al., 1999;Jetten et al., 2009). Even so, it cannot be excluded that well-known PCR biases might 575 also have led to a low coverage of Ca. Scalindua spp. reads. Unequal amplification efficiency of PCR products could result in the preferential amplification of certain 16S rRNA genes, whilst others might be inhibited for amplification (e.g. Pinto & Raskin, 2012).
Ca. Scalindua spp. 16S rRNA gene sequences were also detected in offshore waters. Yet, co-occurrence with BHT-x was limited (only in four of the 19 offshore SPM samples) and the extremely low relative abundance of Ca. Scalindua spp. 580 16S rRNA gene sequences here (0-0.4‰; Fig. 5n) and BHT-x concentrations (factor 10 to 100 lower than at shelf stations) make it unlikely that anammox bacteria formed an active community. Rather, lateral organic matter transport, discussed in section 4.1.3, seems to contribute to the BHT-x concentrations observed offshore. Considerations must thus be taken when interpreting low abundances of BHT-x, as these may inaccurately suggest the presence of living Ca. Scalindua.

Application and constraints on the use of BHT-x ratio as a biomarker for low oxygen conditions 585
In addition to being a useful biomarker for Ca. Scalindua, BHT-x has been applied as a proxy for low oxygen concentration in At BUS shelf stations, when [O2] was >50 µmol L -1 , the BHT-x ratio remained below 0.04, in all but one case (station 5 at 30 mbss). However, at five offshore sites where [O2] was >50 µmol L -1 (up to ~180 µmol L -1 ), BHT-x ratios >0.04 were observed. Likely, transported BHT-x derived from the ODZ on the shelf (see discussion section 4.1.3) and the markedly low BHT concentrations (Table S3), contributed to the relatively high BHT-x ratio signal observed here. When 595 considering both offshore and on-shelf sites, when [O2] was >50 µmol L -1 , the BHT-x ratio remained below 0.2, in all but one case (station 5 at 30 mbss). In addition, a ratio of ≥0.2 corresponded in all cases (except one) with the presence of the Ca. show that that when [O2] is <50 µmol kg -1 , the BHT-x ratio (i.e. BHT-II ratio) is ≥0.2 (except in 1 sample from the Cariaco Basin). 605 Nonetheless, in order to apply this threshold (BHT-x ratio ≥0.2) to infer low oxygen conditions (<50 µmol kg -1 ) in sedimentary records, this signal must be retained in the sediment (i.e. not become diluted by BHT settling from the oxic zone of the water column). Matys et al. (2017) found that BHT II isomer ratios (i.e. BHT-x ratios) observed in surface sediments of the Humboldt current system were comparable to those observed in the OMZ core of the overlying water. In accordance, 625 Berndmeyer et al., (2013) showed that BHPs recorded in the sediment of the Gotland Deep mirrored those of the suboxic zone of the water column. Hence, it is likely that BHT-x ratios observed in the low-oxygen zone of the water column, are retained in the underlying sediments. Moreover, considering the large variety in marine settings (four different upwelling regions and one restricted anoxic basin) and in methodologies, a BHT-x ratio of ≥0.2 is thought to provide a robust threshold in sedimentary records to estimate past low-oxygen conditions (<50 µmol kg -1 ) of the overlying water column, hereby accounting for potential 630 allochthonous BHT-x material.

Conclusion
This study reveals a strong spatiotemporal variability in the presence of anammox bacteria (as reflected by their 16S rRNA gene sequences) and their membrane lipids in the Benguela Upwelling System (BUS), which corresponds to differences in hydrographic characteristics of the water column. By elucidating the distribution of BHT-x across a large oxygen gradient, and 635 comparing it to distributions of ladderane IPLs, ladderane FAs and Ca. Scalindua spp. 16S rRNA gene sequences, we assessed the suitability of BHT-x as a lipid biomarker for Ca. Scalindua spp., as well as its ratio over total BHT as a proxy for lowoxygen water column conditions. On the continental shelf, BHT-x co-occurred with the detection of Ca. Scalindua spp. 16S rRNA genes in all but one cases, further highlighting its suitability as a lipid biomarker for marine anammox in the sedimentary sampled 27 days apart, implied that anammox bacteria only became an established community in the shelf waters at the end of austral summer, when oxygen depletion was most severe. At the offshore stations, ladderane FAs and low concentrations of BHT-x were also observed to accumulate in relatively oxygenated waters ([O2] up to ~180 µmol L -1 ), while ladderane IPLs were constrained to the shelf stations. Calculating the temperature sensitive NL5 index for ladderane FAs, indicated that offshore ladderane FAs were not synthesized in situ and likely originated from the shelf. This must be taken into consideration 645 when using BHT-x and ladderane FAs as lipid biomarkers for in situ water column anammox. Lastly, at shelf stations, when [O2] was >50 µmol L -1 , the BHT-x ratio remained below 0.04, in all but one case. Yet, laterally transported BHT-x resulted in high offshore BHT-x ratio values (>0.04) in oxygenated waters. We therefore suggest to use a BHT-x ratio threshold of ≥0. Author contribution. ZRvK wrote the manuscript; PK and DR were in charge of the research expeditions; ZRvK, DR and PK performed the sample collection; ZRvK performed the laboratory work and data analysis. LV and HJW contributed to the data analysis of the 16S rRNA gene sequences; ECH optimized UHPLC measurements. ECH and DR contributed to the lipid data analysis. DR, JSSD, LV and ZRvK designed and conceptualized the project; All co-authors provided critical feedback and helped shape the research, analysis and manuscript. 670