Rapid abiotic transformation of marine dissolved organic material to particulate organic material in surface and deep waters

Marine particulate organic matter (POM) supports the vertical transport of organics in the oceans, and the ecology of microbes and filter feeders. POM is collected on GFF filters and quantified as particulate organic carbon (POC) and nitrogen (PON). The filtrate contains dissolved organic matter (DOM) that partly is abiotically converted into POM and can be collected on GFF filters. We filtered seawater from cultures and the pelagic ocean, after the initial sample filtration yielding the 10 conventional POC/PON sample (POM1), we refiltered the filtrate yielding POM2, refiltering its filtrate we obtained POM3 and so on till POMi. Refiltered POM tended to the same concentration independent of the sample depth and even after a few refiltrations, independent of the original particulate organic load. POM2/POMi for surface water (<100m) was about 0.1 and for deep water (>1000m) was >1.0. We considered adsorption of DOM and bacteria that had passed through the first filter but concluded that these could not explain the high POM concentration in the filtrates. We suggest that POM in filtrates represent 15 gels formed due to hydraulic stress during filtration. The ratio of POM2/POMi will partly depend on environmental conditions like turbulent energy. We suggest that the increase of POM2/POMi with depth is related to the lower in situ turbulent energy at greater depth. We discuss aspects of POM methodology, including problems with acidification of samples, and the wider ecological implications of our results.

The data of actual samples were less reproducible than standards (Fig. 2), possible reasons may have been sample 130 heterogeneity, time periods between filtration steps, differences in individual filters or hydraulic shear exposure during filtration. The variance does not seem to increase with concentration (homoscedastic behaviour) for either POC or PON, meaning that the coefficient of variation decreases with concentration. We can use two times the standard error of the intercept to estimate the 95% range of values falling within +/-0.51 mol POC L -1 and +/-0.05 mol PON L -1 . Actually, if we use the 12 non-replicate POC and PON values from greater than 1000m depth (Fig.5) and exclude one outlier, we arrive at standard 135 deviations of 0.50 mol POC L -1 and 0.017 mol PON L -1 . This means that the reproducibility of the method is similar to the concentration reproducibility o deep samples and we cannot distinguish between different samples.
After initial trials we decided not to treat our samples with acid because a) we found the traditional acid fume exposure method to be faulty (see below), b) we considered it unlikely that refiltered samples contained inorganic carbon particles, and c) the literature suggests that gels are sensitive to the pH and tend to disintegrate at low pH. We considered the possibility of a low 140 concentration of inorganic carbon in POM 1 samples but we did not want to use a method for POM 1 samples that was different from POM 2 to POM i samples.
We experimented with acid vapor exposure of blank precombusted GFF filters similar to common sample treatment. At the bottom of a glass desiccator, a glass vessel with 11N HCl was placed. The filters were wetted with 0.5 or 1 ml of distilled water and placed in precombusted scintillation vials. Exposure to acid fumes was 24 hours within a glass desiccator. Blank 145 filters were wetted and placed inside precombusted vials before being placed together with the acid treated filters in a drying oven (60 o Celsius). Initially for two experiments, the glass desiccator was sealed with silicon high vacuum grease (Dow Corning). For our experiments with grease free acid vapor treatment, we thoroughly cleaned the glass desiccator initially with acetone, in another experiment with detergent and in a third experiment again with acetone. For all experiments dry precombusted GFF filters were used as blanks. 150 For the processing of Micro Cube instrument data we reviewed all standard and sample peak integrations individually, we then subtracted from the sample peak a blank value obtained by wetting a precombusted GFF filter with 1 mL of sample and otherwise treated the same as the samples. To the difference between sample and blank we applied the regression slope of the acetonitrile standards to calculate the organic carbon and nitrogen of the samples.

Transparent Exopolymer Particles (TEPs) 155
We used the colorimetric method based on Passow and Alldredge (1995) to quantify TEPs (Hakspiel et al, 2017), Samples were collected on 0.45 mm polycarbonate filters (Poretics), stained with 0.02 % Alcian Blue in 0.06 % glacial acid, and dissolved in 80 % sulphuric acid. The spectral peak at 787nm was used to calculate the TEP concentration using a calibration curve of Xanthan Gum (XG) and a TEP carbon equivalence to Xantham Gum of 0.63 gC gXG -1 .

Bacterial counts 160
Bacteria were fixed with buffered formaldehyde (2 % final concentration) before storage in the refrigerator. The sample was incubated with 4′,6-diamino-2-phenylindole (DAPI, final concentration 1 µg mL−1) (Porter and Feig, 1980) and filtered immediately on 0.2 µm black polycarbonate filters (Poretics). A total of more than 300 cells were counted for each sample using a Carl Zeiss epifluorescence microscope with an X100 objective, and a 175W xenon lamp (Lambda LS, Sutter) connected through a liquid light guide. 165 All regressions presented here were of type 2 (Pearson) except where noted. For the comparison of data sets, we used the Mann-Whitney test considering non-normal distribution (Statistica 7.1). No outliers were excluded in the data analysis except when specifically noted.

Acid pretreatment of samples to eliminate inorganic carbon
We investigated the effect of vapor acid exposure to samples to eliminate inorganic carbon. Humidified precombusted GFF filters were exposed to HCl fumes in a glass desiccator (Yamamouro and Kayanne, 1995). The ground glass joint of the desiccator top was sealed dry (experiments 1, 2 and 3) or with silicon high vacuum grease (Dow Corning), experiments 5 and 6. In experiment #3 (Fig. 3) two acid treated filters were accidentally joined in measurement, here we report only one value of 175 half the POC and PON measured. In experiment #5 the HCl was more than 0.5 years old. Experiment #6 used new HCl. Figure 3 i.e. the distilled water wetted GFF samples had on average lower values than the dry GFF used to define the zero. The PON data in Figure 3 showed no significant differences (p>0.05) between acidified (A) and non-acid exposed (B) samples. The average PON of all non-acidified samples was 0.012moles,

Effect of DOC adsorption to a 2nd filter
In part of the experiments we used the 2-filter method to estimate the DOM absorption as has been done before (Menzel, 1967;185 Turnewitsch et al., 2007;Liu et al., 2005Liu et al., , 2009Cetinç et al., 2012). The underlying concept is that the filtrate that had passed the first filter contained only DOM and all organic carbon measured on the second filter underlying the top filter would be adsorbed DOC. The organic carbon of the second filter could then be subtracted from the top filter to correct the top filter sample for the organics adsorbed and thus yield the concentration of organic particles. This experimental setup had only been used for the first filtration step before. Here we repeatedly refiltered the filtrates through a double layer of filters.
The data in Figure 4 clearly show diminishing presence of organic particles in top filters even after repeated double-filter filtrations. The POC retained by the bottom filter was lower than the top filter even after several refiltrations. In Fig. 4C the POC/PON ratio of the top filter was initially below the ratio of the bottom filter but similar after the second filtration. The geometric means of the POC/PON ratio for top filter, was 14 for the 1 st filtration and 21 for the 2 nd to 5 th filtration. The POC/PON ratio for the bottom filter was 130 for the 1 st filtration, and 53 for the 2 nd to 5 th filtration. This change in ratio would 195 be consistent with a reduction in biomass and increase in hydrocarbon gels.
We normalized the data in Figure 4 by dividing the POC of the bottom filter by the POC of the top filter. In Figure 4D the geometric means of the bottom/top filter ratio for POC 1 is 0.11 and for POC 2 it is 0.17. With further refiltrations the ratio is trending towards 1.0.The data suggest that organic particle in form of gels are formed in the filtrate and retained at the next refiltration. One possible explanation is that the aggregation process is helped by the shear stress when the sample is passing 200 through the GFF filter. In this case the POM measured on the underlying second filter could also partly originate from the aggregated DOM that had passed the top filter.

Sequential filtration of samples
In Figure

Transparent extracellular particles, TEPS
Transparent extracellular particles are hydrogels and therefore we expected to find a similar pattern with refiltration as for POC and PON, specifically because both methods use filters with similar lower particle size cut-offs. The initial concentrations 225 (TEP 1 ) in Figure 7A are in the expected range for filters of 0.45m. In contrast to Figure 5, the different depth ranges did not show the same clear differences for the first filtration. All the samples show significant TEP concentrations when the filtrates are refiltered. The 0-100m samples showed a tendency for TEP 2 /TEP 1 to decrease (p < 0.005), but this was not found with deeper samples. The continued presence of TEP in the refiltered samples demonstrates that TEPs can be formed in the filtrates.
In Figure 7B two samples show a strong increase from TEP 2 to TEP 3 , these two samples were kept in the refrigerator during 230 the night to limit bacterial activity, but the low temperature probably promoted the gel aggregation. The formation of TEPs in the filtrates should be an abiotic process because the refiltration of TEP 2 was within about one hour and the presence of bacteria was reduced after the initial TEP 1 filtration. It is known that TEPs exist in a size continuum because when filters with smaller pore size are used to collect TEPs than significantly more TEP is measured (e.g. Hakspiel et al., 2017). The existence of the smaller size class TEPs can then serve as precursors to form the bigger TEP aggregates that are then collected by the 235 conventional pore size of 0.45 m.

The bacterial abundance in refiltered samples
Bacterial biomass should contribute significantly to the POC retained when filtrates are filtered. Lee et al. (1995) reported that 35 to 43% and 49% of the original bacterial concentration passed GFF filters. Figure 8 shows that a small and with successive filtrations diminishing percentage of bacteria are retained by the GFF filter. Our bacterial abundance retention with successive 240 filtration is given for double precombusted GFF filters by Eq. (1) and for single GFF filters by Eq. (1995) did not use pre-combusted filters which might mean that the effective pore size was bigger than for our pre-combusted GFFs. Their retention efficiency lies between our double filter and single filter efficiencies. The bacteria retention efficiency 250 of the first filter, single or in the form of double filters, will depend strongly on the average size of the bacteria and hence on their physiological status. For example, it might be expected that deep ocean bacteria are less efficiently retained by the first filter. As Figure 8 shows, the subsequent filtration steps retain only a few percent of the original bacterial population. It is interesting that the relative filtration efficiency is relatively reduced with each step probably as a result of size selection, with cannot be explained by a phase transition between the collapsed to a swollen state of the gels (Tanaka, 1992), because this transition would only follow a change in sample conditions, for example temperature, pH or ionic concentration, changes that did not happen in the filtrates. Below we interpret our data along the two lines, methodology and ecological implications. 290

Acidification of POC samples
Generally oceanic POC samples are pretreated with acid to eliminate the inorganic particulate carbon. The two methods commonly used are either adding a very small volume of diluted acid to the sample filter, or exposing the sample to acid vapors. There is no consensus about the best method but the latter method is most commonly employed. Publications indicate 295 that for sediment samples the popular acid-vapor method can introduce systematic overestimation (Brodie et al., 2011) and errors in isotopic composition (Schubert and Nielsen, 2000). The latter two publications trace the measurement errors to silicon greased glass desiccator tops used for HCl vapor treatment. Our results show that the acid-vapor treatment using a silicon grease sealed glass dessicator increased the measured POC even in filtered plankton samples, but did not change the PON (Fig.   3). Without using grease, we found some or no increase in POC (Fig. 3). A comparison with global POC data (Martiny et al. 300 2014) shows that the POC added by acid treatment could significantly increase the POC concentration, especially for samples below the epipelagic layer. The relative POC concentration increase would depend on the filtered sample volume, which makes it difficult to correct the POC concentrations reported in data banks without detailed methodological information. Together with the POC, the POC/PON ratio would increase because the PON showed no increase with acidification. We suggest that without greased dessicators tops the acid vapor method would improve but our results do not absolve this method from a 305 possible increase POC concentration.
We took the results in Figure 3 as justification for not having acid-treated our samples. Other reasons included the unlikely possibility that refiltered samples contained inorganic carbon particles; that gels are possibly sensitive to the pH and tend to disintegrate at low pH, and that acidification of POM 1 samples would result in two different methods for POM 2 to POM i samples. 310

Adsorption of organics to the filter surface
One of the methodological artifacts in the POM methodology is the tendency of dissolved organic material to adhere to inorganic filter surfaces (Maske and Garcia, 1994;Moran et al., 1999;Gardner et al., 2003;Rasse et al., 2017) and thus adding a small fraction of dissolved organic matter (DOM) to the POM sample. This adsorbed DOM would form part of the POM blank that has to be subtracted to obtain true POM. Recently Novak et al., (2018) investigated POC retained on GFF filters 315 when different volumes of pre-filtered sample were filtered. He adjusted a global non-linear model suggesting that it would help to correct POC data that were previously not corrected for adsorbed DOC. His results are consistent with our argument that the pre-filtration had caused a phase transition of DOC to POC. For POC previously reported blanks were 1.25 to 1.75 mol C sample -1 (see references in Rasse et al., 2017). Incidentally, considering the results in Figure 3, this would be about the amount of POC concentration added to acid vapor treated oceanic samples if one-liter sample size is considered. Rasse et 320 al. (2017) used two different methods to estimate the contribution of adsorbed DOM, a) filtering different volumes of the same sample and assuming that the amount of adsorbed DOC is independent of the volume filtered and thus indicated by the POC axis intercept; b) refiltering the sample. The latter approach has been discussed by Turnewitsch et al. (2007) and is recommended by the U.S.JGOFS protocol (http://usjgofs.whoi.edu/eqpac-docs/proto-18.html). The refiltration of the sample filtrate for a blank is similar to our approach to estimate the gels formed in the filtrate. Rasse et al. (2017) found that, as the 325 volume of filtrate that was refiltered increased, the measured POC on the filters increased. They did not consider the collection of bacteria or the formation of hydrogels to explain this pattern, but used the POC of an intermediate refiltered volume to indicate adsorbed DOC to the filter. According to this interpretation they found for one cruise that method (b) yielded higher values than method (a), and in another cruise the results were similar. Our results suggest that the POC encountered in refiltered samples has only minor relation with adsorbed organics but is the result of gel aggregation. 330 We consistently found a small amount of organics in a second filters placed directly underneath the top, primary POM filter.
It might be argued that these organics are bacteria that had passed through the first filter (see below) or that they are gel aggregates that are rapidly formed from dissolved organics in the interface between both filters, similar to our POM 2 and subsequent refiltrations. One problem with this interpretation is the very short time available during the filter passage for the gel-precursors to form aggregates. Maske and Garcia (1994) immersed filters in 14 C enriched dissolved organic matter and 335 found that inorganic filters, glass fiber or aluminum oxide, always adsorbed more 14 C than filters made of organic material.
Because they did not pass, the sample through the filters there was no collection of particles on the filter. The collection of 14 C-organics on the filters was within a few seconds and without turbulence that might provoke the formation of gels. Maske and Garcia (1994) interpreted these result as adsorption driven by polarity or electrostatic effects. Therefore, we assume that in our present data (Fig. 5) where the geometric means of the POC ratio of bottom to top filters was 0.11 for POC 1 and 0.17 340 for POC 2 is partially the result of adsorption.

The contribution of bacteria to the POC of refiltered samples
Apart from the adsorption of organics or hydrogel formation, another possible explanation for POM in secondary filters is the retention of bacteria that had passed the top filter and then are trapped with a certain probability in a second filter. A simple conceptual model would consider three processes in a depth filter like GFF, which is really a maze with a range of different 345 channel sizes: 1. Particles are retained because they are too big for the dominant pore size. 2. Smaller particles pass through the bigger channels of the filter maze with a certain probability depending on the maze architecture. 3. These smaller particles get trapped in the smaller channels of the lower filter. This conceptual model is similar to the functioning of size exclusion chromatography except that in the case of GFF filters the maze is made of glass fibers and in SEC by gels. It might be possible that gels trapped in the filters help to retain some of the small particles, for example bacteria.         bacterial abundance in the original sample, i = 2 indicates the relative abundance after passage through double GFF filters or two separate GFF filtrations. The left ordinate: The numerical approximations for the two sets of data (Eq. 1 and 2). On the right ordinate the relative loss in bacterial abundance ((Bi-1-Bi) Fi-1 B0-1) was calculated from Eq. 1 and 2 and graphed against Fi -0.5. 730 Figure 9. Measured POC versus the POC calculated to represent the bacterial biomass retained by the GFF filter. Bacterial POC was calculated from bacterial abundance of the sample and a retention efficiency of 50% per GFF passage.