Changes in phytoplankton and biomineral content of particles during episodic fluxes to abyssal depth

Large episodic pulses of particulate organic carbon (POC) at the deep‐sea (~ 4000 m) time‐series Sta. M in the Northeast Pacific Ocean have increased in frequency and magnitude over the past 32 years. We inferred the ecological drivers of these events by quantifying the phytoplankton and biomineral composition within particles collected by bottom‐moored sediment traps immediately before, during, and after 14 high‐flux events. Samples collected during high‐flux events contained a significantly different phytoplankton community from other sampling periods. These particles contained relatively fewer intact phytoplankton cells and a sustained contribution from fragmented diatom frustules from species typical in coastal blooms. Biomineral fluxes did not appear to be driving high‐flux events. We suggest that most of the observed high‐flux events were generated by offshore transport of coastal diatom blooms, but that these particles were also highly transformed by deep‐sea pelagic food webs before reaching bathypelagic depths.

The rain of particles to the seafloor sustains deep ocean ecosystems and plays a key role in the global carbon cycle by sequestering carbon away from the atmosphere for long time periods. Particulate organic carbon (POC) originates through primary production by phytoplankton in the surface ocean but only a small fraction is exported as sinking particles into the mesopelagic and bathypelagic (Martin et al. 1987). At these depths, zooplankton and microbial communities consume and respire sinking POC, transforming the contents of particles before they reach the seafloor. In some cases, POC can efficiently reach the seafloor if it bypasses these deep ocean food webs in large, rapidly settling particles such as phytoplankton aggregates (Alldredge and Gotschalk 1988;Kemp et al. 2000), specific types of zooplankton fecal pellets Huffard et al. 2020), or animal carcasses (Robison et al. 2005;Smith et al. 2014). Resolving the mechanisms that transport particles to the deep ocean is needed to explain variability in longterm carbon sequestration and to predict how fluxes will change in the future (Siegel et al. 2014).
A long-term increase in POC flux to the seafloor has been observed at Sta. M, a deep-sea time series (4000 m deep) in the California Current (34 50 0 N, 123 00 0 W) (Smith and Druffel 1998;Smith et al., 2017). This increase observed over 32 years is driven by large episodic pulses of particles that account for 25-63% of the total carbon flux to the seafloor annually (Smith et al. 2018). Although a few of these events have been examined in detail Preston et al. 2020), the mechanisms generating most events are still unknown.
Identifying the cause of changing POC flux at 4000 m is difficult because the driving processes occur over large time and space scales, beginning in the surface ocean. Export modeled from satellite observations of chlorophyll in surface waters within a 100 km radius overlying Sta. M correlates with typical background POC fluxes measured in deep sediment traps (Smith et al. 2006(Smith et al. , 2018. However, this model does not capture the large episodic pulses that are driving long-term changes. Particles originating in close proximity to Sta. M could be captured in the sediment traps if they sank between 100 and 1000 m d À1 , taking only a few days or weeks to reach the seafloor ). According to a particle tracking model, slowly sinking particles (e.g., 50 m d À1 ) collected in the deep sediment traps originate up to 1000 km away, spanning the entire region southeast of Sta. M to the northern Baja Peninsula, and could take up to 3 months to reach bathypelagic depths . Mesopelagic and bathypelagic food webs would play a larger role in transforming slowly sinking particles on their way to the deep ocean. If high-flux events are generated by these slowly sinking particles, it would be difficult to link them to any specific feature in the surface ocean.
Clues about the origin of episodic high-flux events can be found in the composition of particles collected in sediment traps. On several occasions, salp fecal pellets have dominated seasonal increases in POC flux ) and helped drive an episodic high-flux event . These large pellets are capable of sinking fast enough to originate from local waters and would transfer relatively unprocessed surface-derived carbon to the bathypelagic (Phillips et al. 2009). Most of the time, the relative flux of fecal pellets actually decreases with increasing total POC flux, which is instead largely composed of amorphous detritus (Wilson et al. 2013). This detritus includes marine snow and phytodetrital aggregates, which typically sink at speeds less than 100 m d À1 (Alldredge and Gotschalk 1988;Trull et al. 2007;Baker et al. 2017) though notably can sink more rapidly if ballasted with biominerals (Lombard et al. 2013).
In this context, we combine analysis of particle chemical composition and phytoplankton community contents to identify ecological processes involved in generating high POC flux events at Sta. M. We specifically assess whether high-flux events at Sta. M were most likely generated by rapidly sinking particles from surface water overlying Sta. M or from more slowly settling particles processed through zooplankton and microbial food webs. We use our findings to explore possible mechanisms of increasing carbon export at this location over the last 32 years.

Sample collection and selection
Samples were collected off coastal California at Sta. M (34 50 0 N, 123 00 0 W, 4000 m depth) in McLane Parflux Mark 78H2 ) sediment trap deployed 600 m above the sea floor (Smith et al. 2018). The sediment trap collected sinking particles (0.5 m 2 funnel collection area) into 21 sequential cups pre-filled with 3-5% buffered formalin that rotated every 10, 17, or 30 d starting in 1989 until present day .
We analyzed samples collected during periods of high POC flux (Smith et al. 2018), defined as fluxes greater than 2 standard deviations above the average POC flux of the time series. High-flux samples were compared to samples collected immediately before and after each event (for brevity referred to as "low-flux" periods, though these fluxes were within the seasonal average value) ( Table 1). Samples collected during two high-flux events (events 8 and 9, Table 1) were unavailable and analyses were not performed on those samples for this study. All data are deposited in the National Centers for Environmental Information database (Michaud et al. 2021).

Chemical flux measurements
Particulate inorganic carbon (PIC), POC, and mass flux were previously measured in each of the samples as part of the Sta. M time-series program (Baldwin et al. 1998;Smith et al. 2018). To determine biogenic silica (BSi) contents of particles, triplicate 100 μL subsamples were measured using the colorimetric protocol of Strickland and Parsons (1972 Cell fluxes Changes in phytoplankton composition and degradation within sinking particles were determined by microscopy. Phytoplankton contents were quantified on an inverted Nikon Eclipse Ts2 microscope using a Sedgewick rafter (Fig. 1). A minimum of 400 cells were counted and then counting continued until a nonparametric estimator of asymptotic species richness (Chao et al. 2009) predicted that at least 80% of the total diversity had been observed. If that metric was never achieved, a maximum of 1000 cells were counted. Cell number fluxes were calculated by multiplying cell concentrations in the sample by the total volume of the sample and divided by the collection time and trap collection area. Cells still embedded within intact aggregates and fecal pellets were not counted by this method. However, we expect most cells were disrupted from within larger particles when collected in the trap cup, when homogenized in the lab, undergoing a freeze and thaw cycle during sample storage, and when aspirating with the pipettor prior to microscopy. Unfortunately, due to a small volume counted larger organisms including Rhizaria, which contain larger quantities of silica, were unable to be quantified.

Cell classification
Cells were classified in major functional types including diatoms, coccolithophores, dinoflagellates, and nanoflagellates (Fig. 1). The nanoflagellate and dinoflagellate categories likely represent a mixture of both phototrophic and heterotrophic taxa. Cells were often fragmented or missing defining features and we further classified cells to the most specific taxonomic category possible (Supporting Information Table S1): species, genus, or morphotype (Tomas et al. 1997), which we refer to as cell type. Cell types were further grouped into broader categories if fewer than five cells within that type were observed, including a single category for rare cells within each phytoplankton functional type. Fragmented and empty diatom cells that did not contain a chloroplast were included in counts if they could be placed in a cell type category but were recorded distinctly from those cells that contained chlorophyll. Only fragments that included at least one valve were counted (i.e., spine fragments were not counted). Functional groups other than diatoms appeared to be intact.

Analysis of particle community composition
All analyses were performed using tools available in python. The community composition within sinking particles was compared among samples using multidimensional scaling (sklearn.manifold.MDS) of Bray-Curtis dissimilarities (sklearn. neighbors.DistanceMetric) and significant differences assessed by a permutational multivariate ANOVA (PERMANOVA, pvalue threshold < 0.001; skbio.stats.distance.permanova). Only intact cells were included in the whole community comparisons among samples. A separate analysis was used to assess changes in the intact and fragmented diatom communities. SIMPER (similarity percentage) analysis was performed, with code adapted in python, to identify cell types driving dissimilarities among samples. Changes in the contribution of phytoplankton to flux were assessed by cell number flux in reference to measured POC and biomineral fluxes. Differences in the contribution of major phytoplankton function groups (intact diatoms, fragmented diatoms, nanoflagellates, and coccolithophores) during high-flux periods compared to lowflux periods (i.e., before and after high-flux periods, Table 1) were assessed using a Mann-Whitney test.

Particle chemical composition during high-flux events
Carbon export at Sta. M has increased over time as a result of increasing frequency and magnitude of episodic high-flux events (Smith et al. 2018;Fig. 2a). POC composed 7.77% AE 3.7% of the total mass flux of sinking particles on average across the entire time series (Fig. 2b). Only two of the 13 analyzed high-flux events (Table 1:  Relative mineral concentration within sinking particles did not significantly increase with increasing POC flux magnitude (Fig. 3). POC flux was positively correlated with opal flux (slope 1.59, p < 0.001, SE = 0.292, R 2 = 0.375), but with large variation during high-flux events (Fig. 3a). POC and opal flux were more strongly correlated during low-flux periods (slope 2.23, p < 0.001, SE = 0.548, R 2 = 0.440). Although the median POC : opal fluxes were often higher during high-flux events, they were not significantly different from samples collected during low-flux periods (Mann-Whitney p = 0.2; Fig. 3c). A weak correlation was detected between POC and CaCO 3 flux (slope = 0.146, p-value = 0.0051, SE = 0.050, R 2 = 0.142) (Fig. 3b). POC : CaCO 3 fluxes were slightly lower during high-flux events compared to low-flux periods, but not significant based on our threshold (Mann-Whitney p = 0.01; Fig. 3d).

Phytoplankton composition within particles during high-flux events
Phytoplankton cell number fluxes were higher during high POC flux events compared to before or after the event in 8 out of the 11 analyzed events (Fig.2c, Supporting Information Table S1). In most samples, phytoplankton fluxes were numerically dominated by diatoms followed by nanoflagellates and coccolithophores (Fig. 2d). Dinoflagellates, silicoflagellates, and heterotrophic protists (e.g., rhizaria, ciliates) were rarely detected (< 10 cells observed per sample or not at all) and not included in this study.
The relative number flux (cell flux : POC flux) of intact diatom, coccolithophore, and nanoflagellate cells decreased during high-flux events (n = 38 samples) compared to low-flux periods (n = 21 samples) by 3.08 AE 0.06, 2.40 AE 0.15, and 2.86 AE 0.20 fold, respectively (Mann-Whitney p < 0.001) (Fig. 4). Unlike intact cells, the relative number flux of fragmented and/or empty diatom frustules within particles was not significantly different between high-flux events and lowflux periods (p-value = 0.02) (Fig. 4D). An increase in the relative number flux of cells was never detected during any of the 11 high-flux events for any phytoplankton functional group.
The taxonomic composition of intact phytoplankton within sinking particles changed during high-flux events compared to low-flux periods (p < 0.001, PERMANOVA, Fig. 5). Changes in the flux of Emiliania-like and Pseudo-nitzschia cells contributed the most to this dissimilarity (46% and 14%, respectively), with lower relative flux of both taxa in particles during high-flux events (2.4-fold lower for Emiliania-like and 3-fold lower for Pseudo-nitzschia).

Diatom fragmentation during high-flux events
Although the number flux of intact diatoms relative to POC flux decreased during high-flux events, the composition of these intact diatom cells did not differ between high-and low-flux periods (PERMANOVA, p > 0.001). By contrast, number flux of fragmented diatoms relative to POC flux did not change during high-flux events compared to low-flux periods, but the composition of these fragments was different (p < 0.001, PERMANOVA). Fragmented Skeletonema and Chaetoceros cells contributed the most to this dissimilarity (36% and 10%, respectively) and their relative number flux did not differ between high-flux events and low-flux periods.

Proposed ecological mechanisms of high-flux events
Microscopic and chemical analyses of sinking particles indicated that distinct ecological mechanisms were involved in generating high-flux events at Sta. M. The relatively consistent bulk POC : mass flux during most of the time series (Fig. 2B), including during 11 out of 13 analyzed high-flux events, indicated that the particles underwent relatively similar levels of processing through epipelagic, mesopelagic, and bathypelagic food webs (see exceptions during two events described in detail below). Particles sinking during high-flux events usually contained relatively fewer intact phytoplankton cells, especially those typically associated with open-ocean and iron-limited environments (Emiliania-like and Pseudonitzschia) (Venrick et al. 2015). In contrast to intact diatom cells, the composition of fragmented diatom frustules changed during high-flux events due to the presence of genera typical in coastal blooms (Skeletonema and Chaetoceros) (Venrick et al. 2015). Because opal content was relatively low and biominerals were not enriched during high-flux events, we suggest that ballasting did not drive high-flux events. Instead, we hypothesize that most high-flux events were driven by the offshore transport of coastal diatom blooms that slowly settled en masse through the water column, interacting with mesopelagic and bathypelagic ecosystems before reaching the seafloor.
High-flux events may be increasing over time due to changes in the physical dynamics of the California Current. Jets and eddies in the California Current transport and fuel coastal diatom blooms offshore beyond the shelf break (Chavez 1991;Ohman et al. 2012;Krause et al. 2015). Frontal regions between these coastal and offshore water masses also enhance POC flux from the surface into the mesopelagic due to subduction and increasing aggregation and zooplankton grazing Barth 2002;Ohman et al. 2012;Stukel et al. 2016). We suggest that the enhanced POC export generated by coastal diatom blooms at offshore fronts continue to transport large quantities of POC to the bathypelagic and seafloor, but these particles sink relatively slowly and are highly processed by the time they reach the deep ocean. Subducted water masses are advected vertically at relatively slow rates (up to 25 m d À1 ) compared to horizontal advection (up to 29 km d À1 ) Washburn et al. 1991), and would transport surface phytoplankton blooms and particles into the mesopelagic far from the site of surface production . Slow transit through the mesopelagic by this mechanism would fuel mesopelagic food webs and particle transformation.
Increasing carbon export to Sta. M may also be influenced by long-term increases in coastal phytoplankton production, in addition to changes in offshore transport and dynamics at ocean fronts. The strength and duration of upwelling in the California Current has increased due to climatological changes in atmospheric pressure gradients (García-Reyes and Largier 2010), increasing the supply of nutrients available to enrich phytoplankton blooms. A corresponding increase in chlorophyll concentrations has also been observed by ocean color satellites over a 15-year period (Kahru et al. 2012).
Two of the 11 analyzed high-flux events appear to have been generated by a different ecological mechanism, based on the anomalously high POC : mass flux of these particles. We suggest that event 5 (15 June 2011-24 August 2011) and event 11 (18 June 2016-5 July 2016) may have been generated by rapidly sinking particles (100-1000 m d À1 ) such as salp fecal pellets. Huffard et al. (2020) observed a massive flux of salp fecal pellets in late spring 2015, when POC : mass flux was also elevated (Fig. 2b). Though this period was not classified as "high POC flux" by our statistical threshold, it does suggest that high POC : mass may be a chemical signature of salp pellet fluxes. Events 5 and 11 also corresponded with especially high fluxes of intact diatoms (Fig. 2d), which may have contributed to the increased POC : mass. Diatoms may have a greater chance of remaining intact within salp pellets due to the relatively gentle packaging by filter feeders and their reduced time transiting through mesopelagic food webs. However, high fluxes of intact diatom cells were not unique to events with high POC : mass (e.g., events 1 and 7 had low POC : mass fluxes) and are probably not by themselves a good marker for fluxes driven by salp fecal pellets. Boxplots represent the quartiles of the data distribution. The orange line represents the median of the data. The whiskers represent the spread of the data and the dots represent the outliers in the data.

Biominerals and ballasting
The California Current is a diatom-rich, intermittently iron-limited ecosystem with silicic acid readily available (Hutchins et al. 1998;Brzezinski et al. 2003Brzezinski et al. , 2015. Consequently, we expected to detect relatively high opal fluxes leading to an increase in POC flux through biomineral ballasting. The ballast hypothesis suggests that particles containing biominerals such as opal and CaCO 3 will sink more quickly and export more carbon to the deep ocean (Klaas and Archer 2002). Similar to Klaas and Archer (2002), we identified a correlation between POC flux and these biominerals at Sta. M. The slope of POC flux to CaCO 3 flux was similar between our study and their study (0.146 vs. 0.126, respectfully). By contrast, the slope of POC flux to opal flux was much steeper in the Sta. M trap samples (1.59) compared to their globally distributed average (0.061), suggesting relatively high silica remineralization and dissolution occurred before reaching the bathypelagic collection depth. Though previous studies show low BSi dissolution in this region during diatom blooms (Brzezinski et al. 2003), slowly sinking particles may undergo high dissolution and cell fragmentation as they sink through the mesopelagic and bathypelagic. Our data may also be impacted by dissolution of BSi within the collection cups, though even at the high end of previously reported rates (30%, Klaas and Archer 2002), dissolution in the collection cups still would not account for the 26-fold difference in slopes between our two studies. In spite of the apparently high BSi dissolution in the deep-water column, the remaining cellular and mineral contents within the samples still provided evidence for differences between high-flux events and low-flux periods. Particles sinking during high-flux events were not enriched in biominerals, suggesting that ballasting was not responsible for high-flux events by the time they sank to the bathypelagic. These results are similar to those found at an abyssopelagic time-series location in the North Atlantic, where high POC fluxes often occurred when mineral fluxes were at their lowest, and with no enrichment in opal fluxes (Lampitt et al. 2009). These results lend support of the hypothesis proposed by Passow and De La Rocha (2006), that total mass flux determines the magnitude of mineral flux, and not vice versa.
This study provides a mechanistic example of how longterm, climate-driven changes in an eastern boundary current can affect carbon export. The data presented here suggests that changes occurring in this ecosystem are generating deep carbon export due to the offshore transport of slow settling of detritus originating from coastal diatom blooms. Although the total magnitude of export to the deep ocean is increasing, the likely mechanisms of export suggest that much of this biomass is still lost to remineralization and/or transferred through deep ocean food webs on its way to the seafloor.