Distinct habitat and biogeochemical properties of low‐oxygen‐adapted tropical oceanic phytoplankton

We use data collected by Biogeochemical Argo (BGC‐Argo) float, over a 5‐year period (2016–2021), to study the dynamics of a unique low‐oxygen‐adapted phytoplanktonic community in the eastern tropical North Pacific. We isolate this community using a model that partitions vertical profiles of chlorophyll a (Chl a) and particulate backscattering into the contributions of three communities of phytoplankton: C1, the community in the mixed‐layer; C2, at the deep Chl a maximum; and C3, in low‐oxygen waters at the base of the euphotic zone. We find that C3 has a similar chl‐specific particulate backscattering to C2, both lower than C1. C2 and C3 contribute significantly to integrated stocks of Chl a, both at around 41%, and both around 30% of integrated particulate backscattering (after removing a background signal attributed to nonalgal particles). Found at depths of around 100 m, the peak biomass of C3 is lower than C2 (located at around 60 m), and yet, C3 makes similar contributions to integrated stocks, because it has a broader peak than C2. In relation to C1 and C2, C3 thrives in a lower temperature, higher density, lower light, lower oxygen, and higher saline habitat. This work illustrates how BGC‐Argo floats, in combination with simple conceptual models, can be used to observe the dynamics of unique communities of phytoplankton in extreme environments. The projected climate‐driven changes in oxygen minimum zones add urgency to understand the vulnerabilities of these communities both in terms of stocks and composition.

Phytoplankton are photosynthesizing, microscopic, unicellular algae that occupy the euphotic zone (Reynolds 2006).Extracting energy from light through photosynthesis and converting inorganic carbon (carbon dioxide) into organic carbon, phytoplankton contribute to around 50% of global net primary production (Longhurst et al. 1995;Field et al. 1998).They modify the total CO 2 concentration within the upper ocean, impacting the air-sea CO 2 flux that regulates Earth's climate (Falkowski 2012).Energy produced by phytoplanktonin the form of organic carbon-is made available to most of the marine ecosystem, sustaining mesopelagic and bathypelagic communities (Reynolds 2006;Martiny et al. 2022).Phytoplankton are considered an essential climate variable (GCOS 2011), imperative to the ocean's biological pump-one of the Earth's major carbon sinks (Basu and Mackey 2018;Brewin et al. 2021).Produced particulate organic carbon (POC) is consumed and processed by microbes, zooplankton, and subsequent trophic levels, and later respired back into inorganic carbon, or transferred into the deep ocean (Basu and Mackey 2018), a process referred to as the biological pump.The efficiency of the biological pump and the amount of organic carbon produced at the surface that is transferred to the deep ocean, is dependent on phytoplanktonic physiology, community structure and species dynamics, distribution, and vertical structure, as governed by environmental conditions (Basu and Mackey 2018).Distinct communities and species of phytoplankton also contribute differently to the cycling of biogeochemical elements other than carbon, such as the nitrogen, silica, iron, and phosphate cycles (Arrigo 2005;Le Quéré et al. 2005;IOCCG 2014), maintaining planetary homeostasis.
Elevated concentrations of phytoplankton are typically found in regions where nutrient concentrations are highincluding coastal regions and upwelling regions, and, during spring in temperate and high latitudes-while lower concentrations are often associated with the central oligotrophic gyres that are governed by nutrient limitation (Longhurst et al. 1995).Comprised of an estimated $ 25,000 species (Marañ on 2009), phytoplankton have adapted to thrive in a wide variety of environments, occupying a range of environmental niches.Examples include the high-light, shallow mixed layer of the tropical ocean, the very low-light, deep mixed layer of high latitudes during winter, and more extreme habitats such as under sea ice and in oxygen minimum zones (OMZs).Phytoplankton have been observed at the base of the euphotic zone in OMZs (Goericke et al. 2000;Cepeda-Morales et al. 2009;Whitmire et al. 2009), where light is limiting to phytoplanktonic growth.Found in regions of strong stratification, subsurface low-oxygen zones occur due to the decomposition of surface-derived sinking organic material by aerobic respiration (Ulloa et al. 2012;Long et al. 2019).OMZs are projected to expand by 2100 (International Panel on Climate Change, IPCC 2018).In some cases, these oxygen-depleted waters-which can be rich in nutrients-intrude into the base of the euphotic zone, establishing a habitat for phytoplanktonic growth (Cepeda-Morales et al. 2009;Fuchsman et al. 2019).
Previous studies (e.g., Goericke et al. 2000;Lavin et al. 2010) have outlined the physiological and genetic adaptivity of phytoplanktonic cells that can grow and thrive in these extreme low-light and low-oxygen environments.They have shown that the community found in these environments is composed of the cyanobacteria Prochlorococcus and Synechococcus (Goericke et al. 2000;Lavin et al. 2010).The dominant abundance of Prochlorococcus at the bottom of the euphotic zone is attributed to its physiological plasticity and genetic diversity.Prochlorococcus-composed of at least 12 distinct lineages-can be divided into high-and low-light adapted ecotypes (Bouman et al. 2006;Biller et al. 2015).Goericke et al. (2000) and Lavin et al. (2010) have shown some low-light lineages to have a competitive advantage in low-oxygen environments below the euphotic depth.Dominance of Prochlorococcus in oxygendepleted waters has been attributed to species emergence in low-oxygen conditions in the ancient ocean, with lineages retaining phycobilisomes as light-harvesting antennae, contributing to their ability to sustain aerobic metabolisms in oxygen-depleted waters (Ulloa et al. 2021).Synechococcus are also reported to reside in the low-light, low-oxygen conditions (Goericke et al. 2000;Lavin et al. 2010) along with Prochlorococcus.However, their concentrations appear consistently lower than those of Prochlorococcus within the OMZ (Goericke et al. 2000;Lavin et al. 2010).
The niche that the low-oxygen, low-light phytoplankton occupy makes it difficult to observe these phytoplankton.Satellite remote sensing of ocean color is recognized as an important tool to monitoring phytoplankton (Moisan et al. 2017;Sathyendranath et al. 2019;Brewin et al. 2022).The capability to monitor phytoplankton abundance on global and synoptic scales allows for an assessment of the entire global surface phytoplankton distribution and abundance (Gordon et al. 1980;Groom et al. 2019).Although effective at representing the ocean surface (< 30 m), satellites are unable to observe phytoplankton abundance within stratified regions at depths of 50-200 m (Brewin et al. 2022).Monitoring phytoplankton communities below the surface layer in these seasonally and permanently stratified regions (that constitute $ 70% of the global ocean) requires in-situ measurements.Although important for studying phytoplankton at deeper depths of the water column, traditional ship-based monitoring cannot provide the high spatial and temporal coverage needed to improve our understanding phytoplankton dynamics and distributions at depth.This is now possible, however, using ocean robotic platforms, such as autonomous profiling floats (Chai et al. 2020;Claustre et al. 2020).
Autonomous profiling floats, including Biogeochemical Argo floats (BGC-Argo floats), are designed to explore the biogeochemical and physical dynamics of the ocean up to depths of 2000 m, collecting data at higher spatial and temporal resolutions than conventical platforms (Ravichandran et al. 2012;Bittig et al. 2019).Alongside collecting environmental data, BGC-Argo floats have optical sensors measuring chlorophyll a (Chl a) pigment concentration and particulate backscattering (b bp ), both of which have been used as measures of phytoplanktonic biomass (Martiny et al. 2022).Brewin et al. (2022) used a conceptual model to partition vertical profiles of Chl a and b bp , collected using a BGC-Argo float in the northern Red Sea, into the contributions of two phytoplanktonic communities (C 1 in the surface mixed layer, and C 2 at the deep chlorophyll maximum), to examine their seasonal dynamics.
Here, the two-community model of Brewin et al. (2022) is extended by adding a third community that thrives in lowoxygen, low-light waters at the base of the euphotic zone.By applying the model to data collected by a BGC-Argo float in the eastern tropical North Pacific (ETNP) over a 5-year period (2016-2021), we explore the environmental niche and dynamics of this unique third phytoplanktonic community.Specifically, this work has the following aims: (1) Quantify the vertical structure of the third phytoplanktonic community, with respect to other communities in the water column; (2) Use a combination of Chl a and b bp data to characterize the third phytoplanktonic community; (3) Assess the contribution of the low-oxygen-adapted community to the stocks of phytoplankton (inferred from Chl a and b bp ) in the water column, and compare these to other phytoplankton communities within the water column; and (4) Quantify the environmental conditions in which the low-oxygen-adapted community thrives, and how these environmental conditions contrast to those where other communities are found.

Study area
Located between 9.8 N-14.5 N, 118 W-108 W, the study area is situated in the eastern tropical North Pacific Ocean (ETNP, Fig. 1).This region is characterized by reduced seasonal variation, sluggish circulation, and weak internal mixing (Fiedler and Talley 2006).The study area is influenced by a strong interannual climate variability, with the El Niño-Southern Oscillation (ENSO) dominating this interannual variability (Timmermann and Jin 2002).The amalgamation of high surface phytoplanktonic productivity, sinking organic matter, strong stratification, ENSO, the Pacific Decadal Oscillation, and the resultant reduced mixing and sluggish circulation, generates extensive subsurface oxygen depletion (Tiano et al. 2014;Duteil et al. 2018).Home to Earth's largest OMZ, the imbalance between supply and demand of oxygen in the OMZ below depths of 100 m has led to the development of an environment that supports the presence of unique planktonic communities (Cepeda-Morales et al. 2009;Stramma et al. 2010;Lam and Kuypers 2011;Fenchel et al. 2012;Long et al. 2019).

Biogeochemical Argo float (WMO 3901531)
Data in this study were collected using a PROVOR BGC-Argo float (Argo 2000) in the ETNP along the geographical locations displayed in Fig. 1 (World Meteorological Organization [WMO] number 3901531), downloaded from the Ifremer Argo data assembly centre (ftp://ftp.ifremer.fr/ifremer/argo/dac/bodc/;Sprof file).Equipped with conductivity-temperature-depth (CTD) sensors, dissolved oxygen (DO) optode, downwelling irradiance radiometer (at wavelengths of 380, 412, and 490 nm) alongside a photosynthetically active radiation (PAR) radiometer, chlorophyll-a fluorometer (Chl a), backscattering meter (b bp ), and an iridium transmission system (GPS), the BGC-Argo provides data from which profiles of temperature ( C), salinity (PSU), DO (μmol kg À1 ), Chl a (mg m À3 ), b bp (m À1 ), and PAR (mol quanta m À2 d À1 ) were measured.Pressure, temperature, PAR, and salinity data were delayed-mode quality-controlled.For each profile, density (kg m À3 ) and the Brunt-Väisälä buoyancy frequency (N 2 in units of s À2 , where N is computed as radians per second), were computed using the Python seawater package (version 3.3).The mixed layer depth Z m ð ) was computed using Holte and Talley (2009) method for each profile, using the temperature algorithm (Python package https:// github.com/garrettdreyfus/python-holteandtalley).For Chl a (Adjusted mode), DO and b bp profiles (both real-time mode) additional processing steps (as described by Schmechtig et al. 2018Schmechtig et al. , 2019) ) were employed to ensure maximum data quality.The adjusted mode Chl a data were processed as follows: the fluorescence measurements were first corrected by subtracting a dark fluorescence value; the dark-corrected fluorescence values were then multiplied by the manufacturers scaling factor to convert to a Chl a concentration and divided by two following recommendations by Roesler et al. (2017) who found a systematic overestimate of Chl a by a factor of two when testing WET Labs ECO sensors; the range of Chl a in the profile was then checked for realistic values (should be between À0.1 and 50 mg m À3 ) and the profile was tested for negative Chl a spikes; and finally the profile was checked for nonphotochemical quenching (NPQ) on Chl a and if required, corrected based on a method adapted from Xing et al. (2012).Schmechtig et al. (2018) provides further details on the Chl a processing.Instantaneous measurements of PAR were converted into daily integrals following Brewin et al. (2022), accounting for day length as a function of the time of year and latitude, and under the assumption of a sinusoidal light cycle (Brewin et al. 2020).The diffuse attenuation coefficient (K d ), representing the attenuation of PAR with depth, was computed for each profile by fitting a Beer-Lambert law to the PAR data in the top 200 m of the water column (Brewin et al. 2022).Profiles where K d could not be computed were removed.

Three-community model development
For this study, the two-community model of Brewin et al. (2022) was extended into a three-community model.First, and for a Chl a given profile, the dimensionless optical depth (τ) was computed by multiplying the geometric depth (m) by K d (m À1 ).Following Brewin et al. (2022), the Chl a profile was then converted into a normalized profile (B*), in which the Chl a concentration (B) at depth (z) was divided by the surface Chl a concentration (B s ), taken as the median Chl a in the first optical depth.This was conducted after the removal of profiles without surface Chl a, profiles with less than 11 measurments, or with surface Chl a values < 0.01 mg m À3 .The normalized Chl a concentration (B Ã ) at a particular optical depth (τ) was then considered as a combination of contributions from three communities of phytoplankton.
Community 1 represents as a group of phytoplanktonic species that principally dwell in the turbulent mixed layer in the surface ocean, and following Brewin et al. (2022), was expressed as, where, P 1 controls the slope of change in B Ã 1 with τ, and τ 1 represents the midpoint of the slope (Brewin et al. 2022).It was assumed that as τ tends to zero, community 1 would dominate the surface Chl a concentration (> 99%).
Community 2 represents the group of phytoplanktonic species that resides in a stable environment below the mixed layer, characteristic of the phytoplanktonic community at the first deep Chl a maximum (DCM1).Following Brewin et al. (2022), B Ã 2 was modeled as a Gaussian function of τ, such that, where B Ã m,2 defines the maximum value of the DCM1 (B Ã 2 ), the dimensionless depth where B Ã m,2 occurs is τ 2 , and the width of the B Ã m,2 peak is defined by σ 2 .Community 3 represents the phytoplanktonic group beneath community 2, living in very low-light and lowoxygen conditions (see Goericke et al. 2000;M arquez-Artavia et al. 2019).Similarly, B Ã 3 is modeled as a Gaussian function of τ, such that, where B Ã m,3 represents the maximum value of the second DCM (B Ã 3 ); the dimensionless depth where B Ã m,3 occurs is τ 3 , and the width of the B Ã m,3 peak is defined by σ 3 .Communities 1, 2, and 3 (Eqs.1-3) can be combined to derive the total normalized chlorophyll profile (B Ã ), expressed as, The total Chl a concentration (B) at depth (z) can be reconstructed by scaling to B s and K d , with knowledge of the eight parameters: ,τ 3 , and σ 3 : Owing to strong stratification in the region, the model parameters were independently derived for each profile, as follows.First, parameters P 1 and τ 1 , which are physical and optical in nature, were estimated empirically from Z m and K d following Brewin et al. (2022), where τ 1 = 0.62 Z m K d + 2.29 and P 1 ¼ 10 0:08τ1þ0:06 .Once τ 1 and P 1 are known (and fixed), the other parameters B Ã m,2 , τ 2 , B Ã m,3 , τ 3 , and σ 3 are estimated by fitting the chlorophyll model to the data.The optimization function used was Levenberg-Marquardt iterative minimization (Python function minimize).Initial guesses of parameters were as follows: σ 2 and σ 3 were set to one, while B Ã m,2 and B Ã m,3 were initialised by extracting the maximum B Ã above and below 5 μmol kg À1 DO depth, and τ 2 and τ 3 were initialised at the depths of these respective maximums.Once model parameters were obtained, Chl a profiles were reconstructed by scaling the normalized profiles by B s and K d .
Following Brewin et al. (2022), the model was further extended to reproduce profiles of particle backscattering at 700 nm (b bp ), in the top 300 m of the water column.To do this, first b bp is normalized by the surface particle backscattering (b bp,s ) defined as the median b bp within the first optical depth.The normalized particle backscattering (b Ã bp ) at a given optical depth (τ) is then considered as a combination of the three phytoplanktonic communities and a background component, attributed primarily to nonalgal particles, such that, where scaling factors, ω 1 , ω 2, and ω 3 , link the three phytoplanktonic communities and b Ã bp,k is a constant background factor (Brewin et al. 2012).Assuming that ω 1 can be expressed as 1 À b Ã bp,k as in Brewin et al. (2022), the backscattering model can be further simplified, such that, ω 3 , and b Ã bp,k required for the backscattering model were computed by fitting Eq. 6 to b bp profiles using the Levenberg-Marquardt method (python function minimize).Initial guesses for ω 2 , ω 3 , and b Ã bp,k were set to 0.3, 0.3, and 0.8, respectively, and constrained to a lower limit of 0. bp,k by surface particle backscattering (b bp,s ).Once model parameters were known, the particulate backscattering profiles were also reconstructed by rescaling the normalized profiles by surface particle backscattering (b bp,s ) and K d .Backscattering profiles were apparently compromised (owing possibly to debris attached to the sensor) in mid 2018, returning to realistic values by late 2018.The compromised data were excluded from the analysis.An example of a model fit to a profile from the float is displayed in Fig. 2. The trapezoid method was used to carry out depth integration in the upper 200 m of the water column.

Extraction of environmental data for the three communities
Environmental conditions where the three communities were present (temperature, salinity, density, Brunt-Väisälä, DO, and PAR) were extracted for each BGC-Argo profile.For community 1 (C 1 ), median values in the mixed layer were computed.Meanwhile, for community 2 (C 2 ) and 3 (C 3 ), environmental data at τ 2 and τ 3 were used, respectively.Data from all profiles were combined and used to evaluate the habitats that the communities reside in.

Performance and fit of model
The three-community Chl a model is seen to successfully capture the dynamics and trends of the BGC-Argo float data (Fig. 3a,b).Differences between in situ and modeled data are close to zero across large areas of the water column (median difference = À0.003mg m À3 , standard deviation = 0.03 mg m À3 ; Fig. 3c, see Supporting Information Fig. S1 for percentage deviations between model and data), in some cases higher, for instances where the model fails to represent the complexity of the Chl a profile.The contributions of the three communities (C 1 community 1; C 2 community 2; and C 3 community 3) to the Chl a profiles are displayed in Fig. 3d-f, with C 1 dominating the mixed layer and having higher concentrations during the winter (November to April).The presence of C 2 is fairly constant over the annual cycle, with the depth of the peak biomass found at around 60 m, fluctuating between 40 and 100 m, being shallower in the winter months due to lower surface light availability (Fig. 3e, Supporting Information Figs.S2, S3).In contrast to C 2 , C 3 thrives below the 5 μmol kg À1 DO boundary, having a peak biomass depth at around 100 m, with some fluctuations between 50 and 200 m, deepening during the spring and summer.Additionally, the three-community model is found to successfully replicate the bulk backscattering fields (Fig. 4), but with a slight tendency to overestimate around the two DCMs, and slightly underestimate above and below the depth ranges, with a median difference of 0.00007 m À1 (standard deviation 0.00012 m À1 , see Supporting Information Fig. S1 for percentage deviations between model and data).

Parameters of communities 2 and 3
Considering the identical mathematical expressions for C 2 and C 3 (Eqs.2 and 3), the parameters of the two communities can be cross compared (Fig. 5).We see that C 2 has a median peak value of 0.49 mg m À3 (B Ã 2,m Â B S ), a median width of peak of 13.2 m (σ 2 =K d ) and a depth of peak of 60.8 m (t 2 =K d ), whereas C 3 has a deeper median depth of peak of 106.4 m (t 3 =K d ), and a lower median peak value of 0.30 mg m À3 (B Ã 3,m Â B S ) when compared with C 2 , but a greater median peak width of 24.7 (σ 3 =K d ).The parameters of the two communities remain different over the course of the year but show some seasonal variability (see Supporting Information Fig. S3).For example, with maximum peak values, and maximum depth of peaks, for both communities (C 2 and C 3 ) in spring, and minimums in winter (Supporting Information Fig. S3).

Contributions to community stocks
The chlorophyll-specific particulate backscattering coefficient is highest for C 1 , with a median value of 0.006 m 2 [mg B] À1 .Both C 2 and C 3 have very similar chlorophyll-specific particulate backscattering coefficients, with C 3 having a slightly higher value than C 2 , of 0.0010 m 2 [mg B] À1 (standard deviation = 0.0006) and 0.0009 m 2 [mg B] À1 (standard deviation = 0.0007) respectively (Fig. 6a).The background backscattering coefficient (b k bp ) has a median value of 0.0004 m À1 .Background backscattering dominates b bp contributing 61% to the integrated particulate backscattering (Fig. 6b).The populations C 2 and C 3 contribute similarly to integrated particulate backscattering, with C 3 contributing 0.7% more than C 2 (10.0% and 10.7%, respectively), though the difference is not statistically significant.When removing the background backscattering coefficient from integrated particulate backscattering, C 1 remains the highest contributor to particulate backscattering with a median of 41%.The median contribution of C 3 is slightly greater than C 2 (27% and 26%, respectively).However, the interquartile range for C 2 overlaps with C 3 (Fig. 6c) indicating no significance in differences between these two communities.
In contrast to the high contribution of C 1 to particulate backscattering, the median value of the contribution of C 1 to integrated Chl a is low, at around 11%, though there are occasional profiles in which C 1 dominates, contributing up to $ 99% integrated Chl a (see outliers in Fig. 6d).Like the patterns identified in Fig. 6a-c, the contributions to integrated Chl a by C 2 and C 3 are statistically similar, with median values of 44% and 41%, respectively.

Temperature
Thriving at differing depths, the thermal niche of each community drastically differs (Fig. 7a).Heat uptake by seawater is highest at the surface due to the absorption of solar radiation (Whitmarsh et al., 2015).Therefore, as expected, C 1 resides in waters with the highest temperature (median 28.2 C).As temperature decreases with depth, C 2 and C 3 are associated with lower temperatures, with C 2 located above the thermocline (see Supporting Information Fig. S2), thriving in slightly cooler waters than C 1 (median

Salinity
As with the thermal conditions surrounding C 1 , C 2 , and C 3 , the salinity of each community is also drastically different (Fig. 7b).The low salinity (median 33.61 PSU) of C 1 coincides with higher temperature outlined in Section 2.4.1.As temperature decreases with depth (Fig. 7a), salinity increases (see Fig. 7b).C 3 resides in waters with the highest salinity (median 34.74 PSU).

Density
Outlined by differing thermal and saline levels described in Sections 2.4.1 and 2.4.2, the density of the environments in which the communities thrive are clearly separable (Fig. 7c).The low density (median 1021.29 kg m À3 ), high temperature (median 28.2 C) and low saline (median 33.61 PSU) surface waters where C 1 resides, and the denser waters (median 1023.19 kg m À3 ) surrounding C 2 are in contrast to the highly dense (median 1025.90 kg m À3 ), cold (median 14.56 C), and saline (median 34.74 PSU) waters in which C 3 thrives.The strong stratification in the region, and the consequent sluggish circulation of this study site supports the formation of the vertical differences in community composition (Barnett et al., 2022).

Brunt-Väisälä
Although density increases with depth, the patterns in stability (Brunt-Väisälä) are more complex (Supporting Information Fig. S2).As expected, owing to atmosphere-ocean interactions, such as wind-generated mixing of the surface layer, C 1 resides in a turbulent mixed layer (median Brunt-Väisälä of 0.000012 s À2 ; Fig. 7d).Interestingly, despite C 3 thriving in the densest waters, it is C 2 that resides in waters with the greatest stability (median Brunt-Väisälä of 0.00074 s À2 ), being located at the top of the thermocline between depths of $ 20-100 m.C 3 is found in waters with lower stability than C 2 (median Brunt-Väisälä of 0.00017 s À2 ), because of its position below the thermocline and oxycline.

Dissolved oxygen
Atmosphere-ocean interaction and oxygen production through phytoplanktonic photosynthesis (having a greater oxygen production rate over respirational removal) places C 1 in waters with the highest DO in the water column, with a median 181.79 μmol kg À1 (Fig. 7e).Community respiration below the mixed layer where C 2 thrives, and lower net primary production (oxygen production), results in lower DO (median 96.63 μmol kg À1 ).The near permanent stratification of the water column and respiration exhausts the oxygen supply at the depths in which C 3 thrives, resulting in extreme oxygen depletion with a median of 0.55 μmol kg À1 (this value should be interpreted with caution given how close it is to the limit of detection of the oxygen sensor) suggesting physiological adaptions of C 3 to extremely low-oxygen conditions.

Photosynthetically active radiation
At the bottom of the euphotic zone, C 3 is characterized by very low light (median of 0.02 mol quanta m À2 d À1 , acknowledging this value should be interpreted with caution given how close it is to the limit of detection of the PAR sensor, $ 0.04% surface irradiance), resulting in a low-light adapted community (Ulloa et al. 2021).C 1 receives the highest level of light (median PAR of 7.77 mol quanta m À2 d À1 in the mixed layer, $20% surface irradiance), due to its position at the ocean surface (Fig. 7f).The C 2 population receives around 0.85 mol quanta m À2 d À1 of PAR ($ 1.2% surface irradiance).In other words, on average the C 3 population receives light fluxes that are more than three orders of magnitude less than what phytoplankton at the very surface receive.

Discussion
Model performance and extension to three communities The model fit Dynamics and trends of the float data are encapsulated by the model, which was successful in detecting the presence of three communities (C 1 , C 2 , and C 3 ) in the water column (between depths of 0-200 m), demonstrating reasonable fit to the data and model performance as indicated by low residual and relative errors (Figs. 3 and 4, Supporting Information Fig. S1).The patterns in the in-situ Chl a data (Fig. 3a) are reflected well in the modeled Chl a data (Fig. 3b)-similar to the b bp data in Fig. 4-with the model successfully reproducing the trends of the in-situ Chl a and b bp data (aim 1 of this study).However, while the model fit generally captures data trends and patterns, there is a slight tendency of the model to overestimate Chl a above the defined peaks of C 2 and C 3 (Fig. 3c).Meanwhile, there is also an occasional slight underestimation below the peaks.In a few cases, where profiles are more complex and the three-community model fails to represent the data well, larger biases between the model and the in-situ data were observed (Fig. 3c).
The three-component model is designed to partition the vertical profiles of Chl a and b bp into three communities of phytoplankton using Gaussian and Sigmoid functions (Brewin et al. 2022).Here, C 1 is defined by a sigmoid function, with its parameters being tied to the mixed-layer depth (Eq.1).Both C 2 and C 3 use Gaussian functions to describe the peak value of both DCMs, the depth of the DCMs peak, and the width of the two DCMs.The mathematical functions (gaussian and sigmoid) used in our model builds on a long history in biological oceanography (see Platt and Sathyendranath 1988;Morel and Berthon, 1989;Uitz et al. 2006;Mignot et al. 2011;Brewin et al. 2017).The model also took into consideration the law of parsimony, such that we strived to minimize the complexity of the model (only having eight parameters for the Chl a model) while maintaining model performance (with the model on average explaining $ 90% of the variability in a Chl a profile).Nonetheless, there were some cases where the model could be improved further.For example, at times we observed a slight asymmetry in the depth distribution of community 2 and 3 (sharper rise than decline as a function of depth, see Fig. 2).There may be scope to refine the mathematics used in the model, or investigate different weighting methods, to improve model fitting.
Following Brewin et al. (2022), our approach makes it feasible to reconcile vertical variations in two independent proxies of phytoplankton biomass, Chl a and b bp, by simulating changes in the chlorophyll-specific particulate backscattering of the total community with depth.However, we do assume that each community has a fixed chlorophyll-specific particulate backscatter, which may not always hold since each community can span a range of environmental conditions.Despite the model successfully capturing the dynamics of the data collected on the BGC-Argo float (WMO 3901531), caution should be taken when fitting the model to other floats and other oceanic regions (Brewin et al. 2022).As community 3 was present in all profiles, we did not consider cases where it was absent.This simplified the tuning of our model for this study.A more flexible parameterization would need to be developed to handle cases where the presence of community 3 is more sporadic, by unifying the two-community model of Brewin et al. (2022) with this three-community model.Additionally, model parameters may need to be revised for future studies, by refining the set of conditions used for model tuning in differing environments.Such consideration will need to be made in regions characterized by complex dynamics (Schneider et al. 2014;Brewin et al. 2022).

BGC-Argo float data
While there is great capacity for data collection using autonomous profiling BGC-Argo floats, there are challenges in ensuring that the data and conclusions drawn from analysis are of sufficient quality.Currently, autonomous profiling floats use Chl a fluorescence as a proxy of Chl a concentration (Barbieux et al. 2017).However, for the same Chl a concentration, the fluorescence yield of phytoplankton can vary depending on environment and growth conditions, and species composition (Ricour et al. 2021;Brewin et al. 2022;Martiny et al. 2022;Petit et al. 2022).Additionally, Chl a fluorescence is sensitive to NPQ (Xing et al. 2016).Although corrections to NPQ are part of standard BGC-Argo processing, additional work is needed to improve and validate them, as well as in the conversion of chlorophyll fluorescence into Chl a concentration.Poorly oxygenated waters can have high concentrations of dissolved organic matter (Loginova et al. 2020).The Chl a fluorescence signal can also be affected by the content and presence of colored dissolved organic matter, producing a deceptive increase in Chl a concentration with depth in some regions (Ricour et al. 2021).Some confidence in the data analyzed here can be gained from the general agreement in patterns of b bp data and Chl a fluorescence data (Figs. 3 and 4).Challenges are also faced when ensuring the quality of b bp data, frequently used as a proxy of the stock of POC (Barbieux et al. 2017).An example is seen in Fig. 4 showing data collected between July and August in 2018, that contained suspiciously high b bp data that had to be removed from the analysis.It is therefore critical that robust protocols be established to maximize data quality and quantify data uncertainties.Future developments in BGC-Argo float design might also lead to the incorporation of detectors for measuring phytoplankton absorption in the red spectrum that will aid the estimation of Chl a concentration, as a compliment to fluorescence-based sensors and methods (Dall'Olmo et al. 2012;Brewin et al. 2016).
The advantage of BGC-Argo floats lies in their ability to sample the interior global ocean in a temporally and spatially continuous way, which compliments satellite data by providing information on the vertical profile of the water column.This provides the opportunity, for the first time, to study phytoplankton populations in the first and second DCM in remote locations, in a systematic manner, while being a rich source of data that can be exploited appropriately and optimally.

Vertical structure of low-oxygen-adapted phytoplankton
Thriving in poorly oxygenated water, C 3 is positioned at a depth of around 100 m (Fig. 5b).In contrast to C 1 and C 2 , found between 0 and 50 m and at around 60 m respectively, there is a clear vertical partitioning of communities with depth (Figs. 3 and 4).Niche-partitioning in the vertical water column indicates differing growth and nutrient environments across communities (Scanlan et al. 2009).Such vertical partitioning is considered instrumental (Cepeda-Morales et al. 2009;Scanlan et al. 2009;Varkey et al. 2015;Thompson et al. 2021) in the development of differing Synechococcus and Prochlorococcus ecotypes and clades.In support of our findings, Goericke et al. (2000) identified a low-oxygen-adapted phytoplanktonic community-located 8 AE 2 N of this study-to thrive at 120 m.Although this depth is slightly deeper than that found in our study (Fig. 5), both studies found this community to reside slightly below the oxycline (Fig. 3 and Supporting Information Figs.S2, S4).Goericke et al. (2000) suggest that this community is dominated by low-lightadapted Prochlorococcus ecotypes (97% contribution to Chl a concentrations) with a minor presence of Synechococcus (< 3% of Chl a concentrations).
Ecological niches of C 3 typically (though not always) appear to manifest themselves with a smaller peak Chl a concentration (median 0.30 mg m À3 ) when compared with the DCM where C 2 dominates (median 0.49 mg m À3 ).The physiological and genetic adaptations required for phytoplanktonic growth at $ 100 m depths and their associated environmental constraints, such as low light (see Section 3.3), seem to reduce peak Chl a concentrations of the second DCM (C 3 ).While the peak is greater for C 2 than C 3 , the width of peak (m) is greater for C 3 (by $ 11 m).Located below the mixed layer at the top of the thermocline, C 2 may have greater temperature sensitivity than C 3 , characteristic of clades within the high-light-adapted Prochlorococcus thought to dominate in the eastern tropical North Pacific (Goericke et al. 2000).

Adaptation and composition of the low-oxygen community
The physiological response of phytoplankton to light availability is tied to depth (Khanna et al. 2005), vertical mixing (Lewis et al. 1984), and the spectral quality of available light (Sathyendranath and Platt 2007).Light availability is thought to be an important factor controlling vertical variations in community composition (M arquez-Artavia et al. 2019).Thriving at depths of $ 100 m, C 3 appears to be acclimated to very low light.The b bp : Chl a ratio has been used as a proxy for the carbon : Chl a ratio (Brewin et al. 2022), and to understand changes in photoacclimation and photoadaption.C 3 has significantly lower chl-specific backscattering than C 1 that resides in the well-lit mixed layer (Fig. 6b).The lower b bp : Chl a ratio of C 3 is consistent with the theoretical understanding of photoacclimation in low-light adapted phytoplankton (Sathyendranath et al. 2020;Brewin et al. 2022).With a low b bp : Chl a ratio, C 3 is composed of low-light adapted phytoplanktonic cells, while C 1 is characterized by cells adapted to high irradiance (Sathyendranath et al. 2020), consistent with the findings of Wojtasiewicz et al. (2021), who studied the phytoplanktonic community within the second DCM in very low oxygen waters of the Arabian Sea.The physiological adaptations to light variability, to achieve optimal growth at depths, have been linked to the phylogenetic distinct clades of the cyanobacteria, Synechococcus and Prochlorococcus (Biller et al. 2015).Considering the suggested dominance of Prochlorococcus within the ETNP, the low chl-specific backscattering may be attributed to the community's composition of primarily low-light-adapted Prochlorococcus ecotypes, similar to the communities identified by Wojtasiewicz et al. (2021) in the Arabian Sea.That said, it could also be argued that differences in the chl-specific backscattering coefficients of C 1 , C 2 , and C 3 could also reflect the depth variation in taxonomy and size of the phytoplanktonic cells (Brewin et al. 2022).Therefore, while the b bp : Chl a ratio is a useful tool to understand the phylogenetic adaptations and photoacclimation of naturally occurring low-light-adapted phytoplankton in the oceans, influence of species composition, taxonomy and cell structure should also be considered.Such consideration is required when comparing the chl-specific backscattering associated with C 2 and C 3 .Being unexpectedly similar, despite differences in depth and environmental conditions (see Section 3.5), differences in species composition of these two low-light adapted phytoplankton communities, could be responsible for the lack of variance between them.For example, the derivation of Chl a from fluorescence in this work assumes no change in the slope factor (linear conversion of fluorescence to Chl a) with depth or community composition.Yet, variations have been observed (Petit et al. 2022), which may have masked differences in the b bp : Chl a ratio between C 2 and C 3 in this work.Nitrogen-producing heterotrophic bacteria are also known to exist in OMZ and impact b bp (Rasse et al. 2020), which may also influence the b bp : Chl a ratio of C 3 , in instances when the heterotrophic bacterial backscattering signal is not captured in the background (b k bp ).

Importance to integrated stocks
The contribution of each community to the integrated stocks of Chl a and b bp provides an indication of their contribution to phytoplanktonic biomass.The contribution of C 1 to integrated Chl a is generally low, at around 11%, though, during blooms, can be high (Fig. 3c).Compared with C 2 and C 3 , C 1 has a higher contribution to integrated b bp , as can be inferred from the differences in the chl-specific particulate backscattering coefficient (Fig. 6a-c, see also Fig. 4 and Supporting Information Fig. S5).The communities C 2 and C 3 contribute more to integrated stocks of Chl a than C 1 , with roughly equal contributions of $ 45 AE 4% (average 44% and 41% respectively-see Fig. 6d).Yet, C 2 and C 3 contribute less than C 1 to integrated b bp , with each contributing $ 30%, compared to the $ 40% contribution to the total from C 1 , once the contribution of the background population of nonalgal particles was removed (Fig. 6c).Brewin et al. (2022) suggest that such patterns would be consistent with depth-dependant variations in the carbon : Chl a ratio of phytoplankton (see Section 3.3).The results presented here indicate C 3 to have important contributions to integrated stocks of Chl a ($ 41%) and b bp ($ 30% after the removal of background b bp values), which have implications for estimating phytoplankton stocks from satellite remote sensing, since like C 2 , this community cannot be observed from space.

Habitat and environmental conditions
Through partitioning Chl a and b bp into three communities, we have quantified differences in their environmental conditions (Fig. 7).Previous research has highlighted phytoplanktonic cells to be phylogenetically influencedspecifically through the influence on the Chl a : carbon ratioby differences in oceanographic gradients, including temperature, salinity, density, oxygen, and light gradients (Sathyendranath et al. 2020;Thompson et al. 2021).At a median depth of $ 106 m, C 3 thrives below the euphotic zone and is adapted to low irradiance (PAR around 0.02 molequanta m À2 d À1 , $ 0.04% of surface irradiance), while C 1 is adapted to high light (PAR around of 7.77 molequanta m À2 d À1 ).These adaptions are further supported by the low chl-specific backscattering of C 3 .Mella-Flores et al. (2012) have demonstrated that Prochlorococcus and Synechococcus strains tune their photosynthetic apparatus to PAR fluctuations.Goericke et al. (2000) suggested that the deep and tropical populations of Synechococcus in the ETNP, co-existing with Prochlorococcus in the C 3 , are physiologically inactive, and that the community is primarily dominated by low-light-adapted strains of Prochlorococcus.As seen in Fig. 3f, C 3 resides at DO concentrations < 5 μ mol kg À1 .Cepeda- Morales et al. (2009) suggested that this might be made possible by an adaptive mechanism that enables the maintenance of an aerobic metabolism in very low oxygen waters.Ulloa et al. (2021) have suggested this to be related to the supplementation of carbon and energy requirements, via organic compounds, that allows survival in dark poorly oxygenated waters, an adaptation indicated to be founded in the evolutionary relationship between Prochlorococcus and oxygen availability.
Strong stratification in the ETNP is mirrored in the oxygen distribution (Supporting Information Fig. 2) within the upper 200 m of an OMZ (Garcia-Robledo et al. 2017).C 1 resides in the turbulent mixed layer, with higher oxygen concentrations (a median of 181.79 μmol kg À1 and a Brunt-Välsälä frequency N 2 of 0.0007 s À2 ), between depths of 0-35 m (see Supporting Information Figs.S2, S4) where production is higher than consumption, and air-sea gas exchange occurs.The oxygen concentration surrounding C 2 deeper in the water column-at a median depth of $ 60 m-reaches values around 100 μmol kg À1 , as oxygen demand and respiration increases, and biomass-specific photosynthesis decreases, reducing oxygen production.Depletion of oxygen within this layer is due to heightened metabolic activity, microbial decomposition, and consumption within a stratified region (Garcio-Robledo et al. 2017;IPCC 2018).In drastic contrast to C 1 and C 2 , C 3 thrives in extremely low oxygen concentrations (< 1 μmol kg À1 ).Immediate consumption by microbial communities of oxygen produced by C 3 , along with the lack of oxygen replenishment within this layer due to sluggish circulation, has been referred to as a cryptic oxygen cycle (Garcia-Robledo et al. 2017).The low irradiance and low oxygen levels where C 3 resides correspond well with previous values identified in the eastern Equatorial Pacific, and also in the poorly-oxygenated waters of the Arabian Sea, both of which are thought to be dominated by the cyanobacterium, Prochlorococcus (Goericke et al. 2000;Cepeda-Morales et al. 2009;Lavin et al. 2010;Garcia-Robledo et al. 2017;Marquez-Artavia et al. 2019;Ulloa et al. 2021).We extracted environmental conditions for communities 2 and 3 at the depths of their maximum abundance.Our justification for this is that these depths likely reflect their environmental niches.However, we acknowledge that their presence is not solely limited to these depths, and that these communities are observed in a range of environmental conditions centered around those at the depths of their maximum.Additionally, seasonal variability in some of these environmental conditions is expected.

Role of low-oxygen phytoplankton in a changing world
Anthropogenic climate change projections reported by the IPCC (2018) highlight the implications of rising sea-water temperatures on the ocean's oxygen inventory.A reported 2% decrease in oxygen saturation between 1960 and 2018, is projected to increase by a further 1%-7% by 2100 under all of IPCC's representative concentration pathways (Keeling et al. 2010;IPCC 2018;Long et al. 2019).Intensification of oxygen depletion in the global ocean (Stramma et al. 2008) is expected to sharpen the oxycline and cause the shoaling of OMZ into and near the euphotic depth, potentially benefitting C 3 (Long et al. 2019).The biomass of C 3 -specifically that of the assumed dominant species Prochlorococcus-may increase as deoxygenation is enhanced globally, furthering their global importance (Agusti et al. 2019).

Future work
This work has been limited to the analysis of a single BGC-Argo float in the ETNP.Quantifying the wider distribution of this low-oxygen-adapted phytoplankton community may help our understanding of biogeochemical cycles in OMZ regions.Fitting the three-community model to multiple floats and datasets from the region (ETNP), combined with linking the parameters to larger scale environmental fields amenable to satellite remote sensing and model simulations, could provide insight into the distribution of C 3 at large scales.We have demonstrated the potential of this model to successfully capture the presence and properties of the C 3 populations; however, the limitations outlined in Section 3.1 should be considered when conducting future work and widescale application on BGC-Argo data.
Vertical partitioning achieved by this model provided an insight into C 3 's contribution to integrated stocks, and a comparison with the other communities, C 1 and C 2 .However, the influence of Chl a : biomass sensitivity (see Section 3.4), and their presence in such low-light conditions, may limit their contribution to integrated fluxes (particularly primary production).Additionally, information on the compositions of these phytoplanktonic communities, and how they may be changing, may provide insight into trophic pathways, secondary production, and trophic energy transfer (Brewin et al. 2022).Quantifying the rates of productivity of this low-light, lowoxygen adapted phytoplankton community, and their nutrient requirements, may further support our understanding on biogeochemical cycles in OMZs, while providing insights into the physiology of C 3 .

Summary
A diverse array of phytoplanktonic species supports the ocean's biogeochemical cycles, occupying and thriving in a wide range of environments within the sunlit regions of the ocean.One such environment is found at the base of the euphotic zone and oxycline, within the OMZ, where a unique community of phytoplankton is known to thrive.Monitoring this deep community of phytoplankton, however, is challenging, and as a consequence, little is known about its dynamics.Here, the two-community model of Brewin et al. (2022) was extended to include a third community (C 3 ) that thrives in a low-light and low-oxygen environment, within the OMZ.The model was tuned to data collected by a BGC-Argo float in the ETNP.Analysis of the model and its parameters revealed the model to reproduce effectively the patterns of both Chl a concentrations and particulate backscattering throughout the water column.C 2 and C 3 were found to have similarly low chl-specific particulate backscattering in comparison with C 1 , indicating low-light adaptation of both communities.The low-light, low-oxygen adapted phytoplankton, C 3 , is responsible for $ 41% of integrated Chl a (at 200 m integration), similar to that of C 2 , and $ 30% to integrated backscattering, after accounting for background concentrations of non-algal particles, consistent with our knowledge on the depth-dependency of the carbon : Chl a ratio of phytoplankton.Vertical partitioning revealed drastic differences in environmental conditions between where these three communities reside, within the water column.Our results emphasize how robotic platforms such as BGC-Argo floats can be utilized to observe unique phytoplanktonic communities that thrive in extremities, and that are otherwise distinctively challenging to observe and monitor.The combination of the BGC-Argo data and simple conceptual models can be used successfully to extract characteristics of these communities.Gaining an understanding of their composition and dynamics will provide an insight into the future of the marine ecosystem under incessant anthropogenic climate change.
05 and upper limit of 0.95, as with Brewin et al. (2022).Chlorophyllspecific particulate backscattering coefficients, b B bp,1 , b B bp,2 , and b B bp,3 , were then derived once ω 2 and b Ã bp,k were known, by dividing the scaling factors-ω 1 , ω 2 , and ω 3 -by the ratio of surface Chl a and surface particle backscattering (B s /b bp,s ).The background particle backscattering coefficient (b k bp )-thought to be dominated by nonalgal particles in the Pacific (Zhang et al. 2020)-was computed by multiplying b Ã

Fig. 2 .
Fig. 2.An example fit (profile collected on 04 June 2017) of the (a) scaled dimensionless variables of chlorophyll a, (b) rescaled fields in absolute units for chlorophyll a, (c) scaled dimensionless variables of particulate backscattering, and (d) rescaled fields in absolute units for particulate backscattering models.

Fig. 3 .
Fig. 3. Contour plots of the Chl a concentrations collected by the BGC-Argo float between 02 December 2016 and 08 July 2021 in the top 200 m of the water column.(a) Total Chl a data extracted from the BGC-Argo float, (b) parameterized and tuned model output of the total Chl a, (c) differences between the total Chl a model output and data, (d) model output for C 1 , (e) model output for C 2 , and (f) model output for C 3 .White line (black line in subplot c) represents the depth where DO falls below 5 μmol kg À1 .

Fig. 5 .
Fig. 5. Boxplots of the (a) Chl a peaks, (b) depth of Chl a peaks, and (c) width of Chl a peaks, for community 2 (C 2 ) and community 3 (C 3 ).

Fig. 6 .
Fig. 6.Boxplots displaying (a) chlorophyll-specific particulate backscattering values for C 1 , C 2 , C 3 , and the background backscattering coefficient (b k bp ), (b) fractions of integrated particulate backscattering for C 1 , C 2 , C 3 , and b k bp , (c) fractions of integrated particulate backscattering for C 1 , C 2 , C 3 having removed b k bp , (d) fractions of integrated Chl a for C 1 , C 2 , and C 3 .
23.9 C), but warmer than C 3 .Further below the mixed layer, at a median depth of 106 m, C 3 thrives in the coldest thermal conditions (median of 14.6 C).Therefore, C 3 has adapted to thrive at depths with cooler temperatures.Varkey et al. (2015) suggest that Synechococcus clades I and IV can thrive in low temperatures alongside Prochlorococcus lowlight-adapted ecotypes.