# PHOSPHORUS ALONG THE SOIL-FRESHWATER-OCEAN CONTINUUM

EDITED BY : Barbara J. Cade-Menun, Solange Duhamel, Rosalind J. Dodd, Christian Lønborg, Chris T. Parsons and William D. Taylor PUBLISHED IN : Frontiers in Marine Science, Frontiers in Microbiology, Frontiers in Earth Science and Frontiers in Environmental Science

#### Frontiers Copyright Statement

© Copyright 2007-2019 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.

The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.

Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.

Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.

As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.

All copyright, and all rights therein, are protected by national and international copyright laws.

The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use. ISSN 1664-8714 ISBN 978-2-88945-829-5 DOI 10.3389/978-2-88945-829-5

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

Frontiers in Marine Science 1 April 2019 | Phosphorus Along the Soil-Freshwater-Ocean Continuum

# PHOSPHORUS ALONG THE SOIL-FRESHWATER-OCEAN CONTINUUM

Topic Editors:

Barbara J. Cade-Menun, Agriculture & Agri-Food Canada, Canada Solange Duhamel, Columbia University, United States Rosalind J. Dodd, Lincoln University, New Zealand Christian Lønborg, Australian Institute of Marine Science, Australia Chris T. Parsons, University of Waterloo, Canada William D. Taylor, University of Waterloo, Canada

Phosphorus (P) is an essential element for all organisms. However, there is a P paradox, whereby P concentrations considered deficient in some environments such as in agricultural soils are considered excessive in water, where they trigger eutrophication. Ensuring adequate P for crop production while minimizing water quality degradation requires consideration of the P continuum from soils to freshwater and oceans. It also requires an international, interdisciplinary approach to monitoring and scientific research. This eBook brings together P studies in soil science, lakes, rivers, estuaries and oceans, with 74 authors from 12 countries in Asia, Europe and North America. The papers assembled here provide important new information to address knowledge gaps, cover P forms and cycling in soil and water, and identify key priorities for future research. Thus, the papers assembled here provide current and interdisciplinary information about P forms and their cycling along the soilfreshwater-ocean continuum, which is essential for environmentally sustainable P use.

Citation: Cade-Menun, B. J., Duhamel, S., Dodd, R. J., Lønborg, C., Parsons, C. T., Taylor, W. D., eds. (2019). Phosphorus Along the Soil-Freshwater-Ocean Continuum. Frontiers Media. doi: 10.3389/978-2-88945-829-5

# Table of Contents

*05 Editorial: Phosphorus Along the Soil-Freshwater-Ocean Continuum* Barbara J. Cade-Menun, Solange Duhamel, Rosalind J. Dodd, Christian Lønborg, Chris T. Parsons and William D. Taylor

#### SECTION 1

#### SOIL PHOSPHORUS FORMS AND MANAGEMENT

*08 Long-Term Land Use Affects Phosphorus Speciation and the Composition of Phosphorus Cycling Genes in Agricultural Soils*

Jin Liu, Barbara J. Cade-Menun, Jianjun Yang, Yongfeng Hu, Corey W. Liu, Julien Tremblay, Kerry LaForge, Michael Schellenberg, Chantal Hamel and Luke D. Bainard


Roland Bol, Gerard Gruau, Per-Erik Mellander, Rémi Dupas, Marianne Bechmann, Eva Skarbøvik, Magdalena Bieroza, Faruk Djodjic, Miriam Glendell, Philip Jordan, Bas Van der Grift, Michael Rode, Erik Smolders, Mieke Verbeeck, Sen Gu, Erwin Klumpp, Ina Pohle, Maelle Fresne and Chantal Gascuel-Odoux

#### SECTION 2

#### PHOSPHORUS FORMS AND CYCLING IN FRESHWATER AND ESTUARIES

	- O. Roger Anderson

Lisa Felgentreu, Günther Nausch, Franziska Bitschofsky, Monika Nausch and Detlef Schulz-Bull


*111 Phosphorus Forms in Sediments of a River-Dominated Estuary* Sheree J. Watson, Barbara J. Cade-Menun, Joseph A. Needoba and Tawnya D. Peterson

#### SECTION 3

#### PHOSPHORUS FORMS AND CYCLING IN OCEANS


Kahina Djaoudi, France Van Wambeke, Laurent Coppola, Fabrizio D'Ortenzio, Sandra Helias-Nunige, Patrick Raimbault, Vincent Taillandier, Pierre Testor, Thibaut Wagener and Elvira Pulido-Villena

# Editorial: Phosphorus Along the Soil-Freshwater-Ocean Continuum

Barbara J. Cade-Menun<sup>1</sup> \*, Solange Duhamel <sup>2</sup> , Rosalind J. Dodd<sup>3</sup> , Christian Lønborg<sup>4</sup> , Chris T. Parsons <sup>5</sup> and William D. Taylor <sup>6</sup>

*<sup>1</sup> Agriculture and Agri-Food Canada, Swift Current Research and Development Centre, Swift Current, SK, Canada, <sup>2</sup> Lamont Doherty Earth Observatory, Division of Biology and Paleo Environment, Columbia University, Palisades, NY, United States, <sup>3</sup> Faculty of Agriculture and Life Sciences, Lincoln University, Lincoln, New Zealand, <sup>4</sup> Australian Institute of Marine Science, Townsville, QLD, Australia, <sup>5</sup> Department of Earth and Environmental Sciences and the Water Institute, University of Waterloo, Waterloo, ON, Canada, <sup>6</sup> Department of Biology, University of Waterloo, Waterloo, ON, Canada*

Keywords: phosphorus, land-water continuum, lakes, estuaries, oceans, soil

#### **Editorial on the Research Topic**

#### **Phosphorus Along the Soil-Freshwater-Ocean Continuum**

Phosphorus (P) is an essential element for all organisms. However, there is a P paradox, whereby P concentrations considered deficient in some environments such as in agricultural soils are considered excessive in freshwater, where they trigger eutrophication (e.g., Sims and Sharpley, 2005 and references therein; Elser and Bennet, 2011; Lougheed, 2011). Geographical imbalances also occur, with excesses in Western Europe and North America and deficiencies in regions with highly weathered soils, such as sub-Saharan Africa. There is a strong link between soil P stores and P mobilization and transfer to receiving waters, termed the P transfer continuum (Haygarth et al., 2005). Ensuring adequate P for crop production while minimizing water quality degradation requires consideration of this continuum and an international, interdisciplinary approach. This research topic brings together P studies in soil science, lakes, rivers, estuaries, and oceans, with 74 authors from 12 countries in Asia, Europe, and North America, and identifies key priorities for future research.

In all ecosystems, P exists in many chemical forms. Inorganic P compounds include phosphate, pyrophosphate, and polyphosphate, with organisms directly using phosphate (H2PO<sup>−</sup> 4 or HPO2<sup>−</sup> 4 at the pH range of most natural ecosystems). Organic P compounds contain a link to carbon (C), and include orthophosphate monoesters [e.g., glucose 6-phosphate and myo-inositol hexakisphosphate (myo-IHP or phytate)]; orthophosphate diesters (e.g., nucleic acids, phospholipids); phosphonates (e.g., aminoethyl phosphonate), and biological polyphosphates (e.g., ATP). In all environments, P cycles geochemically and biologically. Geochemical processes include adsorption/desorption of organic and inorganic P compounds with mineral particles, and precipitation with cations in solution. In biological cycling, phosphate is taken up by organisms and converted to organic P forms such as nucleic acids or is stored in cells as phytate or polyphosphate. This P can be released by secretion or after cell death, and converted back to phosphate by mineralization, primarily an enzymatic process involving P-specific enzymes (phosphatases). For more information on P cycling, a number of review papers are available (e.g., Condron et al., 2005; Pierzynski et al., 2005; Paytan and McLaughlin, 2007; Baldwin, 2013; Karl, 2014; Orihel et al., 2017).

There are similarities and differences in P cycling in terrestrial and aquatic ecosystems. Terrestrial organisms obtain phosphate from the pool that is dissolved in the soil solution, in the space between the solid components of soil (minerals and organic matter). Here, P availability is limited by moisture. Easily-obtained (labile) phosphate is taken up from the soil solution, then replenished by desorption, dissolution, or mineralization.

#### Edited and reviewed by:

*Eric 'Pieter Achterberg, GEOMAR Helmholtz Center for Ocean Research Kiel, Germany*

#### \*Correspondence:

*Barbara J. Cade-Menun barbara.cade-menun@canada.ca*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science*

Received: *20 December 2018* Accepted: *21 January 2019* Published: *25 February 2019*

#### Citation:

*Cade-Menun BJ, Duhamel S, Dodd RJ, Lønborg C, Parsons CT and Taylor WD (2019) Editorial: Phosphorus Along the Soil-Freshwater-Ocean Continuum. Front. Mar. Sci. 6:28. doi: 10.3389/fmars.2019.00028* Because terrestrial plants are fixed in place, strategies have evolved to improve their P acquisition. This includes changes in rooting structure or associations with rhizosphere microbes and symbionts, which can enhance access to phosphate, or production of phosphatases or organic acids to desorb P compounds from mineral surfaces (Richardson et al., 2011). Aquatic environments include sediments, benthic biofilms, and the water column. In sediments, organic and inorganic P compounds will be associated with mineral particles, algal and bacterial biomass (particularly at the sediment-water interface) and organic matter, or occur in aqueous form in pore water. Phosphate and other P forms can exchange with the water column through adsorption/desorption and precipitation/dissolution, and through uptake and remineralization by organisms in the sediment and in the water column. Within the water column, organic and inorganic P compounds can be dissolved, associated with colloids, or associated with larger particulates including plankton (e.g., Paytan and McLaughlin, 2007; Orihel et al., 2017; Taylor and Lean, 2018).

Although it can be the most limiting nutrient, or second to nitrogen (N) in many environments, P cycling in terrestrial and aquatic ecosystems in still poorly understood relative to C or N, because P concentrations are orders of magnitude lower than those of C and N, and due to methodological limitations. Unlike C and N, P has no significant natural gaseous forms, and only one stable isotope (31P), which can hamper tracing P sources in the environment, but it does have two radioisotopes ( <sup>32</sup>P and <sup>33</sup>P) that occur naturally or can be added as tracers. Total P is determined by digestion of solids, or by inductively coupled argon plasma optical emission spectroscopy (ICP-OES) of liquids, and phosphate concentrations in liquids or extracted samples are analyzed by colorimetric techniques such as the molybdate blue method (Murphy and Riley, 1962). However, characterization of specific P compounds requires advanced techniques such as <sup>31</sup>P-nuclear magnetic resonance (P-NMR) spectroscopy or P-X-ray Absorption Near Edge Structure (P-XANES) spectroscopy that are not readily available to all researchers. Information about biological P cycling is also limited, particularly the roles of specific organisms. The links between the P cycle and those of other elements, including C and N, are also poorly understood.

The papers assembled here provide important new information to address knowledge gaps, and cover all aspects of the soil-freshwater-ocean continuum. Liu et al. examined P cycling in agricultural lands in Canada. Bol et al. and Reid et al. addressed P transfer from agricultural land to water, with Reid et al. assessing the components of P loss and assessment tools, and Bol et al. discussing the need for more integrated research efforts into the processes and mechanisms controlling P loss. In aquatic environments, P cycling in lakes was investigated by Thompson and Cotner; Anderson; and Depew et al; in estuaries by Watson et al.; in rivers by Felgentreu et al. and Gomes et al. and in oceans by Björkman et al.; Diaz et al. and Djaoudi et al. Across disciplines, several themes emerged. With respect to methods, colorimetric techniques were frequently used. However, Felgentreu et al. assessed filtration and colorimetric methods in river samples to improve routine monitoring programs, while Djaoudi et al. used techniques to improve the sensitivity of the Murphy and Riley (1962) colorimetric technique to determine nanomolar concentrations of dissolved inorganic P in the Mediterranean Sea. Liu et al. and Watson et al. both used P-NMR to characterize organic P forms, in soils and estuary sediments respectively, and Liu et al. used P-XANES to characterize soil inorganic P species. Oxygen isotope ratios of phosphate were used by Depew et al. to identify sources of P to Lake Erie and pathways of cycling.

Biological P cycling and P bioavailability in different ecosystems was a common theme. Liu et al. identified microbial communities in various agricultural soils, showing that soil disturbance controlled P forms and cycling in these ecosystems. Gomes et al. demonstrated that Amazon and Changjiang river plume phytoplankton communities were shaped by river N:P ratios. Björkman et al. assessed spatial variability in the dynamics of inorganic and organic P compounds in the North Pacific Tropical Gyre with radio tracer techniques, while Diaz et al. estimated the lability of model P compounds by diatom cultures and used proteomics to examine phosphatase diversity. Anderson measured alkaline phosphomonoesterase activities in freshwater microbial communities, while Thompson and Cotner investigated the bioavailability of dissolved organic P in lakes in Minnesota and South Dakota; both studies linked P and C biogeochemical cycles.

Thus, the articles in this research topic provide current and interdisciplinary information about P forms and their cycling along the soil-freshwater-ocean continuum, which is essential for environmentally sustainable P use.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### REFERENCES


the phosphorus efficiency of agriculture. Plant Soil 349, 121–156. doi: 10.1007/s11104-011-0950-4.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Duhamel, Dodd, Lønborg, Parsons, Taylor and Her Majesty the Queen in Right of Canada, as represented by the Minister of Agriculture & Agri-Food Canada. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Long-Term Land Use Affects Phosphorus Speciation and the Composition of Phosphorus Cycling Genes in Agricultural Soils

Jin Liu1,2, Barbara J. Cade-Menun<sup>3</sup> , Jianjun Yang<sup>4</sup> , Yongfeng Hu<sup>5</sup> , Corey W. Liu<sup>6</sup> , Julien Tremblay<sup>7</sup> , Kerry LaForge<sup>3</sup> , Michael Schellenberg<sup>3</sup> , Chantal Hamel<sup>3</sup> and Luke D. Bainard<sup>3</sup> \*

<sup>1</sup> College of Agronomy and Biotechnology, China Agricultural University, Beijing, China, <sup>2</sup> Visiting Scientist, Agriculture and Agri-Food Canada, Swift Current Research and Development Centre, Swift Current, SK, Canada, <sup>3</sup> Swift Current Research and Development Centre, Agriculture and Agri-Food Canada, Swift Current, SK, Canada, <sup>4</sup> Institute of Environment and Sustainable Development in Agriculture, Chinese Academy of Agricultural Sciences, Beijing, China, <sup>5</sup> Canadian Light Source, University of Saskatchewan, Saskatoon, SK, Canada, <sup>6</sup> Stanford Magnetic Resonance Laboratory, Stanford University School of Medicine and ChEM-H-Stanford University, Stanford, CA, United States, <sup>7</sup> Energy, Mining and Environment, National Research Council of Canada, Montreal, QC, Canada

#### Edited by:

Rosalind Jane Dodd, Bangor University, United Kingdom

#### Reviewed by:

John W. Moreau, University of Melbourne, Australia Mustafa Yucel, Middle East Technical University, Turkey

\*Correspondence:

Luke D. Bainard luke.bainard@agr.gc.ca

#### Specialty section:

This article was submitted to Microbiological Chemistry and Geomicrobiology, a section of the journal Frontiers in Microbiology

Received: 31 March 2018 Accepted: 02 July 2018 Published: 20 July 2018

#### Citation:

Liu J, Cade-Menun BJ, Yang J, Hu Y, Liu CW, Tremblay J, LaForge K, Schellenberg M, Hamel C and Bainard LD (2018) Long-Term Land Use Affects Phosphorus Speciation and the Composition of Phosphorus Cycling Genes in Agricultural Soils. Front. Microbiol. 9:1643. doi: 10.3389/fmicb.2018.01643 Agriculturally-driven land transformation is increasing globally. Improving phosphorus (P) use efficiency to sustain optimum productivity in diverse ecosystems, based on knowledge of soil P dynamics, is also globally important in light of potential shortages of rock phosphate to manufacture P fertilizer. We investigated P chemical speciation and P cycling with solution <sup>31</sup>P nuclear magnetic resonance, P K-edge X-ray absorption nearedge structure spectroscopy, phosphatase activity assays, and shotgun metagenomics in soil samples from long-term agricultural fields containing four different land-use types (native and tame grasslands, annual croplands, and roadside ditches). Across these land use types, native and tame grasslands showed high accumulation of organic P, principally orthophosphate monoesters, and high acid phosphomonoesterase activity but the lowest abundance of P cycling genes. The proportion of inositol hexaphosphates (IHP), especially the neo-IHP stereoisomer that likely originates from microbes rather than plants, was significantly increased in native grasslands than croplands. Annual croplands had the largest variances of soil P composition, and the highest potential capacity for P cycling processes based on the abundance of genes coding for P cycling processes. In contrast, roadside soils had the highest soil Olsen-P concentrations, lowest organic P, and highest tricalcium phosphate concentrations, which were likely facilitated by the neutral pH and high exchangeable Ca of these soils. Redundancy analysis demonstrated that IHP by NMR, potential phosphatase activity, Olsen-P, and pH were important P chemistry predictors of the P cycling bacterial community and functional gene composition. Combining chemical and metagenomics results provides important insights into soil P processes and dynamics in different land-use ecosystems.

Keywords: land use, soil, phosphorus, solution NMR, XANES, shotgun metagenomics

# INTRODUCTION

fmicb-09-01643 July 23, 2018 Time: 16:4 # 2

Driven by the increasing demand for agricultural production, land-use change has been widespread globally over the last several decades (Guillaume et al., 2015). In most parts of the world, the original vegetation has been cleared for the expansion of croplands and pastures, both of which are typical land uses crucial for food production (Houghton, 1994). Agricultural areas mostly devoted to either arable croplands or grazed pastures comprise about one third of the land surface globally (FAO/UNEP, 2015). The global effects of land-use change also contribute to global changes in nutrient cycling and dynamics (Houghton, 1994). This is especially significant for phosphorus (P), which often limits the productivity and sustainability of agriculture, requiring fertilization. Rock phosphate sources used to produce fertilizers are globally limited, and there are concerns about their longterm availability (Elser and Bennett, 2011; Sharpley et al., 2013). Additionally, P loss from agriculture can have a negative effect on the aquatic environment. This is expected to continue even if P fertilization is reduced due to the large amount of residual P accumulated in agricultural soils through time in many regions (Garcia-Montiel et al., 2000; Sharpley et al., 2013; Stutter et al., 2015). Efficient P use is therefore a priority when replacing natural ecosystems with managed ecosystems (Stutter et al., 2015).

Conversion of land-use is expected to change soil P dynamics. Changes in P inputs and outputs through management practices that alter soil physical, chemical and biological properties affect the chemical nature of different P species in soils and ultimately their bioavailability (Condron et al., 2005; Maranguit et al., 2017). A full understanding of the effects of land-use and management systems on soil P composition remains obscure, partially due to methodological limitations such as a reliance on the widely used but operationally defined sequential fractionation approach (Guggenberger et al., 1996; Negassa and Leinweber, 2009; Crews and Brookes, 2014; Maranguit et al., 2017). A more useful method, capable of identifying P species, particularly organic P (Po) compounds, in soils at the molecular level, is solution <sup>31</sup>P nuclear magnetic resonance (P-NMR) spectroscopy. With P-NMR, more detailed insights into P<sup>o</sup> species behind the P pools have been revealed in recent years (Condron et al., 2005; McDowell and Stewart, 2006; Stutter et al., 2015). Nevertheless, many published studies have limited peak identification to clearly separated, distinct peaks only (e.g., orthophosphate and pyrophosphate) and have grouped the remaining peaks together into broad compound classes such as orthophosphate monoesters and diesters; this provides limited information about P cycling and availability (Cade-Menun, 2017). Additionally, even within a single broad category such as orthophosphate monoesters or orthophosphate diesters, P<sup>o</sup> forms differ in their bioavailability and reactivity. As such, a full understanding of P cycling in soils requires identifying as many specific P<sup>o</sup> species as possible (Cade-Menun, 2015; Cade-Menun, 2017). Additionally, P K-edge X-ray absorption near-edge structure (P-XANES) spectroscopy provides a new and powerful approach to directly identify inorganic P (Pi) compounds (Prietzel et al., 2013; Liu et al., 2015). Therefore, the combined application of P K-edge XANES and

Land use change will introduce significant changes in vegetation, which can alter soil biology and nutrient cycling. Vegetation changes will change rooting depth, nutrient and water uptake, soil chemistry and symbioses, such as N fixation or mycorrhizae (Bainard et al., 2017; Cade-Menun et al., 2017a). Plants and microorganisms are essential drivers of soil P turnover and dynamics. They enhance the solubilization of P by the release of low molecular weight acids and phosphatases that mineralize P<sup>o</sup> (Richardson et al., 2011). Under conditions of phosphate deficiency, bacteria can induce the phosphate (Pho) regulon to excrete phosphatases to obtain bioavailable orthophosphate (Santos-Beneit, 2015). Land use has been shown to influence specific functional genes (e.g., phytase and phosphatase genes), and the composition of microbial communities associated with P cycling in soils (Jangid et al., 2008; Neal et al., 2017). Few studies to date have generated a thorough insight into the response of soil microbial communities and their P cycling capacity, coupled with soil P chemistry (e.g., soil physico-chemical parameters and P speciation) to land use.

The general objective of this study was to investigate the effects of land use change on P cycling, using advanced chemical techniques and metagenomics sequencing. We chose four agricultural areas in southwestern Saskatchewan, with adjacent sites of four typical agricultural land uses in the region: annual cropland, native grassland, tame grassland and roadsides. The close proximity of these locations and their historical continuity of land use (each > 50 years) kept all otherwise interrelated variables relatively constant (climate, topography, parent material, etc.) except land use. Samples from the four locations were patterned as replicated sites for each land use, providing a unique and valuable research platform to clarify soil P cycling induced by land use change. Combining state-of-theart spectroscopic approaches (solution P-NMR and P K-edge XANES spectroscopy) with metagenomics, the specific objectives of this study were: (1) to characterize the composition of P<sup>i</sup> and P<sup>o</sup> in the soils under various land uses; (2) to investigate the abundance and composition of P functional genes and the microbial community within various ecosystems; and (3) to link these chemical and metagenomics results together for a better understanding of P cycling processes under different land uses.

# MATERIALS AND METHODS

# Field Sites and Sample Collection

The four experimental sites [Auvergne Wise Creek (AWC), Val Marie (VM), Masefield (MF1, MF2)], located in southwestern Saskatchewan, have the same soil type (well-drained Orthic Brown Chernozems; Saskatchewan Environment, and Resource Management [SERM], 1997) and a known history of more than 50 years in each studied land use type (Cade-Menun et al., 2017a). The native grasslands were a mixed grass prairie community, and tame grasslands were crested wheatgrass [Agropyron cristatum (L.) Gaetern.] stands that had been established for over 50 years.

The native and tame grasslands used for this study were in pastures grazed for beef production. Croplands were in dryland (unirrigated) annual wheat-based production. The fourth land use type in this study was roadside ditches, which serve as buffers between roads and fields. More details of the experimental sites and the broader land use study are available in Cade-Menun et al. (2017a). For the current study, soil samples were collected in July 2013 from the 0 to 30 cm depth from four land use types at four locations (n = 16). At each location, six soil cores (1.9 cm diameter) were collected inside four 1 m<sup>2</sup> quadrats that were situated along a 10 m transect for a total of 24 total soil cores per location. This differs from Cade-Menun et al. (2017a) with respect to date, sampling depth and number of study sites. Soil samples were stored in a cooler with ice packs while in the field. At the lab, field-moist soil cores from each location were pooled together and sieved (<2 mm) to form one composite sample per location and to remove rocks, larger roots, and coarse plant material. A portion of each composite soil sample was air-dried and stored at room temperature for chemical analysis, including XANES. An additional sub-sample was immediately stored at −20◦C for molecular analysis, and the remainder was refrigerated (4◦C) for enzyme assays and extraction for P-NMR.

#### Chemical Analysis

Soil pH was measured in CaCl<sup>2</sup> (1:2 w/v; Hendershot et al., 2008). Soils were analyzed for total C, total N, and organic C (after acidification) by dry combustion (Vario Micro Cube, Elementar). Total P was determined by digestion (Parkinson and Allen, 1975), total P<sup>o</sup> was determined by the ignition method (Saunders and Williams, 1955), and Olsen-P was determined with sodium bicarbonate extraction (Sims, 2009), all followed by colorimetric analysis (Murphy and Riley, 1962). Mehlich-3 extraction (Sims, 2009) was used to determine P, Al, Ca, and Fe through analysis of extracts by inductively coupled plasma optical emission spectroscopy (ICP-OES; Thermo Scientific ICAP 6300 Duo). Exchangeable Ca was extracted in ammonium acetate and measured by ICP-OES (Hendershot et al., 2008). The activities of acid and alkaline phosphomonoesterase were assayed with p-nitrophenyl phosphate as substrates with the buffer pH-values adjusted to 6.5 and 11, respectively, phosphodiesterase activity was assayed with bis-p-nitrophenyl phosphate as substrate with the buffer pH 8.0 (Tabatabai, 1994).

# Solution P-NMR Spectroscopy

Refrigerated soils were extracted with NaOH-EDTA in a 1:10 soil: extract ratio for P-NMR as previously described (Liu et al., 2015). Solution P-NMR spectra were collected as described in Cade-Menun et al. (2017b), using a Varian INOVA 600 MHz (202.5 MHz for P) spectrometer with a 10 mm broadband probe at the Stanford Magnetic Resonance Laboratory. The NMR parameters were: 90◦ pulse (30 µs), 0.675 s acquisition time, 4.32 s pulse delay, 20◦C, 2,160–11,520 scans (3–16 h); no proton decoupling. The delay time used was based on the P:(Fe + Mn) concentrations in extracts (McDowell et al., 2006; Cade-Menun and Liu, 2014). To facilitate peak identification, spiking experiments with phytate, α- and β-glycerophosphate and adenosine monophosphate were conducted (Cade-Menun, 2015; Liu et al., 2015; Cade-Menun et al., 2017b). Compounds were identified by their chemical shifts after the orthophosphate peak in each spectrum was standardized to 6.0 ppm during processing. Peak areas were calculated by integration on spectra processed with 7 and 2 Hz line-broadening, using NUTS software (2000 edition; Acorn NMR, Livermore, CA, United States) and manual calculation. Percentages of orthophosphate monoesters and diesters were corrected for degradation of diesters to monoesters during NMR analysis (Liu et al., 2015; Cade-Menun et al., 2017b).

# Phosphorus K-Edge XANES Spectroscopy

Phosphorus K-edge XANES spectra were collected at the Soft X-ray Micro-characterization Beamline (SXRMB) equipped with a InSb(111) double-crystal monochromator at the Canadian Light Source (CLS), Saskatoon, SK, Canada. Detailed information on instrument setting, sample preparation and data collection was described previously (Liu et al., 2014, 2015). In brief, soil samples were thinly spread over a P-free and double-sided carbon tape for the XANES measurements. The soil spectra were collected in partial fluorescence yield (PFY) mode using a four-element fluorescence detector. At least three XANES spectra were collected and averaged for each soil sample to obtain acceptable signal-to-noise level. Radiation damage during XANES experiment was excluded by a good reproducibility of the repeated measurements on the same spot and repeated scans over different spots for each sample. All XANES spectra were analyzed by Athena (Ravel and Newville, 2005). The absolute energy scale was calibrated to 2,149 eV (E0) as the maximum energy of the first peak in the first derivative spectrum of AlPO<sup>4</sup> (Beauchemin et al., 2003). Spectra were background corrected by a linear regression fit through the pre-edge region and normalized total K-edge intensity to one unit edge jump by defining the continuum regions (>50 eV above absorption edge) as the post-edge region. Principal component analysis (PCA) was performed on the set of 16 soil XANES spectra using the program SixPack (Webb, 2005). According to PCA results (Supplementary Table S1), the minimum indicator (IND) suggested that four components were optimal for linear combination fitting (LCF) analysis of these soil samples. Consistently, the variations of the fifth component almost represented random variations due to noise rather than real spectral variations (Supplementary Figure S1). As there were up to four components contributing to 92.5% of spectra variations for all of the investigated samples (Supplementary Table S1 and Supplementary Figure S1), LCF of soil spectra were performed over the spectral energy region from 2,139 to 2,164 eV using all possible binary, ternary, and quaternary combinations of our reference spectra which were collected at the same beamline and reported in our previous studies (Liu et al., 2013, 2015). The E<sup>0</sup> was fixed during LCF analysis and weights of all P standards used were forced to sum 1. Phosphorus forms with proportions <10% were excluded from the fit set and replaced with other possible P species (Weyers et al., 2016). The goodness-of-fit was judged by the Chi-squared values and R values, and P standards yielding the best fit were considered as the most possible P species in the investigated soil samples (Supplementary Figure S2 and Supplementary Table S2).

#### Metagenomic Analysis

fmicb-09-01643 July 23, 2018 Time: 16:4 # 4

Total nucleic acids were extracted from 1 g (2 × 0.5 g) frozen soil from each sample using the PowerSoil DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, United States) following the manufacturer's recommended protocol. The DNA samples were quantified using the Qubit dsDNA BR Assay Kit (ThermoFisher Scientific) and 2,100 Bioanalyzer instrument. Metagenomic libraries were prepared and sequenced on an Illumina HiSeq2500 system on a rapid mode 2 × 150 bp configuration. A total of 20 samples were submitted for metagenome sequencing of which the resulting data (273 Giga-bases) were processed through our metagenomics bioinformatics pipeline (Tremblay et al., 2017). Read count summaries and mapping statistics are provided (Supplementary Table S3). Sequencing adapters were removed from each read and bases at the end of reads having a quality score <30 were cut off (Trimmomatic v0.32; Bolger et al., 2014) and scanned for sequencing adapters contaminants reads using DUK<sup>1</sup> to generate quality controlled (QC) reads. The QC-passed reads from each sample were co-assembled using Megahit v1.1.2 (Li et al., 2015) on a 3 Tera-Bytes of RAM compute node with iterative kmer sizes of 31, 41, 51, 61, 71, 81, and 91 bases (see Supplementary Table S4 for assembly statistics). Gene prediction was performed by calling genes on each assembled contig using Prodigal v2.6.2 (Hyatt et al., 2010). Genes were annotated following the JGI's guidelines (Huntemann et al., 2015) including the assignment of KEGG orthologs (KO; Kanehisa and Goto, 2000). The QC-passed reads were mapped (BWA mem v0.7.15<sup>2</sup> ) against contigs to assess quality of metagenome assembly and to obtain contig abundance profiles. Alignment files in bam format were sorted by read coordinates using samtools v1.2 (Li et al., 2009), and only properly aligned read pairs were kept for downstream steps. Each bam file (containing properly aligned paired-reads only) was analyzed for coverage of called genes and contigs using bedtools (v2.17.0; Quinlan and Hall, 2010) using a custom bed file representing gene coordinates on each contig. Only paired reads both overlapping their contig or gene were considered for gene counts. Coverage profiles of each sample were merged to generate an abundance matrix (rows = contig, columns = samples) for which a corresponding CPM (Counts Per Million–normalized using the TMM method; edgeR v3.10.2; Robinson et al., 2010). Taxonomic summaries were performed using a combination of in-house Perl and R scripts and Qiime v.1.9.1 (Caporaso et al., 2010).

#### Statistical Analyses

One-way ANOVAs were conducted for all the data of each land use (n = 4) separately, using SPSS 13.0 (SPSS, Inc), followed by a least significant difference (LSD) test with α = 0.05. The P-NMR data were clr transformed prior to statistical analysis (Abdi et al., 2015; Liu et al., 2015); other data were transformed as needed

<sup>1</sup>http://duk.sourceforge.net/

<sup>2</sup>http://bio-bwa.sourceforge.net

for normality. Redundancy analysis (RDA) was used to identify the important P chemistry-related drivers of the P-cycling soil bacterial community and functional gene composition. A set of nonredundant predictors of the bacterial community and functional gene composition were selected using the ordistep and envfit functions (vegan package, R 3.4.3). Only the significant variables were included in the final models, excluding collinear variables with a variance inflation factor >10.

# RESULTS

#### Soil Properties

Among the investigated land uses, there were no significant differences in soil pH, total P, total C, organic C, total N, Mehlichextractable P, Al, Fe, and Ca (**Table 1**). The percentage of P<sup>o</sup> was significantly higher for both native and tame grasslands (both 75.4%) than for annual croplands (52.6%) and roadside soils (54.4%, **Table 1**). The highest Olsen-P concentration was observed in roadside soils (12.5 mg kg−<sup>1</sup> ), which was significantly higher than native grasslands (3.8 mg kg−<sup>1</sup> ) and annual croplands (5.6 mg kg−<sup>1</sup> , **Table 1**). The highest exchangeable Ca occurred in roadside soils and the lowest in tame grasslands (**Table 1**).

#### Solution P-NMR Spectroscopy and Soil Phosphatase Activities

Examples of P-NMR spectra for these samples are shown in **Figure 1** and Supplementary Figure S3; chemical shifts of identified peaks are shown in Supplementary Table S5, and proportions of P forms and compound classes determined by P-NMR are shown in **Table 2** and Supplementary Table S6. Extraction with NaOH-EDTA recovered 34.9–50.5% of total P without significant differences among the land use types (**Table 2**). Inorganic P in the NaOH-EDTA extracts was significantly greater in roadside samples (46.3%) than native and tame grasslands (31.5 and 30.9%; **Table 2**), and was mainly orthophosphate (27.8–43.1%) with traces of polyphosphate (2.2–2.4%) and pyrophosphate (0.7–1.3%, Supplementary Table S6). The percentage of phosphonates was low (1.3–1.8%) with no differences among land use types. There were no significant differences among land use types for orthophosphate diesters with (CDiest, 19.5–24.0%) and without (Di, 2.5–4.1%) correction for degradation products (Deg, **Table 2**). In these samples, diester degradation was primarily from RNA to mononucleotides (Nucl, Supplementary Table S6), because the percentages of α- and β-glycerophosphates (degradation products of phospholipids) were low, and peaks were identified in the OthDi1 category where undegraded phospholipids would be observed (Supplementary Table S6). There were significant differences among the land use types for orthophosphate monoesters, with (Mono) and without (Cmono) correction for degradation (**Table 2**), with native and tame grasslands greater than roadside and annual cropland samples.

By NMR, P<sup>o</sup> represented 53.7–69.2% of the NaOH-EDTA extracted P in the tested soils (**Table 2**). The greatest P<sup>o</sup> proportions were in native and tame grasslands (**Table 2**) and the


least was in roadside soils, consistent with general soil chemical results (**Table 1**). The significant differences among the land use types for P<sup>o</sup> predominantly occurred through differences in inositol hexakisphosphates (IHP, **Table 2**) and an unidentified P compound resolved at 4.9 ppm (Supplementary Tables S5, S6). Among the four stereoisomers of IHP, the 4 equatorial/2 axial configuration of neo-IHP in native grasslands (1.1%) was significantly higher than that in annual croplands (0.6%). The unidentified peak at 4.9 ppm, which was significantly higher for tame and native grasslands, may be the 4 axial/2 equatorial configuration of neo-IHP (Turner et al., 2012), but this could not be confirmed with spiking. There were no significant differences among other land use types for the other IHP stereoisomers (Supplementary Table S6), or for the other P<sup>o</sup> compound classes and identified species, although some of them showed higher proportions in grasslands than other land uses (**Table 2** and Supplementary Table S6).

Phosphatases play a key role in catalyzing the hydrolysis of P<sup>o</sup> to release orthophosphate for plant uptake. The activity of acid phosphomonoesterase, produced by plants and microbes (Tabatabai, 1994), was significantly lower in soils from annual cropland than other land use types (Supplementary Figure S4). There were no significant differences among land use types for alkaline phosphatase, which is produced by microbes only, and the activity of this enzyme was lower

fmicb-09-01643 July 23, 2018 Time: 16:4 # 5


than that of acid phosphatase for all but annual cropland soils. Phosphodiesterase activities were lowest for land use types compared to phosphomonoesterase activities, and were significantly lower in annual cropland soils than roadside soils.

#### Phosphorus K-Edge XANES Spectroscopy

Supplementary Figure S5 shows the K-edge XANES spectra of P standards in this study, where similar shapes with both the white line peak (ii) and oxygen oscillation peak (v) are observed. In addition, these P standards exhibited fingerprinting features that allow for identification and quantification of different P species in soil samples. For example, all Ca-associated P (Ca-P) standards exhibit a post-edge shoulder (iii) that is sharper for compounds containing many Ca atoms [HAP and Ca3(PO4)2] than for monocalcium phosphates [Ca(H2PO4)<sup>2</sup> and CaHPO4], and another signal (iv). Iron phosphate (FePO4) displayed a pre-edge feature (i). Aluminum phosphate (AlPO4), without the pre-edge feature (i) seen for FePO4, shows a post-edge feature (iv) similar to Ca-P. These features agree well with previous reports (Beauchemin et al., 2003; Lombi et al., 2006; Prietzel et al., 2013). Spectra of native and grassland soils (**Figures 2C,B**), rather similar across locations, resembled the spectra of IHP (Supplementary Figure S5). As the K-edge XANES spectra of P<sup>o</sup> compounds are mostly featureless and hard to differentiate, the contribution of IHP is interpreted as the presence of all possible organically bound P rather than specifically to IHP in the samples. Roadside soils showed spectra with a broad post-edge feature and slight shoulders (iii and iv, **Figure 2D**), demonstrating the presence of Ca-P in these soils. In contrast, spectra of cropland soils showed more variation than those of other land uses (**Figure 2A**).

The LCF for each soil sample, shown in Supplementary Figure S2 and Supplementary Table S2, indicated that IHP was present in all soil samples, in greater proportions for the native (92%) and tame (82%) grasslands than that for the roadside soils (50%, **Table 3**). The roadside soils from all four locations also contained some Ca-P in the form of tricalcium phosphate (TCP, 42–54%), which was absent from the native grassland samples, the tame grassland samples, and two cropland samples (Supplementary Table S2). Additionally, small amounts of P (<17%, Supplementary Table S2) in the forms of dibasic calcium phosphate (DCP), monobasic calcium phosphate (MCP), hydroxyapatite (HAP), or FePO<sup>4</sup> were also fitted in some samples. This indicated that these P species may be present in the soils, but there were no significant differences of these P species among land uses types (**Table 3**).

#### Metagenomic Analysis

Land use had a strong effect on the taxonomic composition of the soil microbial community. Although the archaeal community did not differ, the bacterial and fungal communities significantly differed among the land use types (Supplementary Table S7). Tame and native grassland soils harbored similar bacterial and fungal communities, whereas annual cropland and roadsides clustered separately in the principle coordinate analysis

fmicb-09-01643 July 23, 2018 Time: 16:4 # 6

C denotes a correction for degradation

 products.

(PCoA, Supplementary Figure S6), indicating they had distinct communities.

The microbial genes coding for various P cycling processes were broadly grouped into six functional categories based on Bergkemper et al. (2016), and these included phosphoesterase, phytase, phosphonate degradation, inorganic phosphate solubilizing, P transporter, and regulation of phosphate starvation genes (**Table 4**). Taxonomic classification of the P cycling genes revealed that the majority of the reads (76.1%) were assigned or classified as bacteria, 0.9% archaea, 0.8% eukaryota, and the remaining 22.1% of the reads were unclassified (**Figure 3**). Actinobacteria and Alphaproteobacteria were the most abundant bacterial classes but did not significantly differ among the land use types. Significant differences were observed for the Betaproteobacteria, Deinococci, Deltaproteobacteria, Gammaproteobacteria, Gemmatimonadetes, and Eukaryota, all of which were the most abundant in annual cropland soils and least abundant in native grassland soils (**Figure 3**). Overall, annual cropland soils had the highest number of reads assigned to P cycling coding genes; native grassland soils had the lowest.

The permanova results showed a significant difference in the composition of the P cycling bacterial communities and composition of the P cycling functional genes among the four land use types (Supplementary Table S7). The PCoA (Supplementary Figure S7A) revealed a similar clustering of the P cycling bacterial communities compared to the total bacterial community (which included taxa that were not linked to P cycling functional genes) among the land use types, with roadside communities being the most distinct and dissimilar to native and tame grassland communities. However, PCoA of the P cycling gene composition revealed that native grassland and annual cropland soils had the most dissimilar P cycling gene compositions, with roadside and tame grassland soils being intermediary (Supplementary Figure S7B).

The abundance of most P cycling genes significantly differed among the land use types (**Table 4**). Genes coding for phosphoesterase enzymes were the most abundant in annual cropland and roadside soils, and least abundant in native grassland soils. Alkaline phosphatase (phoD, phoA, phoX) and glycerophosphoryl diester phosphodiesterase (upgQ) were the most abundant phosphoesterase enzyme coding genes for all land use types, possibly indicating that these enzymes have a higher capacity for P mineralization in the soils from this region compared to acid phosphatase and phytase enzymes. Roadside soils had the highest abundance of genes coding for phosphonate degradation enzymes, and this was particularly evident of genes coding for enzymes involved in the carbonphosphorus (C-P) lyase core complex (phnG, phnH, phnI, phnJ, phnM). Genes coding for inorganic phosphate solubilizing enzymes (ppa, ppx, ppk, gcd) were the most abundant group of P cycling genes for all land use types. Annual cropland and roadside soils appeared to have the highest inorganic phosphate solubilizing capacity based on the abundance of these genes, and native grassland soils the lowest capacity. Annual cropland soils also had the highest abundance of genes coding for P transporter subunits (phosphate-specific transporter, phosphonate transporter, and glycerol-3-phosphate transporter) and genes regulating phosphate starvation (phoB, phoR, phoU). Native grassland and tame grassland soils exhibited a lower phosphate uptake and regulation capacity based on the lower abundance of these genes.


TABLE 3 | Phosphorus K-edge XANES fitting results showing the relative percent of each phosphate species<sup>a</sup> in the studied soils under different land uses (means ± standard errors, n = 4)<sup>b</sup> .

a IHP, inositol hexakisphosphate; TCP, tricalcium phosphate; DCP, dibasic calcium phosphate; MCP, monobasic calcium phosphate; HAP, hydroxyapatite. <sup>b</sup>Values in each column followed by the same lowercase letters are not significantly different according to LSD (P < 0.05).

Supplementary Figure S8 shows the taxonomic composition (at the class level) of each P cycling gene. There are similarities in the taxonomic composition of genes that are functionally related. For example, most genes coding for the C-P lyase multienzyme complex are primarily from Alphaproteobacteria and from Actinobacteria and Betaproteobacteria to a lesser extent. The genes coding for the phosphate-specific transport subunit (pstA, pstB, pstC, pstS), phosphonate transporter subunit (phnC, phnD, phnE), and glycerol-3-phosphate transporter subunit (ugpA, ugpB, ugpC, ugpE) each had similar taxonomic compositions among the genes that make up their respective subunits. In contrast, other genes coding for similar functions such as alkaline phosphatase (phoA, phoX, phoD) and acid phosphatase (phoN, aphA, K01078) had higher variability in their respective taxonomic compositions.

#### Redundancy Analysis

Redundancy analysis (RDA) was used to identify the most important P chemistry predictors of the P cycling bacterial community composition (**Figure 4A**) and P cycling gene composition (**Figure 4B**). Similar P chemistry predictors were identified for both the bacterial community and gene compositions with a few exceptions. Overall, NMR (myo-IHP and IHP), enzyme activity (acid and alkaline phosphatase), Olsen P, and pH were significantly correlated with the composition of both the bacterial and functional gene compositions. In contrast, P-XANES species exhibited a limited relationship as only TCP was identified as a significant predictor of the P cycling bacterial community composition. The P chemistry predictors were able to explain a high proportion of the P cycling bacterial community and functional gene composition as the adjusted r 2 -values for each RDA was 0.34 and 0.75, respectively.

# DISCUSSION

The results by the ignition method, NMR, and XANES consistently indicated that native and tame grasslands had higher total P<sup>o</sup> proportions among the investigated land use types. In grazed grasslands, the majority of the P (∼85%) taken up in the form of orthophosphate by plants is returned to the soil in the form of P<sup>o</sup> through dung, providing a significant P<sup>o</sup> stock (Nash et al., 2014). Specifically, differences among land use types were significant for monoesters, principally total IHP and the neo-IHP stereoisomer. Higher proportions of monoester P in grasslands than cultivated soils have been reported previously (Condron et al., 1990; Stutter et al., 2015), although these studies did not correct for diester degradation or identify specific P forms. Using an improved method of P-NMR spectra interpretation to identify specific P forms, we found a significantly higher abundance of neo-IHP in native grasslands and an unidentified monoester species resolved at 4.9 ppm, which might also be neo-IHP (Turner et al., 2012), in tame grasslands compared to croplands. The abundance of chiro-IHP was extremely high in the tame pasture soil from one location, but the overall abundance of either of the two configurations of chiro-IHP was not significantly different among land use types. The specific origins and function of chiro-IHP and neo-IHP in soils are unknown, but they are thought to be synthesized by soil microbes (Turner et al., 2002, 2014; Giles et al., 2011). The higher abundance of these stereoisomers and lower ratio of myo-IHP:other IHP forms (**Table 2**) in grassland soils indicates that microbes are likely responsible for the higher accumulation of IHP compared to annual cropland and roadside soils.

The higher abundance of neo-IHP in grassland soils may be linked to reduced P turnover or cycling in these land use types. For example, Young et al. (2013) reported significantly higher neo-IHP concentrations under conditions (i.e., poor drainage) that constrain microbial P cycling. In the present study, native and tame grassland soils had the lowest abundance of genes coding for most P cycling processes (except phosphonate degradation) compared to other land use types, which may be an indicator of reduced P cycling capacity. The accumulation of neo-IHP in the grassland soils of this study is consistent with the results of Young et al. (2013), supporting their speculation that this IHP stereoisomer accumulates when microbial P cycling is reduced.

The native grassland soils of this study had a greater abundance of the 4-phytase gene (appA) than cropland soils, although there were no significant differences among land use types for the more abundant 3-phytase genes (K01083). Differing from other P<sup>o</sup> mineralization processes, phytases may not be controlled by the PhoBR two component regulatory system, but likely respond to the presence or absence of phytate in the soil (Lidbury et al., 2016). However, the ability of these genes to degrade other stereoisomers besides phytate (myo-IHP) is unknown, and previous research suggests neo-, chiro-, and

TABLE 4 | Abundance of microbial genes associated with phosphorus cycling in soil.


Mean values (n = 4) represent counts per million of mapped reads for each land use type. Values within a row that have different letters indicate a significant difference between land use types (P < 0.05).∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.

scyllo-IHP are more resistant to phytase hydrolysis than the myo-IHP stereoisomer (Turner et al., 2012). It has been suggested that microbes synthesize these stereoisomers as a potential strategy to preserve phosphorus from competing organisms under limiting P conditions (Turner, 2007). Contrary to the anecdotal evidence that P<sup>o</sup> mineralization in pastures might be high (Nash et al., 2014), our results demonstrated that the abundance of most P cycling genes was lowest in native

grasslands, except for the 4-phytase gene. This suggests that P<sup>o</sup> turnover in grasslands from this region may not be as active as other more intensive land use types, allowing P<sup>o</sup> to accumulate or become immobilized in soils via microbial processes. This is particularly relevant in grassland soils with low P<sup>i</sup> availability (Bünemann et al., 2012). Given the lower degree of disturbance in grasslands as a long-term stable ecosystem (Cade-Menun et al., 2017a), the transformation of P<sup>o</sup> in grasslands is likely more tightly regulated than arable systems and driven by intrinsic soilplant-microbial cycling demands for P. It is clear that P<sup>o</sup> serves as a substantial reserve of P for plant nutrition in grasslands. As such, further investigations linking the supply of orthophosphate from these P<sup>o</sup> species with plant needs warrants further studies (Nash et al., 2014).

In annual croplands, P<sup>o</sup> returns to soils are interrupted at crop harvest, which may account for the limited P<sup>o</sup> accumulation in this study. However, enhanced mineralization of P<sup>o</sup> by microbemediated activities due to low soil Olsen-P cannot be excluded, since annual croplands had the highest total abundance of genes coding for phosphoesterase and phytase enzymes. This contradicts the lower acid phosphatase and phosphodiesterase activity we observed in annual cropland soil despite the higher abundance of genes coding for these enzymes (phoN, ugpQ) compared to the other land use types. One explanation for this observation may be linked to the pH conditions of the enzyme assays as this can affect the relationships between gene abundance and enzyme activity (Fraser et al., 2017).

The high abundance of P cycling genes suggests a greater capacity for P transformation in cropland soils of this study. This may be a strategy utilized by the soil microbial community to adapt to the high temporal variability in P forms and availability (Hedley et al., 1982; Bainard et al., 2016) associated with the intensive practices of annual crop production (e.g., fertilization, tillage, harvest, weed management and fallow). There was considerable variability in crop management practices among the annual cropland sampling sites, including fertilizer and P-containing herbicide inputs (Cade-Menun et al., 2017a), which may have contributed to the high site variance in P<sup>i</sup> species and the insignificant differences among the four land use types. Additionally, the relatively higher TCP in the annual croplands than grasslands may arise from tillage, because the surface of annual croplands could be replenished by the deeper soil containing high levels of carbonates after repeated tillage. The higher abundance of P<sup>i</sup> solubilizing genes in annual cropland soils could indicate a higher capacity for the microbial community to access these P<sup>i</sup> forms compared to the grasslands. Interestingly, glyphosate was commonly used in three of the four annual cropping sites, but these soils had the lowest abundance of genes coding for the polypeptides that make up the core complex or reaction of the C-P lyase pathway, which is one of the primary pathways for the catabolism of phosphonates (Hove-Jensen et al., 2014). However, all polypeptides involved in the C-P lyase pathway (phnCDEFGHIJKLMNOP) are required for the utilization of glyphosate as a phosphate source (Chen et al., 1990; Hove-Jensen et al., 2014), including those that are more abundant in annual cropland soil (phosphonate transporters phnCDE, aminoalkylphosphonate N-acetyltransferase phnO, and phosphoribosyl cyclic phosphodiesterase phnP). Glyphosate has a distinctive peak in P-NMR spectra (Cade-Menun, 2015), which was not detected in the cropland soils, and the concentrations of phosphonates in general were not significantly greater in cropland soils than other land use types in this study. Further investigation is warranted to understand the factors controlling the degradation of agricultural compounds such as glyphosate in the soils of this region.

Annual crop production practices appear to have an impact on the PhoBR two component regulatory system based on the higher abundance of P-starvation-inducible genes compared to the other land use types. This is consistent with the low Olsen P-values for the cropland soils. This system regulates several important P cycling processes controlled by genes coding for phosphoesterase, phosphonate degradation, inorganic phosphate solubilizing, and P transporter enzymes (Furtwängler et al., 2010; Lidbury et al., 2016). Bergkemper et al. (2016) suggested that a high abundance of P-starvation-inducible genes can enable microbial communities to utilize alternative P forms

under P limiting conditions. This may be more important in annual cropland soils compared to perennial grasslands, because grassland ecosystems likely experience fewer disturbances and fluxes in P availability.

Roadside soils differ from the other land use types because they are disturbed environments that have high variability in terms of soil chemical and physical properties and soil moisture gradients due to runoff from adjacent roads and croplands and nutrient removal via haying and mowing (Dai et al., 2013; Cade-Menun et al., 2017a). Roadside soils distinguished from other land use types in this study based on having the highest Olsen P and TCP concentrations and low Po. In a separate study, Cade-Menun et al. (2017a) showed that roadside soils collected from the same sampling locations as the current study at the 0–7.5 cm depth had the greatest percentage of clay, the highest soil pH and higher total C relative to the other land use types. There were generally no differences among land use types at lower depths; thus, sampling at 0–30 cm may have obscured some differences for the current study. Roadsides in Saskatchewan are subjected to snowmelt runoff, in which dissolved reactive P accounted for 97– 100% of the dissolved total P loss from croplands and pastures (Cade-Menun et al., 2013), which would account for the high Olsen P concentrations in these roadside soils. These roadside soils would also experience deposition of clays during wind erosion (Cade-Menun et al., 2017a). This, with the neutral pH (7.1) and high NH4OAC-extracted Ca are expected to facilitate the formation of TCP in these roadside soils (Sato et al., 2005). From the microbial perspective, the relatively high abundance of P cycling functional genes along with high phosphatase activity and low abundance of P<sup>o</sup> indicate that microbe-mediated P dynamics in roadsides are active but highly variable.

To the best of our knowledge, this is the first study to combine state-of-the art spectroscopic methods for P chemistry with shotgun metagenomics to provide an in depth evaluation of the dominant mechanisms involved in P cycling in soils with different land uses. Grasslands had high abundances of monoesters, principally IHP stereoisomers, and high acid phosphomonoesterase activities but lower abundance of genes coding for P cycling processes. In particular, the significantly higher proportion of neo-IHP in the native grasslands than in croplands confirms the important role of microbes in P<sup>o</sup> transformation in grassland soils. In contrast, croplands showed the largest variance of P speciation among the land use types, illustrating the crucial role of specific field management practices within croplands. Furthermore, the conversion of native grassland to annual crop production appears to increase the abundance of P-cycling genes, which may be required in soils that are under intensive management practices. Future studies are warranted to design tailored agronomic practices that directly facilitate functional genes and microbial communities for certain P cycling processes (e.g., P<sup>o</sup> mineralization) to optimize P-use efficiency. Roadside soils had the highest Olsen-P due to inputs from erosion and runoff, and had high proportions of TCP, reflecting clay inputs, neutral pH and high exchangeable Ca concentrations. The RDA results demonstrated that IHP by NMR, enzyme activity, Olsen-P, and pH were important P chemistry predictors of the P cycling bacterial community composition and functional gene composition.

### AUTHOR CONTRIBUTIONS

LB, BC-M, MS, KL, and CH designed and conducted the study. The NMR analysis was performed by CL and interpreted by BC-M. JL and JY conducted the synchrotron experiments. JL, JY, and YH analyzed the data. LB and JT performed the soil microbial and bioinformatic analyses. JL, LB, and BC-M wrote the manuscript with inputs from all authors. All authors read and approved the manuscript.

#### ACKNOWLEDGMENTS

This project was funded by the Agriculture & Agri-Food Canada A-Base project 1154 (Impact of Land Use on Soil Functional

Diversity and Nutrient Cycling in Prairie Ecosystems). We gratefully thank staff and students at the SCRDC for assistance with sample collection and analysis. The NMR analysis was conducted at the Stanford Magnetic Resonance Laboratory at Stanford University, and the metagenomic analysis was done at the Biotechnology Research Institute of the National Research Council of Canada. Synchrotron measurement was carried out at the SXRMB beamline of the Canadian Light Source, which is financially supported by the Natural Sciences and Engineering Research Council of Canada, the National Research Council of Canada, the Canadian Institutes of Health Research, the Province

#### REFERENCES


of Saskatchewan, Western Economic Diversification Canada, and the University of Saskatchewan. We also wish to acknowledge Compute Canada for access to the University of Waterloo High Performance Computing (HPC) infrastructure (Graham system).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.01643/full#supplementary-material




Young, E. O., Ross, D. S., Cade-Menun, B. J., and Liu, C. W. (2013). Phosphorus speciation in riparian soils: a phosphorus-31 nuclear magnetic resonance spectroscopy and enzyme hydrolysis study. Soil Sci. Soc. Am. J. 77, 1636–1647. doi: 10.2136/sssaj2012.0313

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Her Majesty the Queen in Right of Canada, as represented by the Minister of Agriculture and Agri-Food Canada. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Components of Phosphorus Loss From Agricultural Landscapes, and How to Incorporate Them Into Risk Assessment Tools

#### Keith Reid<sup>1</sup> \*, Kimberly Schneider<sup>1</sup> and Brian McConkey<sup>2</sup>

<sup>1</sup> Science and Technology Branch, Agriculture and Agri-Food Canada, Guelph, ON, Canada, <sup>2</sup> Science and Technology Branch, Agriculture and Agri-Food Canada, Swift Current, SK, Canada

Phosphorus (P) loss to surface freshwater is a key driver of environmental degradation, including blooms of both harmful (e.g., microcystis) and nuisance (e.g., cladophora) algae, along with the development of hypoxic zones that could significantly impact fish habitat. Mitigating P losses from agricultural land will require a detailed understanding of the forms (particulate versus dissolved, and chemical speciation), sources (soil erosion, desorption of soil P, dissolved P from fertilizer or manure application, or release from frozen vegetation) and transport pathways (surface runoff, or subsurface runoff through tile drains). This paper describes each of these components in detail, and discusses how this can guide the adoption of appropriate beneficial management practices to effectively reduce P losses. Further, it describes how this component structure has been incorporated into the Canadian national Indicator of Risk of Water Contamination by Phosphorus (IROWC-P) as an example of a risk assessment tool.

#### Edited by:

Chris Thomas Parsons, University of Waterloo, Canada

#### Reviewed by:

Yi Zhang, University of Maryland, College Park, United States Ryuichiro Shinohara, National Institute for Environmental Studies, Japan

\*Correspondence:

Keith Reid keith.reid@canada.ca

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Earth Science

Received: 26 January 2018 Accepted: 20 August 2018 Published: 05 September 2018

#### Citation:

Reid K, Schneider K and McConkey B (2018) Components of Phosphorus Loss From Agricultural Landscapes, and How to Incorporate Them Into Risk Assessment Tools. Front. Earth Sci. 6:135. doi: 10.3389/feart.2018.00135 Keywords: phosphorus, nutrient loss, risk assessment tools, component P index, P source, P transport

# INTRODUCTION

Phosphorus (P) loss to surface freshwater is a key driver of environmental degradation (Sharpley et al., 2003; Jarvie et al., 2013; Scavia et al., 2014), including blooms of both harmful (e.g., microcystis) (Conroy et al., 2014; Steffen et al., 2014; Simic et al., 2017) and nuisance (e.g., cladophora) algae (Auer et al., 2010; Depew et al., 2011; Howell and Dove, 2017), along with contributing to the development of hypoxic zones that impact fish habitat (Bouffard et al., 2013; Scavia et al., 2014). While agricultural runoff is not the only source of P loading to surface water, it is significant in many areas and is implicated as the dominant source to some of the most heavily impacted waters (Sharpley et al., 2003; Michaud et al., 2004; Joosse and Baker, 2011; Bunting et al., 2016).

Given the importance of agricultural runoff in P loading to surface water, there is great interest in finding ways to mitigate these losses (Haygarth et al., 2005; OLEPTF, 2010; Reutter et al., 2011; Osmond et al., 2012a; Kerr et al., 2016). The success of these mitigation activities depends on how well our understanding of the source and transport processes for P represent what is actually happening in an individual field, farm or catchment (Gburek et al., 2002; Osmond et al., 2012b; Sharpley et al., 2012; Radcliffe et al., 2015). Unfortunately, this linkage is often less than perfect for a number of reasons. For example, models may be adapted from other jurisdictions that are dominated by different processes; our understanding of the underlying processes for the transport

**22**

of P may be flawed, or based on outdated research; or the models chosen may be biased by tightly held cultural, political or economic preferences (Kleinman et al., 2015b).

This paper will attempt to outline our current understanding of the dominant sources and transport pathways for phosphorus loss from agricultural land under varying inherent conditions and management practices, and how the relative importance of each may be assessed. The Indicator of Risk of Water Contamination by Phosphorus (IROWC-P) will be discussed as an example of one type of assessment tool, including describing the areas for future improvement.

# THE SOURCE × TRANSPORT PARADIGM FOR UNDERSTANDING P LOSS

The earliest P indexes considered sources of P independently from the risk of transport (Lemunyon and Gilbert, 1993) and did not always assign risk correctly if there was, for example, a large source of P with no transport risk. Gburek et al. (2000) pointed out this flaw, and suggested an alternative model which determined the potential P source for loss, and multiplied it by a transport modifier to assess the risk of P loss to surface water. This forms the basis for our understanding of P losses, where water movement is needed before any P transport can occur. The unfortunate corollary is that some models assume that any water leaving the landscape will carry the same concentration of P, which ignores the differences in the ways that P is released into water traveling by different pathways, the opportunities for mitigation of P traveling through different pathways, and the ways that different forms of P will be transported.

This weakness is addressed by using a component model to understand and predict P losses, where the combinations of P sources and transport pathways are assessed individually, and then summed to determine the total risk of P loss from the field or region being assessed. The focus in this paper is on P transport by water, although atmospheric deposition of P can also occur, and may be a significant fraction of P inputs to some smaller water bodies adjacent to sources of P (Anderson and Downing, 2006), and this could be included in as a component in a P loss model. The factors that make up each of the components are discussed in the following sections.

# PARTICULATE VERSUS DISSOLVED P

Phosphorus exported from terrestrial systems is a combination of dissolved P, and P that is bound to soil as particulate P. Dissolved P is predominantly phosphate ions [also identified as dissolved reactive P (DRP) due to its reactivity with molybdate to form a blue colored complex (Joosse and Baker, 2011)], but it can also contain P sorbed to colloidal particles, organic P compounds (Heathwaite et al., 2005), and non-reactive mineral forms including polyphosphates and phosphonates (Turner and Newman, 2005; Weihrauch and Opp, 2018). The phosphate ion is extremely reactive, and therefore tends to form insoluble or slightly soluble compounds if a suitable counter ion is available (Sharpley, 1995). Most P in the soil is in the particulate form, as precipitates of iron, aluminum or calcium phosphates, bound to soil minerals (clay, calcite, aluminum hydroxides, etc.), or occluded within soil granules (Wang et al., 2010). This has led to the misconception that controlling soil erosion will effectively control P export from agricultural land (Baker et al., 2014), but recent developments have shown that a significant portion of P losses can be in the dissolved form (Baker et al., 2007; Joosse and Baker, 2011).

This is relevant to water quality because of the relative availability of each P form to freshwater algae. Dissolved P, especially PO<sup>4</sup> <sup>3</sup>−, is immediately available to algae, and is absorbed within a time frame of minutes to hours (Barlow-Busch et al., 2006), so many lake water analyses ignore the dissolved fraction completely since the majority is in the algal biomass (Lin and Guo, 2016). The particulate fraction is, generally, more slowly available, as the recalcitrant compounds gradually dissolve. This proportion will vary with the soil chemistry, and with the amount of P in the soil. Estimates of the proportion of total particulate P that is bioavailable vary widely, with Sharpley et al. (1991) reporting a range from 0 to 95%, and Young et al. (1985) reporting a range from 0 to 70% for sediments in the Great Lakes basin. The most frequently used estimates, however, are in the range of 10% (Fang et al., 2002) to 30% (Sharpley and Smith, 1993), which is within the range reported for five lower Great Lakes tributaries by DePinto et al. (1981). A full discussion of the complexity of various P forms in soils and sediments is beyond the scope of this paper, but the topic has been well summarized by Condron and Newman (2011) and Weihrauch and Opp (2018).

The biological response of lakes and streams to P inputs will depend on the total amount of bioavailable P. This can be calculated as the total of dissolved P and the bioavailable portion of particulate P.

# P SOURCES

One of the limitations of P risk assessments is focusing on a limited number of potential sources, which may lead to ignoring sources that are relevant to the conditions in a given area. This is a particular danger if P indexes are adopted from other regions with different dominant P sources. While it may be appropriate to focus on particular sources, this should only be done after an objective assessment of the relative importance of each source within the area of interest, rather than adopting the assumptions made in different geographies.

# Particulate P From Eroded Soil

In many environments, the particulate P fraction represents the majority of total P leaving a field, although it may not represent the largest amount of bioavailable P. Nonetheless, it is important to consider in any assessment of P loss.

The factors affecting particulate P loss include the quantity of eroded soil, and the P concentration within that sediment. It is difficult to measure soil loss directly, but there are a number of tools available to estimate soil erosion from soil, landscape,

climate, and management factors. These are generally based on the Universal Soil Loss Equation (USLE), which is a mixed model that combines both source and transport factors to predict soil delivery to the bottom of a 22.7 m slope (Foster et al., 2003; Kinnell, 2008). This helps to explain why soil erosion is often included as a transport factor rather than a source factor. Variations on USLE, like RUSLE (Renard et al., 1995) or WEPP (Flanagan et al., 2007), use similar equations to estimate the quantity of soil detached but also account for additional landscape factors to improve estimates of P delivery to the edge of field.

Two different methods have been used for estimation of the particulate P concentration within the soil. Some studies have shown a correlation between the soil test P (STP) of the soil and total P concentration (van der Perk et al., 2007; Withers et al., 2009). The second method is to assume that particulate P concentration is a constant fraction of the soil, which is supported by data showing a weak relationship between STP and PP (Uusitalo et al., 2003; Reid, 2011; Glæsner et al., 2013; Borda et al., 2014). It is likely that this proportion varies with mineralogy as well as organic matter content and composition, so there may be significant differences between regions that are independent of STP.

Complicating the estimation of PP losses are the issues of P stratification and P enrichment. P stratification is the accumulation of elevated levels of P at the soil surface, either by the deposition of plant residues on the soil surface or by the surface application of fertilizer or manure (Smith et al., 2017). P stratification is an important consideration because the runoff water interacts with a very shallow layer of soil at the surface (Sharpley, 1985). P enrichment refers to the selective erosion of sediments with a higher P concentration than the bulk soil during small erosion events (Sharpley, 1980).

Some risk assessment tools are attempting to estimate the bioavailable fraction of PP, rather than the total, to better predict the biological consequences of the P loss. This can be estimated as a constant proportion of the PP (see previous section), or related to the STP (Sharpley et al., 1992; Sharpley and Smith, 1993). Ellison and Brett (2006) showed the proportion of bioavailable particulate P (BAPP) ranging from 12 to 29% in runoff from rural areas in Washington State, United States. DePinto et al. (1981) found that the BAPP of suspended sediments in tributaries of Lakes Erie and Ontario were relatively consistent within tributary samples, but varied between tributaries with a range of 6.1–35.8% of total sediment P.

#### Dissolved P Desorbed From Soil

When rain or snow-melt interacts with the soil surface, a small part of the P contained in that soil will dissolve and be transported with runoff water, either across the soil surface in surface runoff or diverted vertically through macropores to tile drains. The exact proportion of soil P that desorbs will vary with P content of the soil, and the soil mineralogy. A number of studies have shown that DP losses are related to the agronomic soil tests that are appropriate for that region, which provides a readily available tool for estimating the risk of DP losses (Vadas et al., 2005; Little et al., 2007; Wang et al., 2012, 2015). Some of these studies have shown greater accuracy when the STP is expressed as a degree of P saturation (DPS), calculated as the STP divided by the P sorption capacity of the soil. The importance of this increased accuracy for routine evaluations is unclear, however, since it would require soil analyses that are not included as part of most STP evaluations, and the increase in accuracy is modest relative to the range of STP values found under field conditions (Vadas et al., 2005; Wang et al., 2010, 2015).

The proportion of desorbed P has generally been assessed using artificial rainfall on small field plots or repacked soil boxes (National Phosphorus Research Project [NPRP], 2001; Wang et al., 2010). This consistent approach has been very good for assessing the relative difference between different soils and STP values, although there is evidence that it may be underestimating the concentration of P in runoff water from small rain events (Shigaki et al., 2007; Srinivasan et al., 2007; Dougherty et al., 2008; Dunkerley, 2017). It is speculated that the intensive rainfall in the rainfall simulators limits the opportunity for rainwater to interact with the soil surface due to the rapid runoff, and that the estimated proportions should be increased for light rain.

#### Dissolved P Desorbed From Applied Fertilizer

Fertilizer applied to the soil surface represents a highly soluble source of phosphorus, which can readily be mobilized in runoff water. Vadas et al. (2008) summarized the state of knowledge regarding interactions of fertilizer with soil and precipitation, noting that fertilizer granules dissolve over time with successive rainfall events. The resulting solution is more concentrated than what would result from the interaction of runoff water with soil, so the soil represents a net sink for the P desorbed from fertilizer. Using the assumption that the P in water which infiltrates into the soil will be retained there, the P losses from fertilizer can be estimated by using a distribution factor to account for the partitioning between runoff and infiltration (Vadas et al., 2008).

It is important to note that losses from applied P are only relevant from fertilizer that remains on the surface. Subsurface banding or incorporation by tillage should, theoretically, eliminate losses from applied P fertilizer, although the reality is that a small amount of the applied P remains on the surface. The degree of incorporation, and the lag time between application and incorporation, if known, should be considered in P loss risk assessments.

# Dissolved P Released From Applied Manure

Similar to fertilizer, manure is a highly available source of P for runoff, but has two key differences. The first is that manure contains a range of organic and mineral forms of P (Johannesson et al., 2017), so not all is immediately soluble although some compounds will degrade to release soluble P over time (Sharpley and Moyer, 2000; Vadas et al., 2011). The rate of release of these compounds can also vary with soil conditions, e.g., redox potential (Turner and Newman, 2005). The second is the physical form of the manure, as there may be immediate infiltration of liquid manure (Vadas, 2006), or reduced effectiveness of

incorporation due to the larger volume applied and the clumpy nature of solid manure (Vadas et al., 2007). The net result is that the immediate release from applied manure may be less than from an equivalent amount of fertilizer P, but the concentration of P in runoff water will not decline as quickly.

Dissolved P release from applied manure can be estimated from the total P application rate multiplied by the water extractable P proportion in the manure, which varies among livestock species (Kleinman et al., 2002, 2005, 2006), minus the amount of liquid manure that has infiltrated (Vadas et al., 2007). This dissolved P is then subject to a similar partitioning between runoff and infiltration as fertilizer, but with a different partitioning coefficient to reflect the slower rate but longer duration of P release from manure (Vadas et al., 2008, 2009).

### Dissolved P Released From Frozen Plant Residue

In areas where annual runoff is dominated by snowmelt, P that is released from frozen plant residue can represent a significant part of annual P losses to surface water (Elliott, 2013). It has been well established that the rate of P release from plant tissue is accelerated by freezing and thawing (Miller et al., 1994; Bechmann et al., 2005; Liu J. et al., 2013; Kirchmann and Wessling, 2017), although this has not translated into increased P losses to surface water under all conditions (Lozier et al., 2017). The most likely explanation for this seeming discrepancy is the difference in water infiltration under different climatic conditions. Where freezing conditions are followed by warm weather and rainfall, the P leached out of the plant residues will have to opportunity to interact with the soil and be adsorbed. In contrast, if there is little or no rainfall between freezing conditions and the onset of winter with snow accumulation, most of the P would remain in the residue in soluble form until it is carried off the field during snowmelt (Roberson et al., 2007).

This can be modeled by assuming the labile portion of the P in crop residue (roughly 50% for most residues) is converted to soluble forms according to an exponential decay function (Damon et al., 2014). This soluble P is then washed out of the residue by rainwater or by "steeping" in the water from snowmelt (Elliott, 2013), and partitions between infiltration and runoff in the same manner as dissolved P from fertilizer (Vadas et al., 2009). Modeling the P in alfalfa residue against actual weather conditions in southwestern Ontario (Harrow) and southern Saskatchewan (Swift Current) showed that most of the P was leached out of the Ontario residue by January with relatively little carried off the field, while significant amounts remained over winter in the prairie environment (Saskatchewan) and was exported in runoff during snowmelt (**Figure 1**).

#### P TRANSPORT PATHWAYS

As already noted, P movement in runoff cannot occur unless there is also water movement (Chardon and Schoumans, 2002). Most models of P loss account for surface runoff, but in regions with extensive tile drainage it is important to consider this pathway, as well (Reid et al., 2012; Jarvie et al., 2017). In some environments, subsurface lateral movement of P can contribute to P losses, but this requires preferential flow pathways in both vertical and horizontal directions, that discharge to surface water (Allen et al., 2012; McGrath et al., 2013; Kleinman et al., 2015a).

It should be noted that, although it is common to refer to the vertical movement of P as "leaching," that this is a different process from the leaching of compounds like nitrate. Both dissolved and particulate forms of P are significantly retarded by contact with the soil matrix, so the vertical movement of P to tile drains is through preferential flow pathways rather than through the soil matrix (Radcliffe et al., 2015). Leaching to deep aquifers is not assessed as part of P loss, both because dissolved P does not pose a direct human health risk, and because the conditions to support the biological response to P loading (i.e., algae growth) are not present in groundwater.

Transport of P through the air to surface water is ignored in most models, because, unlike N, there are no gaseous losses of P from the soil surface and it is considered to be a relatively minor contributor to P loading. There can, however, be significant generation of dust by soil disturbance that can then be deposited on surface water, which can contribute a significant part of P loading to water bodies that are relatively small (Anderson and Downing, 2006).

#### Surface Runoff

Surface runoff will be generated whenever water inputs from precipitation or snowmelt exceed the capacity of the soil to absorb that water. This exceedance may occur because the rate of precipitation exceeds the infiltration capacity of the soil (Infiltration excess runoff, or Hortonian flow), or because the soil is already saturated and cannot absorb any more water (saturation excess runoff). The amount of runoff, and therefore the risk of P loss in that runoff, is driven by the total amount of precipitation, the intensity of that precipitation, and the proportion of that precipitation that becomes runoff. Runoff from any landscape will be a combination of both processes, although the dominant process will depend on climate, topography, and soils.

Infiltration excess is the basis of runoff predictions using the SCS Curve Number approach (Garen and Moore, 2005; Singh et al., 2010). The underlying assumption is that areas with similar infiltration capacity (hydrologic response units, or HRUs), as determined by soil type, crop types, soil management and antecedent moisture, will all generate runoff at the same time when an intense rain event occurs. The corollary of this assumption is that the runoff generated will all run downslope to surface water, so the entire HRU is a source for P transport. Infiltration excess runoff will dominate in areas where soil permeability is low and precipitation is concentrated into intense storm events. Because of the influence of crop type and soil management (particularly tillage system), this form of runoff generation will change in response to management (Garen and Moore, 2005).

Saturation excess flow, in contrast, occurs in parts of the landscape where the water table is close to the surface of the soil. This may be due to topographic position (i.e., low elevation),

or to impermeable layers in (cemented horizons or shallow bedrock) or at the surface (compacted soils) of the soil. It is the basis for variable source areas for runoff (Gburek et al., 2006; Schneiderman et al., 2007; Dahlke et al., 2009). Runoff will generally occur from only a small part of the landscape, but can have a disproportionate effect because they are concentrated close to surface water. The proportion of precipitation that becomes runoff is related to the proportion of the landscape with saturated conditions; this parameter will not be affected by changes in crop type or soil management.

Both of these methods can be calibrated to provide good estimations of total runoff volume from a landscape, but the parts of that landscape that are assumed to be generating runoff can be quite different for each (Lyon et al., 2006). Models based on infiltration excess will predict most of the runoff coming from upland areas within a watershed, while saturation excess models will predict that most runoff comes from the lower reaches. Understanding the relative importance of each process within an area will be critical to the choice of P source mitigation efforts, if the goal is to reduce P losses to surface water.

Runoff from snowmelt is a special case, since large parts of the landscape may be either saturated or frozen. If this occurs, the differences between infiltration and saturation excess disappear. One of the major challenges, from both a modeling and a management perspective, is that there are also conditions where the snow melts gradually over a permeable soil so almost all of it infiltrates (Zhang, 2005; Dutta et al., 2017). This process has not been extensively researched, and the process of infiltration into unfrozen soils is not included in the Cold Regions Hydrologic Model (Pomeroy et al., 2007), although it does include a factor for infiltration into unsaturated frozen soils.

The impact of the interactions between regional climate variation and the seasonality of runoff generation on potential source areas for P loss can be illustrated by some examples. In the northern Great Plains, dry conditions dictate that there is virtually no runoff except during spring snowmelt (Liu K. et al., 2013; Liu et al., 2014; Mahmood et al., 2017), but there is also little opportunity for infiltration of P into frozen soils. Surface applied P during the fall or winter, or P from frozen plant residues, would represent a high risk of P loss. In contrast, runoff generation in much of south-eastern United States is dominated by infiltration excess from summer storms so losses from spring applied nutrients are more problematic than from fall applications (Osmond et al., 2012b; Woodbury et al., 2014). The Eastern Corn Belt is a transition zone, where runoff in the spring freshet mimics infiltration excess, while during the growing season there are seldom storms intense enough to exceed soil infiltration capacity so runoff is dominated by saturation excess (Gburek et al., 2006; Easton et al., 2008).

#### Tile Runoff

Tile drainage is widely used in many humid temperate agricultural areas to manage excess soil moisture and improve crop growth, with some of the most intensive areas of tile drainage in the southern parts of the Great Lakes basin (Tan and Zhang, 2011; Reid et al., 2012; King et al., 2015), the Mississippi River basin (Blann et al., 2009), and northern Europe (Uusitalo et al., 2007; Ulén et al., 2011). Tile drains have a marked effect on hydrology by reducing the volume of surface runoff, lowering the water table within the soil, and changing the timing and duration of peak runoff flows (Sloan et al., 2016), although the total runoff from an individual field may not increase very much

(King et al., 2014) if at all (Singh et al., 2006; Turunen et al., 2013). At the same time, the increased connectivity of the upper parts of the landscape and the greater conductivity of straightened streams and surface drains installed to receive tile water can increase the total water export from the landscape (Blann et al., 2009).

Although tile drains may have limited impact on total water movement off the landscape, they typically carry a significant part of the total annual runoff. King et al. (2014) measured tile contribution to stream flow on a silt loam to clay loam watershed in Ohio that ranged from 0 to 100% of total runoff in any month, with a mean annual value of 41%. This is similar to the 42% of total streamflow from tiles found by Macrae et al. (2007) on a watershed with coarse to medium texture soils in southern Ontario. Van Esbroeck et al. (2016) found a higher proportion of tile runoff at edge of field on three coarse to medium textured sites, ranging from 78 to 90% of total runoff. Tan and Zhang (2011) found that 96% of the total runoff, averaged over 5 years, was through tile drains on a clay soil in Essex County, ON, Canada.

Phosphorus export through tile drains in all of the above studies was less than the proportion of total flow, indicating that either movement of P from the surface to the tile drains was being partially impeded, or that the tile flow was a mix of surface runoff diverted to the tiles through macropores and water that had percolated through the soil matrix. Given the propensity of most soils in the tile drained regions to adsorb P (Richards et al., 1995; Wang et al., 2014), the latter mechanism seems much more likely, and has been proposed by a number of authors as an explanation for the P concentrations found in tile flow (Stamm et al., 1998; Chikhaoui et al., 2008; Van Bochove et al., 2010; Reid et al., 2012; Beven and Germann, 2013; Radcliffe et al., 2015; Jarvis et al., 2016). Further evidence that P transport to tile drains is occurring through macropores is that the proportion of particulate P in the tiles is similar to that in the surface runoff, when averaged over a year (Ball Coelho et al., 2010; Radcliffe et al., 2015; Pease et al., 2017), although there can be significant variability between individual flow events, and the application of fertilizer or manure can temporarily increase the proportion of dissolved P (Macrae et al., 2007). Within this framework, the main determinant of P movement to tile drains will be the partitioning of water reaching the tile drains between macropore flow, with a relatively high P concentration, and matrix flow which carries very little P. This partitioning is not well represented in most models of P transport (Radcliffe et al., 2015), but there has been some progress in empirical determination of macropore flow showing a strong influence of soil texture (Allaire et al., 2009; Eastman et al., 2010; Karahan and Er¸sahin, 2017). The textural influence may be directly affecting the propensity of the soil to form shrinkage cracks, or indirectly affect the population of anecic earthworms, both of which can contribute to macroporosity (van Bochove et al., 2007; Fox et al., 2008; Koestel et al., 2011; Jarvis et al., 2016). The general trend, where there have been direct comparisons, is for greater macropore flow in fine than coarse textured soils (Eastman et al., 2010).

Since the installation of tile drains reduces the volume of surface runoff, proper accounting for the contribution of tile drains to P transport requires estimation of the reduced surface runoff, as well as the macropore portion of tile flow as shown in **Figure 2**. Since the macropores with direct connection to the tiles are concentrated in a narrow strip over the tile drains, the spacing between the tile runs will influence the amount of P that could move to the tile. An additional complication is the extent of many subsurface drainage systems. Within an individual field, the catchment areas for surface runoff and subsurface runoff may correspond fairly closely, but modern tile drainage systems may encompass many fields and extend for a kilometer or more into the landscape. This represents a large increase in the connectivity of the landscape to surface water, and since there is essentially no mitigation of P in the tile flow and the drain outlets are generally directly into surface water, it can bypass the riparian buffers that can reduce the P entrained in surface runoff.

#### Sub Surface Lateral Flow

Preferential flow is generally considered to be vertical, but there are situations where there are horizontal preferential pathways, often where there is an impervious layer beneath the surface that impedes vertical flow and creates a perched water table. Water reaching these lateral flow paths, either through vertical macropores or by matrix flow through a soil that is saturated with P, can be carried horizontally to side-hill seeps (where the impervious layer reaches the soil surface along a slope) or into drainage ditches that intersect the impervious layer. This is the dominant pathway for P loss in the Delmarva Peninsula (Kleinman et al., 2007, 2015a).

# MITIGATION FROM EDGE-OF-FIELD TO EDGE-OF-WATER

The sources of P from agricultural land, and the transport factors that are calculated are operating within the same geography, so the end result is an estimate of P delivery to the edge of the field. If this coincides with the shore of a water body it is a reasonable estimate of risk, but more often the field is at some distance from surface water, and runoff must travel over other fields or buffer areas where mitigation of the P losses could occur. The actual risk to surface water quality, therefore, will depend on the distance from the edge of field to the stream, and on the nature of the soils and vegetation in this area.

Mitigation of P runoff can occur in several ways. Particulate P can be trapped or filtered by growing vegetation, particularly if there is a dense sward of grasses (Butler et al., 2006). Surface runoff can infiltrate into the soil in a buffer zone, which will reduce both dissolved and particulate P (Leguédois et al., 2008; Dosskey et al., 2010). There can even be some mitigation of P in runoff over a field, as the P concentration in the runoff water reaches equilibrium with the soil surface. All of these mechanisms will be most effective if there is sheet flow, to maximize the area of interaction. Concentrated flow will have enough energy to travel through the buffer without any appreciable reduction in P (Dosskey et al., 2002).

This mitigation is generally accounted for in models as a delivery ratio. This may be a single coefficient for buffers of a given width, or there may be more complex models that account for landscape types and vegetation (Li et al., 2011). Neither of these approaches has been well calibrated, and this remains an area requiring further work.

# THE COMPONENT MODEL STRUCTURE

The principle of the component model structure is that each source has a unique transport modifier, and the product of each source × transport calculation is summed to determine the total risk of P loss. If taken to this extreme, the component model would become complex to the point of being impractical. Fortunately, it is possible to lump together many of the sources to simplify the calculations. The transport of dissolved P is going to be similar, for example, whether it came from applied fertilizer or released from frozen vegetation.

Even with the rationalization of transport coefficients, there are many possible combinations of source and transport which could make the model unwieldy. Anyone developing a P Index using this approach should apply judgment as to which components are most important for the conditions within their jurisdiction, and which can be safely excluded. This will mean that indexes will vary between jurisdictions, but if both are using the component structure the results should still be comparable. This structure also encourages an objective assessment of which components should be included or excluded, and a framework for explaining why there are differences across jurisdictional boundaries.

Key considerations in the development of a component P index are consistency of units, and ensuring that the weighting of each of the individual components is correct. Ignoring either of these details can result in a model that is highly skewed because one of the components is generating results on a different scale than the others, and so is off by an order of magnitude or more. One advantage of the component structure is that, while the models are not complex enough to generate P loadings from individual storm events, the results should correlate with water quality measurements over the annual time scale and so can be validated as to whether they meet the goal of being "directionally and magnitudinally correct" (Sharpley et al., 2012).

We will discuss the IROWC-P model as an example of the component structure in the Section "IROWC-P as an Example of P Loss Risk Assessment."

# SETTING CATEGORIES FOR RISK OF P LOSS

The numbers generated by any index do not, without context, provide any useful information, so it is important to assign the results into categories. This will allow users to quickly determine if they are at a high or low risk. For risk indicators that cover a broad area, it will also facilitate visual representation of the data on a map. There are a number of ways to define risk categories,

but two of the most common are to relate the categories to environmental outcomes (or reduction targets), or to use the frequency distribution of the index results to base the categories on percentiles.

The idea of classifying risk categories by environmental outcomes is attractive, but it overestimates the model's ability to quantify the losses from the landscape. There is also a danger that the choice of thresholds, even if the model were accurate, would be set so that the highest level of risk would not exceed environmental outcomes, even though the majority of the catchment is already below that level. In this case, the categories could provide an unrealistically harsh perception of the performance of farmers in a region.

Another approach is to use the frequency distribution to separate the risk categories, either as even quartiles or quintiles, or using a skewed distribution of the percentiles. This latter approach can use the cumulative risk at each level to inform the setting of the categories, since a few very large scores at the top of the range can have a much greater impact than a large number of low scores. This focuses attention on the relatively few operations that contribute disproportionately to environmental impact, and allows for continuous improvement as the frequency distribution moves downward. It depends, however, on knowing the distribution of index scores. Since this is not possible before the use of the index is widespread, the distribution can be estimated by using Monte Carlo simulations as a proxy.

# IROWC-P AS AN EXAMPLE OF P LOSS RISK ASSESSMENT

Developing models to assess the risk of P loss from terrestrial systems is a balancing act between representing all of the source and transport processes as accurately as possible, the availability of input data at an appropriate level of detail to run a processbased model, and the complexity of the model, at a scale that is useful to the end user (Sharpley et al., 2002). One of the key challenges is that individual source and transport factors have been quantified through research, but there are few studies that account for the relative weighting of these factors in describing P losses at the field or landscape scale under varying soil types, topography, vegetation, P inputs and weather. Osmond et al. (2012b) found that P Indices developed using professional judgment about local conditions were more accurate at predicting runoff water quality than process based models. This may stem, in part, from imperfect representation of P loss processes in the model, or from over-parameterization where the errors associated with input variables propagate through the model to skew the final results (Vadas et al., 2013).

IROWC-P is one of the agri-environmental indicators developed by Agriculture and Agri-Food Canada (AAFC) to provide information on the status and trends of environmental performance by Canadian agricultural producers (Reid et al., 2016). Although this indicator operates at a regional rather than field scale – the basic spatial unit is a Soil Landscape of Canada (SLC) polygon, or 1:1,000,000 scale – the calculations used to estimate risk of P loss are parallel to those used in a field scale P Index. Each SLC polygon is treated as a virtual field, with input parameters to the model estimated from publicly available data. The version used in the last report was a multiplicative approach to P risk assessment, with a single transport coefficient, but it is currently being updated to a component model. The discussion in this paper will focus on the updated version, and on the algorithms used for determining risk of P loss but not the estimation procedures for the input parameters.

IROWC-P first determines the risk of P loss at the edge of field, and then applies a delivery ratio to estimate the proportion of this loss that reaches surface water. The source and transport factors for each of the components are summarized in **Table 1**.

# BioAvailable Particulate P

The developers of this indicator chose to express the results in terms of bioavailable P rather than total P to provide better information on the potential impact of P loss in different forms on water quality. The sediment loss to the edge of field is derived from the water erosion component of the AAFC Soil Erosion indicator (Lobb et al., 2016), and the bioavailable portion of this sediment is predicted from the STP levels according to the method of Sharpley and Smith (1993). A P enrichment factor is also included to account for the increased P concentrations in eroded sediment relative to the bulk soil in the field, particularly under small runoff events (Sharpley, 1980). The product of this equation is multiplied by a scaling factor to align the units of the outputs with the rest of the components in IROWC-P (Reid et al., 2016).

Soil erosion models combine sediment source and transport to predict delivery of sediment to the bottom of a slope or, in this case, to the edge of a field. A separate transport modifier is not required, except where tile drains are present. The tile drains reduce the amount of surface runoff, diverting that water into the tile drains, but also carrying the portion of the sediment that is carried with the portion of tile flow that has reached the tile through macropores rather than soil matrix flow. The partitioning of water between surface runoff and tile flow is predicted using the hydrology module of the DNDC model (Kröbel et al., 2010; Dutta et al., 2016; Guest et al., 2017), and the fraction of tile flow from macropores is derived from the probability of burrow flow (through anecic worm burrows) (Dadfar et al., 2010a) and crack flow (Dadfar et al., 2010b). In general, the macropore flow is highest for the fine and medium textured soils, and lowest for the coarse textured soils. Since there is compelling evidence that macropore flow represents diverted surface runoff, and that the ratio of dissolved to particulate P in macropore flow tracks that in surface runoff (Radcliffe et al., 2009; Reid et al., 2012), this partitioning of water into surface runoff, macropore tile flow and matrix tile flow will be used for all of the components.

# Dissolved P Desorbed From Soil

This component represents the release of dissolved P from the soil surface as it interacts with runoff water. Water Extractable P (WEP), derived from the STP values, is used as an indicator of the amount of P that could be desorbed from the soil surface. This is multiplied by an extraction coefficient to convert the TABLE 1 | Source and transport components in IROWC-P.

feart-06-00135 September 4, 2018 Time: 9:5 # 9


potentially desorbed P into actual dissolved P (Wang et al., 2010), and modified by a stratification factor based on tillage system to account for the build-up of P in the shallow surface layer that runoff water interacts with (Sharpley, 1985).

The transport factor for this component is the total volume of runoff, as predicted by the DNDC hydrology module (Dutta et al., 2016; Guest et al., 2017). This is assessed on an annual basis, since the pool of P available for desorption is very large relative to the actual losses and so does not vary over the year. Adjustments for tile drains are the same as for BAPP.

#### Dissolved P From Applied Fertilizer

Because the risk of P loss from applied nutrients is highest immediately after application and declines quickly as they interact with the soil (Vadas et al., 2008), both fertilizer and manure applications are assessed separately for the growing- and non-growing-seasons. Under Canadian climatic conditions, the partitioning of water between infiltration and runoff is markedly different between these two periods (Liu K. et al., 2013; Mahmood et al., 2017). The source factor for this component is the rate of fertilizer P applied, modified by an application factor to account for the proportion of fertilizer remaining on the soil surface. Incorporation has been shown to effectively shield the applied P from immediate losses (Sharpley, 1985), but there will always be a small amount remaining at the surface so the application factor estimates these proportions for different application systems.

The concentration of P in runoff water from applied fertilizer will be high relative to the P desorbed from soil, so the transport factor needs to account for the soil as a sink for this dissolved P. IROWC-P follows the method of Vadas et al. (2008), using a P distribution factor (PDF) multiplied by the ratio of runoff:precipitation to partition the applied P between adsorption by the soil and loss in runoff. The proportion of precipitation that runs off is much greater in the non-growing season under Canadian climatic conditions, so differentiating between P applications in the growing- versus non-growingseason is important for predicting risk of P loss. As noted above, the impact of tile drainage is predicted using the same algorithms as for BAPP.

#### Dissolved P From Applied Manure

The prediction of P losses from applied manure parallels that of applied fertilizer, with three exceptions. The source factor is based on the water extractable P in the manure, so only a portion of the total P applied is considered to be available for loss. The application factors are also modified to reflect the difficulty in achieving complete incorporation of the large volumes normally associated with manure application. On the transport side, the PDF is specific to manure (Vadas et al., 2004), reflecting the difference in the timing of P release from manure compared to fertilizer.

# Dissolved P From Over-Wintering Vegetation

This component only applies to the non-growing season, from the time of killing frost until the spring freshet prior to the next growing season. The source factor for this component is the labile, or easily degraded, portion of the P contained in the above-ground portion of plant residues at the time of killing frost in the fall. This is calculated as the plant biomass times the P concentration times the labile fraction, which is assumed to be 0.50 after Damon et al. (2014). An exponential decay function is used to estimate the conversion of labile P to soluble P within the plant biomass from freezing until the P is leached

out by rainfall. These calculations can be applied to crop residue remaining in the field post-harvest (in which case the harvest date is used as a proxy for plant maturity and the beginning of plant senescence) or for forage regrowth. The current version of the indicator is not assessing the P contributions from cover crops because of insufficient data regarding the area of cover crops, the biomass production and the P concentration, although it would be desirable to include this in future versions.

The transport of P from the frozen plant residues is in two steps, with the first being the leaching of the P out of the plant residue. It is assumed that rainfall must be >3 mm to initiate this process, or that the residue is in contact with meltwater from the accumulated snowpack, and that the dissolution of the soluble P in the residue follows a similar pattern to that of fertilizer P as described in Vadas et al. (2008), but with the duration of rainfall used by Vadas et al. (2008) replaced by the accumulated rainfall. The dissolved P that was then entrained in the runoff water was distributed between soil adsorption and runoff using the PDF for fertilizer (Vadas et al., 2009). Melt water runoff that occurs on frozen soils will carry much of the over-winter accumulation of soluble P from the field, because of limited infiltration.

#### Aggregate Risk of P Loss

Since the components are designed so that all of the outputs are at the same scale, the total risk of P loss at the edge of field is calculated by summing the values for each of the components.

# Delivery Ratio From Edge of Field to Surface Water

To assess the potential impact of agricultural P losses on surface water quality, it is necessary to estimate how much of the P leaving the edge of a field actually reaches surface water. The underlying assumption in assessing connectivity to surface water is that there is mitigation of P losses with greater transport distance to surface water. Processes that could contribute to this mitigation include:


These processes are not well quantified, and have seldom been included in P transport models beyond rudimentary estimates. Gburek et al. (2006), for example, estimated the proportion of the landscape that was contributing surface runoff from the proportion of precipitation that contributed to increased streamflow during storm events. Other models have simply assumed a delivery ratio based on the measured P loading at a watershed scale, relative to modeled or measured P losses within fields in the watershed, but without any understanding in the processes involved in P delivery from the field to the stream.

The AAFC Soil Erosion indicator (Lobb et al., 2016) includes calculations for sediment reaching surface water based on the stream density (defined as the kilometers of shoreline per square kilometer of watershed), and the landform type and slope within the watershed. This has been modified to assume the greatest mitigation of BAPP losses in overland runoff, less mitigation of dissolved P in surface runoff, and no mitigation of either BAPP or dissolved P in tile flow.

#### Additional Considerations

Phosphorus losses in sub-surface lateral flow are not included as part of the IROWC-P model. The conditions favorable for this type of P transport from Canadian agricultural land are considered to be rare enough that ignoring it will not introduce a significant source of error in the estimates of P loss.

One of the desired outcomes from this indicator is an assessment of the impact of changes in management on the risk of P loss from agricultural landscapes, but extreme weather events can completely eclipse the effect of management changes in any given year. This could be addressed by using average weather conditions in model runs, so that regional variations in climate are accounted for but annual weather is de-emphasized. The drawback to this approach is that responses to weather conditions non-linear and losses from areas with more variable weather may be underestimated. Calculations for IROWC-P are conducted using weather data from each year within the 30-year window that makes up the climate norm, and then the results averaged.

# GAPS AND NEXT STEPS

While the authors consider this revision of IROWC-P to be a significant advance in accurately representing the geographic distribution and temporal trends in the relative risk of P losses from Canadian agricultural land, there are still some areas in need of improvement. The individual components have been derived from a large body of scientific research, but not all of them have been validated for Canadian conditions. The aggregate values for edge of field losses are also in need of validation to ensure that the relative contribution from each of the components has been captured correctly. This model does not, nor was it intended to, provide a quantitative estimate of P loss from agricultural land during individual runoff events, so validation needs to be based on whether the relative P losses from contrasting fields matches the relative predicted risk of P loss, over an annual period or longer.

As previously mentioned, the delivery ratios from edge of field to surface water are an area of significant uncertainty. The scope of this issue means that it will require a large research effort to adequately address. The mandate to conduct this research, however, or the funding availability, often falls between the areas of responsibility of organizations with a primary focus on land management, and those focused on water quality impacts.

A perennial challenge for any regional or national scale indicator is the availability of input data to run the models. We do not know what each individual land manager is doing on their properties, so average or typical values are inferred from public data sources. In the case of IROWC-P,

these values are based on interpretation of data from the Census of Agriculture, supplemented by data from surveys (Reid et al., 2016), but it is recognized that the information about nutrient management practices is extremely limited. In the United States, the Conservation Effects Assessment Program (CEAP, 2011) has addressed this issue through detailed interviews with a stratified sample of farmers within a region, but this requires a large resource commitment, and while it provides a more complete picture of individual farmer activities there is limited opportunity to assess how well this represents the population as a whole. We can expect future improvements in some of the input data using remote sensing, which can detect topography and soil moisture, and enumerate areas of different crops or amount of soil cover, but this will be of little use for capturing data about management. Better methods of data collection will improve the accuracy of the model predictions, but they will need to meet the multiple requirements of protecting individual privacy, ensuring accuracy of the collected data, and being economical to collect and store.

The focus to this point has been on understanding the spatial variation in risk of P loss to surface water, and how it is changing over time. The indicator has been designed in such a way, however, that data for individual crops or farming systems can be extracted, although it will have to be at a coarser spatial scale to be meaningful. This creates the opportunity to work with commodity organizations to generate intensity based indicators of risk of P loss per unit of production. An additional benefit to this type of assessment might be the ability for individual farmers to compare the performance of their own farms to regional averages.

#### SUMMARY AND CONCLUSION

Phosphorus loss from a landscape is not a single process, but rather is the aggregate of a number of different source × transport interactions. Which of these components will dominate in a particular area, or on a particular farm, is a function of climate, soils, topography, and both long- and short-term management.

#### REFERENCES


The choice of options to mitigate these losses will, necessarily, be equally complex. It also renders useless the application of static coefficients to predict the effectiveness of mitigation practices, since the effectiveness of each practice will depend on the conditions where it is applied.

Progress is being made in developing risk indicators that address this complexity in a meaningful way, although much work remains to be done. IROWC-P has been presented as one example of how this could be approached, but other examples include models developed in North Carolina (N.C. PLAT Committee, 2005) and Wisconsin (Vadas et al., 2013). It is likely that selecting the best features from each of these component models for the individual conditions within a given jurisdiction will result in the most accurate representation of P loss risk.

#### AUTHOR CONTRIBUTIONS

The majority of the writing in this paper was contributed by KR, with significant additions from KS and BM.

#### FUNDING

Research and development activities described in this paper were funded as part of the Agriculture and Agri-Food Canada internal Research, Development and Transfer project funds.

#### ACKNOWLEDGMENTS

The authors would like to acknowledge the contributions of all of the scientists and professionals who contributed to the development of the Sustainability Metrics for Canadian Agriculture, as well as the programs which preceded it (Agri-Environmental Indicators for Canadian Agriculture, and the National Agri-Environmental Standards Initiative).

Sandusky and Cuyahoga rivers: the importance of bioavailability. J. Great Lakes Res. 40, 502–517. doi: 10.1016/j.jglr.2014.05.001


water use in temperate field cropping systems. Environ. Model. Softw. 80, 9–25. doi: 10.1016/j.envsoft.2016.02.014


feart-06-00135 September 4, 2018 Time: 9:5 # 12




potential phosphorus loss from Ontario soils. Geoderma 241–242, 51–58. doi: 10.1016/j.geoderma.2014.11.001


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Reid, Schneider and McConkey. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Challenges of Reducing Phosphorus Based Water Eutrophication in the

Roland Bol <sup>1</sup> \*, Gerard Gruau<sup>2</sup> , Per-Erik Mellander <sup>3</sup> , Rémi Dupas <sup>4</sup> , Marianne Bechmann<sup>5</sup> , Eva Skarbøvik <sup>5</sup> , Magdalena Bieroza<sup>6</sup> , Faruk Djodjic<sup>6</sup> , Miriam Glendell <sup>7</sup> , Philip Jordan<sup>8</sup> , Bas Van der Grift <sup>9</sup> , Michael Rode<sup>10</sup>, Erik Smolders <sup>11</sup>, Mieke Verbeeck <sup>11</sup>, Sen Gu<sup>2</sup> , Erwin Klumpp<sup>1</sup> , Ina Pohle<sup>7</sup> , Maelle Fresne<sup>3</sup> and Chantal Gascuel-Odoux <sup>4</sup>

<sup>1</sup> Agrosphere (IBG-3), Institute of Bio- and Geosciences, Forschungszentrum Jülich, Jülich, Germany, <sup>2</sup> CNRS, Géosciences Rennes - UMR 6118, FranceOSUR, University of Rennes, Rennes, France, <sup>3</sup> Agricultural Catchments Program, Department of Environment, Soils and Landuse, TEAGASC, Johnstown Castle, Ireland, <sup>4</sup> INRA, Agrocampus-Ouest, UMR1069 SAS, Rennes, France, <sup>5</sup> Norwegian Institute of Bioeconomy Research, Ås, Norway, <sup>6</sup> Department of Soil and Environment, Swedish University of Agricultural Sciences, Uppsala, Sweden, <sup>7</sup> Environmental and Biochemical Sciences Group, James Hutton Institute, Aberdeen, Scotland, <sup>8</sup> Environmental Sciences Research Institute, Ulster University, Londonderry, Ireland, <sup>9</sup> Subsurface and Groundwater Quality, Deltares, Delft, Netherlands, <sup>10</sup> Department of Aquatic Ecosystem Analysis, Helmholtz Centre for Environmental Research-UFZ, Magdeburg, Germany, <sup>11</sup> Division of Soil and Water Management, KU Leuven, Leuven, Belgium

# Agricultural Landscapes of Northwest Europe

#### Edited by:

Rosalind Jane Dodd, Lincoln University, New Zealand

#### Reviewed by:

Sönke Hohn, Leibniz Centre for Tropical Marine Research (LG), Germany Verena Pfahler, Rothamsted Research (BBSRC), United Kingdom

> \*Correspondence: Roland Bol r.bol@fz-juelich.de

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 26 March 2018 Accepted: 20 July 2018 Published: 23 August 2018

#### Citation:

Bol R, Gruau G, Mellander P-E, Dupas R, Bechmann M, Skarbøvik E, Bieroza M, Djodjic F, Glendell M, Jordan P, Van der Grift B, Rode M, Smolders E, Verbeeck M, Gu S, Klumpp E, Pohle I, Fresne M and Gascuel-Odoux C (2018) Challenges of Reducing Phosphorus Based Water Eutrophication in the Agricultural Landscapes of Northwest Europe. Front. Mar. Sci. 5:276. doi: 10.3389/fmars.2018.00276 In this paper, we outline several recent insights for the priorities and challenges for future research for reducing phosphorus (P) based water eutrophication in the agricultural landscapes of Northwest Europe. We highlight that new research efforts best be focused on headwater catchments as they are a key influence on the initial chemistry of the larger river catchments, and here many management interventions are most effectively made. We emphasize the lack of understanding on how climate change will impact on P losses from agricultural landscapes. Particularly, the capability to disentangle current and future trends in P fluxes, due to climate change itself, from climate driven changes in agricultural management practices and P inputs. Knowing that, future climatic change trajectories for Western Europe will accelerate the release of the most bioavailable soil P. We stress the ambiguities created by the large varieties of sources and storage/transfer processes involved in P emissions in landscapes and the need to develop specific data treatment methods or tracers able to circumvent them, thereby helping catchment managers to identify the ultimate P sources that most contribute to diffuse P emissions. We point out that soil and aqueous P exist not only in various chemical forms, but also in range of less considered physical forms e.g., dissolved, nanoparticulate, colloidal and other particulates, all affected differently by climate as well as other environmental factors, and require bespoke mitigation measures. We support increased high resolution monitoring of headwater catchments, to not only help verify the effectiveness of catchments mitigation strategies, but also add data to further develop new water quality models (e.g., those include Fe-P interactions) which can deal with climate and land use change effects within an uncertainty framework. We finally conclude that there is a crucial need for more integrative research efforts to deal with our incomplete understanding

**37**

of the mechanisms and processes associated with the identification of critical source areas, P mobilization, delivery and biogeochemical processing, as otherwise even highintensity and high-resolution research efforts will only reveal an incomplete picture of the full global impact of the terrestrial derived P on downstream aquatic and marine ecosystems.

Keywords: phosphorus, cycling, soil, eutrophication, climate change, colloidal and particulate, water quality

#### INTRODUCTION

The increasing world-wide issue of the eutrophication of our lakes, reservoirs, rivers and coastal waters has highlighted an urgent need for interdisciplinary action across research fields (e.g., Moss, 2012; Withers et al., 2014; Elmgren et al., 2015; Pinay et al., 2017; Charlton et al., 2018). It is now well documented that anthropogenically derived phosphorus (P) and nitrogen (N) pollution are currently the main drivers of eutrophication, with the excessive inputs of these two nutrients into freshwater and estuarine water bodies being considered as one of the most urgent environmental issues that human societies face (Rockström et al., 2009; Steffen et al., 2015; George et al., 2017). Agricultural activities and urbanization can both deliver excess P and N to aquatic ecosystems, which may cause eutrophication of water courses and which in turn may alter the native ecological communities, degrade ecosystem services and directly or indirectly impact water supply, recreational uses, and human health (e.g., cyanobacteria blooms). Already, in the late 1960s a fundamental change started to occur in most developed countries regarding the type of P input (see **Table S1** for P terminology and additionally Haygarth and Sharpley, 2000), namely a very marked decrease of point-source P due to an increased waste water treatment (Persson, 2001; Billen et al., 2007; Grizzetti et al., 2012; Scavia et al., 2014; Minaudo et al., 2015). This decrease has had rapid and marked effects on the P loading and trophic status of downstream water bodies (e.g., a 75% reduction in 25 years of the P flux in the Seine River and Lake Geneva; disappearance of cyanobacteria blooms in some lakes, such as Lake Erie in the USA, Lake du Bourget in France, and Lake Mjosa in Norway; Nesheim et al., 2010**;** Jacquet et al., 2014; Scavia et al., 2014; Romero et al., 2016). However, this reduction in point sources has been counterbalanced by the stagnation or even the increase of diffuse P emissions from agricultural soils (Scavia et al., 2014; Dupas et al., 2015d; Stoddard et al., 2016). This has happened even despite the fact that overall fertilizer usage in NW Europe since the 1980's has decreased significantly (Schoumans et al., 2015). For example in Sweden, lower P fertilizer use is now accompanied by more or less balanced P inputs and outputs in agriculture (Bergström et al., 2015). Both dissolved/colloidal P, i.e., the most bio-available forms of P for algae (Dupas et al., 2015a; Mellander et al., 2016; Gu et al., 2017), as well as particulate P (Bechmann and Deelstra, 2013) are involved in this increase of P emissions from agricultural soils. For Norway increased delivery of particulate P to surface waters can partly be assigned to higher soil erosion due to climate change effects (i.e., increased runoff) (Deelstra et al., 2011). Whereas, for Sweden a modeling study (Arheimer and Donnelly, 2013) suggested that the total mean load to the Baltic Sea will increase for P, but may decrease for N by 2100 due to climate-induced changes. Enhanced particulate P can be formed in the surface waters due to the discharge of iron (Fe) and calcium (Ca)-bearing groundwater and subsequent precipitation of P-rich minerals (Baken et al., 2013; Van der Grift et al., 2014, 2018). Depending on the receiving water body, most of this particulate P could in the long run become available to algae (e.g., Yang et al., 2016; Yao et al., 2017).

Maintaining high levels or even increasing diffuse P losses from agricultural landscapes hampers any expected improvements in water quality, thereby challenging our society's ability to combat eutrophication. The issue here concerns not only lakes, rivers and estuaries, but also the multitude of small water bodies of natural and artificial origin, some which may serve as reservoirs for drinking water production.

For 30 years or more, much knowledge has been gained on the basic physico-chemical processes by which P is mobilized in soils and transported to rivers, as well as on the chemical forms of the mobilized and transported P. For example, the role of wetting-drying cycles as catalysts for the production and transfer of dissolved and colloidal phosphorus is now well known (e.g., Turner et al., 2003; Butterly et al., 2011; Blackwell et al., 2013; Chen et al., 2016; Gu et al., 2018). In particular, the important role of colloids to serve as carrier of P in soils and waters has been highlighted (e.g., Henderson et al., 2012; Gottselig et al., 2014; Liu et al., 2014; Baken et al., 2016; Jiang et al., 2017; Gu et al., 2018; Missong et al., 2018). Also wellconstrained is the capacity of P to bind with manganese (Mn) and iron oxides and the influence the redox state of soils and waters has on P mobility in the environment (Scalenghe et al., 2012; Van der Grift et al., 2014; Jiang et al., 2015a,b; Smolders et al., 2017). Along with these understandings, novel high resolution P water quality monitoring, such as sensor technology, can be used to understand the processes of dissolved and particle-bound substances in waters. For example by using turbidity as a proxy for particulate P we can now reliably achieve high-frequency observations of P losses from agricultural landscapes over multiannual periods (Jordan et al., 2007; Skarbøvik and Roseth, 2014; Rode et al., 2016; Shore et al., 2017). This information, combined with ever advancing analytical techniques (Kruse et al., 2015) as well as improvements in landscape visualization tool kits, is helping to facilitate efforts to better quantify P dynamics in time and space, from plot to catchment and landscape.

Transforming this knowledge into appropriate cost effective mitigation strategies remains a major challenge (Schoumans et al., 2014, 2015; Dodds and Sharpley, 2015; Kleinman et al., 2015; Sharpley et al., 2015; Withers et al., 2015a,b; Rowe et al., 2016). The degree of integration and knowledge is such today that some of the most recent studies can go as far as integrating the costs of management in the reduction of diffuse phosphorus emissions, in connection with national or local legislation or land property issues (e.g., McDowell et al., 2016; Vinten et al., 2017; Zhang et al., 2017).

The EU Water Framework Directive (WFD) has since its emergence in the beginning of this millennium set the scene for water management in Europe. Its implementation has resulted in massive efforts to classify all water bodies according to their ecological status, within this context monitoring of European waters has, at least tentatively, become more harmonized through the Common Implementation Strategy (CIS) guidelines. Achieving Good Ecological Status of waterbodies across Europe under the Water Framework Directive will require an improved understanding of the link between P concentrations, loads and sources in the soil and recipient freshwater ecosystems within the context of multiple stressors (Whitehead and Crossman, 2012; Crossman et al., 2013). However, in this paper, we will not focus on all the aspects of the WFD or other EU legislation, but only few aspects that are linked to the main foci of our review. These include the types of monitoring that can give new insight into catchment processes, including investigative monitoring and source apportionment.

Furthermore, we are taking into account the considerable amount of papers already published on the topic of P transfer and P management in agricultural landscapes, including several recent review papers (e.g., Kleinman et al., 2011; Chowdhury et al., 2014; Schoumans et al., 2014; Sharpley et al., 2015; Kadlec, 2016; McDowell et al., 2016). As well as related contemporary topical reviews on aspects like scenario analysis and mitigation (Roberts et al., 2012; Schoumans et al., 2014; Dodd and Sharpley, 2016; Ahmad et al., 2017; Liu et al., 2017; Wu et al., 2017), policy and implementation (Christen and Dalgaard, 2013; McDowell et al., 2016), novel modeling approaches and management insights Shepherd et al., 2011; Radcliffe et al., 2015; Xie et al., 2015; Ouyang et al., 2017, water quality modeling (Rode et al., 2010; Wellen et al., 2015; Hashemi et al., 2016), erosion (Panagos et al., 2017) and other more general appraisal of future needs and directions (Fernandez-Mena et al., 2016; Garnache et al., 2016). Hence our intention in this article is not to produce an exhaustive review which also covers the state of knowledge in these related research fields. Rather, our paper is aimed at highlighting certain important knowledge gaps and research challenges on P cycling and transfer processes in agricultural landscapes in Northwest Europe in relation to the continued issue of eutrophication of inland and coastal surface waters. Thereby, implicitly also identifying research needs for these 4 key challenges. These main challenges for us concern: (i) a better understanding of the processes and variables controlling P mobilization and P transfer in headwater catchments, (ii) a better consideration of how the basic mechanisms involved in P mobilization in soils at the small scale aggregate themselves at larger scales, and how this aggregation ultimately control P diffuse emissions in agricultural landscapes, (iii) the resolution of ambiguities in the interpretation of river P dynamics for identification of sources in catchments, and (iv) a better understanding and prediction of the effects of climate change on P fluxes and P physical-chemical forms.

#### HEADWATER CATCHMENTS: THE RIGHT PLACE TO BE TO MANAGE AND MONITOR DIFFUSE PHOSPHORUS EMISSIONS

Combating eutrophication relies on effective reduction in P losses from land to water through appropriate management of P sources and pathways in the landscape. The key scientific and management questions are: where is the excess P coming from? How are these P sources mobilized and delivered to surface waters? What is their impact on water bodies? For still unknown reasons, some headwater catchments (surface area <50 km<sup>2</sup> ) in agricultural landscapes appear more resilient (have more buffering capacity) than others with regards to P loading, resulting in huge differences in average annual P concentrations and annual P fluxes, as exemplified by the case of Brittany, one of the most intensively farmed regions of France (Legeay et al., 2015; Abbott et al., 2017). This increased variability of P concentration and P fluxes with decreasing catchment size is as a result of headwater catchments having lower intense cultivation rates. In fact the majority of these Brittany headwater catchments shown in **Figure 1** are intensively cultivated (>80% of arable land) throughout (Legeay et al., 2015; Abbott et al., 2017). They were also found to be insensitive to the general decreased particulate P emissions that has been recorded in Brittany since the beginning of the 1990's, as this decrease left the amplitude of particulate P flux variability in headwater catchments unchanged (see **Figure 1**).

The increased variability of P emissions when moving upstream in headwater catchments is found both for dissolved and particulate P fluxes (**Figure 1**). Land use, and especially the share of arable land within a catchment, is known to have a strong positive relationship on nutrient losses (Evans et al., 2014). Indeed, plotting median TP (total P) losses against the share of arable land in Swedish catchments included in the monitoring program River outlets and Trend Watercourses results in strong (R <sup>2</sup> = 0.79, p < 0.005) and positive relationship (**Figure 2**). However, long term data from another Swedish water quality program, monitoring program arable catchments, also highlighted an inherent great variability in nutrient delivery among small catchments (<50 km<sup>2</sup> ) with high portion of the arable land (**Figure 2**). For these small catchments with high share of arable land, the relationship was not statistically significant at all (R <sup>2</sup> = 0.02, p = 0.41). Consequently, measured P concentrations from some small catchments were remarkably low in spite of intensive agricultural production, whereas other catchments were found to be less resilient and much more vulnerable to P losses.

The high spatial variability observed among headwater catchments appears persistent over time suggesting that occasional synoptic sampling of headwater catchments can provide valuable information for catchment characterization and

Original source data:

Monitoring programs River outlets and Trend Watercourses: http://miljodata.slu.se/

Monitoring programs Arable catchments:;http://jordbruksvatten.slu.se/.

management with regards to P emissions. This finding also raises the question of the spatial threshold at which the landscape splits into poorly and highly contributing headwater catchments with regards to P emissions, which has been recently referred as to the "landscape grain size" concept (Abbott et al., 2017). There are many potential parameters that could modify the capacity of headwater catchments to release P to river networks, and thus create variations in P emission properties at a specific spatial threshold, or given landscape grain size (see also Dodds and Oakes, 2008; Haygarth et al., 2012). Included are the extent of preferential flow paths in soils and aquifers, which determine residence times in different catchment components, as well as the connectivity between land and water which may strongly vary among headwater catchments (Dahlke et al., 2012; Dupas et al., 2015d, 2016; Mellander et al., 2015). Differences in P applied to or present in soils, the biogeochemical cycling in soils, linked with variations in groundwater dynamics, themselves influenced topography, could be also involved (e.g., Gu et al., 2017). Differences in past land-use history and soil properties will likewise will play a role (Stutter et al., 2015; Gu et al., 2017).

Irrespective of the processes and factors that control the ability of headwater catchments to act as sinks or sources for P, the understanding where P is coming from is important both to target and reduce these sources and also to establish who is responsible for their management according to the polluter pays principle (as defined in the WFD). There is however less clarity in WFD who will pay for the monitoring to obtain this information, in fact, the direct monitoring of point sources is often left to the owner of the waste water treatment plants (Skarbøvik et al., 2014).

Knowing the spatial structure and the typical grain size of source and sink headwater catchments with regards to diffuse P emissions in the landscape could improve site selection for targeted management efforts. Should the difference in resilience properties be of high spatial and temporal stability as found by Legeay et al. (2015) and Abbott et al. (2017), then the intervention in those headwater catchments showing the lowest resilience or highest source properties would potentially yield the largest catchment-level improvements at the lowest cost.

Knowing the landscape typical threshold or grain size could be also helpful for rationalizing water quality monitoring networks. Owing to the temporal stationarity of the spatial distribution of poorly and highly contributing headwater catchments, occasional synoptic sampling of headwater catchments would provide valuable information for identifying high and low contributing headwater catchments in agricultural landscapes. The redistribution of agricultural activity based on the difference in P emission properties of headwater catchments could be a cost effective management strategy for decreasing P loads of larger rivers. This could be implemented even in the absence of a clear understanding of the underlying mechanisms and factors that cause differences in headwater catchment properties. The implementation over the long-term of high frequency monitoring devices at the outlet of larger river basins where the river channel integrates multiple headwater catchments would make it possible to assess the effectiveness of the mitigation strategies thus deployed in the most contributive headwater catchment areas.

#### DIFFICULTIES IN DISENTANGLING POINT-SOURCE, DIFFUSE PHOSPHORUS EMISSIONS AND PRE-AGRICULTURAL PHOSPHORUS CONCENTRATION BASELINES

Load apportionment models are often used to distinguish between diffuse (mainly from agriculture) and point P sources (sewage treatments, industry sources, fish farming etc.) (Bowes et al., 2008). In many European countries, monitoring networks (statutory monitoring) were established well before 2000 to monitor pollution from point sources and to assess long-term water quality status (Bieroza et al., 2014; Kyllmar et al., 2014), but recently more new emphasis has been put on such networks as part of the obligations and the verification of compliance with EU legislative directives (e.g., WFD). Furthermore, both the location of the sampling points and the monitoring interval are not always suitable to capture the spatial and temporal dynamics of diffuse sources (Jordan and Cassidy, 2011; Bieroza et al., 2014). Thus, targeting diffuse P is difficult, since the sources operate sparsely in both space and time, except for specific locations in the catchment, so-called Critical Source Areas (CSA), and during specific times e.g., rainfall and snowmelt events (Djodjic and Villa, 2015).

Haygarth and Jarvis (1999) stated that having good quantitative knowledge of the baseline P concentrations is necessary when assessing spatiotemporal trends in water quality in agricultural areas. This is also a requirement in the WFD, which states that the environmental goals should deviate only slightly from the background or reference conditions. However, the baseline P concentrations can be expected to vary widely among regions. For example, Phillips and Pitt (2015) demonstrated the differences between European countries of nutrient boundaries used for the WFD. This may be due to natural variations in soil types and topography, but may also be a result of limitations in catchment understanding.

At any rate, such baseline levels should be based on scientific methods, and not on political concerns. Estimations of baseline P is especially difficult in the lower lying areas where the land has been used for agricultural production for centuries, and where such baseline P concentrations must either be modeled or estimated based on limited data. Additionally, those P sources originating from forests and highlands areas ("background runoff "), which are generally poorly quantified themselves, do also contribute to overall baseline P concentrations in the waters of agricultural and urban catchments. Furthermore, the baseline P values may also change, due to large-scale changes such as climate change or the reductions in acid rain. The latter has led to brownification (increased darkening of the water color, due to the increase of dissolved organic carbon content) of Scandinavian waters, which again may have impacted on P losses (Fölster et al., 2014).

#### CHALLENGES TO MONITOR AND MODEL PHOSPHORUS TRANSFER AT THE LANDSCAPE SCALE

Monitoring and modeling of P transport pathways in the landscape remains a challenge. This is particularly true for the transfer of particulate P, which is a highly dynamic process, and targeting particulate P sources depends on the exact timing of mitigation measures in relation to plant growth, soil management and soil moisture. Modeling approaches based on detailed measurements of topography (e.g., Lidar) are normally used to target both sources and pathways (e.g., erosion gullies) of sediment-bound P (Thomas et al., 2016). On tile drained soils, significant amounts of particulate P can also be transferred through macropores, which are difficult to monitor and manage (Bechmann et al., 2017a). In addition, elucidating the sources and pathways of dissolved P is also a challenge since it involves understanding of sub-surface pathways (e.g., shallow groundwater) and their connectivity with water bodies as well as its interactions with soil chemistry governing P sorption and release. Shallow sub-surface pathways and hyporheic flow have been found to be important for dissolved P delivery to streams (Bieroza and Heathwaite, 2015; Mellander et al., 2015). Recent advances in high-temporal resolution sampling with in situ analyzers and sensors have yielded information to help delineate potential pathways through hysteresis pattern analysis (Bieroza and Heathwaite, 2015; Dupas et al., 2015b). Furthermore, establishing robust correlations between routinely measured parameters (e.g., turbidity) or smart environmental tracers and P fractions can further help to gain improved understanding of processes and assist to reduce the monitoring cost (Bieroza and Heathwaite, 2016; Minaudo et al., 2017; Stutter et al., 2017).

Statutory monitoring data are used to assess chemical and ecological status of water bodies and to evaluate the impacts of diffuse pollution on stream nutrient concentrations. Despite intensive efforts to reduce these negative impacts, many rural catchments still show increasing long-term P concentration trends (Bechmann et al., 2017b), possibly due to the influence of weather events on local landscape P dynamics, highlighting the continued need for a deeper understanding of these weatherrelated processes (Bieroza et al., 2014). Seemingly similar catchments (e.g., percentage arable land in a catchment) are known to show opposing nutrient trends, which in turn may indicate a differing resilience to environmental change (Legeay et al., 2015). The observed disconnection between management practices and water quality in many catchments necessitates further efforts to better target P sources and pathways in space and time. The persistence of P pollution also requires more catchment-tailored approaches to mitigate diffuse pollution. Use of high-resolution mapping of CSAs, including soil chemistry, topography and soil type, can help to identify locations in the landscape that will benefit most from the implementation of mitigation measures (Thomas et al., 2016; Djodjic et al., 2017). Using statutory and high-frequency P monitoring in parallel could further help in identifying the most critical time periods and pathways of diffuse pollution (Dupas et al., 2015c). Additionally, new upscaling modeling approaches are needed to link single location patterns to landscape-wide process understanding.

#### AMBIGUITIES IN THE INTERPRETATION OF RIVER P DYNAMICS FOR IDENTIFICATION OF SOURCES IN CATCHMENTS

In order to limit P transfer and reduce eutrophication, it is important to identify the sources and the mobilization/delivery mechanisms of P into rivers. One commonly used method for source identification is statistical analysis of water quality time series. In particular, load apportionment models (LAMs) based on concentration - discharge (C-Q) relationships (Bowes et al., 2008, 2014, 2015; Greene et al., 2011; Lamba et al., 2015; Crockford et al., 2017; Glendell et al., 2018) or the identification of periods of the year when one source is believed to dominate over other sources, a method is used to disentangle point sources from diffuse sources (Legeay et al., 2015). Equations in LAMs all rely on the assumption that point source emissions are constant in time (leading to negative C-Q relationships due to dilution when discharge increases) while diffuse source emissions increase with discharge (leading to positive C-Q relationships due to mobilization and delivery of P during storm events). Furthermore, analysis of C-Q hysteresis loops during storm events has been used to distinguish proximal (remobilization of stream bed sediments, bank erosion, erosion of riparian area) and distal sources (hillslope erosion, subsurface transfer) (Stutter et al., 2008; Outram et al., 2014, 2016; Bieroza and Heathwaite, 2015; Bowes et al., 2015; Dupas et al., 2015a,b; Perks et al., 2015; Sherriff et al., 2016). Clockwise loops are often interpreted as originating from proximal sources whereas anticlockwise loops are interpreted as originating from distal sources. However the interpretation of P dynamics observed in rivers, and both LAMs and analysis of C-Q hysteresis loops are subject to "ambiguity" problems. Here, the term ambiguity describes situations when several processes (or sources) can lead to the same observed P dynamics in rivers. This can lead to difficulties in inferring the dominant controlling process (or source) from statistical analysis of water quality time series. Jarvie et al. (2012) have shown that remobilization of streambed sediment during winter high flow is often (wrongly) attributed to diffuse source, whereas these sediments may have been enriched by point sources during the summer low flow period, and thus the primary source is predominantly a point source. This ambiguity may lead to overestimation of diffuse sources in LAMs.

The reductive dissolution of Fe oxyhydroxides in sediments of lowland rivers (Smolders et al., 2017) or riparian wetlands in upland systems (Dupas et al., 2017) can lead to soluble reactive phosphorus (SRP) release during summer low flows, increasing summer SRP concentrations in rivers in the same manner as does an undiluted point source. This ambiguity may lead to overestimation of point sources in LAMs. Likewise, ambiguities in river P dynamics have been highlighted in the identification of proximal versus distal sources within hillslopes. Bieroza and Heathwaite (2015) have shown that in the case of two successive storm events, the first event could transport a distal P source to the near-stream zone, which can be remobilized during a second event (exhibiting a clockwise hysteresis loop). Therefore, if only the second event was monitored, there is an ambiguity in the location (proximal/distal) of the primary source, because the primary mobilization of a proximal source can lead to the same observed P dynamics pattern as the secondary mobilization of a distal source (both would exhibit clockwise hysteresis loops). Furthermore, in-situ monitoring of uncultivated soils in the near stream zone have highlighted that these zones represented a significant contribution to annual SRP exports in a temperate agricultural catchment (Gu et al., 2017). However, it is not clear to which extend these zones are primary sources (due to legacy P inputs before their conversion into uncultivated buffers) or secondary sources (as they accumulate and re mobilize P rich eroded soils from cultivated upslope areas). At the scale of the hyporheic zone, Van der Grift et al. (2014, 2018) have shown that

SRP via Fe(II) oxidatie at the groundwater-surface water interface can be followed by remobilization of newly formed particulate P. Thus, there can be an ambiguity in the location (proximal/distal) of the source of the particulate P, which can be attributed to a proximal source (the river bank) whereas it may stem from remobilization of a more distal primary source. Another example of ambiguity is the P release pulses observed in summer from river sediments whose ultimate source can be a mixture of several point and diffuse sources, located further upstream (Cooper et al., 2015a; Dupas et al., 2017; Smolders et al., 2017).

Given the important implications of these ambiguities in the identification of P sources and pathways in rural landscapes, several improved methods can be proposed. Firstly, it is possible to use tracers (e.g., stable isotopes and plant-specific biomarkers) and/or complementary water quality parameters in addition to P (Jarvie et al., 2012; Cooper et al., 2015b; Alewell et al., 2016; Glendell et al., 2018) to trace the primary sediment source. For example, Ahlgren et al., 2012 found Barium (Ba) being a promising tracer element, being present in significantly higher amounts in waters affected by agricultural runoff. Whereas Paruch and co-authors developed microbial source-tracking techniques based on advanced DNA methods to distinguish between different sources of microbial contamination (Paruch et al., 2015; Paruch and Paruch, 2017). Furthermore, recent work do suggest that the composition of P carrier colloids could also be a 'tracer/fingerprint' the source of P (Missong et al., 2018). Secondly, more observational data is needed, combining monitoring at the catchment outlet with in situ observation within functional zones such as riparian wetland or groundwater (Mellander et al., 2016; Gu et al., 2017) to disentangle primary mobilization from secondary delivery/remobilization. Thirdly, interpretation of long term change in water quality dynamics ("trajectory") in catchments where management changes, such as removal of point sources, decrease in soil P status or installation of buffer zones, have been recorded over several decades (Bieroza et al., 2014) may help to resolve ambiguities in river P dynamics. Therefore, meta-analysis of large catchment datasets may help to identify typical situations where one mechanism (e.g., summer sediment P release versus point source P input) dominates over others. In doing so, "typical" situations can be identified without ambiguity and the P dynamics pattern can be transferred to other, less studied areas as a guide to infer what are the key mechanisms controlling P loads. Finally, consideration of time lags and legacy effects is crucial when communicating results of source identification studies to catchment managers, so that they do not only target secondary proximal sources (short term vision) but also primary more distal sources (long term vision).

### DIFFICULTIES IN PREDICTIONS OF DISSOLVED/COLLOIDAL/PARTICULATE CATCHMENT PHOSPHORUS "HOT SPOTS" AND "HOT/CRITICAL" MOMENTS

The difficulty in attributing spatio-temporal variations in P fluxes and P concentrations in streams and rivers draining agricultural landscapes to specific sources or specific mobilization/transfer processes is well known (e.g., Haygarth et al., 2005; Edwards and Withers, 2007; Granger et al., 2010; Trevisan et al., 2012; Hahn et al., 2013; Sharpley et al., 2015). This difficulty mainly arises from the fact that the transfer of P in agricultural landscapes is not often caused by a single process or a single source, rather being the consequence of a series of processes, mobilizing P from several permanent or temporary sources. A good illustration of this difficulty and of the ambiguities it may create in terms of process and source identification was provided by the dissolved/colloidal P release pulses observed in riparian wetland zones. Riparian zones in agricultural catchments are active in transforming poorly mobile particulate P into highly mobile dissolved and colloidal P, which is directly or indirectly bioavailable (Stutter et al., 2009; Dupas et al., 2015a; Gu et al., 2017). Indeed, these P release pulses come from the solubilisation of particulate P eroded from the upland cultivated fields which are temporarily accumulated in wetland zones. This implies that the primary sources of the released P are the upland cultivated fields from which P originates. Quite clearly, such entanglement of processes and existence of temporary sources may lead to ambiguities in the clear identification of the zones and processes contributing most to the risk of P mobilization and P transfer at the catchment scale. Therefore, moving toward full understanding the overall sequence of transfer and retainment of P process chain and how P release capacities vary in space and time is challenging, but it is a knowledge prerequisite to better control and reduce P loss in agricultural landscapes.

In order to better evaluate and quantify how the above mentioned processes control P emissions in catchments, we need to understand how their spatio-temporal occurrence and intensity is controlled by landscape properties, and how the zones where these processes take place are connected with each other and to the river network. Though relevant for particulate P, the issue is particularly relevant for dissolved and colloidal P, which both appear much more significant contributors to P losses from agricultural lands than previously assumed (Kleinman et al., 2011, 2015; Dupas et al., 2015c; Jiang et al., 2015a; Mellander et al., 2015, 2016; Gu et al., 2017), and are also potentially bioavailable P forms. From the perspective of locating dissolved and colloidal P sources in the landscape, the way landscape structures, in interaction with climatic variables, control groundand surface water dynamics is of fundamental importance. Groundwater level changes in the overall landscape, in response to precipitation and evapotranspiration, influence the location, timing and duration of soil water saturation and wetting-drying episodes that in turn cause reduction of soil Fe-oxyhydroxides. Ground- and surface water dynamics are thus anticipated to be the important determinants of hot spots and hot moment release of dissolved/colloidal P in catchments. An example of such complex control of landscape structure on P release dynamics in catchments has been provided in a recent study conducted on riparian wetlands in a small French catchment (Gu et al., 2017). Large differences have been observed with regards to P release dynamics in the riparian zone, depending on the topography and the control it exerts on the location and timing of biogeochemical processes such as soil Fe-oxyhydroxide reduction or P leaching following soil rewetting. This study has led to the development of a coherent, spatially integrated concept of "landscape biogeochemistry" (see **Figure 3**) in which the topography, as a first-order control on spatial variation of hydrological conditions (Sørensen et al., 2006), appears to control spatial and temporal distribution of P release processes, the inputs of P from cultivated soils situated upslope (control of topography on soil erosion) and the organic and inorganic nature of released P through indirect control of topography on the rate of soil organic matter mineralization.

The chemistry of P in the soil-water continuum is often controlled through interactions with iron (Fe). Iron is a redox sensitive element and is commonly present in the soil-groundwater-surface water system. Mechanisms such as reduction of soil Fe-oxyhydroxides or rewetting of dried soils are known to trigger the release of P in soils, notably that of dissolved and colloidal P fractions (Turner and Haygarth, 2001; Stutter et al., 2009; Blackwell et al., 2010, 2013; Obour et al., 2011; Scalenghe et al., 2012; Gu et al., 2017; Missong et al., 2018). Redox gradients also result in the precipitation or dissolution of Fe-rich particles, which have a major impact on the fate and bioavailability of P (Jiang et al., 2017). Particulate or colloidal P is formed at the groundwater-surface water interface in groundwater-fed lowland catchments (Van der Grift et al., 2014, 2018; Baken et al., 2015). The formation of Fe hydroxyphosphate precipitates, with molar P/Fe ratio of 0.5 is the main immobilization process of dissolved P during Fe(II) oxidation (Voegelin et al., 2013; Van der Grift et al., 2016) and therefore a major control on the P retention in natural waters that drain anaerobic aquifers. As a consequence, Febound P can be the dominant P fraction in suspended particulate matter under such conditions (Van der Grift, 2017). The Fe(II) oxidation may also lead to formation of Fe(III) bearing colloids. Furthermore, redox mediated P release from river sediments in lowland rivers has been identified as dissolved reactive P (DRP) release mechanism during summer anoxia (Smolders et al., 2017). Reductive dissolution of ferric Fe oxides was found to be associated with mobilization of P to the water column from sediments with a molar P/Fe ratio >0.4. In contrast, no sediment DRP release was found for a lower P/Fe ratio irrespective of temperature and dissolved oxygen treatments. Hence, the P/Fe molar ratio in sediments is an indicator for P mobilization. Clearly, Fe redox dynamics should be taken into account when describing and predicting P transfer from both soils and sediments to surface water.

Particulate P, the dominant P form in many agricultural areas in Northwest European countries like Norway (Bechmann et al., 2008), may be transferred within the headwater catchments adsorbed onto soil particles, but it may also be released from the soil particles when it reaches the recipient waterbody. In addition, some algae can also use particulate P directly. When soil particles do contain high levels of P (high soil P status), these particles may release high amounts of P when diluted in water. Therefore, the importance of particulate (including colloidal) associated phosphorus for algal growth, both on a short and long-term time-scale requires further investigation.

Locating and apportionment of the sources and fluxes of the dissolved vs. collodial P conveyed by rivers in landscapes remains a longstanding and fundamental challenge for catchment scientists (see McClain et al., 2003; Bol et al., 2016; Gottselig et al., 2017b; Missong et al., 2018), with major implications for land management and nutrient pollution mitigation strategies. Colloidal P is characterized as dissolved P in most routine water quality monitoring programs (Gottselig et al., 2017a), however, it is believed to have reduced bioavailability compared to the truly dissolved P (Baken et al., 2014). Furthermore, colloids which consist of organic matter, Fe/Al oxyhydroxides and clay minerals (Jiang et al., 2015a; Missong et al., 2017) are important P carriers in and to surface waters (Gottselig et al., 2014, 2017a,b) and thus should be more explicitly accounted for in P budgets. These interactions between P and colloids need to be understood to better assess and predict P mobility, availability and eutrophication in surface waters, such as homo- and heteroaggregation behavior of colloidal particles as a controlling factor for P retention. While the mobilization and retention of P cannot be fully understood without understanding its interaction with Fe it is currently omitted from current water quality models and one of their important limitations.

### UNCERTAIN IMPACTS OF WEATHER AND CLIMATE CHANGE ON P STATUS

Human influence on the global climate and a range of future climate projections are now well established (Jenkins et al., 2010; IPCC, 2014). However, less information is currently available in the literature on how climate change will impact on current and future P losses from agricultural landscapes (Ockenden et al., 2017). Similarly, how to disentangle current and future trends in P fluxes due to climate change itself from those related to climate driven changes in agricultural management practices and P inputs (Dupas et al., 2016; Bussi et al., 2017). Furthermore, the responses to climate change could vary widely across geographical scales due to localized differences in the sensitivity of catchment P losses to climatic drivers (Mellander et al., 2018), thus highlighting the need to establish climate vulnerability maps for better assessments of the sensitivity of P water quality trends (such as eutrophication potential) to climatic factors. For example, in Western Europe the weather patterns are influenced by both anthropogenic warming and by decadal trends in the North Atlantic Oscillation (NAO) (Hurrell, 1995). Thus long term changes in weather are not spatially uniform and the effects on water quality are expected to vary for different physical and chemical settings, making an area more or less susceptible to P loss (Mellander et al., 2018). We therefore need to better understand both the heterogeneity in large scale weather changes and the influence on small scale processes of P release, retention and transfer pathways within the landscape, as each pathway has a different impact on water quality at the catchment scale. This knowledge is needed both in order to assess current water quality trends (concentrations and loads) and to model future scenarios for specific regions.

As said, while the impacts of projected climate on river hydrological regimes are widely studied, fewer projections of the likely impact on future P pollution in running waters exist

FIGURE 4 | Overland flow on an arable field with well drained soils (country Wexford, Ireland), following heavy rainfall in November 2014 [photo: Michael Fleming].

(Whitehead et al., 2009; Dunn et al., 2015; Mehdi et al., 2015; Hesse and Krysanova, 2016; Ockenden et al., 2016, 2017). This may be related to the continuing complexity of human activities and its impact on water quality across a range of different spatial scales, including the small headwater catchment scale, while climate scientists traditionally focus their work mostly only at the larger scales (Michalak, 2016). In addition, while it is possible to model the link between precipitation and discharge with reasonable certainty, the relationship between climate change and water quality is subject to uncertainties in our understanding of the potential effects of climate change on the physical and biogeochemical processes along the nutrient source-mobilization-delivery-impact continuum. Furthermore, it is difficult to disentangle direct climate change impacts from the indirect consequences of adaptive land use mediated by climate change (Bussi et al., 2016, 2017). Thus, water quality modeling with respect to climate change impacts is subject to more weaknesses and uncertainties compared to rainfallrunoff modeling (Hesse and Krysanova, 2016). Understanding the impact of climate change on water quality will require an understanding of the link between P concentrations, loads and recipient freshwater ecosystems within the context of multiple stressors (Whitehead and Crossman, 2012; Crossman et al., 2013), including point source and diffuse pollution sources and water abstraction.

Across NW Europe, national scale climate projections vary from relatively straightforward patterns of drier summers and wetter winters in the Atlantic climate across the British Isles (Jenkins et al., 2010), to more complicated regional patterns and higher uncertainty related to both the direction and magnitude of change in precipitation in the more continental climatic regime of Germany (e.g., Van der Linden and Mitchell, 2009; Gädeke et al., 2017). Meanwhile, in Scandinavian countries (Norway and Sweden), climate projections have indicated greater warming in the northern latitudes than in southern latitudes, particularly during winter. The precipitation in Sweden is expected to increase throughout the country but more moderately in summer (Lind and Kjellström, 2008), while in Norway precipitation is expected to increase across the country with the highest increase expected during spring in Mid-Norway (Hanssen-Bauer et al., 2009). Air temperatures affect soil P mineralization, therefore increased temperatures in summer will likely lead to a buildup of labile P pool in the soil (Turner and Haygarth, 2001), ready for transfer to waterbodies by the autumn/winter rains. Thus, areas where both summer temperatures and winter rainfall are projected to increase are likely to become more prone to P mobilization and thus vulnerable to P loss to water (Jordan et al., 2012) (**Figure 4**). Soil rewetting after a dry summer will mobilize P of mostly microbial origin, largely controlled by the soil P status (Dupas et al., 2015a), while prolonged wet periods due to more rain or more frequent rain may further enhance P release due to reductive dissolution of Fe (hydr)oxides (Van der Grift et al., 2016; Gu et al., 2017). In both cases, the mobilized P consists mainly of dissolved (both organic and inorganic) P forms, i.e., of P forms which are potentially highly bioavailable to algae.

Drying-rewetting and freezing-thawing episodes of soils are two of the most common forms of abiotic soil perturbation, resulting in solubilisation and release of P (Blackwell et al., 2009). In response to climate change it is likely that the frequency of such episodes and/or the duration of the dry and frozen periods will change differently in different areas. Furthermore, soil freezing-thawing events can increase the rates of P losses due to freezing of plant material (e.g., Bechmann et al., 2005). In Northern European countries, such as Sweden, projected warmer temperatures will likely shorten the period of persistent snowpack and cause more soil freezing-thawing episodes (Mellander et al., 2007) with increased erosion and loss of particulate P. In German lower mountain ranges snowmelt events are often an important source of sediment and P loss (Ollesch et al., 2005, 2006). Increasing temperatures may therefore shorten periods with snow cover, but could in contrast to Northern Europe also lower surface runoff due to smaller number of snowmelt events (Anis and Rode, 2015). These may mobilize both the dissolved and particulate P. In Norway, the trend in the number of freeze-thaw events depends on altitude, with an increasing number observed in the mountains (Hanssen-Bauer et al., 2009) and a decrease observed in the lower lying agricultural areas along the coast (Bechmann and Eggestad, 2016).

Catchments at high risk of P transfer are particularly responsive to changes in climate. For example, in a hydrologically flashy agricultural catchment in Ireland the total reactive phosphorus (TRP) loss was threefold that of a more hydrologically buffered catchment, despite similar soil P status (Mellander et al., 2015). The inter-annual variability in P loss in the flashy catchment was larger than the difference between the two catchments, highlighting the catchment dependence of P transfer risk and sensitivity to more rain, or more frequent rain, to catchment different hydrological flow paths. In another example from Norway, soil erosion was found to be the main process of P transfer in agricultural catchments with arable land and a flashy hydrology. In these catchments the soil P status appeared to have an influence on the catchment P loss (Bechmann et al., 2008). Thus changing rainfall patterns and rainfall intensity under climate change may affect rainfall erosivity and therefore particulate P mobilization and transfer to watercourses (Panagos et al., 2017; Poggio et al., 2018).

An understanding of the magnitude of different hydrological flow paths, their response to climate change and their potential effects on nutrient and soil loss processes will be necessary for choosing the right mitigation measures under future climate scenarios. While the amount of P lost from catchments increases with runoff, several mitigation measures, such as sedimentation ponds, have been implemented to reduce P loading from agricultural catchments, with an effective annual retention of 8– 35% of TP (average 18%) (Blankenberg et al., 2013). However, a recent study found that current mitigation efforts will not be sufficient to combat the effects of climate change on P losses, particularly in areas where increased winter runoff will lead to an overriding increase in P loads (Ockenden et al., 2017).

The importance of concentrations vs. loads and seasonality of climate change impacts (i.e., summer vs. winter) will vary with the type of receptor waterbodies (e.g., headwater catchments vs. downstream river reaches and lakes) and will have different impacts on freshwater ecology. For example, increased winter river flows may lead to increased total annual P loads (Ockenden et al., 2017), leading to negative impacts on standing waters such as lakes and reservoirs during the ecologically active period of spring and summer (Stamm et al., 2014). Conversely, increased P concentration during reduced summer flows (Bussi et al., 2017), could lead to negative impacts on river ecosystems (Stamm et al., 2014). These likely differential seasonal and water-body effects will be important to consider when modeling the impact of climate change scenarios on aquatic ecology. Thus improvement of our mechanistic understanding at multiple scales along with development of novel methods for accommodating rigorous error analysis are the imperative challenges for the future of integrated water quality modeling (Rode et al., 2010). Meanwhile, modeling the complex interactions between climate change impacts, social and economic adaptation and land use change is not a trivial task and whilst it is critical for future adaptation policies, to our knowledge, only limited examples of such integrated analysis are available to date (Dunn et al., 2012; Mehdi et al., 2015; Sample et al., 2016).

#### CONCLUSION FOR FUTURE RESEARCH

Independent of the variety of issues highlighted in the present paper which do highlight the fact that the reduction of phosphorus (P) based water eutrophication in the agricultural landscapes of Northwest Europe is neither simple nor straightforward. We like to conclude that there is a crucial need for more integrative research efforts to deal with our

#### REFERENCES


incomplete understanding of the mechanisms and processes associated with the identification of critical source areas, P mobilization, delivery and biogeochemical processing, as otherwise even high-intensity and high-resolution research efforts will likely only reveal an incomplete picture of the full global impact of the role of terrestrial derived P on downstream aquatic and marine ecosystems.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### ACKNOWLEDGMENTS

The INSU-CNRS Earth Sciences and Astronomy Observatory of Rennes (OSUR), the Loire-Brittany Water Agency (AELB), and the French Catchment Network Observatory (SOERE Réseau des Bassins Versants) are thanked for their financial support to the international workshop meeting which was held in Rennes (France) between 24th and 26th October 2017 and whose outputs and discussions have served as a basis for constructing this paper.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00276/full#supplementary-material

streamwater in groundwater-fed lowland catchments. Environ. Sci. Technol. 49, 2886–2894. doi: 10.1021/es505834y


ecosystem services and implementation. Biomass Bioenergy 55, 53–67. doi: 10.1016/j.biombioe.2012.09.053


catchments to downstream reaches. Biogeosciences 14, 4391–4407. doi: 10.5194/bg-14-4391-2017


small agricultural catchments: calibration and validation of a parsimonious model. Hydrol. Earth Syst. Sci. 17, 3679–3693. doi: 10.5194/hess-17-3679-2013


patterns of biogeochemistry of the Loire River (1980-2012). Biogeoscience 12, 2549–2563. doi: 10.5194/bg-12-2549-2015


from groundwater into surface water. Hydrol. Earth Syst. Sci. 18, 4687–4702. doi: 10.5194/hess-18-4687-2014


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Bol, Gruau, Mellander, Dupas, Bechmann, Skarbøvik, Bieroza, Djodjic, Glendell, Jordan, Van der Grift, Rode, Smolders, Verbeeck, Gu, Klumpp, Pohle, Fresne and Gascuel-Odoux. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Evidence for Coupling of the Carbon and Phosphorus Biogeochemical Cycles in Freshwater Microbial Communities

#### O. Roger Anderson\*

*Department of Biology, Lamont-Doherty Earth Observatory, Columbia University, Palisades, NY, United States*

Considerable attention has been given to the roles of the carbon and phosphate cycles in aquatic environments, but less attention has been given to an experimental analysis of the coupling of the C and P cycles in freshwater and marine ecosystems. Using laboratory microcosm experiments, prepared with natural pond-water microbial communities, evidence is presented for the coupling of dissolved organic C with microbial production of alkaline phosphatase driving the phosphorus cycle in freshwater microbial communities. The effects of glucose C-supplementation in microcosm microbial communities (including bacteria and heterotrophic nanoflagellates) on gains in microbial C-content and alkaline phosphatase activity (APA) were estimated in relation to control microcosms without C-supplementation. The C-supplementation increased total microbial APA (pmol min−<sup>1</sup> µg <sup>−</sup><sup>1</sup> bacterial C) in the C-supplemented treatment (6.5 ± 0.6) compared to the non-supplemented cultures (5.1 ± 1.7). Microbial-bound APA in the C-supplemented treatment was particularly enhanced (4.4 ± 0.9) compared to control cultures (1.3 ± 0.8), but the amount of free (soluble) APA in the aquatic phase was less compared to the controls (*n* = 5, *p* < 0.001). Alkaline phosphatase activity was highly correlated (*r* = 0.97) with bacterial densities in the C-supplemented cultures, further supporting the hypothesis that C-supplementation can increase phosphorus remineralization through elevated production of microbial alkaline phosphatase. This laboratory-based, experimental study suggests that additional research on the coupling of the C and P cycles in freshwater and marine environments may yield productive insights into the finer details of the roles of these two biogeochemical cycles in aquatic microbial community dynamics.

Keywords: aquatic alkaline phosphatase, bacterial C biomass, biogeochemical cycles, heterotrophic nanoflagellate C biomass, microbial food webs, phosphorus biomineralization, planktonic microbial communities

# INTRODUCTION

There is considerable evidence that eukaryotic microbial communities serve a significant role in fertility of aquatic and terrestrial environments (e.g., Bloem et al., 1989; Adl, 2003; Anderson, 2014b), particularly the contribution of heterotrophic protists such as nanoflagellates that prey on bacteria and promote remineralization of mineral and carbon compounds (e.g., Cole et al., 1977; Clarholm, 1989; Caron, 1994; Selph et al., 2003; Anderson, 2012). Some of this effect can

#### Edited by:

*Christian Lønborg, Australian Institute of Marine Science, Australia*

#### Reviewed by:

*L. Antonio Cuevas, EULA-Chile, Universidad de Concepcion, Chile Renata Zaccone, Istituto per l'Ambiente Marino Costiero (CNR), Italy Casey Michael Godwin, University of Michigan, United States*

> \*Correspondence: *O. Roger Anderson ora@ldeo.columbia.edu*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science*

Received: *11 November 2017* Accepted: *16 January 2018* Published: *05 February 2018*

#### Citation:

*Anderson OR (2018) Evidence for Coupling of the Carbon and Phosphorus Biogeochemical Cycles in Freshwater Microbial Communities. Front. Mar. Sci. 5:20. doi: 10.3389/fmars.2018.00020*

**53**

be attributed to nutrient release during phagotrophic feeding, especially in the microbial loop (e.g., Güde, 1985; Caron, 1994). Further contributions come from the death and lysis of bacteria (Renn, 1937) and by stimulation of bacterial hydrolytic enzymes mediating remineralization of inorganic nutrients such as phosphorus and nitrogen (Hadas et al., 1992; Ferrier-Pagès and Rassoulzadegan, 1994). In heterotrophic microbial communities, organic carbon compounds are a major source of nutrients for bacteria at the base of the food web (e.g., Ducklow, 2000; Anderson and McGuire, 2013). Heterotrophic nanoflagellates (HNAN) are often abundant major predators on bacteria, and are important links in the trophic hierarchy between bacteria and higher biota such as invertebrates (Caron et al., 1985; Anderson, 2012); therefore, the abundance and C-content of HNAN are analyzed in addition to bacteria. Moreover, because there is a focus in this study on C enrichment, and the C and P biogeochemical cycles in microbial communities, the alkaline phosphatase activity (APA) is normalized to bacterial C-biomass, expressed as APA unitsµg <sup>−</sup><sup>1</sup> bacterial C-content.

Although there is substantial published research on the role of organic C in microbial community dynamics, there is less research on the coupling of organic C with the phosphorus cycle, including production of microbial phosphatases, a potentially significant source of phosphate remineralization, especially in aquatic environments (Gibson et al., 1996; Dyhrman et al., 2007; Duhamel and Moutin, 2009; Ruttenberg and Dyhrman, 2012; Duhamel et al., 2014). Because of the potential threats from global warming, considerable research has been done on the coupling of nitrogen and carbon cycles, largely in relation to the effect of nitrogen on primary production, thus enhancing CO<sup>2</sup> sequestration (Esser et al., 2011; Zaehle, 2013). However, there is a relative lack of ecological experimental research on the potential role of organic C in driving the phosphorus cycle as mediated by microbial organo-phosphohydrolytic enzymes within the context of a community level analysis. A complex network of interactions link microbial communities to the biogeochemical changes of C and P (e.g., Anderson et al., 2018), including the role of increased atmospheric C as CO<sup>2</sup> that drives primary production, thus producing more soluble organic nutrients in aquatic ecosystems. Soluble nutrients potentially enhance heterotrophic microbial production, including release of enzymes such as alkaline phosphatases that convert organic phosphates into free phosphate available to support further growth of primary producers, and micro-heterotrophs. Understanding the dynamics and inter-relations of these processes can yield deeper insights into aquatic microplankton interactions.

This is a report of the effects of dissolved organic C (DOC) enrichment on alkaline phosphatase activity in freshwater heterotrophic microbial communities with a goal of estimating the role of DOC in phosphorus remineralization. The major research questions addressed are as follows: (1) What is the effect of emending freshwater microbial cultures with glucose-C on the relative gain in densities and C-content of bacteria and heterotrophic nanoflagellates? (2) To what extent does enrichment with glucose-C affect the alkaline phosphatase activity in these cultures? (3) Is there experimental evidence of coupling of the C cycle and phosphorus cycle in aquatic microbial communities?

# MATERIALS AND METHODS

#### Laboratory Cultures

Laboratory cultures of microbial communities were established using a standard source of pond water distributed by Carolina Biological (Burlington, NC), a filtered and pasteurized water obtained from a local freshwater pond at Burlington, NC. To establish a sufficiently robust microbial community in laboratory cultures, the pond water was inoculated with a suspension of dark, rich sediment obtained from the margin of a freshwater pond on the Lamont-Doherty Earth Observatory Campus (41.004247, - 73.911677). A sample of the surface sediment (0.1 g) was added to 5 ml of the Carolina pond water in a 15-ml plastic conical centrifuge tube and thoroughly suspended using a vortex mixer. One milliliter of the suspension was added to 100 ml of Carolina pond water to be used for the experimental laboratory cultures. Fifty milliliter of this prepared pond water was distributed into each of two sterile 250 ml Falcon culture flasks (Thermo Fisher Scientific, Waltham, MA). One flask served as a control and the other flask served as a C-enrichment culture, emended with reagent grade glucose at a final concentration of 4 mg glucose-C ml−<sup>1</sup> water. The cultures were maintained in the dark at 20◦C (a reasonable growth temperature in a temperate pond) in a constant temperature incubator (Percival Scientific, Inc., Perry, IA) for 6 days before analysis; assuming the glucose remained labile. Experiments were replicated 5 times during 2 weeks. Because a different 0.1 g portion of sediment was used for each replication, the initial composition of the inoculating biota varied. At 6 days, the final biomass varied sufficiently to permit correlation of enzyme activity with densities.

#### Microbial Densities and C-Content

After 6-d incubation, 5 ml of thoroughly suspended culture were withdrawn from each of the control and treatment flasks and fixed by addition of pure glutaraldehyde (Ladd Research, Williston, VT), final concentration of 3% (v/v). Bacterial and heterotrophic flagellate densities and C-content were assessed at 1000x magnification by UV epifluorescent microscopy (Anderson et al., 2001). During microscopic enumeration of bacteria and HNAN, they were sized within 0.5µm using an ocular reticule, deemed sufficiently accurate for the purposes of this research. While counting cocci, the diameter of representative individuals was measured and used in the formula for a sphere to estimate the volume. Likewise, the major and minor axes of bacilli were measured and used in the formula for a prolate spheroid to estimate the cell volume. The length and width of HNANs was assessed during counting and converted to volume using the formula for a prolate spheroid. The C-content of each group of microbes was estimated using a size-based regression equation relating C-content to cell volume (125.3 fg Cµm−<sup>3</sup> ) as previously published by Pelegri et al. (1999). Densities of the microbes and their biomasses were expressed as number ml−<sup>1</sup> andµg ml−<sup>1</sup> , respectively, and reported as means ± SE based on data from the five replicate experiments. The Pelegri et al. coefficient for conversion of volume to C-content was deemed to be adequate for both bacteria and HNAN, because the HNAN are relatively small in the range of 2–3 µm. Thus the differences in size ranges are not likely to cause a major bias in C-content estimates, at least within the precision needed for the limited comparisons made in this study.

#### Alkaline Phosphatase Activity Assays

Alkaline phosphatase activity in the control and treatment cultures was assayed using a Sigma APO100 colorimetric kit with paranitrophenylphosphate (pNPP) substrate as specified by the supplier (Sigma-Aldrich, St. Louis, MO). Blank cuvettes were prepared by adding 980µL of buffer (1.0 M Diethanolamine and 0.50 mM Mg Cl2, pH 9.8). Test cuvettes were prepared by adding 20µL pNPP substrate solution and 960µL of the buffer. The cuvettes were equilibrated to 37◦C before adding 20µL of the microcosm test sample to the test and blank cuvettes, respectively. The contents of the cuvettes were each immediately mixed by inversion and incubated for 30 min. The absorbance of the test sample vs. the blank was measured at a wavelength of 405 nm using a Beckman DU dual beam spectrophotometer (Beckman, Indianapolis, IN). Triplicate measurements were made for each assay and a mean value was calculated. Total enzyme activity in the control and treatment cultures was obtained using unaltered aliquots of samples from each condition. Free activity (non-bacterial bound) was assayed by passing samples of the cultures through 0.22µm pore-sized filters (EMD Millipore Corp., Billerica, MA) before enzyme analysis. The enzyme assays were done immediately after sample aliquots were taken from the microcosms and also immediately after filtration through the micropore filters. Bacterial-bound activity was estimated by subtraction of the free activity from the total activity. Enzyme activity was expressed asµmol min−<sup>1</sup> ml−<sup>1</sup> of culture, or converted to pmol min−<sup>1</sup> µg <sup>−</sup><sup>1</sup> of bacterial C in samples from each culture condition. The enzyme activities were converted from µmol min−<sup>1</sup> ml−<sup>1</sup> to pmol min−<sup>1</sup> µg −1 C, by first converting the volume in ml toµg-C based on the estimated C ml−<sup>1</sup> for the density of bacteria in the sample. The alkaline phosphatase activity was converted from µmol min−<sup>1</sup> to pmol min−<sup>1</sup> by dividing it by 10<sup>6</sup> , the number of pmols in aµmol. APA activity was normalized to only the bacterial Ccontent, because the HNAN C-content was relatively negligible; i.e., as much as two orders of magnitude less. Therefore, for purposes of approximation suitable for this analysis, the APA was normalized only to bacterial C-content. Bacteria can be good proxies for microbial contributions to some ecologically relevant measurements, such as correlation with total soil respiration in some ecosystems (e.g., Anderson, 2014a). Moreover, expressing APA units in terms of bacterial C-content is consistent with prior studies in our research program.

#### Statistical Analyses

Correlation analyses were obtained using StatPlus <sup>R</sup> (AnalystSoft, Inc., Alexandria, VA) for Microsoft Excel (Microsoft Corp., Redmond, WA), Student's t-test (GraphPad Software, La Jolla, CA) was used to assess the statistical significance of differences (p ≤ 0.05) in mean data for the experimental and control samples, because it is an appropriate analysis to use for data from two groups. Additionally, the change in value (gain) for each of the two means was also reported.

# RESULTS

#### Question 1: C Enhancement Effects on Microbial Densities and C-Content

What is the effect of emending freshwater microbial cultures with glucose-C on the relative gain in densities and C-content of bacteria and heterotrophic nanoflagellates? Mean densities and C-content of bacteria and heterotrophic nanoflagellates, including gains resulting from C enrichment (experimental treatment values minus control values), are presented in **Table 1**. As would be expected, the data for heterotrophic nanoflagellates (higher up the trophic hierarchy) were substantially lower than for the bacteria at the base of the food chain. The relative gains in density for the two groups of microbes expressed as percentage increase of the C-supplemented cultures relative to the controls are: bacteria (189%), and HNAN (192%). With respect to the relative gains in C-content, the corresponding proportional increases expressed as percentages are: bacteria (195%), and HNAN (188%).

# Question 2: C Enhancement of Alkaline Phosphatase Activity

To what extent does enrichment with glucose-C affect the alkaline phosphatase activity in these cultures? Mean data for alkaline phosphatase activities (total, free, and bacterial-bound) for the C-enriched and control cultures are presented in **Table 3**. The amount of enzyme activity was larger in the C-enriched cultures compared to the control cultures. The gain for the total enzyme activity, normalized to bacterial C, was 1.4 pmol min−<sup>1</sup> µg <sup>−</sup><sup>1</sup> bacterial C. For free activity, there was a relatively greater amount in the control than the treatment culture with a difference of −1.7. The relative gain in bacterial-bound activity, obtained by algebraic subtraction of the mean free value (−1.7) from the total (1.4), was 3.1 (**Table 3**). The gain in bound APA is larger than the estimated total gain, because the free enzyme activity was lower in the control culture than the C-enriched culture. This indicates that C-enrichment tended to increase the bacterial-bound enzyme activity relative to that of the controls. It is important to note that these values are gains (difference values), hence it is possible for the estimate of the gain in bacterial-bound

TABLE 1 | Mean <sup>±</sup> SE densities (no. ml−<sup>1</sup> ) of bacteria (×10<sup>9</sup> ), heterotrophic nanoflagellates (HNAN) (×10<sup>6</sup> ) in glucose-supplemented (+Gluc.) and non-supplemented (–Gluc.) cultures, and the gain expressed as the mean difference between the two conditions.


*Gains based on differences between means, and t-test for differences in means: <sup>a</sup>Bacteria (t* = *4.24, p* = *0.003, df* = *8), <sup>b</sup>HNAN (t* = *1.26, p* = *0.243, df* = *8).*

TABLE 2 | Mean <sup>±</sup> SE carbon content (µg ml−<sup>1</sup> ) of bacteria and heterotrophic nanoflagellates (HNAN) in glucose-supplemented (+Gluc.) and non-supplemented (–Gluc.) cultures, and the gain in C expressed as the mean difference between the two conditions.


*Gains based on differences between means, and t-test for differences in means: <sup>a</sup>Bacteria (t* = *4.51, p* =*0.002), df* = *8), <sup>b</sup>HNAN (t* = *1.23, p* = *0.255, df* = *8).*

activity to be greater than the gain in total activity. The linear correlation of the enzyme activity with bacterial densities was r = 0.97 (p < 0.001, N = 5). Thus, bacterial densities accounted for ∼94% of the variance in APA based on the square of the correlation coefficient.

#### Question 3: Coupling of the C and Phosphorus Cycles

Is there evidence of coupling of the C cycle and phosphorus cycle in aquatic microbial communities? Based on the foregoing data, evidence indicates that soluble organic C enrichment in the aquatic microbial communities not only increases the C-content of the microbiota, but secondarily also increases the alkaline phosphatase activity (particularly bacterial-bound APA), thus potentially driving phosphorus remineralization due to organophosphate hydrolysis. As reported above, in this experimental preparation, the total APA enzyme activity increased by a factor of 1.4 for everyµg of glucose-C added, when compared to the control; and there was a substantial increase in bacterial-bound APA activity. Based on the statistical analyses in **Tables 1**–**3**, the increase in bacterial abundance and bacterial-bound APA in the C-enriched treatments are statistically significant (p < 0.05).

### DISCUSSION

#### Summary of Conclusions

Glucose enrichment of the freshwater culture medium produced a gain of 111µg bacterial C ml−<sup>1</sup> of culture medium. This represents ∼3% of the amount of glucose-C added ml−<sup>1</sup> (i.e., 4.0 mg ml−<sup>1</sup> ). A positive correlation between alkaline phosphatase activity and bacterial densities was found accounting for ∼94% of the variance in the enzyme activity, estimated from the square of the correlation coefficient. Based on this experiment, for everyµg of C added above that in the control culture, the microcosm alkaline phosphatase activity (pmol min−<sup>1</sup> µg <sup>−</sup><sup>1</sup> bacterial C) increased by a factor of 1.4. Moreover, C-enrichment tends to increase the amount of bacterial-bound alkaline phosphatase, compared to that in control cultures lacking C-enrichment. In this study, the APA was normalized to estimated total bacterial C biomass, because the focus was on the C-budget and relevance of the APA to the C-cycle. However, this does not provide an estimate of the APA relative to the standing stock of metabolically active bacteria. Such estimates can be achieved by using fluorescent staining techniques that permit TABLE 3 | Mean ± SE of total, free, and bacterial-bound alkaline phosphatase activity (pmol min−<sup>1</sup> <sup>µ</sup><sup>g</sup> <sup>−</sup><sup>1</sup> bacterial C) in glucose-supplemented (+Gluc.) and non-supplemented (–Gluc.) cultures, and the gain expressed as the mean difference between the two conditions.


*Gains based on differences between means, and t-test for differences in means: <sup>a</sup>Total (t* = *1.67, p* = *0.13, df* = *8), <sup>b</sup>Soluble free (t* = *4.12, p* < *0.001, df* = *8), <sup>c</sup>Bacterial-bound (t* = *5.76, p* < *0.001, df* = *8).*

microscopic identification of the bacteria exhibiting surface APA activity. Some of these methods are discussed more fully in Section Relationships to prior research.

In this study, glucose was used as the sole C source. Glucose is commonly used in experimental microbe ecological research as representative of smaller, soluble organic nutrients (e.g., Hanson and Snyder, 1980; Shiah and Ducklow, 1994; Frost and Elser, 2002; Meon and Amon, 2004; Valliéres et al., 2008). Moreover, phytoplankton can be a significant source of glucose, in natural aquatic systems (e.g., Fogg, 1971; Moshiri et al., 1979). However, further research using a broader range of organic nutrients is needed to more fully document the effects of soluble Cenrichment on community structure and the production of APA, as well as the fate of soluble P produced as a result of organic phosphohydrolysis. Among other aspects, the molecular size and chemical composition of the organic matter, as well as its state (dissolved or particulate), needs to be examined in relation to the role of microbial communities in the aquatic P cycle.

Both the quality (kind) and quantity of soluble organic matter in natural aquatic ecosystems affect the composition and activity of microbial communities (e.g., Pérez and Sommaruga, 2006; Apple and del Giorgio, 2007). Although the use of a single C source in this study provided initial, experimental evidence of coupling of the C and P cycles, further studies with a broader composition of soluble organic matter may contribute greater predictability of the microbial community activity relative to the natural environment. Moreover microcosm, and other "bottle," experiments can introduce artifacts during laboratory incubation. Among other factors, the composition and activity of the prokaryote community may change during incubation, especially with longer durations such as days (e.g., Massana et al., 2001), especially if predator-prey relationships are not typical of the natural environment. In this study, heterotrophic nanoflagellate predators were present, and their densities and estimated C-biomass were included in addition to estimates for the bacteria (**Tables 1**, **2**).

#### Relationships to Prior Research

These data support the hypothesis of close coupling of the carbon and phosphorus biogeochemical cycles in freshwater aquatic microbial communities and provides a quantitative estimate of the likely gain in phosphorus remineralization that can occur due to soluble organic C enrichment. Organic C derived potentially from primary production, from decaying biota, or from allochthonous sources, can enhance microbial alkaline phosphatase production and thus increase phosphorus remineralization by organo-phosphate hydrolysis. Based on the evidence in this research, it appears that C-enrichment produces less free enzyme and more bacterial-bound enzyme compared to control conditions without C-enrichment. This has also been reported in studies of organic C enrichment in natural environments (e.g., Chróst et al., 1989). It is not immediately clear why this is the case. However, further research is needed to better document changes in the cellular physiology of the microbes, particularly examining the localization of the APA on the bacterial surface, perhaps by using modern fluorescent labeling techniques (e.g., Giepmans et al., 2006). Moreover, additional insights may be obtained by transcriptome analyses to determine if there are changes temporally in the transcription of alkaline phosphatase mRNA and other proteins of biogeochemical significance (e.g., Mark et al., 2005). This avenue of research may be enhanced by examining the effects of different levels of aquatic C enrichment on microbial APA, and possible variations in its transcription.

Although prior research has documented the role of phagotrophic predation in the release of organophosphohydrolases, thus promoting phosphorus remineralization, the data presented here augments these prior findings by adding experimental evidence of C-based increase in APA production within the context of as somewhat broader community-based analysis, including major bacterial predators (e.g., HNAN). Free phosphate is typically assimilated much more rapidly by bacteria than heterotrophic eukaryotic microbes and bacteria may be one of the more important contributors to the phosphorous cycle in marine ecosystems (Duhamel and Moutin, 2009; Popendorf and Duhamel, 2015). Moreover, among lake microplankton, bacteria may account for as much as 94% of the APA (Chróst et al., 1984). Overall, the increased available phosphorus due to C enrichment as observed in the research reported here may enhance bacterial growth, thus providing a positive forcing function on growth of bacterial predators higher up the food web. In some marine environments, however, bacterial remineralization of soluble phosphates may exceed the immediate needs of the bacteria, thus making the excess P directly available to drive primary

#### REFERENCES


production of phytoplankton, and consequently increased biomass of their predators (White et al., 2012).

The research reported here expands our knowledge base of the dynamic role of organic C, beyond its role as a nutrient source, to include a link to increased alkaline phosphatase activity, thus driving phosphorus remineralization in freshwater heterotrophic microbial communities. This is one of the first experimental microcosm studies to provide a quantitative estimate of the C-supplementation effect on APA in microbial communities. It is consistent with prior published research in the natural environment showing that dissolved organic C has significant contributory effects on activity of bacterial alkaline phosphatase and primary productivity in freshwater and marine ecosystems (e.g., Stewart and Wetzel, 1982; Chróst et al., 1989; Boyer et al., 2006). Moreover, similar results of C-enrichment effects on APA were recently reported for laboratory microcosm studies using organic-rich arctic soil microbial communities (Anderson et al., 2018). Elucidation of the finer ecological implications will require additional research on how environmental variables interact with the organic C enhancement effect, including variations in temperature, oxygen concentration, organic and mineral nutrient concentrations (particularly organo-phosphate substrates and available free phosphate), and interactions of the heterotrophic microbiota with photosynthetic protists in the aquatic environment. Further comparative microcosm, and fieldbased studies, across a continuum of different terrestrial and aquatic ecosystems (e.g., coastal and oceanic), correlating organic C concentrations with microbial alkaline phosphatase activity, are needed to provide further evidence of the coupling of carbon and phosphate cycles in the natural environment as suggested by this laboratory experimental study.

#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

#### ACKNOWLEDGMENTS

This is Lamont-Doherty Earth Observatory Contribution number 8179.


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Anderson. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Phosphorus Dynamics and Availability in the Nearshore of Eastern Lake Erie: Insights From Oxygen Isotope Ratios of Phosphate

#### David C. Depew<sup>1</sup> \*, Geoffrey Koehler <sup>2</sup> and Veronique Hiriart-Baer <sup>1</sup>

<sup>1</sup> Watershed Hydrology and Ecology Research Division, Environment and Climate Change Canada, Burlington, ON, Canada, <sup>2</sup> Watershed Hydrology and Ecology Research Division, Environment and Climate Change Canada, Saskatoon, SK, Canada

Blooms of filamentous benthic algae that plagued Lake Erie in the 1950s through 1970s were largely reduced through reductions of phosphorus (P) loading from point sources. Since the mid-1990s, these blooms have returned despite a period of relatively stable external P inputs. While increased loadings of dissolved P have been causally linked to cyanobacterial blooms in some parts of the lake, the impacts of ecosystem changes such as the effect of invasive species on nutrient cycling and availability have not been fully elucidated, leading to uncertainty as to the effectiveness of additional non-point P management actions. Here we use the oxygen isotope ratios (δ <sup>18</sup>OP) of phosphate in concert with measures of water quality along the northern shore of the east basin of Lake Erie to identify sources and pathways of P cycling and infer potential importance in relation to annual blooms of Cladophora that foul the shorelines of eastern Lake Erie. δ <sup>18</sup>O<sup>P</sup> data indicate that potential external source signatures are rapidly overprinted by biological cycling of P by the plankton community and that much of the available phosphate in the nearshore waters is derived from hydrolysis of dissolved organic P compounds. Near the dreissenid-colonized lake bed, δ <sup>18</sup>O<sup>P</sup> was persistently and significantly enriched in <sup>18</sup>O relative to δ <sup>18</sup>O<sup>P</sup> measured in surface waters and was similar to δ <sup>18</sup>O<sup>P</sup> of phosphate excreted by dreissenid mussels in incubations. These results implicate dreissenid mussels as key agents in nearshore P cycling and highlight the importance of considering ecosystem changes in the development of nutrient management strategies designed to ameliorate symptoms of eutrophication.

Keywords: Lake Erie, phosphorus, Cladophora, oxygen isotopes, dreissenids

# INTRODUCTION

In the North American Laurentian Great Lakes, blooms of filamentous green algae (mostly Cladophora sp.) were one of the most visible signs of cultural eutrophication in the early 1950s through to the late 1970s, particularly in Lake Erie (Shear and Konasewich, 1975; Higgins et al., 2008). Phosphorus (P) abatement programs under the Great Lakes Water Quality Agreement (GLWQA) resulted in rapid and substantial reductions in external P inputs, that while primarily aimed at reducing extensive phytoplankton blooms, hypoxia, and fish kills, also appeared to reduce Cladophora blooms although confirmational data are limited (Painter and McCabe, 1987). Despite

#### Edited by:

Christian Lønborg, Australian Institute of Marine Science, Australia

#### Reviewed by:

Michael R. Twiss, Clarkson University, United States Federica Tamburini, ETH Zürich, Switzerland

> \*Correspondence: David C. Depew david.depew@canada.ca

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 28 February 2018 Accepted: 30 May 2018 Published: 19 June 2018

#### Citation:

Depew DC, Koehler G and Hiriart-Baer V (2018) Phosphorus Dynamics and Availability in the Nearshore of Eastern Lake Erie: Insights From Oxygen Isotope Ratios of Phosphate. Front. Mar. Sci. 5:215. doi: 10.3389/fmars.2018.00215 these apparent improvements, reports of expansive blooms of Cladophora increased in frequency in eastern Lake Erie in the mid to late 1990s (Higgins et al., 2005) following P load reductions but also colonization of Lake Erie by exotic zebra (Dreissena polymorpha) and quagga (D. rostriformis bugensis) mussels (Griffiths et al., 1991; Roe and MacIsaac, 1997). It was postulated that the recurrence of these blooms during a period of stable or declining P loads (Dolan and Chapra, 2012) was in part related to structural and functional changes to the ecosystem mediated by large populations of filter feeding dreissenid mussels (Hecky et al., 2004) that increased water clarity (Howell et al., 1996), expanded areas of hard substrate for algal attachment (Haltuch et al., 2000), and enriched the benthic environment through the excretion of metabolic wastes and undigested algae (Arnott and Vanni, 1996; Conroy et al., 2005). At the same time, changing agricultural practices, precipitation patterns, and weather conditions have increased soluble P loadings to western Lake Erie, resulting in a recurrence of cyanobacterial blooms (Michalak et al., 2013). While there is general agreement reductions in P availability are the most practical means with which to reduce the extent and severity of Cladophora blooms (Auer et al., 2010), uncertainty remains as to the relative importance of direct P inputs, P circulating within the lake, and P potentially recycled in the benthic environment by dreissenid mussels (Bootsma et al., 2015). These uncertainties have confounded efforts to develop revised P loading targets for the east basin of Lake Erie to address Cladophora blooms (GLWQA, 2015).

Several recent studies have indicated that the relative abundance of oxygen stable isotopes in phosphate (δ <sup>18</sup>OP) carries information regarding the sources and biological cycling of phosphate (PO4, hereafter Pi) in aquatic environments (Colman et al., 2005; McLaughlin et al., 2006b, 2013; Elsbury et al., 2009; Joshi et al., 2015; Li et al., 2017). The strong bond between oxygen and phosphorus atoms in phosphate are resistant to oxygen exchange under most environmentally relevant conditions (Paytan and McLaughlin, 2012). While isotopic fractionation effects of abiotic processes such as sorption and desorption appear to be relatively minor (Jaisi et al., 2010, 2011), fractionation during biologically mediated enzymatic reactions can be large and at times, substrate and enzyme dependent (Blake, 2005; Liang and Blake, 2006, 2009). The uptake and internal cycling of P<sup>i</sup> by intracellular pyrophosphatases (EC 3.6.1.1) results in complete exchange of all oxygen atoms from P<sup>i</sup> with oxygen atoms from water, leading to a rapid temperature-dependent equilibrium between oxygen in P<sup>i</sup> and intracellular water (Longinelli and Nuti, 1973; Blake, 2005), which is expected to be equal to that of ambient water (δ <sup>18</sup>OW) (but see Li et al., 2016b). This isotopic equilibration has been observed in many environments to be an indicator of P<sup>i</sup> turnover (McLaughlin et al., 2006b, 2013). Hydrolysis of organic phosphorus (OP) by various extra- and intracellular phosphatases and phosphohydrolases (for example, alkaline phosphatase; EC 3.1.3.1, 5' nucleotidase; EC 3.1.3.5, phosphodiesterase EC 3.1.4.1) impart a negative fractionation at the P-O bond site, shifting the δ <sup>18</sup>O<sup>P</sup> of the product (Pi) further away from equilibrium (Liang and Blake, 2006, 2009). Consequently, knowledge of these reactions and pathways may provide a means to identify sources of P or the importance of specific P cycling pathways governing P<sup>i</sup> availability in aquatic ecosystems (Elsbury et al., 2009; McLaughlin et al., 2013; Joshi et al., 2015; Granger et al., 2017b; Li et al., 2017).

Initial application of this approach in central and western Lake Erie indicated the presence of two distinct sources of P<sup>i</sup> ; riverine inputs (∼+12‰) and a second, isotopically heavier source (∼+17‰) of unknown origin (Elsbury et al., 2009). Subsequent investigations have suggested that this P<sup>i</sup> is derived primarily by redox mediated release from anoxic sediments in the central basin (Paytan et al., 2017). Given the apparent success at identifying different P<sup>i</sup> sources in central and western Lake Erie and the residual uncertainty regarding the relevance of external vs. internal sources of P<sup>i</sup> associated with Cladophora blooms, we sought to apply this approach to (1) examine the influence of the Grand River on nearshore phosphorus concentrations (2) investigate the utility of stable isotopes of oxygen in phosphate for differentiating P<sup>i</sup> sources in the nearshore of the eastern basin and (3) to improve our understanding of P dynamics in a dynamic coastal environment. Our results presented here add to the expanding library of freshwater δ <sup>18</sup>O<sup>P</sup> data, particularly at the oligotrophic end of the spectrum. In addition, we provide data from a eutrophic river and to our knowledge, the first measurements of δ <sup>18</sup>O<sup>P</sup> generated by filter feeding bivalves. An improved understanding of the sources of P<sup>i</sup> and the manner in which it is processed and recycled within the nearshore are of critical importance to establishing effective nutrient management strategies to ameliorate Cladophora blooms as required by the amended (GLWQA, 2012).

# MATERIALS AND METHODS

# Study Region

The eastern basin of Lake Erie is the deepest of the three basins (64 m maximum depth) and is considered oligotrophic based on long term monitoring of phosphorus concentrations (mean spring TP ∼0.32µM; Dove and Chapra, 2015). The eastern basin receives ∼40% of its total P load from the Grand River (**Figure 1**) while the remainder comes via exchange with the central basin and other smaller tributaries around the north and south shores of the basin (Maccoux et al., 2016). With prevailing winds from the southwest direction, the north shore littoral region is subjected to a large fetch and this highly exposed shoreline is dominated by limestone bedrock with smaller regions of glacial till and sand (Rukavina and St. Jacques, 1971). Dreissenid mussels have colonized most of the hard substrate along this shore and remain a dominant feature of the benthos (Patterson et al., 2005; Chomicki et al., 2016).

#### Water Sample Collection and Nutrient Analysis

Water samples were collected on four to five occasions in 2013 and 2014 between the months of April and September at 12 sites along the northern shore of the east basin of Lake Erie (**Figure 1**). Vertical profiles of temperature, specific conductivity, pH, dissolved oxygen, the physical structure of

Grand River at York, Ontario.

the water column were collected at each site with a YSI-6600 or EXO-2 water quality sonde prior to initiation of sampling. All water samples were collected using a submersible pump and stored in acid-cleaned 4 L HDPE jugs, placed on ice in coolers until processing for total phosphorus (TP), total dissolved phosphorus (TDP), soluble reactive phosphorus (SRP), ammonia (NH3), nitrate/nitrite (NO3/NO2), total chlorophyll a (CHLA), particulate carbon (PC), and particulate nitrogen (PN). Water samples were processed for the above parameters following (Dove et al., 2009) and analyzed at the National Laboratory for Environmental Testing, Burlington ON. Particulate P (PP) was calculated as TP—TDP and dissolved organic P (DOP) was calculated as TDP-SRP.

For isotopic analyses of water and P<sup>i</sup> , additional samples were collected with a target volume of ∼200 L (to attain a minimum of 10µM Pi) from selected depths using a submersible pump mounted on a tripod (**Table 1**): surface (S; 1.0 m depth) and ∼0.1 m above the lake bottom (LB). To minimize boundary layer disturbance, the pump flow rates were set to ∼50–60 mL min−<sup>1</sup> , resulting in collection times of ∼1 h. Pumped water was passed under low pressure first through a 1µm polypropylene depth cartridge filter (Hytrex <sup>R</sup> ) and then through a 0.2µm pleated polypropylene filter (Flotrex <sup>R</sup> ; FPN921BGS) to remove particles and colloidal material and collected in 10 20 L acid cleaned carboys. To each carboy, ∼50 g of MgCl<sup>2</sup> (Fisher Scientific, ACS grade) was added and dissolved followed by an addition of 100 mL of 1M NaOH (Fisher Scientific, ACS grade) to achieve a loading of ∼0.3% v:v to induce precipitation of brucite (MgOH2) (Karl and Tien, 1992; Thomson-Bulldis and Karl, 1998). Fixed sample bottles were kept darkened and cool until further processing upon return to shore (see below).

#### Phosphate and Water Oxygen Isotope Analysis

To analyze δ <sup>18</sup>OP in P<sup>i</sup> , the brucite floc from each set of carboys was allowed to settle before siphoning supernatant water away. Samples were combined into a smaller number of carboys and this process was repeated until ∼10 L of brucite floc remained. Sufficient 1M HNO<sup>3</sup> (ACS grade) was added to dissolve the brucite floc, but keep the pH of the sample between 5 and 7. 1M NaOH was then added to reprecipitate brucite and reduce the overall sample volume. At least three cycles were required to reduce sample volumes from


TABLE 1 | Oxygen isotopic compositions of water (δ Ow), phosphate (δ OP), and calculated equilibrium phosphate oxygen isotopic compositions (<sup>δ</sup> OP−EQ) for different sites, dates, and depths in eastern Lake Erie and the lower Grand River.

(Continued)

TABLE 1 | Continued


Note δ <sup>18</sup>OP−EQ is calculated following Chang and Blake (2015) using the measured δ <sup>18</sup>O<sup>w</sup> at the time of sampling, while the range in brackets represents the uncertainty window calculated using the minimum and maximum δ <sup>18</sup>Ow.

200 L to ∼50 mL. At each stage, SRP and TDP concentrations in both the sample and supernatant were measured after reagent addition to ensure >99% recovery of SRP (i.e., SRP in supernatant was not detectable). Recovery of SRP was always >99% while recovery of TDP varied between 5 and 53%, depending on the sample and amount of 1M NaOH required.

Initially, we sought to precipitate silver phosphate (Ag3PO4) following the protocol of McLaughlin et al. (2004). However, we were unable to generate Ag3PO<sup>4</sup> that was sufficiently clean of contaminant oxygen using this approach. Consequently samples collected in 2013 were deemed contaminated and are not presented herein. For samples collected in 2014, we adapted the approach of Colman et al. (2005) to remove organic matter prior to undergoing sequential precipitation and crystallization steps. Briefly, concentrated samples (∼50 mL) were run through Oasis HLB resin columns (1 mL min−<sup>1</sup> flow rate) to remove organic matter. P<sup>i</sup> in the now cleaned sample was first precipitated as ammonium phosphomolybdate (APM) and then magnesium ammonium phosphate (MAP) (Kolodny et al., 1983; Tamburini et al., 2010) to separate P<sup>i</sup> from remaining impurities. Following dissolution of the MAP precipitate, cations were removed with BioRad AGx50 cation resin, and samples were evaporated to a total volume of ∼1.0 mL prior to precipitation of Ag3PO4. Ag3PO<sup>4</sup> crystals were cleaned with H2O<sup>2</sup> (15%; Tamburini et al., 2010), washed copiously with distilled water, dried at 60◦C and stored in a dessicator prior to analysis. When possible, samples were processed in duplicate with <sup>18</sup>O enriched reagents (±59.7‰) to assess potential for hydrolysis of organic P compounds. For the limited number of samples with sufficient phosphate recovered that we could process with enriched reagents (n = 9), the difference between spiked and un-spiked samples was on average <1.5‰, thus we conclude that minimal hydrolysis occurred during processing and purification.

Between 0.25 and 0.35 mg of sample and nickelized carbon were added to silver capsules (pre-baked @ 500 C, 4 h) and shipped to the National Hydrology Research Centre in Saskatoon, SK. Samples were run in duplicate or triplicate if sufficient Ag3PO<sup>4</sup> was recovered, but the majority of samples only had sufficient Ag3PO<sup>4</sup> for a single analysis. Oxygen stable isotope analyses were accomplished by thermal conversion of Ag3PO<sup>4</sup> to CO by reaction with glassy carbon at 1450◦C in a Thermo Finnigan TC/EA device. Reaction gasses were separated by gas chromatography and the resultant CO introduced to a Thermo Finnigan Delta V Isotope Ratio Mass Spectrometer (IRMS). We used crimped silver capillaries (Longinelli and Nuti, 1973) containing amount equivalent aliquots of USGS UC03 (δ <sup>18</sup>O = +29.8‰) and IAEA VSMOW (δ <sup>18</sup>O = 0.0‰) to normalize raw δ <sup>18</sup>O values from the mass spectrometer to the SMOW scale. An internal laboratory standard (0.3µM; K2HPO4; Fisher, ACS Grade) was prepared in 10 L of filtered (<0.2µm) Lake Erie water that had been previously stripped of P<sup>i</sup> using brucite flocculation as above. Internal standards were run as samples and averaged +9.6 ± 0.3‰ (un-spiked reagents) and + 9.9 ± 0.1‰ (spiked reagents). Certified reference material B2207 (Isomass Scientific) measured 22.1 ± 0.5‰. Analytical precision averaged 0.6‰. All values are reported using standard delta notation relative to Vienna Standard Mean Ocean Water (VSMOW).

Samples for water δ <sup>18</sup>O<sup>W</sup> were collected whenever samples were taken for δ <sup>18</sup>O<sup>P</sup> by filtering ∼2 mL of lake water through a 0.45µm cellulose acetate syringe filter into 2 mL glass vials. Vials were stored on ice and refrigerated until analysis by Off-Axis Integrated Output Spectroscopy (OA-ICOS) on a Los Gatos Research DLT-100 Liquid Water Isotope Analyser. We used two internal references waters, ROD3 and INV1 (δ <sup>18</sup>O = −1.1 and −28.3‰, respectively), to calibrate the raw values to the SMOW scale. The analytical precision of δ <sup>18</sup>O<sup>W</sup> values was determined by duplicate analyses of samples and internal references and is ± 0.2‰.

# Dreissenid Mussel and Cladophora Sampling

Benthic surveys to collected dreissenid mussels and Cladophora were conducted at similar intervals in 2013 and 2014 at a larger number of stations in the study area as part of a targeted monitoring effort, however we only report data relevant to the stations identified in this study for 2014 as these surveys were conducted within 1 to 4 days of sampling for water chemistry and isotopes. Briefly, mussels and algae were harvested from three 0.15 m<sup>2</sup> quadrats at each station (with the exception of 1 m depth stations where a 0.125 m<sup>2</sup> quadrat was used). Mussels were removed by hand or by scraping while applying an airlift device to collect samples in a mesh bag. Samples were stored cool on ice until return to the laboratory in Burlington, ON and subsequently cleaned of debris and frozen prior to analysis. After lyophilization, mussels were counted, weighed, measured, and soft tissue removed to estimate densities of live individuals and biomass (as shell free dry mass, SFDM m−<sup>2</sup> ). For sites with algae collected concurrently, algal filaments were first cleaned in the field by washing with lake water. Upon return to the laboratory, algal filaments were cleaned under DI water and subsequently frozen. Algal biomass was determined after lyophilization and expressed as dry mass (g DM m−<sup>2</sup> ).

#### Dreissenid Incubations

Live dreissenid mussels were collected in July of 2013 and May of 2014 to aid in characterizing the δ <sup>18</sup>O<sup>P</sup> of P<sup>i</sup> excreted by mussels (δ <sup>18</sup>OP\_DM). Divers harvested intact mussels by hand into a fine mesh bag. Mussels were placed in ziplock bags with damp paper towel and stored on ice for transport to the laboratory in Burlington, ON. Upon arrival, mussels were gently separated from rocks, shell material and other debris, brushed clean with a toothbrush under running DI water. Between 75 and 150 individual mussels (10–30 mm valve length) were placed into acid washed 4 L polypropylene jars filled with filtered (<0.2µm) Lake Erie water. Jars were aerated with air stones, and mussels were allowed to defecate and excrete for 12–36 h under a 16:8 L:D cycle at ± 2 ◦C of lake temperature. SRP concentrations were monitored in the chambers until sufficient P had accumulated for analysis (∼10 µmol). Lakewater was gently decanted and siphoned from the jar, filtered through a 0.45µm cellulose acetate filter, and processed as described above for δ <sup>18</sup>OP.

#### Calculation of Isotopic Equilibrium Values for Phosphate

The theoretical isotopic equilibrium values for P<sup>i</sup> (δ <sup>18</sup>OP\_EQ) can be calculated using P<sup>i</sup> water fractionation equations derived by Chang and Blake (2015) using Equation (1);

$$
\delta^{18}O\_{P\_-EQ} = \left(\delta^{18}O\_\text{w} \pm 1000\right) \times e^{\left[14.43 \times \left(\frac{1000}{T(K)} - 26.54\right) / 1000\right]}\tag{1}
$$

$$-1000\tag{1}$$

δ <sup>18</sup>OP\_EQ values were calculated for each sample using the measured δ <sup>18</sup>O<sup>w</sup> and temperature from the water column profiles. To assess uncertainty, we calculated δ <sup>18</sup>OP\_EQ using the minimum and maximum δ <sup>18</sup>O<sup>w</sup> for the entire study period for lake and river sites respectively.

#### Fraction of River Water and Expected Nutrient Concentrations

To estimate the relative influence of the river on nutrient conditions at the study sites, we used a linear mixing model to estimate the fractional contribution of river water (FRW) at lake sites using the measured values of δ <sup>18</sup>Ow

$$FRW = \frac{\delta^{18}O\_{\text{ws}} - \delta^{18}O\_{\text{wl}}}{\delta^{18}O\_{\text{wr}} - \delta^{18}O\_{\text{wl}}} \tag{2}$$

where δ <sup>18</sup>Ows is the water δ <sup>18</sup>O<sup>w</sup> value at each site, δ <sup>18</sup>Owr is the water δ <sup>18</sup>O<sup>w</sup> value at the river mouth and δ <sup>18</sup>Owl is the highest water δ <sup>18</sup>O<sup>w</sup> value observed at the open lake sites. We elected to use the highest δ <sup>18</sup>O<sup>w</sup> value for each survey period as there was some spatial and temporal variation in δ <sup>18</sup>Ow, possibly due to larger scale circulation features. For each site (i), we calculated the expected concentration of PP, TDP, and SRP assuming conservative mixing between the open lake and river;

$$\mathcal{C}\_{P(\boldsymbol{l})} = \mathcal{C}\_{P(\boldsymbol{l})} \left(1 - FRW\right) \pm \mathcal{C}\_{P(\boldsymbol{r})} FRW \tag{3}$$

Where CP(i) is the expected concentration of P (PP, TDP, or SRP) at each site, CP(l) is the concentration of P for the open lake end member, and CP(r) is the concentration of P at the river mouth. We used the average concentrations for sites with FRW < 0.1 to derive the open lake end member values for each survey, and CP(r) is represented by the phosphorus concentration at the river mouth.

#### RESULTS

#### Influence of the Grand River and Patterns of Nutrient Variation in the Nearshore

δ <sup>18</sup>O<sup>w</sup> values ranged from −12.6 to −8.6‰ at the river mouth but were more constrained at the lake sites, ranging from −7.9 to −6.0‰ (**Table 1**). The calculated fraction of river water (FRW) was variable over all surveys but was highest in late May—early June, reaching up to 0.8 at shallow stations ∼2 km due east of the river mouth (**Figure 2**). While FRW was highest after the spring runoff period, values ranged from near 0 to ∼0.35 between June and August surveys, particularly at stations east of the river mouth (**Figure 2**). FRW tended to increase with increasing river discharge (**Figure 2**) although the correlation was weak (r = 0.23, p < 0.1).

Nutrient concentrations were often higher and more variable in proximity to the Grand River mixing area. Total phosphorus, total dissolved phosphorus and soluble reactive phosphorus (TP, TDP, and SRP) were negatively correlated with distance from the river mouth (r = −0.42, r = −0.39, r = −0.34, p < 0.001, p < 0.001, p < 0.01, respectively). Nitrate (NO3) and dissolved organic carbon (DOC) concentrations were also strongly negatively related to distance from the Grand River mouth (r = −0.48, r = −0.48, p < 0.001, respectively). Ammonia (NH3) was the only nutrient parameter measured that did not

display any spatial gradient in relation to the Grand River (r = −0.17, p > 0.2).

Despite apparently elevated phosphorus concentrations at sites close to the river, deviation from expected concentrations based on a simple two member mixing model indicated strongly non-conservative behavior for both particulate and dissolved fractions in late May—early June. Notably, the deviations were strongest as FRW increased suggesting rapid consumption/settling or uptake processes were active (**Figure 3A**). From mid-June to August, the same general trend was apparent for particulate P, but was somewhat attenuated for TDP and SRP, largely because the concentration difference of TDP and SRP between river mouth station and open lake sites was small (**Figures 3B–D**). Some positive excursions for particulate P were evident at 1 m depth sites (**Figures 3B,C**) suggesting possible resuspension of sediment or sloughed periphyton. During mid-June, July and August surveys, TDP and SRP were generally lower than predicted for LB samples from shallow water (1 and 3 m depth) sites, with the exception of PSN 1371 and 1274 in July (**Figure 3C**). TDP and SRP in LB samples at 10 m sites were often (but not always) elevated relative to expected concentrations predicted from mixing alone (**Figure 3**).

#### Dreissenid Mussel and Cladophora Distribution

Plots of Cladophora and dreissenid mussel abundance and biomass (SFDM m−<sup>2</sup> ) are shown in **Figure 4**. Cladophora biomass ranged from 0 to 189 g DM m−<sup>2</sup> , reaching maximum values at 1 m depth in June and July (**Figure 4**). Cladophora biomass at 3 m depth sites did not exceed 25 g DM m−<sup>2</sup> and at 10 m sites did not exceed 2 g DM m−<sup>2</sup> (**Figure 4**). Dreissenid abundance and biomass varied widely between surveys and locations (6–2131 m−<sup>2</sup> and 0.2–65 g SFDM m−<sup>2</sup> respectively). Dreissenid abundance and biomass was clearly higher at 10 m depth sites than 3 m sites. No strong spatial patterns in dreissenid abundance or biomass were observed, and although Cladophora biomass was high at the 1 m sites close to the Grand River, we do not have data for similar depths at greater distances to the west side of the river, which obscures the lack of spatial patterns observed in other years (D.Depew, unpubl.data).

#### Oxygen Isotopic Composition of River and Lake Samples

δ <sup>18</sup>O<sup>P</sup> and calculated δ <sup>18</sup>OP\_EQ values for all stations and dates are shown in **Table 1**; and their spatial/seasonal and depth distributions in **Figure 5** through **Figure 7**. δ <sup>18</sup>OP\_EQ values at the river mouth were isotopically lighter compared to lake sites (**Table 1**). The seasonal dynamics of δ <sup>18</sup>O<sup>P</sup> at the river mouth are displayed in **Figure 5**. During freshet period (April-late May), despite large variation in river discharge, δ <sup>18</sup>O<sup>P</sup> varied relatively little, ranging between +11.1 ± 0.6 and +11.9 ± 0.7‰ and overlapped the δ <sup>18</sup>OP\_EQ window at times (**Figure 5**). From mid-May to late August, δ <sup>18</sup>O<sup>P</sup> declined steadily (δ <sup>18</sup>O<sup>P</sup> in the river was significantly correlated with day of year; r = −0.95, p < 0.001) through the season reaching a low of +5.7‰ in late August (**Figure 5**).

The spatial and temporal variation of δ <sup>18</sup>O<sup>P</sup> at lake sites are shown in **Figure 6**. δ <sup>18</sup>O<sup>P</sup> at lake sites ranged from +4.7 to +14.9‰ and did not show strong relationships with FRW or with distance from the Grand River (**Figure 6**). Within-transect variation (i.e., variation across different depths) in δ <sup>18</sup>O<sup>P</sup> was often greater than differences from west to east (**Figure 6**). No obvious differences were observed for samples collected near the lake bed at 1 and 3 m sites and surface samples at 10 m sites, but samples collected near the lake bed at 10 m sites (LB10) were consistently isotopically heavier than companion surface samples (paired t-test, t = 11.1, n = 12, p < 0.001) and frequently overlapped the range of δ <sup>18</sup>OP\_DM (**Figure 7**). Notably, all but one lake sample are below the calculated theoretical equilibrium based on water temperature and δ <sup>18</sup>O<sup>W</sup> (**Figure 6**, **Table 1**).

The degree of isotopic offset of δ <sup>18</sup>O<sup>P</sup> from the calculated δ <sup>18</sup>OP\_EQ was strongly correlated to SRP concentration and negatively correlated to the seston N:P ratio, suggesting larger negative excursions for samples associated with low P availability (**Figure 8**). Such patterns are expected if the hydrolysis of DOP is an important mechanism of P<sup>i</sup> regeneration. Indeed, we observed low δ <sup>18</sup>O<sup>P</sup> values and larger negative excursions from the δ <sup>18</sup>OP\_EQ values as the TDP pool became increasingly dominated by DOP (**Figure 8**). The only samples that appeared to not follow this pattern were those collected at the lakebed at 10 m depth sites.

FIGURE 3 | Plot the calculated difference between observed and expected concentrations of particulate P (PP), total dissolved P (TDP), and soluble reactive P (SRP) as a function of the fraction of river water (FRW) at each site for (A) late-May to early-June, (B) June, (C) July, and (D) August. Symbols correspond to surface (S; circles), lakebed 10 m (LB10; triangle), lakebed 3 m (LB3; diamond), and lakebed 1 m (LB1; squares). Dashed line in each panel indicates the difference in concentration that would occur if mixing PP, TDP, and SRP of river and lake water were conservative assuming a two—member mixing model.Expected concentrations were calculated collowing Equation (3).

#### Oxygen Isotopic Composition of Phosphorus Excreted by Dreissenid Mussels

Oxygen isotopic compositions of phosphate collected during dreissenid excretion experiments (δ <sup>18</sup>OP\_DM) ranged from +10.2 to +13.5‰ and averaged +11.9 ± 1.1‰ (**Table 2**). δ <sup>18</sup>O<sup>P</sup> at LB10 and LB3 sites were positively related to dreissenid abundance and shell free biomass (**Figure 9**).

#### DISCUSSION

Variation in water quality and nutrient conditions along the north shore of the eastern basin are shaped to a large degree by the discharge of nutrient rich water from the Grand River (Chomicki et al., 2016) as well as interaction with hydrodynamic features. Wind forcing (He et al., 2006) and large scale circulation features (Rao and Schwab, 2007) often trap river inputs alongshore, creating a band of enriched water that extends to approximately the 5 m depth contour (Chomicki et al., 2016). Estimates of FRW at our study sites are in general agreement with prior studies and existing knowledge and observations of nearshore circulation and river plume dynamics. For example, we observed FRW estimates approaching 80% at sites to the east of the river mouth, with larger values typically encountered at 1 and 3 m depth sites which are <1 km from the shoreline. It is unclear if the high FRW observed ∼20 km to the east represents input from the Grand River as there are a number of smaller tributaries that discharge along the shore between the Grand River and the eastern-most study sites. Previous estimates of river influence extent range from 8 to 10 km (Nicholls et al., 1983; Howell and Hobson, 2003) although these are generally based on seasonal data and use other semi-conservative tracers. Discharge from the Grand River in April and May of 2014 exceeded the

where isotope samples were collected. Note the x-axis in each panel denotes the west (negative) and east (positive) distances of each sampling site from the mouth of the Grand River. All benthic samples were collected within the same week as water quality and stable isotope samples.

FIGURE 5 | Temporal variation of Grand River discharge (m−<sup>3</sup> s −1 ; solid line) and δ <sup>18</sup>O<sup>P</sup> measured at the river mouth during the study period (circles, error bars represent 1 SD of multiple measurements). Shaded gray area represents the equilibrium window (δ <sup>18</sup>OP\_EQ) based on the minimum and maximum measured water δ <sup>18</sup>O<sup>w</sup> at the river site using the equation of Chang and Blake (2015).

long term (1913–2016) 75th percentile of ∼300 m<sup>3</sup> s <sup>−</sup><sup>1</sup> by nearly a factor 2−3 (https://wateroffice.ec.gc.ca/report/historical\_e. html?mode=Graph&type=h2oArc&stn=02GB001&dataType=

Daily&parameterType=Flow&year=2014&y1Max=1&y1Upper= 1&scale=normal), and given that river water temperatures at this time were approximately double those encountered in the lake (∼19.5◦C compared to 11–12◦C), the river plume would have likely been thermally buoyant, spreading thinly over large distances before mixing completely. The FRW was less variable during low flow periods ranging from 0 to ∼20%. FRW was generally higher at shallow stations (i.e., the 1 and 3 m depth sites), consistent with the alongshore entrapment of the plume.

Although nutrient concentrations were generally elevated closer to the river mouth and the river water clearly influenced our study sites, we observed strong non-conservative behavior of all forms of phosphorus across the river mouth—lake mixing area. During late May—early June, when concentrations of PP, TDP, and SRP were highest at the river mouth, the measured concentrations in the lake were considerably lower than would be expected based on simple dilution, even at sites in close proximity (<2 km) with FRW approaching 80% (**Figure 3**). In fact, losses of PP, TDP, and SRP appeared to be strongest at the sites closest to the river mouth, suggesting rapid sedimentation, consumption, and/or uptake (**Figure 3A**). Losses of particulate P could reflect the settling of large and coarser organic P particles owing to reductions in flow velocity and turbulence, or perhaps consumption of organic P by dreissenid mussels (Chomicki et al., 2016). Smaller particles such as non-apatite inorganic P (∼4– 7µm diameter; Chomicki et al., 2016) would likely remain in suspension for a greater duration and be subjected to more widespread mixing and dispersion. Declines in SRP and to an extent TDP may reflect uptake by phytoplankton as chlorophyll a concentrations increased from ∼0.5 µg L−<sup>1</sup> at open lake sites (e.g., 502 and 1350) to a maximum of 15.3 µg L−<sup>1</sup> at PSN 1371, 2 km east of the river mouth. Similar results have been observed

<sup>18</sup>O<sup>w</sup> over the study period

<sup>18</sup>OP\_EQ) was calculated using the minimum and maximum measured water <sup>δ</sup>

across the Milwaukee River mixing zone in Lake Michigan, where biological processes have been implicated in removal of SRP in close proximity to the point of discharge (Lin and Guo, 2016).

(C) mid-July, and (D) late August. The equilibrium window (δ

using the equation of Chang and Blake (2015).

Over the late June to August surveys, losses of particulate P remained consistently strong in close proximity to the river mouth although there were instances of resuspension elevating PP concentrations, particularly at 1 m depth sites (**Figure 3B**). The responses of TDP and SRP were generally consistent, although the magnitude of difference between observed and expected concentrations was small owing to the relatively small difference between TDP and SRP concentrations at the river mouth and lake sites. At times, TDP and SRP concentrations were larger than expected based on the mixing model calculations. Elevated concentrations of SRP were often observed for LB samples at 10 m sites (**Figures 3B–D**), but were also observed at 1 m and 3 m sites in July (**Figure 3C**). The proximate cause of these elevated concentrations could be due to leakage of nutrient rich groundwater (e.g., Robinson, 2015) or perhaps due to the decomposition of Cladophora that had accumulated at the adjacent shoreline as decomposing Cladophora is known to lose phosphorus slowly over a period of several weeks in situ (Paalme et al., 2002).

# Oxygen Isotope Ratios of Phosphate Indicate Extensive Phosphorus Cycling

Compared to marine systems (Colman et al., 2005; McLaughlin et al., 2006a,b, 2013; Goldhammer et al., 2011; Jaisi et al., 2011; Joshi et al., 2015; Li et al., 2017) and soils (Zohar et al., 2010; Angert et al., 2012; Tamburini et al., 2012; Granger et al., 2017a), freshwater environments are poorly characterized with respect to δ <sup>18</sup>O<sup>P</sup> [but see (Elsbury et al., 2009; Granger et al., 2017b; Pistocchi et al., 2017)]. The δ <sup>18</sup>O<sup>P</sup> we measured in the Grand River during April and early May (+9.4 ± 1.7 to +11.9 ± 0.7‰) are similar to values reported for Lake Erie tributaries in the central and western basins over summer and fall (Elsbury et al., 2009), but lower than those reported for the River Taw, UK (Granger et al., 2017b). δ <sup>18</sup>O<sup>P</sup> at the river mouth site was near or overlapped δ <sup>18</sup>OP\_EQ in late April and early May, but declined steadily from +9.4 ± 1.7‰ to a minimum of +5.7‰ through August, with the departure from δ <sup>18</sup>OP\_EQ increasing to the end of August(**Figure 5**). Similarly, δ <sup>18</sup>O<sup>P</sup> at lake sites are considerably lower than those reported by (Elsbury et al., 2009) for the west and central basin of Lake Erie (∼+11 to +17‰), and all but one LB sample were below the calculated δ <sup>18</sup>OP\_EQ based on δ <sup>18</sup>O<sup>W</sup> and water temperature.

It is important to keep in mind that δ <sup>18</sup>O<sup>P</sup> measured at a given site at a given time represents a snapshot of a variety of processes occurring along a continuum of P cycling. For example, the P<sup>i</sup> pool may be affected to varying degrees by the intensity of (1) P<sup>i</sup> that has been equilibrated internally and released by microbes and/or phytoplankton (Blake et al., 1997; Paytan et al., 2002), (2) P<sup>i</sup> excreted by zooplankton (Taylor and Lean, 1981), (3) remineralization of DOP excreted by phytoplankton (Bentzen et al., 1992) or liberated via zooplankton grazing and/or viral lysis (Taylor and Lean, 1981), and (4) mixing with an external P<sup>i</sup> source with a distinct δ <sup>18</sup>O<sup>P</sup> (Paytan et al., 2017). The strong negative deviations from theoretical equilibrium values observed here are anticipated in systems where ambient levels of P are limiting and arise due to the large inheritance effects associated with isotopic fractionation during the incorporation

FIGURE 7 | Vertical distribution of δ <sup>18</sup>OP(circles) and calculated equilibrium <sup>δ</sup> <sup>18</sup>OP\_EQ) (shaded polygons) in the water column at the 10 m depth sites in eastern lake Erie during (A) late-May to early June, (B) June, (C) July, and (D) August of 2014. The equilibrium window was determined by interpolating δ <sup>18</sup>O<sup>w</sup> as function of depth and calculating the range of δ <sup>18</sup>OP\_EQ using the temperatures derived from vertical profiles and the equation Chang and Blake (2015) for the minimum and maximum δ <sup>18</sup>O<sup>w</sup> observed over the study period. The dotted (dashed) vertical line(s) represent the average (range) δ <sup>18</sup>OP\_DM measured in laboratory experiments.

of water oxygen into the P<sup>i</sup> ion during enzymatic hydrolysis (Liang and Blake, 2006, 2009). Forms of DOP commonly found in aquatic ecosystems include phosphomonoesters, phosphodiesters, and phospholipids produced by organisms in situ (Baldwin, 2013) but may also include phosphomonesters, phosphodiesters and inositol phosphates exported from the surrounding catchment (Monbet et al., 2009). Monoesters are generally considered bioavailable due to the presence of enzymes such alkaline phosphatase (EC 3.1.3.1) (Bentzen et al., 1992), but phosphodiesterases and phytases may make diesters and inotisol phosphates potentially bioavailable (Cheng and Lim, 2006). Although we do not have detailed information on the nature and composition of DOP or enzyme activity in eastern Lake Erie (but see North et al., 2012), if we assume that organic P contained within organisms in the lake is at or near equilibrium (e.g., Paytan et al., 2002, based on known fractionation factors for various DOP compounds, the δ <sup>18</sup>O<sup>P</sup> of released P<sup>i</sup> recycled from the DOP pool is expected to be < ∼+14‰, and even lower < +11‰ if substrates such as phytate is not a major component of the bioavailable DOP pool (**Table 3**). While these are generalizations, they illustrate the relative magnitude expected if regeneration of Pi from DOP is an important P recycling pathway. This is further supported by the significant relationships we observed between δ <sup>18</sup>O<sup>P</sup> and other proxies of P availability such as %DOP, SRP concentration and seston N:P ratios (**Figure 8**), and we infer that the recycling of DOP by the plankton community is a major determinant of the isotopic composition of the available P<sup>i</sup> pool in eastern Lake Erie.

# Oxygen Isotope Ratios Cannot Be Used to Track Riverine Input to Eastern Erie

Concentrations of SRP at the river mouth site were high (up to 6.9µM) between April and early June. Despite large variation in river discharge volume, δ <sup>18</sup>O<sup>P</sup> was relatively invariant (**Figure 5**).

FIGURE 8 | Relationships between (A) the offset of measured δ <sup>18</sup>O<sup>p</sup> (‰; symbols) from theoretical δ <sup>18</sup>OP\_EQ and soluble reactive P concentrations, (B) offset of measured δ <sup>18</sup>O<sup>p</sup> (‰; symbols) from δ <sup>18</sup>OP\_EQ and seston N:P ratios, and (C) offset of measured <sup>δ</sup> <sup>18</sup>O<sup>p</sup> (‰; symbols) from δ <sup>18</sup>OP\_EQ and % DOP (calculated as (TDP-SRP)/TDP).

TABLE 2 | Oxygen isotopic composition of phosphate collected δ <sup>18</sup>O<sup>P</sup> in incubations with dreissenid mussels (δ <sup>18</sup>OP\_DM) harvested from eastern Lake Erie in 2013 and 2014.


PSN denotes station ID as in Figure 1.

This is somewhat surprising given that land use in the Grand River watershed is a mixture of intensive agricultural and urban use (Holeton, 2013). Although we do not have a detailed inventory of δ <sup>18</sup>O<sup>P</sup> of different sources within the Grand River watershed, based on existing data it seems likely that δ <sup>18</sup>O<sup>P</sup> might vary according to the relative inputs of P<sup>i</sup> from different sources (e.g., Young et al., 2009). In addition,the short residence time of the river (∼3 d, Rosamond, 2013) and low temperatures (<10◦C) should have acted to minimize any uptake and cycling of P<sup>i</sup> in the river at this time. While it is possible that δ <sup>18</sup>O<sup>P</sup> within the watershed might not vary across different sources, or perhaps simply reflected thorough mixing of source inputs, we note that our SRP concentrations measured at the river mouth site were up to 2 to 6-fold higher compared to concentrations measured the same day at a nutrient monitoring station ∼40 km upstream (Supplemental Information Figure S1), indicating a significant input of P<sup>i</sup> below the monitoring station. While further work will be required to identify the proximate source of this P<sup>i</sup> , we note that this section of the river is characterized by extensive wetland complexes, some of which at times can have very high TDP concentrations (6–190µM; (Gilbert and Ryan, 2007), and may explain lack of variation in δ <sup>18</sup>O<sup>P</sup> during this period.

Despite high concentrations of SRP, a relatively constant δ <sup>18</sup>O<sup>P</sup> and high discharge during the April to early June we did not observe any evidence of mixing between riverine δ <sup>18</sup>O<sup>P</sup> and lake δ <sup>18</sup>O<sup>P</sup> at any of our sites. During the cooler spring months, P uptake by phytoplankton is generally reduced relative to summer months (Lean et al., 1983), yet at sites closest to the river mouth (i.e., those with high FRW), δ <sup>18</sup>O<sup>P</sup> was no different than lake sites far removed from the river mixing zone. Only LB samples from 10 m depth sites had δ <sup>18</sup>O<sup>P</sup> that was reasonably close to riverine δ <sup>18</sup>OP, but these samples had low FRW and are likely affected by recycling of P by dreissenid mussels (see below).

Between June and August, δ <sup>18</sup>O<sup>P</sup> at the river mouth site declined steadily, departing further from δ <sup>18</sup>OP\_EQ (**Figure 7**). Such observations are not uncommon in P–rich systems (Paytan and McLaughlin, 2012), however, δ <sup>18</sup>O<sup>P</sup> moved further away from δ <sup>18</sup>OP\_EQ as SRP concentrations declined from 6.9µM in late May to a low of 0.08µM in July, which is counter to expectation (McLaughlin et al., 2006b; Granger et al., 2017b). On one hand, this could indicate an input of an isotopically light source of P<sup>i</sup> such as waste water treatment plant (WWTP) effluent (Young et al., 2009) or P<sup>i</sup> liberated via the degradation of glyphosate (Li et al., 2016a). During summer base flow periods, WWTP inputs become a larger component of the total P mass load to the river (Holeton, 2013), however, δ <sup>18</sup>O<sup>P</sup> of treated WWTP effluent from plants within the Grand River watershed are reported to be higher than values we measured [+11.4 to +22.8‰ (Morrison, 2014)] and P uptake studies downstream of WWTP discharges indicate relatively short uptake lengths (<5500 m; Barlow-Busch et al., 2006). Similarly, while glyphosate is a common herbicide in southern Ontario, concentrations typically peak during rain events (Struger et al., 2015) and since we observed low δ <sup>18</sup>O<sup>P</sup> at near base flow conditions, this is unlikely to explain the values measured here. Rather, we

speculate that P<sup>i</sup> is extensively processed within the lower river by the attendant phytoplankton community. Previous studies of the lower Grand have indicated that an appreciable portion of dissolved P is converted into phytoplankton biomass prior to entering Lake Erie (Kuntz, 2008). Despite the overall eutrophic conditions, the increasing disequilibrium from δ <sup>18</sup>OP\_EQ at the river mouth during the summer in combination with high chlorophyll a concentrations (25–48 µg L−<sup>1</sup> ) is not inconsistent with a highly productive river estuary where even the use of DOP can be important (Larson et al., 2016; Yuan et al., 2017). Further study of phosphorus sources to the lower river and speciation would be required to confirm such pathways, particularly in order to trace back to specific input sources of P. Nonetheless, even though riverine δ <sup>18</sup>O<sup>P</sup> appears to be altered prior to discharge into the lake, when our δ <sup>18</sup>O<sup>P</sup> data are considered together with evidence of strongly non-conservative mixing behavior of SRP, we infer that most of the P<sup>i</sup> discharged by the river is quickly assimilated into phytoplankton and rapidly overprinted via DOP recycling.

#### Isotopic Enrichment at the Lake Bottom Indicates Benthic P Recycling

Samples collected near the lake bed at 10 m depth sites were persistently isotopically heavier than paired surface samples at all sites on all dates, and most samples overlapped the δ <sup>18</sup>O<sup>P</sup> measured in our excretion experiments (δ <sup>18</sup>OP\_DM) (**Figure 7**). In some cases, differences between lakebed and surface SRP concentrations were large (up to 10-fold) indicating enrichment of the near bottom waters while in others, the differences were smaller. Overall, SRP was significantly higher in LB samples compared to surface samples (paired t-test, t = 2.24, p < 0.05) and chlorophyll a concentrations were lower (paired t-test, t = 2.89, p < 0.05) consistent with known impacts of dreissenid grazing of phytoplankton and excretion of soluble P (Ackerman et al., 2001; Conroy et al., 2005; Ozersky et al., 2009). Differences in TDP displayed similar patterns but were not significant (paired t-test, t = 1.69, p < 0.1) and differences in NH4, NO<sup>3</sup> and PC, PN or PP were not different (data not shown).

The experimentally determined δ <sup>18</sup>OP\_DM was constrained relative to δ <sup>18</sup>O<sup>P</sup> collected at lake sites, regardless of where or when dreissenid mussels were collected for incubations. Dreissenid mussels excrete P<sup>i</sup> directly (Arnott and Vanni, 1996), DOP (Mosley and Bootsma, 2015), and deposit a larger quantity of nutrients (including P) as partially digested pseudofaeces and feces on the surrounding sediments/lakebed (Nalepa et al., 1991). The smaller range of δ <sup>18</sup>OP\_DM compared to δ <sup>18</sup>OP\_from LB samples might be due to differences in the quantities of P<sup>i</sup> derived from direct excretion vs mineralization of pseudofecal and fecal material, as our incubation time required to generate sufficient Pi for recovery (∼10µM; 12–36 h) were of sufficient duration to permit accumulation of feces and psuedofaeces, and although we did not measure microbial activity, decomposition of fecal and pseudofecal material generally occurs over a longer time frame than employed here (Roditi et al., 1997). In contrast, our LB samples could contain P<sup>i</sup> excreted directly by dreissenids, or derived from the re-mineralization of pseudofaeces and/or feces. P<sup>i</sup> excreted directly by dreissenids is likely liberated through the action of acid phosphatases (EC 3.1.3.2) in the acidic gut (Sauey et al., 2016), and based on examples in **Table 3**, released P<sup>i</sup> could range between +7.0 and +14‰. Many of our δ <sup>18</sup>OP\_DM fall within this range, as do δ <sup>18</sup>O<sup>P</sup> from LB samples at 10 m sites. The generally consistent overlap between δ <sup>18</sup>OP\_DM from incubations and δ <sup>18</sup>O<sup>P</sup> from LB sites is further supported by positive relationships observed between the degree of offset of δ <sup>18</sup>O<sup>P</sup> from our experimentally determined δ <sup>18</sup>OP\_DM and dreissenid abundance and shell free biomass (**Figure 9**) and support the role of dreissenid mussels as active recyclers of P in the benthic environment.


TABLE 3 | Examples of mass balance calculations\* for the oxygen isotopic composition of phosphate (δ <sup>18</sup>OP; ‰) released by enzymatic hydrolysis for different enzymes and substrates using literature values for isotopic fractionation (ε).

\*For these calculations, it is assumed that organic P (dissolved or particulate) retains an equilibrium signature (isotope equilibrium value associated with pyrophosphatase P cycling at measured temperatures and δ <sup>18</sup>Ow). The lower and upper values represent the range of possible δ <sup>18</sup>O<sup>P</sup> of released phosphate assuming the δ <sup>18</sup>O<sup>P</sup> of organic P ranges between +15.5 and +18.9, which represent equilibrium values for an average δ <sup>18</sup>O<sup>w</sup> of <sup>−</sup>6.8‰ for eastern Lake Erie, and water temperatures of 23◦ and 4◦C respectively. Mass balance equations for each enzyme and substrate are found in listed references. AcidPase<sup>1</sup> , Potato AcidPase; AcidPase<sup>2</sup> , Wheat germ AcidPase; AlkPase<sup>3</sup> , calf AlkPase; AlkPase<sup>4</sup> , E. coli AlkPase; AMP, adenosine 5′ -monophosphate; GPO4, glycerophosphate; Glu-1-P, glucose-1-phosphate.

Alternative explanations for isotopic enrichment of δ <sup>18</sup>O<sup>P</sup> near the lake bed could include possible desorption of P<sup>i</sup> from clays or iron minerals, or injection of isotopically heavier P<sup>i</sup> from the hypolimnion. Chomicki et al. (2016) suggested that ∼50% of the particulate P leaving the Grand River is non-apatite inorganic P (NAIP), of which iron oxides are a dominant component (Mayer and Manning, 1989). While much of this material likely settles out within a short distance of the river mouth, it is nonetheless subject to resuspension and subsequent dispersal by wind and waves (Chomicki et al., 2016). The structurally complex and rough surfaces created by abundant dreissenid mussels may retain settled material in interstitial spaces and enhanced O<sup>2</sup> demand from fecal material and respiratory activity of mussels (Turner, 2010) could promote release of P<sup>i</sup> from redox sensitive P bearing sediments. While we did not directly measure the δ <sup>18</sup>O<sup>P</sup> of redox sensitive NAIP, we consider this to be an unlikely explanation for the persistently heavy δ <sup>18</sup>O<sup>P</sup> near the lake bed. Jaisi et al. (2010) have shown that P<sup>i</sup> adsorbed to mineral surfaces are constantly exchanged with P<sup>i</sup> in the surrounding water such that fractionation associated with adsorption/desorption is negligible over a matter of hours. Since NAIP particles near the Grand River are small (4–7µm; Chomicki et al., 2016) and likely to stay in suspension for several days (DePinto et al., 1981), we would expect P<sup>i</sup> retained on NAIP particles that eventually settles within mussel beds to be equilibrated with δ <sup>18</sup>O<sup>P</sup> in the water column. If P<sup>i</sup> is released via redox mediated dissolution, we infer that such P<sup>i</sup> would retain δ <sup>18</sup>O<sup>P</sup> closer to that of lake water, and would be unlikely to explain the higher δ <sup>18</sup>O<sup>P</sup> values observed near the lake bed.

Entrainment of cooler water from the hypolimnion could theoretically account for near bed enrichment via (1) remineralization of sinking P by bacteria or (2) release of P<sup>i</sup> from anoxic sediments. In the open ocean, δ <sup>18</sup>O<sup>P</sup> shifts toward the theoretical equilibrium with increasing depth as sinking organic matter is re-mineralized and microbes become carbon limited (Colman et al., 2005). Demand for P<sup>i</sup> by phytoplankton and activity of hydrolytic enzymes should also be reduced in the cooler and darker hypolimnion compared to the illuminated surface layers (Pick, 1987). Cooler temperatures would favor an isotopically heavier δ <sup>18</sup>OP\_EQ. Upwelling has been identified as a potentially significant transport mechanism for hypolimnetic SRP in some years (Valipour et al., 2016) which could affect our measures of δ <sup>18</sup>OP. Wind speed and direction data from buoy #C45142, prior to and during our sampling trips did not indicate the presence of conditions favorable for upwelling (Supplemental Figure S2), nor did we observe any deviations in water temperature profiles during surveys associated with hypolimnetic intrusions (data not shown), thus we infer that upwelling cannot explain the heavier δ <sup>18</sup>O<sup>P</sup> observed near the lake bed. Similarly, Paytan et al. (2017) have linked release of isotopically heavier δ <sup>18</sup>O<sup>P</sup> in the central basinto sediment release during hypoxic events. However, hypolimnetic oxygen concentrations in the much less productive eastern basin rarely fall below 0.3 mM during the stratified period (Carrick, 2004). Consequently, high rates of sediment P release from internal loading is unlikely to explain the δ <sup>18</sup>O<sup>P</sup> measured at our 10 m sites.

The lack of a similar degree of isotopic enrichment in the LB samples at 1 and 3 m sites is not easily explained. On one hand, the relationship between δ <sup>18</sup>O<sup>P</sup> and dreissenid biomass might suggest that dreissenid biomass at these shallow depths was insufficient to generate a strong isotope signal. Rapid uptake of released P<sup>i</sup> by Cladophora might also attenuate an isotopic signal, however, Cladophora biomass in 2014 was relatively low, in large part because water clarity in 2014 was poor owing to high inputs of suspended solids from the Grand River (Valipour et al., 2016). It is also possible that the length of time required (∼1 h) to collect sufficient sample volume (∼200 L) whilst minimizing the disturbance near the lake bed may have compromised our samples to varying degrees if vertical mixing was sufficient to mix surface waters (containing lighter δ <sup>18</sup>OP) with bottom waters (containing heavier δ <sup>18</sup>OP). We evaluated this possibility by calculating the characteristic time scale for diffusive mixing of the water column following Boegman et al. (2008)

$$T\_D^{\text{mix}} = \frac{N^2}{3} \left(\frac{H}{1.4u^\*}\right)^3 \tag{4}$$

where N is the buoyant frequency (calculated as <sup>r</sup> −g/ρ<sup>0</sup> ∂ρ ∂z , H is the water column depth (m), and u∗ is the surface shear velocity (calculated as qτ<sup>w</sup> ρ0 ; where τ<sup>w</sup> can be estimated from CDρAU 2 10e −βz ; with CD<sup>=</sup> 1.3 <sup>×</sup> <sup>10</sup>−<sup>3</sup> , <sup>ρ</sup><sup>0</sup> <sup>=</sup> 1,000 kg m−<sup>3</sup> , ρ<sup>A</sup> = 1.2 kg m−<sup>3</sup> , and U<sup>10</sup> is the average daily wind speed (m s−<sup>1</sup> ) adjusted to a height of 10 m above surface recorded at buoy #C45142). Note since τ<sup>w</sup> was evaluated at surface, the exponential term is neglected (Boegman et al., 2008). For many samples, it is clear that the time to completely mix the water column was shorter than our sample collection time (**Figure 10**)—thus we cannot exclude the possibility that our samples may have been compromised by mixing of lower δ <sup>18</sup>O<sup>P</sup> from surface waters. Furthermore, the degree of δ <sup>18</sup>O<sup>P</sup> offset from δ <sup>18</sup>OP\_DM appeared to be reduced as T mix D increased for samples collected in May and June when density gradients were greatest (**Figure 10**). Although the low sample size likely precludes statistical significance, these calculations are consistent with our hypothesis.

#### CONCLUSIONS

One of our original objectives was to assess the utility of δ <sup>18</sup>O<sup>P</sup> to delineate the impact of riverine P<sup>i</sup> in the nearshore and determine any potential association with Cladophora blooms. This proved to be challenging as the δ <sup>18</sup>O<sup>P</sup> of riverine P<sup>i</sup> appeared to be rapidly altered by in-river and in-lake processes. Based on the nature and magnitude of deviations from δ <sup>18</sup>OP\_EQ, δ <sup>18</sup>OP, we infer that internal recycling of DOP by the plankton community is a dominant pathway determining the overall δ <sup>18</sup>O<sup>P</sup> of the available P<sup>i</sup> pool in eastern Lake Erie. In addition, our

FIGURE 10 | Plot of the offset of δ <sup>18</sup>O<sup>P</sup> in LB samples from the average experimentally determined δ <sup>18</sup>OP\_DM (11.9 <sup>±</sup> 1.1‰) against the estimated time to completely mix the water column (T mix D ) for samples collected 0.1 m above bottom at stationsof 3 m depth.

data provide compelling evidence for the release of P<sup>i</sup> at the lake bed, facilitated by dreissenid mussels. Management actions aimed at reducing the extent and severity of Cladophora blooms in Lake Erie will likely require consideration of linking the magnitude of external P inputs to the in-lake growth response in a more complex framework than prior to the advent of dreissenid populations. The ability of dreissenid mussels to function as a P trap in the nearshore as proposed by Hecky et al. (2004) poses distinct challenges. Dreissenid mussels in eastern Lake Erie are highly reliant on phytoplankton (Campbell et al., 2009), and the open waters represents a large reservoir of P that would have been previously unavailable to Cladophora. At the same time, local inputs of P<sup>i</sup> from tributaries may be rapidly incorporated into phytoplankton, which while formerly unavailable to Cladophora, must now pass through the littoral zone which is dominated by expansive mussel populations and such P may ultimately be trapped and made available in the nearshore through the feeding activity of mussels. While it is clear that reductions in P inputs are the appropriate management tool, further work will be required to better quantify if reductions in Cladophora growth are achievable and whether these changes will require more stringent P controls than already proposed for Lake Erie.

#### AUTHOR CONTRIBUTIONS

DD and VH-B designed the study. DD carried out the study. GK conducted isotope analysis. DD wrote manuscript.

#### FUNDING

This study was funded by Environment Canada through the Great Lakes Nutrient Initiative.

#### ACKNOWLEDGMENTS

The authors thank the Technical Operations Services at Environment and Climate Change Canada for field and laboratory support. Greg Boulton, Karina Tsang, Danielle Bruyns, and Jacqui Milne provided tireless assistance in the field. Alice Dove provided nutrient data collected at York, ON,

#### REFERENCES


and data on dreissenid abundance and biomass Sandra Cooke (Grand River Conservation Authority) provided valuable input on the Grand River ecology. Comments and advice regarding purification protocols from Federica Tamburini, Adina Paytan, and Deb Jaisi are appreciated and were essential for improving sample processing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00215/full#supplementary-material


in the Great Lakes Region. Can. J. Fish. Aquat. Sci. 48, 1381–1388. doi: 10.1139/f91-165


Erie, 2002. J. Gt. Lakes Res. 31, 223–237. doi: 10.1016/S0380-1330(05) 70316-6


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Depew, Koehler and Hiriart-Baer. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Colorimetric Chemical Differentiation and Detection of Phosphorus in Eutrophic and High Particulate Waters: Advantages of a New Monitoring Approach

Lisa Felgentreu<sup>1</sup> \*, Günther Nausch<sup>1</sup> , Franziska Bitschofsky <sup>2</sup> , Monika Nausch<sup>2</sup> and Detlef Schulz-Bull <sup>1</sup>

#### Edited by:

Barbara Cade-Menun, Agriculture and Agri-Food Canada (AAFC), Canada

#### Reviewed by:

Raghab Ray, University of Tokyo, Japan Sheree J. Watson, University of Hawaii at Manoa, United States

\*Correspondence:

Lisa Felgentreu lisa.felgentreu@io-warnemuende.de

#### Specialty section:

This article was submitted to Coastal Ocean Processes, a section of the journal Frontiers in Marine Science

Received: 29 January 2018 Accepted: 29 May 2018 Published: 19 June 2018

#### Citation:

Felgentreu L, Nausch G, Bitschofsky F, Nausch M and Schulz-Bull D (2018) Colorimetric Chemical Differentiation and Detection of Phosphorus in Eutrophic and High Particulate Waters: Advantages of a New Monitoring Approach. Front. Mar. Sci. 5:212. doi: 10.3389/fmars.2018.00212

<sup>1</sup> Department of Marine Chemistry, Leibniz-Institute of Baltic Sea Research, Warnemünde, Germany, <sup>2</sup> Department of Marine Biology, Leibniz-Institute of Baltic Sea Research, Warnemünde, Germany

Phosphorus (P) is a key factor forcing eutrophication in limnic and marine systems, and all monitoring programs for water quality accordingly include P determinations. However, traditional monitoring does not allow an analysis of the different components involved in the P cycle taking place in the water column. Nonetheless, the implementation of measures addressing eutrophication requires a full understanding of the processes involved in the transformation and transport of P, in all its chemical forms. In this study, the P categories present in a river and its estuary in northern Germany, which discharge into the Baltic Sea, were characterized. Using the molybdenum blue method we found that the classification of P into the traditional fractions (DIP, DOP, POP) applied in the ocean cannot be applied to turbid waters such as rivers because interferences between the fractions seems to occur. Therefore a new nomenclature has been introduced. In addition to total phosphorus (TP) and dissolved molybdate-reactive phosphorus (DRP; previously referred to as inorganic phosphorus), dissolved non-molybdate-reactive phosphorus (DNP), particulate molybdatereactive phosphorus (PRP), and particulate non-molybdate-reactive phosphorus (PNP) were distinguished. The high spatial and temporal variations in the proportions of these forms with respect to the TP concentration well-demonstrate the complexity of the P cycle and the involved P fractions and emphasize the need for expanded monitoring approach. The potential of eutrophication could be underestimated if not all P categories were considered. With the new operational nomenclature the common and standardized molybdenum blue reaction could be used to implement the analysis of various P components into regular monitoring programs.

Keywords: phosphorus categories, differentiation, detection, monitoring program, contribution to eutrophication, Baltic Sea

# INTRODUCTION

To address eutrophication in European waters and adopt measures aimed at restoring their good ecological status, the European Water Framework Directive and the European Marine Strategy Framework Directive were established. However, phosphorus (P) is a critical element forcing eutrophication in aquatic systems throughout the world, especially in surface waters (OECD, 1982; Hecky and Kilham, 1988). Moreover, eutrophication is reinforced by the presence of internal P cycles, for example, the P released from the sediment under anoxic conditions.

Both the cycling of P and the intense P transformations in the environment have been characterized and include biotic (uptake and release by organisms) and abiotic (adsorption onto and desorption from particles, precipitation and dissolution, advection and diffusion) processes in which both dissolved and particulate P forms are generated. In the literature, these functionally distinct forms of organic and inorganic P that make up the P cycle have been variably named (Jarvie et al., 2002; Withers and Jarvie, 2008; Worsfold et al., 2016). Traditional terms for the same P component include orthophosphate (ortho-P), dissolved inorganic P (DIP), or simply phosphate (PO3<sup>−</sup> 4 ), but also soluble reactive P (SRP/srP), dissolved reactive P (DRP), filterable reactive P (FRP), and molybdate-reactive P (MRP) are used. While at times confusing, the use of operational (e.g., FRP, used in analytical determinations) and functional (e.g., DIP, as applied to analyses of an ecosystem or microbial forms) terminologies can also be possible. The functional differentiation of organic and inorganic P requires a big effort which takes much time. It is hardly realizable to measure organic P in its entirety rather its single compounds like DNA and RNA or ATP which have a minor part in the DOP (Unger et al., 2013). Furthermore, when using the molybdenum blue method, labile P components other than DIP, e.g., labile organic P or colloidal bound P, may also be bound to the molybdenum blue complex (Rigler, 1968; Stainton, 1980; Haygarth and Sharpley, 2000; Jarvie et al., 2002).

Our studies within the PhosWaM project (phosphorus from the source to the sea) have focused on the Warnow catchment area (**Figure 1**), specifically: (1) P fractions in the Warnow river and its estuary, as a transition zone between an agrarian catchment area and the Baltic Sea, and (2) P fluxes, especially with respect to the transport and exchange of P with coastal waters. Rivers play a key role in transporting and processing P from source to sea (Withers and Jarvie, 2008), especially when diffuse sources are the main factor for eutrophication after the elimination of point sources. During the second half of the twentieth century, environmental inputs of P have increased worldwide (Gustafsson et al., 2012). Despite numerous efforts to reduce nutrient inputs (Backer et al., 2010) into the coastal waters of the southern Baltic Sea, inputs of P have led to a shift in the trophic status and the degree of eutrophication remains high (Nausch et al., 2011). Effective measures are therefore needed to control P inputs reaching rivers and estuaries characterized by high organic loads. Although a large number of monitoring programs targeting the nutrient status of these waters are in place, they typically measure only two forms of P, total P (TP) and DIP (Baldwin, 1998; Jarvie et al., 2002; Jescovitch et al., 2017; Stammler et al., 2017). However, because all P fractions can contribute to eutrophication and all of them can be successfully targeted for reduction, the proportions of each one must be considered. For example, among the various forms of P, dissolved organic P is counted to be the most relevant P source for the growth of algae and bacteria during the productive season (Nausch and Nausch, 2006) and is therefore a major contributor to eutrophication.

Reducing the P that contributes to eutrophication requires a revised approach to P monitoring and is one of the primary aims of the PhosWaM project. In this study we demonstrate the importance of measuring the various components of P and their implementation into regular monitoring measures. Therefore, we introduce a nomenclature for the different operational P forms, one that is tailored to those in the turbid riverine and estuarine waters of the investigated area (lowland agricultural dominated catchment) and applicable to similar areas. Additionally, the nomenclature is more accurate according to the widely common and standardized molybdenum blue colorimetric method.

# MATERIALS AND METHODS

Phosphorus (P) categories were analyzed in surface water samples taken monthly since August 2016 along the river Warnow and its estuary to the Baltic Sea (**Figure 1**). The Warnow river, in which the sampling station "Bützow" is located, has a low current velocity. The Warnow estuary, where the station "Hundsburg" is located, is a typical partially mixed estuary. The sampling station "Baltic Sea" is located at the mouth of the river in the coastal zone of the Baltic Sea.

Our newly established nomenclature differentiates P categories based on their dissolved and particulate forms. The detection of dissolved P is standardized using the molybdenum blue colorimetric method (Murphy and Riley, 1962), which should quantify the amount of DIP, also referred to as ortho-P or SRP. However, when used in a water sample, the molybdenum blue can also bind labile P components as described above (Rigler, 1968; Stainton, 1980; Haygarth and Sharpley, 2000; Jarvie et al., 2002). Using the molybdenum blue method in unfiltered and filtered estuarine and river water samples resulted in remarkable different P values, indicating the presence of a labile particulate P forms, the particulate molybdate-reactive P fraction (PRP). It thus became evident that the traditionally measured and named fractions were not suitable for use in the lowland catchment area investigated in this study. Hence,

**Abbreviations:** DHP, dissolved hydrolysable phosphorus; DIP, dissolved inorganic phosphorus; DNP, dissolved non-molybdate-reactive phosphorus; DP, (total) dissolved phosphorus; DRP, dissolved reactive phosphorus, dissolved molybdatereactive phosphorus (new); DUP, dissolved unreactive phosphorus; FOP, filterable organic phosphorus; FRP, filterable reactive phosphorus; MRP, molybdate reactive phosphorus; ortho-P, orthophosphate; P, phosphorus; PNP, particulate nonmolybdate-reactive phosphorus; PO3<sup>−</sup> 4 , phosphate; PP, particulate phosphorus; PRP, particulate molybdate-reactive phosphorus; SRP/srP, soluble reactive phosphorus; SUP, soluble unreactive phosphorus; TDP, total dissolved phosphorus; TFP, total filterable phosphorus; TP, total phosphorus; TPP, total particulate phosphorus; TRP, total molybdate-reactive phosphorus.

we developed a detection scheme (**Figure 2**) and renamed the traditional terms of P categories (**Table 1**).

#### Analysis of Molybdate-Reactive Phosphorus Categories

For these components the water sample should be analyzed as soon as possible after sampling (Jarvie et al., 2002). Dissolved P and particulate P are separated by filtration (maximum 250 ml) over precombusted (4 h at 450◦C) filters (GF/F-filter, retention range 0.7µm) using a vacuum pressure constantly adjusted to 850 hPa. The P concentrations are measured in duplicates using the molybdenum blue colorimetric method (Murphy and Riley, 1962) in which a blue phosphomolybdate complex is formed. The reaction requires a reductant, typically acidified ascorbic acid solution, and a mixed reagent consisting of ammonium heptamolybdate tetrahydrate-solution, potassium antimony tartrate-solution, and 50% sulfuric acid. The procedure generally follows that of Koroleff (1983) but with the following modifications. After the addition of 1 ml of the mixed reagent to a 50 ml water sample, the absorbance at a wavelength of 885 nm is immediately measured to obtain a turbidity blank. This is followed by the addition of 0.5 ml of the reductant to the sample-reagent mixture. Over a period of 20–60 min (depending on the ambient temperature) the P in the sample reacts with the molybdate to form a bluish complex. The absorbance of this sample is then measured again at 885 nm and the P concentration calculated according to Koroleff (1983).

FIGURE 2 | Scheme used in the detection of the various phosphorus (P) fractions. Mo-reaction, reaction of phosphate with molybdate whereby a blue phosphomolybdate complex is generated; TP, total P; DP, total dissolved P; PP, total particulate P; TRP, total molybdate-reactive P; DRP, dissolved molybdate-reactive P; PRP, particulate molybdate-reactive P; DNP, dissolved non-molybdate-reactive P; PNP, particulate non-molybdate-reactive P.

Rostock), and Baltic Sea (coastal zone of Baltic Sea).

The concentration of the dissolved molybdate-reactive P (DRP) in the filtrate is determined together with the concentration of total molybdate-reactive P (TRP) in the unfiltered water sample. The difference between the TRP and DRP values is the amount of particulate molybdate-reactive P (PRP).

The calibration curve, which is renewed with every new regent and reductant solution, for this method is linear up to 10µM, and in the present study the detection limit was reached at an absorbance of 0.007 (equal to ca. 0.07µM), measured in a 5 cm cuvette. The measured standard (potassium-dihydrogen phosphate) has a concentration of 5µM which fitted in a 10%– variance. At least one blank and one standard was applied at every measurement.

#### Analysis of Total Phosphorus Categories

The determination of TP and total dissolved P (DP) requires a digestion of the sample so that all of the P in the sample is converted to DRP. The previously frozen (−20◦C), stored water samples (exact 40 ml) were thawed and then transferred to Teflon PFA <sup>R</sup> -bottles. The digestion is achieved by the addition of an oxidant (4 ml), usually potassium peroxydisulfate (Koroleff, 1983). The oxidation was done in an alkaline medium.

The digestion was carried out in a lab-microwave. In this study, the CEM MARS Xpress and the following protocol were

TABLE 1 | Comparison of the traditional and new nomenclatures used to describe the different phosphorus (P) categories.


used: After a short initialization phase of a few seconds, the oxidant-containing samples were subjected to a 70 min heating phase during which they were heated to 180◦C. This temperature was held for 50 min, after which the samples were cooled down over 30 min in the microwave to 70–80◦C. The amount of DRP in the completely cooled samples was then measured using the molybdenum blue colorimetric method, in a mixture containing 0.8 ml of reductant and reagent. The samples were analyzed as duplicates.

The TP concentration was determined in the unfiltered water sample, and the DP concentration in the filtrate. The total particulate P (PP) concentration was calculated as the difference between TP and DP.

Every new solution of the oxidant is calibrated with standards concentrated between 0 and 20µM. In each digestion two to three blanks and two 2 µM-standard (glucose-6-phosphate) were included. The detection limit was reached at an absorbance of 0.008 (equal to ca. 0.08µM), measured in a 5 cm cuvette. The variance of the 2 µM-standard was within the 5% of the reference value. The measurement of the standard was linear up to a concentration of 15µM.

#### Analysis of Non-molybdate-reactive Phosphorus Categories

The two non-molybdate-reactive P categories were determined by calculation. Dissolved non-molybdate-reactive P (DNP) was defined as the difference between DP and DRP. Thus, the particulate non-molybdate-reactive P (PNP) was calculated out as the difference between PP and PRP (**Figure 2**).

#### RESULTS

In our area of investigation (**Figure 1**) the concentration of the different P categories vary seasonally and spatially. In general the TP concentrations decreased from the Warnow river to the Baltic Sea (**Table 2**) and thus so did the various P fractions, except PRP,

TABLE 2 | The variability of the concentration ranges of the phosphorus (P) fractions at the different stations of the study vs. that reported in the literature for different study areas.


<sup>a</sup>P categories measured in traditional monitoring.

<sup>b</sup>mean values in winter 2007.

<sup>c</sup>different stations at the Great Lakes (USA) with mixed type of land use in adjacent areas.

<sup>d</sup>Three Gorges Reservoir (China).

which increased toward the Baltic Sea. The contribution to TP of the measured P forms varied over a wide range. Thus, at the river station (Bützow, **Figure 1**) DRP concentrations ranged between 2.4 and 65.7%, DNP concentrations between 0 and 100%, PRP concentrations between 7.5 and 53.4%, and PNP concentrations between 0 and 55.8%. Similarly high variations were measured at the estuarine station (Hundsburg, **Figure 1**), where the DRP concentrations were between 0.3 and 100%, the DNP and PRP concentrations between 0 and 100%, and the PNP concentrations up to 78.6%. At the "Baltic Sea" station (**Figure 1**), DRP and PRP, as molybdate-reactive P fractions, accounted for 0–100%, and DNP and PNP, as non-molybdate-reactive P fractions, for 0–56.5 and 0–56.7%, respectively. According to this we calculated the difference of the mean values of TP and DRP, as the traditional measured P components (**Table 3**) to show the importance of measuring the different P categories. In "Bützow," the river station, the TP and DRP values differed with 1.97µM. At the station in the estuary the difference was with 1.62µM lower and at the "Baltic Sea" station was a TP-DRP-difference of 0.44µM. In addition, the DNP concentrations as mean values also decreased from 0.38µM in the river (Bützow) to 0.26µM at the "Baltic Sea."

Seasonal variations were exemplified by the data from February 2017 and June 2017 (**Figure 3**). Additional abiotic parameters (water temperature, rainfall, wind, and discharge) and biotic parameters (chlorophyll a and seston) were given in **Table 4**. In February 2017 it was stormy during the sampling at the stations "Hundsburg" and "Baltic Sea" and also in June

TABLE 3 | Differences of TP and DRP/DIP as traditional monitoring parameter in comparison with differences of DP and DRP (DNP/DOP) at the three stations of the study area as mean values from September 2016 to 2017.


2017 at the river station in "Bützow." Moreover, the salinity situation was different in the 2 months (**Figure 4**). In February, the water column was locally nearly totally mixed at the different stations. In June, however, the water column was a bit stratified with freshwater (1 to 8 g kg−<sup>1</sup> ) at the surface and saltwater (8 to 10/11 g kg−<sup>1</sup> ) below the surface layer to the bottom layer. Furthermore, the chlorophyll a concentration was higher in February 2017 than in June 2017 at the stations "Bützow" (25.10µg l−<sup>1</sup> in February and 8.76 µg l−<sup>1</sup> in June) and "Baltic Sea" (4.05µg l−<sup>1</sup> in February and 2.63 µg l−<sup>1</sup> in June). In contrast, the Secchi depth at "Hundsburg" and "Baltic Sea" were in February deeper than in June. At all stations TP concentration is higher in June than in February. In February the particulate fractions dominated both at the "Baltic Sea" (∼60% PRP from a TP of 0.89µM) and the river (∼20% PRP and 50% PNP from a TP of 2.01µM) stations. By contrast, at the estuary station the dissolved fractions dominated (∼50% DRP and 30% DNP from a TP of 1.07µM). In June, the conditions were reversed. At the "Baltic Sea" station (TP = 0.95µM), DRP accounted for ∼40% and DNP for ∼30%, and at the river station (TP = 3.03µM) DRP for 60% and DNP for ∼20%. At the estuary (TP = 1.96µM), PNP dominated, with ∼60%.

# DISCUSSION

#### Detection of Different Phosphorus Categories and Their Impact

The nomenclature presented herein is an operational one and was developed to address the large variety of terms used to describe the same P form, and the resulting confusion (**Table 5**). The terms we introduced for the different P categories are consistent for us and our further work in the project. Moreover, it allows a more accurate description of the P fractions usually detected in monitoring programs (**Table 2**), especially in eutrophic waters with high amount of particles like our study area (**Table 3**). It does not replace the functional terms of organic and inorganic P forms, which are chemically-defined on the bonding of P with carbon. Instead, the proposed nomenclature has its antagonist in the traditional nomenclature (**Table 1**). Nevertheless, the functional terms of P categories could not be differentiated



\*Measured where the river change into the estuary.

accurately by the common and standardized molybdenum blue colorimetric method.

While the molybdenum blue colorimetric method for the analytical detection of P categories has been used for many years, discrepancies have often been reported (e.g., Murphy and Riley, 1962; Rigler, 1968; Baldwin, 1998). For example, the terms SRP and ortho-P have been variably used to refer often to the same P fraction (**Table 5**; Jarvie et al., 2002) although they do not describe the same fraction (Rigler, 1968; Stainton, 1980). Their synonymous use may reflect the fact that labile forms of organic P might be hydrolyzed in the acidic milieu of the molybdenum blue reaction (Rigler, 1968; Baldwin, 1998; Jarvie et al., 2002). Additionally, the P associated with colloids could be released during the reaction between molybdate and phosphates (Stainton, 1980; Baldwin, 1998). Therefore, a consistent operational nomenclature is necessary to prevent misinterpretation of results from molybdenum blue method. For example, DNP is indeed DOP but it will be underestimated because an amount of it is measured with the DRP which is dominated by DIP. In literature most studies and environmental reports refer to functional terms of P forms (e.g., organic, inorganic), which are also detected by molybdenum blue method (e.g., Nausch and Nausch, 2006).

In the traditional monitoring praxis as well as in most environmental studies (**Table 2**) only TP and DRP (after introduced nomenclature) are measured. On this basis the ecological status of an aquatic system is evaluated and potential aquatic management measures are developed. The DRP is presumed to be the P category available for phytoplankton and therefore the main source for eutrophication. **Table 3** shows, that with ∼40 to 50% of TP the DRP amount in the aquatic environment is not specified. This results in an underestimation of eutrophication potential, since at least DNP is known to be bioavailable (Nausch and Nausch, 2006). An advantage of the additional measurement of TRP is the calculation of PRP, which, as demonstrated by our results, plays an important role in the P cycle of the study area, especially during regeneration periods. Furthermore, because PRP is bound in labile form to particulate material, it is potentially bioavailable.

A problem of the method occurs with the unfiltered water samples which often had a high turbidity as we worked in waters with a high level of eutrophication (Nausch et al., 2011). The presence of particles that remain after the digestion can result in an underestimation of TP (Jarvie et al., 2002). In this study, this was avoided by exemplarily diluting the samples to a median sample:Milli-Q water ratio of 4:3 and 1:2.

Furthermore, because of the pressure exerted by the vacuum pump during the filtration (Rigler, 1968), the cells might collapse such that P of particulate origin contaminates the dissolved P fractions. This would lead to an underestimation of the particulate P fractions. To avoid this problem, we prevented crowding of the cells as described above.

Apparently, the traditional nomenclature is adaptable in less eutrophic waters or rather in waters with less turbidity like


Murphy and Riley (1962); Rigler (1968); Baldwin (1998); Jarvie et al. (2002); Withers and Jarvie (2008); Worsfold et al. (2016).

the station "Baltic Sea" (**Table 3**). Here, the difference between TP and DRP/DIP are less and the amount of particulate P is negligible. Thus, the proposed P categories could be used in waters with less turbidity but are advisable in waters with a high level of eutrophication and particles.

#### Differentiation of the Phosphorus Categories

The P fractions (TP and DRP) of the traditional monitoring (Nausch et al., 2011; Stammler et al., 2017; Han et al., 2018) and that of the extended programs (Nausch et al., 2017, own data) are highly variable, regardless of the chosen study area (**Table 2**). This was evidenced by the similar ranges of our own data from the estuary and the Baltic Sea and the mean TP and DRP values reported by Nausch et al. (2011). Especially in summer, the export of dissolved P forms out of the Warnow river may support primary production in the estuary and the southern Baltic Sea. Therefore, the proportion of PNP increased. The general trend of a decrease in TP towards the Baltic Sea suggested either the dilution of inflowing Baltic Sea water (**Figure 4**) or smaller inflows in the municipal area of Rostock together with the sedimentation of particulate P forms. In a seasonal comparison, the TP concentrations were higher in June 2017 than in February 2017. The reason therefore is a higher biomass in June than in February. In February the phytoplankton is dominant whereas in June the community consists of phytoplankton and zooplankton. Additionally, the Secchi depth is lower in June than in February which points to higher biomass in June.

In contrast to TP the PRP fraction was larger in February than in June at all three stations. This may have been a consequence of the high turbidity caused by a storm event that occurred during sampling in February, especially at the station "Baltic Sea." Higher turbidity and PRP concentrations could be positively correlated because dissolved P adsorb on particular material like on resuspended sediment (Stainton, 1980). Except for the river station, the amount of PNP in February was less than in June, most likely due to particulate material that was autochthonously built and not exported. Moreover, the chlorophyll a concentration was in "Bützow" in February higher than in June so that the high PNP data could be explained by a possible phytoplankton bloom.

The values of the TP and DRP concentrations reported by Stammler et al. (2017) in their study of the river system associated with the Great Lakes (USA) and by Han et al. (2018), who analyzed P concentrations in the Three Gorges Reservoir (China), were higher than those determined at the Warnow stations representing the Baltic Sea and the Warnow estuary but also analyzed by Nausch et al. (2011). They were also higher than those measured in a portion of the Warnow catchment area (Nausch et al., 2017). This large variety in TP and DRP may reflect regional differences between the different study areas. Additionally, according to Stammler et al. (2017), the deviation in the range of P values is allegeable by the different sampling rhythms. The data used in that study were collected between 1979 and 2011, with monthly sampling during the growing season (April to November). Our data were obtained from monthly samples collected between September 2016 and September 2017. On the other hand, Han et al. (2018) used data from four samplings timed according to the water cycle of the examined reservoir. In that study, sampling was performed in October 2014 (impounding period), January 2015 (high water level period), July 2015 (low water level period), and April 2016 (sluicing period).

Our results demonstrate that the various P forms occurring in aquatic systems depend on the season and the hydrology. Further research is needed to specify the single P components and their function within the P cycle. They also showed that the transformation processes taking place between the single P fractions are not reflected in the traditional monitoring, which detect only TP and DRP concentrations.

#### CONCLUSION

In traditional monitoring programs, only TP and DRP concentrations are measured. This does not reflect the diversity and complexity of P cycle in aquatic environment and underestimates the eutrophication potential. The determination of all P categories is important both for comparison to data reported in the literature and to own data within a single multiyear study. However, because P forms are not stable, spatially and even within the same season, all P categories must be measured if effective nutrient reduction is to be successfully implemented. Additionally, an understanding of the function of the individual P components, such as their bioavailability, is important to determine to identify their contribution to eutrophication.

Furthermore, it is necessary to communicate research results to stake holders and implement the measurement of all P categories into monitoring strategies. Therefore, a simple and standardized method and concise nomenclature is helpful. The molybdenum blue reaction is a common and standardized method to determine P. The operative nomenclature, we presented in this study, based on reactivity of P with molybdenum blue. This should simplify research communication and prevent misinterpretation due to the functional terms of organic and inorganic P given to operational differentiated P components.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

All authors were involved in contributing the conception and design of the study and performed the experiments; FB and LF analyzed the data; LF wrote the first and following drafts of the manuscript; DS-B supported LF during the writing process with ideas for the manuscript design. All authors are involved in manuscript revision, read, and approved the submitted version.

#### ACKNOWLEDGMENTS

We thank Christoph Kamper and Carmen Esser for their technical assistance in the lab. Additionally, we thank Sandra Jahn who aided during the development of the filtration scheme. This study is part of the PhosWaM project (Phosphorus from source to sea—Integrated phosphorus and water resources management for sustainable water protection), funded by the Federal Ministry of Education and Research, Germany (FKZ 033W042A). It also was performed within the scope of the Leibniz ScienceCampus Phosphorus Research Rostock.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Felgentreu, Nausch, Bitschofsky, Nausch and Schulz-Bull. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Influence of Riverine Nutrients in Niche Partitioning of Phytoplankton Communities–A Contrast Between the Amazon River Plume and the Changjiang (Yangtze) River Diluted Water of the East China Sea

#### Edited by:

Christian Lønborg, Australian Institute of Marine Science (AIMS), Australia

#### Reviewed by:

María Froján, Instituto de Investigaciones Marinas, Vigo, Consejo Superior de Investigaciones Científicas (CSIC), Spain Bingzhang Chen, Tokyo University of Marine Science and Technology, Japan Maria Aranguren-Gassis, University of Vigo, Spain

#### \*Correspondence:

Helga do Rosario Gomes helga@ldeo.columbia.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 05 April 2018 Accepted: 06 September 2018 Published: 25 September 2018

#### Citation:

Gomes HdR, Xu Q, Ishizaka J, Carpenter EJ, Yager PL and Goes JI (2018) The Influence of Riverine Nutrients in Niche Partitioning of Phytoplankton Communities–A Contrast Between the Amazon River Plume and the Changjiang (Yangtze) River Diluted Water of the East China Sea. Front. Mar. Sci. 5:343. doi: 10.3389/fmars.2018.00343 Helga do Rosario Gomes<sup>1</sup> \*, Qian Xu<sup>2</sup> , Joji Ishizaka<sup>2</sup> , Edward J. Carpenter<sup>3</sup> , Patricia L. Yager<sup>4</sup> and Joaquim I. Goes<sup>1</sup>

<sup>1</sup> Lamont Doherty Earth Observatory at Columbia, Palisades, NY, United States, <sup>2</sup> Institute for Space-Earth Environmental Research, Nagoya University, Nagoya, Japan, <sup>3</sup> Estuary & Ocean Science Center, San Francisco State University, Tiburon, CA, United States, <sup>4</sup> School of Biology, Georgia Institute of Technology, Atlanta, GA, United States

Riverine nutrients act in concert with local hydrographic conditions to create distinct ecological niches for phytoplankton communities across river-ocean continuums. Here we compare two of the world's largest river-ocean systems, the Amazon River Plume (ARP) which outflows into the Western Tropical North Atlantic and the Changjiang Diluted Water (CDW) which empties into the East China Sea to show how distinctly different N: P supply ratios of their source waters, shape phytoplankton communities along the river-ocean continuum. Sampling in the relatively unpolluted surface waters of the ARP during peak river discharge revealed that phytoplankton communities along the riverocean continuum were strongly limited by Dissolved Inorganic Nitrogen (DIN, nitrate plus nitrite) which was low or beyond detectable, while Dissolved Inorganic Phosphorous (DIP, phosphate) and Silica were not limiting. The resulting low N:P supply ratio allowed diazotrophs to co-exist with non-diazotrophs. Diatom-Diazotroph Associations (DDAs) such as Hemiaulus hauckii-Richelia proliferated, while in the oligotrophic oceanic waters, Trichodesmium spp. thrived. In contrast, in the CDW, anthropogenic nitrogen inputs from human pressures in the Changjiang River system has led to a system where the changing supply rate of the single nutrient (DIP) is responsible for the interannual variability seen in the phytoplankton community structure of the CDW. During years of low discharge, DIP limitation can be ameliorated by on-shelf upwelling of DIP rich Kuroshio Intermediate Waters leading to domination of diatoms and dinoflagellates. Conversely, during years of heavy discharge, the westward flowing CDW plume was severely DIP limited, probably because water column stratification dampened upwelling of subsurface waters. The consequent DIP limitation led to the proliferation of small phytoplankton such as Chlorophytes and Cyanobacteria. The absence of diazotrophs in the CDW, leads us to hypothesize that river-ocean continuums, whose source waters

**85**

are heavily impacted by anthropogenic activities and with high nitrate concentrations often substantially in excess of Redfield ratios, may not support diatoms offshore on account of DIP limitation nor diazotrophy because of excess DIN.

Keywords: Amazon River Plume, Changjiang River Diluted Water, East China Sea, resource competition, nutrient stoichiometry, phytoplankton communities, diazotrophy

#### INTRODUCTION

fmars-05-00343 September 24, 2018 Time: 16:49 # 2

Rivers are the primary conduit transporting weathered, leached, and human-derived material from land to the oceans (Sharples et al., 2017). In addition to carbon, rivers export nitrogen, phosphorus and silica which are the key potentially limiting nutrients required for phytoplankton growth. Thus riverine nutrients can not only augment primary productivity but also contribute to regulating the long-term biological productivity of the ocean and hence ocean carbon storage (Jickells et al., 2017). Several studies show that anthropogenic disturbance of river nutrient loads and export to coastal and ocean marine systems has increased, creating a global problem affecting water quality, and biodiversity (Bouwman et al., 2005). Increasing inputs of nitrogen and phosphorous from human activity, predominantly from land-based activities, now have the potential to modify oceanic, and even global, biogeochemical systems. The total river input of nitrogen and phosphorous to the coastal seas has approximately doubled over the last few hundred years (Seitzinger et al., 2010; Beusen et al., 2016; Jickells et al., 2017). Nitrogen inputs are a consequence of urbanization, sanitation, development of sewerage systems, and lagging wastewater treatment, as well as increasing food production and associated inputs of N fertilizer, animal manure, atmospheric N deposition, and biological N fixation in agricultural systems. P has increased through the use of rock phosphate as fertilizer, detergent additives, animal feed supplements etc. (Seitzinger et al., 2010). Sharples et al. (2017) calculated that globally 75 and 80% Dissolved Inorganic Nitrogen (DIN) and Dissolved Inorganic Phosphate (DIP) respectively, reaches the open ocean although these are considered as the upper limits as estuarine processes were ignored in the study. However, there is a significant and systematic spatial variability in supply materials to the open ocean as well as the amount and kinds of nutrients supplied. Sharples et al. (2017) reported that the proportion of nutrients reaching the open ocean, from tropical and subtropical rivers tend to be the most important for nutrient delivery because the weak Coriolis force allows direct across-shelf movement of river plumes unlike in temperate and polar regions, where the Coriolis force moves freshwater flows along the shelf (Jickells et al., 2017). Hence low latitude rivers such as the Amazon, and the rivers in Southeast Asia including the Changjiang River are major sources of nutrients for the oceans or seas into which they discharge.

As nutrient loads change, so does the stoichiometry of the coastal and open ocean waters resulting mostly in increasing N:P supply ratios and consequent variations in algal communities (Glibert et al., 2014). There has been a large body of work (Anderson et al., 2002; Heisler et al., 2008; Glibert et al., 2010) on eutrophication which report that inland and coastal waters are witnessing increased algal growth and development of high biomass blooms as well as changes in species diversity from increased riverine nutrient loads. However, only a few studies document changes in regions of the open ocean where riverine nutrients are exported and which can influence phytoplankton biodiversity and productivity.

Here we compare two widely disparate river-ocean continuums, the tropical Amazon River Plume (ARP) which outflows into the Western Tropical North Atlantic and the Changjiang Diluted Water (CDW), the plume of low salinity water from the Changjiang River which extends across East China Sea (ECS).

#### Comparison of the Amazon River and the Changjiang River Continuums

The magnitude of discharge (120,000 m<sup>3</sup> s −1 ) from the Amazon River is unique in the global oceans; the Amazon River discharges as much freshwater as the next 8 largest rivers in the world combined (Coles et al., 2013). This tremendous volume of freshwater forms a very dynamic and extensive surface plume that protrudes well offshore into the WNTA (**Figure 1a**). Because the Amazon discharges at the equator and on the western boundary of the ocean, its waters are entrained in energetic boundary currents associated with the North Brazil Current, the North Equatorial Counter Current, and the coastal Guyana Current (**Figure 1a**; Coles et al., 2013). These currents have strong seasonal variations in response to the atmospheric Intertropical Convergence Zone (ITCZ) which is located in the southern position in winter, and northern in summer. Thus, during the winter and spring months, the plume follows the North Brazil Current along the northeastern coast of South America, carrying low salinity water to the Caribbean and in summer and fall, it retroflects to the east, entrained in the North Equatorial Counter Current (Muller-Karger et al., 1988; Coles et al., 2013; Weber et al., 2017). The discharge, introduces tremendous quantities of nutrient-rich water to the Western Tropical North Atlantic, with nutrients being mostly plant derived as it traverses for over 6000 km from the Andes in Peru to the Atlantic Ocean, and covers a watershed that includes the largest tropical rainforest in the world as well as areas of dry grassland, or savannah (Del Vecchio and Subramaniam, 2004).

The Changjiang River, the third largest in the world and the largest outflowing river in China (Yan et al., 2010) discharges into the ECS, one of the world's largest marginal seas (**Figure 1b**) with 70% of its area characterized by a continental shelf slope shallower than 200 m. The freshwater discharge and total

suspended matter from the Changjiang River spreads over the entire shelf (Sukigara et al., 2017), accounting for 90–95% of the total riverine input into the ECS (Zhang et al., 2007). Heaviest discharge is in summer in association with the monsoons with its buoyant plume, known as CDW, spreading eastward over the ECS over an area ca. 85×10<sup>3</sup> km<sup>2</sup> (Zhang et al., 2007). In winter when discharge is lower, the northerly monsoonal winds prevail over the ECS and the Changjiang discharge is restricted to the western side of the ECS with waters moving southward and forming a narrow belt along the coast of China and toward the Taiwan Strait (**Figure 1**; Zhang et al., 2007). Anthropogenic activities appear to be the principal driver of the high DIN and DIP in the coastal environment of the Changjiang River (Zhang et al., 2007) with DIN inputs having increased almost 4-fold from 1970 to 2003 (Yan et al., 2010). In contrast, the Yellow Sea Coastal water flowing from the north, provides only marginal amounts of DIP and DIN (Zhang et al., 2007). High levels of nutrients from land sources brought in by the Changjiang are usually constrained to coastal and inner-shelf regions with nutrient species in surface waters gradually decreasing from eutrophic coastal to oligotrophic open shelf waters depending on the hydrographic stages of the river (Chen, 2008). Highest nitrate concentrations are found near the Changjiang estuary with DIN concentrations often exceeding 50 µM and DIP concentrations >1 µM (Zhang et al., 2007) and in some cases N:P ratios exceeding 100 (Chen and Wang, 1999). However, large phytoplankton blooms primarily of diatoms and dinoflagellates rapidly consume the DIP which is reduced to below 0. 2 µM while DIN is still higher than 1 µM (Chen, 2008) making DIP a limiting nutrient in the shelf waters of the ECS as well as in the mid-shelf and offshore region (Wang and Wang, 2007). Off the shelf break, the northward intrusion of the nutrient-rich, sub-surface Kuroshio waters into the ECS at 50-100m depth also affects the nutrient regime of the ECS (**Figure 1b**). These waters can upwell in summer when winds are favorable (Zhang et al., 2007) alleviating the severe P limitation at least in the ECS shelf region (Yang et al., 2012; Yang et al., 2013; Tseng et al., 2014) Relative to shelf waters, the Kuroshio intrusion waters are high in DIN and DIP (Zhang et al., 2007) providing a much required DIP source to phytoplankton in the euphotic layer.

Compared to the Changjiang River which flows through highly industrialized regions (Beusen et al., 2016) where riverine N and P have increased tremendously, the Amazon river delivers relatively low anthropogenic inorganic nitrogen because of both less intensive catchment agriculture and dilution of nitrogen inputs by the high river flow (Jickells et al., 2017). The nitrate in this source water exists at sub-Redfield ratios, and hence is quickly drawn down by blooms of coastal diatoms that benefit from the DIP and Si-rich water (Stukel et al., 2014). This leaves a plume that extends more than 3000 km from the river mouth (Hu et al., 2004) which is nitrogen-poor but both Si and DIP rich. DIP and Si exhibit patterns that suggest near conservative mixing between high nutrient river water and oligotrophic oceanic water. Additional inputs of DIP from particle leaching both in the

estuary and offshore contribute to the low N:P ratios (DeMaster, 1996) establishing nitrogen limitation early in the life of the plume and propagated offshore (Weber et al., 2017). As the plume extends from the river mouth, it spreads as a thin cap well beyond the river mouth, leading to surface stratification that can impede nutrient inputs and strongly influence the structure of the phytoplankton communities.

While a considerable effort has been made to study the relative impacts of nutrients, physical forcing, and grazing on the phytoplankton niche communities along the Amazon river continuum (Subramaniam et al., 2008; Goes et al., 2014; Stukel et al., 2014; Conroy et al., 2016; Conroy et al., 2017; Weber et al., 2017)., there is sparse information on how the CDW influences the phytoplankton community structure of the ECS. Extensive studies have focused on the nutrient contribution of the Changjiang outflow to the river-ocean continuum as well as the role of sub-surface nutrient rich Kuroshio waters (Zhang et al., 2007; Chen, 2008 and references within) in modulating the nutrient regime of the ECS while others have focused either on the phytoplankton communities of the nearshore Changjiang estuary (Li et al., 2009; Guo et al., 2014) or of the Kuroshio influenced region (Jiao et al., 2005). Our recent study (Xu et al., 2018) improves the situation by examining the phytoplankton community structure in the ECS during the summers of 3 consecutive years from 2009 to 2011 as well in 2013 clearly demonstrating the interannual variability of phytoplankton communities of the ECS associated with variations in the Changjiang River discharge and its spread eastward.

In this study we compare these two highly disparate riverocean continuums to show that variability in the stoichiometry of their waters supports two distinct gradients of phytoplankton communities as the plumes move offshore from nutrient-rich coastal waters to offshore oligotrophic regions.

# MATERIALS AND METHODS

#### Sampling and Hydrographic Properties ARP

In order to investigate the system of phytoplankton communities as they adapt to nutrient limitation and changing N:P ratios we undertook 3 cruises in the ARP (Coles et al., 2013; Goes et al., 2014; Stukel et al., 2014; Weber et al., 2017). This paper describes the findings of the first cruise on board the R/V Knorr from 22nd May to 25th June, 2010 when the discharge was at its maximum.

For a detailed description of the sampling strategies and methods, the reader is directed to Goes et al. (2014) and Weber et al. (2017). Briefly, a total of 25 stations were occupied along a cruise track (**Figure 1a**) which traversed the plume four times and extended well offshore to provide contrast with waters within the plume (Weber et al., 2017). Samples were collected with Niskin <sup>R</sup> bottles attached to a Conductivity-Temperature-Depth (CTD) Sea-Bird Electronics <sup>R</sup> Rosette from 4 depths within the euphotic zone determined from in-water profiles of photosynthetically available radiation (PAR) using a Biospherical Instruments <sup>R</sup> in-water quantum scalar radiometer (Goes et al., 2014). Additionally, surface samples were collected at hourly intervals along the cruise track from the ship's uncontaminated seawater flow-through system, filtered and stored in the same manner as the hydrocast samples. While, samples were also collected from the deep chlorophyll maximum (DCM) and from below the euphotic depth, in this paper we only discuss data from surface samples and information on depth profiles is available in Goes et al. (2014). A highly refined phytoplankton structure of surface waters was obtained using microscopically determined phytoplankton species (Goes et al., 2014) while in this paper some pigment derived phytoplankton classes have also been added.

#### ECS

All studies were conducted on board the T/V Nagasaki Maru in late July of each year (**Figure 1b**). Sampling locations were confined to the ECS, in an area extending from 124.6 to 128.8oE and from 31.4 to 33.0oN. This allowed a better understanding of how phytoplankton composition in the mid-shelf ECS can be impacted by variations in the mixing of different water masses, in particular the flow and the extent of mixing of the CDW and the Kuroshio with their different nutrient contents. A more thorough and detailed description on the methods is available in Xu et al. (2018). Briefly, a CTD was used to obtain profiles of seawater Temperature (T) and Salinity (S) of the upper 80 m. Seawater samples from the surface were collected using an acid washed bucket and analyzed for pigments, and nutrients (DIN and DIP). Nutrient samples were immediately frozen in polyethylene tubes after sampling and transferred under frozen conditions to the shore laboratory for analysis using an Auto-analyzer (AACS-IV, BLTEC).

#### High Pressure Liquid Chromatography (HPLC) Pigment Analysis and CHEMTAX ARP

Total phytoplankton biomass was collected by gently filtering 1–2.5L surface seawater onto 47-mm GF/F filters, which were frozen in liquid nitrogen at sea and transported to shore for analysis by HPLC. Acetone extracted phytoplankton pigments were separated and characterized by methods described in Van Heukelem and Thomas (2001) and Hooker et al. (2005). Currently this method is recognized as one of the most efficient and reliable methods to analyze algal pigments (Serive et al., 2017). HPLC provided separation of almost 15 pigments including the primary pigment, Chlorophyll a (Chl a) and a suite of accessory pigments (carotenoids and chlorophylls) many of which are specific to individual phytoplankton taxa or groups (Mackey et al., 1996; Vidussi et al., 2001; Wright, 2005; Wright and Jeffrey, 2006) and can be used to categorize phytoplankton classes such as diatoms, dinoflagellates, prochlorophytes etc. This was undertaken using CHEMTAX, a free statistical software program that estimates algal class abundances from pigment markers by applying an iterative process to find the optimal ratio of biomarker pigment: Chl a (Mackey et al., 1996; Higgins et al., 2011) and then generates the fraction of the total Chl a

pool belonging to each pigment-determined group (Mackey et al., 1996). Accurate estimates necessitate that the initial or seed ratios be close to those of the phytoplankton populations that are being assessed (Latasa, 2007; Swan et al., 2016). As this constitutes the first report on pigment based phytoplankton distribution of the Amazon plume continuum we do not have pigment ratios exclusively for this region and have relied on the extensive and recently published synthesis of culture and field pigment ratios for major algal groups (**Table 1**; Higgins et al., 2011). These ratios have been used by Swan et al. (2016) for their global analysis and by others for select regions (Armbrecht et al., 2015; Barlow et al., 2016). However, we were also guided in our choice of ratios by microscopy data detailed in our earlier paper (Goes et al., 2014) and the extensive taxonomic report of Wood (1966) for the ARP. Another a priori assumption required for effective CHEMTAX analysis is that the accessory pigment-to-chla ratios should be constant across the data subset under consideration i.e., the phytoplankton community must be relatively homogeneous (Swan et al., 2016). We have minimized this problem by clustering the data into subsets of pigment ratios with similarities (Dandonneau and Niang, 2007; Torrecilla et al., 2011; Wolf et al., 2014) using the statistical package PRIMER (ver. 6.1.13) on log(X + 1) transformed pigment to Chl a ratios from the upper 5m. Considering only linkage distances greater than 50%, our pigment ratios separated into 3 clusters (Bray Curtis similarity index) which comprised of samples from similar salinity ranges as well as coincided with the water types proposed by Goes et al. (2014) based on fluorescence data. Cluster 1 (salinity > 35) comprised mostly open ocean samples, Cluster 2 was made up of mesohaline samples in the extension of the plume (salinity 30–35), and cluster 3 were samples in the plume (salinity < 30). Each data cluster was assumed to be made up of samples of similar phytoplankton populations and CHEMTAX was run separately on each cluster.

The initial ratios or 'true' pigment ratios (**Table 1**; Higgins et al., 2011) were used to generate about 30 'artificial' ratios within a certain range (S. Wright, personal comm.). CHEMTAX was applied to each cluster using these initial 30 matrices. CHEMTAX quantitatively estimates the contribution of algal taxa by iteratively modifying the user-specified Pigment:Chla ratios using a "steepest descent" algorithm to successively reduce the amount of unexplained pigment measured as the RMS of the residuals. The output of the matrix with the smallest RMS error was used to create 30 more matrices. However, prior to selecting this matrix, the ratios were plotted to see that they had stabilized as RMS decreased. When the RMS and the Pigment:Chla ratios had stabilized, the matrices associated with the 6 best RMS were averaged. This 'best' matrix was then compared to the initial matrix to see if the ratios had deviated in which case the matrix was discarded. This procedure was repeated until we were confident that the RMSE was low but more importantly, that the pigment ratios had converged to within known ranges. The output (phytoplankton classes) from the 6 best RMS were averaged to provide the final phytoplankton classes for that cluster (Final Ratios in **Table 1**). Prochlorococcus sp. was quantified exclusively by its unambiguous divinyl Chlorophyll a (DVChla) signature (Chisholm et al., 1992; Swan et al., 2016) and was not included in the CHEMTAX analysis.

#### ECS

Pigment analysis of samples collected in the ECS was undertaken as described in Section 2.1.2 and (Xu et al., 2018). Initial pigment ratios were from (Furuya et al., 2003) and are based on earlier pigment studies in the ECS. CHEMTAX was run as described in Section ARP. Initial and Final Ratios are shown in Xu et al. (2018). For accuracy, (see Section ARP) CHEMTAX was run on surface datasets of 2009, 2010, 2011, and 2013 separately.

#### Phytoplankton Taxonomy **ARP**

Details of sample collection and processing for phytoplankton taxonomy are in Goes et al. (2014). Briefly, phytoplankton cells were filtered onto 8 µm pore size, 47 mm diameter Nuclepore <sup>R</sup> filters and examined under an epifluorescence microscope (Carpenter et al., 1999), and enumerated within 24 h.

TABLE 1 | Initial pigment: Chl a ratios entered into the CHEMTAX program are from Higgins et al. (2011), Swan et al. (2016) and Wright, S. (personal comm.), and Final ratios are the output after completion of CHEMTAX runs.


These ratios are for analysis of pigments from the ARP while similar ratios for the ECS can be found in Xu et al. (2018). Pigment abbreviations are Chlorophyll c3 and c1 (Chl c3, c1), Chlorophyll b (Chl b), Chlorophyll a (Chl a), 19'butanoyloxyfucoxanthin (ButFuc), Fucoxanthin (Fucox), Prasinoxanthin (Prasinox), Violaxanthin (Violax), 19'hexanoyloxyfucoxanthin (Hex-fuco), Zeaxanthin (Zea), and Alloxanthin (Allox).

#### **ECS**

Taxonomy was solely based on pigments and no microscopy was undertaken.

#### RESULTS

#### ARP

The trajectory of the ARP is clearly defined by its salinity gradient (**Figures 1a**, **2a**) as it traveled northward in narrow coastal band, broadening and dispersing off the continental shelf at around 8oN. Salinities < 35 were evident as far north as 18oN and west of 47oW (**Figure 2a**). Based on high frequency fluorescence data we previously delineated three water types along the plume continuum (Goes et al., 2014), which are coincident with the salinity based demarcations of Subramaniam et al. (2008). These water types include an oceanic type of high salinity (salinity > 35), a mesohaline type with salinities from 30-35, and plume waters with salinity < 30.

The ARP clearly influenced the nutrient (DIN, DIP, and SiO3) concentrations of the otherwise oligotrophic Western Tropical North Atlantic. DIN was almost depleted from the waters by large coastal blooms (Goes et al., 2014; **Figure 2b**) and concentrations ranging from 0.1–1.5 µM were found at only 9 locations, 6 of them in the plume. The highest value of 1.5 µM was at 6oN (**Figure 2b**) where depth profiles indicated an upward flux of the nutrient from deeper depths (see vertical profile in Figure 4e of Goes et al., 2014). In contrast, higher concentrations of DIP (**Figure 2c**) and SiO<sup>3</sup> (**Figure 2d**) were measured in the plume than in the surrounding waters indicating riverine influence. Although DIP distribution followed the conservative mixing line at least below 9oN, north of this, there were strong positive deviations along the axis of the plume (Goes et al., 2014; Weber et al., 2017) which have been ascribed to the release of DIP from suspended particles and dissolved organic matter (Goes et al., 2014; Weber et al., 2017). N:P ratios of surface waters were mostly negative indicating severe nitrogen limitation. This is best illustrated using the diagnostic parameter Excess Nitrate (ExN), which measures departure from classic Redfield ratios and is calculated as ExN = DIN–(R∗DIP) (R = Redfield N:P ratio of 16 ). ExN values of < 0 indicate DIP enrichment, while ExN > 0 µM (N:P > 16) indicates the converse (Wong et al., 1998). ExN was consistently between −17.5 and −7.5 as the plume moved northwestward indicating severe DIN limitation. Low ExN arises from a combination of rapid uptake of DIN in the plume and the addition of DIP from leaching (Fox et al., 1986). Surface SiO<sup>3</sup> concentrations (**Figure 2d**) ranged widely from 0.37 to 50 µM and mostly followed a conservative behavior as a function of salinity but with lower values in the plume and the northwestern edge of the plume because of high uptake rates by diatoms (**Figure 2d**).

Pigment derived phytoplankton classes and microscopically determined phytoplankton communities show the formation of niches governed by the stoichiometry of the region. A very dense and mixed bloom of diatoms comprising primarily of Skeletonema marinoi (sensu lato, costatum), Pseudo-nitzchia spp., Thalasiossira alienii, and Chaetoceros spp. occupied the low salinity core of the plume (**Figure 3a**). These blooms were fueled by riverine derived SiO<sup>3</sup> (**Figure 2d**) and cross-shore and upward flux of DIN into the surface layers. Maximum cell counts for some of the diatom species included Pseudo-nitzschia spp (1.5 × 10<sup>6</sup> cells L−<sup>1</sup> ), Thalassiosira spp. (3.5 × 10<sup>6</sup> cells L−<sup>1</sup> ), Guinardia flaccida (9.0 × 10<sup>4</sup> cells L−<sup>1</sup> ), Chaetoceros spp (3.6 × 10<sup>5</sup> cells L −1 ), Coscinodiscus (3.3.8 × 10<sup>6</sup> cells L−<sup>1</sup> ), and Odontella sinensis (4.5 × 10<sup>7</sup> cells L−<sup>1</sup> ). These blooms depleted DIN from surface waters in the mesohaline and open ocean regions (**Figure 2b**).

Consequently, in the mesohaline waters, the moderately high Silicate (28–33 µM) and DIP (0.5–0.9 µM) allowed a very different group of diatoms, the DDAs to emerge (**Figure 3a**). These symbiotic diatom-diazotrophic assemblages are capable of photosynthesizing and growing actively in the absence of DIN, because their nitrogen requirements are met via the N<sup>2</sup> fixing ability of their endosymbiont Richelia intracellularis (Foster et al., 2007; Foster et al., 2011). The dominant DDA was Hemiaulus hauckii with a maximum cell count of 9.7 × 10<sup>5</sup> cells L−<sup>1</sup> and high cell counts on the western side of the plume. Another DDA, Rhizosolenia clevei was also observed but in much smaller numbers, with maximum concentrations of about 800 cells L −1 . Additionally, the free living, non-symbiotic, cyanobacteria, Trichodesmium was also seen but more broadly distributed than the DDAs. Higher Trichodesmium concentrations were seen just north of the plume and in the oligotrophic waters east of the plume (**Figure 3a**). In addition to the microplankton (diatoms, dinoflagellates, and DDAs) (**Figure 3a**), nanoplankton like Cryptophytes, Prasinophytes and Haptophytes were also observed in large numbers in the two regions where DIN was available albeit at low concentrations (**Figure 3b**). This included the plume region and the mesohaline region where DDAs replenish DIN when cells disintegrate. Cryptophytes proliferated only in the low salinity plume waters with numbers ranging from 3.2 × 103–6.3 × 10<sup>5</sup> L −1 . This was also the case with prasinophytes but Haptophytes were seen in both the plume and the mesohaline area coincident with the DDA blooms (**Figure 3b**). Picoplanktonic Synechoccocus (**Figure 3c**) were also seen in large numbers especially in the plume where their numbers ranged from 5.0 × 103–6.9 × 10<sup>5</sup> L −1 . Outside of the plume, Synechoccocus were only seen in the mesohaline, DDA dominated region. Conversely, Prochlorococcus identified by their unique Divinyl Chl a signature were absent from the nutrient rich plume waters but were ubiquitous elsewhere especially in the waters east of the plume (**Figure 3b**), where all three macro nutrients were below the level of detection.

The significance of the ARP in driving changes in the hydrography, chemistry, and phytoplankton communities of the western Tropical North Atlantic is illustrated by Principal Component Analysis (PCA) of hydrographic data and microscopic and CHEMTAX derived phytoplankton communities (**Figure 4a**). PCA enables a visualization of the similarities and differences between samples, as well as the correlations between the variables. The first principle component accounted for 32% of the variance and divided the data set into phytoplankton communities that proliferated in the plume (diatoms, dinoflagellates, cryptophytes and prasinophytes) from those in the mesohaline and oligotrophic waters (DDAs, Trichodesmium, Prochlorococcus, and Haptophytes). The former group correlated with DIN indicating its importance in the growth of these micro and nanoplankton. The latter group which comprised diazotrophs, Haptophytes, and picoplanktonic Prochlorococcus with low nutrient requirements were found in high salinity waters outside of the plume. The PC2 axis separated DIN, the limiting nutrient from DIP and silicate which were not limiting.

#### ECS

Our four-year study clearly demonstrated that the hydrochemical environment of the ECS undergoes large interannual variations linked to the discharge of the Changjiang River and the eastward movement of its plume (CDW) which strongly influences phytoplankton distribution. As in the case of the ARP, and based on T-S plots, Xu et al. (2018) defined three water types viz. CDW (T > 23oC, S < 28) primarily influenced by discharge from the Changjiang River, the Kuroshio surface waters associated with the deeper, higher salinity Kuroshio waters (S > 32.9) and the Shelf waters. The latter is a mixed water mass (Salinity of 30–32.9) of the CDW and Kuroshio surface waters, located in the central part (125–128oE) of our study area. The large interannual variability in temperature, salinity and nutrients of the ESC was mainly from the variable influence of CDW and Kuroshio surface waters as is evident in **Figures 5a–v**. The influence of the CDW was greatest in 2010 with colder, lower salinity waters extending up to 127.5oE (**Figures 5b,f**) followed by the consecutive year of 2011 when the CDW was confined to west of 126oE (**Figures 5c,g**). In contrast, 2009 and 2013 experienced a greater influence of warmer, more saline Kuroshio surface

waters (**Figures 5a,e,d,h**). This resulted in mixed Shelf waters covering the entire study area. During these 2 years, waters in the eastern part of the observation area were warmer (SST > 28.5oC) (**Figures 5a,d**) and more saline (>32.9) (**Figures 5e,h**).

The surface waters of the eastern region of the study area, which are less influenced by riverine discharge and in the path of the Kuroshio extension were generally depleted of nutrients during all 4 years of study. DIN in these waters ranged from 0.02 to 0.11 µM (**Figures 5i–l**), and DIP was less than 0.07 µM (**Figures 5m–p**). In these Kuroshio influenced waters, ExN measured around 0 µM (**Figures 5q–v**) indicating invariant stoichiometry. In contrast, the CDW waters showed distinct interannual differences in their nutrient regime with significantly higher (non-parametric Mann-Whitney U test, p < 0.05) ExN values in 2010 and 2011 (5–25 µM) (**Figures 5t,u**) indicating severe DIP limitation. This is contrast to 2009 and 2013 when ExN was <0 µM (**Figures 5q,v**). As expected, in the former years, the CDW was characterized by extremely high DIN (>10 µM, maximum of 26.1 µM) derived from the anthropogenic load of the Changjiang River (**Figures 5j,k**) but low DIP (<0.05 µM) (**Figures 5n,o**). A contrasting situation emerged in 2009 and 2013 when only one station to the west of 125oE showed high DIN (**Figures 5i,l**). DIP in 2009 and 2013 was higher (up to 0.2 mg m−<sup>3</sup> ) compared to concentrations in 2010 and 2011 when it was almost below the limit of detection. Statistically, the differences in DIN, DIP, and ExN between higher DIP years (2009, 2013) and high lower DIP years (2010, 2011) were highly significant (p < 0.01) indicating the predominant influence of the Changjiang River on the stoichiometry of the ECS. Overall, phytoplankton biomass reflected the large west-east differences in N:P ratios in the ECS. Chl a was higher in the CDW region (0.25–4.0 mg m−<sup>3</sup> ) but decreased in the shelf waters (0.14– 0.9 mg m−<sup>3</sup> ) (**Figures 5w–z**). Lowest values were observed in Kuroshio surface waters where Chl a rarely exceeded 0.2 mg m−<sup>3</sup> .

Similar to phytoplankton communities in the ARP continuum, those in the ECS also adapted to the changing nutrient regime influenced by the Changjiang River discharge. Phytoplankton communities of the CDW (**Figure 6**) showed clear distinctions between low DIP (2010, 2011) and high DIP (2009, 2013) years. Diatoms dominated the latter years and diatom-derived Chl a ranged from 3–2.1 mg m−<sup>3</sup> compared to concentrations of 0.02–0.34 mg m−<sup>3</sup> during low DIP years. In contrast, during 2010 and 2011, cyanobacteria dominated, contributing 6–76% of the Chl a as opposed to 1–20% in 2009 and 2013. A substantial population of Prymnesiophytes also contributed to the Chl a (25–55%) in 2013. Although we have chosen to show only the phytoplankton community structure of the CDW, it is apparent from Figure 4b of Xu et al. (2018) that there were no interannual differences in the phytoplankton structure of the mixed shelf waters. However, diatom populations did show dominance in the high DIP year of 2009 (Chl a contribution 28–83%). A mixture of cyanobacteria and the picoplanktonic prochlorophytes dominated (>61% of the total Chl a) the low Chl a at waters of the oligotrophic Kuroshio surface waters during all 4 years of study (Figure 4c in Xu et al., 2018). Prochlorophytes, the second highest group after cyanobacteria, comprised more than 30% of the phytoplankton community in 2009, 2010, and 2013 and 19% in 2011 (Xu et al., 2018 ). Diatoms were

absent or negligible except in 2011 when 18% of Chl a was from this group. As in the case of the ARP (Goes et al., 2014), large diatom blooms were seen in the coastal waters of the Changjiang River Continuum (Zhou et al., 2008; Guo et al., 2014) advantaged by the abundant nutrients available from the river discharge.

The possible dependence of the various algal groups on Salinity, Temperature and nutrients is also examined by means of PCA (**Figure 4b**). PC1 accounted for 27.3% of the variance in this data separating DIP and DIN. PC1 also divided the data set into large microplankton (diatoms and dinoflagellates) with high P requirements versus smaller nano and picoplankton such as Chlorophytes, Cyanobacteria and Prochlorophytes which can grow at lower DIP concentrations. DIP correlated with large microplankton (diatoms and dinoflagellates) showing it to be the limiting nutrient that determined the transition from small to large phytoplankton in the CDW. Conversely, cyanobacteria, chlorophytes, and prochlorophytes were associated with warmer waters of low DIP concentrations characteristic of Kuroshio surface waters. This is also demonstrated by the positive correlation of smaller phytoplankton with ExN on PCA1 axis and the converse with microplankton. Both PC1 and PC2 show a strong negative correlation between Salinity and DIN indicating riverine inputs of DIN. In contrast to PC1, the PC2 axis which accounted for 24% of the variance in the dataset showed a positive correlation between DIN, DIP, and diatoms but notably a negative correlation with Temperature. This typifies the intrusion of colder, upwelled waters which bring in DIP and result in the domination of diatoms. The latter is elaborated further in the Discussion section.

# DISCUSSION

Revisiting and comparing the previously conducted research in two of the world's largest river-ocean systems gave us an opportunity to understand how fluctuations in nutrient availability and nutrient ratios shape phytoplankton communities along the river-ocean continuum to produce highly distinct and specialized groups that can profoundly affect the food web and carbon fluxes to deeper depths. As we discuss below, it is likely that the N:P supply ratios from the two systems determine whether or not diazotrophs are part of the community, while it is likely that supply of a single limiting nutrient (such as DIP) is responsible for the interannual variability seen in the ECS. We use resource competition theory and resource supply ratio to describe the distribution of phytoplankton communities of the ARP and CDW in relation to changing in N:P supply ratios and changing DIP supplies along the river-ocean continuums.

Simply stated, Tilman (1977)'s resource competition theory states that in an equilibrium environment, each competing phytoplankton type is capable of drawing the limiting nutrient down to distinct low subsistence concentrations, known as R<sup>∗</sup> . The value of R<sup>∗</sup> for any nutrient is set by the ecophysiological characteristics of the phytoplankton type and represents both the concentration of the limiting nutrient at which growth is balanced by mortality (top down) and the minimum nutrient concentration (bottom up) required for survival (Ward et al., 2014). If multiple organism types are present, the ambient resource concentration will be drawn down to the lowest R<sup>∗</sup> and only that class of organisms will survive while other organisms will be excluded over time (Dutkiewicz et al., 2009). This theory then implies that the smallest cells which have higher

nutrient uptake rates should accumulate biomass by drawing down nutrients to the lowest level thus excluding any larger phytoplankton with higher nutrient requirements. However, this is not the case because the smallest and most competitive phytoplankton groups are in fact kept under control by grazing by the small zooplankton, allowing larger size classes to become established when excess nutrients become available (Ward et al., 2012; Ward et al., 2014). In this situations the R<sup>∗</sup> s intersect and stable coexistence is possible. Thus, when nutrient supplies are higher, more classes of larger phytoplankton (with less efficient nutrient uptake rates) could coexist with the smallest types. This seems to the case in the ECS where the CDW waters were DIP limited. However, in years (2009, 2013) when DIP concentrations increased we saw predominance of diatoms while in years when DIP concentrations were low or beyond the level of detection, small cells like cyanobacteria and chlorophytes predominated. Dinoflagellates remained invariant during all 4 years of sampling with Chl a ranging from 0.1–0.4 mg m−<sup>3</sup> of the total Chl a. Many dinoflagellates are mixotrophic, and it is possible that this mixing of trophic levels allows them to exist in all situations.

On the eastern side of our sampling grid, in the oligotrophic waters dominated by the nutrient poor Kuroshio surface waters only picoplanktonic Prochloroccocus proliferated (Xu et al., 2018). The increased DIP concentrations appear to be regulated by the extent of low saline CDW in the western region which as determined by the surface temperature and salinity was limited westward in 2009 and 2013. In our more extensive study (Xu et al., 2018), we have investigated the physical forces driving the higher DIP concentrations in the mid-shelf of the ECS in 2009 and 2013. All evidence points to upwelling of DIP-rich Kuroshio Intermediate Water onto the upper layers of the shelf (Tseng et al., 2014) and its advection eastward by the CDW into the mid-shelf of the ECS. In Xu et al. (2018) we explain the higher surface DIP in 2009 and 2013 and the converse situation in 2010 and 2011 as follows. In the latter, when the Changjiang River discharge was higher, DIP from coastal upwelling was not only diluted as the CDW moved eastward, vertical stratification from density differences also prevented mixing. In contrast, during low discharge years, DIP from coastal upwelling could be transported further eastward into the mid-shelf where it supported the growth of larger phytoplankton.

In case of the ARP, we can also explain the appearance of diazotrophs versus non-diazotrophs (non-nitrogen fixing marine phytoplankton) in terms of the resource supply-ratio theory. Ward et al. (2013) describes the competition between the two (diazotrophs versus non-diazotrophs) through their interaction with the three essential nutrient elements viz. Fe, DIN and DIP. Fe is not limiting for phytoplankton both in the ARP and the CDW largely because of intense weathering and erosion that takes place in the drainage basins of this river system (Edmond et al., 1985; Bergquist and Boyle, 2006). Whereas in the CDW, DIP appears to be the single nutrient that controls the size structure of phytoplankton communities, in the ARP, it is the N: P ratio that will determine whether diazotrophs or nondiazotrophs will establish themselves. Generally, diazotrophs exhibit slower growth rates than non-diazotrophs so they will be outcompeted if both nutrients are limited. But if there is excess DIP relative to the stoichiometric requirements of the non-diazotrophs then this DIP can fuel diazotroph growth,

since it is assumed that diazotrophs can fix N. In other words, in situations where DIP:DIN supply rates are in excess of the non-diazotroph requirements, then diazotrophs can coexist with the non-diazotrophs (Dutkiewicz et al., 2012). In the case of the ARP, away from the inner shelf where sediment load is reduced and light levels are not limiting, large mixed blooms (Shipe et al., 2006; Goes et al., 2014) deplete the DIN and reduce DIP although the riverine Si required for diatoms was not limiting. As the plume flows offshore, DIP is replenished through desorption (Fox et al., 1986; Berner and Rao, 1994) and the system transitions to diazotrophy and the acute shortage of DIN is thus mitigated. As stated earlier, a variety of diatom diazotroph assemblages as well as nonsymbiotic and non-siliceous diazotrophs like Trichodesmium proliferated in the mesohaline waters. Small nitrogenous inputs from symbiotic diatoms supported populations of nanoplankton such as Prasinophytes and Haptophytes which were earlier not reported but are identified in this study using pigment biomarkers. Experimental evidence from other locations (Gulf of California and the subtropical North Pacific) shows that Richelia fixes 81–744% more N than needed for its own growth and up to 97.3% of the fixed N is transferred to the diatom partners (Foster et al., 2011). This same study also showed that N<sup>2</sup> fixation rates of Richelia and another cyanobacterial symbiont Calothrix were 171–420 times higher when the cells were symbiotic compared with the rates for the cells living freely.

It is well known that diazotrophy is inhibited when DIN and/or ammonium concentrations exceed 1 µM (Knapp, 2012) and the resource supply-ratio theory suggests that high N:P supply ratios will not support diazotrophy, so it is not surprising that we did not see either DDAs or non-symbiotic diazotrophs such as Trichodesmium in the ECS during our 4 years of study as DIN was not a limiting nutrient in these waters. What is noteworthy about the results from the ECS is the marked interannual variability in the hydrology, stoichiometry and consequent phytoplankton community structure of ECS modulated by the extent of the Changjiang River and its DIP inputs.

Long term changes being monitored in the ECS using satellitederived salinity algorithms (Bai et al., 2014) show that the spread and direction of CDW plume on the ECS continental shelf are affected by the Changjiang River discharge while other long term studies of summer phytoplankton community in the Changjiang estuary during the past 50 years (Jiang et al., 2014) suggest that anthropogenic loading is leading to increase in smaller phytoplankton and dinoflagellates which could exacerbate harmful algal blooms. Although similar information is not available for the Amazon River continuum, model projections of climate change effects on discharge and inundation in the Amazon basin (Sorribas et al., 2016) show a system in transition (Davidson et al., 2012). A study of all the rivers of South America (van der Struijk and Kroeze, 2010) including the Amazon river shows that exports of DIN and DIP increased between 1970 and 2000 although the increases were more prominent in rivers south of the Amazon. If DIN export were to increase and attain concentrations higher than DIP loading, then we would expect diazotrophy to disappear from the ARP. Elsewhere, nutrient reduction measures such as in the Chesapeake Bay Estuary along the east coast of the US, have resulted in a decrease in diatoms from a decrease in DIN (Harding et al., 2015) whereas the reverse was seen in Patos Lagoon Estuary, Brazil which showed increasing signs of eutrophication (Haraguchi et al., 2015). Not surprisingly and considering the extent of the rapid economic development of China during the past 30 years, the Pearl River which empties into the northern part of South China Sea showed the same situation that we observed in the ECS. While diatoms proliferated in the inner estuary, limited DIP resulted in the proliferation of picophytoplankton in the mesohaline and nearshore oceanic waters (Qiu et al., 2010).

The varied taxonomical diversity of the two river continuums should be reflected in the geochemical fluxes to deeper depth. Variations in the impact of the CDW through its nutrient supply and consequently on the phytoplankton community are also reflected in the carbon fluxes. Sukigara et al. (2017) who operated sediment traps in 2000 and 2011 in tandem with our studies, showed that sinking particles were thrice as high in 2000 when our study showed the dominance of diatoms than in 2011. Particulate carbon and nitrogen contents as well as isotope ratios, showed that the particles that sunk out the euphotic zone in 2010 were primarily from the CDW layer with a secondary contribution from the SCM layer. Concomitant with our ARP studies Chong et al. (2014), showed a distinct axis of POC and biogenic silica deposition on the deep floor aligned with the plume and a footprint of approximately 1 million km<sup>2</sup> of carbon and biogenic silica on the deep sea floor.

Traditionally, the elemental stoichiometry of both phytoplankton biomass and dissolved nutrient pools has been viewed as having limited variability. Expressing the relevance of the canonical Redfield ratio, Falkowski and Davis (2004) wrote that "the uniformity of elemental Redfield ratios in the oceans and the life they contain underpins our understanding of marine biogeochemistry." But just as this nearconstant ratio laid the foundations for the twentieth-century advances in our understanding of marine biogeochemistry, deviations from this ratio are now providing twenty-firstcentury insights into the nutrient dynamics of oceans modern and ancient (Nature Geoscience, 2014). Questions remain with respect to fundamental facts such as the flexibility of phytoplankton stoichiometry and the relationship of internal ratios to resource availability and growth rates (Hillebrand et al., 2013). Relationship between available and internal N:P ratios are not necessarily linear. While Goldman et al. (1979) found that phytoplankton N:P ratios varied widely at low N:P converging to classic N:P ratio of 16 only at high growth rates, Hillebrand et al. (2013) revisited this premise to show that although N:P ratios did converge to an optimal ratio it was different for different species and phylogenetic groups. We think that river continuums with changing spatial and temporal stoichiometry are ideal environments to study the flexibility of phytoplankton stoichiometry vis a vis resource availability and its implications for size and functional traits.

# CONCLUSION

fmars-05-00343 September 24, 2018 Time: 16:49 # 12

We have compared the distribution of phytoplankton communities in the river-ocean continuums of two of the world's largest rivers, the Amazon River, the nutrient content of which is largely governed by forest derived nutrients and the Changjiang River, the waters of which are heavily impacted by human activities. We show that phytoplankton communities along the axes of their plumes are shaped largely by the initial N: P content of the rivers. In the case of the ARP, phytoplankton populations downstream of the river plume were largely controlled by the availability of inorganic nitrogenous nutrients with diatoms dominating upstream where nitrogenous were available either from the source water or from the onshore advection of nutrient rich waters from depth. In the case of the Changjiang River plume, phytoplankton communities were limited significantly by a single nutrient viz. inorganic phosphate and downstream of the plume, diatoms were observed during years when coastal upwelling was more intense and inorganic phosphate from deeper depths advected offshore. One of the largest differences between the ARP and the Changjiang River plume waters was the absence of DDAs and Trichodesmium spp. in the latter, which appears to be tied to different N:P supply ratios.

#### AUTHOR CONTRIBUTIONS

HG wrote the paper and conducted the Amazon plume research. QX conducted the East China Sea research. JG conducted research in the Amazon plume. EC undertook the cell counts for

#### REFERENCES


the Amazon plume research. JI and PY were the lead PIs on the East China Sea and Amazon Plume projects respectively.

# FUNDING

This work was supported by Visiting Professorships at Nagoya University, Japan to HG and JG. Research conducted in the ECS by JI and QX was supported by JAXA GCOM-C and JSPS KAKENHI grant number JP26241009 awarded to JI. Research work in the ARP was supported by grants NSF OCE-0934095 and 1133277 to PY, JG, and HG and NASA grants NNX13AI29A and NNX16AD40G to JG and HG.

#### ACKNOWLEDGMENTS

We are highly indebted to Dr. Simon Wright (Australian Antarctic Division) for the CHEMTAX code and for his constant help and guidance in running CHEMTAX. We would specially like to thank Dr. Stephanie Dutkiewicz (MIT) for her time, effort and invaluable suggestions that improved this paper immensely. We thank the NASA-HPLC Center, for help with analysis of phytoplankton pigments for the ARP study. We thank Victoria Coles (Center for Environmental Science, University of Maryland) for the Salinity distribution map of the ARP and we acknowledge the Salinity map for the ECS provided by Asia-Pacific Data Research Center.


tropical North Atlantic. Limnol. Oceanogr. 61, 825–840. doi: 10.3389/fmicb. 2017.01436



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Gomes, Xu, Ishizaka, Carpenter, Yager and Goes. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Bioavailability of Dissolved Organic Phosphorus in Temperate Lakes

Seth K. Thompson<sup>1</sup> \* and James B. Cotner <sup>2</sup>

*<sup>1</sup> Water Resource Science Program, University of Minnesota, St. Paul, MN, United States, <sup>2</sup> Department of Ecology, Evolution and Behavior, University of Minnesota, St. Paul, MN, United States*

Freshwater aquatic systems are biogeochemical hotspots, with heterotrophic bacteria rapidly cycling the compounds that pass through them. P is a key nutrient that controls primary production in many freshwater ecosystems and is important for understanding eutrophication in lakes. Previous work has often focused on the dynamics of inorganic phosphorus and its impact on primary production, however, the role of nutrients bound in more complex organic forms (such as dissolved organic phosphorus, DOP) in supporting primary production and harmful algal blooms has been neglected. Here, we quantify the bioavailability of dissolved organic carbon (DOC) and DOP in 27 aquatic systems across the Upper Midwest United States. Using exponential decay models, long-term nutrient degradation assays revealed that decay constants for DOP ranged from −0.001 per day to −0.12 per day with a median value of −0.01 per day. These rates were geographically variable and were as high or higher than DOC decay constants, which ranged from −0.003 per day to −0.024 per day with a median value of −0.01 per day. Additionally, total bioavailability of DOP ranged from 0 to 100% with a median value of 78% of the DOP pool, demonstrating that DOP bioavailability was highly variable across systems. In contrast, bioavailable DOC was more tightly constrained with values ranging from 4.37 to 53.81% of the total DOC pool with a median value of 24.95%. DOP bioavailability was negatively correlated with the DOC:DOP of the organic matter pool, suggesting that bioavailable DOP is drawn down in systems that are more likely to be P limited. Finally, we show that including estimates of DOC and DOP bioavailability reduces estimates of elemental imbalance experienced by aquatic bacteria.

#### Keywords: phosphorus, carbon, bioavailability, dissolved organic matter, degradation

# INTRODUCTION

Freshwater systems are incredibly active biogeochemical hot spots, particularly with regards to the processing of organic matter (Cole et al., 2007; Tranvik et al., 2009). Heterotrophic bacteria are major biogeochemical players in aquatic systems (Cotner and Biddanda, 2002) and understanding how these microbes interact with organic matter is fundamental to predicting the flow of energy and nutrients through freshwaters. Dissolved organic matter (DOM) is a major resource pool for aquatic bacteria and has been the focus of numerous studies over the last 20 years. The vast majority of this work has focused on understanding how microbial modifications of DOM influence global carbon cycle processes, but it is important to remember that DOM is not solely composed of carbon. Microbial interactions with DOM are also likely important for understanding other biogeochemical

#### Edited by:

*Solange Duhamel, Lamont Doherty Earth Observatory (LDEO), United States*

#### Reviewed by:

*Monika Nausch, Leibniz Institute for Baltic Sea Research (LG), Germany Martin Tucker Auer, Michigan Technological University, United States*

> \*Correspondence: *Seth K. Thompson thom2587@umn.edu*

#### Specialty section:

*This article was submitted to Freshwater Science, a section of the journal Frontiers in Environmental Science*

> Received: *24 March 2018* Accepted: *07 June 2018* Published: *28 June 2018*

#### Citation:

*Thompson SK and Cotner JB (2018) Bioavailability of Dissolved Organic Phosphorus in Temperate Lakes. Front. Environ. Sci. 6:62. doi: 10.3389/fenvs.2018.00062* cycles in freshwater such as the phosphorus (P) cycle, but our understanding of the role DOM plays in these other nutrient cycles remains limited (Maranger et al., 2018).

Humans have had profound impacts on the global P cycle by increasing the annual flux of P through ecosystems by a factor of 4–8 (Falkowski, 2000; Schlesinger and Bernhardt, 2013). This has important biogeochemical implications and has resulted in the eutrophication of freshwater systems worldwide leading to degraded water quality on a global scale. Fundamentally, eutrophication is a biogeochemical imbalance, where excess nutrients, often P in freshwater (Schindler et al., 2008), result in excessive accumulation of carbon (C) in the form of increased algal biomass. This continued anthropogenic modification of freshwater nutrient and organic matter pools, together with observations of shifts in planktonic community composition and organic nutrient pools (Teubner et al., 2003) suggest that the bioavailability of organic nutrient pools could an important factor affecting auto-heterotrophic coupling as well as harmful algal blooms.

Previous work has often explored the role of a single inorganic nutrient controlling primary production, but our rapidly evolving understanding suggests more complex scenarios are likely involved. Specifically, there is a growing appreciation for the role of nutrients bound in complex organic forms (such as dissolved organic phosphorus, DOP) in acting as important resources for aquatic organisms (Jackson and Williams, 1985; Cotner and Wetzel, 1992; Björkman and Karl, 2003; Nausch and Nausch, 2007; Soares et al., 2017). Therefore, it is imperative to understand the bioavailability of DOP in natural systems in order to better predict its capacity to serve as a resource in the absence of (or supplementary to) inorganic phosphorus. Furthermore, understanding the composition and bioavailability of nutrients and organic matter, not just the total quantities in a system have been shown to have important impacts on the formation and toxicity of harmful algal blooms (Anderson et al., 2002; Donald et al., 2011) providing another compelling reason to further study DOP bioavailability in freshwater systems.

Organic P is the dominant form of P in most freshwater systems with DOP typically comprising 25–50% of the total P pool (Wetzel, 2001). Many studies have clearly demonstrated that at least some forms of DOP can serve as a P source for primary and secondary producers (Cotner and Wetzel, 1992; Björkman and Karl, 2003; Nausch and Nausch, 2007; Li and Brett, 2013) and DOP is therefore likely an important source of P when inorganic P is limited. While there have been some studies on DOP bioavailability specific to freshwater systems (Sonzogni et al., 1982; Boström et al., 1988; Cotner and Wetzel, 1992; Li and Brett, 2013), more studies have examined this topic in marine systems (Björkman and Karl, 1994; Ruttenberg and Sonya, 2005; Dyhrman and Ruttenberg, 2006; Nausch and Nausch, 2006, 2007). Nonetheless, in both marine and freshwaters, a portion of the DOP pool is readily available for assimilation into planktonic organisms. The relative bioavailability of specific DOP compounds can range from almost 0% to over 90% (Li and Brett, 2013). While these studies have clearly suggested that DOP may be an important source of P for aquatic microorganisms, few studies have quantified bioavailable DOP in freshwater systems or examined how organic matter stoichiometry may affect the bioavailability of DOP compounds. In this paper, we present results from DOM bioavailability assays from 27 aquatic systems in the Upper Midwest of the USA. Our goal was to quantify the bioavailability of DOC and DOP from a diverse set of aquatic ecosystems and explore the potential environmental drivers of DOM bioavailability.

### MATERIALS AND METHODS

#### Study Sites and Sample Collection

During the summer season (June–September) water was collected from 27 freshwater systems (24 lakes and 3 streams) in Minnesota and South Dakota (Supplementary Figure 1) from the upper mixed layer (0–2 m of depth) using a Van Dorn water sampler. The 27 systems covered three different Level III ecoregions as defined by the United States Environmental Protection Agency: The Black Hills in Western South Dakota (Middle Rockies), Itasca State Park in North Minnesota (Northern Lakes and Forests), and the Twin Cities greater metropolitan area (North Central Hardwood Forests). The Black Hills and Itasca State Park systems are relatively pristine systems with watersheds primarily dominated by coniferous forest whereas the Twin Cities greater metropolitan area is highly human impacted with watersheds dominated by hardwoods with urban land use or small-scale agriculture. General characteristics of each lake system can be found in **Table 1**.

Samples were collected in acid-washed HDPE amber bottles after pre-rinsing them with ∼100 ml of sample water. Samples were stored on ice until they could be returned to the lab (always less than 4 h), where they were stored at 4◦C until processed (<72 h). For processing, ∼100 ml of water was filtered through pre-combusted 0.7µm nominal pore-size glass-fiber filters (Whatman, GF/F) and collected in pre-combusted borosilicate vials. Twenty milliliters of this sample was acidified with 10% HCL and used to measure total dissolved organic carbon (DOC) and total dissolved nitrogen (TDN) on a Shimadzu TOC-L auto-analyzer with a TNM-L module (CSH/CSN model, Shimadzu Corp). Another 20 ml of filtrate was reserved for absorbance scans (wavelengths from 200 to 800 nm) using a Cary 50 spectrophotometer, which was used to calculate the specific UV absorbance at a wavelength of 254 nm (SUVA). The remaining 60 ml of sample was used to measure total dissolved phosphorus (TDP), and soluble reactive phosphorus (SRP) using a molybdenum blue reaction with and without acid-persulfate digestion (Murphy and Riley, 1962). DOP was calculated as the difference between TDP and SRP. Additionally, the GF/F filters were collected for fluorometric quantification of chlorophyll-a after being extracted in 90% acetone (Standard Methods, 2005).

#### DOM Degradation Assays

DOM degradation assays were performed as long-term dark bottle incubations. Within 72 h of collecting each sample, 900 ml of lake water was filter-sterilized using a 0.22µm pore-size filter (EMD Millipore Steritop Filters) and collected in a precombusted 1 L amber glass bottle. Each bottle was inoculated with 100 ml of water from the same lake that had been filtered TABLE 1 | Table showing general characteristics of each of the 27 sampling sites.


through a 1.6µm pore-size glass fiber filter (Whatman, GF/A). While this approach did not standardize for the absolute inoculum size (i.e., number of bacterial cells), it did provided a consistent relative inoculum source for each assay of 10% by volume, similar to previous work (Wiegner et al., 2006; Lønborg et al., 2009; Vonk et al., 2015). Furthermore, the relative size of the inoculum has been shown to have little to no effect on the overall degradation of DOC (Vonk et al., 2015). These bottles were incubated in the dark at 20◦C for a minimum of 230 days. Periodically throughout the incubation period (approximately monthly for the first 6 months and less often subsequently), 100 ml samples were removed from the incubations. These samples were filtered using pre-combusted GF/F filters and DOC, TDN, DOP, SRP, and absorbance scans were measured as described above.

#### Data Analysis

To calculate DOM degradation rates, data were fitted to 3 unique models (linear, 2 component exponential, and 3 component exponential) and the model fit was compared using Akaike information criterion (AIC) scores. Because model fits were variable for different portions of the DOM pool, relative bioavailability of DOC and DOP was calculated by dividing the maximum nutrient loss (initial concentration minus the minimum measured concentration during the incubation period) by the starting concentration to obtain the percentage of the total pool that was degraded. This approach allowed for the direct comparison of relative lability between the DOP and DOC pools. The relationships between bioavailability and environmental parameters (such as SUVA, nutrients, etc.) were examined using simple linear regression. All analysis was performed using JMP <sup>R</sup> version 13 (SAS Institute Inc., Cary, NC, 1989–2007).

#### RESULTS

#### Bulk Nutrient Analysis

The 27 systems studied covered a trophic gradient with chlorophyll-a values ranging from 0.25 µg/L to 57.19 µg/L and total dissolved phosphorus concentrations ranging from 0.07µM to 2.31µM (**Table 1**). Mean chlorophyll levels were much higher the Twin Cities region compared to the two less human-impacted regions and chlorophyll values were also more variable in the Twin Cities compared to the other two regions. SUVA values were calculated for each system by dividing the specific UV absorbance at 254 nm by the DOC concentration of the system. SUVA can be used as an index of the terrestrial contribution to the DOM pool, with higher SUVA values indicating more terrestrial influence. We sampled systems that exhibited a range of SUVA values (from 0.86 to 3.69) to cover a gradient of terrestrial influence on the DOM pool. Across the three regions sampled, chlorophyll and the relative contribution of DOC to the total dissolved carbon pool (DOC:TDC) showed strong differences, but DOP concentration and the relative contribution of DOP to TDP did not (**Figure 1**). Also, our study sites showed no regional differences in TDP or SRP concentrations, despite the fact that others have shown strong differences in P concentrations across these ecoregions (Heiskary et al., 1987). Furthermore, DOP concentration was not significantly correlated with any of the measured lake characteristics (pH, alkalinity, chlorophyll, or SUVA), but was weakly positively correlated with DOC concentration and TDN concentration (**Figure 2**). Additionally, the Twin Cities dissolved carbon pool had a much higher fractional total organic carbon signature (with a median value of approximately 40% DOC) compared to the Black Hills and Itasca where DOC contributed less (11 and 18.5% respectively) to the dissolved carbon pool (**Figure 1**).

#### Degradation Rates of DOC and DOP

DOC degradation was best fit by an exponential decay model with a non-zero asymptote (Equation 1, model resulted in an R 2 value of 0.998 across all lakes; Supplementary Figure 2). Two lakes, Canyon Lake and Roubaix (both from the Black Hills region), resulted in model fits that had positive k values, despite both having lost DOC over the course of the incubation so they were excluded from analysis of DOC decay rates. In the remaining 25 lakes, k values ranged from −0.003 to −0.024 per day with a median value of −0.009 and quartiles of −0.006 and −0.013 or median turnover time of 111 days (**Figure 3**). DOC degradation rates were not significantly correlated to measured elemental pools (DOC, TDN, TDP, SRP) or lake characteristics (pH, chlorophyll, SUVA) and there were no significant differences in DOC decay rates across region.

$$DOC\_t = BDOC^{kt} + DOC\_R \tag{1}$$

Three parameter exponential decay model that was the best fit for long term DOC incubations where DOC<sup>t</sup> is the concentration of DOC at time t, BDOC is the total pool of bioavailable DOC, k is the degradation rate, t is the time of incubation in days, and DOC<sup>R</sup> is the size of the recalcitrant DOC pool.

TDP, SRP, and DOP incubations revealed turnover times of approximately 150 days. After this period, DOP concentrations tended to increase in the incubations suggesting internal cycling of DOP (see Supplementary Figure 3). Given that there were no external sources of P to these incubations, this increasing concentration of dissolved P likely resulted from the degradation of particulate P that had accumulated over the early portion of the incubation. To examine the degradation of DOP over the course of the incubation, we excluded all data points after 150 days of incubation and excluded points when the calculated DOP concentration was below zero (this only occurred in 6 of the 230 total measurements). A 2-parameter exponential fit model best described the

Cities (urban) region. (B) Shows no significant differences in total DOP concentration across the three study regions and (C) shows no significant differences in the relative contribution of DOP to the total dissolved P pool, however the Twin Cities does show the largest range of relative DOP contribution. (D) Shows the contribution of DOC to the total dissolved carbon pool with the Twin Cities showing a much higher contribution of DOC compared to the other two regions, in other words, inorganic carbon dominates the dissolved carbon pool in the Black Hills and Itasca regions.

DOP data [the same model as Equation (1), omitting the recalcitrant pool] resulting in a model with an overall r 2 value of 0.77.

DOP degradation rates (k) across the 27 lakes largely fell between 0 and −0.025 per day (22 of the 27 lakes). Three lakes had positive modeled k values with two of these systems in the Twin Cities metro area (Fish and Staring North) and one was in the Black Hills (Pactola, see supplementary data file). Because the model estimated k values were positive despite the fact that concentrations of DOP decreased in the incubations, these lakes were excluded from other analysis of DOP degradation rates. Additionally, two lakes form the Itasca region (Boot and Elk) had extreme negative k values of −0.096 and −0.123 respectively (**Figure 3**). The median value for all 27 systems was −0.010 corresponding to a median turnover time of 100 days. DOP degradation rates were not significantly correlated to measured elemental pools (DOC, TDN, TDP, SRP) or lake characteristics (pH, chlorophyll, SUVA).

To compare the relative rate of DOP turnover to DOC turnover, we calculated a kDOC:kDOP value for the 22 systems that had negative k values for both DOC and DOP. Overall, turnover

rates of the two nutrients were remarkably similar with a median kDOC:kDOP of 0.98, upper quartile of 1.83, and lower quartile of 0.50. However, there were several lakes with more extreme values with the most extreme system having a DOP turnover rate nearly 20 times faster than DOC (Elk Lake in Itasca State Park). The Itasca region did have significantly lower kDOC:kDOP values than the other two regions (**Figure 4**, Chi Square Median test, p = 0.0297). The Itasca region also had a typical kDOC:kDOP value less than 1 (Wilcoxon signed-rank, p = 0.014), indicating that for this region DOP degradation constants were significantly higher than DOC degradation constants.

#### Estimates of DOC and DOP Bioavailability

It took approximately 9 months for degradation models to give reasonable predictions for the total size of the recalcitrant DOC pool (i.e., the 3 parameter fit models outperformed the 2 parameter fit models). Prior to 9 months, 2 parameter fit models outperformed the 3 parameter models so a clear asymptote was not indefinable. In contrast, for DOP, the 2-parameter fit model was always a better fit than 3 parameter fit model, so a modeled estimate of the recalcitrant DOP concentration was not possible. Therefore, in order to estimate and compare the relative sizes of the bioavailable pools of DOC and DOP, we calculated these values using the lowest measured value of DOC and DOP over the course of the incubation. By subtracting this lowest value from the starting concentration, we estimated the amount of DOC or DOP that had been degraded during the incubation period and used this as an estimator of the size of the BDOC and BDOP pools. BDOC values in these lakes ranged from ∼19µM to ∼397µM with a median value of 118µM, and relative BDOC values ranged from 4.4 to 53.8% of the total DOC pool with a median value of 25.0% with no regional differences in BDOC. In contrast, a much larger portion of the DOP pool tended to be bioavailable. Eight systems had BDOP values over 95% of the total DOP pool and the median value for all the lakes was 78%. Relative BDOP was also more variable than BDOC and had an interquartile range of 40.8 to 97.5% compared to 21.0 to 30.6% for BDOC (**Figure 5**). Absolute values for BDOP concentrations ranged from 0.01µM to 0.82µM and three-quarters of the samples had BDOP concentrations below 0.26µM. As with BDOC, there were no significant regional difference in BDOP.

In these systems, relative %BDOP was positively correlated to the initial concentration of DOP in the system, suggesting that systems with a larger DOP pools not only had more BDOP, but also had a larger fraction of the DOP pool that was bioavailable (**Figure 6**, p = 0.0043). Additionally, relative BDOP was negatively correlated to the initial DOC: DOP ratio, indicating that DOP was relatively less bioavailable when it was scarce relative to DOC (**Figure 7**, p = 0.0002). However, relative BDOP was not significantly correlated to other individual element pools (DOC, TDN, DIC) nor was it significantly correlated to any of the lake characteristics measured (pH, alkalinity, chlorophyll or SUVA). Temperature data were only available for the 10 lakes in the Itasca region, but within this subset of the data temperature was not a significant predictor of BDOP concentration or relative BDOP percentage. Absolute concentrations of BDOP also showed a strong negative correlation with the DOC:DOP ratio, suggesting that BDOP was being drawn down at high DOC:DOP ratios (**Figure 8**), but absolute BDOP was not significantly correlated to other elemental or lake characteristic measurements.

Relative %BDOC was not significantly correlated with any of the measured elemental parameters (initial DOC, TDN, TDP, SRP, DOP or DOC:TDN, DOC:TDP, DOC:SRP, or DOC:DOP). Interestingly, relative %BDOC was also not significantly correlated to SUVA but the absolute size of the BDOC pool did show a significant positive correlation with SUVA (**Figure 9**; p = 0.0217). This positive trend with SUVA is likely partially explained by the fact that high SUVA systems tend to be high in DOC (and the absolute amount of BDOC was strongly positively correlated to DOC concentration), but in combination with the fact that relative BDOC did not change with SUVA, this suggests that even systems dominated by more terrestrial-like organic matter contain large amounts of bioavailable DOC. Furthermore, absolute BDOC concentration was strongly positively correlated to TDN, TDP, and chlorophyll (**Figure 10**, p < 0.0001), consistent with the accumulation of labile DOC under high nutrient conditions and high productivity.

# Stoichiometry of Bioavailable Nutrients

Overall, ratios of bioavailable C and P were much lower than the bulk chemistry pools. DOC:TDP ratios ranged from 319 to 7122:1 with a median of 1595:1 while DOC:DOP ratios ranged from 679 to 15360:1 with a median of 2449:1 in the systems we examined. In comparison, BDOC:BTDP ranged from 133 to 8848:1 (this high point was an outlier with the next highest value being 2943) and median value of 746:1. BDOC:BDOP ranged from 144 to 9719:1 with a median value of 843:1. Previous work showed that assemblages of aquatic heterotrophic bacteria in lakes have mean biomass C:P ratios around 102:1 (Cotner et al.,

2010), while individual strains can have highly variable biomass composition with values well over 1000:1 (Godwin and James, 2015). Therefore, the stoichiometry of the bioavailable nutrients measured in this study more closely match typical bacterial biomass stoichiometry than measures of bulk nutrient chemistry. Bioavailable nutrient stoichiometry was also positively correlated with bulk nutrient stoichiometry (**Figure 11**, p < 0.0001).

# DISCUSSION

The data on degradation and bioavailability of DOC and DOP provide insights into three important areas. First, across all lakes, median BDOC and BDOP turnover times were similar equal for BDOC and BDOP (∼100 days), but in Itasca State Park lakes, BDOP turnover was significantly faster than BDOC (in

and the original concentration of DOP in the sample (log transformed) for 26 lakes. One lake had to be excluded because the initial SRP concentration was below the method detection limit, so a DOP concentration could not be calculated. A positive relationship shows that as DOP concentration increased among systems, the relative lability of DOP increased.

some cases by as much as 20 times). This spatial variability highlights the need for more empirical measurements of DOP degradation rates from a variety of systems to better understand potential spatial patterns. In our study, degradation rates of DOC and DOP could not be explained by the other elemental pool sizes or environmental characteristics, further emphasizing the need for more work in this area. Second, we show that the portion of the DOP pool that is bioavailable was extremely variable across systems, but often exceeded 50% and was strongly related to the DOC:DOP ratio of the system. BDOP was drawn down in systems with high DOC:DOP ratios (where P was more likely limiting) and therefore it is likely that in many systems, DOP represents a high-quality P resource to supplement inorganic P availability. Thirdly, the bioavailability of the DOM pool suggests that the nutrient stoichiometry of available resources in aquatic systems may be more similar to the biomass demands of aquatic microbes than previously

FIGURE 8 | Scatterplot showing the linear regression function comparing the concentration of BDOP to the DOC:DOP ratio of the initial sample (log transformed). The significant negative relationship demonstrates BDOP is drawn down as DOP becomes scarce relative to DOC.

thought. This has important implications for understanding the experienced nutrient imbalance by heterotrophic bacteria and in turn, understanding how bacteria couple multiple elemental

#### Degradation Rates of DOC and DOP

cycles in aquatic systems.

carbon compounds (high SUVA) still have large pools of BDOC.

Degradation rates of DOC in freshwater systems have been the topic a numerous papers and a recently published metaanalysis showed that DOC decay rates can vary by several orders of magnitude across different systems (Catalán et al.,

2016). Their dataset included 33 bioassay measurements from lakes in a similar climatic region to our study and these 33 lakes had a median kDOC value of 0.0021, the same order of magnitude as our median value of 0.0077. Estimates of DOP degradation rates are scarcer in the literature and are dominated by estimates from marine systems. One study in the Baltic Sea estimated DOP turnover times to be between 3 to 4 days, about twice as fast as the shortest turnover time in our data set (Nausch and Nausch, 2006). In contrast, DOP turnover times from the North and South Atlantic Ocean subtropical gyres were 5.5 months and 10.5 years respectively (Mather et al., 2008). Another study from Station ALOHA, in the North Pacific Subtropical Gyre found that DOP turnover time increased with depth and ranged from 12 to 268 days at a single sampling site (Björkman and Karl, 2003). One major difference between these systems that could explain the differences in turnover times is productivity, with the Baltic being highly productive compared to Station ALOHA. However, DOP degradation rates were not significantly correlated to chlorophyll levels in our study, so within our systems, productivity was not a good predictor of DOP turnover. The variability of literature measurements, along with the variability in DOP turnover times in this study, highlight the need for more direct measurements of DOP degradation across a variety of systems to better constrain typical DOP degradation dynamics and to better understand the factors controlling them.

Here, it is also important to consider the difference in model fits between the DOC and DOP degradation curves. DOC degradation was incredible, consistent across system, allowing for the degradation models to accurately estimate the recalcitrant portion of the DOC pool and degradation rate (Supplementary Figure 2). However, in the case of DOP there was more variation in degradation pattern across systems and the proximity of many of the DOP measurements to a nonzero asymptote made it difficult to use the degradation model to accurately estimate a refractory DOP pool (Supplementary Figure 3). Nonetheless, we elected to use a two parameter decay model to estimate DOP degradation rates because the model provided a reasonable overall fit (R 2 -0.77, Supplementary Figure 3) and gave rates that could be directly compared to the DOC estimates. This approach does by definition infer that a refractory DOP is not present (i.e., 100% bioavailability) and tends to homogenize the overall DOP degradation patterns. A closer view of the DOP degradation plots (Supplementary Figure 3) revealed at least three different general patterns for DOP degradation: (1) DOP was rapidly degraded to a zero intercept, (2) DOP was slowly degraded to a zero intercept, or (3) DOP degraded slowly to a non-zero intercept but a zero intercept is inferred by the fit model. This third case would suggest that there is in fact a refractory DOP pool in these systems and this observation further supports our use of measured concentrations differences rather than DOP degradation models to calculate the size of the bioavailable DOP pool in these systems.

#### Bioavailability of DOC and DOP

Our measurements of the relative bioavailability of DOC are well within the range measured in other aquatic systems (Sondergaard and Middelboe, 1995; Stets and Cotner, 2008; Catalán et al., 2015; Helton et al., 2015; Frey et al., 2016). All but one of the systems we measured had BDOC values less than 50% of the total DOC pool, further supporting the idea that the bulk portion of DOC in freshwater is recalcitrant. However, these relative BDOC measures are quite high compared to marine systems, suggesting

the C:P ratio of the bioavailable pool is lower than the measured C:P of the system. Additionally, the slope of the regression fit is greater than 1 meaning that BDOC:BDOP ratios more closely match bulk DOC:DOP ratios when they are higher (in systems that are DOP poor relative to DOC).

that exports to freshwater represent a younger, more labile carbon source than those found in marine systems. Furthermore, absolute BDOC concentrations were strongly related to TDN, TDP, and chlorophyll concentrations suggesting that nutrient availability is an important control on the accumulation of BDOC. This finding contrasts previous work that showed no significant correlation between BDOC and nutrient conditions (Stets and Cotner, 2008); however, it should be noted that Stets and Cotner also measured positive correlations between BDOC and both TDP and chlorophyll concentrations but the relationships were not statistically significant in the 12 lakes they studied. Nonetheless, our data also suggest that stoichiometry may be an additional constraint to BDOC accumulation. The fact that the slope in **Figure 11B** was more than 1 suggested that BDOC was accumulating disproportionately when DOC:DOP ratios were highest.

It was interesting to note that while the amount of BDOC was positively correlated to SUVA, the relative lability was not (**Figure 9**). SUVA has been shown to be highly correlated with the aromaticity of the DOM pool (Weishaar et al., 2003), so this pattern suggests that increasing aromaticity of the DOM pool does not significantly decrease its bioavailability. Furthermore, SUVA is considered a useful proxy for terrigenous organic matter with higher SUVA systems receiving large terrigenous inputs. SUVA was not strongly correlated to relative or absolute concentrations of BDOP, so we suggest that terrigenous inputs in our study systems represent a labile source of organic carbon but not organic phosphorus.

As with DOP degradation kinetics, estimates of the relative pool size of BDOP are sparse. However, the values reported in the literature are in good agreement with the values we measured here. In a Baltic Sea study, the DOP pool was 75% bioavailable (Stepanauskas et al., 2002) and a similar value (33.2– 60%) was reported for 3 stations in the central Baltic as part of a different study (Nausch and Nausch, 2007). A more recent analysis suggested that ∼40% of the DOP in four boreal lakes was bioavailable (Soares et al., 2017). Our study of 27 unique systems supports the idea of very labile BDOP with a median value of ∼78%, but also highlights the large amount of variability in BDOP across systems. It should be mentioned, however, that our incubations lasted much longer than these other studies (150 days compared to ∼7 days), which should have resulted in higher estimates of BDOP as our incubations would capture both rapidly degrading and slowly degrading DOP compounds. The fact that many of our incubations continued to show DOP losses up until 150 days into the incubations (Supplementary Figure 3) demonstrates the need for longer term incubations to fully describe the BDOP pool. On the other hand, our first sampling period occurred after ∼30 days of incubation, which limited our ability to describe the degradation rates of the fastest degrading DOP pool. Given the rates of degradation documented in the literature and also the fact than many of our incubations showed major losses of DOP within the first 30 days, a stratified sampling method with high frequency measurements over the first few weeks and then less frequent measurements over several months may provide the best overall picture of DOP degradation.

Our findings also suggest that organic matter stoichiometry is an important control on the bioavailability of DOP. We found that the DOC:DOP ratio was a significant predictor of both absolute and relative BDOP (**Figures 7**, **8**), with higher DOC:DOP ratios correlated to lower BDOP concentrations and percentages and a higher BDOC:BDOP ratio relative to the DOC:DOP pool (**Figure 11**). These patterns suggest that relative size of the BDOP pool decreased when the DOP was small relative to DOC. Presumably when the DOC pool size was large relative to DOP, microbes were more likely to be Plimited and consumed any bioavailable P. In lakes with lower DOC:DOP ratios, the organic matter pool would more closely resemble the biomass requirements of aquatic bacteria and/or the microbes are more likely to be limited by organic C rather than P.

#### Stoichiometry of Bioavailable Nutrients

Ecological stoichiometry provides a guide for predicting the cycling of multiple nutrients by examining the elemental balance between organisms and their resources. However, a fundamental problem associated with understanding these imbalances is our capacity to know what the resource availability is that organisms actually experience. Our ability to accurately describe the resource imbalance experienced by bacterial communities in situ is hindered by our lack of simultaneous measurements of the bioavailability of multiple elements (Berggren et al., 2014; Soares et al., 2017). Here, we observed that the experienced BDOC:BDOP resource ratios of aquatic bacterial communities were typically less than measured DOC:DOP pools (**Figure 11**). Therefore, bulk chemical measurements likely overestimate the size of the labile DOC pool or underestimate the size of the DOP pool. The fact that our measurements of BDOP indicated that large fractions of the DOP pool could be bioavailable while there clearly was a non-labile pool of DOC suggests that DOC measurements overestimate the BDOC pool, which many other studies have observed. Nonetheless, previous measurements of imbalance using bulk chemistry data likely overestimate the actual imbalance experienced by these communities particularly in more carbon-limited systems. Furthermore, our results suggest that it is at lower C:P ratios, i.e., more eutrophic systems, where chemical measurements of DOC:DOP are likely to overestimate the bioavailable pool of DOC (**Figure 11**). Although there is more BDOC being produced in these systems, the microbial biomass is more likely to be limited by the availability of organic carbon, resulting in more drawdown and an increased proportion of the DOC pool being recalcitrant. This has important implications for understanding how bacteria couple C and P cycles in freshwater because the experienced imbalance between consumer and resources governs the differential recycling of those nutrients. If heterotrophic bacteria in aquatic systems experience a more balanced resource pool in terms of the C:P ratio than previously thought, this should result in more efficient C cycling as compared to predictions based on bulk chemistry ratios.

# CONCLUSIONS

The 27 aquatic systems examined in this study demonstrate that DOP bioavailability was quite variable across systems but was strongly predicted by the DOC:DOP ratio of the system. The bioavailability of DOC was more tightly constrained due to an increased proportion of recalcitrant material relative to DOP and it was not predicted by organic pool stoichiometry but rather was strongly related to nutrient conditions (both TDN and TDP concentration). Exponential decay models fit the loss of DOC tightly, but were not as strong of a fit for DOP. Given that DOP turnover times were calculated using an exponential model with no asymptote (and therefore assuming 100% bioavailable DOP in all samples) our estimates for turnover times are likely skewed high, particularly for systems with sizeable recalcitrant DOP pools. Despite these potential limitations, our data suggest that DOP turnover time was significantly faster than DOC turnover in the Itasca region (in one case, ∼20 times faster), but DOC and DOP had similar rates of turnover in the other two regions. This suggests that overall DOP in freshwater systems is turning over as quickly or more quickly than DOC. More measurements are needed in order to properly assess this spatial variability and determine if there are any broader geographic patterns in the relative turnover rates of DOC and DOP in freshwaters, particularly because variability in DOC and DOP degradation rates could not be explained by inorganic nutrient pool sizes or lake characteristics measured in this study.

Furthermore, we have shown that organic matter stoichiometry is an important control on the accumulation of bioavailable DOP in aquatic systems. Relative bioavailability of DOP was positively related to the concentration of DOP in the sample and negatively correlated to the initial DOC:DOP ratio, suggesting that DOP accumulates in systems that are less P-limited. In contrast, the initial organic matter stoichiometry was not predictive of relative or absolute BDOC. Instead, BDOC accumulation was associated with high nutrients (TDN and TDP) and high production (chlorophyll). Absolute BDOC was also strongly correlated to SUVA values, providing evidence that terrestrial subsidies represent a labile source of DOC in the systems studied. Finally, incorporating measures of nutrient bioavailability decreased the predicted nutrient imbalance experienced by heterotrophic bacteria in aquatic systems, which has important implications for understanding the coupling of C and P biogeochemical cycles.

# AUTHOR CONTRIBUTIONS

ST performed the field work and data collection. JC and ST jointly contributed to the research design, data analysis, and manuscript writing.

# FUNDING

This work was funded by the Itasca Graduate Student Fellowship and Moos Graduate Fellowship in the aquatic science awarded to ST by the College of Biological Sciences at the University of Minnesota and NSF award 1257571 to JC.

# ACKNOWLEDGMENTS

The authors would like to thank Bri Loeks-Johnson and Dr. Lesley Knoll for their help in the field collecting samples and Andrea Little for her technical support on the project.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvs. 2018.00062/full#supplementary-material


Sea Res. Oceanogr. Res. Papers 32, 223–235. doi: 10.1016/0198-0149(85)90 030-5


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Thompson and Cotner. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Phosphorus Forms in Sediments of a River-Dominated Estuary

Sheree J. Watson<sup>1</sup> \*, Barbara J. Cade-Menun<sup>2</sup> , Joseph A. Needoba1,3 and Tawnya D. Peterson1,3

*1 Institute of Environmental Health, Oregon Health and Science University, Portland, OR, United States, <sup>2</sup> Agriculture and Agri-Food Canada, Swift Current Research and Development Centre, Swift Current, SK, Canada, <sup>3</sup> OHSU-PSU School of Public Health, Oregon Health and Science University, Portland, OR, United States*

Estuaries are biologically productive transition zones between land and sea that play a vital role in transforming, recycling, and sequestering nutrients and organic matter, thus influencing nutrient loading to coastal systems. Yet, the processes involved in phosphorus (P) transformation and cycling among inorganic and organic P forms are poorly known in estuaries. To better understand the potential for P transformation and sequestration, we identified P forms and estimated their contributions to total P in intertidal wetland sediments of a river-dominated estuary (Columbia River, Oregon, USA) using solution <sup>31</sup>P nuclear magnetic resonance spectroscopy (P-NMR). Inorganic P forms dominated sediment P extracts throughout the estuary, with orthophosphate accounting for 71–84% of total extracted P. However, biologically-derived inorganic and organic P forms were also detected. Polyphosphates were found in sediment extracts throughout the estuary, contributing as much as 10% of extracted P. Similar to other wetlands, orthophosphate monoesters and diesters made approximately equal contributions (∼ 20%) to total extracted P. However, monoesters (e.g., phytate) were more abundant in sedimentary environments characterized by low organic matter content, while diesters (e.g., DNA) were more abundant in sedimentary environments with high organic matter, regardless of salinity. Collectively, the data show strong evidence for P transformation in sediments of a large, river-dominated estuary, which influences its transport to the coastal Pacific Ocean via the expansive Columbia River plume.

Keywords: phosphorus, <sup>31</sup>P-nuclear magnetic resonance, sediments, river, estuary

# INTRODUCTION

Phosphorus (P) is an essential but often limiting nutrient in freshwater aquatic and terrestrial ecosystems (Daniel et al., 1998); however, downstream systems are increasingly under threat from nutrient-stimulated eutrophication (Conley et al., 2009; Tiessen et al., 2011). Watershed losses result from inefficient recycling, or poor sequestration, and lead to P transport from soils to surface waters; ultimately, excess P ends up in estuaries and coastal systems where it often results in nutrient-driven eutrophication. Eutrophication can lead to hypoxia, acidification, and harmful algal blooms (Paerl, 2006; Bricker et al., 2008) and represents a growing worldwide threat to human health and well-being (Millenium Ecosystem Assessment, 2006).

Estuaries are biologically productive transition zones comprising a variety of habitats including wetlands, mudflats, and seagrass beds between freshwater and marine systems. They play a vital role in transforming, recycling, and sequestering nutrients and organic matter (OM) prior to entering

#### Edited by:

*Toshi Nagata, The University of Tokyo, Japan*

#### Reviewed by:

*Masahiro Suzumura, National Institute of Advanced Industrial Science and Technology (AIST), Japan Sue Newman, South Florida Water Management District, United States*

> \*Correspondence: *Sheree J. Watson shereew@hawaii.edu*

#### Specialty section:

*This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science*

Received: *28 February 2018* Accepted: *08 August 2018* Published: *04 September 2018*

#### Citation:

*Watson SJ, Cade-Menun BJ, Needoba JA and Peterson TD (2018) Phosphorus Forms in Sediments of a River-Dominated Estuary. Front. Mar. Sci. 5:302. doi: 10.3389/fmars.2018.00302* the coastal ocean (Barbier et al., 2011). Tidal mixing of fresh and saltwater within estuaries leads to particle flocculation (Sholkovitz, 1976) and the localized resuspension of bottom material in turbidity maxima (van Beusekom and Brockmann, 1998; Small and Prahl, 2004). Combined with downstream transport in rivers, estuaries are vulnerable to nutrient pollution because dissolved inorganic P (orthophosphate) desorption from sediment surfaces occurs when anions in seawater compete with phosphate anions for binding sites, leading to P efflux (Fox et al., 1986; Froelich, 1988; House and Warwick, 1999; Monbet et al., 2010). Moreover, the reducing conditions associated with intense OM remineralization increase P desorption, setting up a positive feedback loop whereby eutrophic conditions contribute to P mobilization and limit nutrient sequestration (Howarth et al., 2011; Li et al., 2013). This impairs the critical function of estuaries and wetlands, which is to reduce nutrient loads before they reach coastal waters (Barbier et al., 2011).

Phosphorus retention, or sequestration, in wetlands can occur via chemical, physical, or biological means. For example, orthophosphate is retained in sediment pore waters through adsorption from the water column, particularly in sediments dominated by clays and minerals (Reddy and DeLaune, 2008). Phosphorus (P) buffering is defined as the equilibrium reached between adsorption and desorption of orthophosphate at the water/sediment interface, which reduces variability in orthophosphate concentrations irrespective of input or removal processes (Froelich, 1988). High concentrations of inorganic P and minerals (i.e., Fe, Al, Ca, and Mg) in sediments result in the formation of both soluble and insoluble P compounds, predominantly Fe and Al minerals in acidic soils and Ca compounds in near-neutral to alkaline soils (Stevenson, 1986; Reddy et al., 1999; Richardson and Reddy, 2013). However, seasonality in hydrology, including flooding and drying, produces changes in soil redox conditions, which play a governing role in transformation, storage, and transport of P in wetlands.

In comparison, organic P in wetlands is associated with living systems including plants, microbes, detritus and organic matter, and commonly accounts for large fractions of total P (Vaithiyanathan and Richardson, 1997; Reddy et al., 1999). Although plant storage material is often the dominant source of organic P in wetland soils (Reddy and DeLaune, 2008), soil microbes play a key role in controlling degradation of organic matter and cycling of P in wetlands. For example, microbes secrete enzymes that readily mineralize organic P and can quickly assimilate orthophosphate released into pore water (Richardson, 1985; Kellogg and Bridgham, 2003; Noe et al., 2003). Microbes also regulate P in sediments through biomass, which can account for up to 25–50% of the total P in wetland mineral sediments (Wright et al., 2001; McDowell and Sharpley, 2003) and up to 70% of total P in hydrologically isolated sediments (Williams and Silcock, 2001).

Because the mobility and retention of P depends upon its chemical forms, the determination of chemical structures is critical for identifying possible sources and for estimating the potential for retention, sequestration, and transformation (Celi and Barberis, 2005; Condron et al., 2005; Quiquampoix and Mousain, 2005). The goal of this study was to investigate P forms and their contributions in a mesotidal, river-dominated system characterized by rapid flushing and short water residence times (1–5 d; Chawla et al., 2008; Columbia River estuary, Oregon, USA). Previous work in this estuary showed that watercolumn dissolved inorganic phosphate (DIP) exhibited seasonal non-conservative increases along a salinity gradient, suggesting the presence of P sources that were not accounted for when orthophosphate was measured in river and ocean end members (Gilbert et al., 2013). To address the knowledge gap, our goal was to determine what P forms are present in the peripheral bay sediments that might contribute to fluxes of DIP and to characterize variability in estuarine P forms as a function of sediment properties.

# MATERIALS AND METHODS

#### Field Sites

The Columbia River is the second largest by discharge in the USA (Sherwood et al., 1990), culminating in a river-dominated, salt-wedge estuary with strong seasonality in salinity and biogeochemical properties (Roegner et al., 2011). The U.S. Pacific Northwest coast is strongly influenced by summer upwelling when northerly winds promote the offshore transport of surface waters, which are replaced by nutrient-rich subsurface oceanic water (Huyer, 1983; Grantham et al., 2004). The Columbia River discharge varies with season, peaking at >10,000 m<sup>3</sup> s <sup>−</sup><sup>1</sup> during the snowmelt-driven spring freshet (April–June) and declining to < 2,000 m<sup>3</sup> s −1 in the dry season (July–October). Salt intrusion into the estuary is greatest during periods of low discharge, which typically occur coincidentally with upwelling-favorable conditions between June and September (Chawla et al., 2008). The estuary exchanges the majority of water in north and south paths that run along the peripheral bays, including Baker Bay (north) and Youngs and Cathlamet Bay (south; **Figure 1**).

Compared to the estuary at large, the three peripheral bays included in this study share some similarities in terms of sediment characteristics (Sherwood et al., 1984; Sherwood and Creager, 1990); however, there are differences among them in terms of the composition (i.e., percent silt and clay vs. sand) and grain size. Although Baker and Youngs Bays receive considerable deposits of fine sand, silts, and clays, the sources of sediment deposits differ (Sherwood and Creager, 1990; Simenstad et al., 1990). Baker Bay and Cathlamet Bay receive fine sediments from suspended material, while Youngs Bay received fortnightly inputs of fine sediment from ephemeral estuarine turbidity maximum events (Sherwood and Creager, 1990). In addition, there are seasonal differences in depositional patterns, with the western side of Baker Bay receiving coarser sediment deposits than the east side in February and June. In Youngs Bay, the upstream reaches tend to accumulate a combination of long-term deposits of coarse and fine sediments, with the latter dominating on the fringes. Closer to the estuary's south channel, sediment deposits are coarser than in the other peripheral bay areas. Salinity varies throughout the year in peripheral bays as a function of river discharge; Baker Bay on the northern side of the estuary is mesohaline (i.e., brackish), with a range of 5–20 practical salinity

units (PSU) during periods of relatively low flow (i.e., summer and autumn), while Youngs and Cathlamet Bays on the southern side of the estuary are oligohaline (0–5 PSU). The furthest upriver site, Cathlamet Bay, includes a diverse region of islands, tidal flats and marshes that experience seasonal exposure to salinity at depth.

#### Sample Collection and Chemical Analysis

Sediment texture (percent clay, silt) was determined using the hydrometer method (AgSource Laboratories, OR) on sediment samples from August 2013 prior to the collection of samples for the analysis of P forms. From September 2013–August 2014, sediment samples for the analysis of P forms were collected 1–2 h before low tide at sites within Baker Bay (four sites: BBB, BBI, BBA, BBC), Youngs Bay (two sites: YBB, YBM) and Cathlamet Bay (one site: CBK; **Figure 1**), all on exposed mudflats lacking any emergent vegetation at low tide. Cores were collected by pressing 50 mL Falcon conical centrifuge tubes with the bottom cut off into sediments. A full tube represented a sediment volume of ∼43 cm<sup>3</sup> (with a core diameter of 2.3 cm and a length of 10 cm). The tubes were transported back to the laboratory on ice (∼2–6 h) and kept at −20◦C pending analysis. For this study, sediment cores from March–April and May– August were coarsely sieved to remove large debris, air-dried in a fume hood, ground and homogenized prior to P-NMR extraction (Cheesman et al., 2013). Total P concentrations in the unextracted sediments were determined by ignition followed by extraction with dilute H2SO<sup>4</sup> (SmartChem 170, Unity Scientific, MA). Following combustion, dissolved P from the sediment extracts was determined colorimetrically (Murphy and Riley, 1962) and total OM content was estimated by loss on ignition (LOI) at 500◦C (Nelson and Sommers, 1996).

Samples collected from waters overlying sediments were filtered in the field by syringe (60 mL) through combusted (4 h at 450◦C) 0.7µm, 25 mm GF/F filters (Whatman, NY) into 60 mL acid-cleaned HDPE bottles (Nalgene) and transported to the lab on dry ice in a cooler, where they were stored at −20◦C pending processing. Total nitrogen (N) and total P concentrations were determined on unfiltered water samples. A rapid flow analyzer (Astoria Analyzer, Astoria Pacific, Clackamas, OR) was used for colorimetric determination of dissolved nitrate+nitrite (NOx), ammonium, nitrite, and DIP (here measured as molybdate-reactive orthophosphate) concentrations and for measurement of total N and P after alkaline persulfate digestion (US EPA, 1983a,b; Antweiler et al., 1996; Zimmerman and Keefe, 1997; Patton and Kryskalla, 2003). Further, salinity and temperature measurements were checked during time of sampling by in situ sensor platforms (SATURN04, Cathlamet Bay; SATURN07, Baker Bay; SATURN09, Youngs Bay) in the Columbia River from a publicly available online database (www.stccmop.org) maintained by the Center for Coastal Margin Observation and Prediction (Baptista et al., 2015).

# Extraction of Phosphorus

Using a modification of the Cade-Menun and Preston (1996) procedure, approximately 3.0 g of air-dried sediment was suspended in 30 mL of NaOH-EDTA (0.25 M NaOH, 50 mM Na2EDTA) and sonicated (Branson Digital, Cleveland, OH) for 1 min at 20 kHz (∼55 W) in an ice bath followed by shaking at 150 rpm for 4 h. Samples were centrifuged for 10 min at 1,200 × g and the supernatant was removed, neutralized with 10% HCl and frozen at −80◦C prior to lyophilization. An aliquot of each NaOH-EDTA extract was analyzed for Al, Ca, Fe, Mg, manganese (Mn), and P (mg L−<sup>1</sup> ) by inductively coupled plasma atomic emission spectroscopy (ICP-AES) using a Plasma 400 emission spectrophotometer (PerkinElmer, Waltham, MA).

#### Solution P-NMR

Lyophilized extracts were redissolved in a solution combining 0.6–0.65 mL each of D2O, H2O, NaOH-EDTA solution, and 10 M NaOH, vortexed, centrifuged at 1,500 × g for 20 min, and stored at 4◦C within 12 h of analysis. If precipitate formed (observed in two samples, from BBA and BBC), it was filtered out using 0.2 µm polycarbonate syringe filters prior to analysis. Spectra were collected on a Bruker Avance 500 MHz NMR spectrometer with a 10 mm broadband probe. The NMR delay times were determined based on the ratio of concentrations of P/(Fe+Mn) within the extracts (McDowell et al., 2006). A temperature of 21◦C and a 45◦ pulse were used, with a pulse delay of 4.5 s. Acquisition time was 0.5 s and 2,800–5,830 scans (∼4–8 h) were collected. Samples were analyzed with no spinning and no proton decoupling.

#### Data Analysis and Statistics

The NMR spectra were processed using NMR Utility Transform Software (NUTS; Acorn NMR, Livermore, CA, 2000 edition). Processing included standardizing the orthophosphate peak for each sample at 6.000 ppm. Phosphorus forms (P forms) were identified by their chemical shifts using published libraries for known peaks, with spiking for confirmation (Cade-Menun, 2015). Spectra were processed with 7–10 Hz line broadening for full spectra and 3 Hz line broadening to identify peaks in the orthophosphate monoester and diester regions (Cade-Menun and Liu, 2014). Corrections were made to account for orthophosphate diester degradation by summing the peak areas associated with degradation products (αand β-glycerophosphate and mononucleotides, degraded from phospholipids and RNA respectively), subtracting this sum from the total monoester area, and adding it to the total diester area (Young et al., 2013; Schneider et al., 2016).

A constrained ordination was performed on the transformed P forms (abundance) and sediment environmental data (TotalP, LOI, Fe, Mn, Ca, and Mg). A redundancy analysis (RDA) is an extension of principal components analysis (PCA) designed to identify or reduce complexity of environmental data. The principal components were constrained to be linear combinations of the environmental variables, with the goal being to explain variation of the dependent variables (P form abundance) as much as can be explained by independent variables (sediment properties; Legendre and Legendre, 1998; Ramette, 2007). Phosphorus forms and sediment data were normalized with centered log ratio (clr) transformations (Abdi et al., 2015). Statistical analyses were performed with R using the vegan package (R Core Team, 2015).

# RESULTS

#### Chemical Analysis

Physicochemical characteristics associated with each sample are shown in **Table 1**. It should be noted that the small number of samples analyzed in this study limits spatial and temporal comparisons made across sites and environmental characteristics. Total sediment P concentrations varied by more than two-fold, with the highest value observed at CBK (1,366 mg kg−<sup>1</sup> ) and the lowest at BBC (601 mg kg−<sup>1</sup> ). The OM (LOI%) content mirrored total sediment P (11.3% at CBK, 2.2% at BBC). Sediment texture analysis demonstrated the largest percentage of clay and silt at the western side of Baker Bay (42.4%, BBB) and at the mouth of Youngs Bay (35%, YBM), which come from different sources in the estuary including fine suspended material and estuarine turbidity maxima events, respectively (Sherwood and Creager, 1990).


*†Water collected above the sediments; DIP, dissolved inorganic phosphorus (P); TotP, total P; NO<sup>x</sup> , NO*<sup>−</sup> <sup>3</sup> <sup>+</sup>*NO*<sup>−</sup> 2 *; TotN, total dissolved nitrate/nitrite; NH*<sup>+</sup> 4 *, ammonium; PSU, practical salinity units; and Temp, temperature.*

*‡Total P (sediment) mg kg*−<sup>1</sup> *dry sediment weight, and organic matter was calculated as loss-on-ignition (LOI); Clay and Silt were analyzed for sediments collected in August 2013. §Salinity and temperature collected from local autonomous sensors (SATURN04, Cathlamet Bay; SATURN07, Baker Bay; SATURN09, Youngs Bay).*

Monthly measurements of DIP, NOx, and ammonium (NH4) at the seven sites over a nine-month period (Jan– Sep 2014) are provided in Supplemental Materials (**Figure S2**). Only the measurements that coincide with the analysis of sediment P forms are indicated in **Table 1**. Water column DIP concentrations were low from March–May; in contrast, the sample taken in August at site BBA was one of the highest observed (4.79µM; **Table 1**). The 9-month time series confirmed that DIP levels were low in the winter and spring but increased in the late summer (**Figure S2**). Higher salinity (6–18 PSU vs. 0–4 PSU) and higher DIP concentrations were observed in the August water sample compared to the others. High concentrations of NO<sup>x</sup> in overlying waters were observed in samples collected from March–May, with the exception of the BBC site; NO<sup>x</sup> concentrations were lowest in late summer (**Table 1**, **Figure S2**). Ammonium concentrations mirrored DIP throughout the year (**Figure S2**), with the lowest concentrations observed in spring and the highest value in the late summer (**Table 1**).

The P recovery following sediment NaOH-EDTA extraction varied from a low of 1.5 at site BBA to a high of 24.5% at YBB (**Table 2**). The samples with the lowest OM tended to have the lowest Fe and Mn concentrations in the extracts (e.g., 0.45 and 0.34 mg kg−<sup>1</sup> , respectively at BBC; **Table 2**), and therefore may have required longer delay times to be fully quantitative for extracted P. Higher concentrations of paramagnetic ions in samples (e.g., Fe, Mn) shorten <sup>31</sup>P nuclei relaxation times, reducing the delay time required between pulses (McDowell et al., 2006), but potentially increasing line-broadening of peaks. Since it is unknown whether variation in the efficiency of P extraction among the samples led to any biases due to preferential extraction of certain P forms, quantifications should be interpreted with caution. Of the cations extracted with P, Ca was most abundant, ranging from 110 to 2077 mg kg−<sup>1</sup> across samples (**Table 2**), while Fe concentrations ranged from 1.5 mg kg−<sup>1</sup> (BBA) to 73.7 mg kg−<sup>1</sup> (YBB), however caution should be taken in interpretation as these cation concentrations do not necessarily represent their actual concentrations in sediments.

#### Phosphorus Forms and Their Contributions to P in Sediments

**Figure 2** shows example P-NMR spectra, which demonstrate peaks associated with P forms detected in estuarine sediments of the Columbia River; the remaining spectra are included as Supplemental Data (**Figure S1**). Chemical shifts associated with the peaks are summarized in **Table S1** and the percentages of total extracted P accounted for by identified P forms are in **Table S2**.

Total extracted P was dominated by inorganic forms (P<sup>i</sup> ; **Figure 3A**), which exceeded contributions from organic P (Po) by 2–6 times. Orthophosphate, which varied twentyfold across sites (8.1–231.3 mg kg−<sup>1</sup> at BBA and YBB, respectively), accounted for 70.7–84.4% of total P [Σ (P<sup>i</sup> + Po)] and 87–99% of P<sup>i</sup> (**Table S2**). Both pyrophosphate and polyphosphate were detected at all sites except BBA in the summer (where only polyphosphate was detected). There was little variation in concentration among most sites (**Figure 3B**), with pyrophosphate and polyphosphate accounting for 1–13% of P<sup>i</sup> and 0.8–10.6% of total P, respectively (**Table S2**). The largest total polyphosphate (pyrophosphate + polyphosphate) concentrations were observed at the west end of Baker Bay (6.2 mg kg−<sup>1</sup> , BBI), while the lowest were observed in Cathlamet Bay, located in a southern bay of the estuary (CBK, 0.8 mg kg−<sup>1</sup> ; **Table 3**).

Concentrations of P<sup>o</sup> ranged from 2.9 to 59.5 mg kg−<sup>1</sup> , accounting for 13.5–26.6% of total extracted P in sediments (**Figure 3C**). Orthophosphate monoesters made the largest contribution to Po, where they accounted for 66–94% of P<sup>o</sup> and 9.4–25% of total extracted P. Identified monoesters comprised stereoisomers of inositol hexakisphosphate (IHP), including chiro, myo, neo, scyllo-IHP, as well as α- and β-glycerophosphate and mononucleotides. The IHP stereoisomers were broadly

TABLE 2 | Concentrations of P and selected cations in alkaline extracts from estuary lateral bay sediments. Location Sample Date (Time) NaOH-EDTA extracts P Fe Mn Al Mg Ca P Recovery†% mg kg−<sup>1</sup> Cathlamet Bay CBK 4/29/14 (1:10) 111.0 34.1 60.3 77.0 24.5 1553.5 8.1 Youngs Bay YBM 3/28/14 (18:55) 47.7 1.69 0.56 13.4 36.3 193.6 6.7 YBB 3/28/14 (19:19) 298.8 73.7 34.0 205.1 16.1 1077.7 24.5 Baker Bay BBB 3/29/14 (7:04) 204.3 12.0 6.01 58.5 73.3 1044.1 23.9 BBI 4/29/14 (7:30) 100.1 3.29 1.39 49.6 87.6 575.5 15.9 BBC 5/29/14 (10:15) 33.9 0.45 0.34 16.7 10.6 109.7 5.7 BBA 8/25/14 (7:50) 11.1 1.50 0.19 16.5 12.9 79.4 1.5

*†P Recovery, Percentage of total sediment P extracted by NaOH-EDTA.*

detected in estuarine sediments at concentrations that varied considerably (1.4–54.1 mg kg−<sup>1</sup> ) over short distances, indicating high local heterogeneity (**Table 3**). For example, the lowest and highest IHP concentrations were observed within Baker Bay (BBB and BBI, respectively while BBA had no detectable IHP). Stereoisomers of IHP (chiro, myo, neo, scyllo) contributed approximately equally to IHP with the exception of BBB, where contributions from myo-IHP were small (**Table 3**).

Orthophosphate diesters made up 0.8–7.4% of extracted P, contributing 0.4–28.7 mg kg−<sup>1</sup> of P (**Table S2**; **Figure 3C**). The highest concentration of total orthophosphate diesters was observed at YBB (28.7 mg kg−<sup>1</sup> ). DNA was detected throughout the estuary, contributing as little as 0.2 mg kg−<sup>1</sup> of P (BBC) and as much as 4.5 mg kg−<sup>1</sup> of P (YBB), which also had the highest concentration of diester degradation products (αand β-glycerophosphate and mononucleotides; **Table 3**). Though present at low concentrations, phosphonates were detected throughout estuarine sediments (0.7–1.4% P), ranging from 0.33 to 2.1 mg kg−<sup>1</sup> of total extracted sediment P (**Figure 3C**).

#### Redundancy Analysis

An RDA was performed to explore possible relationships among P forms and sediment characteristics within lateral bay sediments (**Figure 4**). The biplots of the RDA model display ordination between response variables (P form abundance) and explanatory variables (TotalP, LOI, Fe, Mn, Ca, and Mg; **Figure 4A**), sites (non-quantitative centroids; CBK, YBM, YBB, BBB, BBI, BBC, and BBA) and explanatory variables (**Figure 4B**). The dominant constraining variable (RDA1) was explained by variation in organic matter (LOI; 90.6%) and secondarily (RDA2) by Mg concentrations (Mg; 6.7%), which together explained 97.3% of the variation observed in P forms.

An ordination biplot was generated to look for correlative relationships among sediment characteristics and P form abundances (**Figure 4A**). For example, among the response variables (i.e., P form abundances), phytate (IHP)—and to a lesser degree, total monoesters (Mono)—were associated with low organic matter (LOI) found at BBA, BBC, BBI, and CBK (**Figures 4A,B**). In contrast, diesters (Diester), and to a weaker extent DNA were associated with high concentrations of organic matter, Ca, and Mg (LOI, Ca, Mg; **Figure 4A**), observed at YBM, YBB and BBB. Orthophosphate and phosphonates were weakly associated with high OM (LOI), and total polyphosphates (TotPoly; pyrophosphate and polyphosphates) were not associated with any explanatory variables. Differences in overlying water properties (total P, NOx, DIP, temperature, salinity) did not explain patterns in P form abundance (data not shown).

An ordination plot of sample sites and sediment characteristics split up sites into two groups according to their associative sediment characteristics (**Figure 4B**). For example, sites BBA, BBC, CBK and to a lesser extent BBI all had low organic matter, total sediment P and concentrations of cations (Ca, Mg, Fe, and Mn; **Figure 4B**), compared to sites YBB, YBM and BBB. The BBA site separated from the other sites because of its very low organic matter (LOI) and cation concentrations (**Figure 4B**). Correlation of explanatory variables can be obtained by taking the cosine of angles between vectors;

this analysis revealed that sediment Ca and organic matter (LOI) were highly correlated, with angles ≤15◦ between most vectors (R = 0.97; **Figures 4A,B**).

#### DISCUSSION

There are few studies reporting P forms in estuarine environments, and therefore the small data set presented here provides useful information about the potential for P cycling and retention in a river-dominated estuary. The time and costs associated with P-NMR often limit the number of samples that can be analyzed; in addition, samples are often homogenized and pooled, limiting potential replication. In addition, standardization of methods is being stressed in the literature so that P-NMR results can be compared across environments (Cade-Menun and Liu, 2014). For this reason, extracts were prepared from 10 cm diameter cores, which represents the homogenization of at least 1 year's worth of sediment deposition (based on rates of ∼6 cm year−<sup>1</sup> ; Sherwood and Creager, 1990). In addition, the extraction of organic P forms is notoriously difficult (Cade-Menun et al., 2006), therefore we used an alkaline extraction that optimizes the extraction of organic P forms; however, this tends to produce low extraction efficiencies, and this must be kept in mind when interpreting the data. The recovery rates reported here for NaOH-EDTA extracted P (6–24%) are comparable to other P-NMR studies in wetlands with similar sediment characteristics. For example, P recovery was 6–16% in calcareous soils and 16–30% in sediments with low OM and neutral pH (Cheesman et al., 2014). In contrast, higher recoveries (25–84%) have been reported in wetland sediments that are acidic (pH 3.6–4.8) or that have high OM (56–94%; Turner and Newman, 2005; Cheesman et al., 2014). It is likely that the residual P that was not extracted in our protocol was inorganic, alkali-stable, and associated with Ca or Mg minerals, since these forms are not readily extracted using NaOH-EDTA (Turner and Newman, 2005; Defforey et al., 2017). Nevertheless, despite low recovery of P in NaOH-EDTA extracts, we detected a range of P forms in estuarine sediments.

Polyphosphates and pyrophosphates were detected in all estuarine sediment samples, where they contributed as much as 13% of inorganic P (10% of total extracted P). In almost half the samples, polyphosphate concentrations exceeded pyrophosphate, in contrast to other studies (Turner and Newman, 2005; Zhang et al., 2009; Cheesman et al., 2010). This might be explained by the fact that, unlike the previous work cited, we included a neutralization step prior to lyophilization, which increases the stability of polyphosphate (Cade-Menun et al., 2006). Degradation of polyphosphate to pyrophosphate (either in situ or during sediment extraction) may also account for the fact that only a handful of studies report detection of polyphosphate in wetlands (Zhang et al., 2009; Cheesman et al., 2014).

Polyphosphate is biologically produced across the domains of life, including by bacteria, fungi, protozoa, plants, and mammals (Kornberg, 1995; Brown and Kornberg, 2004; Kulaev et al., 2004) and is thought to be indicative of microbial activity rather than linked to particular soil or sediment types (Cheesman et al., 2014). Interestingly, we detected polyphosphate in sediments with different characteristics, from brackish sediments in Baker Bay where concentrations were highest BBI (low OM, low Fe, and Mn) and BBB (high OM, high Fe, and Mn), and further supported by our RDA which found total polyphosphates not affiliated with measured sediment properties. Further, the presence of diesters (e.g., DNA) as an indicator of microbial activity was associated with high OM, Fe, and Mn concentrations and total P in sediments, suggesting that microbial biomass may be a major contributor to organic matter in river-dominated


*† Inorg:Org, ratio of inorganic to organic P; Mono:Di, ratio of Orthophosphate monoesters to diesters without correction for degradation; cMono, cDi, total orthophosphate monoester and diesters corrected for degradation; Total IHP, sum of myo-, scyllo-, neo- and D-chiro-inositol hexakisphosphate; Myo:other IHP, ratio of sum of myo-inositol hexakisphosphate to all sterioisomers (scyllo-, neo-, D-chiro-inositol hexakisphosphate); Deg, degradation compounds include* α*-glycerophosphate,* β*-glycerophosphate, and mononucleotides; DNA, sum of identified DNA peaks; Total Poly, sum of pyrophosphate and polyphosphate; Pyro:Poly, ratio of pyrophosphate to polyphosphate. Standard error (*±*) in P-NMR analysis for samples from native and tame grasslands and croplands (four replicates across land types) was 8.7% for extracted P at* ≥*10%, compared to 20.1% for extracted P at* <*10% (personal communication B. Cade-Menun, Liu et al., 2018).*

estuary sediments. Polyphosphates found throughout sediments produced by microbial activity may be a source of cycled P in dynamic, diel cycled environments such as estuaries, perhaps much more than has generally been assumed and further investigation is certainly warranted.

Although extracted P was dominated by inorganic forms, organic P accounted for ∼30% of total P and was comprised of P forms that are considered both labile (e.g., RNA, phospholipids) and stable (e.g., IHP), despite rapid flushing and a low contribution of organic P compared to other freshwater wetlands (Turner and Newman, 2005; Turner et al., 2006; Cheesman et al., 2010, 2014) or lake sediments (Carman et al., 2000). The dominant organic P form in these estuarine sediments was the orthophosphate monoester myo-inositol hexakisphosphate (myo-IHP, also called phytate), which is derived from plants, and was detected in all but one sample. Inositol phosphates (myo-IHP, and stereioisomers scyllo-, neo-, and D-chiro-IHP), accrue in terrestrial soils and in aquatic sediments, particularly when clay content is high and pH is low; under these conditions, phytate tends to be the dominant form (Turner et al., 2002). In the riverdominated estuary, phytate was associated with low OM in both fresh and brackish sediments, with the highest concentrations occurring in samples from Baker Bay with relatively high salinity (BBA; >15 PSU). Usually, sedimentary phytate concentrations tend to be lower in the presence of salt and under hypoxic conditions, (Suzumura and Kamatani, 1995a,b) due to the destabilization of phytate complexes (Gardolinski et al., 2004) formed through complexation with iron oxides in acidic soils or by minerals in sediments with high mineral content (Turner et al., 2002; Turner and Newman, 2005; Turner and Weckström, 2009; Cheesman et al., 2014). High variability in phytate concentrations in the higher-salinity samples may indicate a varying effect of salinity and/or sediment complexing agents as Baker Bay has the greatest variation in biogeochemical conditions.

#### Other Influences on P Forms and P Retention

Peripheral bay environments in this study are considered depositional environments, where fine material (mainly consisting of clays and silts) is deposited from suspended fluvial material (Sherwood and Creager, 1990). The composition of sediments and grain sizes varies seasonally in the estuary; however, the peripheral bays are consistently sites of long-term deposition of fine sediments. Differences in P forms found at the different sites within these bays could reflect differences in sediment sorting, and/or diagenetic processes. For example, sites differ in terms of the origin of deposited sediments, with Baker and Cathlamet Bay sediment originating from fluvial washload and Youngs Bay sediment originating from a combination of washload and deposits from ephemeral estuarine turbidity maxima.

In addition to sequestration of organic forms, P retention is influenced by abiotic buffering processes, which limit fluctuations in DIP through mineral precipitation with Fe, Mg, Ca, and Mn (e.g., Fe oxyhydroxides, Ca mineral precipitates; Richardson and Marshall, 1986; Froelich, 1988; Reddy et al., 1999; Bridgham et al., 2001). Retention also occurs through abiotic adsorption to sediment surfaces, particularly on fine particles (Li et al., 2013), which dominate the depositional environment of peripheral bays in the estuary. The relatively low and similar DIP concentrations in waters overlying Columbia River estuary sediments (Gilbert et al., 2013; **Figure S2**)—despite variations in total P within the sediments—may be explained by buffering of P (Froelich, 1988), which warrants further exploration as this would affect P retention in these depositional environments. In addition, the limited sample set suggests that YBB and BBB sites had high concentrations of extractable Ca2+, which readily forms mineral precipitates with P if in high concentrations in sediment environments, (e.g., CaHPO<sup>4</sup> from Ca+<sup>2</sup> and HPO−<sup>2</sup> 4 ) at comparable rates with sediment P

BBI, BBA, BBB), and explanatory variables as arrows (LOI, loss-on-ignition, concentrations of Fe, Mn, Ca, Mg, and TotalP, total sediment P).

sorption (Reddy et al., 1999; Li et al., 2013). Further studies should investigate relationships between P forms and total cation concentrations in different sedimentary environments within the estuary to determine how they might affect P mineralization and sequestration.

# CONCLUSIONS

A variety of P forms were identified in river-dominated estuary sediment extracts, which were dominated by inorganic P. Although orthophosphate made the largest contribution to inorganic P, polyphosphate and pyrophosphate were found throughout the lateral bays, constituting up to 13% of extracted inorganic P. Organic forms accounted for ∼30% of total extracted P, and included both labile P forms (e.g., phospholipids, RNA) and resistant forms (e.g., IHP), which were detected at approximately equal concentrations throughout the estuary. Sample-to-sample variability was greatest among organic P forms, with orthophosphate monoesters such as phytate (myo-IHP) being most abundant when OM content was low. In contrast, orthophosphate diesters were more abundant in sediments with high OM, suggesting that microbial activity might play a larger role in P cycling in sediments with high OM. Low orthophosphate concentrations in overlying waters along with variable sediment P concentrations suggest P buffering may occur within peripheral bays. These data demonstrate that despite short water residence times and rapid flushing, sediments of a river-dominated estuary have the potential to sequester P in inorganic and inorganic forms, thus influencing P loading to the coastal ocean via the large Columbia River plume.

# AUTHOR CONTRIBUTIONS

SW collected and analyzed all samples, wrote the manuscript, and managed communication among all the authors. BC-M ran the NMR experiments and analyzed spectra with SW. JN and SW analyzed water samples and JN contributed to sediment P analysis as well as design of the work. TP had substantial contributions to the conception and design of the work and along with BC-M critically revised the work for intellectual content and contributions.

# FUNDING

This work was supported by the National Science Foundation (NSF OCE-0424602), a Science and Technology Center, the Center for Coastal Margin Observation and Prediction (CMOP). Further, travel grants were provided by the National Science Foundation-sponsored Consortium of Universities for the Advancement of Hydrological Science, Inc. (CUAHSI), and an N.L. Tartar Trust Research Award from Oregon Health and Science University (OHSU) for work in Canada.

# ACKNOWLEDGMENTS

The NMR analysis was done at the Saskatchewan Structural Sciences Centre at the University of Saskatchewan, with support funding from Agriculture and Agri-Food Canada; we thank Drs. K. Brown and P. Zhu for their assistance. Total sediment analysis was done with J. Morse at Portland State University.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00302/full#supplementary-material

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Watson, Needoba and Peterson, and Her Majesty the Queen in Right of Canada, as represented by the Minister of Agriculture and Agri-Food Canada. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Spatial and Temporal Dynamics of Inorganic Phosphate and Adenosine-5′ -Triphosphate in the North Pacific Ocean

Karin M. Björkman<sup>1</sup> \*, Solange Duhamel <sup>2</sup> , Matthew J. Church<sup>3</sup> and David M. Karl <sup>1</sup>

<sup>1</sup> Oceanography, University of Hawaii, Honolulu, HI, United States, <sup>2</sup> Lamont Doherty Earth Observatory, Palisades, NY, United States, <sup>3</sup> Flathead Lake Biological Station, University of Montana, Lake County, IL, United States

#### Edited by:

Javier Arístegui, Universidad de Las Palmas de Gran Canaria, Spain

#### Reviewed by:

Marta Sebastian, Instituto de Oceanografía y Cambio Global, Spain Sarah Elizabeth Reynolds, University of Portsmouth, United Kingdom

> \*Correspondence: Karin M. Björkman

> bjorkman@hawaii.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 24 January 2018 Accepted: 18 June 2018 Published: 10 July 2018

#### Citation:

Björkman KM, Duhamel S, Church MJ and Karl DM (2018) Spatial and Temporal Dynamics of Inorganic Phosphate and Adenosine-5′ -Triphosphate in the North Pacific Ocean. Front. Mar. Sci. 5:235. doi: 10.3389/fmars.2018.00235 Temporal variability in dissolved inorganic, organic phosphate (Pi, DOP) and particulate phosphorus (PPO4) concentrations, and microbial utilization of Pi and dissolved adenosine-5′ -triphosphate (DATP) was studied at Station ALOHA (22.75◦N, 158◦W) in the North Pacific Subtropical Gyre (NPSG) over a multi-year period. Spatial variability of the same properties was investigated along two transects, to and from Hawaii, that traversed the NPSG boundaries to the east (2014) and north (2016). Radiotracer techniques were employed to measure the turnover time of Pi and DATP pools to calculate Pi uptake rates and the Pi hydrolysis rates of DATP. Pi concentrations were more variable, both in time and space, than DOP, ranging two orders of magnitude compared to a factor of two for DOP. The DATP pool, while constituting on average <0.15% of the total DOP-P, was as dynamic as Pi (∼1–200 pmol l−<sup>1</sup> ), with lowest concentrations coinciding with Pi depletion. The Pi turnover times ranged from a few hours to several weeks, and were correlated with measured Pi concentrations (r = 0.9; Station ALOHA, n = 28; 2014, n = 14; 2016, n = 12). Pi uptake rates averaged 3.6 ± 1.3 nmol-P l−<sup>1</sup> d −1 (n = 28: Station ALOHA), 9.2 ± 4.7 nmol-P l−<sup>1</sup> d −1 , (n = 15; 2014) and 5.1 ± 2.5 nmol-P l−<sup>1</sup> d −1 , (n = 12; 2016). The turnover time of the DATP pool was typically substantially shorter (0.4–5 days) than for the Pi-pool, and uptake rates ranged from 1 to 115 pmol l−<sup>1</sup> d −1 . However, at very low Pi and ATP concentrations, ATP turnover was longer than Pi turnover and ATP uptake rates lower. Total ATP hydrolysis was high along both transects, exceeding the ATP taken up by the microbial community, resulting in a net release of Pi into the ambient seawater. This net release was positively correlated to Pi concentration. The relative contribution by microbial size classes to total P-uptake depended on whether P was derived from ambient Pi or from DATP, with the <0.6>0.2µm size class dominating the DATP uptake. Our results indicate that during Pi limiting conditions, regenerated P is rapidly consumed, and that Pi limitation occurs locally and transiently but does not appear to be the predominant condition in the upper water column of the NPSG.

Keywords: phosphorus cycling, North Pacific subtropical gyre, phosphate, ATP, station ALOHA

# INTRODUCTION

Phosphorus (P) is essential for all life and is a key component of nucleic acids, cell membrane lipids and in biological energetic processes via e.g., adenosine-5′ -triphosphate (ATP). Inorganic phosphate (Pi) is frequently at very low concentrations in aquatic environments and although the bioavailability of nitrogen (N) appears to be a proximately limiting resource for primary producers in marine ecosystems, P is believed to be the ultimately limiting macronutrient over geological time scales (Falkowski, 1997; Tyrell, 1999). Within the North Pacific Subtropical Gyre (NPSG) inorganic pools of P and N are generally at much lower concentrations than their respective dissolved organic pools (DOP, DON; Karl et al., 2001b), indicative of their preferential exploitation by the microbial community. The utilization of the DOP pool, at least during times of Pi-stress or limitation, should constitute a reservoir or buffer for P. Several studies have shown that microorganisms in marine environments do utilize the DOP both during P stress and under P-replete conditions in nature (Björkman and Karl, 2003; Mather et al., 2008; Lomas et al., 2010; Duhamel et al., 2011, 2017). However, the DOP pool is chemically diverse and still only partially characterized, but consists predominantly of phosphate esters and a substantial amount of phosphonates (Clark et al., 1998; Kolowith et al., 2001; Karl and Björkman, 2015; Repeta et al., 2016). Among the most commonly studied DOP compounds is ATP. ATP has several advantages as a "model" compound, among them that it is available in different radiolabeled forms. This makes it possible to discern preferential uptake of subcomponents of the molecule, and importantly, it is possible to measure its ambient particulate and dissolved concentrations (Bossard and Karl, 1986; Ammerman and Azam, 1991a; Björkman and Karl, 2001) and to follow its utilization by the microbial community (Casey et al., 2009; Björkman et al., 2012). Although ATP may only be a small portion, and not necessarily representative of the average DOP pool constituents, getting a clearer assessment of its flux through the different P-pools will aid in elucidating the behavior of the bioavailable DOP.

In this study we compare temporal and spatial variabilities in the utilization of Pi and ATP by the surface ocean microbial community, as a function of Pi, DOP and dissolved ATP (DATP) concentrations. Radiotracer techniques were used to determine the turnover times of the Pi or ATP pools respectively, as well as the Pi hydrolysis rates from the DATP pool. The temporal study was conducted at or near Station ALOHA (22.75◦N, 158◦W) in the North Pacific Subtropical Gyre (NPSG) over a multiyear period (2005–2015). The spatial variability in these same properties was investigated during an August-September 2014 zonal transect from California to Hawaii, and an April–May 2016 meridional transect from Hawaii to approximately 36◦N along longitude 158◦W. This is, to our knowledge, the first time that all the above P-pools and accompanying rates of utilization have been measured in concert within the oligotrophic oceans.

#### MATERIALS AND METHODS

#### Station Locations

Sampling for rate measurements was conducted at, or near, Station ALOHA (22.75◦N, 158.00◦W) on several Hawaii Ocean Time-series (HOT), or Center for Microbial Oceanography: Research and Education (C-MORE) expeditions during an 11 year time-period from 2005 to 2015. In the late summer/early fall season of 2014 (18 Aug−16 Sep, 2014), a northeast to southwest (hereafter; zonal) transect cruise was conducted onboard the R/V New Horizon (NH1417: Nitrogen Effects on MicroOrganisms—NEMO) originating in San Diego, CA and terminating in Honolulu, HI. This transect cruise occupied 128 stations with near daily samplings for P concentrations, Putilization experiments and primary productivity (n = 26, 20, and 25 respectively; **Table 1**,**Figure 1**), providing an opportunity to assess spatial variability of the inorganic and organic P dynamics over a larger area to complement the temporal investigations at or around Station ALOHA. In the spring of 2016 (April 19–May 4), a meridional cruise onboard the R/V Kaimikai-O-Kanaloa (KOK 1606: SCOPE—Gradients Cruise I), originating and ending in Honolulu, HI, occupied 14 stations between 23◦N and 38◦N along longitude 158◦W, from within the NPSG into the transition zone toward sub-polar waters. Twelve stations were sampled during the meridional transect and comprised the same suite of measurements as the NH1417 cruise but investigated a different region and season than the earlier zonal transect (**Table 1**,**Figure 1**).

#### Hydrography and Sample Collections

Water samples were collected using polyvinyl chloride (PVC) Niskin <sup>R</sup> -type bottles mounted on a 24-place rosette frame, equipped with conductivity, temperature and pressure (CTD) sensors. In addition to the parameters measured with the environmental sensors, discrete seawater samples were also collected. These included samples for soluble reactive phosphate (hereafter called inorganic phosphate [Pi]), total dissolved phosphorus (TDP), particulate P (PPO4), DATP and chlorophyll a. With the exception of DATP samples, which were prefiltered through a 0.2µm polycarbonate filter, the sample collections and subsequent analyses were performed according to the HOT standard protocols for the temporal data collected at HOT (hahana.soest.hawaii.edu/hot/methods/) with slight modifications during the transect cruises. In brief, for chlorophyll a determinations, seawater (150 ml to 2 l) was filtered through a glass fiber filter (GF/F; Whatman) and the filter extracted in 5 ml of 100% acetone. The samples were extracted for 1– 7 days at −20◦C in the dark prior to fluorometric analysis (Turner Designs; model 10-AU, or TD700). Samples for nutrient concentrations were collected into acid washed, deionized water and sample rinsed, high-density polyethylene bottles, stored upright and frozen (−20◦C) until analyzed (Dore et al., 1996).

#### Sample Analyses

The Pi concentrations were measured using the MAGnesium Induced Co-precipitation technique (MAGIC; Karl and Tien, 1992), followed by standard colorimetric assays (Murphy and Riley, 1962) using a 10 cm cuvette cell (Beckman DU 640 spectrophotometer). The Pi samples were treated to reduce arsenate to arsenite to eliminate cross reactivity with the molybdenum blue complex (Johnson, 1971). The analytical precision of this method is ± 1 nmol l−<sup>1</sup> with a detection limit (DL) of ≤ 3 nmol l−<sup>1</sup> , using the definition of DL = 3 x



Station ALOHA data are the mean values at the 25 m horizon for Aug–Sep and April from 2005 to 2015 (n = 20 and 10, respectively), and Sep 2014 and April 2016.

the standard deviation (s.d.) of the analytical precision. TDP was analyzed by the wet persulfate/high temperature oxidation method (Menzel and Corwin, 1965) followed by MAGIC and the standard Pi colorimetric assay. TDP measurements have an analytical precision of ± 5 nmol l−<sup>1</sup> . DOP was calculated as the difference between TDP and Pi concentrations in paired samples.

experiments were conducted, including two additional stations for <sup>14</sup>C-primary production (red triangles; yellow station numbers). Circled is the area along the 30◦<sup>N</sup> meridian where Pi was depleted during the 2014 cruise. Note that Stations 88, 98, and 100 overlap in the figure, but were separated by 8–18 nmiles; the stations closest in space were separated in time by 3 days. The meridional transect (red triangles; black station numbers) occupied 14 stations of which 12 were sampled for P experiments and <sup>14</sup>C-primary production. (Zonal transect: California Current System [CCS] – Stns 1–4; Transition zone [TZ] – Stns 5–43; North Pacific Subtropical Gyre [NPSG] – Stns 46–127: Meridional transect NPSG – Stns 2–5 and 13–14; TZ – Stns 6–12).

DATP concentrations were determined as described in Björkman and Karl (2001) which is based on the co-precipitation of ATP with brucite, based on the same principal as the MAGIC method used for Pi, allowing DATP to be concentrated from seawater before analysis using firefly bioluminescence. Modifications from the original protocol were made to sample volume and concentration factors as follow; triplicate 50 ml subsamples from each field sample were amended with 250 µl 1 N NaOH, mixed and centrifuged at 1,000 × g for 60 min. The supernatant was carefully aspirated and the pellet dissolved with 50 µl 2.5 N HCl, followed by the addition of 250 µl of deionized water. Prior to analyses the sample was mixed with an equal volume of Tris buffer (pH 7.4, Sigma-Aldrich T7693). Samples for the calibration curve were made in surface seawater with the additions of known amounts of ATP and treated as samples. Blanks were made from surface seawater samples without ATP additions and treated with apyrase to hydrolyze any ATP in the sample (Sigma-Aldrich #A6132; stock 10 units ml−<sup>1</sup> , 10 µl per sample:Tris mix). The DL for this assay is 9 pmol DATP l−1, with the precision of ± 3 pmol l−<sup>1</sup> . Data averaged from multiple observations are presented as the mean ± s.d., (n = x) where x is the number of observations. Station ALOHA core data presented here were obtained from the publicly available HOT data archive; HOT Data Organization and Graphical System (HOT-DOGS: http:// hahana.soest.hawaii.edu/hot/hot-dogs/).

#### Incubation Experiments

Seawater was collected into acid washed, deionized water and sample rinsed, clear polycarbonate (PC) incubation bottles (Nalgene: 75–250 ml). All data presented in this study for Station ALOHA and NH1417 were collected from 25 m, and from 15 m during KOK1606. The samples were spiked with tracer amounts of either <sup>32</sup>P or <sup>33</sup>P orthophosphate (MP Biomedicals # 064014L, Perkin-Elmer #NEZ08000; carrier free) or ATP labeled at the gamma position with either <sup>32</sup>P or <sup>33</sup>P (MP Biomedicals # 01350200, Perkin-Elmer NEG302H00; specific activity 111 TBq mmol−<sup>1</sup> ). The final radioactivity of the samples typically ranged from 0.1 to 1 MBq l−<sup>1</sup> depending on experiment. The bottles were incubated in on-deck incubators, cooled with running, surface seawater and shielded with blue plexiglass (Arkema #2069) to achieve the approximate light level and spectral quality of the depth from which the sample water had been collected. The incubation time varied depending on location and expectancy of the rates, but all incubations were conducted during daytime. Experiments were typically performed as time-course incubations with 3–5 sampling events over a 3–6 h period. A subsample (0.1–0.5 ml) was collected from each incubation bottle to measure the total radioactivity added. Particulate activity was determined by filtering a 5–10 ml aliquot through a 0.2µm pore size polycarbonate membrane filter (Nuclepore). Polycarbonate filters of 0.6 and 2µm pore sizes were also included on occasion at Station ALOHA, and at all stations sampled along the two transects, to assess the relative size class contribution to community Pi or DATP pool turnover. The filters were rinsed with 3 × 2 ml of filtered seawater to remove unincorporated radioactivity, placed into a 7 ml plastic scintillation vial (Simport) and 4 ml of scintillation cocktail (Ultima Gold LLT, Perkin Elmer) added. Radioactivity was measured using a liquid scintillation counter (Perkin-Elmer, LSC 2910TR). Data were corrected for activity loss during the counting process due to decay of the shortlived isotopes. Measurements for primary productivity were performed in conjunction with the P-incubation experiments on both transect cruises. Primary productivity was determined by the <sup>14</sup>C-bicarbonate assay (14C-PP) using 75 ml PC-bottles, spiked with <sup>14</sup>C-bicarbonate (MP Biomedicals 117441H) to a final radioactivity of approximately 7.4 MBq l−<sup>1</sup> .

#### ATP Hydrolysis Rates

During the 2014 and 2016 transect cruises, and at Station ALOHA in 2013, the total hydrolysis rates of DATP were assessed from the <sup>33</sup>P-ATP incubation experiments. The protocol used was modified from Ammerman and Azam (1991b). The filtrate from the 0.2µm filter fraction was collected during the time course incubations. Triplicate 1 ml aliquots from each sample were placed into micro-centrifuge tubes and mixed with 0.2 ml activated charcoal slurry (20 mg charcoal ml−<sup>1</sup> in 0.03 N sulfuric acid: The activated charcoal selectively binds organic molecules, such as DATP, while Pi will remain in solution. This allows for the separation of hydrolyzed <sup>33</sup>P-Pi from <sup>33</sup>P-DATP). The samples were thoroughly mixed by vortex and centrifuged for 15 min at 20,800 × g. A subsample (0.75 ml) of the supernatant was placed into a scintillation vial for radioactivity counting. DATP hydrolysis was measured from the increase of radioactivity over time in the supernatant, indicating that Pi had been cleaved from ATP and released into the ambient seawater, i.e., Pi regenerated from DATP. Total DATP hydrolysis was calculated as the sum of the particulate activity retained on the 0.2µm filters (i.e., P from DATP taken up by microbes, hereafter; PATP) and the regenerated, non-incorporated <sup>33</sup>Pi from <sup>33</sup>P-DATP in the charcoal extractions. All rates were based on time course sampling and the rates determined through linear regression over time.

#### Kinetic Pi Rate Experiment

During the 2014 transect at station 100 (30.32◦N, 150.38◦W), an additional experiment was conducted to assess the kinetic response to increasing concentrations of Pi. Very short Pipool turnover times were measured at this location and were presumed to be due to low ambient concentrations of Pi (this was later confirmed by chemical analysis). Seawater samples were amended with non-radioactive Pi to target additions of 0, 5, 10, 25, 50, 75, 100, and 200 nmol Pi l−<sup>1</sup> . A subsample from each concentration step was placed into PC incubation bottles, spiked with <sup>33</sup>P-Pi, incubated and sampled as described above. A separate set of incubation bottles, amended with 0, 25, 50, and 100 nmol Pi l−<sup>1</sup> , was spiked with <sup>33</sup>P-ATP. The remaining seawater was stored frozen (−20◦C) for later analysis of Pi concentrations. Time-course sampling was conducted for the three lowest P concentrations at 5, 25, 75, and 150 min of spiking with <sup>33</sup>P-Pi. Incubations receiving 25 nmol Pi l−<sup>1</sup> or above, and the <sup>33</sup>P-ATP spiked samples, were sub-sampled in triplicate after approximately 3 h of incubation.

#### Calculations of P Uptake Rates and Kinetic Parameters

The Pi or ATP uptake rates and turnover times in days (TOT, d) were calculated as follows: TOT(d) = t/r where t is the total radioactivity added (Bq l−<sup>1</sup> ) and r is the rate of radiolabel uptake into the particulate fraction (Bq l−<sup>1</sup> d −1 ). The rate was determined from linear regression of the incubation time and radioactivity of the filters from the time course experiments. This calculation assumes that the specific activity of the substrate pool is constant during the incubation period. In our experiments <10% of the radiolabel was taken up during the incubation time, with the exception of the samples from stations 88 (30.33◦N, 150.24◦W) and 100 (30.32◦N, 150.38◦W) along the 2014 transect. At station 88; >80% of the radiolabel was taken up by 80 min and at station 100; ∼40% was captured on the filters after 75 min. At such high proportions of the radiolabel taken up, the calculated uptake rate will be biased as a result of recycling, and may lead to underestimates of the actual rate of uptake. However, the finer time-course resolution and additions of Pi at station 100 allowed for the determination of rates at this station (see above Kinetic Pi rate experiment). In this experiment we also calculated the kinetic parameters Vmax and K<sup>m</sup> for the maximum uptake rate and half saturation constant, respectively, using Hanes-Woolf linear transformation of the data. This transformation uses the substrate concentration (S) and uptake velocity (V) to derive Vmax and K<sup>m</sup> as follows: S/V = (1/Vmax) × S +Vmax/Km. Here S/V is the measured TOT in our incubations, the slope of the linear regression is 1/Vmax, and the y-intercept is Vmax/Km.

The rate of Pi or PATP uptake, expressed as nmol l−<sup>1</sup> d −1 or pmol l−<sup>1</sup> d −1 , was calculated from the turnover time of the respective radioactive tracers and the measured concentration of Pi or DATP of the samples.

#### RESULTS

#### Transect Characterizations

The zonal transect in 2014 traversed the California Current System (CCS) for the first few stations (Stations 1–4), characterized by relatively low sea-surface temperatures (SST), high nutrient concentration, high chlorophyll a and <sup>14</sup>C-PP (**Table 1**, **Figure 2B**). Stations 5–43 comprised the transition zone between the CCS and the NPSG (**Figure 1**). This region showed high eddy activity as detected by satellite estimates of sea surface height anomalies (Shilova et al., 2017), with variable chlorophyll a and declining Pi concentrations. Stations 44–127 were all within the NPSG with low eddy activity and unusually high SST, reaching temperatures above 27◦C nearing the Hawaiian Islands. Time-series data from Station ALOHA have only rarely recorded such high temperatures, and in September 2014 was the first time in a decade to do so. The meridional transect in 2016 reached the transition zone between the NPSG and the subpolar front just south of 32◦N (Station 6). This region demonstrated high spatial variability in a variety of parameters (e.g., chlorophyll a, salinity) likely due to entrainment and filaments of mixing water masses belonging to the oligotrophic gyre and more nutrient-enriched water in the transition zone.

#### Chlorophyll a and <sup>14</sup>C-Primary Productivity

Over the past decade at Station ALOHA chlorophyll a concentration at the 25 m horizon varied by approximately a factor of two with the lowest concentrations observed during the summer months (72 ± 16 ng l−<sup>1</sup> ; n = 28) and the highest during winter (119 ± 31 ng l−<sup>1</sup> ; n = 26) with an average concentration of 93 ± 17 ng l−<sup>1</sup> (**Table 1**). Chlorophyll a concentrations along the 2014 zonal transect were highest within the CCS, but even within the NPSG the concentrations varied over 3 fold (Station 81; 66 ± 3 ng l−<sup>1</sup> , Station 69; 249 ± 3 ng l−<sup>1</sup> : **Table 1**), with the highest chlorophyll a found in a phytoplankton bloom (**Figure 1**,**Table 1**). During the spring 2016 meridional transect, chlorophyll a varied two-fold within the NPSG and at the northernmost stations reached concentrations similar to those observed in the CCS. Station ALOHA <sup>14</sup>C-PP, at the 25 m horizon, ranged from 4.5 to 11.8 µg C l−<sup>1</sup> d −1 , with a mean for August-September of 6.4 ± 0.9 µg C l−<sup>1</sup> d −1 (n = 20) and 7.4 ± 1.8 µg C l−<sup>1</sup> d −1 (n = 10) for April. Along the two transects <sup>14</sup>C-PP varied by approximately a factor of 2–3 within the NPSG, but were much higher in the nutrient-enriched regions of the CCS and the northernmost stations of the 2016 transect (**Table 1**). During the zonal transect, within the NPSG, a peak in <sup>14</sup>C-PP was observed at stations 69 and 76 (∼ 14 µg C l−<sup>1</sup> d −1 ). These two stations were also associated with the highest chlorophyll a measured during that transect within the NPSG (**Table 1**).

#### P Concentrations at Station ALOHA and Along Transects

Variability in Pi concentrations at 25 m at Station ALOHA ranged approximately 25-fold over the past decade (2005–2015; 7–195 nmol P l−<sup>1</sup> ; **Figure 2A**), with an average concentration of 70 ± 39 nmol l−<sup>1</sup> (n = 107), and median of 62 nmol l−<sup>1</sup> . It should be noted that Pi inventories over this period displayed a rapid increase beginning in mid-2012, persisting through the end of 2013 before slowly subsiding into 2014 and 2015 (**Figure 2A**). The upper ocean inventories during this 1.5–year period were significantly higher than in the preceding 7–year period (integrated 0–100 m, [2005–2011, n = 67] 4.8 ± 2.4 mmol m−<sup>2</sup> ; [2012–2013, n = 20] 11.3 ± 3.5 mmol m−<sup>2</sup> ) and

this skewed the mean Pi concentrations at the 25 m horizon upwards for the period by approximately 30% from the longer term mean for Station ALOHA. Pi concentrations during the 2014 zonal transect varied 60-fold driven by the high values found within the CCS, but were still variable by a factor of approximately 20 for stations within the NPSG (**Figure 2B**; Stations 46–127; range 7–141 nmol l−<sup>1</sup> ). The stations occupied within the region along 29.3–30.5◦N and between 140 and 150◦W (Stations 61–102), were characterized by low Pi concentrations (range 7–35 nmol l−<sup>1</sup> , mean 15 ± 1 nmol l−<sup>1</sup> , n = 10), and within the more intensely sampled region around 29.3–30.5◦N, ∼150.3◦W (stations 88–102) the mean Pi concentration was even lower at 9 ± 2 nmol l−<sup>1</sup> (n = 6). During the spring 2016 meridional transect, Pi concentrations within the NPSG were 42 ± 8 nmol l−<sup>1</sup> , and increased though the transition zone into the subpolar front to approximately 500 nmol l−<sup>1</sup> at 37.2◦N (**Figure 2C**). The concentrations of DOP were much less variable both in space and time and ranged about a factor of two with concentrations at Station ALOHA averaging 185 ± 31 nmol P l−<sup>1</sup> (n = 85; range 109–248 nmol P l−<sup>1</sup> ; **Figure 2A**), and 2014 transect stations within the NPSG at 182 ± 34 nmol P l−<sup>1</sup> (n = 19; range 119–232 nmol P l−<sup>1</sup> ; **Figure 2B**). The DOP concentrations were significantly lower (t-test, p < 0.001) in the region encompassing stations 61–102 compared to the other stations within the NPSG. The meridional transect in the spring of 2016 showed a more uniform distribution even into the subpolar regions, ranging from a high of 170 nmol P l−<sup>1</sup> at the southernmost site to a low of 114 nmol P l−<sup>1</sup> toward the subpolar front (**Figure 2C**). The relative contribution of DOP to the total dissolved P (TDP) pool at Station ALOHA varied from 45% to near 100%, with an average of 72 ± 11% (n = 85). Along the zonal transect DOP contributed 41–95% of the total pool, when the CCS stations were included, and 62–95% (mean 84 ± 11%, n = 19) for stations sampled within the NPSG. DOP concentrations were relatively low during the meridional transect, but the DOP contribution to the TDP pool within the NPSG was in the range of the long-term mean for Station ALOHA (73 ± 3%, n = 5). The PPO<sup>4</sup> concentrations varied approximately 3-fold during the 10-year period at Station ALOHA (**Figure 2A**; mean 12 ± 4 nmol l−<sup>1</sup> , n = 106; range 6–25 nmol l−<sup>1</sup> ). The mean PPO<sup>4</sup> during 2011–2015 was 16 ± 3 nmol l−<sup>1</sup> (n = 41). During the zonal transect in 2014, PPO<sup>4</sup> peaked within the CCS (**Figure 2B**), whereas concentrations within the NPSG varied approximately 2-fold (mean 21 ± 4 nmol l −1 , n = 19; range 16–33 nmol l−<sup>1</sup> ). Similarly, the meridional transect in 2016 showed the highest PPO<sup>4</sup> concentrations within the transition zone and toward the subpolar front (**Figure 2C**), while concentrations within the NPSG showed small variations averaging 20 ± 2 nmol l−<sup>1</sup> (n = 5). DATP concentrations along the zonal transect ranged from ∼1 to 180 pmol DATP l −1 , with the lowest concentrations coinciding with the lowest Pi pool and shortest Pi-, and ATP-pool turnover times. The highest concentrations were found within the CCS. However, the DATP concentrations were less dynamic than the range in concentrations imply, and were typically ∼100 pmol l−<sup>1</sup> for the majority of stations sampled (91 ± 15 pmol l−<sup>1</sup> , n = 11; excluding stations where Pi < 15 nmol l−<sup>1</sup> ; **Figure 2B**). The meridional transect DATP concentrations ranged from 33 ± 3 to 209 ± 20 pmol l−<sup>1</sup> with the highest values found within the transition zone. The average DATP concentration within the NPSG for the meridional transect 2016 was 60 ± 27 pmol l−<sup>1</sup> (n = 5: **Figure 2C**).

FIGURE 3 | Turnover time (days) of the Pi-pool and dissolved and particulate ATP-pools (DATP and PATP, respectively) as measured by <sup>33</sup>P radiotracer techniques. The Pi and PATP turnover times are based on the P uptake by the microbial community, whereas the DATP turnover time is the sum of the PATP and the net release of Pi originating from the hydrolysis of DATP. (A) zonal transect in the fall of 2014, and (B) meridional transect in the spring 2016. Error bars are derived from the error estimate of the regression of the time-course sampling used to calculate the turnover time. Values for Station ALOHA are mean ± 1 s.d. (Pi n = 30; ATP n = 7). Arrows indicate the stations within the CCS, TZ, and NPSG as described in Figure 1. Note that the scales differ among panels.

#### Pi Rate Measurements

Pi TOT at Station ALOHA varied from a few days to several weeks over a multi-year study with a long term mean of 18 ± 11 days (n = 28) and an uptake rate of 3.6 ± 1.3 nmol l <sup>−</sup><sup>1</sup> d −1 (range 0.6–7.1 nmol l−<sup>1</sup> d −1 ; median 3.3 nmol l−<sup>1</sup> d −1 ). The range observed for the two transects were from a few hours to 3 weeks within the NPSG, and reached 2–3 months in the CCS and subpolar transition zone respectively (**Figures 3A,B**). Very short turnover times (hours to 3 days) were found along the zonal transect around 30◦N (Stations 61–100). Some of the stations (69–76) in this region were associated with a high phytoplankton biomass and enhanced primary productivity

FIGURE 4 | (A,B) Pi-uptake rates (nmol l−<sup>1</sup> <sup>d</sup> ; columns) and the corresponding Pi concentrations (µmol l−<sup>1</sup> ; bullets) and (C,D) the relative contribution from different microbial size classes to the Pi-uptake measured. The zonal transect is shown in (A,C); the meridional transect in (B,D). Station ALOHA values for Pi uptake rates and Pi concentrations are shown to the left in each panel (mean ± 1 s.d; n = 28). Arrows indicate the stations within the CCS, TZ and NPSG as described in Figure 1. Note that the scales differ among panels.

(chlorophyll a >0.2 µg l −1 , <sup>14</sup>C-PP ∼ 14 µg C l−<sup>1</sup> d −1 ; **Table 1**). The Pi-uptake rate along the zonal transect ranged from a few nmol l−<sup>1</sup> d −1 to approximately 15 nmol−<sup>1</sup> d <sup>−</sup><sup>1</sup> within the NPSG, and reached nearly 30 nmol l−<sup>1</sup> d <sup>−</sup><sup>1</sup> within the CCS (**Figure 4A**). However, at the stations where we recorded the lowest Pi concentrations, and associated short turnover times, uptake rates were much enhanced (**Figure 4A**). The experiment to assess the kinetic response to increasing amounts of Pi was conducted at station 100, where there were indications of very low ambient Pi. Although the turnover time grew longer with additional Pi, as would be expected, the calculated rates for the whole water community did not show a clear kinetic response, as the calculated rates were similar for all concentrations tested (**Table 2**). A linear regression analysis gave an uptake rate estimate of 39 ± 1 nmol l−<sup>1</sup> d −1 (n = 8, r <sup>2</sup> = 0.991) for

the whole water community. However, the >2µm size class did show a kinetic response with increased Pi. The Pi uptake rates in this size class tripled between 10 and 75 nmol l−<sup>1</sup> Pi, and although the relative contribution of the >2µm size class to the whole water community was small, it increased from ∼2 to 8% at concentrations above 75 nmol l−<sup>1</sup> Pi, indicating that cells >2µm were P-limited at ambient concentrations at this station. The kinetic analysis indicated that the K<sup>m</sup> for the >2µm size class was 36 ± 7 nmol−<sup>1</sup> (**Table 2**), which was much above the measured ambient Pi concentration of 5 ± 1 nmol l−<sup>1</sup> . Pi-uptake rates along the meridional transect were comparable to the long term mean of Station ALOHA within the NPSG (mean 2.8 ± 0.9 nmol l−<sup>1</sup> d −1 ; n = 5), with higher rates throughout the transition zone (range 5.5–12 nmol l−<sup>1</sup> d −1 ; **Figure 4B**).


TABLE 2 | Kinetic experiment carried out at Station 100 during the east-west transect in 2014. Seawater was amended with increasing concentrations of inorganic phosphate (Pi) to determine the turnover times (TOT) of the Pi- pool and Pi-uptake rates in different size fractions of the microbial community.

The half-saturation constant (Km) and maximum uptake rate (Vmax ) were calculated from Hanes-Woolf linear transformation of the data. N.d, not determined, due to lack of kinetic response to added Pi.

The relative contribution by different microbial size classes to the total community Pi-uptake, although variable along the two transects, showed a clear dominance by cells <2µm with an average of 88% of Pi taken up by this size class within the NPSG and through the transition zone, but was lower in the CCS where cells >2µm represented ∼ 30% of the total Pi taken up (**Figures 4C,D**). Cells <0.6µm became progressively more important to the total Pi uptake when transitioning from high Pi to low Pi environments. During the zonal transect from the CCS, transition zone to NPSG, cells <0.6µm contributed 43 ± 2, 54 ± 12 to 67 ± 18% respectively to the total Pi uptake. The same pattern, although less pronounced, was seen during the meridional transect with 32 ± 13% at the transition zone and 42 ± 12% within the NPSG. At the stations along 30◦N where Pi concentrations were <15 nmol l−<sup>1</sup> (Stations 61, 81, 88, 100) the contributions from the >2µm size class were at their lowest with <5% of the total. However, at station 69, where both chlorophyll a and <sup>14</sup>C-PP were enhanced, indicative of a phytoplankton bloom, the >2µm size class represented ∼ 45% of the total Pi-uptake.

#### ATP Rate Measurements

The TOT of ATP, based on the rate of uptake by the microbial community, was typically much shorter than that measured for the Pi-pool (**Figures 3A,B**) at 1–2 days in most cases (zonal transect range; 0.44 ± 0.05 to 2.17 ± 0.08 days; mean 1.6 ± 0.6 days, n = 20: meridional range; 1.02 ± 0.06 to 11.3 ± 1.9 days; mean 5.17 ± 3.04 days, n = 12: **Figures 3A,B**). Only at the stations along the zonal transect where Pi concentrations were below 10 nmol l−<sup>1</sup> Pi (Stations 61, 88, 100), was the ATP pool turnover longer than that for Pi (**Figure 3A**). The total DATP pool TOT (i.e., the sum of the ATP taken up by microbes and Pi regenerated from the hydrolysis of DATP and released into the ambient seawater) was less than 1 day on average along the zonal transect (0.95 ± 0.30 days: n = 20). The mean DATP turnover time for the meridional transect was 1.43 ± 0.47 days (n = 12: **Figures 3A,B**).

The total DATP hydrolysis rates ranged nearly two orders of magnitude (3–230 pmol l−<sup>1</sup> d −1 ) along the zonal transect, and approximately 10-fold along the meridional transect (**Figures 5A,B**). The very low rates were again in the region along 30◦N where both Pi and DATP pools were depleted. During the zonal transect, excluding the stations where Pi and ATP were depleted, an average of 28 ± 9% (n = 8) of the total DATP hydrolyzed was not incorporated into cells, but regenerated as Pi within the NPSG, and an even higher proportion was regenerated within the transition zone (53 ± 4%, n = 5) and CCS (64 ± 0%, n = 2). The net Pi regenerated from DATP was even greater during the spring meridional transect with 53 ± 8% (n = 5) and 81 ± 8% (n = 7) within the NPSG and transition-sub polar regions, respectively (**Figure 5B**). The proportion (%) of Pi regenerated from DATP to total DATP hydrolysis correlated with ambient Pi concentrations, with an increasing proportion of the total not taken up with increasing Pi concentrations (**Figure 6**). Below 30 nmol Pi l−<sup>1</sup> the regeneration diminished rapidly until Pi release from DATP was no longer measurable at ∼ 5 nmol Pi l−<sup>1</sup> (**Figure 6**).

The relative contribution by different microbial size classes to the total community ATP-uptake, showed a different distribution than that of Pi. A great majority of P derived from ATP was incorporated into the smallest size class (<0.6>0.2 µm) with 80.1 ± 2.7% and 74.0 ± 2.0% of the total particulate uptake along the zonal and meridional transects respectively (**Figures 5C,D**). The >2µm size class contributed <5% of the total uptake. The exceptions were at the Pi depleted stations (88, 100) and the bloom station (69) during the zonal cruise where the relative uptake by the larger size classes was considerably greater (**Figure 5C**). Furthermore, comparing the TOT of Pi and DATP for the different size classes, in paired experiments, revealed that there was no correlation between the two for the whole water community (>0.2µm; **Figure 7A**), whereas the TOT for both Pi and DATP was similar in the >2µm size class (**Figure 7B**,

partitioned into rate of microbial uptake of P from ATP (gray), and the net release rate of Pi from DATP into the surrounding seawater (white). (C,D) Show the relative contribution from different microbial size classes to the microbial ATP-uptake measured. The zonal transect is shown in (A,C); the meridional transect in (B,D). Station ALOHA data for Pi uptake rates and Pi concentrations are shown to the left in each panel as the mean ± 1 s.d. (n = 7). Arrows indicate the stations within the CCS, TZ and NPSG as described in Figure 1. Note that the scales differ among panels.

r = 0.9). In the Pi addition experiments at Station 100, where ATP had longer TOT than Pi at ambient concentrations of Pi, the TOT did not change significantly with additional Pi in the smallest size class (<0.6>0.2µm) at ∼ 1.3 days. However, at Pi additions of 50 and 100 nmol l−<sup>1</sup> , the TOT for DATP was faster than for Pi. The TOT of DATP in the two larger size classes was more similar to Pi with additional Pi.

# Correlation Between TOT, Pi, and DATP Pool Concentrations

The TOT of Pi correlated well with Pi concentrations within the NPSG (**Figure 8**). Correlation analysis of paired measurements of Pi concentrations and TOT at Station ALOHA had an r = 0.9 (n = 28); the zonal transect 2014 r = 0.9 (n = 13) and the meridional transect r = 0.9 (n = 14; **Figure 8**). However, stations within the NPSG during the meridional transect showed weaker correlation at r = 0.7 (n = 5). The relationship between TOT and Pi concentrations showed a better fit to a power function than a linear regression and it is noteworthy that below 20 nmol l−<sup>1</sup> Pi, the slope of a linear regression is comparatively low, indicating that any additional Pi does not have an appreciable effect on TOT, hence additional Pi leads to an increase in uptake rate. Above ∼50 nmol l−<sup>1</sup> Pi the TOT to Pi relationship stabilizes, i.e., additional Pi does not change uptake rate, so TOT instead increases. The same was not true for the TOT of DATP vs. DATP concentration, which was poorly correlated, in fact, the TOT of the DATP pool was better correlated to the Pi pool size than the DATP pool size (r = 0.7 vs. r = 0.5; data not shown).

#### DISCUSSION

#### Characterization of Study Areas

The majority of the work presented here was conducted within the NPSG, an environment characterized by its oligotrophic nature with perennially low inorganic nutrients, low standing stocks of chlorophyll and biomass, and typically low primary productivity, with Station ALOHA serving as a representative for the biome (Karl and Lukas, 1996; Karl and Church, 2014). This ecosystem is microbial based, recycling intensive, with picophytoplankton (<2µm), specifically the cyanobacterium Prochlorococcus, typically the largest contributors to phytoplankton biomass and primary production (Karl, 1999; Karl et al., 2001a). The two transect cruises

FIGURE 6 | Proportion (%) of the total DATP hydrolysis released into the surrounding seawater as dissolved Pi, versus the ambient inorganic phosphate (Pi) concentration. Zonal transect (black bullets), meridional transect (white bullets) and data from BiG RAPA transect cruise in 2010 within the South Pacific Subtropical Gyre (Duhamel et al., 2017).

(zonal, 2014; meridional, 2016), presented an opportunity to compare the temporal variability in the various P-pools and microbial uptake rates of Pi and DATP observed at Station ALOHA, with spatial variability in the NPSG as well as areas within the transition zones into the CCS to the east, and toward the subpolar waters to the north of the North Pacific Ocean. The transition zones showed strong trends in both salinity and temperature with salinities and temperatures increasing from the boundaries to the east and north into the NPSG, as well as higher inorganic nutrients, chlorophyll and primary productivity compared to those of the NPSG.

# P Concentrations at Station ALOHA and Along Transects

Although our understanding and knowledge of the cycling of P in the oligotrophic oceans have increased over the past decades, it is clear that these ecosystems are highly dynamic and not readily predictable. In particular the chemical composition and microbial utilization of the DOP pool, and its relative bioavailability remains an ongoing field of exploration (Karl and Björkman, 2015). Many studies have examined the bioavailability of DOP either by its susceptibility to alkaline phosphatase treatment (Moutin et al., 2008; Suzumura et al., 2012), through bioassays using known DOP compounds (Berman, 1988; Björkman and Karl, 1994; Björkman et al., 2000; Duhamel et al., 2017), or directly via specific ATP pool labeling (Karl and Bossard, 1985; Bossard and Karl, 1986; Björkman and Karl, 2005). In this study, we show the Pi-pool to be highly variable within the NPSG. Nevertheless, the range in concentrations at Station ALOHA and along the transects was similar. This suggests a mosaic upper ocean, with wide, yet limited dynamic range in terms of Pi-availability both temporally and across regions within the NPSG. In contrast, the DOP pool showed much less variability, and interestingly, the PPO<sup>4</sup> pool was relatively invariant within the NPSG, although always at relatively low

FIGURE 7 | Inorganic phosphate (Pi) turnover time (TOT; days) vs. ATP TOT (days) within the North Pacific Subtropical Gyre (NPSG) for (A) the whole water community (>0.2µm), along the two transects and at Station ALOHA, and (B) the >2µm size class. The dashed line indicates the 1:1 relationship.

concentrations. This may indicate that even at very low Piinventories, the particulate P pool is buffered by increased utilization of the DOP pool. However, a reduction of the DOP pool in response to very low Pi is not evident at Station ALOHA, but may not be readily resolved given the analytical precision of the measurement (i.e., ability to detect a change in DOP of ±5%, or 10–15 nmol l−<sup>1</sup> with sufficient certainty). Yet, during the 2014 zonal transect, in the region around 30◦N where Pi was depleted, the DOP pool was significantly reduced relative to other stations along the transect, as well as compared to the long term mean at Station ALOHA. This lower DOP pool size may reflect an increased utilization of DOP due to Plimiting conditions. Nevertheless, the DOP concentrations in this region remained above 100 nM-P, indicating that a substantial fraction of the DOP pool may not be readily available to the microbial community in the surface ocean. Other studies have reached similar conclusions that only a relatively small fraction of the DOP pool appears to be bioavailable, e.g., Moutin et al. (2008) found that only 10–20% of the in situ DOP pool in the South Pacific Subtropical Gyre was hydrolyzable by alkaline phosphatase, similar to what Suzumura et al. (2012) reported for stations in the North Pacific near Japan (22–39%), and in the tropical and subtropical North Atlantic phosphomonoesters, substrates for alkaline phosphatase, constituted 19–37% of the DOP pool (Reynolds et al., 2014). This is also consistent with the degradation of selected DOP compounds at Station ALOHA, as well as the assessment of DOP utilization though bioassays (Björkman and Karl, 1994, 2003). An alternative explanation for high residual DOP during Pi-deplete conditions may be that other resources, such as available nitrogen or iron, are limiting or co-limiting production, and these limitations hamper further utilization of DOP. For example, in the subtropical North Atlantic, zinc has been implied as a limiting factor for the activity of alkaline phosphatase (Mahaffey et al., 2014), indicating that essential microelements may impact DOP degradation processes. The study region for the zonal transect around 30◦N is known to consistently, if not predictably, harbor large, summertime phytoplankton blooms (Wilson, 2003, 2011; Villareal et al., 2012) and satellite imagery showed enhanced chlorophyll in this same area, consistent with such blooms, lasting from mid-July through September 2014. In a previous study within this same region Duhamel et al. (2010) found areas where the microbial community was under P-stress, as implied by the comparatively high alkaline phosphatase activity measured. The 2014 zonal transect coincided with late bloom stages, and areas most likely in post-bloom conditions as indicated by the vanishingly low concentrations of Pi in the surface waters.

The concentration of the DOP pool constituent DATP was quite uniform along the two transects within the NPSG, except where Pi was depleted and where the DATP pool also was drawn down to near the detection limit for our analysis. The concentrations were higher during the fall than spring transect possibly due to seasonal variability in DATP as previously observed at Station ALOHA (Björkman and Karl, 2001). Although DATP-phosphate is but a small fraction (∼0.15%) of the total DOP pool, it appears to be highly bioavailable.

#### P Turnover Times and Uptake Rates

The Pi pool turnover times observed in this study were within the range previously reported from the NPSG (Perry and Eppley, 1981; Björkman et al., 2000), of a few hours to several weeks, again highlighting the variability in Pi-pool dynamics within the NPSG. The Pi-pool turnover time was strongly correlated with measured Pi pool concentrations, but did show concentration dependent kinetics, i.e., increasing uptake rate with increasing Pi concentration, at low ambient Pi concentrations, as described previously for Station ALOHA (Björkman et al., 2012).

The turnover time of the DATP pool was typically substantially shorter than for the Pi-pool, reflecting both its small pool size and that the microbial hydrolysis and uptake of P from DATP proceeds separately from that of the Pi pool (Ammerman and Azam, 1985; Björkman et al., 2012; Duhamel et al., 2017). Total DATP hydrolysis was high along both transects, exceeding the uptake of Pi from DATP by the microbial community, resulting in a net release of regenerated Pi into the ambient seawater. There was a positive relationship between Pi concentrations and the proportion of the DATP hydrolyzed, i.e., net regeneration of Pi from DATP, similar to that observed by Ammerman and Azam (1991a), with a higher fraction of the total hydrolyzed P taken up at lower Pi concentrations. In the South Pacific Subtropical Gyre, where Pi concentrations are perennially higher compared to the NPSG (Moutin et al., 2008), the proportion of regenerated Pi from ATP was also higher than observed in this study (**Figure 6**) emphasizing the decoupling of DATP hydrolysis and Pi uptake at high Pi concentrations (Duhamel et al., 2017). The steepest increase in the proportion of DATP regenerated as Pi was seen at Pi concentrations <100 nmol l−<sup>1</sup> Pi, indicating that the newly regenerated Pi is of increasing importance to P-nutrition the smaller the Pi pool becomes, and the coupling between DATP hydrolysis and uptake becomes increasingly tighter, i.e., the recycling of P intensifies, as Pi concentrations drop below ∼100 nmol l−<sup>1</sup> . Although the smallest components of the microbial community appear to have the highest affinity for Pi (Thingstad et al., 1993; Cañellas et al., 2000; Vadstein, 2000), and typically do not appear to be P-limited, the larger cells can capitalize at higher Pi concentrations with increased Pi-uptake rates as well as capturing a larger proportion of the total community P taken up (Björkman et al., 2012). This was observed during the two transect cruises, and demonstrated in the Pi addition experiment at station 100 where a range of Pi concentrations was used. In this experiment microorganisms in the >2µm fraction increased both Pi uptake rate as well as the relative contribution of the total uptake with increasing Pi. The uptake of P originating from DATP was greatly dominated by the smallest size class (<0.6>0.2µm) and likely driven by heterotrophic bacterial hydrolysis by the enzyme 5′ -nucleotidase and subsequent uptake (Azam and Hodson, 1977; Bengis-Garber and Kushner, 1982; Ammerman and Azam, 1991a,b). Interestingly, at the Pi depleted stations along 30◦N during the zonal transect, Pi from ATP was transferred to a greater extent into the two larger size classes, although the rates were very low. However, this partitioning was not seen in the Pi-uptake, where the largest size classes contributed the least to overall Pi uptake. Taken together, this indicates that this P depleted region was poised for rapid P assimilation, should P become available.

#### CONCLUSION

To gain a better appreciation for the flux of essential nutrients, such as phosphate, through biogeochemical cycles in the vast open ocean biomes, it is necessary to assess the dynamics of the Ppool and how its bioavailability may impact spatial and temporal variability in productivity and community composition. Our study showed that Pi concentrations in the surface waters within the NPSG were highly dynamic both in time and space whereas

#### REFERENCES


DOP and PPO<sup>4</sup> pools were less variable; the particulate pool potentially buffered by the much larger reservoir provided by the DOP pool. Although Pi-stress, or limitation, occurred locally or periodically, P sufficiency appears to be the prevailing condition for the dominant components of the microbial community within the NPSG. DATP was typically more rapidly turned over than the PATP pool, which turned over faster than the Pipool. Microbial uptake of P was dominated by the <2µm size class; however, the turnover time for Pi or ATP in the >2µm fraction were similar, suggesting that microbial utilization of these two phosphate pools are independent of one another within the NPSG and that DATP predominantly is processed by the smallest microbial components of this ecosystem. DATP hydrolysis commonly resulted in the net release of Pi into the ambient seawater and may increasingly contribute to P-nutrition during low Pi conditions.

#### AUTHOR CONTRIBUTIONS

KB and SD designed the experiments, performed the fieldand laboratory work. KB wrote the manuscript. SD, DK, MC contributed significantly to the intellectual content of the manuscript. DK and MC provided funding.

#### ACKNOWLEDGMENTS

We thank the Captains and crew of the R/V Kilo Moana, R/V Kaimikai-O-Kanaloa, R/V New Horizon and R/V Knorr. We thank Lance Fujieki and Benedetto Barone for graphical assistance. Funding was provided by the National Science Foundation for the Hawaii Ocean Time-series program (OCE-1260164, MC, DK) and the Center for Microbial Oceanography: Research and Education (C-MORE, DBI-0424599, DK) and Dimensions in Biodiversity (OCE-124221, MC). Additional support was provided by the Gordon and Betty Moore Foundation: Marine Microbiology Initiative (3794, DK) and the Simons Foundation (SCOPE project ID 329108: DK and MC and Gradients project ID 426570SP: V. Armbrust; U. Washington, with subcontract to DK).


(ATP) in the North Pacific Subtropical Gyre. Front. Microbiol. 3:189. doi: 10.3389/fmicb.2012.00189


role for dissolved organic phosphorus (DOP). Biogeosciences 7, 695–710. doi: 10.5194/bg-7-695-2010


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Björkman, Duhamel, Church and Karl. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dissolved Organic Phosphorus Utilization by Phytoplankton Reveals Preferential Degradation of Polyphosphates Over Phosphomonoesters

Julia M. Diaz<sup>1</sup> \*, Alisia Holland<sup>1</sup> , James G. Sanders<sup>1</sup> , Karrie Bulski<sup>1</sup> , Douglas Mollett1,2 , Chau-Wen Chou<sup>3</sup> , Dennis Phillips<sup>3</sup> , Yuanzhi Tang<sup>4</sup> and Solange Duhamel<sup>5</sup>

#### Edited by:

Wolfgang Koeve, GEOMAR Helmholtz-Zentrum für Ozeanforschung Kiel, Helmholtz-Gemeinschaft Deutscher Forschungszentren (HZ), Germany

#### Reviewed by:

Michael William Lomas, Bigelow Laboratory for Ocean Sciences, United States Monika Nausch, Leibniz Institute for Baltic Sea Research (LG), Germany

> \*Correspondence: Julia M. Diaz julia.diaz@skio.uga.edu

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 13 May 2018 Accepted: 28 September 2018 Published: 25 October 2018

#### Citation:

Diaz JM, Holland A, Sanders JG, Bulski K, Mollett D, Chou C-W, Phillips D, Tang Y and Duhamel S (2018) Dissolved Organic Phosphorus Utilization by Phytoplankton Reveals Preferential Degradation of Polyphosphates Over Phosphomonoesters. Front. Mar. Sci. 5:380. doi: 10.3389/fmars.2018.00380 <sup>1</sup> Department of Marine Sciences, Skidaway Institute of Oceanography, University of Georgia, Savannah, GA, United States, <sup>2</sup> Georgetown College, Georgetown, KY, United States, <sup>3</sup> Proteomics and Mass Spectrometry Core Facility, University of Georgia, Athens, GA, United States, <sup>4</sup> School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, GA, United States, <sup>5</sup> Division of Biology and Paleo Environment, Lamont-Doherty Earth Observatory, Palisades, NY, United States

The nutritionally available pool of dissolved organic phosphorus (DOP) supports marine primary productivity in a range of ocean ecosystems but remains poorly resolved. Here, the relative lability of model phosphorus (P) compounds representing the major P(V) bond classes of marine DOP – phosphomonoesters (P-O-C) and phosphoanhydrides (P-O-P) – was assessed in diatom cultures of the genus Thalassiosira, as well as coastal field sites of the western North Atlantic. In diatom samples, maximum enzymatic hydrolysis rates revealed that the P-anhydride bonds of inorganic tripolyphosphate (3poly-P), followed by the P-anhydride bonds of adenosine 5<sup>0</sup> -triphosphate (ATP), were preferentially degraded relative to the P-monoesters adenosine 5<sup>0</sup> -monophosphate (AMP) and 4-methylumbelliferone phosphate (MUF-P). Consistent with these rate measurements, targeted proteomics analysis demonstrated that the underlying phosphatase diversity present in diatom samples was dominated by P-anhydride degrading enzymes (inorganic pyrophosphatases and nucleoside triphosphatases). Furthermore, biomass-normalized rates of ATP degradation were always suppressed under P-replete conditions in diatom cultures, but the effect of overall P availability on 3poly-P degradation was inconsistent among diatom strains, suggesting that inorganic polyphosphate (poly-P) degradation may persist irrespective of prevailing P levels in the marine environment. Indeed, the majority of field sites examined in the P-replete coastal western North Atlantic exhibited significantly higher maximum rates of inorganic poly-P hydrolysis relative to P-monoester hydrolysis, which was largely driven by phytoplankton dynamics. Based on these results, the possibility that P-anhydride utilization may contribute comparably or even more substantially than P-esters to community-level P demand, phytoplankton growth, and primary productivity should be considered.

Keywords: dissolved organic phosphorus, alkaline phosphatase activity, polyphosphate, phosphoester, phosphoanhydride, phosphorus stress, Thalassiosira, diatom

# INTRODUCTION

fmars-05-00380 October 23, 2018 Time: 14:30 # 2

Marine primary productivity is fundamentally constrained by the availability of nutrients such as phosphorus (P). P is incorporated into vital biomolecules that play essential roles in cellular structure (phospholipids), the storage and transmission of genetic information (DNA and RNA), energy transduction (adenosine 5<sup>0</sup> -triphosphate; ATP), metabolic signaling (inositol trisphosphate, (de)phosphorylated proteins and metabolites), as well as stress response and homeostasis (inorganic polyphosphate; poly-P) (Kornberg et al., 1999; Gray and Jakob, 2015). Although marine primary production is principally limited by either nitrogen or iron over most of the global surface ocean (Moore et al., 2013), P can be present at low or limiting concentrations in many marine systems, from the vast oligotrophic gyres (Rivkin and Swift, 1979; Wu et al., 2000; Moore et al., 2008; Moore et al., 2013) to coastal seas (Berland et al., 1972; Fahmy, 2003; Thingstad et al., 2005; Hardison et al., 2013). Rising anthropogenic inputs of nitrogen have also led to an increasing prevalence of P limitation in many ocean environments, especially in near-shore zones (Harrison et al., 1990; Fisher et al., 1992, 1999; Huang et al., 2003; Kemp et al., 2005; Zhang et al., 2007; Xu et al., 2008; Fu et al., 2012; Laurent et al., 2012; Turner and Rabalais, 2013; Kim et al., 2014). Yet even when P is not limiting, it can be sufficiently scarce to shape microbial physiology and community composition (Moore et al., 2008).

Marine plankton communities can access a variety of P sources to satisfy nutritional P demands. Orthophosphate is considered the most biologically preferred P source, but in the open ocean where orthophosphate is scarce, i.e., <1–10% of total dissolved P (Wu et al., 2000; Lomas et al., 2010), the recycling of dissolved organic P (DOP) supports approximately 90% of gross primary productivity (Karl, 2014). Even in coastal systems that are replete with orthophosphate, a highly labile fraction of the DOP pool is rapidly recycled to support primary production (Benitez-Nelson and Buesseler, 1999). The cycling and biological utilization of DOP is therefore fundamentally tied to the productivity and function of marine ecosystems on a global scale (Letscher and Moore, 2015) across extremes in trophic status.

Natural marine DOP is an operationally defined assemblage of organic, inorganic, and polymeric P-containing molecules. Despite a high degree of heterogeneity at the molecular level, marine DOP contains consistent proportions of three major P-bond classes: P-esters, P-anhydrides, and phosphonates (Young and Ingall, 2010). P-esters contain P in its most common + V oxidation state, typically in the form of the P-O-C (monoester) or C-O-P-O-C (diester) bonds. Representative P-monoesters include nucleotides and phosphosugars, while P-diesters include phospholipids and nucleic acids. Overall, P-esters are the most abundant DOP compound class, representing ∼80% of total marine DOP (Young and Ingall, 2010). Another P(V) bond class is the P-anhydrides, which are characterized by P-O-P bonds like those present in nucleoside di- and triphosphates, inorganic pyrophosphate, and poly-P. P-anhydrides account for ∼10% of total marine DOP and are the most recently recognized DOP bond class, based on advances in techniques to recover the low molecular weight fraction of dissolved organic matter (Diaz et al., 2008; Young and Ingall, 2010). Finally, phosphonates make up ∼10% of DOP (Young and Ingall, 2010). In phosphonates, P is present in its + III oxidation state bound directly to carbon. The identities of natural marine phosphonates are not fully known but may include phosphonolipids and low molecular weight metabolites (Repeta et al., 2016).

The composition of DOP helps determine its bioavailability. For example, while phosphonate utilization may be restricted to certain prokaryotic taxa (Dyhrman et al., 2006; Ilikchyan et al., 2009; Beversdorf et al., 2010; Martinez et al., 2010), P-esters are generally recognized as the major nutritional source of DOP in the ocean (Lin et al., 2016). The widespread importance of P-esters is based on the quantitative dominance of this pool (Young and Ingall, 2010), as well as the environmental prevalence of alkaline phosphatase (AP) (Luo et al., 2009, 2011; Sebastian and Ammerman, 2009). APs are well-known P-esterases, which are active throughout the ocean (Hoppe and Ullrich, 1999; Hoppe, 2003; Mahaffey et al., 2014) and play a vital role in phytoplankton P nutritional physiology, especially in P-deplete environments (Dyhrman and Ruttenberg, 2006; Lin et al., 2016). P-anhydrides may also be a critical, yet under-characterized source of P. For instance, nucleoside triphosphates such as ATP are highly labile P sources (Moore et al., 2005; Alonso-Saez and Gasol, 2007; Michelou et al., 2011; Mazard et al., 2012; Duhamel et al., 2014) in which two-thirds of the available P is contained in P-anhydride bonds. At Station ALOHA in the oligotrophic North Pacific Subtropical Gyre (NPSG), the turnover of dissolved ATP exceeds that of the bulk DOP pool, indicating that it may be preferentially utilized over other DOP sources (Björkman and Karl, 2005). Similarly, depth profiles of inorganic poly-P in the Sargasso Sea suggest that this P-anhydride is preferentially utilized over other forms of P (Martin et al., 2014). Indeed, inorganic poly-P is bioavailable to a diversity of microorganisms, including Prochlorococcus, Synechococcus (Moore et al., 2005), eukaryotic marine phytoplankton (Diaz et al., 2016), and natural phytoplankton communities (Björkman and Karl, 1994).

Despite the widespread importance of DOP and the strong compositional control over its utilization, the relative contribution of specific DOP pools to microbial P demand and primary productivity remains unclear at the community level. Therefore, an improved understanding of compound-specific DOP dynamics is necessary to advance a mechanistic view of overall marine ecosystem productivity and functioning. In particular, a growing appreciation of inorganic poly-P cycling (Diaz et al., 2008; Martin et al., 2014; Saad et al., 2016) brings into question the importance of this P-anhydride source relative to more traditionally recognized P-esters. Therefore, in this study, the relative lability of model P-esters and P-anhydrides was assessed in representative diatom cultures of the genus Thalassiosira, as well as natural samples from the coastal western North Atlantic. To identify underlying mechanisms involved in the compound-specific utilization of DOP, targeted proteomic analysis of diatom culture samples was also conducted.

#### MATERIALS AND METHODS

fmars-05-00380 October 23, 2018 Time: 14:30 # 3

#### Model Diatoms, Growth Conditions, and Culture Sampling

Thalassiosira oceanica CCMP1005, Thalassiosira pseudonana CCMP1335, and Thalassiosira pseudonana CCMP1014 were obtained from the National Center for Marine Algae and Microbiota (NCMA), Bigelow Laboratories, East Boothbay, Maine. Diatoms were grown in batch cultures on autoclaved (121◦C, 20 min) f/2 media (Guillard and Ryther, 1962) prepared using filtered (0.2 µm) natural seawater from the South Atlantic Bight. To examine the effect of overall P availability on the degradation of model DOP sources, diatoms were grown under P-replete (+P) and P-deplete (−P) conditions. In +P media, phosphate was provided at an initial concentration of 36 µM. In −P cultures, no phosphorus source was added. Molar N:P ratios were ∼24 and >882 in +P and −P media, respectively. T. oceanica and T. pseudonana were cultivated at 23◦C and 18◦C, respectively, on a 14 h:10 h light:dark cycle (340 µmol photons m−<sup>2</sup> s −1 ). Phytoplankton growth was monitored daily by measuring in vivo chlorophyll fluorescence with an AquaFluor handheld fluorometer (Turner Designs). Cell counts were conducted using a hemocytometer counting chamber (Karlson et al., 2010).

P hydrolysis rates from a variety of model DOP substrates (see below) were investigated in diatom cultures with two types of experiments. In the first experiment type, the degradation of P sources was monitored in +P and −P cultures of each diatom strain by subsampling the cultures every 3 days across 30 days of cultivation. On a given sampling day, each subsample (6 mL) was directly filtered (0.2 µm, 33 mm) to produce the cell-free filtrate, which was then incubated with a variety of P sources for up to 24 h to determine substrate-specific P hydrolysis rates for that day, as detailed below. In the second type of P hydrolysis experiment, +P cultures of each diatom strain were sampled after 30 days of cultivation, and P hydrolysis rates for each substrate were determined, as detailed below, in whole cultures and cell-free filtrates.

To generate samples for exoproteome analysis, ∼140 mL of cell-free filtrate was sampled from +P cultures after 30 days of growth. These filtrate samples were concentrated and exchanged twice into 20 mM Tris (pH = 8.0) using a 10 kDa Centricon <sup>R</sup> Plus-70 centrifugal filter device (Millipore), to a final volume 0.5 mL.

#### Field Sampling

Surface seawater (5–35 m) was collected in September 2016 during two sampling campaigns in the coastal western North Atlantic (**Supplementary Table S1**). Three sites were sampled aboard the R/V Endeavor using a Niskin rosette sampler and incubated immediately in order to determine rates of P hydrolysis. Two sites accessible by small boat in Woods Hole Harbor and Buzzard's Bay, MA, were sampled utilizing a peristaltic pump. These samples were transported on ice packs and analyzed for P hydrolysis rates within 5–6 h of collection. Additional samples were preserved and analyzed for chlorophyll, bacteria and phytoplankton abundance, and soluble reactive P (SRP), as detailed below.

#### Chlorophyll

In the dark, 250 mL of seawater was filtered onto 25 mm GF/F filters. Samples were stored in the dark at −20◦C until analyzed according to protocols adapted from Strickland and Parsons (1972). Briefly, samples were extracted in 90% acetone in the dark (4◦C, 9 h) and measured using a 10AU fluorometer (Turner). Sample signals were calibrated using a chlorophyll-a standard (Sigma) and were corrected for pheopigments by accounting for the fluorescence of extracts before and after acidification to 0.003 M HCl.

#### Abundance of Bacteria and Phytoplankton

Seawater samples were preserved for flow cytometry with 0.5% glutaraldehyde (final concentration), flash frozen in liquid nitrogen and stored at −80◦C until analysis. Bacteria and group-specific phytoplankton counts were conducted on a Guava EasyCyte HT flow cytometer (Millipore). Instrument-specific beads were used to calibrate the cytometer. Samples were analyzed at a low flow rate (0.24 µL s−<sup>1</sup> ) for 3 min. To enumerate bacteria, samples were diluted (1:100) with filtered seawater (0.01 µm). Samples and filtered seawater blanks were stained with SYBR Green I (Invitrogen) according to the manufacturer's instructions and incubated in a 96-well plate in the dark at room temperature for 1 h. Bacterial cells were counted based on diagnostic forward scatter vs. green fluorescence signals. Major phytoplankton groups were distinguished based on plots of forward scatter vs. orange fluorescence (phycoerythrin-containing Synechococcus sp.), and forward scatter vs. red fluorescence (eukaryotes). Size classes of eukaryotic phytoplankton were further distinguished based on forward scatter (pico-, nano- and large eukaryotes).

#### Soluble Reactive P

Seawater samples were collected into acid cleaned, high density polyethylene bottles. Samples used for determining in situ SRP concentrations were frozen and stored upright at −20◦C until analysis. Field samples and diatom filtrates were both analyzed for SRP using a standard colorimetric method (Hansen and Koroleff, 1999). To determine in situ SRP concentrations in field samples, SRP analysis was conducted using a 4 cm glass spectrophotometry cell on triplicate subsamples, and the detection limit, defined as three times the standard deviation of replicate blank measurements, was 115 nmol L−<sup>1</sup> SRP. For incubations to determine P hydrolysis rates (see below), replicate samples were analyzed in clear 96-well plates on a multimode plate reader (Molecular Devices) with a detection limit of 800 nmol L−<sup>1</sup> P.

#### P-hydrolysis of Model DOP Substrates

Field and culture samples were incubated with a variety of model DOP substrates to determine P hydrolysis rates. Model P substrates were selected to represent the major P(V) bond classes present in natural marine DOP (**Figure 1**). The fluorogenic probe 4-methylumbeliferone phosphate (MUF-P) and the nucleotide adenosine 5<sup>0</sup> -monophosphate (AMP) were utilized as model P-esters. ATP was used as both a representative P-ester and an organic P-anhydride. Finally, two inorganic polyphosphate

bond classes of natural marine DOP. (A) Adenosine monophosphate (AMP), a P-monoester. (B) Adenosine triphosphate (ATP), a P-monoester and P-anhydride. (C) Inorganic polyphosphate (poly-P), a P-anhydride, n ≥ 1. (D) 4-methylumbelliferyl phosphate (MUF-P), a fluorogenically labeled P-monoester.

compounds with an average chain length of 3 or 45 P atoms (3poly-P and 45poly-P, respectively), were utilized as representative inorganic P-anhydrides. All P compounds were obtained from Millipore Sigma.

P hydrolysis rates were determined in whole cultures (MUF-P only), cell-free culture filtrates (MUF-P, AMP, ATP, and 3poly-P), and whole seawater samples (MUF-P, 3poly-P, and 45poly-P). Samples were amended with each substrate at a final concentration of 20 µM P. This concentration was assumed to be rate-saturating based on preliminary experiments with natural samples. Thus, the rates of P hydrolysis reported herein represent maximum hydrolysis rates, consistent with many previous investigations of P-monoester degradation (or alkaline phosphatase activity, APA) in the marine environment [see Mahaffey et al. (2014) and references therein]. P hydrolysis rates were determined through one of the following two methods. In the first method, which was applied to all P substrates, the production of SRP was monitored over time using the colorimetric protocol outlined above. In the second method, the time-dependent hydrolysis of the fluorogenic probe MUF-P was monitored using a standard fluorescence technique (Duhamel et al., 2011). Briefly, hydrolysis of MUF-P to 4-methylumbellierone (MUF) was measured (excitation: 359 nm, emission: 449 nm) and calibrated with a multi-point standard curve of MUF (10–500 nmol L−<sup>1</sup> ). In both methods, samples were corrected for substrate autohydrolysis by accounting for negative controls, which were filtered (0.2 µm) and boiled (99◦C, 15 min) prior to P amendment in order to eliminate enzyme activity.

In culture and field experiments, replicate samples and controls were incubated at room temperature (∼25◦C) in a transparent (SRP method) or black (fluorescence method) 96-well plate. To ensure linearity of the hydrolysis rates, samples and controls were measured at multiple time points during the incubation period on a multimode plate reader (Molecular Devices). Overall incubation times were optimized depending on the sample type, P substrate, and the detection method employed. The SRP and fluorescence techniques were characterized by substantial differences in sensitivity. For example, the lower limit of detection, defined as three times the standard deviation of replicate blank measurements, was 800 nmol P L−<sup>1</sup> for the SRP method and 2 nmol P L−<sup>1</sup> for the fluorescence method. The upper limits of detection for the fluorescence method, defined as the P concentration above which the calibration curves begin to lose linearity, was 500 nmol L−<sup>1</sup> P. Thus, samples exhibiting MUF-P hydrolysis rates within the limits of detection of the fluorescence method had to be incubated at least 1.6 times and up to 400 times longer before reaching detectable levels of SRP. The actual incubation times for the various sample types and analytical methods were as follows: whole seawater samples (fluorescence method: 12–15 h, SRP method: 24–65 h), cell-free culture filtrates (fluorescence method: 1–2 h, SRP method: 30 min to 24 h), whole culture samples (fluorescence method: 5 min to 1 h).

The SRP method was not used with whole culture samples due to interferences associated with the uptake/adsorption of enzymatically released phosphate by the cultures, which was problematic at high cell densities. To verify the lack of such interference in whole seawater samples, SRP calibration curves (2–20 µM P) were prepared in each sample and analyzed at multiple time points throughout the incubation period.

#### Genome Searches

fmars-05-00380 October 23, 2018 Time: 14:30 # 5

Putative APs were identified in the translated genomes of T. pseudonana CCMP1335 (taxid: 296543) and T. oceanica CCMP1005 (taxid: 159749) with NCBI's BLASTP search tool. The following protein sequences were used as queries: PhoA from Escherichia coli (GenBank: AAA83893), PhoD from Bacillus subtilis (UniProt: P42251), and PhoX from Pseudomonas fluorescens (Protein Data Bank: 4A9V). Only hits with an E-value < 1 × 10−<sup>4</sup> are reported.

#### Proteomics

Concentrated exoproteome samples were digested using an insolution tryptic digestion kit (ThermoScientific), according to the manufacturer's instructions. Peptide samples were analyzed at the Proteomics and Mass Spectrometry (PAMS) facility at the University of Georgia on a Thermo-Fisher LTQ Orbitrap Elite mass spectrometer coupled with a Proxeon Easy NanoLC system (Waltham, MA, United States).

The enzymatic peptides were loaded into a reversed-phase column (self-packed column/emitter with 200 Å 5 µM Bruker MagicAQ C18 resin), then directly eluted into the mass spectrometer. Briefly, the two-buffer gradient elution (0.1% formic acid as buffer A and 99.9% acetonitrile with 0.1% formic acid as buffer B) starts with 5% B, holds at 5% B for 2 min, then increases to 25% B in 60 min, to 40% B in 10 min, and to 95% B in 10 min.

The data-dependent acquisition method was used to acquire MS data. A survey MS scan was acquired first, and then the top 5 ions in the MS scan were selected following collision-induced dissociation (CID) and higher-energy collisional dissociation (HCD) MS/MS analysis. Both MS and MS/MS scans were acquired by the Orbitrap mass spectrometer at resolutions of 120,000 and 30,000, respectively. Data were acquired using Xcalibur software (version 2.2, Thermo Fisher Scientific).

Protein identification was performed using Thermo Proteome Discoverer (version 1.4) with Mascot (Matrix Science). The reference database for T. oceanica CCMP1005 consisted of the translated whole genome (NCBI Bioproject PRJNA36595) (Lommer et al., 2012) amended with a list of common contaminants, such as human keratin. The T. pseudonana CCMP1335 database was compiled from the translated nuclear (Bioproject PRJNA191), chloroplast (Bioproject PRJNA20561), and mitochondrial (Bioproject PRJNA15818) genomes (Armbrust et al., 2004; Oudot-Le Secq et al., 2007), and was also amended with common contaminant sequences. Databases included a reversed 'decoy' version for false discovery rate (FDR) analysis. The FDR of identified peptides was ∼2%. Protein matches were also filtered to exclude hits with fewer than two unique peptide identifications. Gene ontology (GO) annotations were integrated into results using the ProteinCenter Annotation node.

# Statistical Analysis

P hydrolysis rates in diatom cell-free filtrates were measured throughout the growth curve of each species cultivated in replete (+P) or P-deficient (−P) media. Differences in hydrolysis rates among the three P sources under each growth condition were assessed using repeated measures analysis of variance (ANOVA) and post hoc testing with Tukey's honest significant difference (HSD) method. Average ratios of ATP and 3polyP hydrolysis were compared using a one-sample T-test. To test whether the degradation of each P source was P-regulated, cell-normalized rates of P hydrolysis were compared in +P and −P cultures using repeated measured ANOVA. Average P hydrolysis rates measured in whole cultures and in cell-free filtrates with different quantification methods were compared using Tukey's HSD test. MUF-P hydrolysis rates determined by the SRP vs. fluorescence methods were assessed using simple linear regression. Field-based rates of polyphosphate and MUF-P hydrolysis were compared at each site using an independent two-sample T-test assuming equal variance. Potential relationships between P hydrolysis rates and in situ chlorophyll concentrations were explored with simple linear regression. All statistical analyses were performed in Microsoft Excel <sup>R</sup> or JMP <sup>R</sup> Pro (V13.0).

# RESULTS

### Degradation of Model P Sources by Diatom Cultures

#### Relative Lability of P Sources in Cell-Free Filtrates

In order to explore the DOP nutritional preferences of Thalassiosira spp., the labilities of model P sources representing the major P(V) DOP bond classes (**Figure 1**) were tested in whole cultures and cell-free filtrates grown in P-replete (+P) and P-deficient (−P) media. Because P hydrolysis rates were quantified based on the production of SRP, cell-free filtrates were primarily utilized instead of whole cultures in order to avoid technical challenges associated with the analysis of SRP at high cell densities (see Discussion). All model P sources were degraded in diatom filtrates (**Figures 2**–**4**) and showed negligible hydrolysis (below detection) in boiled controls, consistent with enzymatic P-hydrolysis. In −P cultures, P hydrolysis rates were substantial within 1 week (T. oceanica and T. pseudonana CCMP1014) (**Figures 2B,F**) or after 2 weeks (T. pseudonana CCMP1335) (**Figure 2D**). In +P cultures, DOP degradation increased dramatically after about ∼2 weeks (**Figure 2**). Hydrolysis rates increased in all cultures over time, consistent with the depletion of SRP and the eventual onset of stationary phase (**Figures 2A,C,E**). Degradation rates among the three P sources were significantly different in each diatom culture and media condition throughout the growth curves (**Figure 2**). In every case, regardless of the diatom strain and the initial level of P availability, the rate of inorganic 3poly-P hydrolysis was significantly highest, followed by ATP, and finally AMP (**Figure 2**). Over the entire growth curve, ATP and AMP hydrolysis rates were statistically similar for both strains

of T. pseudonana (**Figures 2C–E**), except for T. pseudonana CCMP1014 cultivated in −P media (**Figure 2F**). On the other hand, T. oceanica always hydrolyzed ATP significantly more than AMP, regardless of prevailing P availability (**Figures 2A,B**).

A maximum of three orthophosphate molecules can be released from ATP and 3poly-P. However, the complete degradation of phosphoanhydride bonds would produce three orthophosphate groups from 3poly-P but only two from ATP, since the final orthophosphate is bound to the adenosine moiety by a P-monoester linkage (**Figure 1**). Therefore, if the P-anhydride bonds of ATP are degraded, and the P-monoester bond remains intact, the ratio of ATP:3polyP hydrolysis should be 2/3 (or ∼66.7%). However, ratios of ATP:3poly-P hydrolysis over the last 1–2 weeks of growth were significantly less than 2/3 (**Table 1** and **Figures 2A,C–F**), except in the case of T. oceanica (−P) (**Table 1** and **Figure 2B**), in which ATP degradation was statistically similar to 66.7% of 3poly-P degradation.

#### The Effect of Overall P Availability on Substrate-Specific DOP Hydrolysis in Cell-Free Filtrates

The lability of 3poly-P, ATP, and AMP was assessed in the cell-free filtrates of diatoms cultivated in P-replete and P-deficient media. To account for differences in biomass, P hydrolysis rates were normalized to total cell counts in cultures at the respective sampling time. ATP hydrolysis was significantly upregulated under −P conditions in all diatom cultures (**Figures 3G–I**). On the other hand, P availability did not significantly affect

cell-free filtrates of T. oceanica CCMP1005 (A,D,G,J), T. pseudonana CCMP1335 (B,E,H,K), and T. pseudonana CCMP1014 (C,F,I,L) cultivated under P-replete (+P) and P-deficient (–P) conditions. Error bars represent one standard deviation of the mean of three biological replicates. Statistical results from repeated measures ANOVA are provided above each plot and indicate the overall effect of P availability on hydrolysis rates.

AMP hydrolysis in any culture (**Figures 3J–L**). P-deficient conditions induced significant upregulation of 3poly-P hydrolysis by T. oceanica CCMP1005, significant downregulation by T. pseudonana CCMP1014, and no change by T. pseudonana CCMP1335 (**Figures 3D–F**).

#### Comparison to Fluorescence-Based MUF-P Hydrolysis Measurements

For all diatoms, 3poly-P hydrolysis in cell-free filtrates was significantly higher than MUF-P hydrolysis in whole cultures (**Figure 4**). ATP degradation by T. oceanica CCMP1005 cell-free filtrates was also significantly higher than whole culture MUF-P hydrolysis (**Figure 4A**). Rates of P-ester hydrolysis in the cell-free filtrates of all diatoms reflected strong methodological agreement. For example, AMP and MUF-P (fluorescence and SRP methods) were degraded at similar rates (**Figure 4**). In fact, SRP-based estimates of MUF-P hydrolysis systematically underestimated fluorescence-based measurements of MUF-P degradation by about ∼17% (p = 0.0008) in culture filtrates (**Figure 5**).

FIGURE 4 | P hydrolysis rates of model DOP compounds in the presence (whole culture) and absence (filtrate) of cells and using different detection methods. (A) T. oceanica CCMP1005, (B) T. pseudonana CCMP1335, and (C) T. pseudonana CCMP1014. All rates were quantified by measuring the accumulation of phosphate over time (SRP method), except where MUF-P degradation was quantified via the standard fluorometric method (Fl.). Cultures were analyzed after 30 days of growth. Error bars represent one standard deviation of the mean of three (A,B) or two (C) biological replicates. P hydrolysis rates lacking a shared letter are significantly different (p < 0.05; Tukey's honest significant difference).

# Substrate-Specific P Hydrolysis Rates in the Coastal Western North Atlantic

Maximum hydrolysis rates of inorganic poly-P were measured in whole seawater samples from the coastal North Atlantic (**Figure 6A**) using the SRP method with two substrates: 3poly-P and 45poly-P. Poly-P chain length had no effect on P-hydrolysis rates, and the results from both poly-P sources were therefore combined into single average rates of inorganic poly-P hydrolysis. These rates were compared to maximum MUF-P hydrolysis rates determined via fluorescent detection. Average volume-normalized P hydrolysis rates ranged from 3.8–19.3 nmol P L−<sup>1</sup> hr−<sup>1</sup> and 1.4–50.6 nmol P L <sup>−</sup><sup>1</sup> hr−<sup>1</sup> for inorganic poly-P and MUF-P, respectively (**Figure 6B**). Chlorophyll-normalized rates ranged from 5.6– 7.7 nmol P µg chl−<sup>1</sup> hr−<sup>1</sup> and 2.5–20.1 nmol P µg chl−<sup>1</sup> hr−<sup>1</sup> for inorganic poly-P and MUF-P, respectively (**Supplementary Table S1**). SRP levels were above ∼700 nmol L−<sup>1</sup> , except in a phytoplankton bloom along the New Jersey Coast, where SRP was 150 ± 15 nmol L−<sup>1</sup> (avg ± SD, n = 3; **Supplementary Table S1**).

Inorganic poly-P hydrolysis rates were significantly higher than rates of MUF-P hydrolysis (p < 0.05) in four out of the five field sites sampled, by a factor of ∼1.5 to ∼2.7 (**Figure 6B** and **Table 2**). However, in one field site (Buzzard's Bay), MUF-P hydrolysis was ∼2.5-fold higher than inorganic poly-P hydrolysis (p < 0.05; **Figure 6B** and **Table 2**). Inorganic poly-P hydrolysis rates exhibited a significant linear correlation with chlorophyll levels (p = 0.01; **Figure 6C**), which was also evident for rates of MUF-P hydrolysis when the Buzzard's Bay outlier was excluded (p = 0.04; **Figure 6C**). However, P hydrolysis rates did not exhibit a significant relationship with bacterial abundance counts (**Figure 6D**). Bacteria cell counts (cells mL−<sup>1</sup> ) were ∼125 times higher than phytoplankton counts at the Buzzard's Bay site, yet only ∼17 to ∼45 times higher at all other locations (**Supplementary Table S1**).

SRP uptake/adsorption by whole seawater samples can potentially lead to the underestimation of P hydrolysis rates made by the SRP method. The degree of SRP uptake/adsorption was therefore assessed with standard additions of Na2HPO<sup>4</sup> (P = 2–20 µM). Phosphate levels in these amended samples did not change over the course of the 24–65 h incubations.

#### Phosphatase Diversity in Diatom Cultures

The full genome data of two model diatom strains utilized in this study, T. oceanica CCMP1005 and T. pseudonana CCMP1335, are publicly available, enabling the analysis of phosphatase diversity in these two species. First, the genomes were searched for putative homologs of three bacterial alkaline phosphatase isoforms (PhoA, PhoD, and PhoX) using BLASTP analysis. In T. oceanica CCMP1005, results revealed the presence of two putative PhoD homologs, one putative PhoX homolog, but no homologous PhoA sequences (**Table 3**). On the other hand, T. pseudonana CCMP1335 possesses one putative homolog each of PhoA and PhoD, but no homologous sequences to PhoX (**Table 3**).

Since cell-free filtrates were the primary focus of substratespecific P hydrolysis measurements, the diversity of phosphatases was investigated in cell-free filtrates of T. oceanica CCMP1005 and T. pseudonana CCMP1335 (under +P conditions for each culture only). Excluding contaminant sequences (i.e., keratin),

TABLE 1 | Comparison of ATP and 3poly-P hydrolysis rates in diatom cell-free filtrates from Figure 2.


T-scores (T; reported as absolute values) are calculated by subtracting from average experimental ATP:3polyP hydrolysis ratios (Avg) the proportion of ATP:3polyP hydrolysis expected from the complete degradation of phosphoanhydride bonds (66.7%), according to the following equation:

$$\mathcal{T} = \frac{\mathcal{A}\mathcal{W} - \mathcal{G}\mathcal{G}\mathcal{Z}}{\mathcal{S}\mathcal{E}}$$

where SE is the standard error of the average ATP:3poly-P hydrolysis ratio, and df is the degrees of freedom. Two-tailed p-values are reported.

a total of 309 and 133 proteins were identified in extracellular proteomes of T. oceanica CCMP1005 and T. pseudonana CCMP1335, respectively (**Supplementary Tables S2**, **S3**). Among these hits, 24 from T. oceanica CCMP1005 and 9 from T. pseudonana CCMP1335 possess conservative domains consistent with phosphatase activity. The majority of these hits were shared among biological triplicates of each diatom (**Figure 7**). None of the putative bacterial alkaline phosphatase homologs identified through genome searches (**Table 3**) were detected in the cell-free culture filtrates, however. Rather, results indicated a diversity of other putative phosphatases, including P-anhydride- and P-ester-degrading enzymes. In fact, P-anhydride-degrading enzymes dominated the phosphatase diversity in the cell-free filtrates of both microorganisms, including inorganic pyrophosphatase and nucleoside triphosphates (**Figure 7**).

#### DISCUSSION

The goal of this study was to assess the relative lability of model P sources that represent the major marine DOP bond classes. The focus here was on the P(V) DOP bond classes – P-esters and P-anhydrides – because of the potential role that these compound classes have in P utilization by the bulk phytoplankton community. Maximum potential hydrolysis rates of model P-esters and P-anhydrides were assessed in the cell-free filtrates of representative diatom cultures and coastal field sites, while phosphatase diversity was also investigated in model diatom filtrates. Overall, results revealed that an innate capacity for the preferential utilization of P-anhydrides was prevalent across all of these sample types.

#### Degradation of Model P Sources by Diatom Cultures

In model diatoms, maximum rates of P hydrolysis were examined in cell-free filtrates that were prepared via syringe filtration of culture samples. Cell-surface and intracellular phosphatases may have been adventitiously released through the potential disruption and lysis of some cells during this process. Thus, P hydrolysis rates in these cell-free filtrates may overestimate truly dissolved rates of P hydrolysis. However, the aim here was not to quantify cell-free P hydrolysis rates per se, but to avoid technical barriers associated with analyzing the degradation of unlabeled P sources in the presence of cells. In particular, tracking P hydrolysis through the production of SRP becomes difficult at high cell densities because microorganisms have a strong ability to adsorb (Sanudo-Wilhelmy et al., 2004) and assimilate dissolved P.

Cell-free filtrates can only provide a glimpse into the overall DOP preferences of a culture because most DOP degradation takes place in the presence of cells. Thus, MUF-P hydrolysis in whole cultures was measured with a standard fluorometric protocol in order to help contextualize the P dynamics observed in cell-free filtrates. Surprisingly, inorganic 3poly-P degradation in the filtrates vastly exceeded MUF-P hydrolysis levels in whole cultures (p < 0.05), by at least ∼3-fold (T. pseudonana

FIGURE 6 | Inorganic poly-P hydrolysis (SRP method) and MUF-P hydrolysis (fluorometric method) in the coastal western North Atlantic. (A) Samples were collected from the New Jersey coast (green; 39◦24043.5000N, 73◦14056.6400W), offshore Long Island (orange; 40◦53043.9800N, 71◦22049.9200W), Narragansett Bay (yellow; 41◦17012.1200N, 71◦30011.6400W), Woods Hole harbor (blue; 41◦31024.6000N, 70◦40040.8000W), and Buzzard's Bay (red; 41◦32038.2800N, 70◦4008.4600W). (B) Substrate-specific P hydrolysis comparison. The diagonal line depicts the 1:1 ratio of P hydrolysis rates. (C) Relationship of P hydrolysis rates to chlorophyll levels. For inorganic poly-P hydrolysis, the regression analysis was performed on the entire dataset (note the two overlapping data points at ∼3 µg L−<sup>1</sup> ). For MUF-P hydrolysis, the Buzzard's Bay outlier (note broken axis) was excluded from the regression analysis. Chlorophyll correlated significantly with inorganic poly-P degradation (p = 0.01) and MUF-P hydrolysis (p = 0.04). (D) P hydrolysis rates and bacterial abundance. In all panels, error bars represent the standard error of the mean of replicate samples (MUF-P hydrolysis: n = 3; inorganic poly-P hydrolysis: n = 8; chlorophyll: n = 3; bacterial abundance: n = 3).


TABLE 2 | Comparison of inorganic poly-P and MUF-P hydrolysis rates in field samples from Figure 6.

T-scores (T; reported as absolute values) are calculated using an independent two-sample T-test assuming equal variance, according to the following equation:

$$T = \frac{\mathcal{P}\#\_{\text{pro}(\rho,\rho)} - \mathcal{P}\#\_{\text{MJ}(\rho,\rho)}}{\text{SE}\_{\text{pro\text{totort}}}}$$

where PHpoly-P is the average P hydrolysis rate for inorganic polyhosphate, PHMUF−<sup>P</sup> is the average P hydrolysis rate for MUF-P, and SEpooled is the pooled standard error of PHpoly-P and PHMUF-P. Two-tailed p-values are reported.

CCMP1014) and up to ∼9-fold (T. oceanica CCMP1005; **Figure 4**). This finding is probably not a result of artificially elevated 3poly-P hydrolysis due to a filtration artifact, because most extracellular DOP-degrading enzymes localize on the cell surface and thus, most of the DOP degradation capacity is removed by filtering the cells out. Indeed, MUF-P hydrolysis was significantly lower in cell-free filtrates than in whole cultures (p < 0.05; **Figure 4**). Nor are the large differences between whole-culture MUF-P hydrolysis and filtrate-based 3poly-P degradation rates a consequence of the different detection TABLE 3 | Putative homologs of bacterial alkaline phosphatase in the genomes of T. oceanica CCMP1005 and T. pseudonana CCMP1335.


Hits are listed by the GenBank accession number.

fmars-05-00380 October 23, 2018 Time: 14:30 # 11

methods used. In fact, the SRP method is the more conservative approach (**Figure 5**) yet still shows remarkable correspondence with the MUF-P fluorescence method (**Figure 4**), especially given the large difference in sensitivity between these two techniques.

Relatively low levels of P hydrolysis were detected in the cell-free filtrates of actively growing and late log-phase cultures (**Figure 2A**, days 6–18; **Figure 2C**, days 6–15; **Figure 2E**, days 9, 17; **Figures 3D–I**). However, P hydrolysis rates were highest during the later stages of cultivation, including stationary phase and crash (**Figures 2**, **3**). In fact, the hydrolysis of 3poly-P and ATP increased as the cultures matured, even after accounting for increases in biomass through the growth curve (**Figures 3D–I**). These results may be a consequence of differences in the release, stability, or activity of cell-free P hydrolases over the life of the culture. For example, one interpretation is that active P-hydrolyzing enzymes can be released from healthy, growing cells, but these enzymes may be more extensively released from aging and senescent cells. This release could be actively driven by the diatoms themselves and/or mediated by physical mechanisms such as turbulence (e.g., daily culture mixing) or filtering. Like natural seawater, the cell-free filtrates examined in this study likely contain a mixture of actively and passively released proteins. If cell-free P hydrolysis is indeed linked to cell aging and senescence, as our data suggest, then the dissolved P hydrolase activity of seawater may be highest under certain environmental conditions, such as the collapse of phytoplankton blooms.

All P substrates examined (MUF-P, AMP, ATP, and 3poly-P) were actively degraded in diatom filtrates, but inorganic 3-polyP, followed by ATP, were the most rapidly transformed, regardless of overall P availability (**Figures 2**, **3**). Rates of ATP hydrolysis were lower (most cases) or approximately equal (one case) to that expected from the complete degradation of P-anhydride bonds (**Table 1**). Thus, P-ester hydrolysis may not have played a role in ATP degradation, even though P-ester hydrolysis lead to a relatively small level of AMP degradation in the same samples. Assuming that only terminal phosphate groups are enzymatically labile (Huang et al., 2018), we speculate that P-anhydrides may block P-esterases from accessing the P-O-C bond of ATP. Overall, these findings suggest that Thalassiosira spp. may preferentially hydrolyze inorganic poly-P compared to other DOP sources under a wide range of prevailing P availability. A variety of potential mechanisms may underlie the preferential degradation of P-anhydrides in diatom cell-free filtrates, including differences

in the amount, substrate versatility, stability, and diversity of P-hydrolyzing enzymes.

Thalassiosira oceanica exhibited the most versatile response to P amendments, consistent with a diverse nutrient utilization profile expected in this isolate from the P-limited Sargasso Sea. For example, T. oceanica had the broadest potential to utilize both 3poly-P and ATP, regardless of overall P status (**Figure 2**). Furthermore, every culture upregulated the degradation of ATP under P-deplete conditions, but T. oceanica was the only one that also increased its capacity to degrade 3poly-P in response to P scarcity (**Figure 3**). These results suggest that while enhanced ATP utilization may be a widespread adaptation to low P availability among Thalassiosira spp., elevated 3poly-P utilization may not be. Rather, in T. pseudonana, active P-anhydride degradation may be constitutive. Overall, the P utilization potential of T. pseudonana CCMP1014, an oceanic isolate from the NPSG, exhibited more similarity toward T. pseudonana CCMP1335 than T. oceanica. This finding may indicate that species- rather than habitat-level factors may be a more prevalent control over DOP cycling by these diatoms, although each of these strains has been in culture for over 40 years and some genetic drift has likely occurred.

#### P Hydrolysis of Model DOP Sources in the Coastal Western North Atlantic

Although some SRP uptake probably occurred in our field incubations, the results suggest that it was too slow to decrease SRP concentrations by more than the detection limit of 800 nmol L−<sup>1</sup> . This outcome is consistent with SRP uptake rates observed in other coastal systems with similar chlorophyll concentrations as our field sites (Harrison et al., 1977). SRP uptake therefore did not affect measured rates of inorganic poly-P hydrolysis.

Degradation rates of inorganic 3poly-P and 45poly-P were similar in field samples, which is consistent with the chain length-independent utilization of poly-P by some eukaryotic phytoplankton observed in a previous study (Diaz et al., 2016). Maximum rates of inorganic poly-P hydrolysis determined by the SRP method were significantly higher than fluorescence-based measurements of MUF-P hydrolysis in four out of five field sites examined in the coastal North Atlantic (**Figure 6B** and **Table 2**). These differences may be even more extreme, based on the finding that the SRP method for P hydrolysis is systematically conservative relative to the fluorescence-based approach (**Figure 5**).

The majority of studies on marine DOP hydrolysis have focused on P-monoester degradation in oligotrophic waters, where DOP hydrolysis is thought to function primarily as a phosphate-inhibitable phytoplankton response to P depletion (Duhamel et al., 2010; Lin et al., 2016). Above the threshold SRP level of ∼30 nmol L−<sup>1</sup> , P-monoester hydrolysis is typically less than ∼1 nmol L−<sup>1</sup> hr−<sup>1</sup> (∼20 nmol µg chl−<sup>1</sup> hr−<sup>1</sup> ) in these systems (Mahaffey et al., 2014). Consistent with this finding, chlorophyll-normalized rates of MUF-P and inorganic poly-P hydrolysis are less than ∼20 nmol µg chl−<sup>1</sup> hr−<sup>1</sup> in the relatively high-P coastal sites examined here (SRP > ∼150 nmol L−<sup>1</sup> ). However, because total chlorophyll levels are much higher in the waters examined in this study, the volume-normalized rates are much higher (∼4- to ∼20-fold) than in the oligotrophic ocean. Similar rates of volume-normalized P-ester hydrolysis have been previously reported from coastal systems and deep waters (Li et al., 1998; Dyhrman and Ruttenberg, 2006; Nicholson et al., 2006; Davis et al., 2014).

The strong correlation of P hydrolysis rates with chlorophyll levels (**Figure 6C**), and the lack of correlation with bacterial abundance (**Figure 6D**), suggests that the observed P dynamics were driven by phytoplankton. MUF-P hydrolysis (but not inorganic poly-P degradation) at the Buzzard's Bay site was inconsistent with the overall relationship to chlorophyll, which suggests additional, non-phytoplankton associated mechanisms of P-monoester degradation, such as contributions from heterotrophic bacteria. Heterotrophic bacteria utilize phosphatases to remineralize and acquire the carbon and nitrogen contained in DOP (Wilkins, 1972; Wanner and McSharry, 1982; Hoppe and Ullrich, 1999). Because the degradation of inorganic poly-P yields no carbon or nitrogen, heterotrophic bacteria may not contribute extensively to the degradation of inorganic poly-P sources. Thus, in a system dominated by bacterially driven DOP hydrolysis, the inorganic poly-P and P-ester dynamics may be decoupled, such that P-ester hydrolysis rates are substantially higher than inorganic poly-P degradation, consistent with results from the Buzzard's Bay site. In agreement with this hypothesis, the ratio of bacteria to phytoplankton cells was ∼3 fold higher in Buzzard's Bay than at any other location (**Supplementary Table S1**). Furthermore, White et al. (2012) found that 3poly-P was less extensively degraded than P-esters like AMP and glucose-6-phosphate in seawater incubations that had been pre-conditioned to favor bacterial growth.

#### Phosphatase Diversity in Diatom Cultures

Enzymes modulating the P-hydrolysis reactions observed in diatom cultures and coastal Atlantic field samples may include phosphatases such as AP, inorganic pyrophosphatase, and nucleoside triphosphatases. AP's are the best-known group of enzymes involved in DOP utilization by phytoplankton (Lin et al., 2016). These P-esterases are widely distributed in prokaryotic and eukaryotic microorganisms yet show only limited homology across these two groups (Lin et al., 2012). Furthermore, AP's are much better characterized in prokaryotes than microbial eukaryotes (Shaked et al., 2006; Xu et al., 2006, 2010; Lin et al., 2011, 2013, 2015, 2016). For example, major prokaryotic AP isoforms include PhoA, PhoX, and PhoD (Luo et al., 2009, 2011). All of these have been detected in seawater, with PhoD being the most prevalent, followed by PhoX (Luo et al., 2009; Sebastian and Ammerman, 2009; Kathuria and Martiny, 2010; Luo et al., 2011). Substrate specificities of the AP isoforms are not completely characterized, but they may exhibit a wide degree of flexibility. For example, E. coli PhoA is also a polyphosphatase (Lorenz and Schroder, 2001; Huang et al., 2018).

Multiple putative AP's have been recognized in the genome, transcriptome, and proteome of T. pseudonana CCMP1335 (Armbrust et al., 2004; Dyhrman et al., 2012), and at least one of these has been recognized for its resemblance to bacterial PhoA (Lin et al., 2016). Using BLASTP translated genome searches of T. pseudonana CCMP1335 and T. oceanica CCMP1005 with bacterial PhoA, PhoD, and PhoX queries, additional putative AP homologs were identified herein (**Table 3**). However, none of these sequences were apparent in the exoproteome data of either diatom species (**Figure 7**), suggesting that these enzymes may be exclusively cell-associated and ruling them out from participating in the P-hydrolysis dynamics observed in culture filtrates. Exoproteome samples revealed a number of other putative phosphatases, the diversity of which was biased in favor of P-anhydride degradation (**Figure 7**), consistent with the observed preferential hydrolysis of P-anhydrides in culture filtrates (**Figures 2–4**). The mechanisms leading to the appearance of these phosphatases in the exoproteome remain unclear, but may potentially include active secretion, passive dislodging of cell surface-associated enzymes during cultivation, and/or adventitious release during filtrate preparation.

Nearly all of the phosphatase hits present in the T. pseudonana CCMP1335 exoproteome (**Figure 7**) were also detected in its cell-associated transcriptome and proteome, which were reported in a previous study (Dyhrman et al., 2012). The only exception was the P-diesterase (GenBank EED94265). However, this sequence also could not be found in the JGI translated genome assemblies for T. pseudonana CCMP1335 utilized by Dyhrman et al. (2012). The absence of a protein sequence from the reference database precludes its identification in the experimental data. Thus, the absence of EED94265 in the proteome data from this previous study is at least partially a result of the different methods employed. As for the other putative phosphatases detected in the present study, none were upregulated in the T. pseudonana CCMP1335 transcriptome or proteome under P-deplete conditions (Dyhrman et al., 2012). In fact, three hits were upregulated under P-replete conditions: two GTPases (EED91276 in the proteome only; EED90008 in the transcriptome only) and one ATPase (transcriptome and proteome: EED92744). These results are consistent with the lack of upregulation of 3poly-P hydrolysis in the cell-free filtrate of this species under -P conditions (**Figure 3E**). On the other hand, ATP degradation was enhanced (∼2.5-fold) under low P availability in T. pseudonana CCMP1335 cell-free filtrates (**Figure 3H**), but this activity may not be regulated or sufficiently resolved at the transcriptional or translational levels.

Some of the exoproteome hits identified in this study may exhibit phosphatase activities by operating in the reverse direction of their annotated functions (e.g., ATP synthase components). Sequence information can identify candidate enzymes for a given biochemical function, however, these hits must be purified and assessed in vitro to verify their actual enzymatic reactivity. Furthermore, the presence of some protein hits in the exoproteomes may be unexpected. For example, an abundance of proteins affiliated with the ATP synthase complex were found in the exoproteomes of both diatom species. However, in some microorganisms, ATP synthase localizes in the cell membrane, utilizing the proton-motive force produced by the light activated proton-pump proteorhodopsin to generate ATP (Lin et al., 2016). In fact, proteorhodopsin expression may be an active aspect of diatom physiology (Marchetti et al., 2012; Marchetti et al., 2015). The potential localization of ATP synthase on the plasma membrane suggests that it could potentially become adventitiously dislodged without substantial cellular disruption or lysis (e.g., through daily physical mixing of cultures). Indeed, the release of cell-associated material was likely minimal during the preparation of diatom cell-free filtrates, as only 133 protein hits were detected in T. pseudonana CCMP1335 exoprotomes out of 1264 total protein hits reported in a prior study (Dyhrman et al., 2012).

#### Synthesis and Perspectives

The P-anhydride bond is essential to life. It serves as the universal basis of cellular energy transduction through the formation and hydrolysis of ATP and also plays diverse regulatory roles in the form of inorganic polyphosphates that are made by every cell in nature. As a widespread and labile constituent of marine DOP, P-anhydrides may also represent a bioavailable P source capable of supporting phytoplankton growth and productivity in the ocean. This P-anhydride pool has largely been overshadowed by the prevailing assumption that the much more abundant standing stock of marine P-esters is the most important DOP source. Based on results from this study, however, the possibility that P-anhydrides support a high degree, maybe even a majority, of DOP turnover and assimilation in diverse marine environments should be considered.

If P-esters are less preferred than P-anhydrides, then APA measurements may underestimate levels of P stress in oligotrophic systems, where phosphatase activity primarily functions in the acquisition of P by P-starved phytoplankton. In high-nutrient coastal systems, P-anhydrase activity may be constitutively expressed regardless of prevailing orthophosphate availability (Ammerman and Azam, 1991), suggesting that P-anhydrides may be a part of the highly labile DOP pool that is rapidly cycled in these environments (Benitez-Nelson and Buesseler, 1999). For example, in T. pseudonana, degradation of inorganic 3poly-P was just as high or higher under replete P conditions (**Figures 3E,F**). Furthermore, in T. pseudonana CCMP1335, this trend was consistent at the protein level, where each of the potential enzymes involved in 3poly-P degradation (**Figure 7**) were just as prevalent or more prevalent under P-rich conditions examined in a previous study (Dyhrman et al., 2012).

The degradation of dissolved P-anhydrides may be cryptic, with the relatively low abundance of this pool (Young and Ingall, 2010) masking rapid recycling rates. In situ P-anhydride degradation rates are dependent on the inherent degradation kinetics of the bulk P-anhydrase pool, as well as the in situ concentration of P-anhydride substrates. The maximum

hydrolysis rates reported herein do not necessarily reflect in situ degradation rates, because actual P-anhydride levels may not be rate-saturating. In situ DOP utilization rates have been estimated based on standing stock DOP concentrations (Duhamel et al., 2011), but cryptic cycling commonly evades detection by concentration-based approaches (Canfield et al., 2010; Hansel et al., 2015; Berg et al., 2016). Rather, cryptic transformation rates may be more readily quantified by using high-sensitivity radiotracers amended at levels similar to in situ substrate concentrations. Biologically mediated cryptic pathways can also be illuminated by investigating the community-level metabolic potential, as manifest in environmental genome, transcriptome, and proteome sequence data. As a step toward this goal, the present study has identified a diversity of diatom-derived phosphatases that may drive P-anhydrase reactions in the marine environment.

In addition to radiotracers and bioinformatics approaches, the study of P-anhydride degradation in marine systems would also be advanced by the availability of representative labeled substrates. For example, P-ester degradation (conventionally referred to as APA) is routinely assessed using sensitive fluorogenic and chromogenic substrates (Hoppe, 2003; Dyhrman and Ruttenberg, 2006; Mahaffey et al., 2014; Davis and Mahaffey, 2017). By quantifying the inert, signal-bearing probe that is released during P hydrolysis, these substrates can be used to track P-hydrolysis rates independently of P dynamics, which is especially important at high cell densities. Synthesis of labeled inorganic poly-P has been described (Choi et al., 2010; Hebbard et al., 2014), but these substrates were unreactive in positive controls tested in this study (data not shown).

Future work should focus on methods to assess the P hydrolysis of diverse DOP compounds by whole cells. The application of such methods should confirm whether the DOP dynamics observed here in diatom cell-free filtrates reflect broader DOP nutritional preferences. For example, the P hydrolytic potential of diatom cell-free filtrates implies that these microorganisms have a capacity, and potentially a preference in the case of 3poly-P, to utilize a diversity of P sources for growth. If cultures had been amended with 3poly-P during stationary phase, the elevated potential for 3polyP hydrolysis (**Figure 2**) may have helped the cultures avoid crashing, although this was not tested. The P-hydrolysis of DOP sources is not necessarily coupled to the biological assimilation of enzymatically released P, however (Ammerman and Azam, 1985; Duhamel et al., 2017). For instance, heterotrophic bacteria are thought to degrade organic P sources as a carbon and/or nitrogen utilization strategy, and phosphate may accumulate in seawater as a result (Hoppe and Ullrich, 1999; Hoppe, 2003). Furthermore, if phosphatases are not actively degraded, then their activity can lead to the accumulation of dissolved inorganic P when other factors besides P availability limit plankton growth. In such a case, the accumulation of enzymatically released P may have other implications for marine ecosystem functioning. For example, the primary pathway for long-term P removal is the authigenic (or in situ) formation of marine calcium phosphate minerals (Ruttenberg, 2014). The direct precipitation of these minerals is kinetically inhibited in most marine environments but may involve inorganic poly-P as a reactive intermediate (Schulz and Schulz, 2005; Diaz et al., 2008). In fact, the inherently high capacity for P-anhydride degradation documented in diatom cultures and natural samples in this study suggests that pulsed or spatially-heterogeneous poly-P inputs could lead to the formation of transient, submicron-scale "hotspots," which may be kinetically conducive to the formation of authigenic calcium phosphate minerals.

#### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript are deposited at the Biological and Chemical Oceanography Data Management Office (http://bco-dmo.org) under project number 747715.

#### AUTHOR CONTRIBUTIONS

JD, SD, and YT conceived the experiments and overall study design. JD, AH, JS, KB, DM, C-WC, and DP generated results and conducted the data analysis. All authors contributed to manuscript preparation.

# FUNDING

This work was supported by the National Science Foundation under grants 1559124 (JD), 1736967 (JD), 1737083 (SD), and 1559087 (YT), as well as a Junior Faculty Seed Grant from the University of Georgia Research Foundation (JD). The Thermo-Fisher LTQ Orbitrap Elite Mass Spectrometer was purchased with funding from National Institutes of Health under grant S10RR028859.

# ACKNOWLEDGMENTS

We thank Colleen Hansel, Carl Lamborg and the science party and crew of the R/V Endeavor during the TORCH I cruise. We thank James Morrissey for his generosity in providing fluorogenic polyphosphate for our research, and Catherine Baker for synthesizing these substrates. We also thank Lee Ann DeLeo for her help with the preparation of graphics for this paper, and we acknowledge Mohit Jain for technical assistance with proteomics data analysis.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00380/full#supplementary-material

# REFERENCES

fmars-05-00380 October 23, 2018 Time: 14:30 # 15


trace metal addition experiments. Front. Mar. Sci. 1:73. doi: 10.3389/fmars.2014. 00073


fmars-05-00380 October 23, 2018 Time: 14:30 # 16


influence of the Pearl River. Deep Sea Res. I Oceanogr. Res. Pap. 55, 1330–1342. doi: 10.1016/j.dsr.2008.05.007


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Diaz, Holland, Sanders, Bulski, Mollett, Chou, Phillips, Tang and Duhamel. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

, Fabrizio D'Ortenzio<sup>2</sup>

,

# Sensitive Determination of the Dissolved Phosphate Pool for an Improved Resolution of Its Vertical Variability in the Surface Layer: New Views in the P-Depleted Mediterranean Sea

#### Edited by:

Kahina Djaoudi <sup>1</sup>

Christian Lønborg, Australian Institute of Marine Science (AIMS), Australia

#### Reviewed by:

Masahiro Suzumura, National Institute of Advanced Industrial Science and Technology (AIST), Japan LeAnn Whitney, University of Rhode Island, United States

#### \*Correspondence:

Kahina Djaoudi kahina.djaoudi@mio.osupytheas.fr

#### Specialty section:

This article was submitted to Marine Biogeochemistry, a section of the journal Frontiers in Marine Science

Received: 01 February 2018 Accepted: 18 June 2018 Published: 10 July 2018

#### Citation:

Djaoudi K, Van Wambeke F, Coppola L, D'Ortenzio F, Helias-Nunige S, Raimbault P, Taillandier V, Testor P, Wagener T and Pulido-Villena E (2018) Sensitive Determination of the Dissolved Phosphate Pool for an Improved Resolution of Its Vertical Variability in the Surface Layer: New Views in the P-Depleted Mediterranean Sea. Front. Mar. Sci. 5:234. doi: 10.3389/fmars.2018.00234 Sandra Helias-Nunige<sup>1</sup> , Patrick Raimbault <sup>1</sup> , Vincent Taillandier <sup>2</sup> , Pierre Testor <sup>3</sup> , Thibaut Wagener <sup>1</sup> and Elvira Pulido-Villena<sup>1</sup> <sup>1</sup> Aix-Marseille Univ., Université de Toulon, Centre National de la Recherche Scientifique, IRD, MIO UM 110, Marseille, France,

\*, France Van Wambeke<sup>1</sup>

<sup>2</sup> Sorbonne Université, CNRS, Laboratoire d'Océanographie de Villefanche, LOV, Villefranche-sur-mer, France, <sup>3</sup> Sorbonne Université, UPMC, Université Paris 06, Centre National de la Recherche Scientifique, IRD, MNH, Laboratoire d'Océanographie et de Climatologie, IPSL, Paris, France

, Laurent Coppola<sup>2</sup>

An accurate understanding of the biogeochemistry of dissolved phosphate pool in the upper waters of P-depleted oceanic regions is constrained by the low sensitivity of routine phosphate measurements. In this study, by using the sensitive Liquid Waveguide Capillary Cell method, we report the first extensive cross-basin survey of nanomolar dissolved inorganic phosphate (DIP) and dissolved organic phosphate (DOP) concentration in P-depleted surface waters of the Mediterranean Sea during the stratification period. In the north western Mediterranean Sea (NWMS), DIP above the mixed layer depth (MLD) ranged between 4.9 and 26.5 nM. Along an E-W transect crossing Ionian and Tyrrhenian Seas (E-W transect), DIP above the MLD was lower, ranging between 0.9 and 11.4 nM. Contrarily to the traditional view of a depleted and invariant surface dissolved phosphate pool, a significant vertical variability of DIP and DOP was revealed in upper waters. A positive gradient of DIP was observed above the phosphacline, between the MLD and the deep chlorophyll maximum (DCM) depth, suggesting a potential diffusion of new phosphate to near-surface waters, even under stratified conditions. Interestingly, despite this apparent DIP availability, a significant negative gradient of DOP concentration was observed in the same layer. Finally, the positive gradient in DIP coincided with a significant increase in N:P ratio, suggesting a higher rate of increase of N than of P. The results obtained in this study indicate that acquiring nanomolar DIP data is a sine qua non condition for the comprehension and prediction of the biogeochemical functioning of P-depleted oceanic regions, such as the Mediterranean Sea.

Keywords: the Mediterranean Sea, dissolved inorganic phosphate, dissolved organic phosphate, N:P ratios, liquid waveguide capillary cell

#### Djaoudi et al. Nanomolar Phosphate in Mediterranean Sea

# INTRODUCTION

Nitrogen (N) and phosphorus (P) are essential macronutrients for sustaining marine organisms (Redfield, 1958). By controlling the efficiency of the biological pump, N and P determine the strength of the oceanic carbon uptake (Balino et al., 2001). Over the vast oligotrophic ocean, upper waters are often subjected to N and P exhaustion due to a strong vertical stratification and an active drawdown by biological activity (Tyrrell, 1999). N has traditionally been viewed as the limiting nutrient of primary production in marine waters (Ryther and Dunstan, 1971). However, with an increased understanding of the N cycling in the ocean, there is a growing awareness that P has also the potential to limit primary production in certain regions of the oligotrophic ocean (i.e., Wu et al., 2000; Moutin et al., 2008; Lomas et al., 2010). The Redfield N:P ratio of 16:1 has been considered as a fundamental tenet in marine biogeochemistry. Deviations from this canonical ratio have been used to provide insights, for instance, on the efficiency of carbon sequestration in the ocean (Sigman and Boyle, 2000), or on nutrient limitation of primary production (i.e., Falkowski and Raven, 1997; Moore et al., 2013). In some oligotrophic regions, such as the North Atlantic subtropical gyre, N:P ratios in near-surface waters are higher than 16:1, suggesting a depletion of P before that of N (Wu et al., 2000; Ammerman et al., 2003).

In such P-depleted oligotrophic regions, the quantification of dissolved inorganic phosphate (DIP) is a challenging task because concentrations drop to nanomolar (nM) levels, so unquantifiable using the conventional automated colorimetric procedure (CACP) which presents a limit of detection (LOD) of 20 nM, at best cases. Alternative analytical procedures have been introduced to improve the LOD and precision of DIP analysis (see review by Patey et al., 2008). The pre-concentration by magnesium-induced co-precipitation (MAGIC) method substantially improved the amount of DIP data at nM levels (Karl and Tien, 1992; Rimmelin and Moutin, 2005). During the last decade, the Liquid Waveguide Capillary Cell (LWCC) method, have gained a foothold as a reliable method for nM DIP measurements (i.e., Zhang and Chi, 2002; Patey et al., 2008). By using these high sensitive methods, studies in both the North Pacific (Karl et al., 2001 and references therein; Suzumura et al., 2012; Hashihama et al., 2013) and North Atlantic (Ammerman et al., 2003; Lomas et al., 2010) subtropical gyres have revealed DIP concentrations as low as 0.2 nM (Wu et al., 2000). Moreover, since the determination of dissolved organic phosphate (DOP) relies upon DIP measurements, some of the above cited studies have also provided accurate DOP data (i.e., Mather et al., 2008; Suzumura et al., 2012), crucially enhancing our knowledge on the biogeochemical functioning of P-depleted oceanic regions, where DOP represents a P source for sustaining community productivity (Karl and Björkman, 2015 and references therein; Sato et al., 2013).

The Mediterranean Sea is one of the world's most oligotrophic oceanic regions. It is divided into two sub-basins, western and eastern, exhibiting distinct hydrodynamics and biogeochemical features. The Mediterranean Sea exhibits low nutrient concentrations, with a west to east decreasing gradient (Krom et al., 1991; Moutin and Raimbault, 2002; De Madron et al., 2011), partly linked to its anti-estuarine thermohaline circulation (Crispi et al., 2001). Available nutrients in the Atlantic surface inflow entering the Mediterranean Sea are consumed along the way to the eastern basin and exported to deep waters (Crise et al., 1999; Lazzari et al., 2012). Consequently, the western Mediterranean basin is oligotrophic while the eastern basin is ultraoligotrophic (Moutin and Raimbault, 2002).

In the Mediterranean surface waters, DIP concentration has been shown to limit primary production, nitrogen fixation and heterotrophic bacterial activity (Van Wambeke et al., 2002; Thingstad et al., 2005). These findings are supported by low DIP concentrations measured in upper waters and a deep N:P ratio which diverges greatly from the canonical Redfield ratio, with values ranging between 21 and 23 in the Western basin (Bethoux et al., 1992) and even higher in the Eastern basin (Krom et al., 1991). Surface DIP concentration in surface waters is systematically below 20 nM (LOD of the CACP) during the stratification period, dramatically constraining data coverage in this oceanic region (i.e., Pujo-Pay et al., 2011; Céa et al., 2015). The depth at which DIP concentration becomes measurable by the CACP is usually used to define the phosphacline and it has been reported to be between 45 m and 120 m, during the stratified period, in the western and eastern Mediterranean Sea, respectively (Moutin and Raimbault, 2002). To date, lack of DIP data above this phosphacline has precluded resolving vertical variability in this key layer as has led to the traditional view of an entirely depleted and invariant dissolved phosphate surface pool. Few studies, however, by using sensitive determination of DIP, have reported low but detectable increases in nanomolar DIP concentration above the phosphacline adding evidence against this traditional view (Van Wambeke et al., 2002; Krom et al., 2005; Djaoudi et al., 2017). Nevertheless, the scarcity and the sparse nature of available nM DIP data in the Mediterranean Sea still precludes a complete assessment of DIP dynamics in the upper waters at the basin scale.

As for DIP, a critical gap in accurate DOP data (i.e., based on nanomolar DIP measurements) persists in the upper waters of the Mediterranean Sea (Moutin and Raimbault, 2002; Pujo-Pay et al., 2011). Indeed, only two studies have reported nanomolar DIP and DOP data simultaneously (Krom et al., 2005; Djaoudi et al., 2017). In the Mediterranean Sea, when DIP is exhausted, DOP becomes a possible alternative source of P as confirmed by high activities of alkaline phosphatase suggesting an active utilization of the DOP pool in upper waters (Van Wambeke et al., 2002, 2009; Thingstad and Mantoura, 2005). How and to which extent such biogeochemical dynamics induce changes in DOP concentration above the phosphacline remains unknown.

In this study, we conducted cross-basin observations of nM DIP and DOP concentration in the Mediterranean Sea with emphasis on upper waters. The main goals of this work are: (1) to extend the nM dataset of accurate paired DIP/DOP data in the upper Mediterranean Sea under stratification conditions, (2) to examine the vertical variability of both DIP and DOP above the phosphacline, under stratification conditions and, (3) to better resolve the DIN:DIP and DON:DOP ratios in the upper waters of this P-depleted oceanic region.

### MATERIALS AND METHODS

#### Study Area and Sampling

This study was performed in the framework of two research cruises carried out in 2015 in the Mediterranean Sea, during the stratification period, MOOSE-GE (doi.org/10.18142/235) and BioArgomed (doi.org/10.17882/51678) (**Figure 1**).

The MOOSE-GE (Mediterranean Ocean Observing System for the Environment– Grande Échelle) cruise took place on board the R/V Le Suroit between 10 and 26 July in the oligotrophic North Western Mediterranean Sea (NWMS). The BioArgoMed cruise took place on board the R/V Tethys II between 13 May and 03 June along an east to west transect (E-W transect) crossing the ultraoligotrophic Ionian and Tyrrhenian Seas. Therefore, these cruises covered the two Mediterranean sub-basins, western and eastern, which exhibit distinct hydrological and biogeochemical features (i.e., De Madron et al., 2011).

During the two cruises, oceanographic casts were acquired using a conductivity-temperature-depth profiling system (SBE911 + CTD unit). The density profiles were calculated from temperature and salinity profiles, and linearly interpolated at 1 m depth intervals. Fluorescence profiles were provided by a Chelsea Aquatracka 3 fluorimeter interfaced with the CTD unit.

Seawater samples were collected during up casts with a carousel of Niskin bottles for analyses of dissolved phosphate (dissolved inorganic phosphate, DIP; total dissolved phosphate, TDP) and dissolved nitrogen (dissolved inorganic nitrogen, DIN; total dissolved nitrogen, TDN). In the present study, DIN includes nitrate + nitrite concentration and does not include ammonium. Pujo-Pay et al. (2011) reported NH<sup>+</sup> 4 concentrations in the Mediterranean Sea between 0 and 10 nM, indicating that DIN data reported in this study can suffer from up to 10 nM underestimation.

Samples were collected in duplicate and filtered on board through a pre-cleaned (500 mL of ultrapure water) 0.2µm syringe filter. Samples were then disposed into acid washed HDPE-60 mL bottles and were stored frozen (−20◦C) until analysis. To check for potential N and P contamination through filtration, DIP and DIN concentration were analyzed on ultrapure water filtered the same way as samples. DIN concentration in such blanks was below the limit of detection. DIP concentrations in the blanks was 2.8 ± 0.6 nM (n = 5), not significantly different from the reagent blanks (2.2 ± 0.3 nM, n = 5).

During the MOOSE-GE cruise, seawater samples were collected at 25 stations (**Figure 1**), between the surface and 100 m at 4 different depths. Sampling depths were chosen according to the density and fluorescence depth profiles. Density profiles were used to determine the depth of the mixed layer, in which a sampling point was defined (depth 1, **Figure 2**). From the fluorescence depth profiles, which exhibited the characteristic subsurface maximum (deep chlorophyll maximum, DCM) reported for the western Mediterranean Sea during the stratification period (Lavigne et al., 2015), 3 sampling points were defined: one point at the DCM (depth 3, **Figure 2**) and two points above and below the DCM (depths 2 and 4, **Figure 2**). During the BioArgoMed cruise, seawater samples were collected from 6 stations (**Figure 1**) at 6 different depths between the surface and 120 m.

### Dissolved Nutrient Analysis

DIP concentration was determined by using both the CACP (Aminot and Kérouel, 2007) and the LWCC method (Zhang and Chi, 2002). Both spectrophotometric methods are based on the molybdenum blue reaction, described by Murphy and Riley (1962). In the LWCC method, according to the Beer-Lambert law, the sensitivity of the spectrophotometric measurement is improved by increasing the optical path length of the measurement cell from 1 cm (CACP) to 2.5 m (LWCC). The LOD, estimated as three times the standard deviation of ten measurements of the blank, was 20 and 0.8 nM for the CACP and the LWCC method, respectively. For the LWCC method, the linearity of absorbance vs. DIP concentration was assessed, and the dynamic range was determined to be between 0.8 and 143 nM. Therefore, for DIP concentrations above such range, the DIP value provided by the CACP was used. Between 20 and 143 nM (range of concentrations for which both methods are comparable), the relationship between the two methods followed a linear model (r = 0.96; n = 80) with a slope of 1.14 ± 0.04, not significantly different from 1 (p > 0.05).

Total dissolved phosphate (TDP) concentration was analyzed by using the CACP after sample processing by the wet oxidation method (Valderrama, 1981). DOP concentrations were then estimated by subtracting nM LWCC-DIP from TDP concentrations, with an analytical precision of 7%.

Dissolved inorganic nitrogen (DIN) and total dissolved nitrogen (TDN) concentrations were measured by using the CACP with a LOD of 50 nM. As for TDP, samples for TDN were processed according to the wet oxidation method (Valderrama, 1981). Dissolved organic nitrogen (DON) concentrations were then estimated by subtracting DIN from TDN concentrations.

#### Data Treatment and Calculations

The mixed layer depth (MLD) was estimated from the density profiles as the depth where the difference of density from the surface reference (i.e., density at 10 m depth) was 0.03 kg m−<sup>3</sup> (De Boyer Montégut, 2004).

The depth of the phosphacline was calculated as the average of 4 phosphacline depth values determined based on 4 different reported criteria following Van Wambeke et al. (2009). Phosphacline<sup>1</sup> and phosphacline<sup>2</sup> were calculated as the shallowest depths in which DIP concentrations exceeded the mixed layer value by 10 and 50 nM, respectively (Van Wambeke et al., 2009); phosphacline<sup>3</sup> and Phosphacline<sup>4</sup> were calculated as the depths in which DIP concentrations exceeded the concentration at the immediate previous depths by 20 and 100 nM, respectively (Van Wambeke et al., 2009; Pujo-Pay et al., 2011).

The Wilcoxon signed-rank test was used to assess the overall significance of the vertical differences in DIP and DOP concentration for each cruise. For each individual profile, vertical differences in DIP and DOP concentration where considered significant when they were higher than the LOD and precision of the analytical methods.

stations, see Table 1 for references).

#### RESULTS

Over the study area and for both cruises, a stratification of the water column was observed. The MLD ranged between 6 and 30 m with an average of 13 ± 5 m (n = 25) during the MOOSE-GE cruise (hereafter, NWMS: for NW Mediterranean Sea) and between 14 and 20 m with an average of 17 ± 3 m (n = 6) during the BioArgoMed cruise (hereafter, E-W transect) (**Figure 3**). The phosphacline depth in the NWMS ranged between 30 and 89 m with an average of 56 ± 14 m for 21 stations and was higher than 100 m for the remaining 4 stations (**Figure 3**). Along the E-W transect, the phosphacline depths were higher than 120 m (maximum sampled depth) except for one value at 70 m (**Figure 3**).

#### Dissolved Phosphate Pools

By using the LWCC, all DIP concentrations in the upper waters of the Mediterranean Sea obtained in this study were above the LOD of 0.8 nM. Compared to the CACP, the sensitive LWCC method resulted in a 75% increase of valid DIP data and therefore of DOP data in the study area.

Surface (0–100 m) DIP concentration in the NWMS ranged over two orders of magnitude, between 3.6 and 251 nM (**Figure 3A**). Surface (0–120 m) DIP concentrations along the E-W transect were lower, ranging between 0.9 and 14.6 nM except for one single value at 127.5 nM (**Figure 3B**). DIP concentrations above the MLD ranged between 4.9 and 26.5 with an average of 12.8 ± 6.8 (n = 28) in the NWMS (**Figure 3A**) and between 0.9 and 11.4 nM with an average of 5.5 ± 3.6 nM (n = 11) along the E-W transect (**Figure 3B**).

Concerning vertical variability, DIP concentration in the NWMS exhibited a first significant increase between the MLD and the DCM depth (Wilcoxon signed-ranks test, Z = 3.2, p < 0.01, n = 25) (**Figure 4**). It then increased more sharply between the DCM depth and 100 m (Wilcoxon signed-ranked test, Z = 4.0, p < 0.001, n = 25) (**Figure 4**). In order to get a deeper insight on the observed DIP vertical variability, a gradient of DIP was calculated by dividing the difference in DIP concentration by the corresponding difference in depth. In the NWMS, the first DIP positive gradient (between the MLD and the DCM depth) was recorded in 21 out of 25 of the sampling stations (Figure S1) and ranged between 0.04 and 3.2 nM m−<sup>1</sup> with an average of 0.4 ± 0.7 nM m−<sup>1</sup> . The second positive gradient

FIGURE 3 | Vertical section of dissolved inorganic phosphate concentrations (DIP) in the North Western Mediterranean Sea (between the surface and 100 m, A) and along the E-W transect crossing Ionian and Tyrrhenian seas (between the surface and 120 m, B). The red plot represents the mixed layer depth. The orange plot represents the phosphacline depth. Note that the scale of the color bar has been tightened to point out the positive gradient of DIP in the North Western Mediterranean Sea.

bounded line the standard deviation from the mean. The blue dashed line corresponds to the limit of detection (LOD) of the conventional automated colorimetric

procedure (20 nM). The green plot represents the fluorescence profile (mean ± standard of deviation).

(between the DCM depth and 100 m) was recorded in 21 out of 25 of the sampling stations (Figure S1) and ranged between 0.03 and 5.7 nM m−<sup>1</sup> with an average of 2.6 ± 1.6 nM m−<sup>1</sup> .

Likewise, along the E-W transect, a significant increase in DIP concentration was observed between the MLD and the DCM depth (Wilcoxon signed-ranks test, Z = 2.2, p < 0.05, n = 6) (**Figure 4**). This DIP positive gradient was recorded for all stations and ranged between 0.01 and 0.08 nM m−<sup>1</sup> with an average of 0.04 ± 0.03 nM m−<sup>1</sup> (Figure S1). Because in the E-W transect the maximum sampling depth was above the phosphacline, the second increase in DIP concentration below the DCM was not observed.

Surface DOP concentration in the NWMS ranged between 4 and 149 nM with an average of 44 ± 25 nM (n = 100) (**Figure 5A**). As for DIP, DOP concentrations were lower along the E-W transect, ranging between 15 and 64 nM with an average of 33 ± 12 nM (n = 36) (**Figure 5B**). DOP concentration above the MLD ranged between 17 and 139 nM with an average of 47 ± 36 (n = 28) in the NWMS (**Figure 5A**) and between 20 and 62 nM with an average of 35 ± 11 nM (n = 11) along the E-W transect (**Figure 5B**).

Concerning the vertical variability, DOP concentration in the NWMS exhibited a significant decrease between the MLD and the DCM (Wilcoxon signed-ranks test, Z = 2.1, p < 0.05,

FIGURE 5 | Vertical section of dissolved organic phosphate concentrations (DOP) in the North Western Mediterranean Sea (between the surface and 100 m, A) and along the E-W transect crossing Ionian and Tyrrhenian seas (between the surface and 120 m, B). The red plot represents the mixed layer depth. The orange plot represents the phosphacline depth. Note that the scale of the color bar has been tightened to point out the negative gradient of DOP in the North Western Mediterranean Sea.

n = 25) (**Figure 6**). As for DIP, a DOP gradient between the MLD and the DCM was calculated by dividing the difference in DOP concentration by the corresponding difference in depth. The negative gradient of DOP in this layer was recorded in 15 out of 25 of the sampling stations (Figure S2) and ranged between 0.05 and 2.7 nM m−<sup>1</sup> with an average value of 0.6 ± 0.7 nM m−<sup>1</sup> .

Along the E-W transect, no significant difference in DOP concentration was observed between the MLD and the DCM depth (Wilcoxon signed-ranks test, Z = 0.94, p > 0.05, n = 6).

# N:P Molar Ratios in the Upper Mediterranean Sea

In the NWMS, surface (0–100 m) DIN:DIP ratios ranged between 2 and 165 (**Figure 7**) and DON:DOP ratios ranged between 55 and 525 (**Figure 8**). Above the mixed layer depth, DIN:DIP ratios ranged between 1 and 65 with an average of 20 ± 19 and DON:DOP ratios between 113 and 312 with an average of 182 ± 55. Along the E-W transect, 50% of DIN data were below the LOD of the CACP. When DIN data were available, the surface (0–120 m) DIN:DIP ratios in this area ranged between 10 and 289 (**Figure 7**) and DON:DOP ratios ranged between 67 and 252 (**Figure 8**). Along the E-W transect, DIN:DIP and DON:DOP molar ratios have not been calculated above the MLD due to insufficient valid DIN data.

Regarding vertical profiles, in the NWMS, there was a significant increase of DIN:DIP (Wilcoxon signed-ranks test, z = 3.4, p < 0.001, n = 19) and of DON:DOP ratios (Wilcoxon signed-ranks test, z = 2.4, p < 0.05, n = 17), between the MLD

Redfield molar ratio of 16:1.

and the DCM (**Figures 7**, **8**). The increase in the DIN:DIP ratio was observed for 15 out of 19 stations (Figure S3) and ranged between 0.2 and 5.7 m−<sup>1</sup> . The increase in the DON:DOP ratio within this layer was observed for 12 out of 17 stations (Figure S4) and ranged between 0.2 and 9.3 m−<sup>1</sup> . Along the E-W transect, no assessment of the vertical variability of N:P ratios was made due to insufficient valid DIN data.

#### DISCUSSION

In the Mediterranean Sea, the existent dataset of nM measurements of DIP at both spatial and temporal scales is severely scarce. The few reported nM concentrations of DIP are limited to two cross-basin cruises (PROSOPE cruise in 1999, Moutin et al., 2002: 7 values reported; Van Wambeke et al., 2002: 2 profiles reported and BOUM cruise in 2008: Talarmin et al., 2015, 5 profiles reported) and data acquired as part of the CYCLOPS project in the eastern basin of the Mediterranean Sea (Krom et al., 2005, 3 profiles reported). Two additional studies reported 1-year time series of surface nM DIP in the NW Mediterranean Sea at fixed observation stations: DYFAMED (Pulido-Villena et al., 2010) and ANTARES (Djaoudi et al., 2017). By reporting an important amount of vertical profiles along a longitudinal gradient, the present study is the first to extensively address the distribution of nM DIP in the P-depleted Mediterranean surface waters.

The nM DIP data obtained in this study are within the range reported in the above cited few previous studies, despite the use of different methods of analysis (**Table 1**). Regarding the longitudinal pattern of nM DIP, as previously reported by Moutin et al. (2002) for bioavailable DIP concentration in the upper layer (between 10 and 15 m), the present study shows a decreasing trend toward the East within the mixed layer with lower concentrations along the E-W transect (5.5 ± 3.6 nM) compared to the NWMS (12.8 ± 6.8 nM). This longitudinal gradient extends below the mixed layer (between the surface and 100 m), in contrast to Talarmin et al. (2015) study in which DIP concentrations (between the surface and 130 m) did not show any clear longitudinal trend crossing a W-E transect.

Regarding the vertical variability of surface DIP, the few existing data revealed a low but detectable increase of DIP concentration in the layer encompassing the DCM depth (Van Wambeke et al., 2002; Krom et al., 2005). In the present study, a significant vertical gradient of DIP but also of DOP was observed in both the NWMS and the E-W transect, allowing the generalization of this vertical pattern to the whole Mediterranean Sea. The positive gradient of DIP occurred above the phosphacline depth (Figure S1), suggesting that the phosphacline may be shallower than previously thought in both the NWMS and along the E-W transect. This result confirms that DIP in surface waters is not just a depleted invariant pool, highlighting the need to refine the criteria of the phosphacline determination.

The observed positive gradient of DIP, both in the NWMS and along the E-W transect, was higher than the bioavailable orthophosphate (<1–3 nM) reported by Moutin et al. (2002) in the upper waters (10–15 m). This means that the positive gradient of DIP observed in this study may be sufficient to sustain a biological activity in the surface waters, particularly by promoting microbial assemblages with very high-affinity transport systems and living very close to the theoretical diffusion limit for uptake. Indeed, the success of Synechococcus in the oligotrophic Mediterranean Sea has been linked to its ability to acquire phosphate even at low (<5 nM) concentrations (Moutin et al., 2002). However, the lack of biological data in this study precludes any deeper interpretation of the observed DIP gradient in a biological context.

The DIP concentration in surface waters would be the net resultant between biological and physical sources and sinks. The appearance of a positive gradient in upper waters suggests a balance in favor of DIP supply. This positive gradient of DIP could indeed arise from vertical phosphate diffusion from deep layers, suggesting the potential ability of new DIP to reach the upper waters even under stratification conditions in spite of an active biological uptake. The positive gradient of DIP above the phosphacline was, on average, higher in the NWMS than along the E-W transect, in agreement with reported differences in oligotrophy degree between the two sub-basins. Differences in DIP diffusion from deep layers, but also in rates of DOP hydrolysis could explain the observed differences in DIP gradient between the two cruises.

DOP data reported in this study are within the range of previous studies (**Table 1**), despite using different digestion methods (wet persulfate oxidation vs. UV digestion). Regarding the vertical profiles, this study revealed an important variability of DOP concentration in surface waters. Indeed, in the NWMS, a significant negative gradient of DOP was observed between the MLD and the DCM depth (av. 0.6 ± 0.7 nM m−<sup>1</sup> ), which could be the result of its potential bioavailability. High alkaline phosphatase activities have been reported in the NW Mediterranean basin (Van Wambeke et al., 2002; Thingstad and Mantoura, 2005), suggesting an active DOP hydrolysis of phospho-ester bonds which could explain the DOP decrease observed in our study. Interestingly, the negative DOP gradient coincided with the DIP positive gradient, suggesting a DOP utilization even under available DIP conditions. These findings could suggest a possible simultaneous biological uptake of both DIP and DOP in this layer. In a cross-basin cruise in the Mediterranean Sea, Van Wambeke et al. (2002) showed that heterotrophic prokaryotes switched from P to labile carbon limitation at depths where DIP was still undetectable by the conventional technique. Thus, another explanation for the observed DOP decreasing gradient at the DCM depth coincident with sufficiently available DIP is a possible limitation by organic carbon, based on the hypothesis that another role of alkaline phosphatase is to decompose dissolved organic matter (Nausch and Nausch, 2004; Luo et al., 2011). Therefore, if the diffusion is enough to maintain a DIP gradient in the nanomolar range and, thus, remove P limitation, then the observed DOP decrease may be a consequence of the shift from P to C limitation. Again, further studies combining nanomolar DIP and DOP measurements with biological data are needed to confirm the biological meaning of the observed gradients.

#### N:P Ratios

N:P ratios in the upper waters reported in this study showed a high variability in both inorganic and organic pools. This contrasts with the deep waters of the Mediterranean Sea in which N:P ratios are constrained to a narrow range of values, between 21 and 22 in the western basin, and somewhat higher in the eastern basin (Krom et al., 1991; Bethoux et al., 1992). The vertical profiles of DIN:DIP ratio exhibited the classical shape reported for the Mediterranean Sea surface waters, during the stratification period, with a marked increase of the DIN:DIP ratio in subsurface water (Diaz et al., 2001; Moutin and Raimbault, 2002; Van Wambeke et al., 2009). Based on the conventional definition of nutriclines, this increase would be due to a nitracline shallower than the phosphacline, as previously reported (Moutin and Raimbault, 2002; Van Wambeke et al., 2009; Pujo-Pay et al., 2011). In this study, the use of the sensitive LWCC technique for DIP analysis revealed that this increase of the N:P ratio occurred between the MLD and the DCM depth. As said previously, a positive gradient of nM DIP was recorded in this layer, indicating the existence of a higher gradient of DIN compared to DIP. This observation stresses the need to extend nanomolar measurements to DIN concentration for a better understanding of the controlling factors of DIN:DIP ratios variability as on other oceans, DIN also can be the first limiting nutrient (i.e., Gruber, 2008)

The DON:DOP molar ratios in the surface waters of both NWMS and along the E-W transect were higher than the corresponding DIN:DIP ratios pointing to a DOP depletion TABLE 1 | Summary of published data on nanomolar dissolved inorganic phosphate (DIP) and dissolved organic phosphate (DOP) concentration in the Mediterranean Sea.


Depth of the corresponding measurements as well as the methods used for both DIP analysis and DOP hydrolysis are presented. MAGIC and LWCC relate to the magnesium induced co-precipitation and the liquid waveguide capillary cell methods, respectively.

relative to DON. This could be the result of a preferential remineralization of DOP with depth as reported by Aminot and Kérouel (2004) in the Mediterranean Sea and by Letscher and Moore (2015) at the scale of the global ocean. Similarly to the DIN:DIP molar ratios, an increase of the DON:DOP was observed at the DCM depth. This finding could be related to a biological utilization of DOP, since in this layer a negative gradient of DOP was observed. Nevertheless, accurate measurement of DON, implying nM measurements of nitrogen, would help to a better constraining of the DON:DOP variability in upper waters.

#### CONCLUSION

This study reports the first cross-basin distribution of nM DIP and DOP in the P-depleted Mediterranean surface waters. Surface DIP and DOP concentration showed an important variability on both longitudinal and vertical axis. Vertical profiles of DIP exhibited a significant positive gradient above the phosphacline, suggesting the potential of new phosphate to reach the near-surface waters by diffusion even under stratified conditions, and giving evidence against the traditional view of a depleted invariant DIP pool in the surface Mediterranean Sea. Despite this evidence of sufficient P availability, a significant negative gradient of DOP concentration was observed between the mixed layer depth and the DCM. Finally, the positive gradient in DIP was accompanied by a concomitant increase in N:P ratios, suggesting the existence of a higher gradient in DIN compared to DIP. The results obtained in this study indicate that acquiring nanomolar DIP data is a sine qua non condition for the comprehension and prediction of the biogeochemical functioning of the P-depleted Mediterranean Sea.

# AUTHOR CONTRIBUTIONS

KD, EP-V, and FV contributed to the conception and design of the study. KD and TW did seawater sampling and sample conditioning during the MOOSE-GE and BioArgoMed cruises, respectively. KD and SH-N analyzed nutrient concentrations. VT, LC, and PT acquired hydrological data (CTD). KD organized the database and wrote the first draft of the manuscript. EP-V and FV wrote sections in the manuscript. LC and PT were chief scientists during the MOOSE-GE cruise. VT and FD were chief scientists during the BioArgoMed cruise. PR is the principal investigator of the MOOSE project. FD is the coordinator of the Mediterranean work package of the NAOS project. All authors contributed to the manuscript revision, read and approved the submitted version.

#### ACKNOWLEDGMENTS

The authors would like to thank the MOOSE project and the NAOS project (ANR J11R107-F, funded by the French Equipement d'Avenir) in the framework of which this study was performed. Captains and crews of the R/V le Suroit and the R/V le Tethys II are warmly acknowledged for their cooperative work at sea. The work leading to this publication has received funding from European FEDER Fund under project 1166-39417. This work is a contribution to the Labex OT-Med (n◦ ANR-11-LABX-0061) funded by the French Government

#### REFERENCES


Investissements d'Avenir program of the French National Research Agency (ANR) through the A∗MIDEX project (n◦ ANR-11-IDEX-0001-02). Two reviewers are acknowledged for their relevant comments and suggestions on the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00234/full#supplementary-material


predominance of the cyanobacterium Synechococcus? Limnol. Oceanogr. 47, 1562–1567. doi: 10.4319/lo.2002.47.5.1562


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Djaoudi, Van Wambeke, Coppola, D'Ortenzio, Helias-Nunige, Raimbault, Taillandier, Testor, Wagener and Pulido-Villena. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.