Comprehensive Modeling of Agrochemicals Biodegradation in Soil a Multidisciplinary Approach to Make Informed Choices to Protect Human Health and the Environment

Comprehensive modeling of agrochemicals biodegradation in soil A multidisciplinary approach to make informed choices to protect human health and the environment

Daniele la Cecilia

School of Civil Engineering Faculty of Engineering The University of Sydney

2019

A thesis submitted in fulﬁlment of the requirements for the degree of Doctor of Philosophy

Supervisor: Ass. Prof. Federico Maggi Auxiliary Supervisor: Prof. Chengwang Lei

Abstract

Numerical models are relied upon by risk assessors to predict the dynamics of potentially hazardous pesticides in soil. Those models may account for fundamental processes affecting pesticide dynamics, such as environmental and edaphic conditions, water flow, degradation, and sorption. However, those models lack the ability to account for complex biogeochemical and ecological feedbacks, and thus create challenges in achieving robust predictions. In particular, no attention has been paid on the coupled mechanistic description of microbial dynamics and soil organic matter cycling and the implications on agrochemicals biodegradation and soil and groundwater quality. This thesis aims to provide this description by developing a comprehensive framework through a multidisciplinary approach. Microbiological regulation of pesticide dynamics was investigated by coupling theoretical and numerical approaches with experiments carried out in our environmental laboratory or sourced from the literature. We propose the use of reaction networks to highlight the possibly multiple pesticide degradation pathways and the feedbacks with macronutrient cycles. Biochemically-similar pathways are mediated by a specific microbial functional group, which represents the microbial community carrying out particular functions; these functions are biodegradation of pesticides and metabolism of carbon-, nitrogen-, and phosphorus-containing molecules. We describe biochemical reactions by means of Michaelis-Menten-Monod (MMM) kinetics. Indeed, MMM parameters fully encompass the microbial life strategies including rapid growth, high affinity for substrates, or high substrate consumption efficiency. Michaelis-Menten terms allow us to include microbial competition for substrates, growth inhibition, and the memory-associated catabolite repression herein presented, which all can alter agrochemicals biodegradation effectiveness and macronutrients cycling. Because each biogeochemical process is mechanistically characterized in our approach, its uncertainty amd relevance can be quantified by means of sensitivity analyses. The latter are therefore crucial to explore the range of likely outcomes under a suite of scenarios, thus allowing risk managers to make informed decisions. We numerically show that a relatively small variability in MMM kinetic parameters and soil hydraulic parameters can result in large variability in agrochemicals environmental concentration. These results are in line with monitoring campaigns worldwide reporting agrochemicals accumulation in soil and water resources, despite currently-enforced first-order kinetic models predict quick and complete biodegradation. The proposed high-level process coupling introduced using a multidisciplinary approach is urged to develop sustainable plans in accordance with Nature-based strategies to cope with environmental changes and provide robust evidence to make informed choices.

Statement Originality

I, Daniele la Cecilia, hereby certify that to best of my knowledge, the intellectual content and writing embodied in this thesis are my own work and that any collaborations which contributed to this doctoral studies have been appropriately described and acknowledged.

I also certify that this thesis has not been submitted anywhere else for any degree or other purposes.

Name: Daniele Surname: la Cecilia

Signature

Date 25/02/2019

Authorship Attribution Statement

Chapter 1 of this thesis reﬂects views of the author. Chapter 2 Sections 2.2 and 2.5 of this thesis was submitted as a special issue1 to Mathematics and Computers in Simulation. I carried out the literature review and wrote the manuscript. Chapter 3 of this thesis presents laboratory experiments carried out by the author, which results have not been published in any format anywhere else. Chapter 4 of this thesis was published as research articles2,3,4 in Journal of Environmental Management, Journal of Contaminant Hydrology, and Environmental Pollution, respectively. I contributed to develop the reaction networks and I carried out the calibrations, analyzed the data, and wrote the manuscripts. Chapter 5 of this thesis was published as research articles3,5 in Journal of Contaminant Hydrology and Water Research. I carried out the numerical simulations, analyzed the data, and wrote the manuscripts. The source code of the BRTSim-v2 solver used to conduct the simulations was developed by the co-author, Ass. Prof. Maggi. Chapter 6 of this thesis was published as research article6 in Advances in Water Resources, conference paper7 in International Congress on Modelling and Simulation, and conference abstract8 in Computational Methods in Water Resources and submitted as special issue1 in Mathematics and Computers in Simulation. In 6,8 I contributed by explaining the results of the sensitivity analyses and writing both the manuscript and the conference abstract. The source code of the BRTSim-v2 solver used to conduct the simulations was developed by the co-author, Ass. Prof. Maggi, while the source code used to conduct the sensitivity analyses was developed by the author, Prof. Porta. In 1,7 I carried out the numerical simulations, analyzed the data, and wrote both the manuscript and the conference paper. The source code of the BRTSim-v2 solver used to conduct the simulations was developed by the co-author, Ass. Prof. Maggi, while the source code used to conduct the sensitivity analyses was developed by myself. Chapter 7 of this thesis was published as research article9 in Soil Biology & Biochemistry.I developed the kinetic framework, carried out the numerical simulations, analyzed the data, and wrote the manuscript. The source code of the BRTSim-v2 solver used to conduct the simulations was developed by the co-author, Ass. Prof. Maggi. List of publications: 1 la Cecilia, D. and Maggi. F (Under Review). Inﬂuential sources of uncertainty in glyphosate biochemical degradation in soil. Mathematics and Computers in Simulation. Manuscript Num- ber: MATCOM-D-18-00335. 2 la Cecilia, D. and Maggi, F. (2016). Kinetics of Atrazine, Deisopropylatrazine, and Deethylatrazine soil biodecomposers. Journal of Environmental Management, 183, pp. 673- 686, 10.1016/j.jenvman.2016.09.012. 3 la Cecilia, D. and Maggi, F. (2017). In-situ atrazine biodegradation dynamics in wheat (Triticum) crops under variable hydrologic regime. Journal of Contaminant Hydrology, 203, pp. 104-121, http://dx.doi.org/10.1016/j.jconhyd.2017.05.004.

vi 4 la Cecilia, D. and Maggi, F. (2018). Analysis of glyphosate degradation in a soil microcosm. Environmental Pollution. 233, pp. 201-207, https://doi.org/10.1016/j.envpol.2017.10.017. 5 la Cecilia, D., Tang, F.H., Coleman, N.V., Conoley, C., Vervoort, R.W., and Maggi, F. (2018). Glyphosate dispersion, degradation, and aquifer contamination in vineyards and wheat ﬁelds in the Po Valley, Italy. Water Research, 146, pp. 37-54, 10.1016/j.watres.2018.09.008. 6 Porta, G., la Cecilia, D., Guadagnini, A., and Maggi, F. (2018). Implications of uncertain biogeochemical parameters on a complex reaction network of atrazine biodegradation in soil. Advances in Water Resources, 121, pp. 494-498, 10.1016/j.advwatres.2018.08.002. 7 la Cecilia, D. and Maggi, F. (2017). Stochastic sensitivity analysis of glyphosate biochemical degradation. In Syme, G., Hatton MacDonald, D., Fulton, B. and Piantadosi, J. (eds) MODSIM2017, 22nd International Congress on Modelling and Simulation. Modelling and Simulation Society of Australia and New Zealand, December 2017, pp. 257 - 263, isbn 978-0-9872143-7-9, https://www.mssanz.org.au/modsim2017/B3/lacecilia.pdf. 8 la Cecilia, D, Porta, G., Riva, M., Vervoort, RW., Coleman, NV, Tang, FH, and Maggi, F. (2018). Propagation of ecohydrological uncertainty in a complex biogeochemical network of Glyphosate dispersion and degradation. Computational Methods in Water Resources (CMWR) XXII. Bridging gaps between data, models, and predictions, p. 154. 9 la Cecilia, D., Riley, WJ, and Maggi, F. (2019). Biochemical modeling of microbial memory eﬀects and catabolite repression on soil organic carbon compounds, Soil Biology & Biochemistry, 128, pp. 1 - 12, 10.1016/j.soilbio.2018.10.003.

vii Hereby, I (Daniele la Cecilia) confirm that I am the first and the corresponding author of the publications listed above, and that I have acknowledged the first and the corresponding author (Ass. Prof. Giovanni Michele Porta) of the research outputs delivered as a result of collaborative projects with the Politecnico di Milano.

Title: Mr. Surname: la Cecilia Name: Daniele

Signature

Date 25/02/2019

As supervisor for the candidature upon which this thesis is based, I can confirm that the authorship attribution statement above is correct.

Title: A/Prof Surname: Maggi Name: Federico Signature

Date 25 February 2019

viii

Acknowledgments

The scientific advancements achieved during the past 3 years have been embodied in this thesis. Many more words shall be written to pursue the overall aim to understand how Nature responds and copes with pollutants and man-made environmental changes. This thesis outlines experimental and numerical frameworks which have the potential to achieve such objectives. Those works were successfully completed thanks to the continuous professional and emotional support by a large number of people. I wish to begin with thanking my supervisor A/Prof Federico Maggi, who helped me to make the most out of these years by sharing with me his expertise in biogeochemistry, by providing me with a powerful reactive transport simulator, by improving my scientific writing, by involving me in the preparation of scientific proposals, by letting me attend national and international conferences, by introducing me to outstanding collaborators, and by trusting my capability to work in remote, which allowed me to visit international collaborators in Milano and Bolzano and spend fantastic quality-time with my girlfriend, family, and friends at home in Modena. I wish to explicitly thank A/Prof Nicholas Coleman and the whole lab team for their assistance in the microbiology laboratory, and for welcoming me at the lab meeting where ideas were shared, feedbacks were provided, and laughs were complimentary. I thank A/Prof Gio- vanni Porta, Prof Alberto Guadagnini, and Prof Monica Riva for hosting me in Milan twice and whose expertise in uncertainty propagation and sensitivity analyses were a perfect match with bioreactive transport models developed in these doctoral studies. Thank to Dr. William Riley for his valuable collaboration and expertise on numerical modeling of soil microbial dynamics. I express my gratitude to my progress review panel composed by Prof. John Patterson and Dr. Kapil Chauhan, whose feedbacks contributed to improve my research. Thank to my colleagues Dr. Fiona Tang, Chiara Pasut, and Ha Nguyen, who shared happy and sad moments, as well as large amounts of caffeine, with me during the PhD life. Thank to the postgraduate coordinators Daniela Entenmann and Jingping Wu for making our life less stressful and more cheerful. Thank to The University of Sydney for awarding me the scholarship to undertake these PhD studies and the travel grants, which allowed me to showcase my research and network with esteemed researchers. Thank to the anonymous reviewers who with their great and critical reviews improved my papers and therefore my research. Thank to the three anonymous reviewers who reviewed my PhD thesis; their comments contributed to the scientific value of this work and their feedbacks and perspectives encouraged me to keep up the good work. I wish to thank Dr. Nicoletta Leonardi, who I met when she was a passionate PhD student at Boston University; without knowing it, she inspired me to pursue such hard but rewarding path. I want to thank my previous supervisors, Stefano Della Chiesa, Prof. Sergio Fagherazzi, and A/Prof. Marco Toffolon for having been outstanding academic and real-life mentors. Finally, I acknowledge all my teachers from the elementary school to the high school, who used their expertise to challenge my strengths and filling my weaknesses (I still cannot draw though), and tailored their teaching methods to ease the learning process; I ever since realized that it was ashame to disappoint a caring "boss".

x Speaking of people who care, I will never ever be enough thankful to my family for their unconditional love, for their support, for the lessons taught, and for the soft skills they transferred to me. My mother with her creativity, my sister with her attention to details, and my father with his objective-orientated mentality. Thank you all for tolerating my curiosity for the unknown both when I asked "why" and when the "because" was very far away from home. My beloved grandparents, who taught me the importance of growing strong bonds, of reliability, of respect, of chivalry, of ethics, and of persistence. I am grateful to a long list of dear friends for keeping on enriching my life with beautiful memories and having an enormous positive impact in my life. Last but not least, Elisa, the fantastic woman who accepted to marry me on the shores of Lake Takepo, and therefore I will reveal my loving words in my marriage vows next year. Just to cite Pablo Neruda, "I want to do with you, what spring does with the cherry trees".

xi Contents 1 Introduction 1 1.1 Background and scope ...... 1 1.2 Research aims and objectives ...... 2 1.3 Thesis outline ...... 2 2 Literature review 5 2.1 Deﬁnition, uses, and classiﬁcations ...... 5 2.2 Dispersion processes and reactive transport modeling ...... 6 2.2.1 1-D movement of chemicals in soil ...... 7 2.2.2 Sorption processes ...... 9 2.2.3 Chemical and biological degradation ...... 10 2.3 Soil microorganisms: evolution, adaptation, survival, activity, and biogeochemical feedbacks ...... 13 2.4 Good modeling practices ...... 15 2.4.1 Uncertainty and sensitivity analyses ...... 16 2.4.2 Tailoring model development and output communication and visualization 19 2.4.3 Data share and content update ...... 20 2.5 Regulatory approach under the precautionary principle ...... 21 2.6 Summary ...... 22 3 Experimental approach 27 3.1 Introduction ...... 27 3.2 Methods ...... 27 3.2.1 Sampling site description ...... 27 3.2.2 Soil sampling procedure ...... 28 3.2.3 Enrichment cultures ...... 28 3.2.4 Analysis of GLP and AMPA by HPLC ...... 29 3.2.5 GLP and AMPA standards and calibration curves ...... 30 3.2.6 Media transfer ...... 31 3.2.7 Cultures isolation and test for GLP biodegradation ...... 31 3.3 Results and discussions ...... 32 3.3.1 GLP biodegradation dynamics: original enrichment cultures (T0) . . . . 32 3.3.2 GLP biodegradation dynamics: transferred cultures ...... 34 3.3.3 GLP and AMPA re-appearance ...... 35 3.3.4 Bacteria isolation ...... 36 3.4 Summary ...... 36 4 Mechanistic reaction networks development 41 4.1 Introduction ...... 41

xii 4.2 Method of parameters estimation ...... 41 4.3 Speciﬁc biomass aﬃnity ...... 42 4.4 Atrazine ...... 44 4.4.1 Introduction ...... 44 4.4.2 ATZ biodegradation pathways ...... 45 4.4.3 Results ...... 50 4.4.4 Discussion ...... 51 4.5 Glyphosate ...... 58 4.5.1 Introduction ...... 58 4.5.2 GLP biochemical degradation pathways ...... 58 4.5.3 Results ...... 63 4.5.4 Discussion ...... 64 4.6 Summary ...... 66 5 Modeling 67 5.1 Introduction ...... 67 5.2 Bioreactive Transport Simulator BRTSim ...... 67 5.3 Atrazine ...... 69 5.3.1 Introduction ...... 69 5.3.2 Methods ...... 69 5.3.3 Results and discussion ...... 73 5.4 Glyphosate ...... 83 5.4.1 Introduction ...... 83 5.4.2 Methods ...... 83 5.4.3 Results and Discussion ...... 90 5.5 Biochemical reaction and MMM kinetic parameters ...... 102 5.6 Summary ...... 104 6 Sensitivity analyses 105 6.1 Introduction ...... 105 6.2 Atrazine ...... 105 6.3 Glyphosate ...... 109 6.3.1 Reaction path model in a batch-type system ...... 109 6.3.2 1-D real-case scenario ...... 114 6.4 Summary ...... 118 7 Carbon consumption 119 7.1 Introduction ...... 119 7.2 Methods ...... 119 7.3 Results ...... 122 7.4 Discussion ...... 123

xiii 7.5 Summary ...... 133 8 Conclusions and perspectives 135 8.1 Conclusions ...... 135 8.2 Perspectives ...... 136 A Parameter estimation of ATZ reaction network 165 B Parameter estimation of GLP reaction network 171

xiv

List of Figures

1 Sources of uncertainty in mathematical modeling...... 24 2 Herbicides (re)approval process under the precaustionary principle ...... 25 3 Scheme of enrichment cultures preparation ...... 29 4 Scheme of sample preparation for HPLC analysis ...... 30 5 Calibration of GLP and AMPA measurements by HPLC ...... 31 6 Chromatograph of GLP and AMPA from HPLC ...... 32 7 Biodegradation of GLP 1 mM by soil microcosms ...... 34 8 Biodegradation of GLP 5 mM by soil microcosms ...... 35 9 GLP and AMPA biodegradation in transferred enrichment cultures ...... 38 10 Isolated strains from Cam-Cr3-GLP5-T1 ...... 39 11 Isolated strains from soil K3 GLP 5 mM ...... 39 12 Atrazine biochemical reaction network...... 54 13 Scatter plot of MMM parameters for ATZ, DIATZ, and DEATZ biodecomposition 55 14 Glyphosate degradation reaction network in soil...... 59 15 Atrazine reaction network in soil coupled with the N cycle ...... 77 16 Adsorption isotherms for ATZ, HOATZ, DIATZ, and DEATZ ...... 78 17 Hydrological boundary conditions at West Wyalong, NSW, Australia...... 78 18 ATZ partitioning and transport in soil ...... 79 19 Breakthrough curves of ATZ degradation time ...... 79 20 Breakthrough curves of ATZ biodegraded fraction ...... 79

21 Scatter plot between ATZ half-life and Φ0 ...... 80 22 ATZ eﬀect on soil pH ...... 80 23 Groundwater contamination by ATZ ...... 80 24 GLPmodeling1 ...... 84 25 Aadsorption isotherms for GLP and AMPA ...... 85 26 Month-of-the-year hydrological boundary conditions at Modena, Italy . . . . . 87 27 Daily hydrological boundary conditions and groundwater dynamics at Modena, Italy ...... 87 28 Average Hazard Quotient for GLP and AMPA ...... 91 29 Hazard Quotient for GLP and AMPA ...... 92 30 GLP and AMPA biodegradation eﬀectiveness ...... 93 31 GLP and AMPA partitioning and transport in soil ...... 93 32 GLP and AMPA soil residues ...... 94 33 GLP age and turnover time ...... 95 34 Breakthrough curves of GLP degradation time ...... 96 35 Breakthrough curves of GLP biodegraded fraction ...... 97

36 Eﬀect of boundary conditions and CH2O on GLP biodegradation ...... 98

37 Eﬀect of boundary conditions and CH2O on GLP biodegradation over time . . . 99 38 ATZ modeling scenario ...... 106

xvi 39 Sample probability density functions for ATZ, DIATZ, DEATZ, DIDEATZ, CLHOATZ, and their total mass ...... 107 40 Sample probability density functions for ATZhyd, ATZoxi, AER, and their total mass ...... 107 41 Principal sensitivity indices evaluated for ATZ, DIATZ, DEATZ, and DIDEATZ soil residues ...... 108 42 Analysis of variance associated with single and multiple reactions ...... 108 43 Frequency distribution of GLP residues ...... 112 44 Eﬀect of increasing variance on GLP residues ...... 112 45 Sensitivity measure of GLP residues ...... 113 46 GLP modeling scenario ...... 115 47 Time variation of biodegraded GLP ...... 115 48 Time variation of GLP concentration ...... 116 49 Global sensitivity analysis of exceedance time ...... 116 50 Time evolution of AMAE ...... 116 51 Time evolution of AMPA ...... 117 52 MMM-based kinetic schemes developed for CCR experiments ...... 127 53 MACR-C-based kinetic schemes developed for CCR experiments ...... 128 54 MACR-C modelling scenario ...... 128 55 Calibrations outcome for biomass-based MMM kinetics ...... 129 56 Calibrations outcome for enzyme-based MMM kinetics ...... 130 57 Calibrations outcome for biomass-based MACR-C kinetics ...... 131 58 Calibrations outcome for enzyme-based MACR-C kinetics ...... 131 59 Stochastic sensitivity analysis for enzyme-based MACR-C kinetics ...... 132 60 Sensitivity analyses of MACR-C response to varying nutrients bioavailability . 132 A1 Graphs showing calibration outcome of ATZ biodegradation along P1R1 . . . . 166 A2 Graphs showing calibration outcome of ATZ biodegradation along P1R1b . . . 166 A3 Graphs showing calibration outcome of ATZ biodegradation along P1R2 . . . . 167 A4 Graphs showing calibration outcome of ATZ biodegradation along P1R3 . . . . 167 A5 Graphs showing calibration outcome of ATZ biodegradation along P2R1 and P3R1 ...... 167 A6 Graphs showing calibration outcome of DIATZ biodegradation along P2R2 . . 168 A7 Graphs showing calibration outcome of DIATZ and DEATZ biodegradation along P2R2 and P3R2 ...... 168 A8 Graphs showing calibration outcome of CYA biodegradation along P4R1 . . . . 168 A9 Graphs showing calibration outcome of BIU biodegradation along P4R2 . . . . 169 A10 Graphs showing calibration outcome of ALP biodegradation along P4R3 . . . . 169 A11 Graphs showing calibration outcome of ACT biodegradation along P5 . . . . . 169 B1 Graphs showing calibration outcome of GLP biodegradation along P1R1 . . . . 171 B2 Graphs showing calibration outcome of GLP, AMPA, and MTH biodegradation 171

xvii B3 Graphs showing calibration outcome of MTH and SRC biodegradation . . . . . 172 B4 Graphs showing calibration outcome of SRC, MTH, and GLY biodegradation . 172 B5 Graphs showing calibration outcome of GLY biodegradation ...... 172 B6 Graphs showing calibration outcome of GLP and AMPA chemical degradation . 172

xviii

List of Tables

3 Calibration curve for GLP and AMPA measurements ...... 31 4 Average kinetic parameters for ATZ, DIATZ, and DEATZ biodegradation. . . . 53 5 MMM kinetic parameters for ATZ reaction network ...... 57 6 MMM kinetic parameters for GLP reaction network ...... 65 7 Estimated adsorption parameters for ATZ, HOATZ, DIATZ, and DEATZ . . . . 71 8 Soil and hydraulic parameters at West Wyalong, NSW, Australia ...... 73 9 ATZ biodegradation eﬀectiveness ...... 74 10 Numerical framework for the ATZ reaction network ...... 82 11 Estimated adsorption parameters for GLPa and AMPA ...... 85 12 Soil and hydraulic parameters at Modena, Italy ...... 86 13 Yearly cumulative ecohydrological ﬂuxes at Modena, Italy ...... 88 14 Glyphosate applications dose and date...... 89 15 Statistics of GLP disappearance from the top 10 cm of soil ...... 96 16 Numerical framework for the GLP reaction network ...... 103 17 Estimated kinetic parameters for each corresponding developed scheme . . . . 134

Acronyms

ACT Acetate ALP allophanate AMA sensitivity index developed by Aronne, Monica, Alberto AMAE AMA for Estimated value AMAE AMA for Variance AMAP AMA for Probability to exceed a safety threshold

AMAγ AMA for skewness AMPA aminomethylphosphonic acid ATZ atrazine B biomass mass within depth interval

BCcycle Non-repetitive time-series relative to an environmental variable BIU biuret BRZ Below Root Zone

BAER aerobic soil bacteria

BANAER anaerobic soil bacteria

BATZhyd atrazine hydrolizer bacteria

BATZhoxi atrazine oxidizer bacteria

BAOB ammonia oxidizer bacteria

BDEN denitriﬁer bacteria

BHyO glyphosate hydrolizer and oxidizer bacteria

BNOB nitriﬁer bacteria C Carbon

CH2O one-carbon carbon soource CLHOATZ 2-chloro-4-hydroxyl-6-amino-1,3,5-triazine CR Catabolite Repression CR-C Catabolite Repression for Carbon CYA cyanuric acid DEATZ deethylatrazine DEHA deethylhydroxyatrazine DHOATZ 2,4-dihydroxy-6-amino-1,3,5-triazine DHONATZ 2,4-dihydroxy-6-N-ethylamino-1,3,5-atrazine DIATZ deisopropylatrazine DIDEATZ deisopropyldeethylatrazine DIHOATZ deisopropylhydroxyatrazine EFSA European Food and Safety Authority EPA Environmental Protection Agency ETA ethylamine FAO Food and Agriculture Organization

xxii FOE First-order equation FRC Fructose GLP glyphosate GLX glyoxylate GLY glycine HOATZ hydroxyatrazine HPLC high-performance liquid chromatography IPA isopropylamine IPP isopropanol M solute mass within depth interval MACR-C Memory-Associated Catabolite Repression for Carbon MTH methylamine MM Michaelis-Menten MMM Michaelis-Menten-Monod N Nitrogen NIPA N-isopropylammelide OC organic carbon OXL Oxalate P Phosphorus PPP Plant Protection Product RO water Reversed Osmosis water RZ Root Zone SCC Succinate SM Sensitivity Measure SOM soil organic matter SRC sarcosine TCA Tricarboxylic acid TFA Triﬂuoroacetic acid TOC total organic carbon TsCl p-toluenesulphonyl chloride USEPA United States of America Environmental Protection Agency

xxiii Symbols

B Biomass concentration

BS l exponent to calculate correction factor due to soil moisture content

Bs Total mass of target microbial functional group b pore volume distribution index

Ce equilibrium solute concentration

Cthr threshold concentration D Diﬀusion coeﬃcient E Enzyme concentration ET evapotranspiration

ET0 potential evapotranspiration F mass ﬂux rate exiting a depth interval g gravity acceleration HQ Hazard Quotient Irr Irrigation J mass ﬂux K Michaelis-Menten constant k absolute permeability of soil material

Ka,Langmuir Langmuir adsorption constant

KC crop coeﬃcient

Kd linear sorption constant

Kd,Langmuir Langmuir desorption constant

Keq equilibrium constant kFOE disappearance rate constant

KI Inhibition constant

Klinear linear sorption constant

KOC organic carbon-water partition coeﬃcient

kr relative permeability M Memory signal concentration

Mn Total mass of target molecules NRMSE normalized root mean squared error percent P total pressure

PI inﬁltrating precipitation

Pc capillary pressure

PG gas pressure

Qe moles of adsorbed solute per unit mass of adsorbent ◦ Q10 factor quantifying increase in degradation rate following an increase in temperature by 10 C

qm,Langmuir Langmuir maximum adsorption constant R2 coeﬃcient of determination

xxiv R biodegraded mass rate along a pathway r first-order reaction rate constant rB bacteria recovery rate rE enzymatic reaction rate constant S Sulphur S Solute concentration s stoichiometric coefficient S e effective saturation

S l,REF reference soil moisture T temperature

TREF reference temperature

Tt turnover time t time tˆ exceedance time t1/2 half life U Sobol’ index Y biomass yield coeﬃcient z soil depth

αS l correction factor for soil moisture

αT correction factor for temperature δ cell mortality rate

δE enzyme degradation rate

δM memory signal degradation rate Φ specific biomass affinity Φ0 specific biomass affinity per unit biomass concentration µ reaction rate constant

µD dynamic viscosity

Ψs air entry potential at saturation ρ density σ standard deviation ad adsorbed solute aq aqueous solute

NP non-preferred substrate

P Preferred substrate

xxv 1. Introduction

1.1. Background and scope

In the recent decades, the human population has been growing at an unprecedented rate. Food producers believed that the growing demand for food and fibers could not have been met by means of traditional farming practices, which required a continuous and relentless land management. Instead, newly-developed synthetic molecules have provided an effective and cheap alternative. As a result, the agrochemical industry has dramatically changed after the Green Revolution. Agricultural lands have been oversimplified by reducing biodiversity; this substantially decreased plants and soils resilience to cope with stressors. This scenario was fertile ground for an escalation in the use of agrochemicals targeting specific crop issues. In particular, plant protection products (PPPs) are used for protecting crops health against diseases and in- festations. There exist more than 1300 PPPs active ingredients worldwide and a total of nearly 4.1 millions of tonnes were applied in 2016 according to the last data presented by the Food and Agricultural Organization (FAO) (FAO, 2013). However, PPPs may pose a threat to human health and the environment because of their potential ecotoxicity and persistence. As a consequence, effort has been put to understand the predominant processes driving PPPs dynamics in natural systems after their applications onto agricultural soils. In parallel, numerical models are developed to simulate those processes and predict PPPs environmental concentrations for given scenarios. These numerical models find many applications. Usually, they are relied upon by risk management officers appointed by regulatory bodies to assess the environmental sustainability of active ingredients, who may or may not allow the use of such molecules. This thesis contributes to deepen the understanding of ecohydrological and soil biogeochemical feedbacks on PPPs environmental dynamics. This is achieved by proposing a different mathematical framework to describe PPPs degradation by microorganisms in soil. The framework is suitable to account for microbial responses to PPPs exposure and the feedbacks with soil biogeochemical processes. Indeed, biochemical degradation is effective only if suitable biogeochemical conditions are met. Not without reason, bacteria themselves invest resources to modify their surroundings. It is of uttermost important to capture these multifaceted interactions, which require to ask ourselves: What are the processes occurring in pristine and contaminated lands in relation to PPPs dynamics? Are they all relevant? Are they mathematically correctly described? Are the parameter values representative of local conditions? The proposed framework is numerically tested under real-case scenarios for the herbicides atrazine (ATZ) and glyphosate (GLP) to quantify mass flows along different degradation pathways and estimate environmental concentrations of ATZ, GLP, and their corresponding metabolites. Not only these herbicides are amongst the most used active ingredients in the world, but also produce toxic and persistent metabolites. Finally, sensitivity analyses relative to the available scenarios were run to identify driving processes and explore variability in the ATZ and GLP models. It is worth noting that an increasing body of literature reports interactions, sometimes

1 detrimental, amongst PPPs pollution, microbial activity, and soil organic matter (SOM) cycling. Carbon (C), nitrogen (N), and phosphorus (P) are crucial elements in the environment; when correctly balanced they promote sustainability and fertility, otherwise pollution. To add an additional piece to a more robust description of SOM cycling, a 60-year-old knowledge amongst microbiologists is presented to environmental modellers, who may have not explored it, yet. The innovative framework captures bacteria strategies to thrive in a sometimes predictable environment given previous growth conditions. This thesis can be a benchmark for future models aiming at incorporating and mechanistically describe the soil microbiota-plants nexus in the contexts of but not limited to sustainability and remediation.

1.2. Research aims and objectives

This research aims to mechanistically characterize the biochemical reactions relative to the degradation of the herbicides ATZ and GLP and their metabolites. The first objective is to describe degradation processes using Michaelis-Menten (MM) kinetics for chemical degradation and MMM kinetics for biological degradation. This objective was achieved by retrieving laboratory experiments showing molecules degradation over time. The second objective is to develop reaction networks relative to ATZ and GLP biochemical degradation, and characterize the corresponding kinetics. This was accomplished by retrieving suitable laboratory experiments from the literature showing biochemical degradation over time and identifying the produced metabolites. The third objective is to assess ATZ and GLP dynamics under real-case scenarios. For this objective, the ATZ and GLP reaction networks were coupled with a simplified description of C and N cycling in soil and were integrated within a bioreactive transport simulator. The fourth objective is to assess the robustness of the models. This was achieved by propa- gating uncertainties and carrying out sensitivity analyses on the developed models. The fifth objective is to propose an additional inhibitory metabolic process, which contributes to explain C-containing substrates consumption by microorganisms in soil. This was achieved by retrieving laboratory experiments from the literature, which were used to test several hypothesis, and eventually, to formulate the innovative mechanistic framework.

1.3. Thesis outline

This thesis is composed of 8 chapters and 2 appendices. Chapter 1 provides a brief introduction to the overall thesis. Chapter 2 presents a literature review about the main processes driving pesticides fate in the environment and the numerical approaches and limits to the simulation of those processes. A brief overview is also given on the PPPs (re)approval process in European Countries. Finally, the chapter summarizes the major knowledge gaps in PPPs interactions with soil microbiota. Chapter 3 introduces to laboratory experiments used to isolate GLP biodegraders from soil samples and to broaden the understanding of microbial responses to contam-

2 inants. Chapter 4 shows the estimations relative to biodegradation of ATZ and GLP and their metabolites. The chapter also describes the developed reaction networks. Chapter 5 presents the numerical results obtained for ATZ and GLP biodegradation and dispersion under real-case scenarios. Chapter 6 shows the sensitivity analyses carried out to assess parameters uncertainty relative to ATZ and GLP biodegradation and dispersion under real-case scenarios. Chapter 7 introduces to the newly developed framework of microbial preferential substrate consumption and memory of previous growth conditions. Chapter 8 gathers the major achievements in these doctoral studies. The chapter also outlines future research needs to improve the robustness of current environmental models used to predict PPPs dynamics in soil and their feedbacks on soil biogeochemistry. Appendices A contains the graphs showing the parameter estimation outcomes of ATZ reaction network. Appendices B contains the graphs showing the parameter estimation outcomes of GLP reaction network.

2. Literature review

This chapter introduces to the most important characteristics of PPPs, which aﬀect their fate in the environment, the numerical frameworks available to model solutes reactive transport, and good modeling practices aiming at exploring simulation outcomes variability. The focus is on the governing processes occurring in the soil matrix including transport, sorption, biochemical degradation, and key microbiological concepts relevant to PPPs biodegradation, while fundamental processes, either underinvestigated or not integrated in numerical solvers, are acknowledged. Although the reviewed processes generally apply to all PPPs, most of the references cite studies on the molecules chosen in this doctoral research, viz the herbicides ATZ and GLP.

2.1. Deﬁnition, uses, and classiﬁcations

The United Nations Food and Agricultural Organization (FAO) defines herbicides as any substance or mixture of substances intended for preventing, destroying or controlling any [...] unwanted species of plants [...] causing harm during or otherwise interfering with the production, processing, storage, transport or marketing of food, agricultural commodities, wood and wood products or animal feedstuffs [...] (FAO, 2003). Trends in herbicides production show a steady mass production increase since 1950 (Tilman et al., 2002), and after 1970 herbicides have become the most used PPPs worldwide (Agrios, 2005). The cropping system strongly influence the herbicide use pattern (Bos et al., 1995) in terms of active ingredients used and application rate, frequency, and timing. Moreover, given recent biotechnological advances (e.g., Roundup Ready© crops), plants can be genetically modified to become resistant to specific herbicides, which may promote a greater herbicides use (Coupe & Capel, 2016). Herbicides can be grouped according to different characteristics including the chemical class and the mode of action. Amongst some of the most applied herbicides in agriculture (FAO, 2013), (1) 2,4D is a phenoxy-carboxylic acids, it belongs to the synthetic auxins, which mimic the action of an hormone naturally produced by plants, thus interfering with plants normal growth (Tu et al., 2001), (2) glyphosate is a glycine-like molecule, it belongs to the inhibitors of enolpyruvlshikimate-3-phosphate (EPSP) synthase, which prevents the formation of vital aromatic aminoacids (Schönbrunn et al., 2001), and atrazine is a trazine, it belongs to inhibitors of the photosystem II (Purcell et al., 1990). Each group may be most suitable to use in particular periods of the agricultural season (pre-sowing or post-emergence applications). Herbicides may be either selective, viz they target particular species of broadleaf and grass weeds, while saving the crop, or they may indiscriminately suppress all the vulnerable plants; in this latter case they are known as knock-down products, and glyphosate is an example. Herbicides can also be distinguished between those that are foliar or soil applied, and they usually reflect the mechanism of chemicals uptake by plants, which would occur via leaves adsorption or via roots and shoots, respectively (UCDavis, 2015). The mechanism of application may be related to the site of action of the herbicide. To

5 intervene on crops in post-emergence conditions, a foliar solution might be preferred because the chemical would be applied directly onto the target unwanted weed leaves reducing soil and crops contamination to a minimum. Whereas soon after crops harvesting, a soil solution might be employed in order to maintain clear the agricultural plots for the necessary period of time. The type of application may strongly determine the fate of the herbicide in the environment. In case foliar solutions are being used, chemicals may volatilize or be drifted by winds for long distances in the atmosphere from where they can precipitate onto otherwise uncontaminated geographical locations (Goolsby et al., 1997). Again, it may be washed from leaves during extreme rainfall events (Martin, 1977), and a part of it may end up in soil, while the other may be transported by runoﬀ. Soil-applied herbicides can be separated further in two classes depending on the chemicals phase, which can be either solid or aqueous. In regard with soil processes, herbicides may eventually be distinguished based on their persistence. The so called "residual herbicides" usually refers to persistent chemicals applied in soils soon after seeds sowing, which allows for a long-lasting protection against unwanted weeds (UCDavis, 2015). The likelihood that those residuals can pose a threat to soil and groundwater pollution is high given their intrinsic characteristic to remain active in the environment for long times.

2.2. Dispersion processes and reactive transport modeling

A wide range of coupled and highly nonlinear processes influence herbicide dynamics in soil. Once the physics of a phenomenon is understood, its dynamics can be predicted by means of mathematical equations. Reactive transport models provide therefore a fundamental tool to explore herbicide dynamics under different biogeochemical conditions (Maillard et al., 2016), ecohydrological (la Cecilia & Maggi, 2017a; la Cecilia et al., 2018a), climatic and land management scenarios (Steffens et al., 2015), and to answer environmentally-relevant questions including, for example, nutrients cycling (Riley et al., 2014), environmental bioremediation (Bao et al., 2014), and natural attenuation (Mayer et al., 2001). Those studies could be carried out thank to the full coupling of processes studied by different disciplines (Li et al., 2017). Simi- larly, biochemical processes affecting herbicides degradation are recently being mechanistically distinguished in models (la Cecilia & Maggi, 2017a, 2018; Wang et al., 2016). There exist some recognized important processes that must be accounted for predicting herbicides environmental concentrations, such as water movement, chemicals sorption, biogeochemical transformation and degradation, and metabolites production. At the soil surface, air drift (Silva et al., 2017), runoff (Lefrancq et al., 2017), and photodegradation (Katagi, 2004) are other important processes, of which some pose a risk of long-range pesticides transport far from the source points, while others enhance pesticides removal. In soil, herbicides may be taken up or released by plants through their roots system (Laitinen et al., 2007), ingested by earthworms (Tejada et al., 2016), biodegraded by fungi (Singh, 2006), as well as chemically degraded by mineral particles (Paudel et al., 2015; Li et al., 2015), organic matter (Khan. S.U., 1978), and

6 extracellular enzymes (Fragoeiro & Magan, 2005). However, soil is a complex media. Trans- port can be increased in the presence of preferential flows (e.g., through macropores, cracks, fractures, etc...) (Eguchi & Hasegawa, 2008) or slowed down by bacterial processes (Maggi & Porporato, 2007) and biofilm formation (Volk et al., 2016). Transport and pesticides sorption can be modified by surfactants (Mobbs et al., 2012), which are active molecules largely used in the agricultural sector. Numerical models are continuously developed to encompass the increasing number of relevant processes as well as to couple different reaction networks with the aim to capture feedbacks on the modeled system. This is important because transformed and degraded substances may not only contribute to environmental pollution but also have different physical-chemical properties from the original molecule, and therefore, they would show different dynamics under the same conditions. Moreover, molecules toxicity to specific microbial populations may inhibit pesticides biodegradation (See Section 3.3) and cause detrimental effects at the ecosystem level. For instance, pathogenic fungi prevailed over beneficial ones after glyphosate applications (Rosenbaum et al., 2014), while sulfonylurea herbicides substantially impaired the soil nitrogen cycle (Rose et al., 2016). The next sections present the most used approaches and mathematical equations to simulate water transport, sorption, and biochemical degradation.

2.2.1. 1-D movement of chemicals in soil

Numerical codes have been developed to simulate the transport of chemicals in their gaseous and aqueous phases by water and air in porous soils. Those codes may work under specific assumptions, which were made to simulate particular systems; generally, the more complex the system to describe the more detailed the code. For example, groundwater flow models often assume that the vapor behaves as an ideal gas, which may hold true for agricultural purposes. Yet, water displacement by gas-saturated soil pores maybe relevant in agricultural settings but may not be incorporated in groundwater flow models. A good tradeoff between simplicity and efficacy in agricultural settings is provided by codes that simulate two-phase (liquid and gas), two-component (water and air) systems. The soil matrix would be the third component representing the third phase (solid); the properties of the latter phase are sometimes assumed not to change over time. The prediction of movement of both gaseous and aqueous chemical components is fundamental to describe biological processes and herbicide fate in soil. One approach is to compute gas and liquid fluxes; next, the n chemical components undergo diffusion and advection according to the calculated fluxes. In this thesis, I refer to dispersion as the coupling of these two transport processes. Solutes diffusive fluxes can be described using the Fick’s law written as

n n n Jβ = −Dβ∇S β (1)

7 where J is the flux of the chemical component n in its phase β, S is the concentration, and D is the diffusion coefficient. Solutes advection fluxes can be described using the Darcy’s law written as

n krβ Jβ = −k ∇P + ρβg∇z (2) µDβ where k is the absolute permeability of the soil material, krβ is the relative permeability of the phase β, µDβ is the dynamic viscosity, P is the total pressure given by the sum of the liquid e capillary pressure Pc (Pc is a function of the eﬀective saturation S β and can be calculated using the model by Brooks & Corey (1962), by Van Genuchten (1980), or others) plus the gas pressure

PG (calculated under the ideal gas assumption) plus possibly other addends such as the osmotic pressure, ρ is the density, g is the gravity acceleration, and z is the soil depth relative to a reference system. Aqueous and gaseous components may be partitioned into primary and secondary ones (Lichtner, 1985) to reduce computational costs. Under this approach, the user identiﬁes the primary components as the building blocks of the biogeochemical system of interest; these components can undergo equilibrium and kinetic reactions. Instead, concentrations of important secondary components are calculated using the mass action law assuming thermodynamic equilibrium with primary ones as

Keq(T) S s = (3) βSEC QnPRI snPRI 1 S βnPRI

s where S βSEC is the concentration of the secondary component SEC in its phase β with stoichiom- s etry coefficient , Keq(T) is the equilibrium constant at temperature T, nPRI is the number of snPRI primary components with which the secondary one is at thermodynamic equilibrium, and S βnPRI s is the concentration of the primary component in its phase β with stoichiometry coefficient nPRI. The prediction of water movement in agricultural plots can be challenging in some situations. Topographic variability, lateral flows and interactions with surface waters, soil properties heterogeneity, plant physiology, and peculiar land and water management operations can largely influence water dynamics. Suitable models have been developed such as the Agricultural Pro- duction Systems Simulator (APSIM, Keating et al. 2003). This simulator also accounts for water runoff at the soil surface. This process must be accounted for when soils may experience flooded conditions as a result of crop and land management practices and environmental conditions (e.g., rice grown in paddy or flooded fields, flood irrigation, intense rainfall on drylands, etc...). Indeed, a predominant fraction of pesticides are usually found within the top 5 cm of soil, and they are therefore susceptible to be washed away through runoff. Soil properties heterogeneities and preferential pathways contribute to increased solutes transport (Šim˚uneka et al., 2003). In fact, water can flow more quickly through macropores and cracks in soil (Eguchi & Hasegawa, 2008), and can therefore transport greater amounts of solutes including pesticides. Due to the high complexity of the soil matrix, it is difficult to accurately simulate pesticides

8 transport by means of preferential ﬂows. Approaches to integrate this process rely on either more or less complex mathematical models or sensitivity analyses. In the former case, dual porosity models can be chosen to account for diﬀerences in soil matrix permeability (Gerke & Van Genuchten, 1993); other conceptual models were comprehensively reviewed in Šim˚uneka et al. (2003). In the latter case, the soil hydraulic properties can be changed in a number of simulations n, which can be used to identify worst and best case scenarios, for example, in terms of pesticides concentration in groundwater (la Cecilia et al., 2018b).

2.2.2. Sorption processes

Solutes including herbicides can undergo sorption onto the soil organic matter and mineral surfaces, biological surfaces (Tejada et al., 2016), and absorbed into extracellular polymeric substances (Flemming, 1999; Decho & Gutierrez, 2017). Sorption has great impacts on herbicides transport and availability to undergo biogeochemical processes. Strong adsorption reduce dissolved herbicides leaching, and therefore, groundwater contamination. However, strongly bound herbicides may result in a lower activity as they cannot easily migrate into weeds via the root system. Herbicide residues at the surface can strongly contribute to the risk of contaminated spray drift (Silva et al., 2017), and runoff and leaching as macroaggregates through soil macropores (Lefrancq et al., 2017). Yet, bioavailability to micro- and macro-organisms may be reduced, and it becomes more and more difficult to access aged adsorbed herbicides and nutrients in general (Lerch et al., 2009). Though, biodegradation of adsorbed herbicides have been suggested to occur either directly or by facilitating their bioavailability (Park et al., 2003). As an advantage, strong adsorption decrease herbicides toxicity to soil organisms because the latter would be exposed to lower concentrations. Finally, adsorption is fundamental to start chemical reactions (Paudel et al., 2015; Li et al., 2015) as the molecule must first make contact with the reactive sites available at the surface of soil organic matter and mineral particles. In contrast, weak adsorption may result in high herbicides leaching risk, whereas their activity toward weeds may not be affected. Yet, the risk for air spray drift, runoff, and transport of macroaggregates may be reduced as herbicides would not be likely found at the soil surface or bound to particles. Bioavailability would not be a issue in this case, but toxicity may be have profound consequences on soil ecosystem and functioning (Rose et al., 2016). There exist different mathematical descriptions of solute sorption onto solid surfaces, which have also been applied to soil organic matter and mineral particles. A solute Saq in its aqueous phase is in a dynamic equilibrium with its adsorbed phase Saq as a function of temperature (Atkins & De Paula, 2005). The simplest mathematical description is the linear function (e.g., Hedges, 1977; Cantrell et al., 2002) written as

S ad = Klinear × S aq (4)

where Klinear is the adsorption linear coeﬃcient. However, saturation capacity is not accounted for with this mathematical description; hence, the equation may work well at low adsorbent

9 concentration, but may predict unrealistic concentrations of adsorbed solutes. Therefore, nonlinear formulations were developed to more realistically describe solute sorption. For example, despite inherent limitations of mathematical models, the Langmuir model accounts for a maximum adsorption per unit mass (qm,Langmuir) and adsorption (ka,Langmuir) and desorption (kd,Langmuir) rates as (Langmuir, 1918; Atkins & De Paula, 2005)

dS ad = k × S × (q − S ) − k × S (5) dt a,Langmuir aq m,Langmuir ad d,Langmuir ad

Sorption remains a very uncertain process in numerical simulations (FOCUS, 2000). Sorp- tion models that are computationally fast and can be applied across a wide range of scenarios with minimal location-speciﬁc calibration may not account for long-term increase of adsorption of aged residues (Green & Karickhoﬀ, 1990) as well as fundamental dynamic biogeochemical feedbacks among the molecule characteristics, soil conditions (e.g., soil parent material, soil mineral composition, soil organic matter, pH, reactivity of metal oxides, cation exchange capacity, etc.), and land management practices (legacy of applied amendments such as phosphate fertilizers) (Vereecken, 2005).

2.2.3. Chemical and biological degradation

The speciﬁc chemical stability of a molecule determines its tendency to be degraded physically, chemically, and biologically. Amongst these three mechanisms, photodegradation in soil is usually negligible as light intensity needed to break chemical bonds cannot penetrate through the soil (USEPA, 1987), while it may be relevant soon after foliar applications (Bos et al., 1995). Chemical and biological degradation reactions are relevant mechanisms for herbicides removal in soils (Torstensson, 1980). Herbicides degradation may be completely described by using twelve biologically-mediated reactions, which include hydroxylation, dealkylation, ring ﬁssion, hydrolysis, oxidation, and decarboxylation; the latter three reactions can also be chemically mediated. This thesis focused on herbicide degradation by minerals and microorganisms, but it is acknowledged that a fundamental role can also be played by humic acids (Khan. S.U., 1978), fungi (Krzysko-Łupicka´ & Orlik, 1997), and macroorganisms (e.g., earthworms, Kersante et al. 2006 and Tejada et al. 2016). Chemical degradation is catalyzed by reactive sites provided by organic matter and mineral particles. Organic matter can play an important role in transforming and degrading herbicides. In particular, dissolved organic matter contains carboxylic acids, which are highly reactive with positively charged herbicides (Khan. S.U., 1978). Therefore, this pathway may be fundamental to herbicide degradation in agricultural soils rich in organic acids. Mineral particles can also play an important role in degrading herbicides. In particular, highly reactive manganese oxides, such as birnessite, can fast degrade the potentially toxic herbicide glyphosate to non toxic end products such as phosphate, ammonium, and one-carbon substrates (Paudel et al., 2015; Li et al., 2015). However, dissolved metals in soil such as Cu2+ can compete with herbicides for catalytic sites, therefore limiting degradation rates (Barrett & McBride, 2005). Note that metals

10 are largely available in agricultural soils due to soil management practices for crop protection (e.g., copper sulphate) and yield optimization. Chemical degradation is often mathematically modeled by means of ﬁrst-order equations (FOE) or Michaelis-Menten (MM) kinetics. FOE are written as

1 dS (t) i = kFOE × S (t) (6) s dt i i

th where the subscript i refers to the generic i substrate with concentration S and stoichiometric FOE th coefficient s, ki is the disappearance rate constant relative to the i substrate, and t is time. MM kinetics are written as 1 dS i(t) S i(t) = µi (7) s dt S i(t) + Ki −1 th where µi (s ) is the reaction rate constant relative to the i substrate and Ki (M) is the half- saturation constant, or MM constant. Note that, MM kinetics approximate FOE at low substrate concentration as S i(t) µi lim µi ' × S i(t) S i(t)→0 S i(t) + Ki Ki However, the MM framework is theoretically valid when substrate concentration is not limiting the reaction (i.e., [S i(t)] >> Ki). Notwithstanding, the assumption is usually disregarded in numerical models and the MM equation is used without changes. Biological degradation is carried out by adapted microorganisms, which possess the necessary enzymes to break herbicides down. Bacteria can actively take up herbicide inside the cell where either specific or non specific enzymes mediate the catalytic reaction. The former are enzymes with a narrow substrate affinity and the chemical structure of the herbicide fits the active site. Specific enzymes are often the result of bacterial evolution following long exposure to a substrate. In contrast, non specific enzymes are characterized by active sites with broad substrate affinity; that is, the active site can fit a suite of substrates. It follows that specific enzymes allow for effective degradation of the corresponding molecule, while non specific enzymes may allow for fortuitous degradation of newly synthesized molecules. However, uptake may be the limiting step for herbicide biodegradation; Tang & Riley (2013) have proposed a mathematical formulation to account for the uptake of molecules with applications in soil biogeochemistry. Once herbicides (and their metabolites) are taken up, suitable intracellular enzymes may degrade these chemicals along either cometabolic or metabolic reaction pathways. In the former case, the reaction proceeds in the presence of a necessary co-substrate (e.g., an additional energy source, electron donor/acceptor, etc...). In the latter case, the co-substrate is not needed, and the molecule can be used as a source of energy, nutrients, or both. Note that not all taken up molecules are biodegraded (See Section 3.3.3). Of the multiple mathematical formulations to simulate microbial biodegradation dynamics, FOE or pseudo-first order equations are often preferred. However, these approaches implicitly assume that the concentration of bacteria actively degrading the contaminant is constant, thus they neglect microbial biomass growth (Pagel et al., 2014). Hence, a powerful approach is to couple MM kinetics with the Monod growth model to account for microbial biomass dynamics.

11 Considering a simple system with one substrate, the MMM kinetic equations can be written as

1 dS (t) S (t) B(t) = µ × × (8) s dt S (t) + K Y dB(t) 1 dS (t) = × Y − δB(t) (9) dt s dt where B(t) is the microbial biomass concentration (mg-wet-Biomass L−1), Y (mg-wet-Biomass moles-Substrate−1) is the biomass yield coefficient, and δ (s−1) is the cell mortality rate. The system of equations indicates that the herbicide degradation rate changes with the biomass concentration; the latter increases over time as bacteria consume their energy and nutrient source, while decreases due to cells mortality. Low K values predict that the overall reaction rate would be optimal already at low substrate concentration and it thus varies substantially within a narrow range of low concentration values but little otherwise. In contrast, for high K values the optimal overall rate would be achieved only at high substrate concentration and it thus changes largely over time at physiological environmental concentrations. The biomass yield coefficient is usually characteristic given both the bacteria strain and the substrate because different strains may more or less efficiently grow on particular substrates under identical laboratory conditions. Although MMM kinetic parameters are considered to be constant for environmental modeling purposes, they may change consequently to many factors including microbial maintenance (van Bodegom, 2007) and enhanced biodegradation (Krutz et al., 2008).

Environmental factors such as soil temperature and soil moisture can substantially aﬀect microbial activity (Dechesne et al., 2014), and therefore, the overall biodegradation rate dS /dt as in Eqs. 7 and 8. Several mathematical functions are available to describe the above mentioned feedbacks. One option to correct the overall biodegradation rate dS /dt for the temperature eﬀect is to multiply it by a correction factor αT (Bennett, 1984) written as

◦ (T−TREF)/10 C αT = Q10 where Q10 is a factor usually assumed to be equal to 2 for microbial biodegradation after Suárez

(2005), T is the simulated temperature, TREF is the reference temperature TREF at which the reaction rate (kFOE or µ) was measured. One option to correct the overall biodegradation rate dS /dt for the soil moisture is to multiply by a correction factor αS l , calculated using the Walker equation (Walker, 1974), written as

!BS l S l αS l = S l,REF where S l is the soil moisture, S l,REF is the soil moisture at which the degradation rate was measured (i.e., ﬁeld capacity), and BS l is an exponent taken equal to 0.7 after Gottesbüren, B. (1991).

Bacteria and fungi also release extracellular enzymes mediating herbicides degradation (Park et al., 2003; Fragoeiro & Magan, 2005). This degradation pathway could be predicted

12 but with high uncertainty because the fate of enzymes in the soil matrix is still quite underinvestigated.

2.3. Soil microorganisms: evolution, adaptation, survival, activity, and biogeochemical feedbacks

Bacteria are highly complex living organisms despite their micro size. They have survived for millions of years in harsh environments and changing conditions. They have gained a reper- toire of strategies to either adapt to the surrounding environment or reshape it to their benefit. Eventually, bacteria have co-evolved with the surrounding environment to gain advantages from environmental periodic variability and to cope with its randomness. Can we predict the feedbacks between herbicides persistence and soil microbiology and functioning? The predominance of microbial populations change periodically; some communities take over others as environmental conditions change. One driver is the availability of nutrients in soil (Zheng et al., 2018). A very interesting experiment carried out in hydroponics, measured the community changes driven by availability of specific root exudates released by the plant over different growth stages (Zhalnina et al., 2018). Root exudates can also affect the production of extracellular enzymes (Shi et al., 2018). It can be inferred that the soil microbiological component has fine-tuned its development to make the most out of the available resources. These observations together are fundamental as they showed that the SOM cycling below-ground is connected with ground and above-ground processes.

Herbicides have the potential to disrupt those fine-tuned interactions amongst different environmental compartments by inhibiting susceptible bacteria. For instance, bacteria using the shikimate pathway to produce essential aromatic aminoacids are affected by glyphosate because it inhibits the EPSPS enzyme (Schönbrunn et al., 2001). However, both the soil matrix and commensalism within microbial communities may supply the needed substances, thus alleviat- ing inhibitory or detrimental consequences (Nielsen et al., 2018). Otherwise, herbicides may provide an additional source of nutrients for some specific population. For example, glyphosate promoted the growth of fungal species, which took over other microbial populations, thus affecting soil functioning (Kremer & Means, 2009). These examples may explain the contradictory findings relative to the small and temporary versus detrimental and persistent toxic effects to soil bacteria (Rose et al., 2016). Bacteria have evolved different strategies to survive against toxic chemicals. One of them is not to let toxic chemicals entering the cell. Neither the chemical can diffuse in the cell through the membrane nor through active transporters. This protection mechanism stresses the importance to account for herbicide uptake in risk models. A second strategy is to excrete the taken up toxic chemical by means of efflux pumps (Adebusuyi et al., 2012). Yet, Kurenbach et al. (2015) showed that the simultaneous availability of herbicides and antibiotics activated the efflux pumps. This mechanism has important implications in the context of water economy

13 because polluted agricultural water could substantially reduce herbicide biodegradation effectiveness and promote multidrug resistance. A third strategy is to transform the toxic chemical into a non-toxic metabolite. For instance, Shushkova et al. (2016) reported that Achromobacter sp. Kg 16 can acetylate glyphosate to acetyl-glyphosate as a detoxification mechanism; the metabolite was not metabolized further. A fourth strategy is provided through natural selection by which microbes lacking of susceptible target sites can survive and transfer the beneficial trait to daughter cells. The fifth strategy can be very important as it implies the formation of biofilm after exposure to a pollutant as a defense mechanism. The biofilm may contain biogenic nano oxide minerals (Toner et al., 2005). It may be speculated that also glyphosate - more generally pesticides - induces this response after Kremer & Means (2009) observed an increase in manganese-oxidizers EPS-producers bacteria in soil after exposure to glyphosate. On the one hand, this mechanism may contribute to decrease pollution because abiotic degradation of many organic pollutants is mediated by oxides. On the other hand, it may also contribute to modify bacteria metabolic requirements (Wan et al., 2018), soil biogeochemistry, and air and water movement through the soil matrix. The sixth possibility, and to the best of my knowledge also the last one, is biodegradation. Bacteria may possess/evolve specific and/or non-specific enzymes that mediate the breaking down of the herbicide or increase the production of the degrading enzyme (Sviridov et al., 2012).

Adapt bacteria may eventually use herbicides as nutrients and energy sources. For this reason, metabolic regulatory mechanisms may eventually apply to herbicides too. For example, glyphosate can be used as a phosphorus source (Pipke & Amrhein, 1988b); the authors also showed that the availability of the bacterial most preferred phosphorus source (i.e., phosphate) inhibits the pathways through which microorganisms can scavenge phosphorus from glyphosate. It is possible that similar regulatory mechanisms apply to the other nutrient sources including carbon (Dijkhuizen et al., 1980; Mukherjee & Ghosh, 1987), nitrogen (Farrell et al., 2011), and sulfur (Kertesz et al., 1994). Such biological regulations are fundamental for a balanced growth at the individual cell level and for building metabolic co-dependencies at the community level, which ultimately contribute to individual cell survival. The selection pressure experienced by soil bacteria can also affect their response to exposure. From an ecosystem perspective, the range of values of the MMM parameter µ, K, and Y are likely a consequence of bacterial life strategy determined by the referred to as "survival triad" (Panikov, 2010). From field and laboratory observations, it was clear that soil bacteria could be identified as either µ or K strategists as different metabolic strategies to benefit from available resources, where µ implies high reaction rates and K implies high substrate affinity. The same grouping can be applied for bacteria found in pristine and anthropic environments (Tang et al., 2019). Bacteria adapted to high-herbicide concentration conditions are more likely µ strategists, so that they can quickly consume the large availability of nutrients. Yet, these bacteria would be able to survive otherwise possibly toxic concentrations. Vice versa, low herbicide concentrations more likely select for K strategists, so that they can scavenge the little available resources. These strategies have important repercussions in terms of biodegradation efficiency and can be used to

14 predict the implications of higher or lower herbicide application rates than the usually applied rate. In fact, low herbicide application rates where µ strategists are present might result in poor biodegradation because of the poor enzyme aﬃnity for the substrate. Similar results might be found for high application rates where K strategists are; in this case, the aﬃne enzyme would be saturated and would not be able to process the substrate excess. Furthermore, bacteria selected under low herbicide concentrations may not survive under higher ones (See Section 3.3).

2.4. Good modeling practices

Reactive transport models may not account for important feedbacks to the model output. It is important that experts from different disciplines communicate to improve the mechanistic description of process-based models. By differentiating the individual process contribution to the overall system behavior, it is possible to make more robust prediction of herbicides dynamics, to design targeted mitigation and bioremediation strategies, and to develop data-driven regulatory limits. The modeler may find that one process can be described by a suite of solvers (Figure 1a, yellow panel). Selection of the most suitable solver may be achieved based on expert judgment, who assesses the mathematical modeling and knowledge gaps, and by carrying out preliminary analyses to test model robustness against available observations, as well as to rank parameters contribution to outcome variability. Later, the selected models each describing a single process are integrated in order to solve biogeochemical reactions coupled with hydrological forcing, at the requested detail over space and time. In case more than one model can adequately describe one process, then several multi-models may be framed and each multi-model will be separately validated. Also, each single model should be independently characterized using specific observations to quantitatively allocate the contribution of single processes to a target outcome, such as herbicide concentration (Figure 1). This would allow to increase confidence in the formulated multi-model. Some code perform better in analyzing certain systems because they can account for processes neglected or poorly implemented in other models. The user should pay great attention to choose the most suitable code depending on the task to accomplish. In specific, if microbial processes largely depend on nutrients availability and these two aspects together substantially drive degradation pathways, then codes for simulating reactive transport by means of MMM kinetics such as CrunchFlow (Steefel & Lasaga, 1994), Geochemist’s Work- bench (Bethke, 1996), Phreeqc (Parkhurst, D. L. and Appelo, C. A. J., 1999), PECCAD (Pagel et al., 2014), and BRTSim (Maggi, 2015) may be preferred. Other highly-developed simulators, but implementing first-order reactions to describe solutes degradation are SWAT (Arnold et al., 1998), TOUGHREACT (Xu et al., 2011), and HYDRUS (Langergraber & Šim˚unek,2005). For reliability, models require to be calibrated and validated and different approaches can be used. For example, the available dataset (e.g., herbicide concentration at some locations and over time) is separated in two or more sets: the calibration set is used to calibrate the parameters and to assess the correctness of the model structure, while the calibrated model will be used to

15 predict the observations contained in the validation set. For this, field surveys to measure herbicide levels and soil characteristics are necessary. However, these studies can be expensive and time-consuming; because of the lack of resources to carry out extensive monitoring campaigns, field data are usually poor in spatial and temporal resolution. Thus, modelers usually apply biochemical reaction developed under controlled conditions and couple them with hydrological boundary conditions, which are more likely to have been measured and validated. Modeled herbicide concentrations may then be compared with values reported at sites with similar meteorological and hydrological conditions. After model validation, simulations are run to meet the objectives and requirements from the stakeholders. Typical outputs show the concentrations of the herbicide and its metabolites over space and time, partitioning of the molecules in their aqueous, adsorbed, and gaseous phases, and the molecules degradation effectiveness. At this stage, a good modeling practice is to carry out both uncertainty and sensitivity analyses (Figure 1). The former quantifies the model confidence in terms of output variability and can be represented using probability density functions (pdf ). The latter ranks the contribution of each parameter input to the output variability, thus identifying and ranking dominant processes in the model system. In other words, uncertainty analyses can provide policy makers and stakeholders with a quantitative decisional tool for herbicides approval, while sensitivity analyses can provide a wider insight into sustainable land management planning.

2.4.1. Uncertainty and sensitivity analyses

The modeler may be able to simulate the predominant interacting processes driving herbicides fate in soil, as introduced in the previous sections of this chapter. However, those interactions may vary spatially and temporally, while additional processes contributing to the regulation of herbicide dynamics may only be brought to light at a future time. Simulation of complex systems surely need to be endowed with uncertainty and sensitivity analyses to rank and account for multiple sources of uncertainty and errors in data collection, parameter values estimation, and model structure, which usually result in nonlinear model responses and unforeseen outcomes (Dai et al., 2018; Refsgaard et al., 2007; Saltelli & Tarantola, 2002; Zhao et al., 2011). Other benefits from using this approach regard improvements in model structure resulting in a more robust or simplified model than the one previously conceived, a better understanding of the system, and the capability to design effective land management plans and bioremediation strategies (Dai et al., 2018; Mamy et al., 2005; Muñoz-Carpena et al., 2009; Refsgaard et al., 2007; Zhao et al., 2011). Given the important role numerical models can play, their confidence should be consistently evaluated (Bennett et al., 2013). Sources of uncertainty Uncertainty may refer to error in laboratory procedures, parameter estimation in the calibration phase, and lack of knowledge of all the possible processes and their spatial and temporal variability in the modeling phase (Refsgaard et al., 2007). Examples regard the lack of detailed quantitative description of microbial processes (e.g., microbial dynamics affected by varying

16 environmental conditions and nutrients availability Ermakova et al. 2017; Tang et al. 2019, or exposure to exogenous and endogenous stressors such as additional toxic molecules Adebusuyi et al. 2012; Kurenbach et al. 2015, etc.) or the relationships within microbial communities that may affect microbial activity towards an environmental objective (Smith et al., 2005; Smith & Crowley, 2006). Uncertainties will result in variability of modeling outcome and deviation from expectation may be large. To account for laboratory uncertainties, experiments are usually carried out in triplicates, and results are reported with their standard deviation. For example, the output of a herbicide biodegradation experiment is sketched in Figure 1a as Process 1, where concentrations are monitored over time. In this case, the modeler would choose some kinetic model to describe the observed average concentrations to estimate the model parameters. The reported variability in measured concentrations may be taken into account in the estimation procedure. The typical approach for parameter estimation is by inverse problem solution, where parameter values are fine-tuned by minimizing the error between observations and predicted values. This numerical procedure allows to calculate some calibration statics, such as parameter uncertainties and cross correlation amongst parameters. These statistics already provide an indication about the robustness of the model. Single reactions integrated in biochemical networks or each parameter part of the equation may contribute to output uncertainty to different extents. Each parameter is inherently associated with a probability distribution, which may be assumed based on the statistics generated after parameter estimation. In this thesis, it has been assumed that kinetic parameters can follow several distributions including Gaussian (la Cecilia & Maggi, 2017b) and uniform (Porta et al., 2018). Gaussian distributions may well represent laboratory studies where bacteria achieve similar kinetic performances, while uniform ones may well encompass environmental variability. MMM parameter values relative to one particular herbicide can be largely different for different microbial strains tested both under different experimental conditions (a review of atrazine biodegradation experiments was presented in la Cecilia & Maggi (2016), while a review of glyphosate biodegradation experiments can be found in Tang et al. (2019)) and similar ones (See experiments relative to glyphosate biodegradation in Section 3.3). While local conditions may select for microorganisms within a small range of parameter values variability, no explicit studies have investigated this in real agricultural conditions. More studies are needed to enhance knowledge and provide reasonable assumptions of natural systems to overcome this incapability of prediction. When coupling biogeochemical processes with others, boundary conditions may strongly affect the expected outcome. These driving forces include, but are not limited to, meteorological and hydrological conditions, changes in land use and land management, and spatial variability in soil characteristics. If more processes are deemed fundamental to accurately describe one natural system, the uncertainty associated with the model structure increases, which should be thoroughly investigated. Model uncertainty assessment: methods and insights The practice of model uncertainty assessment is becoming more important over time (Fer- retti et al., 2016). Many model uncertainty assessment techniques exist and are presented and

17 reviewed in Refsgaard et al. (2007) and Razavi & Gupta (2015). Those techniques can be applied to assess reactive transport processes. Once the biogeochemical model is calibrated, and possibly validated, a distribution is assigned to input parameters, from which parameter values are extracted using some technique. Random sampling is one option, but advanced sampling methods (Campolongo et al., 2007, 2011) allow to adequately sample the input parameter space, thus allowing to reduce the number of simulations needed to obtain a robust outcome in terms of a deﬁned model target output(s) (e.g., concentration of herbicide or microbial biomass). Note that, cross-correlation amongst parameters should be speciﬁed and accounted for in the sampling.

Different approaches to perform sensitivity analyses should be followed depending on the model output space. Monotone spaces, or simply ordered datasets, may be assessed using differential analysis. With this technique, the modeler calculates multiple model outcomes from a small neighborhood of the input parameters (Local sensitivity analysis). Input values are generally varied one at a time so that the partial derivative of the output with respect to the input can be calculated. Because this approach resorts to small perturbations of input parameter values, exceptional results may remain unpredicted; hence, Local sensitivity analysis are not reliable for studying complex output spaces. Model systems characterized by nonlinear responses and when the modeler aims to investigate the model’s entire input space, then a Global sensitivity analysis is preferred. Usually, this technique explore perturbations of input parameters using a Monte Carlo analysis followed by variance-based methods to identify the most influential parameter(s). The influence of one or more parameters on the variance of the output can be quantified through Sobol’s indices (Sobol’, 1993), AMAV index (Dell’Oca et al., 2017), the Fourier Amplitude Sensitivity Test (Cukier et al., 1973), and others. An advantage of the family of AMA indices (Dell’Oca et al., 2017) is that they allow to quantify parameter(s) influence on other moments (e.g., average, skewness, etc...) of the output distribution (Porta et al., 2018). Bioreactive transport systems may contain many uncertain parameters. To decrease computational time, it could be therefore convenient to carry out two-steps sensitivity analyses. In the first phase, referred to as parameter screening phase, modelers identify and neglect those parameters with low influence to the output. It is common to resort to a differential analysis because this technique is less time-consuming and generates less data than a Monte Carlo analysis, but it may miss to identify extraordinary results. The second step consists of the realization of the variance based global sensitivity analysis on the predominant parameters.

In most real-case modeling applications, the modeler assembles multiple single models to develop a multimodel system. For example, when biochemical reaction networks are coupled with hydrological models. Global sensitivity analyses allow one to rank each process to outcome variability and to assess the correctness of model structure in cases where predictions can be compared against observations. Because a suite of solvers may be used to describe the same process, uncertainty analyses are useful to identify the most appropriate solver for each process. The correctness of the model structure can be assessed by means of the process sensitivity index proposed by Dai et al. (2018) and the Framework for Understanding Structural Errors applied

18 in Borgonovo et al. (2017). Finally, sensitivity analyses may allow one to determine the set of parameters that if optimized would minimize herbicides concentration at a speciﬁc location. The sensitivity of the model output with respect to particularly important parameters may require further work. In case parameter variability results in a wide range of possible outcomes, then the modeler may want to carry out additional investigations with the aim to reduce the parameter uncertainty. On the contrary, a narrow range of possible outcomes may induce the modeler to simplify the model. Model simpliﬁcation can be achieved by reducing the number of redundant or negligible parameters or by creating a surrogate model. A surrogate model is a simple mathematical function, typically a polynomial, that approximate the response of the numerical model given the input, within a prescribed tolerance. Both methods would result in a simpler system, less computationally demanding and time-consuming.

2.4.2. Tailoring model development and output communication and visualization

Multiple classes of stakeholders such as farmers, policy makers, and the public audience are potentially involved in the process of herbicide formulation, (re)approval, use, and monitoring. Modelers can potentially collaborate with stakeholders in each process to build robust and accurate mechanistic representations of herbicide dynamics in the environment (Mamy et al., 2005; EFSA et al., 2018). However, a robust model can be complex to be understood by each stakeholder, and what is more, they may seek different technical support from the modeler. Therefore, the modeler should aim to tailor the communication of model formulation and output to address stakeholders knowledge and inquiries. For example, farmers may be interested in practical advice on what herbicide to apply, when, and at what rate to guarantee crop protection while avoiding exceeding maximum herbicide concentration in food. Land management officers may be interested in comparing different crop management plans to find the most advantageous one considering multiple objectives including health, sustainability, and environmental protection. Environmental Protection Agencies (EPAs) worldwide may be interested in building an efficient monitoring network and may use modeling to localize the most sensitive and informative sites where to collect data on herbicide environmental levels; these sites may provide an early warning in case of contamination. Numerical solvers have thus become more user-friendly as a result of the close collaboration between software developers and end-users. Typical improvements regard:

• Ease of change inputs, boundary conditions, settings, and parameters;

• Ease of integrate additional processes aﬀecting the reaction network, which can be achieved by developing an open source software or by providing technical support;

• Availability of clear software documentation to provide insights on mathematical modeling, hence allowing to understand the conﬁdence and the validity range of the model, and to

19 resolve unforeseen predictions or to explain mismatch between predicted and observed environmental concentrations;

• Presentation of model outputs in informative manner both textually and graphically to address users inquiries.

Information about herbicide concentration in the environment and their effect to human health and ecosystem services is vast but often contradictory. Stakeholders may therefore be interested in clear, explicit, simple, concise, informative, and comprehensive documents reporting predicted concentrations under prescribed environmental scenarios. Comparative numerical analyses may be included to capture the implications of scenario alternatives, while uncertainties may be difficult to communicate. Agencies are putting effort in trying to overcome this issue by promoting discussions amongst experts in the sector.

2.4.3. Data share and content update

Because of the complexity in investigating soil processes, it is desirable that information available to distinguish and mechanistically describe the interaction of each process with herbicide dynamics would be shared. For example:

• Digitalized soil properties (https://www.isric.org) allow one to account for nutrients feedback on microbial activity and estimate soil hydraulic parameters;

• Toxicological values (i.e., inhibitory concentrations) can be very important to forecast bacterial survival, healthy soil functioning, and optimal land management. Values relative to microorganisms are sparsely available in the literature, but they usually show large variability across diﬀerent geographical and environmental settings. In fact, toxicity may depend on history exposure and may be alleviated by the presence of biodegraders together with susceptible microorganisms. The USEPA-managed database ECOTOXnet (USEPA, 2016) provides toxicology information relative to terrestrial and aquatic living organisms including crops, orchards, earthworms, and other organisms, but not on microbes;

• Biochemical reaction networks and their corresponding kinetic parameters should be kept up-to-date. Interested parties can be universities, research centers, and private consultants because both pursue interest in discovering new biochemical mechanisms and developing new strategies to optimize some desired process. Current comprehensive information can beneﬁt farmers, policy makers, and the public audience. While degradation pathways can be predicted (EAWAG, 2019), a greater sharing of kinetic data for biological reactions (e.g., la Cecilia & Maggi (2016, 2017a, 2018)) is key to the success of continuous development of mathematical models;

• Additional fundamental data which feed environmental models. While the state of California is a virtuous example for the public sharing of detailed data relative to pesticides application (https://calpip.cdpr.ca.gov/infodocs.cfm?page=aboutpur), such data are

20 most of the time diﬃcult to access and process. The USGS (USGS, 2017) provided such data for The United States aggregated by state until 2009. Farmers in Europe must keep records relative to crop protection operations but for privacy reasons, data are aggregated by country before being disseminated (EC Regulation 1107/2009, 2009). Moreover, active ingredients are additionally aggregated by pesticide class. The Food and Agricul- ture Organization of the United Nations (FAO) disseminates yearly amounts at the global level, grouped by pesticide class (FAO, 2013).

Fortunately, the current trend showed an increasing eﬀort and willingness in sharing valuable data, and the World Soil Information program is one of the best example (https://www. isric.org), where researchers worldwide develop digital soil maps to share as they are available. Hopefully, trust between research institutes and farmers, the actual land caretakers, will foster new collaborations (Della Chiesa et al., 2019). In fact, herbicide residues in soil are also the result of land management operations and crop history, which might be recorded too (Steﬀens et al., 2015), but often not shared with modelers.

2.5. Regulatory approach under the precautionary principle

"All things are poison, and nothing is without poison, the dosage alone makes it so a thing is not a poison" (Paracelsus, 1965). All chemical substances inevitably present some level of toxicity to humans and the environment. Toxicity typically follows a dose-response relationship in an organism, that is, increasing the level of exposure will result in an increase of the undesired effect. I believe that in a similar manner, and by their own nature, herbicides at concentrations above certain thresholds can cause health and environmental issues. Safety pesticides environmental concentrations must safeguard consumers, non-target organisms, and the environment as a whole as it provides fundamental services. Precautionary measures must be taken because humans develop and input in the environment newly synthesized molecules at an unprecedented rapid pace; not all the impacts of those molecules neither as single active ingredients nor as mixtures can be foreseen, but the need for sustainable practices in the farming sector has been advocated for quite a long time now (Tilman et al., 2002). Because herbicides may have unforeseen effects to human health and ecosystem services (Rose et al., 2016), modeling can make a large positive difference where countries, such as those in the European Union, follow precautionary principles, stating that in the absence of scientific evidence about safety, one situation can pose a risk. In contrast, other countries usually wait for evidence of harm before applying stricter regulations. In the former case, the herbicide (re)approval process can be more tedious, it is generally multi-step and iterative, and it involves Research and Development (R&D) centers, designated experts of concerned countries, safety authorities, public audience, and policy makers (Figure 2). The process is described under the Regulation (EC) No 1107/2009 of the European Parliament and of the Council of 21 October 2009 concerning the placing of plant protection products on the market and repealing

21 Council Directives 79/117/EEC and 91/414/EEC (http://data.europa.eu/eli/reg/2009/ 1107/2014-06-30). R&D laboratories formulate new herbicides and collect preliminary data about their biogeochemical characteristics under controlled conditions, in the laboratory and in the ﬁeld (Figure 2). Next, modeling of molecule dynamics under prescribed scenarios is carried out. The results are documented and submitted to risk assessors, who evaluate the completeness of the data provided, carry out their own risk assessments, and, in cooperation with other stakeholders, produce a peer reviewed report to be submitted to regulatory bodies. The general public can access available documents on herbicides and have an opinion, which may have a role in questioning the licensing of herbicides (Figure 2). Yet, all organizations and authorities, at national and international level, may contribute by providing an additional portfolio of evidence with regard to the herbicide under assessment. The Europe Union provides an exemplary framework where researchers can actively contribute towards the safety of citizens and the environment. Scientiﬁc evidence is collected to develop regulations stating the maximum acceptable concentrations of known contaminants in water resources (EC Directive 2006/118/EC, 2006), air (EC Directive 2008/50/EC, 2008), and food (EC Regulation 396/2005, 2005). In case the residues exceed safety thresholds they may suggest to deliberate more stringent thresholds and to adopt more sustainable management practices aiming at reducing the residues in the workplace and along the food chain. It is not rare that the European Commission reassesses the approval of herbicides, which themselves or their metabolites are found to persist or be highly toxic in the environment (91/414/EEC, 2004; 2007/629/EC, 2007). However, surprisingly, no safety limits exist in the context of soil contamination. In the meanwhile old, and sometimes banned, molecules can still be found in soil (Silva et al., 2019).

2.6. Summary

This chapter reviewed the main processes controlling herbicides dynamics in soil and it introduced to underinvestigated microbial responses to herbicides exposure. These processes were described by means of mathematical equations so that their dynamics can be forecast. Though a few major knowledge gaps for a robust modeling of herbicides dynamics emerged in this literature review: • Microbiological reactions are not generally well implemented in numerical solvers. While soil water movement and solute sorption are generally described by a number of parameters, biological reactions are generally described by only one parameter. Instead, MMM kinetics would provide a ﬂexible framework, which allows to fully capture microbial dynamics and the feedbacks between herbicides and soil biogeochemistry;

• There is no equation to predict the MMM parameter values and biodegradation pathways as a function of exposure history. Instead, the rate of biodegradation reactions may increase up to a maximum over time and biodegradation pathways producing less toxic molecules may become more important;

22 • Herbicides data accessibility is an issue. Legacy data relative to herbicide applications are not easily accessible, if not lacking, the few available databases may not be easily combined, and consistent long-term monitoring plans relative to soil residues have not been implemented. The paucity of such fundamental input data hinders the development of robust predictions of herbicides environmental concentrations.

• Models used for herbicide (re)approval lack to account for fundamental interactions amongst herbicides, soil microbiology, and soil nutrients. Missing these interactions may result in erroneous forecasts, which would bias herbicide (re)approval, with potentially detrimental consequences to all the environmental spheres.

The following chapters investigate the identiﬁed knowledge gaps. In particular, the different, and sometimes unexpected, response of soil microcosms to glyphosate is highlighted by means of laboratory experiments in chapter 3, the complex degradation pathways of atrazine and glyphosate are developed in chapter 4, while the advantages of using MMM kinetics to better quantify herbicides biodegradability under real-case scenarios, to highlight switches in degradation pathways under varying environmental conditions, and to underpin microbial response to herbicides exposure and nutrients availability are shown by means of numerical simulations in chapters 5,6, and 7.

23 26/04/2018

(a) Uncertainty Sensitivity Model Parameter Parameter number analysis analysis Data Model Selection Estimation Model 1 2 … n Model 1_1 1_1 Process 1 Model 1_2 1_2 Model 1_… 1_… Model 1_m 1_m

Process 2

Process …

Model Parameter Parameter number Estimation Data Model Selection Model 1 2 … n Model p_1 p_1 Process p Model p_2 p_2 Model p_… p_… Model p_m p_m

(b) Model Structure Driving Processes

Single Model Multi Model Future Scenarios Policy Scenarios

1_... 1_... p_1 1_m 1_... p_n Climate 1_m p_1 Hydrology Land Management 1_m p_n Natural Variability

(c)

Uncertainty Analyses

Data Scenarios Sensitivity Analyses Model structure

Figure 1: Sketch of good modeling practices, which nearly follow the structure of this thesis, where application of (a) is shown in chapter 4, (b) is in chapter 5, and (c) is in chapter 6. (a) Sketches of observed data for different processes over time (left gray boxes). Multiple numerical models may be used to describe the same process (yellow 1 boxes). Models may contain a different number of parameters that should be estimated. Parameter uncertainty analyses carried out for each model assess the confidence of the model in reproducing the data (green boxes). Parameter sensitivity analyses assess the contribution of each parameter variability to the model output (right gray boxes). (b) The most appropriate models selected to describe each process are coupled together to develop a multi- model. Modeling scenarios include other driving processes; simulations are run. (c) Uncertainty analyses show the probability density functions of the likely outcome due to variability in parameters. Sensitivity analyses allocate the sources of uncertainty amongst the components of the model structure.

24 Policy Makers Role (Re)evaluate available evidence to make an informed decision R&D Modelling Role Molecule development Data collection (lab and Outcome Yes/Not (Re)Approve

field) molecule (Re)Approval Modelling Prediction of molecule dynamics under Public and Associations prescribed scenarios Role Read available evidence Outcome Report to have an opinion Modelling Outcome Reports and express concerns in participatory Food and Safety Centres Services for Farmers and Land events Environment Officers Role Initial Assessment. Monitor emergence Use needed molecules Monitor emergence contamination in following recommended contamination in environment practices humans, food, and Establish dose- feedstuffs response relationships Establish dose-response to identify safety relationships to identify thresholds safety thresholds Modelling Prediction of molecule Optimization Support in-situ best land dynamics under monitoring networks management practices prescribed scenarios Predict herbicide Actions to reduce degradation and humans and livestock dispersion exposure by reducing Planning activity for residues in harvest protection, exposure reduction, and remediation Outcome Reports and Guidelines Reports and Reports Guidelines

Figure 2: Solid black lines represent the possible interactions amongst stakeholders during the process for herbicides approval, while dashed black lines indicate interactions after approval. Scheme assuming adoption of precautionary principle, that is, preliminary information concerning herbicides safety must be available. Names, roles, and actions of some identiﬁed stakeholders are indicated, as well as the corresponding potential use of numerical modeling to support actions.

3. Experimental approach

3.1. Introduction

This chapter presents the procedure followed in these doctoral studies for gaining insights both regarding microbes adaptation to the herbicide GLP in terms of degradation pathways and corresponding kinetics and for isolating the corresponding GLP biodegraders enriched from agricultural soil samples. This chapter may be of interest and applicability to a broad audience including environmental modelers, soil microbiologists and ecologists. The chapter did not receive formal peer review but it was intended to provide a forum for discussion of the novelties emerged in the experiments. As a consequence, statements in this chapter reﬂect views of the author. The laboratory experiments were planned with the supervision of Ass. Prof. Nicholas V. Coleman at The University of Sydney.

3.2. Methods

3.2.1. Sampling site description

Soil samples were collected from a total of 7 locations within the agricultural research fields managed by the University of Sydney. Sampling sites were chosen with the assistance of the corresponding land managers, Mr. Peter Bell (Camden site) and Mr. John Bell (Narrabri site). The soil samples mirror different history exposure to the herbicide GLP, vegetation cover, land management operations, and soil and climatic conditions. Four samples were collected nearby the Plant Breeding Institute in Camden, NSW, Australia (34◦01’05.0”S; 150◦39’49.1”E), on August the 29th 2018. The sample sites were the same as those used by the colleagues in Tang et al. (2019) for reproducibility. Samples were named Cam-Iw (Camden-Irrigation line wet, as it was on the runoff pathway), Cam-Id (Camden-Irrigation line dry, as it was not on the runoff pathway), Cam-Cr (Camden-Crop), Cam-Pa (Camden-Kikuyu Pasture). Cam-Iw soil was collected on the runoff pathway in the irrigation line next to a cropping field where GLP (Roundup Ultra Max, 900 g-a.i. ha−1) was boom-sprayed once every three months for four years, with the last application dated May the 30th 2018, hence nearly 2 months before soil sampling. Cam-Id soil differed from the former because it was chosen a consistently dry area outside the runoff pathway of the irrigation line. Cam-Cr soil was collected in the cereal-clover cropping field where GLP (Roundup Ultra Max, 900 g-a.i. ha−1) was applied once every two years and was last sprayed on May the 31st 2017, hence nearly 15 months before soil sampling. Cam-Pa soil was collected in kikuyu grass pasture, which had no GLP application in the last ten years. However, the area is in the vicinity of managed croplands, and therefore, may had been affected by GLP spray drift. Three more samples were collected within the Watson Grains Research Centre located in

27 Narrabri, NSW, Australia (30◦ 16’ 22”S; 149◦ 48’ 24” E), on September the 26th 2018. Sam- ples were named Nar-Ba (Narrabri-Bare soil), Nar-Iw (Narrabri-Irrigation line wet), Nar-Wh (Narrabri-Wheat covered). Replicates were not available for Nar-Iw and Nar-Wh. Nar-Ba soil received a GLP application 3 days before sampling. No information on the formulation was provided. No information on GLP application timing was provided for Nar-Iw and Nar-Wh soils. Note that, the Narrabri site was experiencing a severe drought, in which rainfall had not been substantial for the previous 6 months.

3.2.2. Soil sampling procedure

One (for Nar-Iw and Nar-Wh soils) or three (other soils) 50 mL samples from the top 5 cm of soil were collected at each location by directly scooping up the sample into sterile polypropylene Falcon tubes. Replicas were spaced approximately 50 cm between each other. The samples were stored at 4 ◦C until the preparation of enrichment cultures. To identify each replica, the samples were labeled as [location]-[soil condition][number replica] (e.g., Cam-Iw1).

3.2.3. Enrichment cultures

Enrichment cultures were set up to have GLP as the only C source. First, 250 mL glass bottles containing either ∼ 48 mL (for the 1 mM test) or ∼ 42 mL (for the 5 mM test) of Minimal

Salt Media (MSM, for 1 L of deionized water: 2.27 g of K2HPO4, 0.95 g of KH2PO4, 0.67 g of (NH4)2SO4, and 2 mL of metals solution, which results in a 7.0 ± 0.2 pH media; 1 L of the trace metals solution contained: 6.37 g of Na2EDTA · 2 H2O, 1.0 g of ZnSO4 · 7 H2O, 0.5 g of CaCl2 · 2 H2O, 2.5 g of FeSO4 · 7 H2O, 0.1 g of NaMoO4 · 2 H2O, 0.1 g of CuSO4 · 5 H2O,

0.2 g of CoCl2 · 6 H2O, 0.52 g of MnSO4 · H2O, and 60 g of MgSO4 · 7 H2O) were autoclaved. MSM was chosen to mimic poor nutrients availability. Next, 0.5 g of soil were transferred into each bottle. Note that, 0.1 mL of the ﬁlter sterilized metals solution were added to each bottle after autoclave. Finally, either ∼ 2 mL (for the 1 mM test) or ∼ 8 mL (for the 5 mM test) of GLP (Sigma Aldrich, 96% purity) was added as an aqueous stock solution (5 g L−1, ﬁlter sterilized). Inside the bottle it remained an air headspace of 200 mL. The glasses were also loosely capped to maintain aerobic conditions and incubated at room temperature (20–23 °C) with orbital shaking at 150 rpm, thus guaranteeing homogeneous conditions, for longer than 100 days. Controls were set up for each enrichment culture by adding the soil before autoclav- ing the glass bottles. The enrichment cultures amended with GLP at 1 mM concentration were labeled as [name soil][number replica]-[GLP at 1 mM concentration] (e.g., Cam-Iw1-GLP1), while those amended with GLP at 5 mM concentration were labeled as [location]-[soil condition][number replica]-[chemical name][glyphosate concentration in the enrichment culture] (e.g., Cam-Iw1-GLP5).

28 SAMPLE CONTROL Soil MSM Trace metals GLP GLP Trace metals Soil MSM

Autoclave Autoclave

Test 1 mM Test 5 mM

250 mL 250 mL

Air Air

50 mL GLP 50 mL 48.2 mL GLP MSM MSM 41.6 mL + + Trace metals Trace metals 0 mL 0 mL

Figure 3: At the top left, the steps for enrichment cultures preparation, while at the top right, the steps for control preparation. On the bottom left, the procedure for GLP at 1 mM; on the bottom right, the procedure for GLP at 5 mM.

3.2.4. Analysis of GLP and AMPA by HPLC

GLP and AMPA concentration measurements were carried out following the method published in Tang et al. (2019) after Kawai et al. (1991). It is here briefly described the method relatively to the 1 mM test (Figure 4 pictures the method for both cases). 550 µL of samples were taken from each enrichment culture at time intervals for analysis of GLP and AMPA concentrations. Sampling frequency changed accordingly to degradation dynamics. Each sample was centrifuged for 4 minutes at 14,600 rpm. Then, 500 µL of supernatant were mixed with 250 µL of Na2HPO4 buffer (0.4 M, pH 11) in an eppendorf vial. After vortexing to assure mixing, vials + were incubated at 30 °C for at least 18 hours; this allowed excess NH4 to be removed through + volatilization of NH3 because NH4 compete with GLP and AMPA for the derivitizing agent p-toluenesulphonyl chloride (TsCl). Two control vials were filled with 750 µL (900 µL for the 5 mM case)of RO water and incubated with the sample vials. By weighting the control vials before and after the incubation period, it was possible to calculate the volume of evaporated water, which was added back to all the vials as reverse osmosis (RO) water. Next, 100 µL of TsCl solution (10 g/L in acetonitrile) were added to the vials. After vortexing to assure mixing, the vials were heated in a water bath at 50 °C for 5 minutes. The samples were filtered through 0.45 µm nylon filters before injecting into the high-performance liquid chromatography (HPLC) system. An Agilent 1100 series HPLC system equipped with an ultraviolet detector was used for analyses of GLP and AMPA derivatives. The separation was conducted at room temperature (21 °C) using a reversed-phase µBondapak C18 column (3.9 mm × 150 mm, particle size 10 µm) equipped with a guard column (4.6 mm × 7.5 mm, particle size 5 µm). Trifluoroacetic acid

29 (TFA, 9.5 mM in RO water at pH 2.1) and acetonitrile were used as the mobile phases; these were filtered through a 0.45 µm nylon membrane before use. The separation was conducted on a gradient from 0%–40% acetonitrile over 10 min at a flow rate of 1.0 mL min−1 and a detection wavelength of 240 nm. The injection volume was 20 µL. The TsCl derivative of GLP (TsCl- GLP) eluted as one, two, or three adjacent peaks because GLP is a zwitterion and TsCl-GLP can be differently charged, thus possibly resulting in multiple peaks. Tang et al. (2019) verified by liquid chromatography-mass spectrophotometry (LC-MS) both the identity of HPLC peaks as either GLP or AMPA and that the adjacent peaks had the same mass spectra, consistent with TsCl-GLP. The TsCl derivatives of GLP eluted at 7.7 and 7.8 minutes, while that of AMPA eluted at 7.4 minutes (Table 3 and Figure 6).

Test 1 mM Test 5 mM

550 µL 500 µL 120 µL 200 µL 250 µL Na2HPO4 480 µL H2O 300 µL Na2HPO4

Oven 30 °C for 18 hours

RO-H2O 100 µL TsCl 120 µL TsCl RO-H2O

HPLC vial

Figure 4: Scheme of sample preparation for HPLC analysis. On the left, the procedure for GLP at 1 mM; on the right, the procedure for GLP at 5 mM

3.2.5. GLP and AMPA standards and calibration curves

Five standards of GLP and AMPA (at 1, 10, 50, 100, and 150 mg L−1) were prepared by dissolv- ing GLP and AMPA (≥99% purity, Sigma Aldrich) in RO water. A power law was used to ﬁt the GLP and AMPA peak areas corresponding to the tested standard concentrations. The power law was used to calculate the GLP and AMPA concentrations throughout the experiments. Due to the very low pH of the liquid phase TFA, the HPLC column deteriorated and was therefore replaced on the 5th of October 2018. Thus, one set of parameter values was calibrated for each HPLC column.

30 Column molecule Elution time a n (minutes) Old GLP 6.4 55.0 0.91 New GLP 7.8 28.5 0.99 Old AMPA 5.7 91.8 0.70 New AMPA 7.6 36.2 0.94

Table 3: Parameters of the calibration curve to convert HPLC measurements to GLP and AMPA concentrations and elution times.

8000 6000 (a) (b) 5000 6000 4000

4000 3000 ] (mAU s) ] (mAU s)

2000 GLP GLP [ 2000 [ 1000

0 0 0 50 100 150 200 0 50 100 150 200 [ GLP] (mg L-1) [ GLP] (mg L-1) 4000 6000 (c) (d) 5000 3000 4000

] (mAU s) 2000 ] (mAU s) 3000

2000 AMPA AMPA

[ 1000 [ 1000

0 0 0 50 100 150 0 50 100 150 [ AMPA] (mg L-1) [ AMPA] (mg L-1)

Figure 5: Calibration GLP (50 mg L−1) and AMPA (50 mg L−1) measurement by HPLC. (a) Old column GLP; (b) New column GLP; (c) old column AMPA; (d) new column AMPA. Circles represent observations, while lines show the ﬁtted equation.

3.2.6. Media transfer

When nearly all GLP was biodegraded, 0.5 µL of the solution were transferred into glass bottles containing fresh media (MSM+GLP) prepared as described in Section 3.2.3. The GLP concentration was kept the same as the corresponding prior enrichment culture. To keep track of the number of transfers experienced by the enriched culture, the label was further revised as [location]-[soil condition][number replica]-[chemical name][glyphosate concentration in the enrichment culture]-T[number of transfer] (e.g., Cam-Iw1-GLP1-T0).

3.2.7. Cultures isolation and test for GLP biodegradation

Concurrently with each transfer, 200 µL of the prior enrichment culture were withdrawn and diluted by up to 5 orders of magnitude to culture possible GLP biodegraders on Petri dishes. Two diﬀerent medium were used. R2A-broth (1 L of solution contained: 0.5 g Casein acid hydrolysate, 0.5 g yeast extract, 0.5 g proteose peptone, 0.5 g dextrose, 0.5 g soluble starch, 0.3 g dipotassium phosphate, 0.024 g magnesium sulphate, 0.3 g sodium pyruvate, ﬁnal pH

31 Figure 6: Chromatograph of GLP and AMPA from HPLC. (a) Old column GLP; (b) New column GLP; (c) old column AMPA; (d) new column AMPA.

7.2) was added to AGAR AGAR as a low nutritional content culture medium, which supports growth of heterotrophic microorganisms normally inhabiting natural water. Otherwise, the same MSM+GLP medium was mixed with AGAR AGAR. After bacterial colonies appeared, single colonies were streaked on fresh Petri dishes prepared with the same original medium (either R2A-broth or MSM+GLP). Single colonies were then restreaked to make sure a pure strain was being isolated. Finally, the resulting isolates were inoculated in glass bottles containing fresh liquid MSM+GLP medium to test whether the strains could biodegrade GLP.

3.3. Results and discussions

3.3.1. GLP biodegradation dynamics: original enrichment cultures (T0)

GLP biodegradation dynamics showed consistent similarities and differences both across the soils sampled from different locations at the same GLP concentration and between the soils sampled from the same location at different GLP concentration (Figures 7 and 8). In the control

32 samples, GLP concentration was constant, thus suggesting that either chemical degradation and adsorption did not occur or contributed to a negligible extent (data not shown). At the lowest GLP concentration (1mM), GLP was biodegraded by all the enrichment cultures and the time delay occurred between microbes exposure to GLP and the onset of observed biodegradation (i.e., lag-time) was shorter than 15 days. However, the kinetics were very different; generally, samples collected from bare soils showed the slowest biodegradation kinetics (Figures 7 a,b,d,e), whereas samples collected from soils covered with a crop showed the fastest ones (Figures 7 c,f). All the cultures biodegraded GLP and produced AMPA stoichiometricly. It was peculiar the case Cam-Pa3-GLP1 (Figure 7 d), which represented bare soil with minimal previous exposure to GLP and showed relatively fast biodegradation without stoichiometric AMPA production. In contrast, the other two replicas, Cam-Pa1-GLP1 and Cam-Pa2-GLP1, showed very slow GLP biodegradation. Similarly, Tang et al. (2019) observed that two replicas from soil Cam-Pa biodegraded GLP at 0.6 mM, while one replica showed no biodegradation. These results highlighted that the C-N bond of GLP was preferably cleaved over the C-P bond except for bacteria from soil LE3 (See detailed information about GLP biodegradation in Sec- tion 4.5.2). Those dynamics could very likely indicate the presence of different microbial communities adapt to cope with the different in-situ soil functioning. Moreover, it was speculated that GLP biodegradation could vary within distances as short as 50 cm in some cases. Produced AMPA was not biodegraded in all cases (Figure 7), hence adaptation to GLP did not result in adaptation to AMPA. AMPA biodegraders may unlikely be selected using AMPA as the C source in enrichment cultures because of its low C content, whereas they may likely use it as a P source in the presence of other C sources. Unexpected AMPA dynamics were observed in soil Nar-Wh after a lag-time of nearly 40 days after GLP was depleted (Figure 7 f), when AMPA suddenly disappeared and reappeared in the corresponding soil microcosm. Similar dynamics were also observed for GLP in transfer cultures (Figure 9 g), and therefore, they are discussed further in Section 3.3.3. At the highest GLP concentration (5 mM), enrichment cultures which fast biodegraded GLP could still biodegrade GLP, even though after nearly 10-day-longer lag-times (Figure 8 c,f). Instead, GLP biodegradation was not observed in those that slowly biodegraded GLP 1mM. Remarkably, all the replicas in Cam-Pa-GLP5 and Nar-Iw-GLP5 failed to biodegrade GLP (Figure 8 d). The experimental data suggested that preexposure to tolerable GLP concentrations promoted adaptation of microbial communities to biodegrade GLP, whereas they may not be able to withstand high concentrations. Methodological and instrumental issues affected the observations to different extents. Point GLP and AMPA concentrations fluctuated over time because the samples had to stay at 30 °C for at least 18 hours. Hence, likely there were errors with the volumes of RO water used to replace the evaporated solution. This was even more evident with measurements for tests exposed to GLP 5 mM because 120 µL of sample had to be diluted with 480 µL of RO water. Moreover, the experiments lasted longer than 100 days; GLP and AMPA concentration were not corrected to account for evaporation from the glass bottles, which resulted in a small steady

33 increase of GLP and AMPA concentrations over time. The aging of the first HPLC column affected the measurements as elution times for GLP and AMPA became closer and closer over time (Figure 6). The issue was solved by replacing the column. Notwithstanding, the procedure might be improved to better separate the GLP and AMPA peaks. For instance, it could be worthy to fine-tune the gradient of the liquid phases in proximity of the AMPA and GLP elution times.

(a) Cam-Iw1-GLP1-T0 (b) Cam-Id1-GLP1-T0 (c) Cam-Cr1-GLP1-T0 Cam-Iw2-GLP1-T0 Cam-Id2-GLP1-T0 Cam-Cr2-GLP1-T0 Cam-Iw3-GLP1-T0 Cam-Id3-GLP1-T0 Cam-Cr3-GLP1-T0 1.4 Cam-Iw1-AMPA1-T0 1.4 Cam-Id1-AMPA1-T0 1.4 Cam-Cr1-AMPA1-T0 Cam-Iw2-AMPA1-T0 Cam-Id2-AMPA1-T0 Cam-Cr2-AMPA1-T0 1.2 1.2 1.2 Cam-Iw3-AMPA1-T0 Cam-Id3-AMPA1-T0 Cam-Cr3-AMPA1-T0 1 1 1 0.8 0.8 0.8 0.6 0.6 0.6

Concentration (mM) 0.4 Concentration (mM) 0.4 Concentration (mM) 0.4 0.2 0.2 0.2 0 0 0 0 50 100 150 0 50 100 150 0 50 100 150 t (day) t (day) t (day) (d) Cam-Pa1-GLP1-T0 (e) Nar-Ba1-GLP1-T0 (f) Cam-Pa2-GLP1-T0 Nar-Ba2-GLP1-T0 Nar-Iw1-GLP1-T0 Cam-Pa3-GLP1-T0 Nar-Ba3-GLP1-T0 Nar-Wh1-GLP1-T0 1.4 Cam-Pa1-AMPA1-T0 1.4 Nar-Ba1-AMPA1-T0 1.4 Nar-Iw1-AMPA1-T0 Cam-Pa2-AMPA1-T0 Nar-Ba2-AMPA1-T0 Nar-Wh2-AMPA1-T0 1.2 1.2 1.2 Cam-Pa3-AMPA1-T0 Nar-Ba3-AMPA1-T0 1 1 1 0.8 0.8 0.8 0.6 0.6 0.6

Concentration (mM) 0.4 Concentration (mM) 0.4 Concentration (mM) 0.4 0.2 0.2 0.2 0 0 0 0 50 100 150 0 50 100 150 0 50 100 150 t (day) t (days) t (days)

Figure 7: GLP and AMPA biodegradation at T0 by enrichment cultures exposed to 1 mM GLP concentration from soils: (a) Cam-Iw; (b) Cam-Id; (c) Cam-Cr; and (d) Cam-Pa; (e) Nar-Ba; (f) Nar-Iw and Nar-Wh.

3.3.2. GLP biodegradation dynamics: transferred cultures

Samples Cam-Cr1-GLP1-T0, Cam-Cr3-GLP1-T0, Cam-Pa3-GLP1-T0, Nar-Iw1-GLP1-T0, and Nar-Wh1-GLP1-T0 and Cam-Cr1-GLP5-T0, Cam-Cr3-GLP5-T0, and Nar-Wh1-GLP5-T0 completely biodegraded the added GLP, and were next transferred into fresh medium (Figure 9). The capability to biodegrade GLP was not always conserved in the transferred enrichment cultures. Unlikely, this was due to the lack of GLP biodegraders in the inoculate, although GLP did not promote substantial growth, which was evident by the lack of change in turbidity in the glass bottles an by the extremely poorer growth on GLP plates than R2A-broth ones (Figure 11). A second likely explanation was the loss of microbial consortia in the inoculate, which generally exchange necessary micro- and macro-nutrients amongst each other to promote overall microbial survival. A very likely possibility was the inability of some bacteria to initially cleave the C-N bond of GLP in the lack of a additional energy (from the soil used to inoculate soil bacteria, stored within the cell, or amended in the enrichment culture). In fact, an enzyme is

34 7 7 7 (a) (b) (c) 6 6 6

5 5 5

4 4 4 Cam-Iw1-GLP5-T0 Cam-Id1-GLP5-T0 Cam-Cr1-GLP5-T0 3 Cam-Iw2-GLP5-T0 3 Cam-Id2-GLP5-T0 3 Cam-Cr2-GLP5-T0 Cam-Iw3-GLP5-T0 Cam-Id3-GLP5-T0 Cam-Cr3-GLP5-T0 2 Cam-Iw1-AMPA5-T0 2 Cam-Id1-AMPA5-T0 2 Cam-Cr1-AMPA5-T0 Concentration (mM) Cam-Iw2-AMPA5-T0 Concentration (mM) Cam-Id2-AMPA5-T0 Concentration (mM) Cam-Cr2-AMPA5-T0 1 Cam-Iw3-AMPA5-T0 1 Cam-Id3-AMPA5-T0 1 Cam-Cr3-AMPA5-T0

0 0 0 0 50 100 150 200 0 50 100 150 200 0 50 100 150 200 t (days) t (days) t (days) 7 7 7 (d) (e) (f) 6 6 6

5 5 5

4 4 4 Cam-Pa1-GLP5-T0 Nar-Ba1-GLP5-T0 3 Cam-Pa2-GLP5-T0 3 Nar-Ba2-GLP5-T0 3 Nar-Iw1-GLP5-T0 Cam-Pa3-GLP5-T0 Nar-Ba3-GLP5-T0 Nar-Wh1-GLP5-T0 2 Cam-Pa1-AMPA5-T0 2 Nar-Ba1-AMPA5-T0 2 Nar-Iw1-AMPA5-T0 Concentration (mM) Cam-Pa2-AMPA5-T0 Concentration (mM) Nar-Ba2-AMPA5-T0 Concentration (mM) Nar-Wh2-AMPA5-T0 1 Cam-Pa3-AMPA5-T0 1 Nar-Ba3-AMPA5-T0 1

0 0 0 0 50 100 150 200 0 50 100 150 0 50 100 150 t (days) t (days) t (days)

Figure 8: GLP and AMPA biodegradation by enrichment cultures exposed to 5 mM GLP concentration at T0 from soils: (a) Cam-Iw; (b) Cam-Id; (c) Cam-Cr; and (d) Cam-Pa; (e) Nar-Ba; (f) Nar-Iw and Nar-Wh. required to liberate the C source glyoxylate. Note that the little amount of soil initially used to prepare the enrichment cultures may have supplied additional nutrients, which may have been used by GLP biodegraders to initially synthesize the necessary enzymes. In other cases, GLP biodegradation was observed but still after a long lag-time. One possible explanation was that the bacterial population was diluted 100 times in the fresh medium and it needed some time to increase the biomass concentration. Another possible reason was the paucity of both stored and bioavailable C to enhance GLP biodegradation. The transferred enrichment Nar-Wh-GLP1-T0, Cam-Cr3-GLP5-T1, and Nar-Wh-GLP5-T1 showed a very interesting and never reported time-series of GLP and AMPA concentration (Fig- ure 9 g). In particular, in soil Cam-Cr3-GLP5-T1, GLP concentration quickly decreased and increased within nearly 20 days. These dynamics repeated 3 times until GLP concentration steadily decreased and AMPA concentration increased (See Section 3.3.3 for further discussions).

3.3.3. GLP and AMPA re-appearance

In Cam-Cr3-GLP5-T1 (Figure 9 g), GLP disappeared and reappeared at least three times. These never-reported-before dynamics may have at least two diﬀerent plausible explanations: (1) a su- perposition of concurrent processes, which include bioaccumulation and release after cell lysis, (2) biosorption, or (3) uptake and release by protozoa, each followed by microbial biodegradation. In the ﬁrst situation, some cells may have rapidly taken up GLP through available

35 transporters (Sviridov et al., 2015), and bioaccumulated it. Because there were not C sources, starved cells eventually died and released cellular organic molecules back into the system after cell lysis, including GLP. Note that, Dornelles & Oliveira (2014) showed that a GLP-based mixture increased by at least 5 times the oxidation of the lipid layer of the cell membrane in bullfrog tadpoles, thus leading to irreversible cell damage. The extent of uptake decreased over time because of the decreasing number of viable cells involved in GLP bioaccumulation. In the second case, GLP could have been biosorbed either onto bacteria wall cell (after Grimes & Mor- rison (1975) and Geller (1979) observed this process for chlorinated hydrocarbon insecticides and atrazine, respectively) or onto bacteria-produced biofilm (after Kremer & Means (2009) observed an increase in biofilm-producing bacteria following exposure to GLP); however, biofilm was not seen in the enrichment culture. The last hypothesis involves the role of protozoa after Gulde et al. (2018) observed that these eukaryotes, which may have been contained in the enrichment culture, trapped ionic amines (GLP belongs to this class) from the liquid culture inside the cells but without degrading them. Finally, GLP concentration decreased more and more rapidly and AMPA was produced stoichiometrically, which can be explained by the increasing abundance of GLP degrading bacteria. Numerical works are currently being carried out with the aim to investigate these peculiar dynamics.

3.3.4. Bacteria isolation

Seven strains were isolated from Cam-Cr3-GLP5-T1 using R2A-broth (Figure 10). Note that, the inoculum to be spread was taken around day 20, that is after GLP disappeared but before it reappeared in the glass bottle (Figure 9g). All the strains grew very fast on the solid media (within 2 days) except for the small pink circular colonies-forming bacteria (within 7 days) (Figure 10 d). Notwithstanding none of those cell colonies were able to biodegrade GLP after being inoculated in bottle ﬂasks containing fresh MSM and GLP at 5 mM concentration (data not shown); very little AMPA was produced only after nearly 50 days. Other agar plates were prepared from Cam-Cr3-GLP5-T1 using either R2A-broth or GLP as the C source (Figure 11). In this case, the spread inoculum was taken around day 85 after GLP was depleted (Figure 9g). Three of the seven strains previously isolated quickly grew again on the fresh plates prepared with R2A-broth (Figure 11 a), while unrecognized bacterial communities very slowly grew with GLP (Figure 11 b). This suggested that GLP is a poor C source, but also corroborated that GLP biodegraders were still alive in the glass bottle. The latter bacteria may be responsible for GLP biodegradation and experimental works are currently being carried out to test this.

3.4. Summary

Some of the results presented hereafter are in accordance with the literature reporting glyphosate biodegradation, while others are quite innovative; these latter open up to new research questions

36 and highlight some lacking knowledge in existing bioreactive models. For example:

• The availability of GLP biodegraders should be assessed in-situ before assuming the contrary so not to overestimate GLP persistence. For a similar reason, GLP toxicity to in-situ available bacterial communities should be assessed;

• In-situ monitoring of soil GLP and AMPA concentrations are suggested as (a) it was here shown that GLP may undergo unexpected dynamics, which do not involve biodegradation, and hence, a reduction in environmental pollution, and (b) GLP was biodegraded to AMPA, which is a toxic and persistent molecule.

37 1.4 1.4 Cam-Cr1 (a) Cam-Cr3 (b) 1.2 1.2

1 1

0.8 GLP1-T0 0.8 GLP1-T0 GLP1-T1 GLP1-T1 0.6 GLP1-T2 0.6 GLP1-T2 AMPA1-T0 AMPA1-T0 0.4 AMPA1-T1 0.4 AMPA1-T1 Concentration (mM) AMPA1-T2 Concentration (mM) AMPA1-T2 0.2 0.2

0 0 0 20 40 60 80 100 0 20 40 60 80 100 t (day) t (day) 1.4 1.4 Cam-Pa3 (c) Nar-Iw1 (d) 1.2 1.2

1 1 GLP1-T0 0.8 0.8 GLP1-T1 GLP1-T0 GLP1-T2 0.6 GLP1-T1 0.6 GLP1-T3 AMPA1-T0 AMPA1-T0 0.4 AMPA1-T1 0.4 AMPA1-T1 Concentration (mM) Concentration (mM) AMPA1-T2 0.2 0.2 AMPA1-T3

0 0 0 20 40 60 80 100 0 20 40 60 80 t (day) t (day) 1.4 6 Nar-Wh1 (e) Cam-Cr1 (f)

1.2 5

1 4 0.8 GLP5-T0 3 GLP5-T1 0.6 AMPA5-T0 AMPA5-T1 GLP1-T0 2 0.4

Concentration (mM) GLP1-T1 Concentration (mM) AMPA1-T0 0.2 1 AMPA1-T1 0 0 0 50 100 150 0 20 40 60 80 100 120 t (day) t (day) 6 6 Cam-Cr3 (g) Nar-Wh1 (h)

5 5

4 GLP5-T0 4 GLP5-T1 GLP5-T0 3 GLP5-T2 3 GLP5-T1 GLP5-T2b AMPA5-T0 AMPA5-T0 2 2 AMPA5-T1 AMPA5-T1 Concentration (mM) AMPA5-T2 Concentration (mM) 1 AMPA5-T2b 1

0 0 0 20 40 60 80 100 0 20 40 60 80 100 t (day) t (day)

Figure 9: GLP and AMPA biodegradation in transferred enrichment cultures

38 Figure 10: Isolated strains on R2A-broth from Cam-Cr3-GLP5-T1 using an inoculum taken around day 20 before GLP depletion.

Figure 11: Isolated strains from Cam-Cr3-GLP5-T1 using an inoculum taken around day 85 after GLP depletion. (a) R2A-broth and (b) MSM+GLP.

4. Mechanistic reaction networks development

4.1. Introduction

This chapter presents the procedure used to estimate the MMM kinetic parameters (i.e., µ, K, and Y in Eq. 8) relative to the biodegradation of the herbicides ATZ and GLP and their metabolites using laboratory experiments published in peer-reviewed journals. Those experiments were also used to develop the reaction networks of the herbicides ATZ and GLP by means of the validation by construct (McCarl & Apland, 1986). This procedure implies that biochemical reactions can be linearly coupled together, as it is the case for biochemical systems where one product is the reagent in other reactions. The ATZ and GLP reaction networks are used for environmental modeling purposes in the next chapter 5, while the uncertainty of the estimated parameters is assessed in Chapter 6.

4.2. Method of parameters estimation

Laboratory experiments such as those described in Section 3.3 can inform researchers on biodegradation rates. In the MMM framework (Eqs. 8 and 9), at least three parameters (i.e., µ, K, and Y) need to be estimated for each biodegradation reaction, as well as the initial biomass concen- −6 −1 tration B0 and the cell mortality δ if they are not known. However, cell mortality δ = 10 s was assumed to be constant for all microbial strains (Gastrin et al., 1968; Salem et al., 2006). As far as cometabolic reactions are concerned, also the substrate(s) affinity K corresponding to the active byproduct(s) need(s) to be estimated. Moreover, unless explicitly expressed in the literature, inhibition and competition terms were not taken into account when these mechanisms were not detected in the retrieved experiments. O2 concentration was assumed to be not limiting and was not explicitly included as a MM term in aerobic reactions. The unknowns were estimated by inverse problem solution against experimental values such as substrates, byproducts, and products concentrations. The direct numerical solution was obtained by an explicit finite difference method solving the MMM kinetic equations (Eq. 8 was adapted on a case by case basis to account for multiple substrate in cometabolic reactions, competition terms, and inhibition terms). Numerical convergence and mass conservation were tested and verified in each time step. Parameter values were fine-tuned using the nonlinear Levenberg-Marquardt least-square fitting algorithm implemented in the PEST environment (Doherty et al., 2016) until the modeled concentrations converged to observations within a tolerance. The advantage of using PEST is that it provides confidence intervals and cross-correlation results at the end of the calibration procedure. These information should be used to assess the robustness of the best estimates. It is also strongly recommended to compare the estimated values with those reported in the literature, if available.

41 In addition to this conventional calibration method to find the best estimate of a parameter value, the Reader may be interested in exploring the Bayesian calibration approach. This allows one to calculate the most likely uncertainty distributions relative to each model parameter that result in a model output uncertainty distribution which most likely describes the measured data (Muehleisen & Bergerson, 2016). Hence, the Bayesian calibration offers a tool to account for parameter uncertainty in numerical predictions. However, it is my opinion that uncertainty distributions may be different when calculated using real world data or laboratory-controlled observations. This means that uncertainty distributions of parameters determined under laboratory conditions may bias the model outcomes of real-case scenarios. In the lack of field data, it may be good practice to use a conventional calibration method for parameter estimation, and later assign a uniform distribution to the parameter space in order to carry out sensitivity analyses. In fact the uniform distribution indicates that no prior information is available relative to the uncertainty of the parameter value. This approach was applied to assess the robustness of the atrazine reaction network developed in Section 4.4.2 Goodness-of-fit was measured with the coefficient of determination R2 and normalized root mean squared error percent NRMSE defined as:

σS S R2 = mod pre , σS mod σS pre q Pno − i=1(S modi S prei) NRMSE = no · 100, max(S pre) − min(S pre) where σ is the standard deviation, S mod and S pre are the model and experimental concentrations, respectively, while no is the number of observations.

4.3. Speciﬁc biomass aﬃnity

Assuming the case of one substrate with stoichiometric coefficient s=1 and a constant microbial biomass concentration, the MMM equation (Eq. 8) with S in mol L−1, B in mg-wet-biomass L−1, and Y in mg-wet-biomass mol-substrate−1 can be rewritten as a pseudo first-order equation as dS S B µB = µ × × → × S dt S + K Y Y(S + K) . The latter can be used to calculate the specific biomass affinity Φ (la Cecilia & Maggi, 2016) in the limit

µB µB∗ Φ = lim = , (10) S →0 mol/L Y(S + K) YK B→B∗ which represents the capability for an enzyme produced by a microorganism with a biomass concentration B∗ to bind to and degrade a substrate S also when S concentration is low. This definition of Φ corresponds to that of the specific affinity a in chemical reactions (Button, 1983; Reay et al., 1999), but includes the biomass concentration B and the yield coefficient Y for that

42 microorganism and, therefore, is more appropriate than a to quantify the affinity for a substrate in a biochemical reaction. Blok (1994) arrived to a similar formulation but introduced a factor to distinguish between situations of zero- and first-order kinetics. Note that the unit of Φ is [T−1]. High Φ values indicate that S will be degraded rapidly as compared to low Φ values. Low Y values, peculiar of microorganisms that do not need large quantities of substrate to grow, imply high Φ values if other parameters are unchanged. Φ can therefore inform on the degradation rate of a substrate with low concentration for any specific degrader. Using a biomass ∗ −1 0 concentration B = B0 leads to Φ0 = µB0/YK; when B0 = 1 mg L , then Φ0 = Φ = µ/YK can be used as a standard measure particularly suitable in laboratory experiments to compare diverse microorganism species in terms of their specific biomass affinity per unit biomass concentration.

43 4.4. Atrazine

In this section, the biodegradation reactions for ATZ and its metabolites are described. The contents come from the article la Cecilia & Maggi (2016)1 published in the Journal of Envi- ronmental Management and the article la Cecilia & Maggi (2017a)2 published in the Journal of Contaminant Hydrology.

4.4.1. Introduction

Atrazine (ATZ) is an herbicide introduced in extensive agriculture since 1958 (IPSC, 1990) and is still used in most countries of the world except in the European Union where it was banned in 2004 due to its persistence (91/414/EEC, 2004). To the best of our knowledge, the last offi- cial figure regarding triazines yearly input worldwide dates to 2013 and reports 14,990 tonnes of active ingredient (FAO, 2013). ATZ is used to suppress weeds in different contexts from agricultural areas to public gardens and households, although glyphosate has now become more common worldwide. After ATZ is applied, it may undergo different fates, including runoff in surface waters or leaching through the ground. There is evidence that ATZ is persistent in these environments, where it can be relatively slowly biodecomposed (Shapir & Mandelbaum, 1997). Studies report the detrimental implications and toxicological effects of ATZ to aquatic microorganisms (Graymore et al., 2001), amphibians (Hayes et al., 2002), and fishes (Fan et al., 2007). Mankind is not spared from serious health issues, including cancer oncogenesis in the reproductive apparatus (Fan et al., 2007) and babies health problems during gestation (Winchester et al., 2008), which may occur even after a single exposure to peak concentrations (Wu et al., 2010). In terrestrial and aquatic ecosystems, temperature, pH, and microbial adaptation (Krutz et al., 2008), electron donor and acceptor availability, and bacteria community composition (Kolic´ et al., 2007; Smith et al., 2005; Smith & Crowley, 2006) may introduce large variability in the decomposition rates of ATZ and its first phytotoxic metabolites DIATZ and DEATZ (Winkel- mann & Klaine, 1991). Note that, ATZ may be biodegraded to another first metabolite HOATZ, which is not as harmful as DIATZ and DEATZ because it does not contain chlorine (Winkel- mann & Klaine, 1991). On the one hand, ATZ half-life is reported to span from 4 to 385 days in soils (Eisler, 1989) or longer in deep soil aquifers (Agertved et al., 1992; Lapworth & Gooddy, 2006; Morvan et al., 2006; Tappe et al., 2002), while it could range from 10 to 105 days in surface waters (Scott, 2008). On the other hand, laboratory standardized tests report consistent average half-life values of 4 days in aerobic conditions (Mandelbaum et al., 1995; Radosevich et al., 1995; Smith et al., 2005) and 1 day in anaerobic conditions (Katz et al., 2000). Clearly, laboratory experiments diverge from non-optimal conditions occurring in the field; therefore,

1la Cecilia, D. and Maggi, F. (2016). Kinetics of Atrazine, Deisopropylatrazine, and Deethylatrazine soil biodecomposers. Journal of Environmental Management, 183, pp. 673-686, 10.1016/j.jenvman.2016.09.012. 2la Cecilia, D. and Maggi, F. (2017). In-situ atrazine biodegradation dynamics in wheat (Triticum) crops under variable hydrologic regime. Journal of Contaminant Hydrology, 203, pp. 104-121, http://dx.doi.org/10.1016/j.jconhyd.2017.05.004.

44 a great effort has been put in the experimental characterization of ATZ biodecomposition and its byproducts by means of bacterial isolates as well as communities of soil microorganisms in different laboratory conditions that could mimic the environmental variability in temperature and pH (Katz et al., 2000; Wang & Xie, 2012), bacteria strains or moisture (Krutz et al., 2008), soil depth (Krutz et al., 2010a), and oxygen (Katz et al., 2000). This has produced a relatively large body of data that may be put to new uses such as suggested in Debasmita & Rajasimman (2013) and Struthers et al. (1998). The objective of this section was to develop the ATZ reaction network, which is made of 18 metabolites, and estimate the corresponding kinetic parameters following the procedure explained in Section 4.2. The information provided herein can be integrated in a comprehensive environmental model used to predict the outcomes of current ATZ uses or to setup site- and scenario-specific studies to assess and mitigate the environmental contamination risk.

4.4.2. ATZ biodegradation pathways

Laboratory experiments show that soil microorganisms can biodegrade ATZ to cyanuric acid (CYA), ethylamine (ETA), isopropylamine (IPA), and C sources along 3 pathways, which are eventually catabolized to additional C sources, NH3, HCl, and CO2 (Figure 12). The occurrence of these biologically-mediated hydroxylation or oxidation reactions depends on the presence of specialized bacteria and environmental conditions (Eisler, 1989).

Pathway P1. It consists of 3 sequential hydrolytic ATZ biodegradation reactions to CYA. Ra- dosevich et al. (1995) carried out experiments of ATZ biodecomposition via P1R1a pathway using Ralstonia Basilensis M91-3 in three different liquid growth media in aerobic conditions. In the first (I), ATZ was used as the only carbon (C) and nitrogen (N) source. In the second (II), – glucose was added as the C source, while also NO3 was added as a N source in the third (III). Mandelbaum et al. (1995) showed that ATZ was biodecomposed to HOATZ by Pseudomonas sp. ADP with citrate as the C source in aerobic conditions. Katz et al. (2000) performed an analysis on ATZ biodegradation to HOATZ by Pseudomonas sp. ADP in the presence of citrate – as the C source in both aerobic (I) and anaerobic conditions. In the latter case, NO3 was added into the liquid growth medium as the electron acceptor (II and III with citrate/nitrate mass ratio equal to 6 and 7.5, respectively). Smith et al. (2005) evaluated the rate of ATZ decomposition in aerobic conditions by means of a community of 8 bacterial strains isolated from an agricultural soil without additional C or N sources. In those experiments, eight 7-member artificially constructed communities were tested for ATZ biodecompositon; among all microbial isolates, Nocardia sp. was the only one able to decompose ATZ to HOATZ with a rate that slightly depended on the community composition. In a following experiment, Smith & Crowley (2006) tested the contribution of fructose as a C source on ATZ biodecomposition using either the same microbial composition as in (Smith et al., 2005) or factitiously creating new pools of degraders. – Excluding the experiment with NO3 in Katz et al. (2000), those laboratory tests showed that ATZ biodecomposition to HOATZ during aerobic respiration on some C source was carried out

45 by Pseudomonas sp. ADP, Ralstonia Basilensis M91-3, Agrobacterium tumefaciens, Caulobac- ter crescentus, Pseudomonas putida, Sphingomonas yaniokuyae, Nocardia sp., Rhizobium sp., Flavobacterium oryzihabitans, Variovorax paradoxus, Rhizobium leguminosarum, Flavobac- terium sp., and Arthrobacter sp. Because these bacteria were able to grow on both ATZ and a C source, two simultaneous, independent reactions for these processes can be written as:

P1R1a C8H14ClN5 + H2O −→ C8H15N5O + HCl, (11) ATZ HOATZ

CH2O + O2 −→ H2O + CO2, (12)

where CH2O was used as the C source in Eq. (12) in place of other sources reported above for simplicity. – In anaerobic conditions, Katz et al. (2000) showed that NO3 was used by Pseudomonas sp. ADP as the electron acceptor in the presence of citrate as an additional electron donor. ATZ biodecomposition to HOATZ and denitriﬁcation occurred therefore simultaneously, and reactions describing these observations can be written as:

P1R1b C8H14ClN5 + H2O −→ C8H15N5O + HCl, (13) ATZ HOATZ − − 2 NO3 + CH2O −→ 2 NO2 + CO2 + H2O, (14) − 2 NO2 + 2 CH2O −→ N2 + 2 CO2 + 2 H2O, (15)

where CH2O was used for simplicity as before. Bacteria were able to grow on both ATZ hy- – – droxylation in Eq. (13), and NO3 and NO2 denitrification in Eqs. (14) and (15) with CH2O as the electron donor. In Eqs. (14) and (15), ATZ was not considered as an electron donor in the denitrification process. Kumar & Singh (2016) tested a community of Bacillus, Pseudomonas, and Burkholderia bacteria for HOATZ biodegradation to N-isopropylammelide (NIPA) and ETA, first without additional C and nitrogen (N) sources, next with sucrose as an additional C source, and finally with (NH4)3PO4 as an additional N source, every time under aerobic conditions (pathway P1R2, Figure 15). The microbial culture was able to degrade HOATZ according to the reaction (Shapir & Mandelbaum, 1997)

P1R2 C8H15N5O + H2O −→ C6H10N4O2 + C2H7N, (16) HOATZ NIPA ETA

CH2O + O2 −→ CO2 + H2O, (17) 3 NH + + O −→ NO − + H O + 2 H+. (18) 4 2 2 2 2 where CH2O was used as the C source in Eq. (17) in place of sucrose for simplicity. Boundy-Mills et al. (1997) used an Escherichia Coli with the gene AtzC extracted from

46 Pseudomonas sp. ADP responsible for NIPA biodegradation to CYA and IPA (pathway P1R3, Figure 15) in aerobic conditions and without any additional C source via a hydrolytic reaction as (Shapir & Mandelbaum, 1997)

P1R3 C6H10N4O2 + H2O −→ C3H3N3O3 + C3H9N. (19) NIPA CYA IPA Pathway P2. It consists of 1 oxidative and 3 hydrolytic ATZ biodegradation sequential reactions to CYA. P2R1 is the ATZ oxidative dealkylation to deisopropylatrazine (DIATZ) and acetone mediated by Rhodococcus strains TE1 (Behki et al., 1993; Shao et al., 1995) and B30 (Behki & Khan, 1994) and by Enterobacter cloacae strain JS08.Deg01 (Solomon et al., 2013) in aerobic conditions (pathway P2R1, Figure 15). 3 Pathway P2R1 can be written as

3 P2R1 C8H14ClN5 + O2 −→ C5H8ClN5 + 3 CH2O. (20) ATZ 2 DIATZ where CH2O was used in place of acetone for simplicity. DIATZ is next hydroxylated by Rhodococcus sp. TE1 containing the plasmid pP18 in the presence of glycerol as an additional C source, by Rhodococcus Corallinus (Shao et al. 1995 and Cook & Huetter 1984), or by Enterobacter cloacae strain JS08.Deg01 (Solomon et al., 2013) to deisopropylhydroxyatrazine (DIHOATZ) and HCl (pathway P2R2, Figure 15) and the reaction can be written as

P2R2 C5H8ClN5 + H2O −→ C5H9N5O + HCl. (21) DIATZ DIHOATZ

No studies have been found for the biodegradation of DIHOATZ down to CYA; however, Kumar & Singh (2016) proposed that DIHOATZ is hydrolyzed to 2,4-dehydroxy-6-N- ethylamino-1,3,5-atrazine (DHONATZ) and NH3, while DHONATZ is hydrolyzed to CYA and ETA (pathways P2R3 and P2R4, respectively, Figure 15). Those two reactions can be written as

P2R3 C5H9N5O + H2O −→ C5H8N4O2 + NH3, (22) DIHOATZ DHONATZ P2R4 C5H8N4O2 + H2O −→ C3H3N3O3 + C2H7N. (23) DHONATZ CYA ETA

Enterobacter cloacae strain JS08.Deg01 may be responsible for DIHOATZ degradation to CYA according to Eqs. (22) and (23) as suggested by Solomon et al. (2013). The kinetic parameters corresponding to these uncharacterized biological reactions were inferred from those of chemical compounds with similar C, N, and Cl atoms number, similarly to what could be done using bioinformatic tools.

3Solomon et al. (2013) assessed the capability of 9 bacterial strains to biodegrade ATZ in aerobic conditions, with ATZ as the only C and N source. All strains, isolated from an agricultural soil, were tested at diﬀerent pH with optimal biodecomposition rate at pH ' 7. Of the 9 strains, isolate JS08.Deg01 degraded ATZ at the fastest rate. This strain was found similar to Enterobacter cloacae ATCC13047TT and E. cloacae LMG.

47 Pathway P3. It consists of 2 oxidative and 3 hydrolytic ATZ biodegradation sequential reactions to CYA. In the ﬁrst oxidation, the Rhodococcus strains TE1 (Behki et al., 1993; Shao et al., 1995) and B30 (Behki & Khan, 1994), and the Enterobacter cloacae strain JS08.Deg01 (Solomon et al., 2013) break ATZ down to deethylatrazine (DEATZ) in aerobic conditions (pathway P3R1, Figure 15), and the reaction can be written as

P3R1 C8H15ClN5 + 3 O2 −→ C6H10ClN5 + 2 H2O + 2 CO2. (24) ATZ DEATZ

DEATZ is oxidized to deisopropyl-deethylatrazine (DIDEATZ) and acetone by strains of the genus Rhodococcus (Cook & Huetter, 1984; Shao et al., 1995) and by Nocardia sp. in a liquid growth medium amended with glucose as the C source in aerobic conditions (Giardi et al., 1985) (pathway P3R2, Figure 15). Noting that acetone (C3H6O) can be substituted with

CH2O also in this case, the reaction can be written as

3 P3R2 C6H10ClN5 + O2 −→ C3H4ClN5 + 3 CH2O. (25) DEATZ 2 DIDEATZ

No studies have been found for the biodegradation of DIDEATZ down to CYA; however,Kumar & Singh (2016) proposed that DIDEATZ is broken down to CYA via 3 hydrolytic reactions in aerobic conditions. First DIDEATZ is hydrolyzed to 2-chloro-4-hydroxy-6-amino-

1,3,5-triazine (CLHOATZ) and NH3, next CLHOATZ is hydroxylated to 2,4-dehydroxy-6- amino-1,3,5-triazine (DHOATZ) and HCl, and eventually DHOATZ is hydrolyzed to CYA and

NH3 (pathways P3R3, P3R4, and P3R5, respectively, Figure 15) as

P3R3 C3H4ClN5 + H2O −→ C3H3ClN4O + NH3, (26) DIDEATZ CLHOATZ P3R4 C3H3ClN4O + H2O −→ C3H4N4O2 + HCl, (27) CLHOATZ DHOATZ P3R5 C3H4N4O2 + H2O −→ C3H3N3O3 + NH3. (28) DHOATZ CYA where Enterobacter cloacae strain JS08.Deg01 may mediate reactions in Eqs. (26) to (28) (Solomon et al., 2013). The kinetic parameters corresponding to these uncharacterized biological reactions were inferred from those of chemical compounds with similar C, N, and Cl atoms number, similarly to what could be done using bioinformatic tools. ATZ can be simultaneously be biodegraded to DIATZ (P2R1) and DEATZ (P3R1), which were in turn biodegraded, likely to DIHOATZ (P2R2) and DIDEATZ (P3R2), respectively (Solomon et al., 2013). In this instance, a system of 5 equations describing pathways P2R1,

P3R1, P2R2, P3R2, and CH2O metabolization should be solved; this experiment stresses the importance to develop reaction networks where switches can be identiﬁed depending on in-situ conditions. However, Shao et al. (1995) reported that DEATZ was more slowly degraded than DIATZ, likely because DIATZ undergoes hydroxylation more rapidly than DEATZ undergoes dealkylation, or because DIATZ may inhibit DEATZ decomposition. Given the uncertainty,

48 inhibition terms were not introduced. DEATZ hydroxylation can result in the formation of deethylhydroxyatrazine (DEHA) (Arthur et al., 1997). While this process does occur in water (Lerch et al., 1998), it is uncertain whether it takes place in soil; therefore, this reaction is included for completeness in Figure 15 but it was not modeled numerically.

Pathway P4. It consists of 3 sequential hydrolytic biodegradation reactions. In three different experiments, Martinez et al. (2001) used a mutant of Escherichia Coli to first hydrolyze CYA to biuret (BIU), next BIU was hydrolyzed to allophanate (ALP), and finally ALP was hydrolyzed to NH3, without additional C and N sources, and in aerobic conditions (pathways P4R1, P4R2, and P4R3, respectively, Figure 15) via the reactions

P4R1 C3H3N3O3 + H2O −→ C2H5N3O2 + CO2, (29) CYA BIU P4R2 C2H5N3O2 + H2O −→ C2H4N4O3 + NH3, (30) BIU ALP P4R3 C2H4N4O3 + H2O −→ 2 NH3 + 2 CO2. (31) ALP where the genes atzD, atzE, and atzF from Pseudomonas sp. ADP were implanted in the Es- cherichia Coli to mediate Eqs. (29), (30), and (31), respectively. Note that BIU appears in its non-ionic form in Eqs. (29) and (30).

Pathway P5. Levering et al. (1984) used Arthrobacter P1 to show ETA biodegradation to acetaldehyde without additional C and N sources under aerobic conditions (pathway P5, Figure 15). Berg et al. (2002) and Levering et al. (1984) report that acetaldehyde can be metabolized within bacteria cells; therefore, it was substituted with CH2O in Eq. (32) for simplicity, and pathway P5 can be written as

P5 C2H7N + O2 −→ 2 CH2O + NH3. (32) ETA

Pathway P6. IPA biodegradation pathway is uncertain (Cerniglia & Perry, 1975; de Azevedo & Susana, 2001). Cerniglia & Perry (1975) showed that IPA is deaminated by Mycobacterium convolutum strain NPA-1 to isopropanol (IPP) and NH3 (pathway P6R1, Figure 15) and the reaction can be written as

P6R1 C3H9N + H2O −→ C3H8O + NH3. (33) IPA IPP

Steﬀan et al. (1997) used a community of soil microorganisms to show IPP oxidation to acetone without additional C and N sources under aerobic conditions (pathway P6R2, Figure

15). With CH2O in place of C3H6O, the reaction can be written as

3 P6R2 C3H8O + O2 −→ 3 CH2O + H2O. (34) IPP 2

49 Similarly to DIHOATZ, DHONATZ, DIDEATZ, CLHOATZ, and DHOATZ the kinetics parameters for IPA and IPP biodegradation were inferred from ETA biodegradation given their similar atomic composition.

4.4.3. Results

Goodness of the fit between model and experiments The experiments retrieved from the literature and used to develop the ATZ reaction network also allowed us to estimate the corresponding kinetic parameters, specific biomass affinity, and goodness-of-fit (Table 5.3.1). The comparison between observed and predicted concentrations are reported in Figures A1 to A11. The index in the top corner of each plot identifies the corresponding laboratory experiment (Test in Table 5.3.1). Experiments relative to the same reaction pathway were grouped in the same figure. Notably P1R1a pathway has been the most investigated one, while ATZ biodecomposition to DIATZ and DEATZ the least. Modeled concentrations achieved R2 ranging between 0.41 and 0.99, and NRMSE spanning from 0.31% to 32%, respectively (Table 5.3.1, columns 14 and 15). Excluding the case of anaerobic ATZ degradation by Pseudomonas sp. ADP, which led to R2 = 0.41 and NRMSE = 32% (Table 5.3.1, tests 6 and 7, respectively), parameter estimations returned R2 > 0.85 and NRMSE < 13% in all other cases. These R2 and NRMSE values suggest that the kinetic equations and parameters were appropriate to describe the biodegradation reactions along each pathway. ATZ, DIATZ, and DEATZ kinetic parameters Regardless of the consumed substrate or the microorganism that carried out the reaction, there was no evident relationship among all kinetic parameters (Figure 13). The maximum specific growth rate µ showed about two orders of magnitude variability and a weak correlation against K (R = 0.03, p > 0.05, Figure 13a), while Y values showed about 4 orders of magnitude variability and a weak correlation against Φ0 (R = -0.17, p > 0.05, Figure 13b). However, a negative exponential relationships for Y and Φ0 couples of the same reaction pathway could be detected as prescribed by Eq. (10). This aligns with our hypothesis that microbial functional groups with high Y values have small Φ0 values. A clearer distinction could be drawn between aerobic (P1R1a, P2R1, and P3R1) and anaerobic (P1R1b) ATZ decomposition pathways, where a lack of O substantially decreased K (where S stands for ATZ, DIATZ, or DEATZ) and K 2 S CH2O (Figure 13). Specifically to ATZ degradation by Pseudomonas sp. ADP, KS decreased more than µS when comparing aerobic to anaerobic breakdown, while YS did not substantially change (Table 5.3.1, tests 5 and 6), thus the degradation rate was higher in anaerobic conditions than in aerobic ones. Analogously, Y slightly varied between aerobic and anaerobic conditions, CH2O while both µ and K decreased in the anaerobic conditions (Table 5.3.1, tests 5 and 6); CH2O CH2O therefore, anaerobic CH2O metabolization was faster than the aerobic one. While parametric variability within a pathway can be explained by experimental and modeling uncertainty, variability between pathways can be better highlighted by analysis of average values. To this purpose, averages were calculated by grouping parameters within each reac-

50 tion pathway assuming that each one was performed by microorganisms with similar metabolic features (Table 4). Average parameters underline that the variability between pathways may be related more likely to the different reaction and microbial functional groups rather than experimental and parametric statistical uncertainty. For example, average K values of functional groups that either hydroxylate (P1R1a) or oxidize ATZ (P2R1 and P3R1) differed by 2 and 4 orders of magnitude, respectively. Moreover, the different oxygen requirement for P1R1a and P1R1b resulted in K and Y values varying by 2 and 1 orders of magnitude, respectively. Gener- ally, functional groups showed µ values varying within one order of magnitude, and Y, Φ0, and K values varying by up to 3, 4, and 7 orders of magnitude, respectively.

4.4.4. Discussion

The calibration procedure of 31 experimental sets of ATZ and its metabolites biodegradation in laboratory conditions has proved to be accurate to estimate the Michaelis-Menten-Monod kinetic parameters given the high correlation values and low errors. Despite being valuable, experiments are heterogeneous because they were carried out in different instances, with different experimental procedures and instruments and, therefore, possible variability in the concentration measurements are to be expected. Moreover, a residual uncertainty in the determination of the kinetic parameters may have arisen from model structure because it is possible that not all processes were adequately described by the kinetic equations used here, such as the anabolic reactions. Bacteria have the aptitude for consuming all bioavailable nutrients (Morita, 1988) and the energy liberated through the breakdown of these molecules is used for growth and cell maintenance (Rittmann & McCarty, 2001). Although the biomass yield Y may depend on the rate of substrate consumption (Rittmann & McCarty, 2001), in this paper Y was assumed to be constant as experimentally determined in Monod (1949). Another source of uncertainty in the model regards cells mortality, which was assumed to be constant (δ = 10−6 1/s) for each experimental set and regardless of the microbial species in consideration as compared to values ranging from 6.9 × 10−7 s−1 to 2 × 10−6 s−1 (Salem et al., 2006). Rather, nutrients availability may have affected microbial survival and the potential biodegradation rate of these compounds (Salem et al., 2006), thus leading to parametric variability detected in our exercise. In particular, addition of C and N sources into the liquid growth medium of the experiments used to determine ATZ degradation kinetic parameters broadened the types of reactions that bacteria could carry out in order to produce energy for their maintenance and growth. For example, an additional C source may have provided the bacteria with the energy to scavenge nitrogen from ATZ ring for anabolic purposes. This mechanism may be an explanation for the faster ATZ degradation in the presence of an additional electron donor (Mandelbaum et al., 1995) as well as experimental – variability. Additional NO3 together with a C source did not affect aerobic ATZ degradation rate (Katz et al., 2000). In contrast, the simultaneous presence of C and N sources in anaerobic conditions led to fast ATZ degradation. The affinity for the C source increased (K decreased CH2O by 4 orders of magnitude in anaerobic against aerobic conditions, tests 5, 6, and 7, Table 5.3.1) likely because the energy yield from the C source was low without oxygen as electron acceptor,

51 and the C source had to be degraded more rapidly to support metabolic reactions. This may have – – led the bacteria to prefer NO3 to ATZ for their nitrogen requirements given NO3 availability. The high initial biomass concentration and the enhanced biomass anaerobic respiration during denitrification reactions contributed to the high ATZ degradation rate in experiments carried out in anaerobic conditions (tests 5, 6, and 7, Table 5.3.1). In all experiments retrieved from the literature, other important electron acceptors were added into the liquid growth medium such as phosphate and sulphate, but they were not considered in our approach; missing phosphate and sulphate substrate metabolization may have contributed to parameter uncertainty. Relative to our results, specific biomass affinity values of ATZ degradation were higher for isolates than consortia of bacteria tested in the experiments retrieved from the literature. This could be due to enhanced microbial adaptation to use ATZ in metabolic reactions in instances where ATZ may have been present at high concentration in agricultural soils; when possible, ATZ degraders may have obtained an advantage in terms of competition over other soil microorganisms. Note, however, that ATZ degradaders have been found in all ATZ-contaminated soils and not elsewhere (Udikovic-Koli´ c´ et al., 2007). Bacteria capable of fast ATZ degradation are rare (Smith et al., 2005) but it is possible to speculate that more species may become capable of ATZ degradation (De Souza et al., 1998) and that ATZ degradation rates in highly contaminated soils may also increase (Krutz et al., 2008). The ability of bacteria to initiate and improve new breakdown pathways in response to synthesized molecules (Copley, 2009) can be exploited toward effective bioremediation inter- ventions, and new genetic metatranscriptomics techniques seems promising in predicting the activity level of a specific enzyme endowed by a microbial community (Helbling et al., 2012). A more straightforward method to estimate the activity level of a bacterial strain may come from the use of MMM kinetic parameters to calculate the bacterial specific biomass affinity Φ0. This parameter may be used to shortlist bacteria with high specific biomass affinity for ATZ as they may be the favorite candidates for bioremediation of ATZ-contaminated soil. As ATZ is degraded faster than in earlier times, its kinetic parameters should be updated. For bioremediation purposes, in addition to nutrient requirements, soil characteristics can deeply affect ATZ biodecomposition and persistence in soils (Krutz et al., 2008; Zablotowicz et al., 2006). Environmental variables such as pH and soil moisture influence ATZ binding to soil particles. Dry alkaline soils (Eisler, 1989; Krutz et al., 2008) decrease the ATZ fraction available to microorganism, thus lengthening ATZ persistence in the soil. Moreover, bacteria inhabiting soils are active only at pH ranging between 4 and 8 (Boon & Laudelout, 1962; Rittmann & McCarty, 2001). Soil temperature also contributes to define the niche of ATZ degraders; wet soil with pH about 7 and temperature between 20 ◦C and 30 ◦C correspond to short ATZ persistence (Krutz et al., 2008; Zablotowicz et al., 2006).

52 Kinetic Parameters µ KYY Φ0 1 mol g-C-Bio mg-wet-Bio 1 Exp. condition Pathway Substrate s L g-C-Subs mol-Subs s 1 2 3 4 5 6 7 8 1 AER P1R1a ATZ (Eq. 11) (3.67 ± 2.49) × 10−5 (3.89 ± 4.24) × 10−4 (3.10 ± 2.82) × 10−1 (2.98 ± 2.71) × 105 (0.77 ± 2.81) × 10−4 2 ANAER P1R1b ATZ (Eq. 13) (2.31 ± 1.75) × 10−6 (3.43 ± 3.31) × 10−6 (3.19 ± 3.21) × 10−2 (3.06 ± 3.08) × 104 (6.14 ± 7.02) × 10−5 3 AER P2R1 ATZ (Eq. 20) (1.69 ± 1.83) × 10−4 (4.61 ± 3.81) × 10−3 (1.46 ± 1.36) × 10−2 (2.77 ± 2.66) × 104 (5.62 ± 8.91) × 10−6 4 AER P3R1 ATZ (Eq. 24) (4.33 ± 6.30) × 10−4 (2.87 ± 1.78) × 10−3 (3.19 ± 2.95) × 10−2 (6.08 ± 5.66) × 104 (1.17 ± 1.70) × 10−5 5 AER P2R2 DIATZ (Eq. 21) (0.85 ± 1.10) × 10−5 (1.03 ± 1.44) × 10−3 (2.40 ± 3.64) × 10−3 (2.78 ± 4.50) × 103 (2.39 ± 2.29) × 10−5 6 AER P3R2 DEATZ (Eq. 25) (5.21 ± 2.32) × 10−6 (3.67 ± 3.68) × 10−3 (7.04 ± 7.91) × 10−3 (1.01 ± 1.40) × 104 (6.78 ± 9.51) × 10−6 ± × −5 ± × −3 ± × −2 ± × 3 ± × −3 7 AER P1R1a CH2O (Eq. 12) (6.22 7.49) 10 (0.74 1.07) 10 (3.82 5.60) 10 (6.68 7.87) 10 (1.44 3.20) 10 ± × −5 ± × −7 ± × −1 ± × 4 ± × −3 8 ANAER CDEN1 P1R1b CH2O (Eq. 14) (7.91 0.83) 10 (2.55 0.46) 10 (5.13 2.65) 10 (6.16 3.18) 10 (5.73 2.57) 10 ± × −6 ± × −7 ± × −2 ± × 3 ± × −3 9 ANAER CDEN2 P1R1b CH2O (Eq. 15) (9.23 0.32) 10 (8.87 0.59) 10 (1.38 1.15) 10 (1.65 1.37) 10 (9.28 7.10) 10 – ± × −5 ± × −3 10 ANAER CDEN1 P1R1b NO3 (Eq. 14) (7.91 0.83) 10 (4.70 0.04) 10 --- – ± × −6 ± × −3 11 ANAER CDEN2 P1R1b NO2 (Eq. 15) (9.23 0.32) 10 (8.60 0.57) 10 --- Table 4: Average kinetic parameters by reaction and microbial functional groups for ATZ, DIATZ, and DEATZ biodecomposition.

53 Atrazine Head compound C8H14ClN5 Intermediate ATZ compound Aerobic Aerobic

H O Anaerobic 2 O Final Aerobic 2 product P1R1b O2 H2O P2R1 H+ P3R1 Cl- P1R1a CH2O H2O - + Cl Denitrification H CO2 DIATZ DEATZ Deisopropyl- Deethylatrazine Hydroxyatrazine H O O2 2 atrazine C H ClN C8H15N5O 6 10 5 P2R2 C5H8ClN5 P3R2 HOATZ CH2O Organochloride H2O - + DIDEATZ NIPA Cl H DIHOATZ P1R2 Deisopropyl- N-isopropylammelide Deisopropyl- deethylatrazine H2O C6H10N4O2 H2O hydroxyatrazine C2H7N C3H4ClN5 C5H9N5O H2O P2R3 P3R3 NH P1R3 3 C H N NH3 Aromatic 3 9 DHONATZ CLHOATZ H O 2 2-Chloro-4-hydroxyl C2H7N 2,4-Dihydroxy-6-(N-ethyl) -6-amino,1-3-5-triazine Amino-1,3,5 atrazine H O Aliphatic C H ClN O 2 C5H8N4O2 3 3 4 P2R4 P3R4 - Cl H+ Inorganic CYA DHOATZ P3R5 H2O Cyanuric acid 2,4-Dihydroxy-6- C3H3N3O3 H O amino-1,3,5-triazine 2 NH 3 C3H4N4O2 BIU P4R1 --- Compound H2O Biuret CO2 H2O short name C2H5N3O2 X Deethylhydroxy- NH3 Cl- Reaction P4R2 atrazine P-R- C H N O coordinates H+ 6 11 5 ETA IPA ALP DEHA Solid line Hydrolization Ethylamine Isopropylamine Allophanate Dashed line Oxidation C2H7N C3H9N C2H4N2O3 Solid line End pathway H2O O2 P6R1 Dashed line H2O P5 NH3 Uncertain P4R3CO2 CH O 2 IPP X Uncharacterized O2 H2O Isopropanol CH2O C3H8O P6R2

Available as electron donor NH Cl- (e.g., C source and 3 denitrification)

- - N NO3 NO2 2 H2O CH2O CO H O 2 CH2O 2 P1R1b CO2 P1R1b

Figure 12: Atrazine biochemical reaction network. Bacterial strains carryout out a speciﬁc reaction pathway are listed in Table 5.3.1. Deethylatrazine hydrolysis to Deethylhydroxy-atrazine was found to occur in water but it is uncertain whether it occurs in soil too. No laboratory experiments have shown the kinetics of reaction pathways P2R3, P2R4, P3R3, P3R4, and P3R5; these hydrolysis have been suggested to occur by Solomon et al. (2013) and Kumar & Singh (2016).

54 2 7 10 10 ATZ, P1R1a ATZ, P1R1a

ATZ, P1R1b ) ATZ, P1R1b

ATZ, P2R1 −1 ATZ, P2R1 0 10 ATZ, P3R1 ATZ, P3R1 DIATZ, P2R2 DIATZ, P2R2 DEATZ, P3R2 105 DEATZ, P3R2 CH O, P1R1a CH O, P1R1a 10−2 2 2 ) CH O, P1R1b CH O, P1R1b 2 2 −1 (s µ

10−4 103

10−6 (mg−wet−Biomass mol−Substrate Y

−8 a) 1 b) 10 10 −8 −6 −4 −2 0 2 10−8 10−6 10−4 10−2 100 102 10 10 10 10 10 10 −1 K (mol L−1) Φ’ (s )

Figure 13: Representation of (a) maximum specific growth rate µ against Michaelis-Menten concentration K, and (b) biomass yield Y against biomass specific affinity Φ0. Data are grouped by reaction pathway, and oxic (red markers) and anoxic (blue markers) conditions. The 1:1 line was plotted to test whether the specific affinity a = µ/K (s−1 / mol L−1) in chemical reactions (Button, 1983; Reay et al., 1999) may follow such slope despite the two variables have different units.

55 Kinetic Parameters µ KYYB Φ0 t R2 NRMSE ini 1/2 1 mol g-C-Bio mg-wet-Bio mg 1 Test Source Exp. condition Pathway Substrate s L g-C-Subs mol-Subs L s (d) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 1 (a) Pseudomonas sp. AER P1R1a ATZ (Eq. 11) 2.87 × 10−5 5.42 × 10−4 1.09 × 10−3 1.05 × 103 1.80 × 10−1 5.04 × 10−5 0.44 0.99 5.52 ADP × −5 × −4 × −4 × 1 × −1 × −3 AER P1R1a CH2O (Eq. 12) 4.39 10 4.22 10 3.37 10 4.05 10 1.80 10 2.57 10 0.36 - - 2 (b) Ralstonia basilensis AER P1R1a ATZ (Eq. 11) 6.17 × 10−5 1.93 × 10−4 5.42 × 10−1 5.21 × 105 1.92 × 10−1 6.16 × 10−7 2.90 0.99 1.98 M91-3 3 (b) Ralstonia basilensis AER P1R1a ATZ (Eq. 11) 1.67 × 10−5 7.43 × 10−4 7.65 × 10−3 7.35 × 103 3.98 × 10−1 3.05 × 10−6 1.30 0.99 2.37 M91-3 × −5 × −4 × −2 × 4 × −1 × −6 AER P1R1a CH2O (Eq. 12) 2.51 10 3.68 10 1.84 10 1.33 10 3.98 10 5.11 10 2.15 - - 4 (b) Ralstonia basilensis AER P1R1a ATZ (Eq. 11) 1.65 × 10−5 1.70 × 10−3 1.73 × 10−2 1.66 × 104 3.98 × 10−1 5.84 × 10−7 1.13 0.96 8.45 M91-3 × −5 × −4 × −3 × 2 × −1 × −4 AER P1R1a CH2O (Eq. 12) 6.10 10 3.46 10 7.07 10 8.48 10 3.98 10 2.08 10 2.69 - - 5 (c) Pseudomonas sp. AER P1R1a ATZ (Eq. 11) 7.87 × 10−5 3.03 × 10−4 6.16 × 10−2 5.91 × 104 8.36 × 10−1 4.39 × 10−6 0.26 0.99 6.25 ADP × −4 × −3 × −1 × 4 × −1 × −6 AER P1R1a CH2O (Eq. 12) 2.30 10 2.04 10 1.61 10 1.93 10 8.36 10 5.85 10 0.28 0.99 1.07 6 (c) Pseudomonas sp. ANAER P1R1b ATZ (Eq. 13) 3.55 × 10−6 5.77 × 10−6 5.46 × 10−2 5.24 × 104 2.55 × 101 1.17 × 10−5 0.32 0.99 4.87 ADP × −5 × −7 × −1 × 4 × 1 × −3 (h) ANAER P1R1b CH2O (Eq. 14) 7.33 10 2.23 10 7.01 10 8.41 10 2.55 10 3.91 10 --- – × −5 × −3 × 1 (h) ANAER P1R1b NO3 (Eq. 14) 7.33 10 4.47 10 --- 2.55 10 - 0.96 7.49 × −6 × −7 × −2 × 3 × 1 × −3 (h) ANAER P1R1b CH2O (Eq. 15) 9.45 10 8.45 10 2.19 10 2.62 10 2.55 10 4.26 10 --- – × −6 × −3 × 1 (h) ANAER P1R1b NO2 (Eq. 15) 9.45 10 8.19 10 --- 2.55 10 - 0.41 29.81 7 (c) Pseudomonas sp. ANAER P1R1b ATZ (13) 1.07 × 10−6 1.09 × 10−6 9.23 × 10−3 8.86 × 103 1.22 × 101 1.11 × 10−4 0.33 0.97 12.91 ADP × −5 × −7 × −1 × 4 × 1 × −3 (h) ANAER P1R1b CH2O (Eq. 14) 8.50 10 2.88 10 3.26 10 3.91 10 1.22 10 7.54 10 - 0.99 3.59 – × −5 × −3 × 1 (h) ANAER P1R1b NO3 (Eq. 14) 8.50 10 4.64 10 - - 1.22 10 - - 0.97 6.11 × −6 × −7 × −3 × 2 × 1 × −2 (h) ANAER P1R1b CH2O (Eq. 15) 9.00 10 9.29 10 5.66 10 6.79 10 1.22 10 1.43 10 - 0.99 3.59 – × −6 × −3 × 1 (h) ANAER P1R1b NO2 (Eq. 15) 9.00 10 9.00 10 - - 1.22 10 - - 0.44 32.00 8 (d) Consortium without AER P1R1a ATZ (Eq. 11) 1.82 × 10−5 7.33 × 10−5 3.72 × 10−1 3.57 × 105 3.57 × 10−2 6.96 × 10−7 6.65 0.99 3.03 Agrobacterium 9 (d) Consortium without AER P1R1a ATZ (Eq. 11) 1.57 × 10−5 4.48 × 10−5 4.39 × 10−1 4.22 × 105 4.56 × 10−2 8.29 × 10−7 6.60 0.99 3.27 Caulocobacter 10 (d) Consortium without AER P1R1a ATZ (Eq. 11) 3.12 × 10−5 2.24 × 10−4 7.26 × 10−1 6.97 × 105 1.54 × 10−1 2.00 × 10−7 6.00 0.99 4.15 Flavobacterium 11 (d) Consortium without AER P1R1a ATZ (Eq. 11) 4.70 × 10−5 3.69 × 10−4 5.54 × 10−1 5.32 × 105 7.43 × 10−2 2.40 × 10−7 6.00 0.99 4.20 Pseudomonas 12 (d) Consortium without AER P1R1a ATZ (Eq. 11) 2.50 × 10−5 2.08 × 10−4 5.67 × 10−1 5.44 × 105 1.98 × 10−1 2.22 × 10−7 6.50 0.99 6.35 Rhizobium 13 (d) Consortium without AER P1R1a ATZ (Eq. 11) 5.95 × 10−5 6.35 × 10−4 7.03 × 10−1 6.75 × 105 1.60 × 10−1 1.39 × 10−7 6.65 0.98 4.66 Sphingomonas 14 (d) Consortium without AER P1R1a ATZ (Eq. 11) 2.45 × 10−5 2.50 × 10−5 2.45 × 10−1 2.35 × 105 1.49 × 10−3 4.17 × 10−6 5.45 0.99 2.63 Variovorax 15 (d) Consortium AER P1R1a ATZ (Eq. 11) 8.60 × 10−5 3.80 × 10−4 6.19 × 10−1 5.94 × 105 1.81 × 10−1 3.81 × 10−7 2.95 0.99 0.81 16 (e) Consortium with AER P1R1a ATZ (Eq. 11) 1.04 × 10−5 9.41 × 10−5 2.67 × 10−3 2.56 × 103 4.48 × 10−3 4.33 × 10−5 2.80 0.98 5.17 Arthrobacter × −5 × −4 × −3 × 2 × −3 × −4 AER P1R1a CH2O (Eq. 12) 1.66 10 2.54 10 1.40 10 1.68 10 4.48 10 3.88 10 2.70 - - 17 (e) Consortium with AER P1R1a ATZ (Eq. 11) 9.34 × 10−6 9.96 × 10−6 8.62 × 10−4 8.27 × 102 1.18 × 10−2 1.13 × 10−3 1.30 0.99 3.57 Leguminosarum × −5 × −3 × −3 × 2 × −2 × −5 AER P1R1a CH2O (Eq. 12) 3.06 10 3.07 10 1.01 10 1.21 10 1.18 10 8.24 10 1.70 - - 18 (e) Consortium AER P1R1a ATZ (Eq. 11) 5.77 × 10−5 6.78 × 10−4 9.98 × 10−2 9.58 × 104 4.76 × 10−1 8.87 × 10−7 1.95 0.98 5.65 × −5 × −5 × −2 × 4 × −1 × −5 AER P1R1a CH2O (Eq. 12) 1.43 10 6.19 10 8.77 10 1.05 10 4.76 10 2.19 10 2.35 - - 19 (f) Community AERO P1R2 HOATZ (Eq. 16) 3.77 × 10−5 3.14 × 10−5 4.77 × 10−2 9.16 × 104 4.62 × 10−3 6.06 × 10−8 4 0.99 2.26 20 (f) Community AERO P1R2 HOATZ (Eq. 17) 2.25 × 10−4 2.12 × 10−2 9.60 × 10−2 1.85 × 105 9.51 × 10−2 5.47 × 10−9 4.45 0.99 2.63 × −5 × −4 × −3 × 2 × −2 × −5 AERO P1R2 CH2O (Eq. 17) 2.23 10 4.42 10 1.46 10 3.51 10 9.51 10 1.36 10 3.5 - - 21 (f) Community AERO P1R2 HOATZ (Eq. 18) 6.13 × 10−4 3.44 × 10−3 3.01 × 10−2 5.77 × 104 2.03 × 10−2 6.28 × 10−8 10.03 0.98 5.44 + × −5 × −4 × −2 AERO P1R2 NH4 (Eq. 18) 2.23 10 2.85 10 - - 2.03 10 - 0.01 - - 22 (g) E. coli with gene AERO P1R3 NIPA (Eq. 19) 3.89 × 10−4 7.80 × 10−4 1.13 × 10−3 1.63 × 103 8.52 × 10−2 2.60 × 10−5 0.16 0.99 1.16 AtzC 23 (i) E. cloacae strain AER P2R1 ATZ (Eq. 20) 3.67 × 10−4 5.27 × 10−3 2.28 × 10−3 4.38 × 103 4.38 × 10−2 1.59 × 10−5 1.8 0.99 3.57 JS08.Deg01 AER P3R1 ATZ (Eq. 24) 6.12 × 10−5 2.67 × 10−3 3.83 × 10−4 7.36 × 102 4.38 × 10−2 3.11 × 10−5 1.8 0.99 3.57 AER P2R2 DIATZ (Eq. 21) 2.12 × 10−5 1.76 × 10−4 6.60 × 10−3 7.97 × 103 4.38 × 10−2 1.51 × 10−5 - 0.81 10.36 AER P3R2 DEATZ (Eq. 25) 3.57 × 10−6 1.07 × 10−3 1.71 × 10−4 2.46 × 102 4.38 × 10−2 1.35 × 10−5 - 0.66 12.20 24 (l) Rhodococcus strain AERO P2R1 ATZ (Eq. 20) 1.34 × 10−4 8.05 × 10−3 1.15 × 10−2 2.21 × 104 2.14 × 10−1 7.55 × 10−7 1.7 0.99 4.26 TE1 AERO P2R2 DIATZ (Eq. 21) ------0.90 9.08 AERO P3R1 ATZ (Eq. 24) 1.16 × 10−3 4.74 × 10−3 3.58 × 10−2 6.88 × 104 2.14 × 10−1 3.56 × 10−6 1.7 0.99 4.26 AERO P3R2 DEATZ (Eq. 25) ------0.96 6.10 AERO CH2O (Eq. 17) 1.00 × 10−9 7.73 × 10−5 6.50 × 10−2 1.54 × 104 2.14 × 10−1 8.38 × 10−10 --- 25 (m) Rhodococcus strain AERO P2R1 ATZ (Eq. 20) 5.79 × 10−6 5.11 × 10−4 2.95 × 10−2 5.67 × 104 1.24 × 10−1 2.00 × 10−7 0.9 0.99 1.22 B30 AERO P2R2 DIATZ (Eq. 21) ------0.96 5.03 AERO P3R1 ATZ (Eq. 24) 7.79 × 10−5 1.19 × 10−3 5.90 × 10−2 1.13 × 105 1.24 × 10−1 5.81 × 10−7 0.9 0.99 1.22 AERO P3R2 DEATZ (Eq. 25) 6.85 × 10−6 6.28 × 10−3 1.39 × 10−2 2.00 × 104 1.24 × 10−1 5.44 × 10−8 - 0.96 5.036 AERO CH2O (Eq. 17) 1.38 × 10−4 2.92 × 10−5 2.03 × 10−3 4.87 × 102 1.24 × 10−1 9.70 × 10−3 0.34 - - 23 (i) E. cloacae strain AER P2R2 DIATZ (Eq. 21) 2.12 × 10−5 1.76 × 10−4 6.60 × 10−3 7.97 × 103 4.38 × 10−2 1.51 × 10−5 - 0.81 10.36 JS08.Deg01 AER P3R1 ATZ (Eq. 24) 6.12 × 10−5 2.67 × 10−3 3.83 × 10−4 7.36 × 102 4.38 × 10−2 3.11 × 10−5 1.8 0.99 3.57 AER P2R1 ATZ (Eq. 20) 3.67 × 10−4 5.27 × 10−3 2.28 × 10−3 4.38 × 103 4.38 × 10−2 1.59 × 10−5 1.8 0.99 3.57

56 AER P3R2 DEATZ (Eq. 25) 3.57 × 10−6 1.07 × 10−3 1.71 × 10−4 2.46 × 102 4.38 × 10−2 1.35 × 10−5 - 0.66 12.20 26 (n) Rhodococcus AER P2R2 DIATZ (Eq. 21) 3.36 × 10−6 2.20 × 10−4 5.11 × 10−4 3.06 × 102 2.77 × 10−2 4.99 × 10−5 2.60 0.99 1.73 Corallinus × −5 × −4 × −2 × 4 × −2 × −5 AER P2R2 CH2O (Eq. 17) 3.51 10 1.19 10 9.83 10 1.18 10 2.77 10 2.49 10 3.20 - - 27 (n) Rhodococcus sp. AER P2R2 DIATZ (Eq. 21) 1.08 × 10−6 2.70 × 10−3 1.00 × 10−4 6.02 × 101 2.04 × 10−1 6.62 × 10−6 1.83 0.99 2.31 strain TE3 × −5 × −5 × −1 × 4 × −1 × −5 AER P2R2 CH2O (Eq. 17) 4.73 10 7.02 10 4.70 10 5.64 10 2.04 10 1.19 10 1.56 - - 22 (i) Enterobacter cloacae AER P3R1 ATZ (Eq. 24) 6.12 × 10−5 2.67 × 10−3 3.83 × 10−4 7.36 × 102 4.38 × 10−2 3.11 × 10−5 1.8 0.99 3.57 strain JS08.Deg01 AER P3R2 DEATZ (Eq. 25) 3.57 × 10−6 1.07 × 10−3 1.71 × 10−4 2.46 × 102 4.38 × 10−2 1.35 × 10−5 - 0.66 12.20 AER P2R1 ATZ (Eq. 20) 3.67 × 10−4 5.27 × 10−3 2.28 × 10−3 4.38 × 103 4.38 × 10−2 1.59 × 10−5 1.8 0.99 3.57 AER P2R2 DIATZ (Eq. 21) 2.12 × 10−5 1.76 × 10−4 6.60 × 10−3 7.97 × 103 4.38 × 10−2 1.51 × 10−5 - 0.81 10.36 24 (l) Rhodococcus strain AERO P3R1 ATZ (Eq. 24) 1.16 × 10−3 4.74 × 10−3 3.58 × 10−2 6.88 × 104 2.14 × 10−1 3.56 × 10−6 1.7 0.99 4.26 TE1 AERO P3R2 DEATZ (Eq. 25) ------0.96 6.10 AERO CH2O (Eq. 17) 1.00 × 10−9 7.73 × 10−5 6.50 × 10−2 1.54 × 104 2.14 × 10−1 8.38 × 10−10 --- AERO P2R1 ATZ (Eq. 20) 1.34 × 10−4 8.05 × 10−3 1.15 × 10−2 2.21 × 104 2.14 × 10−1 7.55 × 10−7 1.7 0.99 4.26 AERO P2R2 DIATZ (Eq. 21) ------0.90 9.08 25 (m) Rhodococcus strain AERO P3R1 ATZ (Eq. 24) 7.79 × 10−5 1.19 × 10−3 5.90 × 10−2 1.13 × 105 1.24 × 10−1 5.81 × 10−7 0.9 0.99 1.22 B30 AERO P3R2 DEATZ (Eq. 25) 6.85 × 10−6 6.28 × 10−3 1.39 × 10−2 2.00 × 104 1.24 × 10−1 5.44 × 10−8 - 0.96 5.036 AERO P2R1 ATZ (Eq. 20) 5.79 × 10−6 5.11 × 10−4 2.95 × 10−2 5.67 × 104 1.24 × 10−1 2.00 × 10−7 0.9 0.99 1.22 AERO P2R2 DIATZ (Eq. 21) ------0.96 5.03 AERO CH2O (Eq. 17) 1.38 × 10−4 2.92 × 10−5 2.03 × 10−3 4.87 × 102 1.24 × 10−1 9.70 × 10−3 0.34 - - 22 (i) Enterobacter cloacae AER P3R2 DEATZ (Eq. 25) 3.57 × 10−6 1.07 × 10−3 1.71 × 10−4 2.46 × 102 4.38 × 10−2 1.35 × 10−5 - 0.66 12.20 strain JS08.Deg01 AER P3R1 ATZ (Eq. 24) 6.12 × 10−5 2.67 × 10−3 3.83 × 10−4 7.36 × 102 4.38 × 10−2 3.11 × 10−5 1.8 0.99 3.57 AER P2R1 ATZ (Eq. 20) 3.67 × 10−4 5.27 × 10−3 2.28 × 10−3 4.38 × 103 4.38 × 10−2 1.59 × 10−5 1.8 0.99 3.57 AER P2R2 DIATZ (Eq. 21) 2.12 × 10−5 1.76 × 10−4 6.60 × 10−3 7.97 × 103 4.38 × 10−2 1.51 × 10−5 - 0.81 10.36 25 (m) Rhodococcus strain AERO P3R2 DEATZ (Eq. 25) 6.85 × 10−6 6.28 × 10−3 1.39 × 10−2 2.00 × 104 1.24 × 10−1 5.44 × 10−8 - 0.96 5.036 B30 AERO P3R1 ATZ (Eq. 24) 7.79 × 10−5 1.19 × 10−3 5.90 × 10−2 1.13 × 105 1.24 × 10−1 5.81 × 10−7 0.9 0.99 1.22 AERO P2R1 ATZ (Eq. 20) 5.79 × 10−6 5.11 × 10−4 2.95 × 10−2 5.67 × 104 1.24 × 10−1 2.00 × 10−7 0.9 0.99 1.22 AERO P2R2 DIATZ (Eq. 21) ------0.96 5.03 AERO CH2O (Eq. 17) 1.38 × 10−4 2.92 × 10−5 2.03 × 10−3 4.87 × 102 1.24 × 10−1 9.70 × 10−3 0.34 - - 28 (o) E. coli with gene AERO P4R1 CYA (Eq. 29) 2.14 × 10−3 6.70 × 10−1 8.50 × 10−4 6.13 × 102 6.72 × 10−1 3.51 × 10−6 1.38 0.99 2.66 AtzD 29 (o) E. coli with gene AERO P4R2 BIU (Eq. 30) 3.41 × 10−4 8.67 × 10−2 6.60 × 10−5 3.18 × 101 7.40 × 10−2 9.16 × 10−6 0.83 0.99 2.66 AtzE 30 (o) E. coli with gene AERO P4R3 ALP (Eq. 31) 9.26 × 10−5 1.11 × 10−1 9.10 × 10−5 4.40 × 101 1.23 × 100 2.33 × 10−5 0.34 0.99 3.38 AtzF 31 (p) Arthrobacter P1 AERO P5 ETA (Eq. 32) 9.34 × 10−5 8.38 × 10−1 7.60 × 10−5 3.67 × 101 3.84 × 100 1.16 × 10−5 0.66 0.89 10.81 Table 5: Estimated kinetic parameters for ATZ, HOATZ, DIATZ, DEATZ, NIPA, CYA, BIU, ALP, and ETA biodecomposition grouped by microbial isolate or community for reaction pathways represented in Figure (12). The specific biomass affinity Φ0 is also tabulated together with the goodness-of-fit of predicted against observed ATZ, DIATZ, and DEATZ concentrations. Experimental temperatures were not reported in the original sources; thus standard conditions were assumed (T = 25◦). Experimental data retrieved from: (a) Mandelbaum et al. (1995), (b) Radosevich et al. (1995), (c) Katz et al. (2000), (d) Smith et al. (2005), (e) Smith & Crowley (2006), (f) Kumar & Singh (2016), (g) Boundy-Mills et al. (1997), (h) Tang (2016), (i) Solomon et al. (2013), (l) Behki et al. (1993), (m) Behki & Khan (1994), (n) Shao et al. (1995), (o) Martinez et al. (2001), (p) Levering et al. (1984). AER and ANAER refer to aerobic and anaerobic conditions, respectively.

57 4.5. Glyphosate

In this section, the biodegradation reactions for GLP and its metabolites are described. The contents come from the article la Cecilia & Maggi (2018)4 published in Environmental Pollution.

4.5.1. Introduction

GLP is a broad spectrum herbicide, which first reached the market in 1974, and has since revolutionized the farming sector (Duke & Powles, 2008). GLP-based herbicides provide a low-cost weed control treatment because it is effective also at low active ingredient doses. GLP use has increased further worldwide after some crops have been genetically modified to be GLP-resistant (Coupe & Capel, 2016). Under laboratory conditions, GLP has been shown to be fast degraded biologically (Balthazor & Hallas, 1986; Jacob et al., 1988; Mcauliffe et al., 1990; Moore et al., 1983) and chemically (Barrett & McBride, 2005; Li et al., 2015; Paudel et al., 2015), thus posing a low risk for persistence. Finally, GLP has been found to be less toxic than other herbicides (Fishel et al., 2015). As a result, GLP has become the most used herbicide in agriculture. However, field surveys have found that an increasing number of GLP-resistant weeds have made their appearance (Heap, 2016). GLP has been found in soil (Aparicio et al., 2013; Silva et al., 2017, 2019), streams (Paris et al., 2013, 2016; Lefrancq et al., 2017), and groundwater (Paris et al., 2016), thus indicating GLP mobility and persistence in the environment, which may eventually result in a risk to groundwater quality (Simonsen et al., 2008). Note that, GLP oxidation has been shown to produce AMPA, a toxic compound more persistent than GLP (Grandcoin et al., 2017). The comparison between Paris et al. (2013) and Paris et al. (2016) provides further evidence consistent with the wide spreading groundwater contamination by GLP and AMPA. Indeed, contaminated sites out of tested sites increased from 0.9% to 5.8% for GLP and from 0.9% to 4.8% for AMPA in the period 2010-2013. The objective of this section was to develop the GLP reaction network and estimate the corresponding kinetic parameters following the procedure explained in Section 4.2. The information provided herein can be integrated in comprehensive environmental model used to predict the outcomes of current GLP uses or to setup site- and scenario-specific studies to assess and mitigate the environmental contamination risk.

4.5.2. GLP biochemical degradation pathways

Fifteen experimental data sets were retrieved from the existing literature relative to the reactions involved in the GLP soil biochemical reaction network. In soil, GLP degradation involves biotic and abiotic processes. The former are mediated by

4la Cecilia, D. and Maggi, F. (2018). Analysis of glyphosate degradation in a soil microcosm. Environmental Pollution. 233, pp. 201-207, https://doi.org/10.1016/j.envpol.2017.10.017

58 Head GLP compound Aq, Glyphosate p Intermediate compound C3H8NPO5 CH2O CO2 + O2 H Final H2O Aerobic A E product P1R1s Aerobic Aerobic P2R1c O2 CH2O Organophosphate A O2 P1R1 3- PO4 A P2R1s H+ CO2 AMPA aq, Compound phase GLX p Aq, (aqueous, protected) Glyoxylate aq Amino-methyl- phosphonic acid p C2H2O4 CH6NPO3 Sarcosine aq Aliphatic P1R2c C3H7NO2 Aerobic E SRC D O2

CH O Aerobic Aerobic Inorganic 2 H+ B O2 Anaerobic P2R2a 3- PO4 A P1R2s GLY

H+ CH --- Compound short name 2

aq O CO2 Glycine MTH C2H5NO2 Phosphono- aq P2R2b P-R- Reaction coordinates Methylamine aq C formaldehyde CO2 Solid line CH5N H2O O2 Biotic reaction

CHPO4 O 2 P2R3b P2R3a Dashed line O2 C B H2O CH CO2 Abiotic reaction Anaerobic Aerobic D B H2O P1R3b C P1R3a Anaerobic Dashed line Uncertain CH2O CO2 CH4

CH2O Inhibitor + H Aerobic 3- PO4 A BHyO

Available as electron donor B BAER (e.g., denitrification, aq 3- Aq, aq sulphate reduction) C B PO4 p ANAER NH3 CH2O D Ochrobactrum anthropi GPK 3 E Birnessite (Mn3+, Mn4+ containing mineral) Figure 14: Glyphosate degradation reaction network in soil. one microbial functional group along two pathways, P1 and P2, which are characterized by oxidative reactions, either cometabolic or not, and cometabolic hydrolysis reactions, respectively (Figure 14). Abiotic processes are oxidations catalyzed by manganese (Mn) ions contained in minerals such as birnessite. The two biotic pathways can be concurrent or not depending on C and P availability, and include intermediate reactions that have been identiﬁed only recently and are described below. The end products of GLP aerobic degradation are CO2, formaldehyde

CH2O, and NH3, while acetate and CH4 are end products in anaerobic conditions. Integration and accounting for all chemical and biological species and corresponding reactions in the network of Figure 14 are described in detail below together with an extended list of biochemical reactions.

Pathway P1. starts with GLP aerobic oxidation to AMPA and GLX mediated by Flavobac- terium sp. GD1 in the presence of gluconate and pyruvate as carbon (C) sources (Balthazor & Hallas, 1986), Ochrobactrum anthropi GPK 3 in the presence of an additional C source as co-substrate (Sviridov et al., 2012), and Pseudomonas sp. LBr in the presence of gluconate as a C source (Jacob et al., 1988) (pathway P1R1s, Figure 14). As suggested by those authors, because these bacteria were able to grow on GLP as a P source being provided an additional C source, one cometabolic reaction for this process was written as:

P1R1s C3H8NPO5 + 2 O2 + CH2O −−−−→ CH6NPO3 + C2H2O4 + CO2 + H2O, (35) GLP AMPA GLX where CH2O was used as the C source in Eq. (35) in place of other sources reported above for simplicity. Agrobacterium radiobacter and Achromobacter Group V D can degrade GLP as the only 3 – C source (pathway P1R1, Figure 14), in the presence of ortophosphate PO4 as a P source

59 (Mcauliﬀe et al., 1990). As suggested by those authors, because these bacteria were able to grow on GLP as a C source, one independent reaction for this process was written as:

P1R1 C3H8NPO5 + O2 −−−→ CH6NPO3 + C2H2O4. (36) GLP AMPA GLX

While GLX is part of metabolic pathways within the cell (Levering et al., 1981; Mcauliﬀe 3 – et al., 1990; Jacob et al., 1988), AMPA can be hydrolyzed to methylamine (MTH), PO4 , and H+ in aerobic conditions by Arthrobacter atrocyaneus ATCC 13752 (Pipke & Amrhein, 1988a) and Flavobacterium sp. GD1 (Balthazor & Hallas, 1986) (pathway P1R2s, Figure 14).

Given that AMPA concentration decreased during GLP biodegradation in Balthazor & Hal- 3 – las (1986), those observations were also used to assess AMPA biodegradation to MTH, PO4 , and H+ by Pseudomonas sp. PG2982 in aerobic conditions via pathway P1R2s. GLP was the only P source, while gluconate was used as a C source. As suggested by those authors, because this bacteria was able to grow on GLP and AMPA being provided an additional C source, two simultaneous, independent reactions for these processes can be written as:

P1R1s C3H8NPO5 + 2 O2 + CH2O −−−−→ CH6NPO3 + C2H2O4 + CO2 + H2O, (37) GLP AMPA GLX P1R2s 3− + CH6NPO3 + H2O + CH2O + O2 −−−−→ CH5N + PO4 + 3 H + CO2 + H2O, (38) AMPA MTH with CH2O as in Eqs. (35) and (45).

Balthazor & Hallas (1986) and Talbot et al. (1984) proposed an alternative biodegradation pathway for AMPA in the presence of pyridoxal phosphate and pyruvate to phosphonoformaldehyde and alanine; this reaction was observed in Ochrobactrum anthropi GPK 3, and the same 3 – + bacteria also hydrolyzed phosphonoformaldehyde to formaldehyde, PO4 , and H (Sviridov et al., 2012). Despite this alternative pathway for AMPA was shown to occur in laboratory conditions, its occurrence in ﬁeld is uncertain and was not explicitly accounted for in our analytical work.

3+ Alternatively, AMPA can adsorb onto the birnessite mineral surface ((Na0.3Ca0.1K0.1)(Mn , 4+ 3 – + Mn )2O4 · 1.5H2O) and be chemically oxidized by Mn ions to MTH, PO4 , and H (Li et al., 2015) (pathway P1R2c, Figure 14). Li et al. (2015) and Paudel et al. (2015) assessed AMPA 3 – + chemical degradation to MTH, PO4 , and H in the presence of birnessite via pathway P1R2c. As reported by the authors, AMPA ﬁrst underwent adsorption onto birnessite, next Mn ions 3 – contained in the mineral catalyzed the chemical reaction. Finally, the ion PO4 was adsorbed onto birnessite as well. Sorption was here described according to the Langmuir kinetic model (Langmuir, 1918) implemented as in Atkins & De Paula (2005), while AMPA degradation was modeled using the Michaelis-Menten model. Therefore, in total three simultaneous, independent reactions for these processes were written as

60 R1 CH6NPO3 ←→ CH6NPO3, (39) AMPA(aq) AMPA(ad)

1 P1R2c 3− + CH6NPO3 + O2 −−−−→ CH5N + PO4 + H , (40) AMPA(ad) 2 MTH

3− R3 3− PO4 (aq) ←→ PO4 (ad), (41)

MTH can be either oxidized by Arthrobacter P1 (Levering et al., 1984) to CH2O and NH3 in aerobic conditions, or hydrolyzed aerobically by Methanosarcina barkeri (Hippe et al., 1979) to CO2, CH4, and NH3 (pathway P1R3a and P1R3b, respectively, Figure 14). Note that CH2O was considered to be a model C source to all microbial functional groups while a number of organic compounds may have an equivalent function (Levering et al., 1981). MTH can be either oxidized or hydrolyzed via P1R3a in aerobic conditions, or in anaerobic conditions via P1R3b; GLY can be aerobically oxidized to CO2, NH3, and H2O by Pseu- domonas Ovalis (Appleyard & Woods, 1956), or anaerobically hydrolyzed to acetate and NH3 by Clostridium purinolyticum (Därre & Andreesen, 1982a) (pathways P2R3a and P2R3b, respectively, Figure 14).

Levering et al. (1984) assessed MTH metabolization to formaldehyde and NH3 by Arthrobac- ter P1 in aerobic conditions via pathway P1R3a. Formaldehyde was used by the microorganisms as a C source; therefore, two simultaneous, independent reactions for these processes can be written as:

1 P1R3a CH5N + O2 −−−−→ CH2O + NH3, (42) MTH 2

CH2O + O2 −−−−−→ CO2 + H2O. (43)

with CH2O as in Eqs. (35), (45), and (38).

Hippe et al. (1979) assessed MTH metabolization to CO2, CH4, and NH3 by Methanosarcina barkeri in anaerobic conditions via pathway P1R3b and the reaction for this process can be written as:

1 P1R3b CH5N + H2O −−−−→ CH4 + CO2 + NH3. (44) MTH 2

3 – + Pathway P2. starts with GLP aerobic cometabolic hydrolysis to SRC, PO4 , and H mediated by Achromobacter sp. MPS 12A (Sviridov et al., 2012), Arthrobacter sp. GLP-1 (Pipke et al., 1987), Arthrobacter atrocyaneus ATCC 13752 (Pipke & Amrhein, 1988a), Arthrobacter sp. GLP-1/Nit-1 (Pipke & Amrhein, 1988b), Pseudomonas PG2982 (Moore et al., 1983), and Streptomycete StC (Obojska et al., 1999) (pathway P2R1s, Figure 14).

61 3 – + Moore et al. (1983) assessed GLP biodegradation to SRC, PO4 , and H by Pseudomonas sp. PG2982 in aerobic conditions via pathway P2R1s; GLP was the only P source, while gluconate was used as a C source. As suggested by those authors, because this bacterial strain was able to grow on GLP being provided an additional C source, one reaction for this process was written as:

P2R1s 3− + C3H8NPO5 + H2O + CH2O + O2 −−−−→ C3H7NO2 + PO4 + 3 H + CO2 + H2O, (45) GLP SRC with CH2O as in Eq. (35).

Similarly to AMPA, GLP can undergo chemical degradation via pathway P2R1c after adsorption onto birnessite mineral surface; depending on the GLP-to-birnessite mass fraction ratio, diﬀerent byproducts can be produced in varying fractions. Here, it was assumed that SRC, 3 – + PO4 , and H were the only byproducts along this pathway, but also AMPA and other uncharacterized phosphate-containing chemicals were found in small amounts (Li et al., 2015) (pathway P2R1c, Figure 14). Again, GLP ﬁrst underwent adsorption onto birnessite, next Mn 3 – ions contained in the mineral catalyzed the chemical reaction. Finally, the ion PO4 was adsorbed onto birnessite as well. Sorption was described according to the Langmuir kinetic model (Langmuir, 1918) implemented as in Atkins & De Paula (2005), while GLP degradation was modeled using the Michaelis-Menten model. Therefore, in total three simultaneous, independent reactions for these processes were written as:

R1 C3H8NPO5 ←→ C3H8NPO5, (46) GLP(aq) GLP(ad)

1 P2R1c 3− + C3H8NPO5 + O2 −−−−→ C3H7NO2 + PO4 + H , (47) GLP(ad) 2 SRC

3− R3 3− PO4 (aq) ←→ PO4 (ad), (48)

Next, SRC can be oxidized either to glycine (GLY) and formaldehyde by Pseudomonas

Ovalis in aerobic conditions (Appleyard & Woods, 1956), or to MTH, CO2, and acetate by Eubacterium acidaminophilum using formate as the e – donor in anaerobic conditions (Hormann & Andreesen, 1989) (pathway P2R2a and P2R2b, respectively, Figure 14). Also in this case, acetate was converted into an equivalent number of CH2O moles, and the overall reaction was rewritten to account for the use of CH2O as a model C source in place of formate.

Appleyard & Woods (1956) assessed exogenous O2 uptake during SRC metabolization to

GLY and NH3 by Pseudomonas Ovalis in aerobic conditions via pathway P2R2a. As suggested by those authors, the reaction describing this process can be written as:

62 1 P2R2a C3H7NO2 + O2 −−−−→ C2H5NO2 + CH2O, (49) SRC 2 GLY

CH2O + O2 −−−−−→ CO2 + H2O. (50)

with CH2O as in Eqs. (35), (45), (38), and (43).

Hormann & Andreesen (1989) assessed SRC metabolization to MTH, acetate, and CO2 by Eubacterium acidaminophilum in anaerobic conditions via pathway P2R2b in the presence of formate (FRM) as the electron donor, which was replaced with CH2O for simplicity. Because the cells were able to grow on SRC the reaction for this process can be written as:

P2R2b + C3H7NO2 + CH2O + H2O −−−−→ CH5N + 2 CH2O + CO2 + 2 H . (51) SRC MTH with CH2O as in Eqs. (35), (45), (38), (43), and (50).

Appleyard & Woods (1956) assessed exogenous O2 uptake during GLY metabolization to

CO2, NH3, and H2O by Pseudomonas Ovalis in aerobic conditions via pathway P2R3a. As suggested by those authors, the reaction describing this process can be written as:

3 P2R3a C2H5NO2 + O2 −−−−→ 2 CO2 + NH3 + H2O. (52) GLY 2

Därre & Andreesen (1982a) assessed GLY metabolization to acetate and NH3 by Clostrid- ium purinolyticum in anaerobic conditions via pathway P2R3b. As suggested by those authors, the reaction describing this process can be written as:

1 P2R3b 3 C2H5NO2 + H2O −−−−→ CH2O + CO2 + NH3. (53) GLY 2 2

4.5.3. Results

Goodness of the fit between model and experiments The experiments retrieved from the literature and used to develop the GLP reaction network also allowed us to estimate the corresponding kinetic parameters, specific biomass affinity, and goodness-of-fit (Table 6). The comparison between observed and predicted concentrations are reported in Figures B1 to B6. The index in the top corner of each plot identifies the corresponding laboratory experiment (Test in Table 6). Modeled concentrations achieved R2 ranging between 0.14 and 0.99, and NRMSE spanning from 0.71% to 40.33%, respectively (Table 6, columns 14 and 15). Excluding the case of aerobic formaldehyde consumption by Arthrobacter P1, which led to R2 = 0.14 and NRMSE = 40.33% (Table 6, test 6), parameter estimations returned R2 > 0.98 and NRMSE <7.74% in

63 all other cases. These R2 and NRMSE values suggest that the kinetic equations and parameters were appropriate to describe the biodegradation reactions along each pathway.

4.5.4. Discussion

Few experiments were available to estimate the MMM kinetic parameters corresponding to GLP biodegradation. More importantly, there was only one experiment for reactions P1R1 and P2R1s; therefore, variability amongst different strains could not be investigated. The MMM parameters corresponding to AMPA biodegradation may be susceptible of high uncertainty because they were estimated from an experiment that showed AMPA production and incomplete biodegradation (Balthazor & Hallas, 1986). Only one experiment was available also for the description of GLP and AMPA chemical degradation along P2R1c and P1R2c, respectively. Li et al. (2015) and Paudel et al. (2015) reported that multiple uncharacterized metabolites containing phosphorus were produced and degraded during their experiments. Their toxicity was unknown and they may not quickly degrade under environmental conditions, where they might persist. In the GLP reaction network, biotic and abiotic degradation reactions are directly con- 3 – nected by feedbacks through which PO4 produced along P2R1s, P1R2s, P2R1c, and P1R2c can both self- and cross-inhibit the reactions just listed. This consideration is extremely important for a robust prediction of GLP and AMPA fate in the environment. Other possible sources of uncertainties related to the MMM parameters of the GLP reaction network were discussed in Section 4.4.4 relatively to the ATZ reaction network. Those uncertainties may regard the assumption of constant values for the MMM parameters and cell mortality. Section 4.4.4 also discussed that bacteria may enhance biodegradation as a result of evolutionary processes driven by exposure to herbicides; the same reasoning could apply to the GLP reaction network. In 3 – particular, the loss of the inhibitory effect of PO4 on GLP biodegradation along P2R1s and AMPA biodegradation along P1R2s may substantially affect the switch between pathways P1 and P2. Yet, the appearance of bacteria which can use AMPA as the only C source would eliminate the dependence on additional C sources for effective AMPA biodegradation.

64 Kinetic Parameters

0 2 µ KKI YYB0 Φ t1/2 R NRMSE

Test Source Exp. condition Pathway Substrate (s) (M) (M) (g-C-Bio g-C-Subs−1) (mg-wet-Bio mol-Subs−1) (mg L−1) (s) (d)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1 (a) Flavobacterium sp. AERO P1R1s GLP (Eq. 35) 2.95 × 10−5 4.12 × 10−4 8.31 × 10−2 1.57 × 10−6 1.94 0.99 3.07 GD1

× −5 × −4 × −2 × 3 × −2 × −6 AERO P1R1s CH2O (Eq. 35) 2.95 10 1.38 10 1.58 10 3.80 10 8.31 10 4.67 10 --- AERO P1R2s AMPA (Eq. 38) 5.04 × 10−6 2.08 × 10−3 2.53 × 10−4 8.31 × 10−2 4.87 × 10−8 - 0.99 4.28

× −6 × −4 × −2 × 3 × −2 × −7 AERO P1R2s CH2O (Eq. 38) 5.04 10 1.38 10 1.73 10 4.14 10 8.31 10 7.31 10 ---

2 (b) Pseudomonas sp. AERO P1R1s GLP (Eq. 35) 3.40 × 10−5 1.67 × 10−3 2.14 × 100 9.60 × 10−7 2.05 0.99 1.81 LBr

× −5 × −4 × −1 × 4 × 0 × −6 AERO P1R1s CH2O (Eq. 35) 3.40 10 1.14 10 1.89 10 4.54 10 2.14 10 1.04 10 ---

3 (c) Agrobacterium AERO P1R1 GLP (Eq. 36) 3.35 × 10−5 4.05 × 10−3 3.86 × 10−2 2.78 × 104 1.53 × 10−1 4.55 × 10−8 3.96 0.98 7.74 radiobacter and Achromobacter Group V D

4 (d) Pseudomonas AERO P2R1s GLP (Eq. 45) 3.34 × 10−5 1.09 × 10−4 1.89 × 10−2 1.58 × 10−7 2.76 0.99 2.08 PG2982

× −5 × −4 × −1 × 4 × −2 × −8 AERO P2R1s CH2O (Eq. 45) 3.34 10 2.12 10 1.52 10 3.65 10 1.89 10 8.14 10 ---

5 (a) Flavobacterium sp. AERO P1R1s GLP (Eq. 35) 2.95 × 10−5 4.12 × 10−4 8.31 × 10−2 1.57 × 10−6 1.94 0.99 3.07 GD1

× −6 × −4 × −2 × 3 × −2 × −7 AERO P1R2s CH2O (Eq. 38) 5.04 10 1.38 10 1.73 10 4.14 10 8.31 10 7.31 10 ---

6 (e) Arthrobacter P1 AERO P1R3a MTH (Eq. 42) 1.39 × 10−4 2.15 × 10−4 2.18 × 10−5 2.66 × 10−3 6.39 × 102 3.91 × 100 3.95 × 10−3 0.49 0.99 2.59

AERO P1R3a FRMH (Eq. 43) 1.86 × 10−2 1.13 × 10−1 2.12 × 10−2 5.10 × 103 3.91 × 100 1.26 × 10−4 0.63 0.14 40.33

7 (f) Methanosarcina ANAERO P1R3b MTH (Eq. 44) 1.17 × 10−4 5.38 × 10−1 1.29 × 10−3 3.09 × 102 1.60 × 100 1.13 × 10−6 1.90 0.99 1.58 barkeri

8 (g) Pseudomonas Ovalis AERO P2R2a SRC (Eq. 49) 5.73 × 10−3 3.88 × 10−5 2.29 × 10−3 1.65 × 103 8.09 × 10−3 7.24 × 10−4 0.01 0.99 0.71

AERO P2R2a FRMH (Eq. 50) 5.23 × 10−4 2.21 × 10−5 5.61 × 10−4 1.35 × 102 8.09 × 10−3 1.42 × 10−3 ---

9 (g) Pseudomonas Ovalis AERO P2R2a SRC (Eq. 49) 5.13 × 10−3 3.66 × 10−5 2.76 × 10−3 1.99 × 103 8.45 × 10−3 5.97 × 10−4 0.01 0.99 0.98

AERO P2R2a FRMH (Eq. 50) 3.93 × 10−4 4.48 × 10−5 1.16 × 10−3 2.78 × 102 8.45 × 10−3 2.67 × 10−4 ---

10 (g) Pseudomonas Ovalis AERO P2R2a SRC (Eq. 49) 1.39 × 10−3 2.57 × 10−5 2.46 × 10−3 1.77 × 103 1.65 × 10−2 5.07 × 10−4 0.02 0.99 1.01

AERO P2R2a FRMH (Eq. 50) 4.05 × 10−4 3.89 × 10−5 2.35 × 10−3 5.64 × 102 1.65 × 10−2 3.06 × 10−4 ---

11 (h) Eubacterium aci- ANAERO P2R2b SRC (Eq. 51) 5.36 × 10−5 6.87 × 10−2 4.46 × 10−3 3.21 × 103 5.43 × 101 1.32 × 10−5 0.65 0.99 3.45 daminophilum

ANAERO P2R2b FRM (Eq. 51) 4.39 × 10−3 2.06 × 10−4 0.71 0.99 2.30

12 (g) Pseudomonas Ovalis AERO P2R3a GLY (Eq. 52) 1.45 × 10−4 9.98 × 10−5 7.45 × 10−4 3.58 × 102 3.85 × 10−3 1.56 × 10−5 - 0.99 4.63

13 (g) Pseudomonas Ovalis AERO P2R3a GLY (Eq. 52) 9.90 × 10−5 1.13 × 10−4 2.98 × 10−4 1.43 × 102 5.70 × 10−3 3.49 × 10−5 0.23 0.99 4.63

14 (i) Clostridium puri- ANAERO P2R3b GLY (Eq. 53) 2.20 × 10−4 2.94 × 10−1 9.25 × 10−5 4.44 × 101 2.08 × 10−1 3.50 × 10−6 0.61 0.99 1.91 nolyticum

r r r r t R2 NRMSE ads des1 des2 1/2 (M s−1) (M s−1) (M s−1) (M s−1) (d)

15 (l) birnessite R1 GLP (Eq. 46) - 2.08 × 10−2 1.03 × 10−2 1.17 × 10−2 3.95 - -

P2R1c GLP (Eq. 40) 2.67 × 10−3 0.06 - -

3 – × −2 × −2 × −3 R3 PO4 (Eq. 48) 2.01 10 2.99 10 1.36 10 - 0.99 7.46

16 (l) birnessite R2 AMPA (Eq. 39) - 2.63 × 10−1 1.59 × 10−4 1.47 × 10−2 3.95 - -

P1R2c AMPA (Eq. 40) 1.51 × 10−5 0.44 - -

3 – × −3 × −5 × −4 R3 PO4 (Eq. 41) 1.65 10 9.90 10 1.83 10 - 0.99 5.87

Table 6: Estimated kinetic parameters for GLP and AMPA degradation and MTH, SRC, and GLY metabolization grouped by microbial isolate or community for reaction pathways of the GLP degradation reaction network. The specific biomass affinity Φ0 (la Cecilia & Maggi, 2016) is also tabulated together with the goodness-of-fit against experiments from: (a) Balthazor & Hallas (1986); (b) Jacob et al. (1988); (c) Mcauliffe et al. (1990); (d) Moore et al. (1983); (e) Levering et al. (1984); (f) Hippe et al. (1979); (g) Appleyard & Woods (1956); (h) Hormann & Andreesen (1989); (i) Därre & Andreesen (1982a); (l) Li et al. (2015). Temperatures T ranged between 25 and 37◦C in the experiments. AERO and ANAERO refer to experiments carried out in aerobic and anaerobic conditions, respectively, while birnessite is the mineral containing Mn ions, which catalyzed GLP and AMPA chemical degradation.

65 4.6. Summary

This chapter reported the development of the reaction networks for the herbicides ATZ and GLP and the calibrated values relative to the MMM kinetic equations describing each of the biologically-mediated reactions. The reaction networks are characterized by the presence of interactions and switches expressed through substrate competitive consumption and inhibitory effects. MMM equations were able to fit well the observed data relative to bacteria-mediated biodegradation reactions because they explicitly account for microbial dynamics (i.e., microbial biomass growth), while first-order kinetics may yield a poorer fit because they implicitly assume a constant microbial biomass concentration. Generally, the MMM framework is suitable for mechanistically describing interactions and switches, and their use is therefore suggested in more comprehensive environmental models.

66 5. Modeling

5.1. Introduction

In previous chapters, it was discussed that living microorganisms may evolve strategies to make use of herbicides for their benefits, which can substantially affect herbicides fate. Some numerical simulators allow one to adequately describe microbial dynamics and their activity towards herbicides biodegradation. Relative to GLP and AMPA degradation, first-order kinetics have been used in earlier numerical approaches in Ghafoor et al. (2011), Guijarro et al. (2018), and Simonsen et al. (2008), and in the PELMO and MACRO models to carry out the initial risk assessment for regulating GLP use in Europe (EFSA, 2015). However, Cheyns et al. (2010) already suggested that MMM kinetics should be used in place of first-order kinetics in pesticide fate model. The authors used atrazine biodegradation in soil mini columns to prove that first-order kinetics may underestimate pesticide leaching. It was evident that additional kinetic mechanisms can control pesticide degradation in soil, which are not necessarily included in MMM-type kinetics, and are certainly omitted in first-order kinetics; these may include multiple inhibitions and competitions for substrates and byproducts, which can be accounted for by means of MM terms. It follows that reactive transport solvers solving MMM kinetics provide a better tool to support robust decision-making for farmers, regulatory bodies, and EPAs to sustainably manage and protect humans health and the environment. This chapter introduces the bioreactive transport solver used to carry out the numerical simulations relative to the ATZ and GLP reaction networks developed in Sections 4.4.2 and 4.5.2; their corresponding kinetic parameters were averaged by reaction pathways as further explained in the Methods in the next sections. The reaction networks were coupled with the N cycle proposed by Maggi et al. (2008) and accounted for a continuous release of an additional carbon source to provide a more comprehensive, ecologically-orientated framework to assess herbicides biodegradation and contamination under real case scenarios described in the following sections. Simulations were run using the model BRTSim (a description of the crucial capabilities of the software are presented in Section 5.2). It is worth to mention that other scientists are moving toward a similar approach proposed in this thesis such as the model PECCAD (PEsticide degradation Coupled to CArbon turnover in the Detritusphere) presented in Pagel et al. (2014).

5.2. Bioreactive Transport Simulator BRTSim

BRTSim-v2.2 (based on Maggi, 2015) is a 1-D general-purpose multiphase and multicomponent model suitable to simulate biogeochemical reaction-advection-diffusion processes in unsaturated soil systems. The soil moisture dynamics are dealt with using a finite volume scheme that solves the Richards equation along the vertical direction. BRTSim can account for any number of chemical and biological species. Primary and secondary chemical species can be defined, the latter being in equilibrium with the corresponding primary species. Equilibrium

67 reactions can be deﬁned for aqueous complexation, ion exchange, gas dissolution, and mineral adsorption and are calculated in BRTSim using the mass-action law. Transport of chemical species is accounted for by the Darcy’s advection and Fick’s diﬀusion. Note that advection of gas species in the gas phase was neglected given the time scales of interest in this work. Chemical and biological reactions involving primary species and microbial functional groups, described in this solver as primary species, are accounted for by means of the MM and MMM kinetic equations, respectively (Bekins et al., 1998; Belser, 1979; Monod, 1949).

68 5.3. Atrazine

In the following subsections, the numerical results showing in-situ biodegradation and dispersion of ATZ and its metabolites are presented. Contents of this chapter come from the article la Cecilia & Maggi (2017a)5 published in the Journal of Contaminant Hydrology.

5.3.1. Introduction

Using the ATZ reaction network in soil developed in Section 4.4.2, this chapter shows comprehensive modeling results, which highlight the high-level of process coupling under real-case ATZ application scenarios in agricultural lands. In particular, this chapter highlights the feedbacks amongst ecohydrological boundary conditions, ATZ biodegradation, and leaching in soil. Reaction pathways were assumed to be carried out by microbial functional groups; the latter gathered together bacterial strains performing the same process, which kinetics were described using averages of reaction pathway-speciﬁc MMM parameter values estimated in Section and reported in Table . Additional processes were included to account for physical, biogeochemical, and ecological feedbacks on molecules fate and microbial dynamics. Those included linear sorption, aqueous complexation, gas dissolution, the addition of three microbial functional groups involved in the N cycle (Maggi et al., 2008), and explicit competition amongst microbial functional groups for the additional C source in the form of CH2O. The model structure levered the validation-by construct principle (McCarl & Apland, 1986) and the estimated parameters were used without adjustments. MM and inhibition terms were used to explicitly account for

O2(aq) and pH effects on aerobic and anaerobic reactions. Sensitivity analyses were run to predict the biogeochemical response to increasing ATZ application rates and availability of an additional C source. All the equilibrium and kinetic reactions implemented in this biochemical system are reported in Table 10 with their corresponding parameters. There were a number of processes which were not distinguished in the presented complex system because of the lack of experiments suitable for developing a mechanistic model. The processes are briefly discusses in the next sections and possibly concerned toxicological effects by ATZ on soil microbial communities which would inhibit biodegradation, biodegradation of adsorbed ATZ, biodegradation by fungi and plants, and chemical degradation.

5.3.2. Methods

ATZ reaction network coupled with N cycle. A comprehensive biodegradation reaction network of atrazine (ATZ) and its 18 metabolites was coupled with the nitrogen cycle as proposed by Maggi et al. (2008) (Figure 15) and integrated in the computational solver BRTSim (Maggi,

5la Cecilia, D. and Maggi, F. (2017). In-situ atrazine biodegradation dynamics in wheat (Triticum) crops under variable hydrologic regime. Journal of Contaminant Hydrology, 203, pp. 104-121, http://dx.doi.org/10.1016/j.jconhyd.2017.05.004.

69 2015) to assess the in-situ biodegradation eﬀectiveness and leaching along a 5 m deep soil cultivated with wheat in West Wyalong, New South Wales, Australia.

Equilibrium adsorption reactions. Clay & Koskinen (1990) and Vryzas et al. (2007) measured ATZ, HOATZ, DIATZ, and DEATZ sorption in Waukegan and Ves soils, and a clayey alluvial soil used for cropping maize, respectively. Those six laboratory experiments reporting adsorption isotherms were used to estimate the corresponding linear equilibrium constants Kd (Table 7) between the substrate dissolved and adsorbed phases by means of inverse problem solution (Figure 16). First, observations were ﬁtted with the Langmuir model as

Ce Qe = Qmax · KL · , 1 + KL · Ce −1 −1 whereQe (mg kgdry-soil) is the mass of adsorbed solute per unit of adsorbent, Qmax (mg kgdry-soil) −1 is the mass of solute required to form a monolayer and ﬁll the surface, KL (L mol ) is the −1 Langmuir equilibrium constant, and Ce (mol L ) is the solute equilibrium concentration in the solution. Next, nonlinearities in the adsorption process were approximated with the tangent line to the Langmuir isotherm in the low adsorption-low concentration range as

Qe Kd = , Ce −1 where Qe (mol kgdry-soil) is the number of moles of adsorbed solute per unit mass of adsorbent, −1 and Ce (mol kg ) is the equilibrium solute concentration. Parameters relative to the nonlinear H2O Langmuir model together with their goodness-of-ﬁt, which was measured by means of the co- eﬃcient of determination (R2) and the normalized root mean squared error percent (NRMSE), and to the linearized approach are reported in Table 7.

A linear Kd was used because the concentrations of those compounds in soils are very low and the tangent of a ﬁt on Langmuir model in the low concentration-low adsorption range was representative of soil buﬀer capacity for those compounds.

The equilibrium constant Kd for CYA, ETA, and IPP (Table in 10) was estimated from their corresponding organic carbon-water partition coeﬃcient KOC retrieved from the literature as

SOM K = K × TOC% = K × d OC OC 1.72 where TOC% is the percent average total organic carbon in the soil, and it was estimated from the available soil organic matter in the test site (Hoyle, 2015) (Table 8).

Experimental site. Assessment of ATZ biodegration, integrated with the N cycle as proposed in Maggi et al. (2008), was numerically estimated in an agricultural plot at West Wyalong (about 470 km West of Sydney, 3305500S; 14701300E), New South Wales, Australia. The soil properties were retrieved from the Harmonized World Soil Database (FAO et al., 2012) and are reported in Table 8. The soil texture in the test site is clay in the top 1.5 m, loamy sand from 1.5 m to 5 m depth, and shale down to 50 m. The shale separates the intermittent shallow aquifer

70 Test Source Soil Sand-Silt-Clay OC Substrate Q K K R2 NRMSE max,Langmuir ads ,Langmuir d (%) g mg mg (-) kgsoil g l 1 2 3 4 5 6 7 8 9 10 11 1 (a) Waukegan silt loam 17-60-23 26.4 ATZ 5 × 10−2 5.39 × 10−2 4.73 × 10−1 0.99 3.22 2 (a) Ves clay loam 43-30-27 27.7 ATZ 3.98 × 10−2 1.192 × 10−1 5.55 × 10−1 0.99 3.71 3 (b) Alluvial deposit 19-17-64 0.82 ATZ 5.60 × 10−3 6.708 × 10−1 5.15 × 10−1 0.99 4.23 4 (b) Alluvial deposit 19-17-64 0.82 HOATZ 7.01 × 10−1 6.500 × 10−3 6.55 × 10−1 0.99 2.63 5 (b) Alluvial deposit 19-17-64 0.82 DIATZ 1.49 × 100 1.500 × 10−3 4.09 × 10−1 0.99 2.97 6 (b) Alluvial deposit 19-17-64 0.82 DEATZ 5.00 × 10−4 9.473 × 100 6.58 × 10−2 0.89 13.17

Table 7: Estimated adsorption parameters for ATZ, HOATZ, DIATZ, DEATZ of nonlinear Langmuir equilibrium Qmax,Langmuir and Kads,Langmuir, and the corresponding linearized equilibrium constant Kd. OC refers to organic carbon. Parameters are tabulated together with the goodness-of-ﬁt against experiments in Figure 16 from: Clay & Koskinen (1990) and Vryzas et al. (2007). Experiments were assumed to be carried out at standard conditions (T = 25◦).

from the deep aquifer. Given the negligible interaction between the two aquifers, the shale was assumed impermeable after Bilge (2012) and no ﬂow was considered below 5 m depth. The soil hydraulic parameters were estimated using the closed-form equations by Van Genuchten (1980). The average rainfall and potential evapotranspiration rates were 462 mm/year and 1770 mm/year, respectively, in the period 1990-2015 (Bureau of Meteorology, 2016). The potential

evapotranspiration (ET0) was calculated using the approach in Allen et al. (1998) (Figure 17).

The actual crop evapotranspiration was calculated as ETC = ET0 ×KC, with the time-varying

crop coeﬃcient KC for wheat equals to 0.5, 0.7, 0.8, and 0.6 in June, July, from August to

October, and in November, respectively, and with KC = 0.3 during the fallow season (Grain,

2008). Weekly averaged precipitation, actual evapotranspiration (ETC), and irrigation were used as upper boundary conditions (Figure 17). Crop emergence and harvest were assumed to occur at the end of April and November, respectively (Stapper, 2007). The average root depth was equal to 0.3 m and the roots density was distributed with a negative exponential + – function down to 1.5 m depth (Bowdena et al., 2008); evapotranspiration regulated NH4 , NO3 , – – + HCO3 , Cl , and H plant uptake. Irrigation was applied during the crop season (Hope, 2003). The analyses were carried out after the biochemical system reached an equilibrium over the root zone, which required nearly 70 simulated years. Therefore, meteorological observations spanning 26 years were repeated 5 and 25 times to construct boundary conditions of 120 and 500 years, respectively; daily precipitations that exceeded 15 years return time were substituted with the mean value to avoid repetition of unlikely events. Rainfall loss by crop interception was considered equal to 20%. Detailed data on the microbial ecology at the test site were not available, though it is likely that ATZ biodecomposers included in Figure 15 inhabited the top soil where ATZ is currently

periodically applied. Although a C source was continuously supplied as CH2O, the biomass concentrations of a microbial functional group j may become 0 for some MMM kinetic parameter values combination (Porta et al., 2018). To maintain the microbial diversity over time, each functional group was set to have a biomass concentration greater or equal than a minimum value.

This was achieved by resorting to a biomass background recovery rate rB, j calculated neglecting

microbial growth from the breakdown of ATZ and its byproducts as dB j/dt = −δ jB j + rB, j = 0, ∗ ∗ hence returning rB, j = δ jB j, with B j the equilibrium concentration for B j in the absence of ATZ-

71 ∗ speciﬁc C sources. Values of B j were chosen such that BAER/BATZhyd = BAER/BATZoxi = 1000

(Alvey & Crowley, 1997), while BAOB/BAER = BAOB/BNOB = 5 and BDEN/BAOB = 10 (de Boer & Kowalchuk, 2001; Fukumoto et al., 2006; Roux-Micholleta et al., 2008; Tang, 2016; Tatti et al., 2014; Wertz et al., 2012). The minimum soil biomass concentration was therefore imposed, while the instantaneous biomass concentration varied according to substrates availability and consumption.

Competition was included in all the reactions where the same functional group B j was able to consume more than one substrate. In particular, BATZoxi competed for ATZ in P2R1 and

P3R1, and for DEATZ in P3R2, while BATZhyd competed for ATZ in P1R1a and P1R1b, and for HOATZ, NIPA, DIATZ, DIHOATZ, DHONATZ, DIDEATZ, CLHOATZ, and DHOATZ in

P1R2, P1R3, P2R2, P2R3, P2R4, P3R3, P3R4, and P3R5, respectively. BAER competed for CYA, BIU, ALO, ETA, IPA, and IPP in P4R1, P4R2, P4R3, P5, P6R1, and P6R2, respectively.

Competition for O2(aq) and CH2O were not explicitly included. The eﬀect of pH on microbial activity was explicitly taken into account in all biologically-mediated reactions using a MM term for high pH with constant 10−8 M, and an inhibition term for low pH with constant 10−6 M, −4 respectively (Boon & Laudelout, 1962). Where O2(aq) consumption occurred a K = 1.5×10

M was used (Gerritse et al., 1992). O2(aq) inhibition on oxidative reactions P2R1, P3R1, and P3R2 was explicitly included using a MM term with constant 2.5×10−6 M (Gerritse et al., 1992; Hao et al., 1983; Stolper et al., 2010). Finally, the soil organic matter (SOM), calculated using Eq. (5.3.2) from the total organic −1 −1 carbon (TOC) measured at the location (Table 8), was converted from g kgdry-soil to mol L and + assumed to slowly release CH2O and NH4 with a ﬁrst-order kinetic reaction with rate constant −11 −1 + r = 1 × 10 s . That is, the molecular mass of 1 mol of CH2O plus 0.039 moles of NH4 have a molecular mass of 30.702 g mol−1; 1 kg of soil contains 6.71 g of SOM (from TOC in Table 8), and therefore, 0.22 moles of SOM; given the soil properties of Table 8 and assuming a soil 3 water saturation of 0.25, the numerical node of volume 0.30 m can contain 452 kgdry−soil and −1 35.25 L of H2O, and therefore 2.79 mol L of SOM. SOM was assumed to be constant over time as a result of recharge from root mortality, exudates, and other debris (Riley et al., 2014).

Note that CH2O was also released by ATZ, DEATZ, ETA, and IPP biodegradation reactions (P2R1, P3R2, P5, and P6R2, Figure 15). Generally, bacteria are known to adapt via genetic mutations to degrade new anthropogenic substances and to enhance their degradation eﬃciency next. In the model, kinetic parameters relative to adapt bacteria were used and those values were kept constant because modeled scenarios covered longer time scales than those involved in acquiring enhanced ATZ-degrading capabilities (Shaner et al., 2007; James et al., 2010; Krutz et al., 2010b). Plants can dealkylate or hydrolyze ATZ in their roots and shoots after passive uptake. Shimabukuro (1967) showed that nearly 35% of applied ATZ was degraded in wheat in optimal conditions, 25% was hydrolyzed to HOATZ by benzoxazinone produced by wheat, while 10% encompassed uncharacterized byproducts as well as DIATZ and DEATZ. Wheat-mediated ATZ degradation was not taken into account because not only it would contribute to a smaller

72 extent than microbiological degradation but also because it depends on benzoxaninone levels in the plant.

West Wyalong Deptha (m) 0-1.5 1.5-5 Soil taxonomya Calcisols Calcisols Soil texturea Clay Loamy sand Sand-Silt-Clay fractiona 15-28-57 80-10-10 Organic C fractiona (g-C/kg-soil) 3.9 3.9 Mineral densityb (kg/m3) 2849 2849 Bulk densitya (kg/m3) 1370 1350 Porosityb 0.47 0.50 Pore size distributionb 1.43 1.74 Air entry suctionb (m) 2.14×10−4 2.37×10−4 Permeabilityb (m2) × 10−12 0.16 1.58

Table 8: Soil and hydraulic parameters at West Wyalong, NSW, Australia. a From the Harmonized World Soil Database (FAO et al., 2012). b Estimated using (Van Genuchten, 1980).

Analysis of ATZ application scenarios. Scenarios were implemented to assess whether nonlinear degradation occurred as a function of different ATZ application rates. In the beginning of December, after wheat harvesting and 4 months before sowing, a reference yearly application rate of 2 kg/ha was used. All the organochlorides were tracked to determine their distribution and their phase partitioning over the soil profile including the vadoze zone and the intermitted superficial aquifer above the shale layer. Tracking was repeated also for ATZ application rates equal to 60%, 80%, 125%, 150%, 200%, and 300% of the reference rate. Numerical simulations covered 120 years; analyses were carried out using modeled concentrations relative to the last 26 years, when the biochemical system was at equilibrium within the root zone.

Statistical analysis of ATZ breakthrough curves. The 26 repetitions used in this analysis were characterized by ecohydrological boundary condition corresponding to a diﬀerent year during dynamic equilibrium. ATZ breakthrough curves at four soil depths within the root zone were deﬁned as the time history of ATZ mass concentration CATZ(z) after a peak CATZ0 (z) was found at each depth z. Times t(z) when CATZ(z)/CATZ0 (z) was between 0.99 and 0.01 were recorded.

ATZ half-life t1/2(z) was then calculated as the time necessary by the biochemical system to meet the condition CATZ(z)/CATZ0 (z) ≤ 0.5. Byproducts HOATZ, DIATZ, and DEATZ masses were monitored to calculate the percentage of biodegraded ATZ at the same times t(z) relative to the ATZ annual reference application.

5.3.3. Results and discussion

ATZ biodegradation rate and Speciﬁc biomass aﬃnity. In the reference simulation (2 kg/ha/year), aerobic ATZ hydrolysis (P1R1a) was the main mechanism for ATZ biodegradation (79.2%); anaerobic hydrolysis (P1R1b) biodegraded 18.2% of applied ATZ, while oxidative reactions P2R1 and P3R1 contributed nearly 0.11 and 0.15%, respectively (Table 9). HOATZ, which does not contain Cl, was the main byproduct of ATZ hydrolysis (P1R1a and P1R1b), and organochlorides mass was reduced to 2.6% of the total applied mass. Krutz et al. (2010b)

73 assumed the formation of HOATZ, DIATZ, and DEATZ to be 71, 7, and 22% of the applied ATZ, respectively, in soils with a long history of ATZ, while HOATZ mass fraction could be smaller than 10% in soils with no previous exposure. 0 −5 −1 ATZ hydrolyzers (BATZhyd, P1R1a and P1R1b) had an average Φ = 1.12 × 10 s , while 0 −6 −1 ATZ oxidizers (BATZoxi, P2R1 and P3R1) had a lower average Φ = 1.21 × 10 s (Table 9). Despite the lowest Φ0, P1R1a was the dominant pathway in the modeled environmental conditions, which may have favored the growth of BATZhyd over BATZoxi because BATZhyd growth did not depend on O2 concentration.

Reaction Pathway Biodegraded Φ0 = µ/K × Y type mass fraction (%) (s−1) Hydrolysis P1R1a 79.2 3.17 × 10−7 Hydrolysis P1R1b 18.2 2.20 × 10−5 Oxidation P2R1 0.1 9.90 × 10−7 Oxidation P3R1 0.1 1.43 × 10−6 Total 97.7

Table 9: ATZ biodegraded mass fractions relative to the total applied mass after 100 years of 2 kg/ha/year ATZ applications. Note that, Φ0 is calculated using B =1 mg/l.

ATZ distribution sensitivity analyses. A range of ATZ yearly applications rates were used to highlight nonlinearities in the distribution of undegraded Cl-containing molecules (ATZ, DI- ATZ, DEATZ, DIDEATZ, and CLHOATZ) and ATZ biodegradation rate along the soil profile. Over 100 years, about 19.4% by mass of those compounds leached and accumulated below the root zone as adsorbed (about 12.1%) and dissolved (about 7.3%) compounds (Figure 18a); about 80.6% by mass of organochlorides remained in the root zone (57.0% adsorbed and 23.6% dissolved). Biodegradation of adsorbed compounds was not accounted for in this work after Johnson & Truex (2006), Ogram et al. (1985), Riley et al. (2014), Xu et al. (2011), and Yu & Zheng (2010); Park et al. (2003) suggested that microorganisms may metabolize adsorbed substrates as well, thus we recognize that this aspect should be better investigated as it may affect compounds distribution and residence time in soil. A 2.6% by mass of Cl was still included in organochlorides resulting from ATZ application and biodegradation (Figure 18b, note the log-scale in the y-direction); of this, 66.7% was ATZ, 32.7% was DEATZ, and only about 0.04% was DIATZ. Phytotoxic DIDEATZ and CLHOATZ byproducts were 0.28 and 0.27% of the organochlorides mass in soil (Figure 18b). Field measurements in similar environmental conditions corroborate that ATZ is the main organochloride in soil; however, DEATZ concentration was found to be maximum three times greater than DIATZ (Rattray et al., 2008). Overall, an increasing ATZ application rate did not greatly affect organochlorides vertical distribution and partitioning into dissolved and adsorbed phases (Figure 18a). ATZ biodegraded mass fraction increased slightly but nonlinearly from 97% to 98% with greater ATZ application rates, resulting in a lower risk of ATZ leaching from the root zone to the intermitted aquifer

74 (Figure 18b) (Bowmer, 1991). Those trends reversed for ATZ application rates greater than 400% of the reference rate; however, a 8 kg/ha/year ATZ application rate is not common, and this case was not investigated further.

ATZ tracking and breakthrough curves. ATZ degradation was the fastest at 0.45 m depth (Fig- ure 19). ATZ breakthrough curves showed great variability between minimum and maximum values. The average time necessary to remove 95% of ATZ relative to the mass peak ﬁrst decreased from 322 days down to 186 days near surface and at 0.45 m depth, respectively, then it increased up to 348 days at 1.05 m depth (Figure 19b). ATZ did not undergo biodegradation in lower soil depths, including in the intermitted aquifer. On average, ATZ biodegradation significantly contributed to ATZ mass peak removal (Figure 20b); when 95% of the ATZ mass peak was removed, biological processes contribution decreased from 38% down to 34% at 0.15 m and 0.45 m depth, respectively, while it increased up to 48% at 1.05 m depth. This highest ATZ biodegraded mass fraction may be due to ATZ long residence time at this depth, which allowed for a persistent ATZ biodegradation. In this framework, ATZ biodegradation time scale may be overestimated after Zablotowicz et al. (2008) reported that 80% of applied ATZ can be mineralized within 30 days in soils previously exposed to ATZ. Note that enhanced ATZ biodegradation was not taken into account in this work; yet, modeled environmental conditions play an important role in determining the microbial activity and concentration.

ATZ half-life and underlying processes. The ATZ minimum and maximum half-life ranged from 150 to 247 days in the soil surface, and from around 52 to 168 days at 0.45 m depth (thick vertical black line in Figure 19a and c). These values are higher than those reported in the literature from simulated and observed data (Krutz et al., 2010b), but in the same order of magnitude with values relative to sandy soils in semi-arid areas (Kookana & Baskaran, 1998; Bowmer, 1991). Both biodegradation and advection processes affect ATZ removal time scale. Microbial biomass was effective in degrading ATZ within the root zone (Figure 20) but a 4.5% residual mass of applied ATZ (not degraded) underwent advection towards lower soil depths. The different contributions of advection and biodegradation processes to ATZ removal affected ATZ degradation time scale. Generally, slow ATZ transport through the soil allowed for longer ATZ half-life values but greater amounts of ATZ were biodegraded, thus reducing the risk of ATZ accumulation at lower depths. Dynamic ecohydrological boundary conditions may explain this wide range in t1/2 in the top soil.

The biomass concentration B0(z) at time t = t0 and depth z, corresponding to the occurrence of the peak ATZ mass concentration CATZ0 (z), was used to calculate Φ0(z) of Eq. (10) in each diﬀerent ecohydrological boundary conditions. The relationship between ATZ half-life t1/2 and

Φ0 for both ATZ hydrolyzers and oxidizers was uncertain in the top soil (R = -0.53 and R

= -0.53, respectively) (Figure 21). The Φ0 for BATZhyd and BATZoxi diﬀered by 3 orders of magnitude, and this may explain the greater contribution of BATZhyd to ATZ degradation than

BATZoxi (Table 9). Therefore, measurements of ATZ byproducts concentration in soil can be used to detect the functional group involved in ATZ degradation.

75 Effect of ATZ degradation on pH. Two simulations with and without yearly ATZ applications at the reference rate were run to understand whether ATZ and its degradation can affect soil pH. The system was initialized at pH = 7 and simulations covered a 120-year-long period. ∆pH ≤ 0.6 were found, thus highlighting that soil acidity is not substantially affected by ATZ and related biodegradation processes (Figure 22c).

ATZ contamination of the aquifer. One 500-year-long simulation was run to assess the impacts of yearly ATZ applications at the reference rate for 100 years to the aquifer. Modeling showed that the intermittent aquifer rose to 2 m below ground level. ATZ reached this depth with a −6 concentration of nearly 1×10 mg/kgdry-soil after 10 years following yearly applications (Figure 23), then it was slowly advected to deeper soil depths. In 200 years, ATZ reached the shale −6 −6 formation with a concentration of nearly 1×10 mg/kgdry-soil and it increased up to 4×10 mg/kgdry-soil. Modeling showed that degradation processes were negligible at these depths; the only mechanism to decrease ATZ contamination was by ATZ advection downwards through the soil proﬁle. More importantly, leaching poses a risk of groundwater contamination. Our results are in line with ATZ concentrations above 0.0001 mg/l being measured in shallow aquifers in certain areas in Italy nowadays (ISPRA, 2016), where ATZ has been banned since 1991.

76 aq, Atrazine p Head compound C8H14ClN5 Intermediate ATZ compound Aerobic Aerobic

H O Anaerobic 2 O Final Aerobic 2 product A P1R1b O2 H2O B P2R1 H+ B P3R1 Cl- Nitrogen P1R1a A CH2O H2O - + Cl H CO2 cycle DIATZ DEATZ aq, Deisopropyl- aq, Deethylatrazine aq, Hydroxyatrazine H O O2 p 2 atrazine p C H ClN p C8H15N5O 6 10 5 C H ClN Organochloride P2R2 A 5 8 5 B P3R2 HOATZ CH2O H2O - + DIDEATZ NIPA Cl H DIHOATZ Compound phase A aq, P1R2 Deisopropyl- aq, N-isopropylammelide aq, Deisopropyl- aq, g, (aqueous, gaseous, p deethylatrazine p H2O p C6H10N4O2 H2O hydroxyatrazine p C2H7N protected) C3H4ClN5 A C5H9N5O A H2O P2R3 P3R3 NH3 Aromatic A P1R3 C H N NH3 3 9 DHONATZ CLHOATZ H O 2 aq, 2-Chloro-4-hydroxyl aq, C2H7N 2,4-Dihydroxy-6-(N-ethyl) Aliphatic p -6-amino,1-3-5-triazine p A Amino-1,3,5 atrazine H O C H ClN O 2 P2R4 C5H8N4O2 3 3 4 A P3R4 - Inorganic Cl H+ CYA DHOATZ P3R5 H2O Cyanuric acid aq, 2,4-Dihydroxy-6- aq, Mixed matter p A p C3H3N3O3 H O amino-1,3,5-triazine 2 NH 3 C3H4N4O2 --- Compound BIU C P4R1 H O short name aq, 2 Biuret CO2 H2O p C Reaction C2H5N3O2 X Deethylhydroxy- aq, P-R- NH3 Cl- P4R2 atrazine p coordinates C H N O H+ 6 11 5 Solid line ETA IPA ALP DEHA Hydrolization Dashed line Ethylamine aq, Isopropylamine aq, Allophanate aq, Oxidation C H N p C H N p C H N O p 2 7 3 9 2 4 2 3 Solid line End pathway SOM O H2O Dashed line 2 P6R1 C C Uncertain H2O C P5 NH3 P4R3CO2 A B CH O ATZhyd 2 IPP O2 H2O X R11 Isopropanol aq, aq B BATZoxi p C CH2O C3H8O P6R2 C BAER

D BAOB H+ R17 R21 Available as electron donor E B aq, aq, NOB + - (e.g., denitrification, NH3 g, NH4 p Cl Cl2 p D sulphate reduction) F + BDEN H2O aq, p H H2O g O2 R4 H+ O Available to plant uptake 2 X Uncharacterized H2O R10 Chemical Reaction + CH O H O H 2 2 H2O H2O + R5 O2 H CO CH2O CO 2 CO2 CH2O 2 E - - F N O N NO3 NO2 NO F 2 F 2 A F aq, p aq, p R7 aq, g R8 aq, g R9 aq, g R6 H O CH2O 2 P1R1b P1R1b CO 2 A H2O CH2O CO2

Figure 15: Atrazine biological reaction network in soil extended after la Cecilia & Maggi (2016) and coupled with the nitrogen cycle proposed by Maggi et al. (2008).

77 −5 −5 −6 x 10 x 10 x 10 2.5 2.5 6 ATZ, exp. (a) ATZ, exp. (b) ATZ, exp. (c) ATZ, model ATZ, model 5 ATZ, model

) 2 ) 2 )

soil soil soil 4

1.5 1.5 3 (mol/kg (mol/kg (mol/kg

e e e 2

Q 1 Q 1 Q 1 Waukegan silt loam Ves clay loam alluvial: Clay 64%, Silt 17% 0.5 0.5 0 0 0.5 1 1.5 0 2 4 6 8 0 0.5 1 1.5 2 C (kg /mol) −5 C (kg /mol) −6 C (kg /mol) −6 e H O x 10 e H O x 10 e H O x 10 2 2 2

−5 −6 −6 x 10 x 10 x 10 5 6 2.5 HOATZ, exp. (d) DIATZ, exp. (e) DEATZ, exp. (f) HOATZ, model DIATZ, model DEATZ, model 4 5 ) ) ) 2

soil soil 4 soil 3 3 1.5 2 (mol/kg (mol/kg (mol/kg

e e 2 e

Q Q Q 1 1 1 alluvial: Clay 64%, Silt 17% alluvial: Clay 64%, Silt 17% alluvial: Clay 64%, Silt 17% 0 0 0.5 0 0.25 0.5 0.75 1 0 1 2 3 0 1 2 3 4 C (kg /mol) −5 C (kg /mol) −6 C (kg /mol) −6 e H O x 10 e H O x 10 e H O x 10 2 2 2

Figure 16: Adsorption for ATZ. (a) and (b) from Clay & Koskinen (1990); (c) to (f) from Vryzas et al. (2007).

Figure 17: (a) Monthly-averaged precipitation in the period from 1990 to 2015 (Bureau of Meteorology, 2016), and irrigation (estimated from Hope, 2003); (b) Monthly-averaged potential and actual evapotranspiration rates ET0 and ETC. The growing season is highlighted in light green background color.

78 Cl Cl Cl Above Root Zone (aq) Below Root Zone (aq) biodegraded DIATZ DIDEATZ

Cl Cl Cl Above Root Zone (ad) Below Root Zone (ad) ATZ DEATZ CLHOATZ

1 100

0.8 10−1

0.6 10−2

0.4 10−3

0.2 10−4

(a) (b) 0 10−5 60% 80% Ref 125% 150% 200% 300% 60% 80% Ref 125% 150% 200% 300%

Figure 18: (a) Cl mass fraction partitioning between dissolved and adsorbed phases within and below the root zone; (b) Cl mass fraction in various organochlorides relative to the total Cl-ATZ mass applied.

0.15 600 0.15 600 0.15 600 150 250 50 150 250 105 305 50 500 500 500

0.45 400 0.45 400 0.45 400

300 300 300

(day) (day) (day) t t t

Depth (m) (m) Depth Depth (m) 0.75 200 0.75 200 0.75 200

half−concentration intercept 100 intercept half−concentration 100 half−concentration intercept 100 250

(a) (b) (c) 1.05 0 1.05 0 1.05 305 505 705 0 99 85 71 57 43 29 15 1 99 85 71 57 43 29 15 1 99 85 71 57 43 29 15 1 M /M (%) M /M (%) M /M (%) ATZ ATZ ATZ ATZ ATZ ATZ 0 0 0

Figure 19: ATZ breakthrough curves showing (a) minimum, (b) average, and (c) maximum time necessary by the biochemical system to decrease the ATZ mass peak (MATZ0 ) among 26 repetitions. The thick vertical black line corresponds to ATZ residual mass fraction of 50% relative to the mass peak.

0.15 90 0.15 5 15 25 90 0.15 90 5 35 15 80 80 80 70 70 70 0.45 60 0.45 60 0.45 60 50 50 50 40 40 40

Depth (m) Depth (m) Depth (m) 0.75 30 0.75 30 0.75 30 50

ATZ biodegraded (%) ATZ biodegraded (%) ATZ biodegraded (%) 20 20 20

half−concentration intercept half−concentration intercept half−concentration intercept 70

10 10 10 30 10 (a) (b) (c) 90 1.05 0 1.05 0 1.05 0 99 85 71 57 43 29 15 1 99 85 71 57 43 29 15 1 99 85 71 57 43 29 15 1 M /M (%) M /M (%) M /M (%) ATZ ATZ ATZ ATZ ATZ ATZ 0 0 0

Figure 20: ATZ biodegraded relative to the annual ATZ reference application equal to 0.2 g as a function of ATZ mass fraction removed relative to the corresponding mass peak at depth z. The thick vertical black line corresponds to ATZ residual mass fraction of 50% relative to the mass peak.

79 −3 x 10 2 1 ATZ biodegraded < 40% B ATZ biodegraded < 40% B 1.8 ATZ biodegraded > 40% ATZhyd 0.9 ATZ biodegraded > 40% ATZoxi

1.6 0.8

1.4 0.7

1.2 0.6

(1/s) 1 (1/s) 0.5 0 0 Φ Φ 0.8 0.4

0.6 0.3

0.4 0.2

0.2 0.1 (a) (b) 0 0 140 160 180 200 220 240 260 140 160 180 200 220 240 260 t t 1/2 (day) 1/2 (day)

Figure 21: Relationship between ATZ half-life t1/2 and speciﬁc biomass aﬃnity Φ0 in the top layer of the root zone for (a) ATZ hydrolyzers and (b) ATZ oxidizers.

0.15 7 0.15 7 0.15 0.6 0.45 0.45 0.45 0.75 6.8 0.75 6.8 0.75 1.05 6.6 1.05 6.6 1.05

6.4 6.4 1.95 1.95 1.95 0.4 6.2 6.2

6 6 pH pH pH 3.15 3.15 3.15 ∆ Depth (m) 5.8 Depth (m) 5.8 Depth (m) 0.2 5.6 5.6 4.05 4.05 4.05 No ATZ application 5.4 Reference ATZ application 5.4 Difference 5.2 5.2 (a) (b) (c) 5.25 5 5.25 5 5.25 0 J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N D Month of the year Month of the year Month of the year

Figure 22: Monthly-averaged pH isolines as a function of time and soil depth for (a) no ATZ application and (b) reference ATZ application. (c) represents the diﬀerence of maps in (b) and (a). The horizontal dashed black line delimits the root zone.

−4 x 10 0.15 2

0.45

0.75 1 0.01

1.05 0.1

0.05 )

1.95 dry−soil

Depth (m) 3.15 max [ATZ] (mg/g 4.05

0.01 5.25 0 0 100 200 300 400 500 Arrival time (years)

Figure 23: ATZ concentration as a function of time and soil depth. Black lines show the contours of ATZ concentrations

80 Pathway Kinetic Parameters Functional group Kinetic biological aqueous reaction µ K (MM) K YY I 1 mol mol g-C-Bio mg-wet-Bio s l l g-C-Subs mol-Subs (a) → + – ± × −5 ± × −4 ± × −1 ± × 5 P1R1a C8H14ClN5 + H2O(aq) C8H15N5O + H + Cl (3.67 2.49) 10 (I) (3.89 4.24) 10 (1.55 1.41) 10 (2.98 2.71) 10 BATZhyd ATZ HOATZ (b) → + – ± × −6 ± × −6 ± × −2 ± × 4 P1R1b C8H14ClN5 + H2O(aq) C8H15N5O + H + Cl (2.31 1.75) 10 (I) (3.43 3.31) 10 (1.59 1.60) 10 (3.06 3.08) 10 BATZhyd ATZ HOATZ (b) – → – ± × −5 ± × −7 × −6 ± × −1 ± × 4 P1R1b CH2O(aq) + 2 NO3 2 NO2 + CO2 + H2O (7.91 0.83) 10 (2.55 0.46) 10 2.50 10 (2.56 1.32) 10 (6.16 3.18) 10 BATZhyd −3 −6 (4.70 ± 0.04) × 10 2.50 × 10 -- BATZhyd (b) – → ± × −6 ± × −7 × −6 ± × −3 ± × 3 P1R1b 2 CH2O(aq) + 2 NO2 N2(aq) + 2 CO2 + 2 H2O (9.23 0.32) 10 (8.87 0.59) 10 2.50 10 (6.90 5.70) 10 (1.65 1.37) 10 BATZhyd −3 −6 (8.60 ± 0.57) × 10 2.50 × 10 -- BATZhyd (c) → × −5 × −5 × −2 × 4 P1R2 C8H15N5O + H2O(aq) C6H10N4O2 + C2H7N 3.77 10 (I) 3.14 10 4.72 10 9.16 10 BATZhyd HOATZ NIPA ETA (d) → × −4 × −4 × −3 × 3 P1R3 C6H10N4O2 + H2O(aq) C3H3N3O3 + C3H9N 3.89 10 (I) 7.80 10 1.13 10 1.63 10 BATZhyd NIPA CYA IPA (e) 3 → ± × −4 ± × −3 ± × −2 × 4 P2R1 C8H14ClN5 + 2 O2(aq) C5H8ClN5 + 3 CH2O (1.61 0.0003) 10 (II) (2.25 0.006) 10 (3.71 0.3) 10 7.22 10 BATZoxi ATZ DIATZ ( f ) → + – ± × −5 ± × −4 ± × −3 × 3 P2R2 C5H8ClN5+ H2O(aq) C5H9N5O + H + Cl (1.23 0.00001) 10 (I) (1.98 0.000009) 10 (3.44 0.04) 10 4.96 10 BATZhyd DIATZ DIHOATZ (g) → ± × −5 ± × −4 ± × −3 × 3 P2R3 C5H9N5O+ H2O(aq) C5H8N4O2+ NH3(aq) (1.23 0.00001) 10 (I) (1.98 0.000009) 10 (3.44 0.04) 10 4.96 10 BATZhyd DIHOATZ DHONATZ (g) → ± × −5 ± × −4 ± × −3 × 3 P2R4 C5H8N4O2+ H2O(aq) C3H3N3O3 + C2H7N (1.23 0.00001) 10 (I) (1.98 0.000009) 10 (3.44 0.04) 10 4.96 10 BATZhyd DIHONATZ CYA ETA (e) → ± × −4 ± × −3 ± × −2 × 4 P3R1 C8H14ClN5 + 3 O2(aq) C6H10ClN5 + 2 H2O(aq) + (1.51 0.0002) 10 (II) (2.09 0.0007) 10 (2.63 0.17) 10 5.06 10 BATZoxi ATZ DEATZ 2 CO2(aq) ( f ) 3 → ± × −6 ± × −3 ± × −3 × 3 P3R2 C6H10ClN5 + 2 O2(aq) C3H4ClN5 + 3 CH2O (5.21 0.000005) 10 (II) (3.68 0.014) 10 (7.00 0.19) 10 8.44 10 BATZoxi DEATZ DIDEATZ (h) → ± × −6 ± × −3 ± × −3 × 3 P3R3 C3H4ClN5 + H2O(aq) C3H3ClN4O + NH3(aq) (5.21 0.000005) 10 (I) (3.68 0.014) 10 (7.00 0.19) 10 8.44 10 BATZhyd DIDEATZ CLHOATZ (h) → + – ± × −6 ± × −3 ± × −3 × 3 P3R4 C3H3ClN4O + H2O(aq) C3H4N4O2 + H + Cl (5.21 0.000005) 10 (I) (3.68 0.014) 10 (7.00 0.19) 10 8.44 10 BATZhyd CLHOATZ DHOATZ (h) → ± × −6 ± × −3 ± × −3 × 3 P3R5 C3H4N4O2 + H2O(aq) C3H3N3O3 + NH3(aq) (5.21 0.000005) 10 (I) (3.68 0.014) 10 (7.00 0.19) 10 8.44 10 BATZhyd DHOATZ CYA (i) → × −3 × −1 × −4 × 2 P4R1 C3H3N3O3 + H2O(aq) C2H5N3O2 + CO2 2.14 10 (III) 6.70 10 8.50 10 6.13 10 BAER CYA BIU (i) → × −4 × −2 × −5 × 1 P4R2 C2H5N3O2 + H2O(aq) C2H4N2O3 + NH3(aq) 3.41 10 (III) 8.67 10 6.60 10 3.18 10 BAER BIU ALP (i) → × −5 × −1 × −5 × 1 P4R3 C2H4N2O3 + H2O(aq) 2 NH3(aq) + 2 CO2(aq) 9.26 10 (III) 1.11 10 9.10 10 4.40 10 BAER ALP (l) 1 → × −5 × −1 × −5 × 1 P5 C2H7N + 2 O2(aq) 2 CH2O + NH3(aq) 9.34 10 (III) 8.38 10 7.60 10 3.67 10 BAER ETA (m) → × −5 × −1 × −5 × 1 P6R1 C3H9N + H2O(aq) C3H8O + NH3(aq) 9.34 10 (III) 8.38 10 7.60 10 3.67 10 BAER IPA IPP (m) 3 → × −5 × −1 × −5 × 1 P6R2 C3H8O + 2 O2(aq) 3 CH2O + H2O(aq) 9.34 10 (III) 8.38 10 7.60 10 3.67 10 BAER IPP (a) → ± × −5 ± × −3 ± × −2 ± × 4 R1 CH2O(aq) + O2(aq) H2O(aq) + CO2(aq) (5.60 6.69) 10 (0.75 1.06) 10 (4.70 7.60) 10 (1.25 1.79) 10 BATZhyd (n) → ± × −5 ± × −3 ± × −2 ± × 4 R2 CH2O + O2(aq) H2O(aq) + CO2(aq) (5.60 6.69) 10 (0.75 1.06) 10 (4.70 7.60) 10 (1.25 1.79) 10 BATZoxi (n) → ± × −5 ± × −3 ± × −2 ± × 4 R3 CH2O + O2(aq) H2O(aq) + CO2(aq) (5.60 6.69) 10 (0.75 1.06) 10 (4.70 7.60) 10 (1.25 1.79) 10 BAER (o) → – + × −5 × −4 × 4 R4 NH3 + 3/2 O2(aq) H2O(aq) + NO2 + H 1.0694 10 3.0410 10 - 0.5 10 BAOB (o) – → – × −5 × −4 × 4 R5 NO2 + 1/2 O2(aq) NO3 3.5966 10 (IV) 2.9840 10 - 0.4 10 BNOB (o) – → – + – × −4 × −4 × −6 × 4 R6 NO3 +1/2 CH2O(aq) 1/2 HCO3 + 1/2 H + NO2 4.0704 10 2.0677 10 2.50 10 - 0.75 10 BDEN – → – + – × −4 × −4 × −6 NO3 +1/2 CH2O(aq) 1/2 HCO3 + 1/2 H + NO2 4.0704 10 2.0703 10 2.50 10 -- BDEN (o) – + → – × −5 × −4 × −6 × 4 R7 NO2 +1/4 CH2O(aq) + 3/4 H 1/4 HCO3 + NO(aq) 9.6768 10 (IV) 7.4892 10 2.50 10 - 0.375 10 BDEN + 1/2 H2O – + → – × −5 × −4 × −6 NO2 +1/4 CH2O(aq) + 3/4 H 1/4 HCO3 + NO(aq) 9.6768 10 1.3599 10 2.50 10 -- BDEN + 1/2 H2O (o) → + – × −4 × −4 × −6 × 4 R8 NO(aq) +1/4 CH2O(aq) 1/4 H + 1/4 HCO3 + 1/2 8.0080 10 1.7551 10 2.50 10 - 0.375 10 BDEN N2O(aq) → + – × −4 × −5 × −6 NO(aq) +1/4 CH2O(aq)* 1/4 H + 1/4 HCO3 + 1/2 8.0080 10 6.2404 10 2.50 10 -- BDEN N2O(aq) (o) → + – × −5 × −5 × −6 × 4 R9 N2O(aq) +1/2 CH2O(aq) 1/2 H + 1/2 HCO3 + 1 1.6846 10 5.1698 10 2.50 10 - 0.75 10 BDEN N2(aq) → + – × −5 × −5 × −6 N2O(aq) +1/2 CH2O(aq) 1/2 H + 1/2 HCO3 + 1 1.6846 10 8.8059 10 2.50 10 -- BDEN N2(aq) Kinetic chemical aqueous reaction µ K (MM) (o) – + → – × −11 × −4 R10 NO2 +2/3 H 1/3 H2O +1/3 NO3 +2/3 NO(aq) 1.0742 10 1.129 10 (p) → + × −11 R11 SOM CH2O + 0.039 NH4 1.0 10 - ◦ Equilibrium aqueous complexation reactions (at T = 25 ) log10(K(aq)) (q) – + −−−* R12 OH + H )−−− H2O(aq) 13.99 (q) + −−−* + R13 NH4 )−−− H + NH3(aq) -9.24 (q) −−−* + – R14 CO2(aq) + H2O )−−− H + HCO3 -6.34 ◦ Equilibrium gas dissolution reactions (at 25 ) log10(K(g)) (q) −−−* R15 O2(aq) )−−− O2(g) 2.8980 (q) −−−* + – R16 CO2(g) + H2O(aq) )−−− H + HCO3 -7.8136 (q) + −−−* + R17 NH3(g) + H )−−− NH4 11.038 (q) −−−* R18 N2(aq) )−−− N2(g) 3.2451 R19(q) NO(aq) )−−−−−−* NO(g) 2.7609 (q) −−−* R20 N2O(aq) )−−− N2O(g) 1.6021 (q) −−−* – + R21 Cl2(g) + H2O )−−− 1/2 O2(aq) +2 Cl + 2 H 1.5516 ◦ Equilibrium protection reactions (at 25 ) log10(K(p)) (r) + −−−* + R22 NH4 )−−− NH4 (p) -0.336 (s) – −−−* – R23 NO3 )−−− NO3 (p) -1.072 (t) – −−−* – R24 NO2 )−−− NO2 (p) -1.072 R25(u) Cl – )−−−−−−* Cl – (p) -1.072 R26(v) ATZ −−−)−−−* ATZ(p) 0.515 R27(z) HOATZ )−−−−−−* HOATZ(p) 0.655 R28(aa) DIATZ )−−−−−−* DIATZ(p) 0.409 R29(bb) DEATZ )−−−−−−* DEATZ(p) 0.066 R30(cc) NIPA )−−−−−−* NIPA(p) 0.655

81 R31(dd) ETA )−−−−−−* ETA(p) -0.9133 R32(cc) DIHOATZ )−−−−−−* DIHOATZ(p) 0.655 R33(ee) DIDEATZ )−−−−−−* DIDEATZ(p) 0.409 R34(cc) DIHONATZ )−−−−−−* DIHONATZ(p) 0.655 R35(ee) CLHOATZ )−−−−−−* CLHOATZ(p) 0.409 R36( f f ) DHOATZ )−−−−−−* DHOATZ(p) -1.18 R37(gg) CYA )−−−−−−* CYA(p) 0.0050 R38(hh) IPA )−−−−−−* IPA(p) -0.9133 R39(ii) BIU −−−)−−−* BIU(p) 0.0050 R40(ii) ALP −−−)−−−* ALP(p) 0.0050 R41(ll) IPP )−−−−−−* IPP(p) -1.5823

BATZhyd encompasses the genus Bacillus, Pseudomonas, and Burkholderia, and the strains Pseudomonas sp. ADP and Nocardia sp.; BATZoxi encompasses the genus Rhodococcus and Enterobacter and Rhodococcus strains TE1 and B30; BAER encompasses the strain Pseudomonas sp. ADP, Arthrobacter P1, and Mycobacteriumconvolutum NPA-1, and an unidentiﬁed community of soil microorganisms; BDEN encompasses the genus Pseudomonas and Thiobacillum; BAOB encompasses the genus Nitrosomona and Nitrosospira; BNOB encompasses the genus Nitrobacter and Nitrospira.

(a) Parameters estimated in la Cecilia & Maggi (2016) against experiments in Katz et al. (2000); Mandelbaum et al. (1995); Radosevich et al. (1995); Smith et al. (2005); Smith & Crowley (2006); (b) Parameters estimated in la Cecilia & Maggi (2016) against experiments in Katz et al. (2000); (c) Parameters estimated against experiments in Kumar & Singh (2016); (d) Parameters estimated against experiments in Boundy-Mills et al. (1997); (e) Parameters against experiments in Behki et al. (1993); Behki & Khan (1994); Solomon et al. (2013); (f) Parameters estimated against experiments in Behki & Khan (1994); Shao et al. (1995); Solomon et al. (2013); (g) Parameters assumed to be similar to those in P2R2; (h) Parameters assumed to be similar to those in P3R2; (i) Parameters estimated against experiments in Martinez et al. (2001); (l) Parameters estimated against experiments in Levering et al. (1984); (m) Parameters assumed to be similar to those in P5; (n) Parameters assumed to be similar to those in R1; (o) Parameters estimated in Maggi et al. (2008) against experiments in Venterea & Rolston (2000); (p) Parameters assumed in Maggi et al. (2008) afterDon & Schulze (2014); (q) Parameters from EQ3/6 Wolery (1992); (r) Parameters estimated in Tang (2016) against experiments in Ding et al. (2010); (s) Parameters estimated in Tang (2016) against experiments in Li & Bowman (2001); (t) – Parameters assumed to be similar to those of NO3 as in Tang (2016); (u) Parameters estimated in Tang (2016) against experiments in Back & Waring (1979); (v) Parameters estimated against experiments in Clay & Koskinen (1990); Vryzas et al. (2007); (z) Parameters estimated against experiments in Vryzas et al. (2007); (aa)Parameters estimated against experiments in Vryzas et al. (2007); (bb)Parameters estimated against experiments in Vryzas et al. (2007); (cc) Parameters assumed to be similar to those of HOATZ; (dd) Parameters converted from Hansch et al. (1995); (ee) Parameters assumed to be similar to those of DIATZ; (ﬀ) Parameters assumed to be similar to those of Sarcosine, which were estimated in Tang (2016) against experiments in Friebele et al. (1980); (gg) Parameters converted from Yalkowsky (2003); (hh) Parameters assumed to be similar to those of ETA; (ii) Parameters assumed to be similar to those of CYA; (ll) Parameters converted from Meylan et al. (1992). The Michaelis-Menten (MM) constants are listed in the order of appearance of corresponding reactants. The Roman number in front of the MM constant refers to the list of MM terms used to account for substrate competition, the MM constant for the reaction was removed from the list accordingly: (I) ATZ in P1R1a and P1R1b, HOATZ in P1R2, NIPA in P1R3, DIATZ in P2R2, DIHOATZ in P2R3, DIHONATZ in P2R4, – DIDEATZ in P3R3, CLHOATZ in P3R4, DHOATZ in P3R5; (II) ATZ in P2R1 and P3R1, DEATZ in P3R2; (III) CYA in P4R1, BIU in P4R2, ALP in P4R3, ETA in P5, IPA in P6R1, IPP in P6R2; (IV) NO2 in R5 and R7. All kinetic + −8 −6 biological aqueous reactions also include a MM term and an inhibition term relative to H concentration with constants 10 and 10 M, respectively. Oxidative reactions also include a MM term relative to O2(aq) concentration with × −4 × −6 constant 1.50 10 in P2R1, P3R1, P3R2, P5, P6R2, R4, and R5. Anaerobic reactions also include an inhibition term relative to O2(aq) concentration with constant 2.5 10 in P1R1b, and from R6 to R9. The microbial mortality was −6 −1 −6 −1 −6 −1 −7 −1 assumed to be 10 s for BATZhyd,BATZoxi, and BAER, 2.69 × 10 s for BAOB, 1.61 × 10 s for BNOB, 1.22 × 10 s for BDEN . Table 10: Equilibrium and kinetic reactions implemented in the ATZ biochemical reaction network and corresponding parameters.

82 5.4. Glyphosate

In this section, the numerical results showing in-situ biodegradation and dispersion of GLP and its metabolite AMPA are presented. Contents of this chapter come from the article la Cecilia et al. (2018a)6 published in Water Research.

5.4.1. Introduction

After developing the GLP reaction network in soil in Section 4.5.2, this chapter shows comprehensive modeling results, which highlight the high level of process coupling in real-case GLP application scenarios in agricultural lands. In particular, BRTSim was used to predict the dynamics of GLP and AMPA under the effects of C, N, and P release from SOM and fluctuating ecohydrological boundary conditions in two different agricultural systems. Field data of precipitation and water table dynamics were used to constrain the model parameters. The modeling approach was introduced in Section 5.3.1. Briefly, reaction pathways are carried out by microbial functional groups and the kinetics were described using averaged MMM parameter values. The same physical, biogeochemical, and ecological feedbacks on molecules fate and microbial dynamics were included, except for competition amongst microbial functional groups for CH2O given the small number of C sources in the reaction network. The model structure levered the validation-by construct principle (McCarl & Apland, 1986) and the estimated parameters were used without adjustments. MM and inhibition terms were used to explicitly account for O2(aq) and pH effects on aerobic and anaerobic reactions. Sensitivity analyses were run to predict the biogeochemical response to increasing GLP application rates and availability of CH2O. All the equilibrium and kinetic reactions implemented in this biochemical system are reported in Table 16 with their corresponding parameters. The biological processes which may affect GLP fate in the environment and were not accounted for in the overall system are discussed in detail in Sections 5.4.2, 5.4.3, 5.4.3, and 5.4.3.

5.4.2. Methods

GLP reaction network coupled with N cycle. The GLP reaction network proposed in Section 4.5.2 and published in la Cecilia & Maggi (2018) was expanded to include a SOM pool, and was coupled with the N cycle earlier described in Maggi et al. (2008) (Figure 24). The SOM pool released nutrients that support the microbial soil communities; this allowed us to investigate nonlinearities in GLP and AMPA soil biodegradation due to varying carbon (C) source bioavailability, in the form of CH2O. Yet, the GLP reaction network and the N cycle share + CH2O, O2, NH3, and H , thus highlighting the level of process coupling that was accounted for. The biochemical reactions are reported in Table 16 together with their corresponding reaction

6la Cecilia, D., Tang, F.H., Coleman, N.V., Conoley, C., Vervoort, R.W., and Maggi, F. (2018). Glyphosate dispersion, degradation, and aquifer contamination in vineyards and wheat ﬁelds in the Po Valley, Italy. Water Research, 146, pp. 37-54, 10.1016/j.watres.2018.09.008.

83 rate constant µ (s−1), half-saturation concentration K (mol L−1), biomass yield coeﬃcient Y (mg- −1 wet-biomass mol-substrate ), and equilibrium constant (Kaq, Kg, and Kad for aqueous complexation, gaseous dissolution, and adsorption). The coupled biochemical system was implemented in the BioReactive Transport Simulator (BRTSim) following the “validation by construct" approach (McCarl & Apland, 1986). GLP and AMPA chemical degradation by birnessite mineral (Li et al., 2015; Paudel et al., 2015) represented in Figure 24 via pathways P2R1c and P1R2c, respectively, was not included because it is uncertain whether it occurs in ﬁeld conditions (i.e., Barrett & McBride 2005 observed that Cu2+ ions preferentially adsorbed onto birnessite, thus inhibiting GLP and AMPA degradation).

GLP

Aq, Glyphosate ad SOM C3H8NPO5 CH2O + Head CO2 H aq, Compound phase O2 H2O Aerobic compound p, A E g (aqueous, adsorbed, gaseous) P1R1s Aerobic Intermediate Aerobic P2R1c CH O O2 2 compound Aliphatic A O2 P1R1 3- PO4 A P2R1s H+ Final CO AMPA 2 product Inorganic GLX Glyoxylate aq Amino-methyl- Aq, Nitrogen ad C H O phosphonic acid Mixed matter 2 2 4 aq cycle CH6NPO3 Sarcosine C3H7NO2 Aerobic P1R2c D E SRC Organophosphate --- O2 Compound short name

Aerobic Aerobic CH2O + O H B 2 Anaerobic P2R2a PO 3- 4 A P1R2s GLY

H+ CH 2

Glycine aq O Reaction coordinates CO2 P-R- MTH C2H5NO2 Phosphono- aq P2R2b Solid line Biotic reaction Methylamine aq C formaldehyde CO2 Dashed line CH N H O O CHPO 5 2 2 Abiotic reaction 4 O 2 P2R3b P2R3a O + C B Dashed line H O 2 CH H 2 Aerobic CO2 Uncertain Anaerobic H O D C B 2 P1R3a Anaerobic + P1R3b H+ H + CH2O Inhibitor H CO2 CH4

CH2O H+ A Aerobic BHyO 3- PO4 B X R9 BAER C BANAER D Ochrobactrum anthropi GPK 3 Available as electron donor H+ R15 (e.g., denitrification, E Birnessite (Mn3+, Mn4+ containing Aq, aq, Aq + 3- aq sulphate reduction) NH g, NH ad PO4 mineral) 3 4 CH2O ad F H O 2 Available to plant uptake F BAOB O2 R2 H+

H2O G BNOB R8 + CH O H O H B H 2 2 H2O H2O DEN O + CH O R3 2 H CO2 2 CO CO 2 CH2O 2 X Uncharacterized G - - H N O N NO3 NO2 NO H 2 H 2 H Chemical Reaction aq, ad aq, ad R5 R6 aq, g R7 aq, g R4 aq, g CH2O H2O CO2

Figure 24: Expanded GLP biodegradation reaction network in soil from la Cecilia & Maggi (2018) coupled with the N cycle from Maggi et al. (2008). Extended biochemical reactions, the corresponding kinetic parameters, and the microbial functional groups mediating the reactions are reported in Table 16.

Equilibrium adsorption reactions. Two laboratory experiments reporting GLP and AMPA adsorption isotherms (5.4.2, Sidoli et al., 2016) were used to estimate the corresponding linear equilibrium constants Kd (Table 11) between the substrate dissolved and adsorbed phases by means of inverse problem solution (Figure 25). First, observations were ﬁtted with the Lang- muir model (5.3.2).

84 2 Test Source Soil Sand-Silt-Clay OC Substrate Qmax,Langmuir Kad,Langmuir log10(Kd) R NRMSE −1 −1 −1 (%-%-%) g kgdry-soil mg g mg L (-) 1 2 3 4 5 6 7 8 9 10 11 1 (a) chromic cambisol 42.8-42.3-11.8 23.1 GLP 4.68 × 10−4 1.01 × 105 1.67 0.99 0.94 2 (a) chromic cambisol 42.8-42.3-11.8 23.1 AMPA 5.12 × 10−4 1.18 × 105 1.67 0.99 2.24

Table 11: Estimated adsorption parameters for GLP and AMPA of nonlinear Langmuir equilibrium Qmax,Langmuir and Kad,Langmuir, and the corresponding linearized equilibrium constant Kd. OC refers to organic carbon. Param- eters are tabulated together with the goodness-of-ﬁt against experiments in Figure 25 from: Sidoli et al. (2016). Experiments were carried out at T = 20◦.

−1 Langmuir equilibrium constant, and Ce (mol L ) is the solute equilibrium concentration in the solution. Next, nonlinearities in the adsorption process were approximated with the tangent line to the Langmuir isotherm in the low adsorption-low concentration range as

Qe Kd = , Ce −1 where Qe (mol kgsoil) is the moles of adsorbed solute per unit mass of adsorbent, and Ce (mol −1 kg ) is the equilibrium solute concentration. A linear Kad was used because GLP and AMPA H2O concentrations in soils are very low and the tangent of the Langmuir model fit in the low concentration-low adsorption range was representative enough of soil buffer capacity for those compounds. Parameters relative to the nonlinear Langmuir model together with their goodness- of-fit, which was measured by means of the coefficient of determination (R2) and the normalized root mean squared error percent (NRMSE), and to the linearized approach are reported in Table 11. 3 – PO4 adsorption onto soil particles was not included in the model; therefore, the competition 3 – between GLP, AMPA, and PO4 for adsorptive sites was not modeled after Sprankle et al. 3 – (1975) showed a low sensitivity of adsorbed GLP over a wide range of PO4 concentrations. Also Munira et al. (2016) showed more recently that P-based fertilizers applied at a rate up to 80 kg-P ha−1y−1 did not substantially affect the adsorbed GLP.

2.5 OC: 2.3%, Clay 11.8%, Silt 42.3% (a) OC: 2.3%, Clay 11.8%, Silt 42.3% (b)

2 ) -4

10 1.5

-1

1 (mol kg e

Q 0.5 GLP, exp. AMPA, exp. GLP, model AMPA, model 0 0 2 4 6 0 2 4 6 C -1 -6 C -1 -6 e (kg mol 10 ) e (kg mol 10 )

Figure 25: (a) Adsorption for GLP; (b) Adsorption for AMPA. Experiments from Sidoli et al. (2016). GLP, exp and AMPA, exp represent the experimental data, while GLP, model and AMPA, model represent the numerical solution.

85 Experimental site. The GLP reaction network was numerically estimated in a vineyard (40◦4502200N; 10◦410500E) and a wheat field (44◦4005700N; 10◦5704800E) in Reggio Emilia and Modena munic- ipalities in the Po Valley, Italy. Soil characteristics were retrieved from SGSS (2016) and are reported in Table 12. The bottom soil layer in both sites was extended down to 5 m depth to predict GLP and AMPA dispersion in the aquifer. Assuming a 20% rainfall loss by crop inter- −1 −1 ception, the infiltrating precipitation PI averaged 540 mm y in the vineyard and 515 mm y in the wheat field in the 2006-2016 period (data post-processed after Arpae-Simc, 2016). The average potential evapotranspiration rate was 1133 mm y−1 in the vineyard and 828 mm y−1 in the wheat field during the same period (data post-processed after Arpae-Simc, 2016). The daily reference crop evapotranspiration (ET0) was calculated using the approach in Allen et al.

(1998) (Figure 26). The daily actual crop evapotranspiration was calculated as ETC = ET0 ×KC, with the time-varying crop coefficient KC taken from Allen et al. (1998) and represented in Fig- ure 26. Daily irrigation rates (Irr) were not available; hence these were estimated depending on the crop and fine-tuned to match in-situ groundwater observations from Chiari et al. (2016) (Figure 27). The vineyard was irrigated during its typical growing season between April and October, while the wheat field was irrigated all year around. Daily precipitation, irrigation, and crop evapotranspiration rates were used as upper boundary conditions (Figure 27) while yearly cumulative rates were used for analyses described later (Table 13). In the vineyard, 60% of the roots system was uniformly distributed within the first 0.6 m depth, while the remaining 40% was uniformly distributed down to 1.4 m (Seguin, 1986). The average roots depth in the wheat field was 0.3 m and the roots density was distributed with a negative exponential function down to 1 m depth (Bowdena et al., 2008). Passive uptake + – – 3 – + of NH4 , NO3 , HCO3 , PO4 , and H was described as driven by evapotranspiration and nutrient concentrations in solution. Meteorological observations spanned 11 years from 2006 to 2016 and constituted one cycle of boundary conditions (BCcycle). The BCcycle was repeated to construct boundary conditions for longer time frames. The biochemical system reached steady state in the top 1 m of soil after four BCcycle.

Deptha Soil textureb Sand-Silt-Clay fractiona SOM fractiona Mineral densityc Porosityc Pore size distributionc Air entry suctionc Permeabilityc −1 −3 2 × −13 (m) (%-%-%) (g kgdry-soil) (kg m ) (-) (-) (m) (m ) 10 Vineyard 0-0.5 Silt loam 16-60-24 20 2848 0.469 6.73 -0.47 1.66 0.5-0.8 Silt loam 9-66-25 20 2855 0.478 6.88 -0.58 1.29 0.8-1.2 Silt loam 12-73-15 10 2852 0.474 5.29 -0.53 1.44 1.2-1.5 Silt loam 12-68-20 10 2852 0.474 6.09 -0.53 1.44 1.5-5.0 Silt loam 12-68-20 10 2852 0.474 6.09 -0.53 1.44 Wheat ﬁeld 0-0.6 Silt loam 19-56-25 20 2845 0.465 6.88 -0.43 1.84 0.6-1.0 Loam 36-43-21 10 2829 0.444 6.25 -0.26 3.35 1.0-1.4 Loam 49-34-17 6 2817 0.427 5.61 -0.17 5.30 1.4-3.4 Loam 36-48-16 6 2829 0.444 5.45 -0.26 3.35 3.4-5.0 Loam 36-48-16 6 2829 0.444 5.45 -0.26 3.35

Table 12: Soil and hydraulic parameters at two test areas in the Po Valley, Italia. a From SGSS (2016). b USDA soil texture classiﬁcation system. c Estimated using Cosby et al. (1984).

Modelling framework. Soil microbial communities change over time and depends on environmental conditions such as water content, nutrients availability, O2 levels, etc. Detailed data on microbial communities present at the test sites were not available, though it is likely that GLP

86 7 (a) Vineyard (b) Wheat field P (mm d-1) I 6 Irr (mm d-1) ET (mm d-1) 5 0 ET (mm d-1) 4 C K 3 C

0 JFMAMJJASOND JFMAMJJASOND Month of the year Month of the year

Figure 26: Monthly-averaged rates of precipitation (PI), irrigation (Irr), and potential (ET0) and actual (ETC) evapotranspiration, in the period from 2006 to 2016 (Arpae-Simc, 2016) in (a) vineyard and (b) wheat ﬁeld. The time-varying crop coeﬃcient (KC) is plotted in gray. ) Vineyard ) −1 4 (a) Irr 40 −1 2 P 20 (mm d (mm d 0 I 0 I P Irr 0 Observed 1 (b) Numerical 2 3 4 GW depth (m) 5 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 t (y) ) Wheat field ) −1 16 (c) Irr 60 −1 8 P 30 (mm d (mm d 0 I 0 I P Irr 0 Observed 1 (d) Numerical 2 3 4 GW depth (m) 5 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 t (y)

Figure 27: Daily rates of precipitation and irrigation, and the resulting groundwater table ﬂuctuations, interpolated at available sample times, at the (a) vineyard and (b) wheat ﬁeld. Black circles represent the in-situ groundwater observations from Chiari et al. (2016).

and AMPA biodecomposers (BHyO, Figure 24) inhabited the root zone where GLP had been periodically applied as earlier observed in Dick & Quinn (1995). Five other microbial functional groups were included to describe the following reactions: non-toxic metabolites of GLP and

AMPA were consumed by BAER in aerobic conditions and by BANAER in anaerobic conditions, two-steps nitriﬁcation was carried out by BAOB and BNOB, and denitriﬁcation was carried out by BDEN. Although a C source was continuously supplied as CH2O, the biomass concentrations of a microbial functional group j may become 0 for some MMM kinetic parameter values combination (Porta et al., 2018). To maintain the microbial diversity over time, each functional group was set to have a biomass concentration greater or equal than a minimum value. This was achieved by resorting to a biomass background recovery rate rB, j calculated neglecting micro-

87 Year Setting 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

−1 PI (mm y ) vineyard 425 428 578 534 702 401 455 686 734 493 504 −1 PI (mm y ) wheat ﬁeld 344 533 624 480 676 333 495 579 579 484 549 Irr (mm y−1) vineyard 252 472 252 180 132 174 186 172 99 132 186 Irr (mm y−1) wheat ﬁeld 706 727 290 718 380 913 821 814 526 871 667

−1 ETC (mm y ) vineyard 582 598 585 598 560 602 607 573 547 604 581 −1 ETC (mm y ) wheat ﬁeld 782 808 757 802 770 819 819 763 771 824 791

Table 13: Yearly cumulative ecohydrological ﬂuxes in the upper boundary representing eleven diﬀerent boundary conditions of the BCcycle.

bial growth from the breakdown of C sources as dB j/dt = −δ jB j + rB, j = 0, hence returning rB, j ∗ ∗ ∗ = δ jB j, with B j the equilibrium concentration for B j in the absence of C sources. Values of B j were chosen such that BAER/BHyO = 1 (Aristilde et al., 2017; Newman et al., 2016; Smit et al.,

2001), while BAOB/BAER = BAOB/BNOB = 5 and BDEN/BAOB = 10 (Tang, 2016). The minimum soil biomass concentration was therefore imposed, while the instantaneous biomass concentration varied according to substrates availability and consumption. Cell mortality rate was assumed to be δ = 10−6 s−1 for all microbial functional groups. Biodegradation of adsorbed compounds was excluded with the assumption that multiple forms of adsorption can prevent enzyme degradation effectiveness (Riley et al., 2014). Substrates consumption competition was not included in the GLP reaction network due to the small range of different C sources available to each microbial group. Microbial competition for O2(aq) and CH2O was implicitly captured by the model thanks to the different MMM parameter values corresponding to each different biological reactions. The effect of pH on microbial activity was explicitly taken into account in all biologically-mediated reactions using a MM term for high pH with constant 10−9 M, and an inhibition term for low pH with constant 10−5

M, respectively (Boon & Laudelout, 1962). The MM constant for O2(aq) consumption was −5 K = 1.4×10 M (Button & Garver, 1966). O2(aq) inhibition on anaerobic reactions P1R3b, P2R2b, and P2R3b was explicitly included using an inhibition term with constant 3.125 ×10−6 M (Kindred & Celia, 1989). The total organic carbon (%TOC), measured at the selected locations (SGSS 2016, Table 12), was used to estimate the soil organic matter (%SOM) as %SOM = 1.72 × %TOC (Hoyle, −1 −1 2015), which was next converted from g kgdry-soil to mol L and assumed to slowly release + 3 – −11 −1 CH2O, NH4 , and PO4 with a ﬁrst-order kinetic reaction with rate constant r = 10 s and stoichiometry taken from Tipping et al. (2016) (reaction R9, Table 16). That is, the molecular + 3 – mass of 1 mol of CH2O plus 0.039 moles of NH4 plus 0.0011 moles of PO4 have a molecular mass of 30.806 g mol−1; 1 kg of soil contains 20 g of SOM (Table 12), and therefore, 0.65 moles of SOM; given the soil properties of Table 12 and assuming a soil water saturation of 0.65, the 3 numerical node of volume 0.10 m can contain 151 kgdry−soil and 31.25 L of H2O, and therefore

88 3.146 mol L−1 of SOM. SOM was assumed to be constant over time as a result of recharge from root mortality, exudates, and other debris (Riley et al., 2014). Note that CH2O was also released by sarcosine, glycine, and methylamine biodegradation reactions (P2R2a and P2R2b, P2R3b, 3 – and P1R3a, Figure 24). Aqueous PO4 inhibited GLP oxidation (Pipke & Amrhein, 1988b) and AMPA oxidation (Balthazor & Hallas, 1986).

Analysis of GLP application scenarios. A range of GLP yearly application rates and CH2O release rates from SOM were used to investigate nonlinearities in GLP and AMPA biodegradation dynamics. GLP yearly applications of 0.72 kg ha−1 in April and October in the vineyard, and 2 kg ha−1 in November in the wheat field were used as the reference for our analyses (Table 14). It was assumed that all applied GLP leached into the soil; because some GLP absorb into leaves in reality, our predictions overestimate GLP soil concentrations and the results are therefore cautionary. GLP and AMPA distribution along the soil profile and their phase partitioning were calculated using masses averaged over the 5th BCcycle, when the biochemical system was at steady state in the top 1 m of soil. These analyses were repeated along the soil profile also for GLP application rates equal to half (GLPlow) and three times (GLPhigh) the reference rate ref (GLP ), and for the stoichiometry coefficient relative to CH2O release from SOM equal to half low high ref (CH2O ) and twice (CH2O ) the reference value (CH2O ).

GLP application dose GLP application date kg-GLP ha−1 (day/month) Vineyard 0.72 01/04 0.72 01/10 Wheat ﬁeld 2.00 01/11

Table 14: Glyphosate applications dose and date.

Assessment of aquifer contamination by GLP and AMPA. The Hazard Quotient (HQ) by Suter (2007) was used to assess the aquifer contamination by GLP and AMPA as

Cpredicted(t) HQ = , (54) Cthreshold where Cpredicted(t) is the GLP or AMPA aqueous concentration, or their sum, as a function of time t, and Cthreshold is the European safety limit to ensure the maintenance of an overall good −1 groundwater quality (EC Directive 2006/118/EC, 2006). Cthreshold = 0.1 µg L is used for a −1 single pesticide and Cthreshold = 0.5 µg L for a pesticide mixture that causes similar adverse health eﬀects. The HQ was averaged over the 5th BCcycle and was represented as a function of soil depth and time (Figure 29). Note that, the contribution of microbial dynamics on contaminant dispersion can easily be quantiﬁed by calculating the biodegradation potential parameter

ψB (Tang et al. 2017, based on la Cecilia & Maggi 2016) in simplistic models that do not account for biodegradation kinetics.

Analysis of GLP age and turnover time. A molecule tracking method was used to calculate GLP age at each depth z as the time necessary for that molecule to move from depth z = 0 m to

89 depth z. Tracking was used for a single application of non-reactive GLP after the biochemical system reached steady state. GLP age variability due to diﬀerent hydrological conditions was investigated by releasing and tracking the molecule for each of the eleven years of the BCcycle (Table 13).

GLP turnover time was averaged at each depth over the 5th BCcycle as:

GLPaq(z) + GLPad(z) Tt,GLP(z) = , (55) RP1R1(z) + RP1R1s(z) + RP2R1s(z) + FGLP,out(z) where Tt,GLP(z) is GLP turnover time, GLPaq(z) is dissolved GLP, GLPad(z) is adsorbed GLP,

RP1R1(z) is biodegraded GLP mass rate along P1R1, RP1R1s(z) is biodegraded GLP mass rate along P1R1s, RP2R1s(z) is biodegraded GLP mass rate along P2R1s, and FGLP,out(z) is GLP mass ﬂux exiting the node at depth z. Together with the overall turnover time, the contribution of each of the biological and physical processes was calculated; for example the turnover time corre- GLPaq(z)+GLPad(z) sponsing to GLP biodegraded along P1R1 was calculated as: Tt,GLP-P1R1(z) = . RP1R1(z)

Statistical analysis of GLP breakthrough curves. The biogeochemical system was exposed to eleven diﬀerent ecohydrological boundary conditions of the BCcycle (Table 13). Between two consecutive GLP applications, a GLP breakthrough curve in the top 10 cm of soil was deﬁned as the time history of normalized GLP mass MGLP(t), that is, the instantaneous GLP mass subtracted from the applied GLP mass MGLP0 . Times t when MGLP(t)/MGLP0 was between

0.99 and 0.01 were recorded. GLP half-life t1/2 was then calculated as the time necessary by the biochemical system to meet the condition MGLP(t)/MGLP0 = 0.5. Metabolites AMPA and sarcosine masses were monitored to calculate the percentage of biodegraded GLP at the same times t relative to applied GLP.

Inter-annual variability to GLP biodegradation. The relationship between cumulative water input and GLP biodegraded fraction between two consecutive GLP applications under diﬀerent hydrological boundary conditions was investigated using outputs from the breakthrough curves.

Cumulative water inputs and GLP biodegraded fractions were mapped for MGLP(t)/MGLP0 equal to 0.01.

5.4.3. Results and Discussion

Assessment of soil water contamination by GLP and AMPA. The water table ﬂuctuated all along the soil column (as shown in Figure 27). GLP and AMPA aqueous concentrations were −1 nearly 200 µg L in the top 0.5 m of soil for both sites in the 5th BCcycle of the reference simulation, and corresponded to HQ ' 2000 (Figure 28). Maximum GLP and AMPA concentrations of nearly 1 and 2 µg L−1, respectively, were measured in groundwater samples in Italy (Paris et al., 2016). Similar concentrations were numerically predicted at 1.5 m depth. Our predictions highlight a risk (HQ ≥ 1 ) of harmful exposure to one or a mixture of contaminants within 1.5 m depth (Figure 28).

90 Vineyard Wheat field 0 GLP GLP 0.5 aq aq AMPA AMPA aq aq 1 GLP +AMPA GLP +AMPA aq aq aq aq 1.5 2 = 1 = 1

(m) 2.5 z

3 HQ HQ 3.5 4 4.5 HQ HQ≥ HQ HQ≥ no risk, <1 risk, 1 (a) no risk, <1 risk, 1 (b) 5 10−6 10−4 10−2 100 102 104 10−6 10−4 10−2 100 102 104 Hazard Quotient (−) Hazard Quotient (−)

Figure 28: Representation of the Hazard Quotient for dissolved GLP and AMPA as a function of soil depth for the reference simulation in (a) vineyard and (b) wheat ﬁeld. Safety levels refer to the European standards for good-quality groundwater, and they are equal to 0.1 µg L−1 for a single pesticide and to 0.5 µg L−1 for a mixture of pesticides.

The Hazard Quotient (HQ) as a function of depth over time for GLP, AMPA, and their mixture is represented in Figure 29. Because GLP leaching was higher in the vineyard’s shallower aquifer than in the wheat field’s deeper aquifer, the safety level HQ = 1 was exceeded down to greater depths in the former setting than in the second one. The effects of groundwater fluctuations on HQ dynamics is particularly evident below the root zone at low concentrations of GLP and AMPA.

GLP and AMPA biodegradation rate. GLP oxidation via P1R1 and P1R1s was the main mechanism for GLP biodegradation for the vineyard in the 5th BCcycle of the reference simulation (38% and 10%, respectively, of GLP biodegraded mass), while hydrolysis via P2R1s accounted for 43% of GLP biodegraded mass (Figure 30a). Thus, 48% of applied GLP was converted into AMPA; only 23% by mass of produced AMPA underwent biodegradation. Similar results were found for the wheat ﬁeld; GLP and AMPA biodegradation increased up to 45% via P2R1s and up to 34% via P1R2s (Figure 30b). These predictions are in line with previous observations by Zablotowicz et al. (2009), who reported GLP mineralization rates as high as 70% within 35 days after application. However, AMPA poses an increasing environmental issue due to its slow degradation rate (Al-Rajab & Hakami, 2014; la Cecilia & Maggi, 2018). This supports the importance of current practices according to which herbicide (re)approval must account for the release of toxic metabolites. GLP and AMPA birnessite-mediated chemical degradation was not accounted for in this numerical study. Although it could be a predominant degradation pathway (la Cecilia & Maggi, 2018), processes inhibiting GLP and AMPA chemical degradation, such as competition for catalytic sites between Cu2+ ions and GLP (Barrett & McBride, 2005), have not been investi-

91 Vineyard 0 1000 3 1000 450 100 2 1 10 1000 100 1 1 100 10 0 1 ) (−) 2 −1 (m) HQ (

z −2

3 −5 10 −5 −5 1 −3

1 1 log 4 −4 −5 (a) GLP (b) AMPA (c) GLP + AMPA 5 −6 90 92 94 96 98 100 90 92 94 96 98 100 90 92 94 96 98 100 t (y) t (y) t (y)

Wheat field 0 1000 1000 600 3 600 2 10 1000 1 100 1 1 100 10 1 0 1 ) (−) 2 −1 (m) HQ (

z −2

3 10 −5 −3

−5 −5 1 log 4 1 1 −4 −5 (d) GLP (e) AMPA (f) GLP + AMPA 5 −6 90 92 94 96 98 100 90 92 94 96 98 100 90 92 94 96 98 100 t (y) t (y) t (y)

Figure 29: Hazard Quotient (HQ) over depth and time at the investigated sites: (a) and (b) represent the HQ for GLP and AMPA in a vineyard, (c) represent the HQ for the mixture GLP and AMPA in a vineyard, (d) and (e) represent the HQ for GLP and AMPA in a wheat field, (f) represent the HQ for the mixture GLP and AMPA in a wheat field. Thick black line highlights HQ = 1. Blue line represents the groundwater table. The threshold value for the calculation of HQ is 0.1 µg L−1 for the single compounds and 0.5 µg L−1 for their mixture. Results are for the reference simulation. gated in field conditions. A beneficial aspect of GLP chemical degradation is that any produced 3 – intermediate organophosphate is rapidly degraded to PO4 (Paudel et al., 2015).

GLP and AMPA distribution. Adsorbed GLP and AMPA represented 19% and 76%, respectively, of the total mass of phytotoxic molecules above the root zone of the vineyard in the 5th

BCcycle of the reference simulation, while 2% was AMPA in the aqueous phase (Figure 31a). Adsorbed AMPA was nearly 3% by mass below the root zone. Adsorbed GLP and AMPA represented 23% and 73% of the total mass of phytotoxic molecules above the root zone of the wheat field, while 2% was AMPA in the aqueous phase (Figure 31b). Adsorbed AMPA was nearly 3% by mass below the root zone. These results are in line with lysimeter studies and phase partitioning along the profile in Al-Rajab & Hakami (2014). It has been shown that phosphorus (P) fertilization can promote GLP and AMPA leaching, and that the leaching mass is greater when GLP applications are concomitant with rainfall events (Sasal et al., 2015). How- ever, P application rates of 80 kg-P ha−1y−1 can decrease GLP adsorption capacity by 25% to 44% (Munira et al., 2016). Since GLP can adsorb very firmly to soil particles, a reduction by 3 – 44% may not result in an adsorption strength lower than that of PO4 . Moreover, Munira et al. (2016) showed that most GLP dispersion results from wind and transport of GLP-sediment ag-

92 Vineyard Wheat field CH Olow CH Oref CH Ohigh CH Olow CH Oref CH Ohigh 2 2 2 2 2 2

GLP (P1R1) (a) GLP (P1R1) (b) GLP (P1R1s) GLP (P1R1s) GLP (P2R1s) GLP (P2R1s) AMPA (P1R2s) AMPA (P1R2s) 100

40 Biodegradation fraction (%)

0 0.7 1.4 2.9 0.7 1.4 2.9 0.7 1.4 2.9 1.0 2.0 4.0 1.0 2.0 4.0 1.0 2.0 4.0 GLP application rate (kg ha−1 y−1) GLP application rate (kg ha−1 y−1)

Figure 30: GLP and AMPA biodegradation fraction relative to applied GLP and produced AMPA, respectively, averaged over the 5th BCcycle, as a function of GLP application rate and CH2O release from SOM in (a) vineyard and (b) wheat ﬁeld. gregates rather than leaching and runoﬀ of dissolved GLP. As a comparison, GLP and AMPA spreading at a rate of about 20 mg ha−1 y−1 due to the former processes was found by Silva et al. (2017).

Vineyard Wheat field CH Olow CH Oref CH Ohigh CH Olow CH Oref CH Ohigh 2 2 2 2 2 2 (a) (b) AMPAbRZ 100 ad GLPaRZ ad AMPAbRZ ad AMPAaRZ aq GLPaRZ 10 aq Mass fraction (%) 1

0.1 0.7 1.4 2.9 0.7 1.4 2.9 0.7 1.4 2.9 1.0 2.0 4.0 1.0 2.0 4.0 1.0 2.0 4.0 GLP application rate (kg ha−1 y−1) GLP application rate (kg ha−1 y−1)

Figure 31: GLP and AMPA mass fraction in aqueous (aq) and adsorbed (ad) phases, above (aRZ) and below (bRZ) the root zone, as a function of GLP application rate and CH2O release from SOM in (a) vineyard and (b) wheat ﬁeld.

GLP and CH2O sensitivity analyses. Increasing GLP application rates did not aﬀect GLP and

AMPA biodegradation in the test areas (Figure 30), while increasing CH2O release from SOM

93 not only enhanced GLP and AMPA biodegradation, but also enhanced the fraction of GLP biodegraded along the beneficial pathway P2R1s. Similarly, varying GLP application rates did not affect GLP and AMPA partitioning along the soil profile in the test areas (Figure 31), while increasing CH2O release from SOM decreased the aqueous GLP mass fraction to below 0.1%. This result suggests that AMPA can become an even more important emergent contaminant if greater GLP applications will be adopted to control GLP-resistant weeds. Despite the generally high GLP biodegradation, GLP and AMPA residues reached 6 mg −1 −1 −1 −1 kgdry-soil and 17 mg kgdry-soil, respectively, in the vineyard and 7 mg kgdry-soil and 15 mg kgdry-soil in the wheat field (Figure 32). These concentrations fall within the values measured in agricultural soils (Silva et al., 2017). Increasing CH2O release from SOM resulted in GLP and −1 −1 −1 AMPA residues of 3 mg kgdry-soil and 14 mg kgdry-soil in the vineyard and of 4 mg kgdry-soil and −1 19 mg kgdry-soil in the wheat field (Figure 32). Interestingly, total GLP average soil residues decreased nonlinearly as CH2O increased (orange and blue stacked bars in Figure 32), while

AMPA showed a maximum average concentration at the reference CH2O release rate (brown and yellow stacked bars in Figure 32). This is explained by the interplay between P1R1 and P2R1s in GLP biodegradation.

Vineyard Wheat field CH Olow CH Oref CH Ohigh CH Olow CH Oref CH Ohigh 2 2 2 2 2 2

(a) (b) AMPAaRZ 100 aq+ad GLPaRZ aq+ad AMPAbRZ ) aq+ad 10 BaRZ HyO

−1 dry−soil GLPbRZ aq+ad 1 BbRZ HyO

0.1 Concentration (mg kg 0.01

0.001 0.7 1.4 2.9 0.7 1.4 2.9 0.7 1.4 2.9 1.0 2.0 4.0 1.0 2.0 4.0 1.0 2.0 4.0 GLP application rate (kg ha−1 y−1) GLP application rate (kg ha−1 y−1)

Figure 32: Average concentrations of GLP and AMPA biodegraders (BHyO), GLP, and AMPA, in the aqueous (aq) and adsorbed (ad) phases, above (aRZ) and below (bRZ) the root zone, as a function of GLP application rate and

CH2O release from SOM in (a) vineyard and (b) wheat ﬁeld.

GLP age and turnover time. The overall GLP age in the reference simulation was similar for the two sites (Figure 33a and b); however, a substantial discontinuity was found in the vineyard, corresponding to a drop in the clay fraction at 0.8 m depth. Also, the vineyard experienced slightly larger groundwater ﬂuctuations, which exacerbated the discontinuity because of higher frequency of dilutions and downward transport.

The overall GLP turnover time Tt,GLP was similar for the two sites (Figure 33c and d); however, the diﬀerent contribution played by physical or biological processes was noticeable. In

94 line with the previous results, the GLP turnover time was shorter for the vineyard and longer for the wheat field, and was explained by the same effect of groundwater fluctuations as above. To the best of our knowledge, this mechanism of GLP mobilization remains yet to be systematically investigated in the field. Amongst the biological pathways, P1R1 and P2R1s provided a similar contribution in the top 10 cm of soil, while P2R1s allowed for the shortest turnover time along the soil profile. This was likely due to a higher affinity (lower K, Table 16) of BHyO to scavenge small GLP concentrations along P2R1s, an effect that decreased with increasing soil depth.

Vineyard Wheat field 0 Average GLP age Soil 1 Soil 1 GLP age standard deviation 0.5 Soil 2 Soil 2 1 Soil 3 Soil 3 1.5 Soil 3 Depth (m) Soil 3

2 (a) (b) 2.5 10−2 10−1 100 101 102 10−2 10−1 100 101 102 Age (y) Age (y)

0 T Soil 1 t,GLP Soil 1 T 0.5 t,GLP−physical T Soil 2 t,GLP−P1R1 Soil 2 T 1 Soil 3 t,GLP−P1R1s Soil 3 T t,GLP−P2R1s 1.5 Depth (m) Soil 3 Soil 3 2 (c) (d) 2.5 100 101 102 103 104 105 100 101 102 103 104 105 Turnover time (y) Turnover time (y)

Figure 33: (a) and (b) represent GLP age in a vineyard and a wheat ﬁeld, respectively; (c) and (d) represent the overall GLP turnover time and the turnover time associated to speciﬁc causative physical and biological processes. Dashed lines indicate the interfaces between soil layers. Results are for the reference simulation.

GLP tracking and breakthrough curve. The ecohydrological conditions corresponding to the

11 years of the BCcycle inﬂuenced the time necessary to remove the applied GLP mass in the ﬁrst 10 cm of soil (Figure 34). The minimum, average, and maximum removal times corresponding to MGLP(t)/MGLP0 = 1% are tabulated in Table 15. Bento et al. (2016) found that it took 280±40 days to remove 90% of applied GLP mass, which is in accordance with our predictions. The minimum, average, and maximum half life values corresponding to MGLP(t)/MGLP0 = 50% are tabulated in Table 15. The predicted half life values are well within or in accordance with the value of 151 days found by Bergström et al. (2011). The biodegraded GLP mass corresponding to GLP disappearance from the top soil over time is represented in Figure 35, while minimum, average, and maximum values are tabulated

95 in Table 15. This result stresses that the biological activity explained up to 32% of GLP mass removed from the top soil, while the remaining percentage was advected to lower soil layers depending on the ecohydrological conditions MGLP,advected(t)% = (1-MGLP(t)/MGLP0 )×100 -

MGLP,biodegraded(t) %.

GLP removal time GLP fraction biodegraded GLP half life GLP fraction biodegraded (M /M = 1%) (M /M = 50%) GLP GLP0 GLP GLP0 min average max min average max min average max min average max (d) (d) (d) (%) (%) (%) (d) (d) (d) (%) (%) (%) Vineyard 125 150 166 19 26 31 60 84 103 11 16 20 Wheat ﬁeld 290 328 360 29 30 32 145 157 170 14 15 18

Table 15: GLP removal times and corresponding fraction of GLP biodegraded mass and GLP half life values and corresponding fraction of GLP biodegraded mass in top 10 cm of soil.

Vineyard ) 10 360 −1 40 80 40 80 40 80 160 270

1 180 (d) t

90 O : SOM (mol mol 2 (a) (b) (c)

CH 0.1 0 Wheat field ) 10 360 −1 60 60 60 120 180 120 180 120 180 360 270

1 180 (d) t

90 O : SOM (mol mol 2 (d) (e) (f)

CH 0.1 0 99 75 50 25 1 99 75 50 25 1 99 75 50 25 1 M /M (%) M /M (%) M /M (%) GLP GLP GLP GLP GLP GLP 0 0 0

Figure 34: GLP breakthrough curves in the top 10 cm of soil as a function of CH2O release from SOM. (a) minimum, (b) average, and (c) maximum GLP disappearance times in the vineyard amongst 11 repetitions representing diﬀerent boundary conditions of the BCcycle. Similarly, (d) minimum, (e) average, and (f) maximum GLP disappearance times in the wheat ﬁeld. Thick vertical black lines correspond to the condition MGLP(t)/MGLP0 = 50%.

Inter-annual variability in GLP biodegradation. The cumulative inter-annual GLP biodegraded fraction was found to decrease for increasing cumulative water inputs (Figure 36). This relationship was weaker at low and normal CH2O bioavailability for both sites when irrigation water was provided. Note that GLP was applied twice in the vineyard, and only the ﬁrst was followed by irrigation. Water transport likely played a smaller role in GLP removal from the top soil in the not-irrigated vineyard, where a longer GLP exposure to biodegraders resulted in higher GLP biodegraded fractions. Irrigation increased GLP leaching from the top 10 cm of soil as compared to no irrigation in the vineyard, thus dry conditions caused GLP accumulation in the top soil as shown by the fewer realizations for which MGLP(t)/MGLP0 ≤ 0.01 between two

96 ae al yais(uvsi rysaei iue3) riainicesdGPleaching GLP increased Irrigation 37). Figure in scale and gray biomass in and (curves forcing dynamics meteorological as table such water processes, several on depends fraction graded di amongst resources. water surface and ground nearby of vulnerability the pref- of via consequence (Munira transport runo aggregates a GLP-sediment and GLP as macropores trigger likely soil may zone through events flows root extreme erential However, below leaching strength. GLP adsorption of high risk GLP the rainfall increase intense contrast, not In did biodegradation. events GLP and growth bacteria favor events rainfall quent 37). (Figure dynamics table water and biomass as di well of as result a practices as application year GLP hydrological each to specific are that variabilities field. show wheat biodegraded and GLP vineyard in the increase both two-fold in a scenario nearly reference and the soil to of compared cm as 10 fraction the in 40% least at to fraction used was application GLP one only where field, frac- wheat biodegraded the GLP 36c). for between (Figure found relationship was similar input A water and b). tion and 36a (Figure applications GLP condition the to correspond lines black vertical Thick field. wheat di representing the repetitions in 11 times CH BC amongst of the vineyard function the of a in conditions as times boundary soil disappearance ent of GLP cm maximum 10 (c) top and average, the (b) in mass biodegraded GLP 35: Figure −1 −1

h eainhpbtenGPboerddfato n ae nu aidwti and within varied input water and fraction biodegraded GLP between relationship The Aslam inputs water corresponding the and fractions biodegraded cumulative GLP time-varying The biodegraded GLP cumulative the increased bioavailability source C in increase two-fold A CH O : SOM (mol mol ) CH O : SOM (mol mol ) 2 2 0.1 0.1 10 10 99 1 1 (d) (a) tal. et ﬀ M rn yrlgclyas ntesm er h ieeouino h L biode- GLP the of evolution time the year, same the In years. hydrological erent 75 10

GLP 10 21)sgetdta pia olmitr otn eutn rmml n fre- and mild from resulting content moisture soil optimal that suggested (2015) /M 50

GLP 20 20 0 (%) 25

cycle 40 1 iial,()mnmm e vrg,ad()mxmmGPdisappearance GLP maximum (f) and average, (e) minimum, (d) Similarly, . 99 tal. et (e) (b)

M 10 75

06;teepoessmyb eeatfridentifying for relevant be may processes these 2016); , 10 Wheat field GLP Vineyard

/M 20 20 50 97 GLP ﬀ 0 (%) 25 fdsovdGP(Lefrancq GLP dissolved of 40

1 40 99

(f) (c)

2 10 ees rmSM a minimum, (a) SOM. from release O M M 10 75 GLP GLP 20 20 ( /M t ) 50 / GLP M ﬀ GLP rn riainand irrigation erent 0

(%) 40 25 0

= 40 tal. et 50%. 50 1

07 or 2017) , 0 20 40 60 0 20 40 60 ﬀ

er- GLP biodegraded (%) GLP biodegraded (%) Vineyard - Not irrigated Vineyard - Irrigated Wheat field - Irrigated 60 CH Ohigh (a) (b) (c) 2 50 CH Oref 2 40 CH Olow 2 30 20 10 0 Biodegraded percentage (%) 0 200 400 600 800 0 200 400 600 800 0 500 1000 1500 Cumulative water input (mm)Cumulative water input (mm)Cumulative water input (mm)

Figure 36: GLP biodegraded mass fraction in the top 10 cm of soil as a function of water inputs and CH2O release from SOM and GLP applied at the reference rate, amongst 11 repetitions representing different boundary conditions of the BCcycle. Circles represent the realizations for which MGLP(t)/MGLP0 ≤ 0.01. (a) Not-irrigated vineyard in the period between 01-Oct and 31-Mar after the autumn GLP application, (b) irrigated vineyard in the period between 01-Apr and 30-Sep after the spring GLP application, and (c) irrigated wheat field in the period between the autumn GLP applications. from the top 10 cm of soil in the vineyard as shown by the higher occurrence of realizations with MGLP(t)/MGLP0 = 0.01 (black bullets in Figure 37a-f). In contrast, drier conditions caused GLP accumulation in the top soil as no further GLP mass disappeared at the time of the next GLP application (not all curves terminate with a black bullet in Figure 37a-f). The wheat field displayed patterns similar to the irrigated vineyard, with some hydrological years resulting in either complete disappearance of GLP or accumulation. Amongst different years, a negative relationship was found between GLP biodegraded fraction and water input for each GLP disappeared mass fraction (each single scatter of bullets with the same color in Figure 37). The relationship was significant at high CH2O bioavailability but it may turn out to be not significant at lower CH2O bioavailability or ratios of MGLP(t)/MGLP0 down to 0.5. This analysis shows that for a given GLP disappeared mass fraction, an increase in water input results in a decrease in GLP biodegradation, with the remaining GLP mass being transported to lower soil depths.

Optimization of microbial biodegradation pathways. In this work, biological reactions and their corresponding kinetic parameters were estimated from laboratory experiments and were kept constant over time. However, this may have led to underestimation of biodegradation and overestimation of soil GLP and AMPA concentrations, as well as GLP removal times. Although newly synthesized herbicides may be very difficult to biodegrade, some may have a chemical structure similar to compounds already biodegradable, and very small genetic adjustments would be needed for bacteria to consume them (Arbeli & Fuentes, 2007). This is referred to as "cross acclimation", that is, the capability to consume a new substrate elicited by the previous capability to consume a similar one. Early inefficiencies in degradation pathways may be off- set via cometabolization of an additional C source for supplying the energy necessary to break chemical bonds. These cometabolic reactions may allow scavenging of crucial macronutrients such as P, N, and S. Similarly, microorganisms can cometabolically use GLP as a P source (Dick & Quinn, 1995) and a N source (Balthazor & Hallas, 1986). Some bacteria have adapted to use

98 Figure 37: Time evolution of GLP biodegraded mass fraction in the top 10 cm of soil as a function of water inputs and CH2O release from SOM, amongst 11 repetitions representing diﬀerent boundary conditions of the BCcycle. Results refer to scenarios with GLP applications at the reference rate. Same colors represent same fraction of GLP disappeared mass from the top soil as a result of reactive transport. (a) low, (b) reference, and (c) high CH2O release from SOM in the vineyard after the autumn GLP application. Similarly, (d) low, (e) reference, and (f) high

CH2O release from SOM in the vineyard after the spring GLP application. Similarly, (g) low, (h) reference, and (i) high CH2O release from SOM in the wheat ﬁeld after the autumn GLP application. MGLP,adv stands for advected GLP mass, while MGLP,biod stands for biodegraded GLP mass.

GLP as a C source (Mcauliffe et al., 1990) with no need of a co-substrate. Microorganisms can also manifest enhanced contaminants biodegradation, and more references can be found in Arbeli & Fuentes (2007). This process is stimulated by continuous exposure to a contaminant, which may trigger selection pressure and an increase in the number of best-fitted degraders in the soil community. Evidence of enhanced biodegradation should promote the characterization of new or optimized reaction pathways and their corresponding kinetic parameters via laboratory experiments, so that contaminants reaction networks are always up-to-date. This effort will result in more reliable modeling frameworks, which may be relied upon by authorities to update policies on pesticides uses and best management practices. Other mechanisms for enhancing contaminants biodegradation may involve gene regulation optimization, such as for the metabolism of organophosphates (McGrath et al., 2013) possibly including also GLP and AMPA. Laboratory experiments suggested that biodegradation of GLP 3 – (Pipke & Amrhein, 1988b) and AMPA (Balthazor & Hallas, 1986) liberating PO4 were in-

99 3 – hibited by the availability of exogenous PO4 . However, Duke (2011) noticed that GLP and 3 – AMPA were biodegraded in agricultural soils, which very likely contained PO4 ions. Sup- 3 – pression of PO4 inhibition of GLP and AMPA biodegradation via P2R1s and P1R2s may result in higher biodegradation rates. It has been generally assumed that adsorbed contaminants are not bioavailable, thus do not undergo biodegradation. However, it is not uncommon that adsorbed pesticides can be biodegraded, albeit at a very slow rate. For example, Park et al. (2003) found that some bacteria consumed adsorbed atrazine, while others could biodegrade adsorbed GLP (Schnürer et al., 2006; Eberbach, 1998). A first attempt to predict adsorbed GLP was carried out by Zaranyika & Nyandoro (1993) for sediment particles in the aquatic environment. Later, Eberbach (1998) estimated half life values relative to the biodegradation of dissolved and adsorbed GLP in soil to be 9 and between 222 to 835 days, respectively. Microorganisms can exploit different strategies to consume adsorbed compounds, such as those suggested in Park et al. (2003), Huang & Schnitzer (1986), and Schnürer et al. (2006), as well as produce new extracellular enzymes or enhance the activity of old ones. Biodegradation of adsorbed compounds will affect the compounds distribution and decrease their residence time in soil. In this work, it was assumed that neither GLP nor AMPA were toxic to microorganisms. Therefore, no "killing" or inhibition terms were accounted for at high GLP or AMPA concentrations. However, GLP was shown to cause adverse effects on some non-target species including soil microorganisms (Aristilde et al., 2017; Newman et al., 2016; Nguyen et al., 2016), especially in areas with no previous history of GLP contamination. Toxicity could result in no or low biodegradation. However, GLP and AMPA degraders are abundant in agricultural soils that have experienced GLP applications (Dick & Quinn, 1995). Absence of toxicity by GLP and AMPA in these bacteria is corroborated by GLP half life as low as 1.5 days (Bento et al., 2016). Once again, selective pressure may have acted to increase the number of GLP degraders to the detriment of species diversity (Dick & Quinn, 1995). Available bioreactive models neglect microbial adaptation to contaminants; as a consequence, those models may overestimate half life values (Krutz et al., 2010b). In the past, typical time scales for bacteria to achieve such enhanced capabilities were 10 to 20 years (Krutz et al., 2008). However, nowadays such enhanced degradation may take very short time scales to manifest because microorganisms can easily biodegrade new xenobiotics as a result of cross acclimation (Arbeli & Fuentes, 2007). Thus we can assume that biodegradation pathways are, as a matter of fact mediated by specific enzymes (Raillard et al., 2001), which catalyze degradation reactions at optimal rates. When this does not hold true, such as when microorganisms do not undergo cross acclimation or require a new enzyme to degraded xenobiotics (like for AMPA, Hove-Jensen et al., 2014), adaptation should be described mechanistically. However, such quantitative treatment is still missing. Because enhanced degradation cannot be predicted, periodic monitoring studies aiming to keep data up-to-date on the population of degraders, biodegradation pathways and their kinetics are recommended.

100 Other biological processes inﬂuencing GLP reaction network. Higher organisms as well as other culturable and unculturable microorganisms can degrade GLP (Forlani et al., 1999), even if to a smaller extent as compared to characterized bacteria (Borggaard & Gimsing, 2008). For example, ten fungal strains were found to use GLP as C or P source (Krzysko-Łupicka´ & Orlik, 1997); also some crop plants can oxidize GLP to AMPA, but still to a small extent (Duke, 1988, 2011). Since plants were found to substantially degrade other herbicides (Coleman et al., 2002), it may be worthy to monitor whether GLP-resistant weeds (Heap, 2016) will develop enhanced GLP and AMPA degradation capabilities, thus contributing to decrease the risk for GLP and AMPA persistence in the environment.

101 5.5. Biochemical reaction and MMM kinetic parameters

All the kinetic and equilibrium reactions implemented in this biogeochemical system are reported in Table 16 with their corresponding parameters.

Pathway Kinetic Parameters Functional group

Kinetic biological aqueous reaction µ KKI YY (s−1) (M) (M) (g-C-Biomass g-C-Substrate−1) (mg-wet-Biomass mol- Substrate−1)

(a) → × −5 × −3 P1R1s C3H8NPO5 + CH2O(aq) + 2 O2(aq) CH6NPO3 + 3.17 10 1.04 10 BHyO GLP AMPA C2H2O4 + CO2(aq) + H2O(aq) GLX 1.26 × 10−4 1.03 × 10−1 2.46 × 104 1.40 × 10−5 (b) → × −5 × −3 × −2 × 4 P1R1 C3H8NPO5 + O2(aq) CH6NPO3 + C2H2O4 3.35 10 4.05 10 3.86 10 2.78 10 BHyO GLP AMPA GLX 1.40 × 10−5 (c) → + × −5 × −4 3 – × −4 P2R1s C3H8NPO5 + CH2O(aq) + O2(aq) C3H7NO2 + 3 H + 3.34 10 1.09 10 PO4 2.53 10 BHyO GLP SRC 3 – PO4 + CO2(aq) 2.12 × 10−4 1.52 × 10−1 3.64 × 104 1.40 × 10−5 (d) → + × −6 × −3 3 – × −4 P1R2s CH6NPO3 + CH2O(aq) + O2(aq) CH5N + 3 H + 5.04 10 2.08 10 PO4 2.53 10 BHyO AMPA MTH 3 – PO4 + CO2(aq) 1.38 × 10−4 1.73 × 10−2 4.14 × 103 1.40 × 10−5 (e) 1 → × −4 × −4 × −3 × 2 P1R3a CH5N + 2 O2(aq) CH2O(aq) + NH3(aq) 1.39 10 2.15 10 2.66 10 6.39 10 BAER MTH 1.40 × 10−5 ( f ) 1 → × −4 × −1 × −6 × −3 × 2 P1R3b CH5N + 2 H2O(aq) CH4(aq) + CO2(aq) + NH3(aq) 1.17 10 5.38 10 O2(aq) 3.125 10 1.29 10 3.09 10 BANAER MTH (g) 1 → × −3 × −5 × −3 × 3 P2R2a C3H7NO2 + 2 O2(aq) C2H5NO2 + CH2O(aq) 4.08 10 3.37 10 2.50 10 1.80 10 BAER SRC GLY 1.40 × 10−5 (h) → × −5 × −4 × −6 × −2 × 4 P2R2b C3H7NO2 + CH2O(aq) + H2O(aq) CH5N + 5.36 10 2.95 10 O2(aq) 3.125 10 6.87 10 4.95 10 BANAER SRC MTH + 2 CH2O(aq) + CO2 +2 H 4.39 × 10−3 (g) 3 → × −4 × −4 × −4 × 2 P2R3a C2H5NO2 + 2 O2(aq) 2 CO2(aq) + NH3(aq) + 1.22 10 1.06 10 5.21 10 2.50 10 BAER GLY H2O(aq) 1.40 × 10−5 (i) 1 → 3 × −4 × −1 × −6 × −5 × 1 P2R3b C2H5NO2 + 2 H2O(aq) 2 CH2O(aq) + CO2(aq) + 2.20 10 2.94 10 O2(aq) 3.125 10 9.25 10 4.44 10 BANAER GLY NH3(aq) (l) → × −5 × −4 × −2 × 4 R1a CH2O + O2 CO2(aq) + H2O(aq) 2.55 10 1.55 10 9.36 10 2.25 10 BHyO 1.40 × 10−5 (m) → × −8 × −7 × −1 × 4 R1b CH2O + O2 CO2(aq) + H2O(aq) 5.60 10 1.00 10 4.00 10 1.44 10 BAER 1.40 × 10−5 (n) → – + × −5 × −4 × 4 R2 NH3 + 3/2 O2(aq) H2O(aq) + NO2 + H 1.0694 10 3.0410 10 - 0.5 10 BAOB 1.50 × 10−4 (n) – → – × −5 × −4 × 4 R3 NO2 + 1/2 O2(aq) NO3 3.5966 10 2.9840 10 - 0.4 10 BNOB 1.50 × 10−4 (n) – → – + – × −4 × −4 × −6 × 4 R4 NO3 +1/2 CH2O(aq) 1/2 HCO3 + 1/2 H + NO2 4.0704 10 2.0677 10 O2(aq) 2.50 10 - 0.75 10 BDEN 2.0703 ×10−4 -- (n) – + → – × −5 × −4 × −6 × 4 R5 NO2 +1/4 CH2O(aq) + 3/4 H 1/4 HCO3 + NO(aq) 9.6768 10 7.4892 10 O2(aq) 2.50 10 - 0.375 10 BDEN + 1/2 H2O 1.3599 ×10−4 -- (n) → + – × −4 × −4 × −6 × 4 R6 NO(aq) +1/4 CH2O(aq) 1/4 H + 1/4 HCO3 + 1/2 8.0080 10 1.7551 10 O2(aq) 2.50 10 - 0.375 10 BDEN N2O(aq) 6.2404 ×10−5 -- (n) → + – × −5 × −5 × −6 × 4 R7 N2O(aq) +1/2 CH2O(aq) 1/2 H + 1/2 HCO3 + 1 1.6846 10 5.1698 10 O2(aq) 2.50 10 - 0.75 10 BDEN N2(aq) 8.8059 ×10−5 -- Kinetic chemical aqueous reaction µ K (s−1) (M) (n) – + → – × −11 × −4 R8 NO2 +2/3 H 1/3 H2O +1/3 NO3 +2/3 NO(aq) 1.0742 10 1.129 10 (o) → + 3 – −11 R9 SOM CH2O + 0.039 NH4 + 0.0011 PO4 10 - ◦ Equilibrium aqueous complexation reactions (at T = 25 ) log10(Kaq) (p) – + −−−* R10 OH + H )−−− H2O(aq) 13.99 (p) + −−−* + R11 NH4 )−−− H + NH3(aq) -9.24 (p) −−−* + – R12 CO2(aq) + H2O )−−− H + HCO3 -6.34 ◦ Equilibrium gas dissolution reactions (at 25 ) log10(Kg) (p) −−−* R13 O2(aq) )−−− O2(g) 2.8980 (p) −−−* + – R14 CO2(g) + H2O(aq) )−−− H + HCO3 -7.8136 (p) + −−−* + R15 NH3(g) + H )−−− NH4 11.038 (p) −−−* R16 N2(aq) )−−− N2(g) 3.2451 R17(p) NO(aq) )−−−−−−* NO(g) 2.7609 (p) −−−* R18 N2O(aq) )−−− N2O(g) 1.6021 ◦ Equilibrium adsorption reactions (at 25 ) log10(Kad)

102 R19(q) GLP(aq) )−−−−−−* GLP(ad) 1.67 (at 20◦ C) R20(q) AMPA(aq) )−−−−−−* AMPA(ad) 1.67 (at 20◦ C) (r) + −−−* + R21 NH4 )−−− NH4 (ad) -0.336 (s) – −−−* – R22 NO3 )−−− NO3 (ad) -1.072 (t) – −−−* – R23 NO2 )−−− NO2 (ad) -1.072

BHyO encompasses Achromobacter Group V D, Agrobacterium radiobacter, Arthrobacter sp. GLP-1, Flavobacterium sp. GD1, Pseudomonas sp. LBr, and Pseudomonas PG298; BAER encompasses Arthrobacter P1 and Pseudomonas Ovalis;BANAER encompasses Clostridium purinolyticum, Methanosarcina barkeri and Eubacterium acidaminophilum;BDEN encompasses the genus Pseudomonas and Thiobacillum;BAOB encompasses the genus Nitrosomona and Nitrosospira;BNOB encompasses the genus Nitrobacter and Nitrospira. (a) Parameters averaged from estimations against experiments in Balthazor & Hallas (1986); Jacob et al. (1988); (b) Parameters estimated against experiments in Mcauliﬀe et al. (1990); (c) Parameters estimated against experiments in Moore et al. (1983); (d) Parameters estimated against experiments in Balthazor & Hallas (1986); (e) Parameters estimated against experiments in Levering et al. (1981); (f) Parameters estimated against experiments in Hippe et al. (1979); (g) Parameters averaged from estimations against experiments in Appleyard & Woods (1956); (h) Parameters estimated against experiments in Hormann & Andreesen (1989); (i) Parameters estimated against experiments in Därre &

Andreesen (1982b); (l) BHyO was assumed to grow on CH2O as an independent reaction, with MMM kinetic parameters averaged from estimations against experiments in Balthazor & Hallas (1986); Jacob et al. (1988); Moore et al. (1983). (m) Parameters reported in Riley et al. (2014) for Monosaccharides were converted for CH2O; (n) Parameters estimated in Maggi et al. (2008) against experiments in Venterea & Rolston (2000); (o) Parameters assumed in Maggi et al. (2008) afterDon & Schulze (2014) and stoichiometry taken from Tipping et al. (2016); (p) Parameters from EQ3/6 Wolery (1992); (q) Parameters estimated against experiments in Sidoli et al. (2016); (r) Parameters estimated in Tang – (2016) against experiments in Ding et al. (2010); (s) Parameters estimated in Tang (2016) against experiments in Li & Bowman (2001); (t) Parameters assumed to be similar to those of NO3 as in Tang (2016). 3 – The Michaelis-Menten (MM) constants are listed in the order of appearance of corresponding reactants. PO4 inhibition was assumed for GLP and AMPA biodegradation via P2R1s and P1R2s, respectively. All kinetic biological aqueous + −9 −5 reactions also include a MM term and an inhibition term relative to H concentration with constants 10 and 10 M, respectively. Oxidative biological reactions also include a MM term relative to O2(aq). All anaerobic biological −6 −1 reactions include an inhibition constant relative to O2(aq) concentration. The microbial mortality was assumed to be 10 s for all the microbial functional groups.

Table 16: Kinetic and equilibrium reactions implemented in this biogeochemical system together with their corresponding parameters.

103 5.6. Summary

The simulations presented in this chapter highlighted complex nonlinear interactions amongst microbial dynamics and varying boundary conditions. The use of coupled physical, chemical, and biological processes describing herbicides biodegradation network under varying ecohydrological boundary conditions allowed us to quantify dispersion, degradation, and aquifer contamination by ATZ, GLP, and their metabolites in relevant agroecosystems worldwide. The proposed framework sets a crucial benchmark to develop models for environmental risk assessments under different land management practices that alter both microbial ecology and feedbacks amongst soil biogeochemical processes. Groundwater pollution was forecast to occur after decades of herbicide applications; at that time, the root zone already contained high amount of residues. Therefore, even if herbicide use was to be discontinued, pollution would be a persistent issue due to likely remobilization and leaching. This insight support the idea to isolate bacteria with high affinity for ATZ, GLP, and their metabolites in order to effectively remediate polluted soils from small residual concentrations. The sensitivity analyses suggested the importance of a surplus of simple C-sources (i.e., CH2O), which substantially enhanced pollution control. Future studies may quantify uncertainties related to model structure and parameter estimation by means of sensitivity analyses to evaluate the confidence in the models. It was here suggested the importance to understand the catalytic potential of Mn-oxides towards GLP and AMPA removal under field conditions. GLP and AMPA chemical degradation would be crucial below the root zone where microbial activity is generally negligible. Actions to strengthen collaborations and discussions amongst stakeholders with modelers are strongly recommended in order to frame comprehensive systems and modeling scenarios, and to address stakeholders inquiries as much and as early as possible.

104 6. Sensitivity analyses

6.1. Introduction

This Chapter highlights nonlinearities in the ATZ and GLP reaction networks fully-coupled with biogeochemical systems forced by varying eco-hydrological boundary conditions as numerically assessed in the previous Sections 5.3.1 and 5.4.1.

6.2. Atrazine

Contents of this chapter come from the article Porta et al. (2018)7 published as the result of a collaboration project with the Politecnico di Milano. Porta et al. (2018) propagated the uncertainty of the kinetic parameters corresponding to the ATZ reaction network (Section 4.4.2) under the real-case scenario presented in Section 5.3.1. Twenty-thousands simulations were used to predict the residues of toxic chlorine-containing molecules in the soil column given one-at-a-time changes of the parameter values within a range (the system setup is displayed in Figure 38). The distribution of the ATZ mass followed a unimodal probability density function, whereas the distribution of the four metabolites were multimodal (Figure 39). The latter result highlighted that diﬀerent metabolites can be found in soil due to the kinetic parameters assigned to the microbial functional groups of ATZ biodegraders and soil carbon utilizers. This is a consequence of the presence of important switches regulating the diﬀerent biodegradation pathways within the reaction network. The study used the Sobol’ indices (Sobol’, 1993) to quantify the contribution of selected kinetic parameters to the variance of the distribution of target model outcomes f (p) as

R 2 E f (p)|pk] − E[ f (p) ρ(pk)dpk U = Γk , (56) k Var[ f (p)]

th where Γk is the domain of variability of the k parameter p, and ρ(pk) is the probability density function associated with pk. Index Uk defines the fraction of Var[ f ] that is associated with the uncertainty of pk and is typically defined as principal Sobol’ index. The joint effect of two parameters, e.g., [pk1 , pk2 ], can be evaluated as R E f (p)|p , p ] − E[ f (p)2 ρ(p )dp Γ k1 k2 k1,k2 k1,k2 U = k1,k2 − U − U . (57) k1,k2 Var[ f (p)] k1 k2 Yet, the study quantified the relative contribution of single kinetic parameters to the average of the distribution of target model outcomes by means of the AMAE index (AMA index for expected value E) (Dell’Oca et al., 2017) and to the variance by means of the AMAV index

7Porta, G., la Cecilia, D., Guadagnini, A., and Maggi, F. (2018). Implications of uncertain biogeochemical parameters on a complex reaction network of atrazine biodegradation in soil. Advances in Water Resources, 121, pp. 494-498, 10.1016/j.advwatres.2018.08.002.

105 (AMA index for variance value V) (Dell’Oca et al., 2017) deﬁned as

R E f (p)|pk] − E[ f (p) ρ(pk)dpk AMAE = Γk , (58) k |E[ f (p)]|

R Var[ f (p)] − Var f (p)|pk ρ(pk)dpk AMAV = Γk , (59) k Var[ f (p)] where E and V indicate that the index measures the importance of pk to the average (expected value) and variance of a model output, respectively. One important ﬁnding of the study was that the capability of ATZ biodegraders to quickly grow on soil carbon substantially promoted ATZ biodegradation thank to their higher overall biomass concentration (Figure 40). Multimodality in metabolites masses was suggested to arise because of the presence of competing bioreactive processes in a carbon-limited setting, which may be typical of conventional agrosystems. The last important result regarded the diﬀerent type of information provided by the AMAE index and the Sobol one. The set of AMA indices allowed us to quantify the impacts of parameters variability to multiple statistical moments of the model target outcome distribution, whereas the Sobol index did not. The advantage lays in the provision of a more comprehensive understanding of the overall system, and therefore, in the possibility to identify meaningful connections. For example, the growth of BAER on CH2O, described as reaction R3 in Table 10, had a substantial impact on the average ATZ mass in soil but not on its metabolites, while it had an impact on the variance of ATZ and its metabolites.

Although, the Sobol index captured an eﬀect on the variance of ATZ mass only (reaction R4,2 in Figure 41), it allowed to quantify the larger contribution to the output uncertainty resulting from the combined variability of two or more kinetic parameters rather than single ones (Figure 42). Again, variability of multiple processes over space and time is to be expected in conventional agrosystems.

ATZ applica*on

0 0 0 0 1.5m

0.5 0.5 0.5 0.5 Root 1 1 1 1 zone 1.5 1.5 1.5 1.5 O by SOM 2

degrada*on 2 2 2 2 CH

2.5 2.5 2.5 2.5

3.5m 3 3 3 3 Depth [m] Depth

3.5 3.5 3.5 3.5

4 4 4 4

4.5 4.5 4.5 4.5

5 5 5 5 0 0.5 1 0 0.5 1 0 0.2 0.4 0 1 2 #10-5 #10-4

Sl ATZ BATZoxi CH2O [mol/L] [mg/L] [mol/L]

Figure 38: On the left, the sketch illustrates the soil column and its division in root zone (RZ) and below root zone

(BRZ). ATZ was applied at the soil surface, while CH2O was released by SOM in the first 90 cm of soil depth. On the right, the simulated profiles of the liquid saturation (S l), and the concentrations of ATZ, ATZ oxidizers, and CH2O. Profiles over time are colored in blue, while those at the final time step = 100 years are colored in red.

106 TT aso l hs irba ucinlgop)rsligfo h 000simulations. 20,000 the from resulting groups) functional microbial these all for of functions mass density (TOT) probability Sample 40: Figure simulations. 20,000 the from resulting molecules these all of mass for (TOT) functions total the density or CLHOATZ probability Sample 39: Figure

pdf pdf B M s 107 M B (where n (where s n [mg/m [mol/m s tnsfrAZy,AZx,AR n h total the and AER, ATZoxi, ATZhyd, for stands n tnsfrAZ IT,DAZ DIDEATZ, DEATZ, DIATZ, ATZ, for stands 2 ] 2 ] a) b)

R4,3 R4,3

R4,2 R4,2

R4,1 R4,1

R3,2 R3,2

R3,1 R3,1 R2,1 R2,1

R1,3 R1,3 R1,2 R1,2 R1,1 R1,1 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 AMAE AMAV U AMAE AMAV U c) d)

R4,3 R4,3

R4,2 R4,2

R4,1 R4,1

R3,2 R3,2

R3,1 R3,1 R2,1 R2,1

R1,3 R1,3 R1,2 R1,2 R1,1 R1,1 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 0 0.1 0.2 AMAE AMAV U AMAE AMAV U

Figure 41: Principal sensitivity indices evaluated for a) MATZ, b) MDIATZ, c) MDEATZ, d) MDIDEATZ. Given i and i j the index of the reaction and j and the index of the pathway composing the ATZ reaction network in Figure 15 and Table 10, green bars indicate sensitivity indices AMAE(i, j), AMAV(i, j), U(i, j) to the specific biomass affinity −1 (i, j) (i, j) (i, j) Φi, j, as defined in Eq. (10) with B0 = 1 mg L . Blue bars indicate indices AMAE , AMAV , and U . For (i, j) (i, j) (i, j) each Ri, j we display three bars which express sensitivity to µ (bottom bar), Y (middle bar) and K (top bar). (i, j) Only sensitivity to µ is expressed for R6,1. Note that, each of the three bars within a row Ri, j corresponds to a unique AMAEk, AMAVk, and Uk as they appear in Eqs. 58, 59, and 56.

BT nR =1

BAER nR =2

BAT Zoxi nR > 2

BAT Zhyd

M AT Z

M DIATZ

M DEATZ

M DIDEATZ

Sobol Indices

(i, j) Figure 42: Analysis of variance: sum of principal indices U associated with a single reaction (nR = 1, blue bars), sum of indices accounting for joint variations of Φi, j associated with two (nR = 2, green bars) or more (nR i j > 2, red bars) reactions. i and the index of the reaction and j and the index of the pathway composing the ATZ reaction network in Figure 15 and Table 10

108 6.3. Glyphosate

Two different exercises were carried out to investigate the uncertainty of biological parameters in a 0-D simplified scenario and one exercise to investigate the uncertainty of soil hydraulic parameters in a 1-D real-case scenario. These analyses highlighted that likely pathways can be predicted using simplified analyses; however, environmental pollution can only be robustly predicted in comprehensive models accounting for soil heterogeneities, biogeochemical feedbacks, and boundary conditions.

6.3.1. Reaction path model in a batch-type system

Contents of this chapter come from scientific works carried out during PhD candidature of the author, such as the manuscript la Cecilia & Maggi (Under Review)8 submitted to Mathematics and Computers in Simulation9 and the conference paper la Cecilia & Maggi (2017b)10. The influence of the MMM kinetic parameters corresponding to the GLP reaction network developed in Section 4.5.2 on the soil concentration of GLP and AMPA was assessed in a 0- D scenario using steady boundary conditions. GLP and AMPA dynamics were numerically investigated in a 1 L bioreactor. GLP at 0.003 M concentration and an additional carbon source −1 (CH2O) at 0.001 M concentration were released at a Q = 0.0036 L h flow rate in an aqueous −1 solution without and with birnessite mineral at 1.20 g kgdry-soil concentration, with constant −1 −1 pH = 7 and O2 levels = 3 mg L . The bacteria mortality rate δ (s ) was assumed to be −6 −1 3 – constant and equal to 10 s after Gastrin et al. (1968). Phosphate (PO4 ) inhibitory effect on GLP and AMPA biodegradation along P1R1 and P1R2, respectively, was accounted for −4 using an inhibition value KI = 2.53 × 10 M estimated against observations in Balthazor & Hallas (1986). Substrate competition was not included in this work due to the limited variety of substrates available. O2 consumption in aerobic reactions was accounted for using a MM value −5 −6 K = 1.40 × 10 M after Button & Garver (1966), while an inhibition value KI = 3.125 × 10

M was used for O2 inhibition on anaerobic processes (adapted from Kindred & Celia (1989)). The pH effect on biological activity was accounted for by using a K = 10−9 M for high pH and −5 an inhibition value KI = 10 M for low pH, respectively, after Boon & Laudelout (1962). The wide spectrum of interconnected catabolic reactions, each occurring at a different rate, as well as uncertainties in kinetic parameters estimation, suggest variability in modeling outcomes, which were quantified by means of a sensitivity analysis. A suite of sensitivity analyses

8la Cecilia, D. and Maggi, F. (2017). Stochastic sensitivity analysis of glyphosate biochemical degradation. In Syme, G., Hatton MacDonald, D., Fulton, B. and Piantadosi, J. (eds) MODSIM2017, 22nd International Congress on Modelling and Simulation. Modelling and Simulation Society of Australia and New Zealand, December 2017, pp. 257 - 263, isbn 978-0-9872143-7-9, https://www.mssanz.org.au/modsim2017/B3/lacecilia.pdf. 9la Cecilia, D. and Maggi. F (Under Review). Inﬂuential sources of uncertainty in glyphosate biochemical degradation in soil. Mathematics and Computers in Simulation. Manuscript Number: MATCOM-D-18-00335. 10la Cecilia, D. and Maggi, F. (2017). Stochastic sensitivity analysis of glyphosate biochemical degradation. In Syme, G., Hatton MacDonald, D., Fulton, B. and Piantadosi, J. (eds) MODSIM2017, 22nd International Congress on Modelling and Simulation. Modelling and Simulation Society of Australia and New Zealand, December 2017, pp. 257 - 263, isbn 978-0-9872143-7-9, https://www.mssanz.org.au/modsim2017/B3/lacecilia.pdf.

109 were run to assess the uncertainty to GLP and AMPA equilibrium concentrations resulting from a specific group of MMM kinetic parameters (i.e., µ, K, or Y) or a specific biological reaction (where the reactions P1R1s, P2R1s, P1R1, and P1R2s described in Table 16 were here labeled EQs 1, 2, 3, and 4, respectively). To this aim, the MMM kinetic parameters relative to one group and to EQs 1 to 4, were randomly chosen from a Gaussian distribution with mean equal to the corresponding experimentally retrieved parameter and standard deviation (σ) equal to 5, 10, 15, 20, 25, and 30% of that value, per each analysis. For the stochastic sensitivity analysis, 2000 simulations were run for each group and for each σ. Simulations were repeated with and without accounting for the effect of chemical degradation after Barrett & McBride (2005) observed that ions may inhibit GLP and AMPA degradation by Mn-oxides. The difference between GLP equilibrium concentration predicted in each model run (GLPc,sto and GLPsto, with and without birnessite respectively) and the concentration predicted using experimentally retrieved parameter values (GLPc,ref and GLPref, with and without birnessite respectively) was used as the sensitivity measure (SMc,GLP = GLPc,sto - GLPc,ref and SMGLP = GLPsto - GLPref). The same approach was repeated for AMPA; therefore, the difference between AMPA equilibrium concentration predicted in each model run (AMPAc,sto and AMPAsto, with and without birnessite respectively) and the concentration predicted using average parameter values (AMPAc,ref and AMPAref) was calculated as SMc,AMPA = AMPAc,sto - AMPAc,ref and SMAMPA = AMPAsto -

AMPAref.

Under this simpliﬁed batch-type scenario, GLP and AMPA concentrations showed mono- modal distributions (Figure 43). When abiotic catalytic reactions were not accounted for, output distributions were more skewed, GLP and AMPA concentrations were higher, and output ranges were larger. AMPA concentrations were higher than GLP ones, highlighting that produced AMPA was slowly biodegraded and suggesting that AMPA can be a more concerning pollutant than GLP in the environment. GLP and AMPA distribution skewness was opposed, meaning that GLP biodegradation to AMPA rather than SRC was the preferential pathway in the reaction network because the more GLP was degraded the more AMPA was produced.

Chemical and biological processes collaborated to fast degrade GLP (Figure 44). Lower µ resulted in slower biodegradation rates, which were flanked by the catalytic action of birnessite mineral. The lowest µ values caused the mineral surface to become saturated; in this case, GLP concentration increased. In the lack of birnessite, the increasing variability in µ resulted in a nonlinear increase in GLP concentration. Biotic processes alone could fast degrade GLP; low µ resulted in a substantial increase in GLP concentration, while high µ did not substantially decrease it. Increasing variability in K resulted in lower GLP concentration both with and without birnessite. This is because GLP application concentration was similar to K; low K substantially increased the biodegradation rate, while high K did not decrease it likewise. This result stressed the importance to consider herbicides application rate together with the available biodegraders for effective pollution control and bioremediation. Similarly, increasing variability in Y resulted in lower GLP concentration. In the presence of birnessite, bacteria consumed small amounts of substrate; therefore, varying Y did not substantially affect GLP. In the lack of birnessite, high

110 Y resulted in a trade oﬀ between a slower degradation rate but a higher biomass concentration; conversely, low Y resulted in faster rates but lower biomass concentration. Therefore, GLP concentration did not change on average. As far as biotic processes were concerned, reaction P1R1 mostly drove the GLP reaction network because the average of SMc,GLP and SMGLP substantially changed as the parameter values relative to EQ3 changed (red horizontal lines in Figure 45a and c, boxplots in 3rd, 7th, and 11th column); P1R1s contributed little to the reaction network, while P2R1s and P1R2s did not aﬀect the reaction network (Figure 45a, boxplots in 1st, 2nd, and 4th column, respectively). The in-silico analysis in la Cecilia & Maggi (2018) suggested that a higher availability of an additional C source would have enhanced GLP biodegradation along the cometabolic pathways P1R1s and P2R1s as well as AMPA biodegradation along P1R2s. Results from EQ3 showed that higher GLPc,sto (therefore greater positive SMc,GLP) resulted from lower µ values (or high

K or Y values, Figure 45a) and corroborated that Y did not affect GLPsto, that is when there was no birnessite mineral (Figure 45c, 9th to 12th column). EQ3 also decreased the model output variability as indicated by the smaller SMc,GLP and SMGLP range for EQ3 compared to those relative to EQs 1, 2, and 4 (Figure 45a and c). EQ4 influenced the least the reaction network. In fact, this reaction involves AMPA biodegradation, which poorly contributes to GLP biodegraders growth (i.e., Y relative to AMPA is 1 order of magnitude lower than Y relative to GLP as reported in Table 6) and occurs at a slow rate (Figure 45b and d). GLP biodegradation to AMPA described by EQ3 was found to be the most important regulatory process on the reaction network; therefore, it was expected that EQ3 influenced SMc,AMPA and SMAMPA as well. A faster AMPA production was not followed by the same increase in AMPA degradation rate, thus it accumulated. In the event that microorganisms degrade GLP to AMPA, then AMPA would pose an even more serious risk to the environment.

111 µ K Y σ = 10%

200 (a) (c) (e) GLP c,sto mean(GLP ) 150 c,sto GLP /5 sto mean(GLP )/5 100 sto GLP c,ref GLP /5 50 ref number of observations

0 6 8 10 12 6 8 10 12 6 8 10 12 [GLP] (g kg−1 ) × 10−4 [GLP] (g kg−1 ) × 10−4 [GLP] (g kg−1 ) × 10−4 dry−soil dry−soil dry−soil 200 (b) (d) (f) AMPA c,sto mean(AMPA ) 150 c,sto AMPA /40 sto mean(AMPA )/40 100 sto AMPA c,ref AMPA /40 50 ref number of observations

0 1 1.5 2 1 1.5 2 1 1.5 2 [AMPA] (g kg−1 ) × 10−3 [AMPA] (g kg−1 ) × 10−3 [AMPA] (g kg−1 ) × 10−3 dry−soil dry−soil dry−soil

Figure 43: Distribution of GLPc,sto and GLPsto around GLPc,ref and GLPref, respectively, in (a), (c), and (e) and AMPAc,sto and AMPAsto around AMPAc,ref and AMPAc,ref, respectively, in (b), (d), and (f). σ = 10%. Number of bins were chosen according to Freedman-Diaconis’ rule.

µ K Y 15 GLP c 10 GLP

Relative change (%) −5 10 AMPA 8 c 6 AMPA

Relative change (%) −2 5% 10% 15% 20% 25% 30% 5% 10% 15% 20% 25% 30% 5% 10% 15% 20% 25% 30% σ σ σ

Figure 44: Relative change in GLPc,sto and GLPsto with respect to GLPc,ref and GLPref, respectively, as a function of σ in the upper panel; relative change in AMPAc,sto and AMPAsto with respect to AMPAc,ref and AMPAref, respectively, as a function of σ in the lower panel.

112 −5 × 10 With Birnessite × 10−4 With Birnessite 4.5 4.5

0 0 ) ) −4.5 −4.5

4.5 4.5 −1 dry−soil −1 dry−soil

0 0 (g kg (g kg −4.5 −4.5

4.5 4.5 c,GLP c,AMPA

SM 0

0 SM CDF<0.33 0.330.66 CDF<0.33 0.330.66 −4.5 µ K Y (a) µ K Y (b)

−3 × 10 Without Birnessite × 10−3 Without Birnessite 1.5 1.8

0 0 )

) −1.8 −1.5

1.5 1.8 −1 dry−soil −1 dry−soil 0 0 (g kg

(g kg −1.8 −1.5

GLP 1.5 1.8 AMPA SM 0 SM 0 CDF<0.33 0.330.66 −1.8 CDF<0.33 0.330.66 −1.5 µ K Y (c) µ K Y (d) 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1: P1R1s 2: P2R1s 3: P1R1 4: P1R2s

Figure 45: Boxplot showing the outcome variability in SMc,GLP (a), SMGLP (c), SMc,AMPA (b), and SMAMPA (d), grouped horizontally by parameter quantiles (i.e., Q1 = 33th and Q2 = 66th) and vertically by MMM kinetic parameter, and organized by equation number. Black horizontal lines indicate SMc,GLP, SMGLP, SMc,AMPA, and SMAMPA equal to 0. σ = 10%.

113 6.3.2. 1-D real-case scenario

Contents of this chapter come from scientific work carried out by Giovanni Porta and presented at the international conference CMWR201811 as a result of a collaborative project with the Po- litecnico di Milano. The analyses have provided material to be presented at the International Conference on Uncertainty in Risk Analysis, Berlin, 20-22 February 2019, organized by the German Federal Institute for Risk Assessment (BfR) and the European Food and Safety Author- ity (EFSA) and to draft a scientific paper, which is expected to be submitted to a peer-reviewed journal soon. The GLP reaction network developed in Section 4.5.2 and tested for the real-case scenario presented in Section 5.4.1 relative to the wheat field was used to quantify GLP biodegradation and groundwater contamination under uncertainty of three soil hydraulic parameters: the soil absolute permeability k (m2), the pore volume distribution index b (-), and the air entry potential at saturation Ψs (m). The latter were used to calculate soil water dynamics according to the Brooks and Corey model (Brooks & Corey, 1962). Because water carries solutes, an increase in downward fluxes would promote GLP leaching. Five-thousands simulations were used to predict the residues of GLP and AMPA in the soil column given one-at-a-time changes of the parameter values within a range (the system setup is displayed in Figure 46). Two regions were identified in the soil column depending on the presence of wheat roots, which are the root zone (RZ) and the below root zone (BRZ). After a lag time of approximately 10 years, the microbial functional group of GLP biodegraders achieved a biodegradation efficiency of nearly 1, meaning that all applied GLP was biodegraded. Microbial activity and biodegradation efficiency were influenced by varying ecohydrological boundary conditions; sometimes, the availability of GLP residues from previous years resulted in biodegradation efficiency greater than 1 (Figure 47). Variability in biodegradation efficiency due to parameter values uncertainty was therefore negligible. Despite the high biodegradability of GLP, Figure 48 showed that the concentration of the herbicide exceeded the safety threshold set to 0.1 µg L−1 by the European Commission (EC Directive 2006/118/EC, 2006) relative to predicted aqueous concentrations of one single pesticide at 1 m depth. The time needed to exceeds the safety limit ranged between 13 and 20 years depending on parameter values. The sensitivity analysis revealed nonlinear interactions between soil hydraulic properties and contaminant dispersion and biodegradation. The parameter k(RZ) and k(BRZ) played the larger contribution to the variance of GLP concentrations below the root zone; the parameter b(RZ) and Ψs(BRZ) played a smaller but still significant contribution (Figure 49a). All the parameters were important relatively to the skewness of GLP concentrations below the root zone (Figure 49b). The importance of the parameters suggested by the AMAE index was also found when the relationship between the exceedance time and parameter variability was investigated

11la Cecilia, D, Porta, G., Riva, M., Vervoort, RW., Coleman, NV, Tang, FH, and Maggi, F. (2018). Propagation of ecohydrological uncertainty in a complex biogeochemical network of Glyphosate dispersion and degradation. Computational Methods in Water Resources (CMWR) XXII. Bridging gaps between data, models, and predictions, p. 154.

114 (Figure 49c-f). Increasing k(RZ) and k(BRZ) resulted in decreasing exceedance times, whereas increasing b(RZ) and Ψ(BRZ) resulted in increasing exceedance times. The analysis of the AMAE index over time revealed that the importance of parameter variability relatively to GLP concentration below the root zone changes over time as well (Figure 50). While k(BRZ) was the most influent parameter throughout the simulations, the parameters relative to the root zone quickly lost their influence. Finally, the overall probability to exceed the safety threshold can be calculated by means of the AMAP index, which was newly developed by Giovanni Porta for this research. Given Pthr = P g(p) > thr the probability that the quantity g(p) exceeds the user-defined threshold thr (Figure 51, left hand side), the AMAP index is written as Z AMAPk = Pthr − P g(p)|pk > thr ρ(pk)dpk, (60) Γk which quantifies the impact of the kth parameter p on the probability to exceed thr given a parameter space Γ (Figure 51, right hand side). The European Commission can usually grant approval of (re)use of herbicides for the following 10 or 15 years. The AMAP index could be used to quantitatively determine the approval period, and possibly inform environmental protection agencies on contamination expected time at different soil depths.

Figure 46: The sketch illustrates the soil column and its division in root zone (RZ), below root zone (BRZ), and aquifer. GLP was applied at the soil surface.

Figure 47: Time variation of biodegraded GLP normalized by applied GLP in the root zone. For each scenario the continuous line represent the mean value, and the dashed lines identify variations of one standard deviation about the mean.

115 Figure 48: Time variation of GLP concentration in groundwater. The continuous line represent the mean value, and the dashed lines identify variations of two standard deviations about the mean. Light gray curves represent the individual realizations for the two models and black horizontal lines indicate the threshold concentration value according to [2006/118/EC, 2006].

a) b) γ AMAE AMAE AMA

K b ψS K b ψS K b ψS K b ψS

c) d) e)

[y]

t ˆ

K(RZ) b(RZ) ψs(RZ) f) g) h)

[y]

t ˆ

K(BRZ) b(BRZ) ψs(BRZ)

Figure 49: Global sensitivity analysis of exceedance time tˆ: a) AMAE, b) AMAγ, while c) to h) show scatterplots of tˆ for each model realization (grey bullets), their conditional average for 10 subgroups (green circles), and unconditional average ' 16 years (black dashed horizontal line).

K(RZ) b(RZ)

ψs(RZ) K(BRZ)

AMAE AMAE tˆ b(BRZ)

ψs(BRZ)

Time [y]

Figure 50: Time evolution of AMAE sensitivity indices computed for the GLP concentration below root zone.

116 Figure 51: Probability of exceedance and time evolution of AMAP sensitivity index computed for GLP concentration below root zone with Cthr = 0.1 µg/L.

117 6.4. Summary

The sensitivity analyses presented in this chapter highlighted complex nonlinear interactions amongst multiple microbial functional groups and between biotic and abiotic processes. The capability to rank different processes with regard to herbicide biodegradation and dispersion was possible thanks to the availability of a comprehensive mechanistic model. ATZ and GLP biodegraded along different pathways. The latter can result in either toxic or non toxic metabolites depending on in-situ conditions, such as bioavailability of C, N, P, and O2 and the presence of microorganisms optimally adapted to the given herbicide soil concentration, which is a consequence of the application rate. It was found important to maintain ecological balance of the microbial community to allow growth and activity of biodegraders; the latter may have been outcompeted for some parameter values, and consequently herbicide biodegradation was reduced. The surplus of simple C-sources (i.e., CH2O) was substantially beneficial towards pollution control and soil remediation. System models made of many coupled equations mechanistically describing abiotic processes, bacteria dynamics, and their feedbacks support a robust explanation of in-situ conditions. Some of the parameters describing those equations may substantially contribute more than others. Identification of the ones driving biochemical reaction networks can help to make decisions about optimal land management. This chapter also reports the applicability of the sensitivity index AMAP as a tool to support regulatory bodies with quantitative information to determine the (re)approval period for active ingredients. AMAP can also be used by EPAs to plan the setup of monitoring campaigns relative to newly approved active ingredients.

118 7. Carbon consumption

7.1. Introduction

Contents of this chapter come from the article la Cecilia et al. (2018c) published in Soil Biology & Biochemistry12 during PhD candidature of the author. The previous chapters showed the fundamental contribution of microbes to biodegrade toxic xenobiotics and how the availability of additional C sources enhanced microbial activity, and therefore, biodegradation. The MMM kinetic framework was shown to be representative of the microbial adaptation to physiological concentration of nutrient and energy sources and their response to environmental stress (e.g., soil pH different from neutral, O2 depletion, etc...). Yet, during the development of numerical models, it has often been assumed that microorganisms can biodegrade and metabolize all the available substrates. However, this assumption may result in the wrong model structure with substantial effect on substrates consumption. In general, a variety of nutrient and energy sources would be bioavailable at the same time. Microorgan- isms have survived and thrived in highly competitive environments and they have adapted to cope with paucity of those substrates. The resilience to harsh conditions is the result of precise metabolic regulatory mechanisms. This chapter introduces a new framework to mechanistically explain microbial consumption of nutrient and energy sources. In fact, microbial metabolism is controlled by catabolite repression (CR, Magasanik, 1961) which leads microbes to grow on preferred substrates first. In particular, Catabolite Repression for Carbon (CR-C) defines the hierarchical preference of bacteria for particular C sources. This control depends on the presence of signal molecules conferring bacteria with a memory for recent growth conditions on less preferred C sources. The combined effect of catabolite repression and microbial memory (called here Memory-Associated Catabolite Repression for Carbon, MACR-C) has not yet been investigated in detail. From the modeling perspective, this chapter shows that even classic MMM kinetics would not work to predict metabolic dynamics when multiple carbon sources are bioavailable but the MMM framework is flexible enough to allow to predict those dynamics after addition of necessary terms. In particular, we formulate MACR-C using three CR-C experiments with two C sources retrieved from the literature (Dijkhuizen et al., 1980; Mukherjee & Ghosh, 1987). Finally, we performed a suite of sensitivity analyses to assess the long-term effect of MACR-C to pulse and continuous C applications into the two 2 C-source systems.

7.2. Methods

Biomass-based and enzyme-based MMM kinetics. Seven schemes of microbial growth on multiple substrates obtained using classic MMM kinetics are shown in Figure 52 together with

12la Cecilia, D., Riley, WJ, and Maggi, F. (2019). Biochemical modeling of microbial memory eﬀects and catabolite repression on soil organic carbon compounds, Soil Biology & Biochemistry, 128, pp. 1 - 12, 10.1016/j.soilbio.2018.10.003

119 their MMM kinetic equations. The MMM framework requires the estimation of 3 parameters to predict the consumption rate of a substrate with concentration S (M) as a function of the microbial biomass concentration B (mg L−1). These MMM kinetic parameters are: the reaction rate constant µ (s−1), the half-saturation concentration constant K (M), and the biomass yield −1 coeﬃcient Y (mg-wet-Biomass mol-C-Substrate ). The terms (1 + S P/KP) and (1 + S NP/KNP), where subscripts P and NP indicate the preferred and non-preferred substrate, respectively, account for competitive consumption (dashed black lines in Figure 52), which increases K and reduces the reaction velocity. The terms [KI,P/(S P + KI,P)] and [KI,NP/(S NP + KI,NP)], where

KI,P (M) and KI,NP (M) are the inhibition constants relative to the preferred and non-preferred substrate, respectively, account for inhibition (solid red lines in Figure 52) and also reduce the reaction velocity. Generally, biogeochemical models predict biochemical reactions as a function of microbial biomass. Here, we develop and test both biomass- and enzyme-based frameworks to explicitly account for cells and intracellular enzyme dynamics. The Enzyme (E) (with E in mg L−1) is −1 −1 synthesized by B at rate rE (L mol s ) depending on the corresponding substrate availability −1 (i.e., inductive enzymes), and degrades at a rate δE = δ, with δ (s ) the bacteria mortality rate;

δE and δ are generally diﬀerent (Schimel et al., 2017) but were assumed here to be the same for −1 simplicity. Enzyme production was inhibited by E with parameter KI,E (mg L ) assumed equal to 1% of the maximum biomass concentration reached in the experiments. We investigated whether classic MMM kinetics can be used to adequately replicate CR-C observations both in the biomass-based and enzyme-based frameworks. For the latter case, it was tested if CR-C aﬀected substrate catalysis (Figures 52d, e, and f) or enzyme production (Figure 52g).

MACR-C kinetics. The schemes in Section 7.2 were further developed into MACR-C kinetics to take into account the production of a memory signal following substrate consumption, which represents the memory of previous growth conditions and presumably aﬀects CR-C dynamics

(Figure 53). To this end, we assumed that the memory signals MP relative to SP consumption inhibited SNP consumption, and vice versa for MNP relative to SNP and SP. MACR-C kinetics were tested in both biomass- and enzyme-based frameworks (Figures 53a and b, respectively).

MP and MNP were assumed to be produced stoichiometrically with SP and SNP, and degraded −1 according to a ﬁrst-order reaction with rate δM (s ). EP and ENP production was assumed to undergo an additional inhibition by MP and MNP, respectively, with constants KI,MP and KI,MNP .

Experimental data of 2 C-source systems. Experimental observations of CR-C in an acetate (ACT) and oxalate (OXL) mixture with Pseudomonas oxalaticus OX1, and in a succinate (SCC) and fructose (FRC) mixture with Azospirillum brasilense were retrieved from Dijkhuizen et al. (1980) and Mukherjee & Ghosh (1987), respectively. In the ﬁrst experiment, ACT and OXL were the only C sources consumed in aerobic conditions. Azospirillum brasilense inoculum was pregrown either in ACT or OXL before incubation in two diﬀerent experiments. Pseudomonas oxalaticus OX1 can grow on both ACT and OXL, but prefers ACT.

120 In the second experiment, SCC and FRC were the only C sources consumed in aerobic conditions. Azospirillum brasilense can grow on both SCC and FRC, but prefers SCC. Those experiments are interesting because substrates were consumed hierarchically and because they allowed us to test the memory-eﬀect on the microbial CR-C hypothesis.

Method of parameter estimation. The unknown parameters µ, K, KI, KIM , rE, and δM were estimated following the procedure described in Section 4.2. The biomass yields Y = ∆B/∆S were calculated from observed changes in concentrations of biomass ∆B and substrate ∆S . Enzyme −6 −1 degradation rate (δE = 10 s ) was assumed to be the same as cell mortality. Combinations of competition and inhibition kinetics were tested as per schemes in Figures 52 and 53.

Stochastic Sensitivity Analyses on MACR-C parameters. In Dijkhuizen et al. (1980), the inoculum was pregrown in the non-preferred C-source OXL, and was next grown in the ACT-OXL mixture. In the latter medium, bacteria grew on OXL for a short time before switching to ACT until they depleted ACT, and ﬁnally they resumed OXL consumption. To our knowledge, that experiment is the most complete evidence of MACR-C; we conducted stochastic sensitivity analyses on this experiment to highlight variability in substrate consumption dynamics using the enzyme-based framework (Figure 53b). In particular, the eﬀects of δM, KIM , and rE on substrate consumption were investigated. For each parameter, 1000 simulations were run with values randomly sampled from independent Gaussian probability distributions with mean equal to the corresponding estimated value and standard deviation equal to 20% of the mean.

Steady-state sensitivity analyses on MACR-C dynamics. MACR-C kinetics were numerically explored for the two C-source systems using the scheme in Figure 53b with the parameters estimated for the case where the inoculum was pregrown with the corresponding preferred C source, ACT and SCC. One analysis elucidated to what extent MACR-C affects the C source concentrations at steady-state given varying continuous incoming C fluxes (e.g., root exudates, SOM cycle, etc...). −1 Each system was continuously amended with SP and SNP at 0.1 mL s flow rate in a 2.5 L control volume containing 1 L of water as solvent (Figure 54a); substrate concentrations ranged from 10−7 to 10−3 M s−1 for ACT, from 10−7 to 1 M s−1 for OXL, from 10−7 to 10−2 M s−1 for SCC, and from 10−7 to 1 M s−1 for FRC. Excess water was extracted from the control volume at the same flow rate to keep the water volume constant. Ten increments were used within each order of magnitude in concentration. The other analysis elucidated MACR-C effects on the short-term microbial response to sudden pulses of C substrates (e.g. leaching, etc...). MACR-C kinetics were tested for the two −1 C-source systems subject to a single pulse of SP and SNP lasting 15 s at 0.1 mL s flow rate in a

2.5 L control volume containing 1 L of water as solvent (Figure 54b). SP amendment was varied −3 −1 in concentration and was time-shifted with respect to SNP. OXL concentration was 10 M s , while ACT concentration ranged between 10−4 and 10−2 M s−1. FRC concentration was 1 M s−1, while ACT concentration ranged between 0.1 and 10 M s−1. Forty-ﬁve increments were used within each order of magnitude in concentration. The system was amended with SP at times

121 ranging from 1 to 40 days using an incremental step of 4 hours, while SNP was amended at day

10. The SNP degradation times in the presence of SP were compared to that resulting from SNP consumption alone, that is, SP was not amended in the system. The time necessary to decrease one substrate concentration below 1% of the initial concentration will be denoted t99%. Our analyses can provide mechanistic information relative to the consumption of chemically similar and diﬀerent nutrients within a mixture (e.g., ACT-OXL and SCC-FRC carbon-source systems).

7.3. Results

Biomass- and enzyme-based MMM kinetics. Classic biomass-based MMM kinetics accounting for substrate competitive consumption, non-competitive inhibition, and their combination, did not capture the observed lag phase when bacteria switched between the two C sources, which is characteristic of CR-C (Figures 55a, b, and c for ACT-OXL mixture and Figures 55d, e, and f for SCC-FRC mixture, goodness-of-fit in Table 17). Classic enzyme-based MMM kinetics did not accurately describe observed CR-C either (Figure 56a to g), except when inhibition of enzyme production was accounted for in the SCC- FRC mixture (Figure 56h, goodness-of-fit in Table 17). However, FRC consumption appeared to be slightly anticipated, with an overall rate slower than observed. Despite the additional description of enzyme production and degradation (Figures 52d, e, f, and g), our results suggest that one or more additional processes had substantial effects but were not represented by the tested MMM kinetics.

Biomass- and enzyme-based MACR-C kinetics. The MACR-C schemes in Figure 53 replicated the observations when using both biomass- and enzyme-based kinetics (Figure 57, goodness- of-ﬁt in Table 17). Accounting for substrate competitive consumption in MACR-C did not improve the ﬁtting against observations (result not shown) and added an unnecessary parameter; therefore, this mechanism was not accounted for in MACR-C kinetics.

The parameters µ, K, rE, and KI,M estimated after pregrowing the inoculum in ACT captured CR-C also after pregrowing the inoculum with the non-preferred C-source OXL in the enzyme- based framework, but not in the biomass-based framework (Table 17). Note that pregrowing bacteria on OXL rather than ACT resulted in lower Y and δM. The initial concentrations of enzymes and memory signals were calibrated because they must had been produced during previous growth conditions. Indeed, bacteria started consuming OXL thanks to the memory

MNP of pregrowth conditions (Figure 58), but they sensed ACT availability and switched to ACT within approximately 2.4 hours. OXL consumption was therefore repressed in the presence of ACT, and was resumed only after ACT was depleted. Note that none of the schemes in Figure 52 captured these dynamics (results not shown) because the corresponding kinetic equations do not explicitly account for the memory of previous growth conditions that allow microbes to consume part of the non-preferred substrate in the presence of the preferred substrate.

122 MACR-C kinetics captured CR-C dynamics both in the biomass- and the enzyme-based frameworks, and can be considered a robust representation of how memory of previous growth conditions can aﬀect substrate consumption.

Stochastic Sensitivity Analyses on ACT-OXL mixture. Stochastic sensitivity analyses on all enzyme-based MACR-C kinetic parameters (i.e., δM, KI,M, and rE) showed large variability in ACT and OXL consumption at any time after switching between C sources (Figure 59). The inhibition constant KI,M returned the largest variability in MACR-C response.

Steady-state sensitivity analysis on MACR-C dynamics. The steady-state response of the ACT and OXL and the SCC and FRC carbon-source systems to continuous substrate application at varying concentrations were different and highly nonlinear. Generally, the preferred C source concentration (ACT and SCC in Figures 60a and b, respectively) was always negligible and decreased under increased concentrations of the amended substrates. In contrast, the non-preferred OXL substrate concentration (Figure 60a) quickly built up at high ACT and OXL application concentrations. The non-preferred FRC substrate concentration (Figure 60b) also increased at increasing concentrations of amended FRC for [SCC] > 0.001 M, but drastically decreased for [SCC] < 0.001 M. The two systems responded differently and nonlinearly to single C source pulses at varying concentrations and time-shifted between each other (Figures 60c and d). OXL (SNP) consumption was slow in the lack of ACT (SP); OXL concentration took nearly 17.5 days to decrease below 1% of the initial value. ACT pulses did not substantially affect OXL kinetics; early and late SP pulses lengthened and shortened SNP consumption time by 10% (Figure 60c).

FRC (SNP) consumption occurred in nearly 7 days in the lack of SCC (SP). SCC pulses shortened the time necessary to decrease FRC concentration below 1% of the initial concentration; MACR-C eﬀect was unexpectedly strong with SP pulses shortening SNP consumption time by up to 7 times (Figure 60d).

7.4. Discussion

MMM kinetics and the MACR-C eﬀect. MMM kinetics are widely used to model biochemical processes and are being incorporated in bioreactive transport models (e.g., TOUGHREACT, Xu et al. 2011; MODFLOW-PHT3D, Prommer et al. 2001; and HYDRUS, Yu & Zheng 2010). However, this work shows that MMM kinetics are not directly applicable to model substrate consumption when regulated through catabolite repression. While a review of modeling approaches to represent catabolite repression was carried out in Kremling et al. (2015), we have shown here that CR-C can be adequately described by introducing a memory signal to inhibit substrate consumption. As reported by Stock & Zhang (2012), the memory can be a molecule, and it is characterized by production and degradation rates. We found a lower memory degradation rate in the experiment where the inoculum was pregrown on the non-preferred C source (Ta- ble 17). We hypothesize this result implies a mechanism through which bacteria are less prone

123 to switch between C sources in fluctuating environments. As suggested in Lambert & Kussell (2014), a memory may also be correlated to the availability of a substrate-specific enzyme. Be- cause both biomass-based and enzyme-based frameworks accurately captured bacterial growth dynamics in the ACT-OXL mixture after pregrowth in OXL, our analysis did not reject the idea of enzyme-carried memory, in addition to microbial memory for preferred substrates. Thus, we found need for a model containing additional parameters to test Lambert & Kussell (2014)’s suggestion. However, the available experiments were insufficient to address this question because enzyme-associated memory may play a role over longer timescales. Notwithstanding, although it is more "natural" to describe metabolic processes using enzyme-based models for biochemical reactions inside cells, this approach may result in redundant parameters and, importantly, the problem formulation can be implemented in only a few available bioreactive transport models, such as BRTSim and TOUGHREACT.

The biological role of MACR. CR is a common bacterial strategy to select the one substrate amongst many which allow optimal growth (Chu & Barnes, 2016); cells process the information in the surrounding environment, and prepare and activate certain metabolic pathways based on some sort of cost-benefit balance and cell requirements (Wang et al., 2015). We assumed those pathways involve inducible enzymes so that bacteria would not be wasting resources to produce enzymes that may seldom be used for an unknown period of time. In the proposed regulatory mechanism, bacteria can still produce any enzyme in sufficient amount to quickly switch from one substrate to another, thus gaining a fitness advantage in fluctuating environments. Yet, the timing for switching between substrates is regulated by antagonist memory signals, which inhibit the consumption of other substrates. A possible explanation for this inhibition is that cells have earlier invested resources to produce the enzymes involved in one pathway, and only under some circumstances would it be convenient to prematurely invest resources to activate a new pathway. Although we have only focused on C, other macronutrients consumption may be regulated through catabolite repression, which was neglected in this work because of the lack of suitable laboratory experiments. However, Farrell et al. (2011) reported that peptides are preferred over aminoacids as an organic N source; therefore, the Memory-Associated CR for Nitrogen

(MACR-N) could exist and should be investigated. Yet, Roca & Olsson (2001) showed that NH3 – repressed NO3 in Pseudomonas fluorescens DF57 nitrogen metabolism. Cross-talk amongst macronutrient cycles have been underpinned and reviewed by Santos-Beneit (2015), focusing on how C and N metabolisms are regulated within cells. This metabolic coupling should be accounted for in SOM models. Yet, the different MACR response to consumption of chemically similar or different nutrients, as shown here for the two C-source systems, may represent an additional level of uncertainty to the robust prediction of labile SOM cycles. The estimated kinetic parameters for the memory signal degradation rate, the substrate consumption inhibition constant, and the enzyme production rate showed large variability between these two cases. The biological explanation can be revealed by analyzing the environments the bacteria inhabit. Soil bacteria evolved to prefer tricarboxylic acid (TCA) intermediates (Mukherjee & Ghosh, 1987).

124 This evolution explains the hierarchies found between SCC and FRC consumption; the hierarchy is corroborated by Iyer et al. (2016)’s comprehensive review on SCC mediated catabolite repression of monosaccharides consumption. Despite the greater biomass yield on FRC than SCC, the former very likely requires the production of many enzymes to be broken down to pyruvate, which can eventually enter the TCA cycle and be used for energy production. This energetically-costly process may become convenient when benefits are larger than costs, that is, when FRC concentration greatly exceeds SCC concentration. This inhibitory mechanism is fully implemented in MACR-C kinetics. We hypothesize that knowledge of bacteria evolutionary trajectories can help identify representative molecules to be encompassed in comprehensive models of SOM dynamics. For instance, Riley et al. (2014) divided SOM into 11 groups, where "Organic Acids" could represent the preferred SOM group, especially when it includes TCA molecules. Such an approach would be supported by recent laboratory investigations highlighting soil bacteria preference to 101 metabolites released as root exudates (Zhalnina et al., 2018). These authors found that aromatic organic acids were the bacteria preferred substrate, despite a great variability in uptake by bacteria from the exudate mixture was observed. High uptakes were also found for amino acids, sugars, and quaternary amines. The authors also observed that microbial communities could be either positively or negatively affected in response to metabolites availability. The two groups of responders were characterized by different substrate preference and may therefore inhabit different soil niches (Zhalnina et al., 2018). As a consequence, it is possible that MACR-C kinetics coupled with environmental and edaphic conditions can provide a robust predictive tool to niche colonization and substrate consumption. This work focused on elucidating unexplored mechanisms regulating C preference in soil bacteria. Experimental evidence (Zhalnina et al., 2018) and our modeling results both suggest that MACR-C may play a pivotal role in soil C dynamics. However, it cannot be excluded that bacterial strains lacking CR-C regulatory mechanisms will differently consume available substrates (Johnson et al., 2017). Yet, different types of environments might select bacteria exploiting different mechanisms to thrive amongst other competitors. For instance, the enteric bacteria E. Coli constitutively produces enzymes involved in glucose metabolism, thus conferring an advantage in an environment where glucose is almost always available in large amounts.

Physiological meaning of MACR-C kinetic parameters. The stochastic sensitivity analysis of

MACR-C showed that KI,M, δM, and rE produced large variability in substrates concentration over time and control therefore the hierarchy for substrate consumption and the adaptation time to newly available substrates. These regulatory processes can be embedded in bacterial genetic material and constitute bacteria long-term memory (Stock & Zhang, 2012). In contrast, the initial memory signal M0 and initial enzyme concentration E0 had a negligible eﬀects on substrate consumption (not shown). We suggest M0 and E0 may represent the bacteria short-term memory produced to provide an advantage in the current environmental conditions. These parameters together quantify the strategies exploited by microorganisms to express their preference to a substrate and cope with its absence.

In this work, enzyme degradation rate δE was assumed to be constant and equal to bacteria

125 mortality δ, while some enzymes can last longer within the cell (Lambert & Kussell, 2014). A longer enzyme persistence would enhance bacterial readiness to consume the corresponding substrate over the long-term. Bacteria memory exerted a powerful inhibitory effect on substrate consumption, but degraded 2 to 3 orders of magnitude more rapidly than enzymes, thus having major effects in the short term. Despite bacteria having a preference for some substrates, this memory does not hinder their capability to grow on other substrates in case the former are not available. Our simulations were designed to investigate nonlinearities in substrates consumption under MACR-C kinetics, and for continuous amendments with C sources after steady-state and for pulse amendments with C sources. Both scenarios revealed interesting features. The bacterial long-term memory represented by KI,M and δM regulated substrate consumption for varying substrate application concentrations, which may be seen as continuous nutrients arrival from a multitude of processes including advection, SOM decomposition, or necromass recycling. Our results showed that bacteria favor consumption of those substrates allowing for the great- est growth based on availability. Next, SP pulses at varying concentrations and time-shifted with respect to a SNP pulse simulated the sudden availability of different C sources. Modeling suggested that early and late SP pulses had small but distinct effects on the consumption of chemically similar and different SNP, resulting in longer or shorter consumption times, respectively. For chemically different substrates, consumption times were always substantially shorter with SP pulses. We speculate that although bacteria prefer to use some C sources, they maintain readiness to switch to others when necessary. In contrast, there is no benefit to switch metabolic pathway for chemically similar C sources as these may have similar nutritional properties. We could not directly assess the role of short-term memory on substrate consumption, which would require the bacteria to be exposed to multiple substrates pulse. With new technologies such as microfluidic devices, bacteria adaptation to non-preferred C sources and the contributions of memory in this process can be studied experimentally (Lambert & Kussell, 2014).

MACR-C and exoenzyme regulation. The metabolic regulatory processes occurring within the cell are very complex. To reduce this complexity so that it is computationally tractable, MACR- C included 2 kinetic equations to model enzyme and memory dynamics for each substrate. Some more complex C sources such as cellulose have to be ﬁrst extracellularly decomposed to more labile substrates to become bioavailable to microorganisms. Microorganisms can liberate exoenzymes in the soil matrix to decompose SOM. Some theory for the regulatory mechanisms behind exoenzymes production (De Nobili et al., 2001; Schimel & Weintraub, 2003) suggest that small amounts of labile substrates signal the presence of SOM; these molecules trigger microbial growth and the production of exoenzymes to keep on decomposing fresh SOM until the bioavailable C is enough to balance the expenditure for exoenzymes production. MACR- C, or more generally MACR, may therefore be involved in regulating exoenzyme production depending on the characteristics of available SOM, which will be decomposed to preferred and non-preferred substrates.

126 (a) Biomass-based Competitive SP + B P d푆 휇 푆 consumption 푃 = − 푃 퐵 푃 K d푡 푌 푆 P 푃 푆푃 + 퐾푃 × 1 + 푁푃ൗ 퐾푁푃 KNP d푆 휇 푆 푁푃 = − 푁푃 퐵 푁푃 d푡 푌 푆 SNP + B P 푁푃 푆푁푃 + 퐾푁푃 × 1 + 푃ൗ 퐾푃

(b)

Biomass-based Inhibition d푆푃 휇푃 푆푃 SP + B P = − 퐵 d푡 푌푃 푆푃 + 퐾푃 K d푆 휇 푆 퐾 I,P 푁푃 = − 푁푃 퐵 푁푃 퐼, 푃 d푡 푌 푆 + 퐾 푆 + 퐾 푁푃 푁푃 푁푃 푃 퐼, 푃 SNP + B P

(c) Biomass-based Competitive d푆푃 휇푃 푆푃 SP + B P = − 퐵 consumption d푡 푌 푆 푃 푆푃 + 퐾푃 × 1 + 푁푃ൗ K 퐾푁푃 and P K d푆푁푃 휇푁푃 푆푁푃 퐾퐼 푃 I,P = − 퐵 , inhibition KNP d푡 푌 푆푃 푆 + 퐾 푁푃 푆푁푃 + 퐾푁푃 × 1 + ൗ 푃 퐼, 푃 퐾푃 SNP + B P

(d) d푆푃 휇푃 푆푃 KI,E = − 퐸 P d푡 푌 푃 푆 Enzyme-based Competitive 푃 푆푃 + 퐾푃 × 1 + 푁푃ൗ퐾 SP + EP 푁푃 consumption d푆푁푃 휇푁푃 푆푁푃 K = − 퐸 P d푡 푌 푁푃 푆 푁푃 푆푁푃 + 퐾푁푃 × 1 + 푃ൗ B + P 퐾푃 K d퐸푃 퐾퐼 퐸 NP = 푟 퐵 푆 , 푃 − 훿 퐸 d푡 퐸푃 푃 퐸 + 퐾 퐸푃 푃 S 푃 퐼, 퐸푃 NP + ENP d퐸 퐾 푁푃 = 푟 퐵 푆 퐼, 퐸푁푃 − 훿 퐸 KI,E d푡 퐸푁푃 푁푃 퐸 + 퐾 퐸푁푃 푁푃 NP 푁푃 퐼, 퐸푁푃 (e) K d푆 휇 푆 Enzyme-based Inhibition I,EP 푃 = − 푃 퐸 푃 d푡 푌 푃 푆 + 퐾 SP + EP 푃 푃 푃 d푆 휇 푆 퐾 푁푃 = − 푁푃 퐸 푁푃 퐼, 푃 d푡 푌 푁푃 푆 + 퐾 푆 + 퐾 K B + P 푁푃 푁푃 푁푃 푃 퐼, 푃 I,P d퐸 퐾 푃 = 푟 퐵 푆 퐼, 퐸푃 − 훿 퐸 d푡 퐸푃 푃 퐸 + 퐾 퐸푃 푃 S + E 푃 퐼, 퐸푃 NP NP d퐸 퐾 푁푃 퐼, 퐸푁푃 K = 푟퐸 퐵 푆푁푃 − 훿퐸 퐸푁푃 I,ENP d푡 푁푃 퐸 + 퐾 푁푃 푁푃 퐼, 퐸푁푃 (f) d푆푃 휇푃 푆푃 KI,E = − 퐸 P d푡 푌 푃 푆 Enzyme-based Competitive 푃 푆푃 + 퐾푃 × 1 + 푁푃ൗ SP + EP 퐾푁푃 d푆 휇 푆 퐾 consumption 푁푃 푁푃 푁푃 퐼, 푃 KP = − 퐸푁푃 d푡 푌 푆푃 푆 + 퐾 and B + P 푁푃 푆푁푃 + 퐾푁푃 × 1 + ൗ 푃 퐼, 푃 퐾푃 KI,P K d퐸푃 퐾퐼 퐸 inhibition NP = 푟 퐵 푆 , 푃 − 훿 퐸 d푡 퐸푃 푃 퐸 + 퐾 퐸푃 푃 S + E 푃 퐼, 퐸푃 NP NP d퐸 퐾 푁푃 퐼, 퐸푁푃 K = 푟 퐵 푆푁푃 − 훿 퐸푁푃 I,ENP d푡 퐸푁푃 퐸 + 퐾 퐸푁푃 푁푃 퐼, 퐸푁푃 (g) d푆 휇 푆 KI,E 푃 푃 푃 P = − 퐸푃 Enzyme-based Enzyme d푡 푌푃 푆푃 + 퐾푃 SP + EP production d푆푁푃 휇푁푃 푆푁푃 = − 퐸푁푃 inhibition B + P d푡 푌푁푃 푆푁푃 + 퐾푁푃 K I,P d퐸 퐾 푃 = 푟 퐵 푆 퐼, 퐸푃 − 훿 퐸 d푡 퐸푃 푃 퐸 + 퐾 퐸푃 푃 SNP + ENP 푃 퐼, 퐸푃 d퐸푁푃 퐾퐼 푃 퐾퐼 퐸 KI,E = 푟 퐵 푆 , , 푁푃 − 훿 퐸 NP d푡 퐸푁푃 푁푃 푆 + 퐾 퐸 + 퐾 퐸푁푃 푁푃 푃 퐼, 푃 푁푃 퐼, 퐸푁푃

d퐵 d푆 d푆 All schemes include = − 푃 푌 − 푁푃 푌 − 훿퐵 d푡 d푡 푃 d푡 푁푃

Preferred C Reaction Non-Preferred C Biomass Competition

Product (unspecified) Enzyme Inhibition

Figure 52: Graphical representation and mathematical description of biomass- and enzyme-based MMM kinetics: (a) Substrate competitive consumption; (b) Non-competitive inhibition of non-preferred C source by preferred C source; (c) Substrate competitive consumption and non-competitive inhibition of non-preferred C source by preferred C source; (d) Substrate competitive consumption; (e) Non-competitive inhibition of non-preferred C source by preferred C source; (f) Substrate competitive consumption and non-competitive inhibition of non-preferred C source by preferred C source; (g) Enzyme production inhibition by preferred C source. Enzyme production was under negative feedback regulation by enzyme concentration. 127 (a) KI,M NP d푆 휇 푆 퐾 푃 = − 푃 퐵 푃 퐼, 푀푁푃 d푡 푌 푆 + 퐾 푀 + 퐾 Biomass-based Inhibition SP + B MP 푃 푃 푃 푁푃 퐼, 푀푁푃 d푆 휇 푆 퐾 푁푃 = − 푁푃 퐵 푁푃 퐼, 푀푃 P d푡 푌 푆 + 퐾 푀 + 퐾 푁푃 푁푃 푁푃 푃 퐼, 푀푃

SNP + B MNP

K I,MP (b) d푆 휇 푆 퐾 푃 푃 푃 퐼, 푀푁푃 Enzyme-based Inhibition K = − 퐸푃 I,MP d푡 푌 푆 + 퐾 푀 + 퐾 K 푃 푃 푃 푁푃 퐼, 푀푁푃 I,EP d푆푁푃 휇푁푃 푆푁푃 퐾퐼 푀 S + EP M = − 퐸 , 푃 P P d푡 푌 푁푃 푆 + 퐾 푀 + 퐾 푁푃 푁푃 푁푃 푃 퐼, 푀푃

P B d퐸푃 퐾퐼 푀 퐾퐼 퐸 + = 푟 퐵 푆 , 푃 , 푃 − 훿 퐸 d푡 퐸푃 푃 푀 + 퐾 퐸 + 퐾 퐸푃 푃 푃 퐼, 푀푃 푃 퐼, 퐸푃 S + E M d퐸푁푃 퐾퐼 푀 퐾퐼 퐸 NP NP NP = 푟 퐵 푆 , 푁푃 , 푁푃 − 훿 퐸 퐸푁푃 푁푃 퐸푁푃 푁푃 K d푡 푀푁푃 + 퐾퐼 푀 퐸푁푃 + 퐾퐼 퐸 I,ENP , 푁푃 , 푁푃 K I,MNP

d퐵 d푆 d푆 All schemes include = − 푃 푌 − 푁푃 푌 − 훿퐵 d푡 d푡 푃 d푡 푁푃 d푀 d푆 푃 = − 푃 − 훿 푀 d푡 d푡 푀푃 푃 d푀 d푆 푁푃 = − 푁푃 − 훿 푀 d푡 d푡 푀푁푃 푁푃

Preferred C Biomass Reaction Non-Preferred C Enzyme Competition

Product (unspecified) Memory Inhibition

Figure 53: Graphical representation and mathematical description of MACR-C kinetics. (a) Biomass-based and (b) enzyme-based substrate non-competitive inhibition by antagonist memory. Enzyme production was under negative feedback regulation by enzyme concentration and on byproducts formation.

(a) H O + S H O + S ) 2 NP 2 P 1 -1 -1 -

Qin=0.1 mL s Qin=0.1 mL s NP

Supply Steady - State (post-

andS processing) P

MACR-C S 0

Time (days)

Continuous concentration (M s (M concentration

H2O -1 Control Volume=2.5 L Qout =0.2 mL s Water Volume=1 L

(b) H O + S H O + S

2 NP 2 P ) 1 -1 -1 - S Injection time Qin=0.1 mL s Qin=0.1 mL s P (for 15 s) (for 15 s) SNP Injection

P (post-

Supply S - processing)

MACR-C Pulse 01 10 40

Time (days) concentration (M s (M concentration

H2O Qout

Figure 54: Model scheme for (a) continuous SP and SNP amendments and (b) single pulse of SP and SNP. The manipulated variables are enclosed within brackets.

128 Biomass−based Competition Inhibition Competition & Inhibition 0.03 (a) (b) (c)

0.02

Concentration 0.01

0 0 5 10 0 5 10 0 5 10 ACT, exp. (M) OXL, exp. (M) B, exp. (mg L−1×105) ACT, model (M) OXL, model (M) B, model (mg L−1×105)

0.03 (d) (e) (f)

0.02

Concentration 0.01

0 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 t (h) t (h) t (h)

SCC, exp. (M) FRC, exp. (M) B, exp. (mg L−1×106) SCC, model (M) FRC, model (M) B, model (mg L−1×106)

Figure 55: Biomass-based kinetics. ACT-OXL and SCC-FRC consumption with the inoculum pregrown with the preferred C source ACT and SCC, respectively. (a) and (d) Substrate competitive consumption corresponding to the scheme in Figure 52a; (b) and (e) Substrate non-competitive inhibition corresponding to the scheme in Figure 52b; (c) and (f) Substrate competitive consumption and non-competitive inhibition corresponding to the scheme in Figure 52c. Estimated parameters are in Table 17.

129 Enzyme−based Inhibition on r Competition Inhibition Competition & Inhibition E 0.05 10 (a) 6 (b) 6 (c) 10 (d) 4 4 0.04 5 5 2 2 0.03 0 0 0 0 0 7 14 0 7 14 0 7 14 0 7 14 0.02 Concentration 0.01

0 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 ACT, exp. (M) OXL, exp. (M) B, exp. (mg L−1×105) ACT, model (M) OXL, model (M) B, model (mg L−1×105) E , model (mg L−1) E , model (mg L−1) ACT OXL 0.05 40 (e) 30 (f) 30 (g) 100 (h) 20 20 0.04 20 50 10 10 0.03 0 0 0 0 0 7 14 21 0 7 14 21 0 7 14 21 0 7 14 21 0.02 Concentration 0.01

0 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 0 3 6 9 12 15 18 21 t (h) t (h) t (h) t (h) SCC, exp. (M) FRC, exp. (M) B, exp. (mg L−1×106) SCC, model (M) FRC, model (M) B, model (mg L−1×106) E , model (mg L−1) E , model (mg L−1) SCC FRC

Figure 56: Enzyme-based kinetics. ACT-OXL and SCC-FRC consumption with the inoculum pregrown with the preferred C source ACT and SCC, respectively. (a) and (e) Substrate competitive consumption corresponding to the scheme in Figure 52d; (b) and (f) Substrate non-competitive inhibition corresponding to the scheme in Figure 52e; (c) and (g) Substrate competitive consumption and non-competitive inhibition corresponding to the scheme in Figure 52f; (d) and (h) Substrate inhibition term on enzyme production corresponding to the scheme in Figure 52g. Estimated parameters are in Table 17. Insets in the top right corners show enzyme dynamics.

130 Biomass−based Enzyme−based Inhibition by M Inhibition by M ACT, exp. (M) 0.05 OXL, exp. (M) 0.02 1 10 B, exp. (mg/L×105) ACT, model (M) 0.04 0.01 0.5 5 OXL, model (M) B, model (mg/L×105) 0.03 0 0 0 0 7 14 0 7 14 0 7 14 INSET left M , model (M × 10) 0.02 ACT M , model (M × 10)

Concentration OXL 0.01 INSET right (a) (b) E , model (mg L−1) 0 ACT 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 E , model (mg L−1) OXL

SCC, exp. (M) 1 1 100 0.05 FRC, exp. (M) B, exp. (mg/L×106) 0.04 0.5 0.5 50 SCC, model (M) FRC, model (M) 0.03 0 0 0 B, model (mg/L×106) 0 7 14 21 0 7 14 21 0 7 14 21 INSET left 0.02 M , model (M × 10) Concentration SCC M , model (M × 10) 0.01 FRC

Figure 57: MACR-C kinetics on ACT-OXL and SCC-FRC consumption with the inoculum pregrown with the preferred C source ACT and SCC, respectively. (a) and (c) Memory inhibition to biomass-based substrate consumption corresponding to the scheme in Figure 53a; (b) and (d) Memory inhibition to enzyme-based substrate catalysis corresponding to the scheme in Figure 53b. Estimated parameters are in Table 17. Insets in the top left and right corners show memory and enzyme dynamics, respectively.

Biomass−based Enzyme−based Inhibition Inhibition 0.04 1 0.06 10

0.04 0.03 0.5 5 0.02

0 0 0 0.02 0 5 10 15 0 5 10 15 0 5 10 15 Concentration 0.01

(a) (b) 0 0 3 6 9 12 15 0 3 6 9 12 15 t (h) t (h) ACT, exp. (M) OXL, exp. (M) B, exp. (mg L−1×105) ACT, model (M) OXL, model (M) B, model (mg L−1×105) M , model (M × 10) M , model (M × 10) INSET LEFT ACT OXL E , model (mg L−1) E , model (mg L−1) INSET RIGHT ACT OXL

Figure 58: MACR-C kinetics on ACT-OXL consumption with the inoculum pregrown with the non-preferred C-source OXL. (a) Memory inhibition to biomass-based substrate consumption corresponding to the scheme in Figure 53a; (b) Memory inhibition to enzyme-based substrate catalysis corresponding to the scheme in Figure 53b. Estimated parameters are in Table 17.

131 0.02 δ K r M I E

0.01 Concentration

(a) (b) (c) 0 0 5 10 15 0 5 10 15 0 5 10 15 t (h) t (h) t (h) ACT, exp. (M) OXL, exp. (M) B, exp. (mg L−1×105) ACT, model (M) OXL, model (M) B, model (mg L−1×105)

range(ACT), model (M) range(OXL), model (M) range(B), model (mg L−1×105)

Figure 59: Stochastic sensitivity analysis for enzyme-based MACR-C kinetics. Eﬀect on substrate consumption due to variability of: (a) memory signal degradation rate δM; (b) memory signal inhibition constant KI,M; (c) enzyme production rate rE. Colored areas show the range of predicted substrate concentrations in 1000 simulations.

10−1 1e−0 6 1e−0 8 10 −1 −7 10 1e−1 1e−1 5 7 10−7 1e−2 ) 1e−2 6 ) −1 −5 −1 4 10 10−3

1e−3 (M) 1e−3 (M) 5

0 0 −8 3 10

(B) (mg L −4 (B) (mg L 1e−4 1e−4 10 4 [OXL]

[FRC] 10 10−4 10

1e−5 −8 2 log 1e−5 log −8 −7 10 3 10 10 10−6 −5 1e−6 10 10 1e−6 10 1 2 −6 −7 (a) (b) 1e−7 1e−7 1e−7 1e−6 1e−5 1e−4 1e−3 1e−2 1e−7 1e−6 1e−5 1e−4 1e−3 1e−2 [ACT] (M) [SCC] (M) 0 0 S application S application NP NP 0.01 20 10 0.25 7

19 1 1 5 18 0.5

(M) (M)

0 0

] 0.001 ] 1 4

99% 99%

P P

t t

[S 0.9 17 [S 3 1.1 0.75 16 2 (c) (d) 0.0001 15 0.1 1 1 10 30 1 10 20 S application time (d) S application time (d) P P

Figure 60: Biomass and substrates concentrations at steady-state as a function of substrates’ release concentration at 0.1mL s−1 ﬂow rate: (a) ACT and OXL carbon-source system; (b) SCC and FRC carbon-source system. Green and red contour lines represent the preferred substrate concentration and non-preferred substrate concentration, respectively, in the control volume. Non-preferred substrate consumption time as a function of preferred substrate application time at varying concentrations: (c) ACT (SP) and OXL (SNP at 0.001 M) carbon-source system; (d) SCC −1 (SP) and FRC (SNP at 1 M) carbon-source system. Pulses lasted 15 s at 0.1 mL s ﬂow rate. Thick black vertical line indicates SNP amendment at day 10. Black contour lines represent the ratio between t99% in the presence of SP over t99% without SP.

132 7.5. Summary

The theoretical and numerical approaches developed in this chapter dealt with the robust prediction of microbial metabolic dynamics when multiple carbon sources are bioavailable. The developed mechanistic model relied on MMM-type kinetics, but accounted for underinvestigated concepts in soil biology such as bacterial memory for previous growth conditions, catabolite repression, and microbial substrate preference. The framework can be applied to explore uncertainties in the biodegradation of single and "cocktails" of xenobiotics in the presence of additional nutrients under environmental conditions. This chapter stressed the importance to provide an adequate description of microbial dynamics in order to develop robust numerical models, which ﬁnd applications in simulation of agricultural systems, mitigation and bioremediation of environmental contaminants, and prediction of environmentally-relevant ﬂuxes of chemical compounds between the air-soil interface.

133 1 2 3 4 5 6 7 8 9 10 11 12 13 Scheme SubS pregrowth Kinetic Parameters 2 µ KKI Ywet Ydry δM rE KIM R NRMSE s−1 mol L−1 mol L−1 mg-wet-Bio mol-Subs−1 g-C-Bio g-C-Subs−1 s−1 L mol−1 s−1 mg L−1 Biomass-based, Competition ACT ACT 1.67×10−4 2.60×10−6 2.59×10+5 5.39×10−1 0.97 10.86 Scheme in Figure 52a OXL 1.66×10−5 1.61×10−5 5.50×10+4 1.15×10−1 0.97 14.27 Result in Figure 55a B 0.97 7.21 Biomass-based, Competition SCC SCC 9.13×10−5 7.07×10−5 9.39×10+5 9.78×10−1 0.99 2.28 Scheme in Figure 52a FRC 1.35×10−4 2.96×10−2 1.31×10+6 9.07×10−1 0.99 6.91 Result in Figure 55d B 0.99 3.53 Biomass-based, Inhibition ACT ACT 9.22×10−5 2.60×10−6 2.59×10+5 5.39×10−1 0.99 3.43 Scheme in Figure 52b OXL 2.05×10−5 1.61×10−5 3.71×10−5 5.50×10+4 1.15×10−1 0.99 10.59 Result in Figure 55b B 0.99 6.09 Biomass-based, Inhibition SCC SCC 9.23×10−5 7.07×10−5 9.39×10+5 9.78×10−1 0.99 2.39 Scheme in Figure 52b FRC 1.45×10−4 2.50×10−2 1.05×10−5 1.31×10+6 9.07×10−1 0.99 12.93 Result in Figure 55e B 0.99 6.08 Biomass-based, Competition & Inhibition ACT ACT 2.07×10−4 2.60×10−6 2.59×10+5 5.39×10−1 0.95 9.56 Scheme in Figure 52c OXL 4.92×10−5 1.61×10−5 1.78×10−5 5.50×10+4 1.15×10−1 0.95 0.70 Result in Figure 55c B 0.95 4.12 Biomass-based, Competition & Inhibition SCC SCC 9.84×10−5 7.07×10−5 9.39×10+5 9.78×10−1 0.99 4.52 Scheme in Figure 52c FRC 8.36×10−5 2.09×10−2 1.82×10−6 1.31×10+6 9.07×10−1 0.99 10.29 Result in Figure 55f B 0.99 5.23 Biomass-based, MACR-C ACT ACT 8.87×10−5 2.60×10−6 2.79×10−1 2.59×10+5 5.39×10−1 4.14×10−4 0.99 4.23 Scheme in Figure 53a OXL 5.85×10−5 1.06×10−5 1.06×10−5 5.50×10+4 1.15×10−1 1.57×10−3 0.99 1.07 Result in Figure 57a B 0.99 2.51 Biomass-based, MACR-C SCC SCC 8.67×10−5 7.07×10−5 3.52×10−1 9.39×10+5 9.78×10−1 4.62×10−5 0.99 2.81 Scheme in Figure 53a FRC 8.18×10−4 2.23×10−2 7.03×10−4 1.31×10+6 9.07×10−1 1.25×10−4 0.99 4.09 Result in Figure 57c B 0.99 5.16 Biomass-based, MACR-C ACT 1.84×10−4 2.67×10−6 5.36×10−5 1.80×10+5 3.75×10−1 7.71×10−5 0.99 4.66 Scheme in Figure 53a OXL OXL 8.07×10−5 1.62×10−5 8.12×10−5 4.58×10+4 9.55×10−2 1.01×10−3 0.99 2.71 Result in Figure 58a B 0.99 4.40 Enzyme-based, Competition ACT ACT 5.07×10+1 2.60×10−6 2.59×10+5 5.39×10−1 6.98×10−5 0.97 7.20 Scheme in Figure 52d OXL 1.18×10+1 1.61×10−5 5.50×10+4 1.15×10−1 9.86×10−6 0.97 12.37 Result in 56a B 0.97 6.39 Enzyme-based, Competition SCC SCC 1.19×10+0 7.07×10−5 9.39×10+5 9.78×10−1 3.75×10−4 0.99 1.83 Scheme in Figure 52d FRC 4.59×10+0 1.89×10−2 1.31×10+6 9.07×10−1 7.76×10−6 0.99 7.01 Result in 56e B 0.99 3.75 Enzyme-based, Inhibition ACT ACT 2.59×10+1 2.60×10−6 2.59×10+5 5.39×10−1 6.51×10−5 0.99 1.17 Scheme in Figure 52e OXL 2.06×10+1 1.61×10−5 6.10×10−4 5.50×10+4 1.15×10−1 5.89×10−6 0.99 10.86 Result in 56b B 0.99 5.67 Enzyme-based, Inhibition SCC SCC 1.44×10+0 7.07×10−5 9.39×10+5 9.78×10−1 2.90×10−4 0.99 1.75 Scheme in Figure 52e FRC 5.61×10+0 1.17×10−2 2.15×10−4 1.31×10+6 9.07×10−1 4.49×10−6 0.99 6.61 Result in 56f B 0.99 3.53 Enzyme-based, Competition & Inhibition ACT ACT 5.66×10+1 2.60×10−6 2.59×10+5 5.39×10−1 5.13×10−5 0.98 9.40 Scheme in Figure 52f OXL 2.07×10+1 1.61×10−5 5.09×10−4 5.50×10+4 1.15×10−1 8.87×10−6 0.98 5.65 Result in 56c B 0.98 5.11 Enzyme-based, Competition & Inhibition SCC SCC 1.24×10+0 7.07×10−5 9.39×10+5 9.78×10−1 3.78×10−4 0.99 3.10 Scheme in Figure 52f FRC 3.57×10+0 1.11×10−2 3.73×10−4 1.31×10+6 9.07×10−1 6.73×10−6 0.99 7.73 Result in 56g B 0.99 6.32 +0 −6 +5 −1 −4 Enzyme-based, rE Inhibition ACT ACT 8.20×10 2.60×10 2.59×10 5.39×10 1.93×10 0.99 8.16 Scheme in Figure 52g OXL 2.04×10+1 1.61×10−5 5.50×10+4 1.15×10−1 1.60×10−5 2.96×10−5 0.99 3.98 Result in 56d B 0.99 3.90 +0 −5 +5 −1 −4 Enzyme-based, rE Inhibition SCC SCC 1.27×10 7.07×10 9.39×10 9.78×10 3.90×10 0.99 2.31 Scheme in Figure 52g FRC 2.26×10+0 2.00×10−2 1.31×10+6 9.07×10−1 5.57×10−5 6.69×10−0 0.99 5.44 Result in 56h B 0.99 1.58 Enzyme-based, MACR-C ACT ACT 1.14×10+2 2.60×10−6 8.82×10−6 2.59×10+5 5.39×10−1 2.33×10−4 1.18×10−4 3.70×10−3 0.99 4.16 Scheme in Figure 53b OXL 2.76×10+1 1.61×10−5 2.14×10−4 5.50×10+4 1.15×10−1 4.59×10−3 1.11×10−5 1.91×10−3 0.99 1.93 Result in 57a B 0.99 2.99 Enzyme-based, MACR-C SCC SCC 9.85×10−1 7.07×10−5 3.26×10−1 9.39×10+5 9.78×10−1 1.34×10−4 3.91×10−4 1.50×10−1 0.99 2.09 Scheme in Figure 53b FRC 3.10×10+0 2.00×10−1 1.61×10−4 1.31×10+6 9.07×10−1 3.81×10−3 5.05×10−5 2.70×10−5 0.99 4.00 Result in 57c B 0.99 2.47 Enzyme-based, MACR-C ACT 1.14×10+2 2.60×10−6 8.82×10−6 1.80×10+5 3.75×10−1 6.59×10−5 1.18×10−4 3.70×10−3 0.99 4.20 Scheme in Figure 53b OXL OXL 2.76×10+1 1.61×10−5 2.14×10−4 4.58×10+4 9.55×10−2 9.93×10−4 1.11×10−5 1.91×10−3 0.99 2.65 Result in 58b B 0.99 4.17

Table 17: Estimated kinetic parameters for ACT and OXL metabolism using the schemes in Figures 52 and 53. Calibration curves against observations are plotted in Figures 55, 56, 57, and 58. Tabulated goodness-of-ﬁt are against experiments from Dijkhuizen et al. (1980) and Mukherjee & Ghosh (1987). Diauxic growth of Pseu- domonas oxalaticus on ACT and OXL was conducted in aerobic conditions, pH 7.5, and T = 30◦; estimated initial biomass concentration was 256 mg-wet-Bio L−1 and 360 mg-wet-Bio L−1 when the inoculum was pregrown in ACT and OXL, respectively. Diauxic growth of Azospirillum brasilense on SCC and FRC was conducted in aerobic conditions, pH 7.6, and T = 32◦; initial biomass concentration was 205 mg-wet-Bio L−1 with the inoculum pregrown in SCC.

134 8. Conclusions and perspectives

8.1. Conclusions

An improved numerical approach to describe herbicides biochemical degradation in soil and groundwater in accordance with microbial processes was proposed and extensively investigated in these doctoral studies. Note that this thesis did not aim to provide a robust description of sorption kinetics (i.e., feedbacks by pH, metal oxides, and other variables) and solute transport in soil macropores, which would be essential for a more accurate description of contaminants dispersion at the plot scale. Rather, the developed approach was intended to set a benchmark for mechanistic tools in support of a more comprehensive and ecological-orientated understanding of the feedbacks between anthropic activities and physical, hydrological, mineral, chemical, and biological processes. Laboratory experiments aiming at isolating glyphosate (GLP) biodegraders revealed important and unexpected soil microcosms responses after exposure to GLP, which may be relevant in polluted areas and are not accounted for in traditional environmental models. These inlude:

• Diﬀerent rates of GLP biodegradation were found given diﬀerent GLP application history and soil cover;

• Longer adaptation times occurred at the highest GLP concentration;

• It is possible that unidentiﬁed bacteria or protozoa bioaccumulated GLP and the metabolite AMPA, and released them back untransformed after a substantially long period of time. This biological response to GLP and AMPA has not previously been reported,it does not involve biodegradation, and hence, a reduction in environmental pollution;

• AMPA biodegradation did not occur even after 100-day-long adaptation time, and therefore, it is an emerging recalcitrant pollutant. This result corroborated the importance of accounting for metabolites production in environmental risk assessments.

Numerical analyses simulating real-case scenarios of the herbicides atrazine (ATZ) and GLP applications in agricultural soils suggested that hydrological, chemical, and biological processes were highly nonlinearly interrelated. Varying rainfall rates strongly affected herbicides biodegradation and dispersion in soil and groundwater, where higher rates reduced biodegradation efficiency and promoted leaching. Similar implications arose when soil hydraulic parameters where changed to simulate a more and more permeable soil. Reactive minerals in soil may interfere with some microbial strategies to better use some pollutants given that steady- state GLP concentrations were insensitive to varying kinetic parameters values. However, at optimal conditions, reactive minerals were found to chemically degrade GLP and its metabolite AMPA at significantly higher rates than the microbial component. Yet, sensitivity analyses highlighted that if bacteria in-situ have different kinetic parameters than the estimated ones, then different degradation pathways may be triggered. This result is of fundamental importance

135 because unexpected hazardous metabolites can be liberated in the environment, thus contributing to environmental pollution, for which monitoring programs were not designed. Similarly, different soil nutrients availability or the lack of beneficial pollutant biodegraders in-situ would result in different degradation rates and pathways. For example, the increasing availability of an additional carbon source enhanced biodegradation processes by promoting co-metabolism of GLP as well as generally increasing the biodegraders biomass concentration. The advantage of mechanistically accounting for such important variables can allow one to predict the outcomes of land management practices and to transfer the scientific knowledge to different geographical settings. Finally, the type and concentration of nutrients and pollutants used in agricultural soils may play a crucial role in enhancing or repressing microbial activity, and more specifically the activity of pollutants biodegraders. Not only preferred organic and inorganic molecules may inhibit consumption of less preferred nutrients, but also they may enhance the growth of competitors of beneficial microbes, and generally alter soil microbial community abundance and structure. Overall, the outlooks set out in these numerical simulations suggested that sooner or later soils and aquifers will become polluted as safety concentrations were always exceeded by active ingredients, their metabolites, or both. Strong policies and heavy regulations aiming for more sustainable land management practices shall therefore need to be implemented to curb the pressure of intensive agricultural systems on the environment and the effects on people’s health.

8.2. Perspectives

The knowledge developed in these doctoral studies can influence related scientific disciplines as briefly outlined in this section. The comprehensive mechanistic approach allowed us to distinguish the contribution of physical, hydrological, chemical, and biological processes and to keep track of all the chemical species involved in the reaction networks and nutrients cycling. The reaction networks for GLP and ATZ would already be applicable for analyses at the global scale. It would be necessary to couple them with in-situ conditions described by soil properties, soil nutrients availability, ecohydrometeorological boundary conditions, and crop type and its management and protection practices. The latter data are difficult to access; therefore, collaboration amongst stakeholders is key to achieve this task. The model robustness could be further enhanced by coupling multiple reaction networks together to account for either enhancing or repressing effects on microbial activities caused by other inorganic and organic nutrients as well as pollutants. This effort will allow one to more accurately carry out environmental risk assessments as unforeseen detrimental side-effects and feedbacks relative to newly synthesized molecules on soil health and functioning would be predicted. It is possible that old synthetic molecules can be dismissed in favor of new ones. From this perspective, a model capable to describe bacterial response to lower and lower concentrations of the former molecules as well as the adaptation to the latter ones would be fundamental to infer degradation pathways and fine- tune the corresponding rates depending on exposure. As shorter adaptation times should be expected for chemically similar molecules, it should be worrying that both bacteria can become

136 resistant to a wider suite of bactericides and humans are interfering with chemical signaling used by bacteria to manage their inter- and intra-relationships. Microbes defense mechanisms following exposure to pollutants should mechanistically be examined as they may affect biodegradation efficacy and alter the soil matrix physical-chemical properties. For example, bacteria exposed to glyphosate can produce a biofilm and precipitate mineral particles in it. Given the accumulation of an increasing number of anthropogenic molecules in soils the examination of the "mixture effect" on soil microbiology and the feedbacks with the soil matrix is fundamental. Considering the large computational power scientists can resort to nowadays, it should not be a concern to increase model complexity to distinguish amongst biological responses to pollutants and environmental changes. Such innovative description of microbial strategies to cope with anthropized environments would pave the way for policies driven by an holistic point of view, which will eventually protect the natural environment and people’s health.

137

References

2007/629/EC, EC Directive. 2007. 2007/629/EC: Commission Decision of 20 September 2007 concerning the non-inclusion of triﬂuralin in Annex I to Council Directive 91/414/EEC and the withdrawal of authorisations for plant protection products containing that substance (no- tiﬁed under document number C(2007) 4282).

91/414/EEC, EC Directive. 2004. 2004/248/EC: Commission Decision of 10 March 2004 concerning the non-inclusion of atrazine in Annex I to Council Directive 91/414/EEC and the withdrawal of authorisations for plant protection products containing this active substance (Text with EEA relevance) (notiﬁed under document number C(2004) 731).

Adebusuyi, A.A., Smith, A.Y., Gray, M.R., & Foght, J.M. 2012. The EmhABC efflux pump decreases the efficiency of phenanthrene biodegradation by Pseudomonas fluorescens strain LP6a. Appl Microbiol Biotechnol, 757 – 766.

Agertved, Jeanette, Rügge, Kirsten, & Barker, James F. 1992. Transformation of the Herbicides MCPP and Atrazine under Natural Aquifer Conditions. Ground Water, 500–506.

Agrios, George. 2005. Plant pathology. Academic Press.

Al-Rajab, A.J., & Hakami, O.M. 2014. Behavior of the non-selective herbicide glyphosate in agricultural soil. American Journal of Environmental Science, 10, 94–101.

Allen, R.G., Pereira, L.S., Raes, D., & Smith, M. 1998. FAO - Food and Agriculture Organization of the United Nations, Natural Resources and Envi- ronment, Rome, 1998. ISBN 92-5-104219-5. http: // www. fao. org/ land-water/ databases-and-software/ eto-calculator/ en/ .

Alvey, S., & Crowley, E. 1997. Survival and Activity of an Atrazine-Mineralizing Bacterial Consortium in Rhizosphere Soil. Environ. Sci. Technol, 30(5), 1596–1603.

Aparicio, V.C., De Gerónimo, E., Marino, D., Primost, J., Carriquiriborde, P., & Costa, J.L. 2013. Environmental fate of glyphosate and aminomethylphosphonic acid in surface waters and soil of agricultural basins. Chemosphere, 93, 1866 – 1873.

Appleyard, G., & Woods, D. D. 1956. The pathway of creatine catabolism by Pseudomonas ovalis. J. Gen. Microbiol., 14, 351–65.

Arbeli, Z., & Fuentes, C.L. 2007. Accelerated biodegradation of pesticides: An overview of the phenomenon, its basis and possible solutions; and a discussion on the tropical dimension. crop prot, 26(12), 1733 – 1746.

Aristilde, L., Reed, M.L., Wilkes, R.A., Youngster, T., Kukurugya, M.A., Katz, V., & Sasaki, C.R.S. 2017. Glyphosate-induced speciﬁc and widespread perturbations in the metabolome of soil Pseudomonas species. Frontiers in Environmental Science, 5, 34.

139 Arnold, J.G., Srinivasan, R., Muttiah, R.S., & Williams, J.R. 1998. Large area hydrologic modeling and assessment part I: Model development. J Am Water Resour As, 34, 73–89.

Arpae-Simc. 2016. database accessed on 01.02.2017 at "http: // www. smr. arpa. emr. it/ dext3r/ ". Arpae emilia-romagna. Servizio IdroMeteoClima. Manuale Dext3r. Versione 0.95.

Arthur, E. L., Anhalt, J. C., Anderson, T. A., & Coats, J. R. 1997. Enhanced degradation of deethylatrazine in an atrazine-history soil of Iowa. J Environ Sci Health, Part B, 32(5), 599–620.

Aslam, S., Iqbal, A., Deschamps, M., Recous, S., Garnier, P., & Benoit, P. 2015. Eﬀect of rainfall regimes and mulch decomposition on the dissipation and leaching of S-metolachlor and glyphosate: a soil column experiment. pest manag sci, 71, 278–291.

Atkins, P., & De Paula, J. 2005. Elements Of Physical Chemistry. 4th edition edn. Oxford University Press, USA.

Back, A.S., & Waring, S.A. 1979. Adsorption of nitrate, chloride and sulfate by some highly weathered soils from south-west Queensland. Aust. J. Soil Res., 17, 271–282.

Balthazor, T. M., & Hallas, L. E. 1986. Glyphosate-degrading microorganisms from industrial activated sludge. Appl. Environ. Microbiol., 51, 432–434.

Bao, C., Wu, H., Li, L., Newcomer, D., Long, P.E., & Williams, K. H. 2014. Uranium Bioreduc- tion Rates across Scales: Biogeochemical Hot Moments and Hot Spots during a Biostimula- tion Experiment at Riﬂe, Colorado. Environmental Science & Technology, 48, 10116–10127.

Barrett, K. A., & McBride, M. B. 2005. Oxidative degradation of glyphosate and aminomethylphosphonate by manganese oxide. env sci tec, 39, 9223–9228.

Behki, R., & Khan, S.U. 1994. Degradation of Atrazine, Propazine, and Simazine by Rhodococ- cus Strain B-30. J. Agric. Food Chem., 42(5), 1237–1241.

Behki, R., Topp, E., Dick, W., & Germon, P. 1993. Metabolism of the Herbicide Atrazine by Rhodococcus Strains. Appl Environ Microbiol, 59(6), 1955–1959.

Bekins, B., Warren, E., & Godsy, E. 1998. A comparison of zero-order, ﬁrst-order and Monod biotransformation models. Ground Water, 36(2), 261–268.

Belser, L. W. 1979. Population ecology of nitrifying bacteria. Annual Review of Microbiology, 33, 309–333.

Bennett, A.F. 1984. Thermal dependence of muscle function. Am. J. Physiol., 247, 217 – 219.

140 Bennett, Neil D., Croke, Barry F.W., Guariso, Giorgio, Guillaume, Joseph H.A., Hamilton, Serena H., Jakeman, Anthony J., Marsili-Libelli, Stefano, Newham, Lachlan T.H., Nor- ton, John P., Perrin, Charles, Pierce, Suzanne A., Robson, Barbara, Seppelt, Ralf, Voinov, Alexey A., Fath, Brian D., & Andreassian, Vazken. 2013. Characterising performance of environmental models. Environ Modell Softw, 40, 1 – 20.

Bento, C.P.M., Yang, X., Gort, G., Xue, S., van Dam, R., Zomer, P., Mol, H.G.J., Ritsema, C.J., & Geissen, V. 2016. Persistence of glyphosate and aminomethylphosphonic acid in loess soil under diﬀerent combinations of temperature, soil moisture and light/darkness. sci tot env, 572, 301 – 311.

Berg, J. M., Tymoczko, J. L., & Stryer, L. 2002. Biochemistry. New York : W.H. Freeman.

Bergström, L., Börjesson, E., & Stenström, J. 2011. Laboratory and lysimeter studies of glyphosate and aminomethylphosphonic acid in a sand and a clay soil. J. Environ. Qual., 40, 98–108.

Bethke, Craig M. 1996. Geochemical Reaction Modeling: Concepts and Applications. New York, Oxford University Press.

Bilge, H. 2012. Upper Lachlan Groundwater Flow Model.

Blok, J. 1994. Classiﬁcation of biodegradability by growth kinetic parameters. Ecotoxicology and Environmental Safety, 27, 294–305.

Boon, B., & Laudelout, H. 1962. Kinetics of nitrite oxidation by Nitrobacter Winogradskyi. biochem j, 85, 440–447.

Borggaard, O.K., & Gimsing, A.L. 2008. Fate of glyphosate in soil and the possibility of leaching to ground and surface waters: a review. pest manag sci, 64, 441–456.

Borgonovo, E., Lu, X., Plischke, E., Rakovec, O., & Hill, M. 2017. Making the most out of a hydrological model data set: Sensitivity analyses to open the model black-box. Water Resour Res, 53, 7933–7950.

Bos, Jacqueline, Jolley, Andelys, Johnstone, Patricia, Research, Grains, (Australia), Develop- ment Corporation, of Agriculture, Victoria Department, & Agmedia. 1995. Herbicide persistence in Australian winter cereal cropping systems : the triazines, the sulfonylureas, the dinitroanilines. AGMEDIA.

Boundy-Mills, K. L., Souza, M. L. D. E., & Mandelbaum, R. T. 1997. The atzB gene of Pseudomonas sp . strain ADP encodes the second enzyme of a novel atrazine degradation pathway. Appl Environ Microbiol, 63(3), 916–923.

Bowdena, P., Edwards, J., Ferguson, N., McNee, T., Manning, B., Roberts, K., Schipp, A., Schulze, K., & Wilkins, J. 2008. Wheat growth & development. NSW Department of Primary Industries.

141 Bowmer, K.H. 1991. Atrazine Persistence and Toxicity in two Irrigated Soils of Australia. Aust. J. Soil Res, 29(2), 339–350.

Brooks, RH, & Corey, AT. 1962. Hydraulic properties of porous media, Hydrology Papers, A14407-599281, Colorado State University, Fort Collins, CO, USA.

Bureau of Meteorology. 2016. http://www.bom.gov.au/climate/data/index.shtml. accessed 26.05.2016.

Button, D. K. 1983. Diﬀerences between the kinetics of nutrient uptake by micro-organisms, growth and enzyme kinetics. Trends Biochem. Sci., 121–124.

Button, D K, & Garver, J C. 1966. Continuous culture of Torulopsis utilis: a kinetic study of oxygen limited growth. J. Gen. Microbiol., 45, 195–204.

Campolongo, Francesca, Cariboni, Jessica, & Saltelli, Andrea. 2007. An eﬀective screening design for sensitivity analysis of large models. Environ Modell Softw, 22, 1509–1518.

Campolongo, Francesca, Saltelli, Andrea, & Cariboni, Jessica. 2011. From screening to quantitative sensitivity analysis. A uniﬁed approach. Comput Phys Commun, 182, 978–988.

Cerniglia, C. E., & Perry, J. J. 1975. Metabolism of n-propylamine, isopropylamine, and 1, 3-propane diamine by Mycobacterium convolutum. J. Bacteriol., 124(1), 285–289.

Cheyns, K., Mertens, J., Diels, J., Smolders, E., & Springael, D. 2010. Monod kinetics rather than a ﬁrst-order degradation model explains atrazine fate in soil mini-columns: Implications for pesticide fate modelling. Environ Pollut, 158, 1405 – 1411.

Chiari, G., Genovesi, R., Raimondi, S., Sarno, G., & Tarocco, P. 2016. database accessed on 01.02.2017 at "http: // cloud. consorziocer. it/ FaldaNET/ retefalda/ index ". CG and GR from Consorzio di Bonifca di secondo grado per il Canale Emiliano-Romagnolo; RS from Coop. I.ter; SG from Regione Emilia-Romagna. Servizio ricerca, innovazione e pro- mozione del sistema agroalimentare; TP from Regione Emilia-Romagna. Servizio Geologico, Sismico e dei Suoli.

Chu, D., & Barnes, D.J. 2016. The lag-phase during diauxic growth is a trade-oﬀ between fast adaptation and high growth rate. Scientiﬁc Reports, 6(25191).

Clay, S. A., & Koskinen, W. C. 1990. Characterization of Alachlor and Atrazine Desorption from Soils. Weed Sci., 38(1), 74–80.

Coleman, J. O. D., Frova, C., Schröder, P., & Tissut, M. 2002. Exploiting plant metabolism for the phytoremediation of persistent herbicides. environ sci pollut r, 9, 18–28.

Cook, A.M., & Huetter, R. 1984. Deethylsimazine: bacterial dechlorination, deamination, and complete degradation. J. Agric. Food Chem., 32(3), 581–585.

142 Copley, Shelley D. 2009. Evolution of Eﬃcient Pathways for Degradation of Anthropogenic Chemicals. dev. biol., 5, 559–566.

Cosby, B.J., Hornberger, G.M., Clapp, R.B., & Ginn, T.R. 1984. A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils. Water Resources Research, 20, 682–690.

Coupe, R.H., & Capel, P.D. 2016. Trends in pesticide use on soybean, corn and cotton since the introduction of major genetically modiﬁed crops in the United States. pest manag sci, 72, 1013–1022.

Cukier, R.I., Fortuin, C.M., Shuler, K.E., Petschek, A.G., & Schaibly, J.H. 1973. Study of the sensitivity of coupled reaction systems to uncertainties in rate coeﬃcients. J Chem Phys, 59, 3873–3878.

Dai, Heng, Ye, Ming, Walker, Anthony P., & Chen, Xingyuan. 2018. A new process sensitivity index to identify important system processes under process model and parametric uncertainty. Water Resour Res, 53, 3476–3490.

Därre, P., & Andreesen, J. R. 1982a. Selenium-dependent growth and glycine fermentation by Clostridium purinolyticum. J. Gen. Microbiol., 128, 1457–1466.

Därre, P., & Andreesen, J. R. 1982b. Selenium-dependent growth and glycine fermentation by Clostridium purinolyticum. J. Gen. Microbiol., 128, 1457–1466. de Azevedo, W., & Susana, I. 2001. Isopropylamine degradation in strain Pseudomonas sp. KIE171. de Boer, W., & Kowalchuk, G.A. 2001. Nitriﬁcation in acid soils: micro-organisms and mechanisms. Soil Biol Biochem, 33(7-8), 853–866.

De Nobili, M., Contin, M., Mondini, C., & Brookes, P.C. 2001. Soil microbial biomass is triggered into activity by trace amounts of substrate. Soil Biology and Biochemistry, 33, 1163 – 1170.

De Souza, M. L., Seﬀernick, J., Martinez, B., Sadowsky, M. J., & Wackett, L. P. 1998. The atrazine catabolism genes atzABC are widespread and highly conserved. J. Bacteriol., 1951– 1954.

Debasmita, N., & Rajasimman, M. 2013. Optimization and kinetics studies on biodegradation of atrazine using mixed microorganisms. Alexandria Eng. J., 499–505.

Dechesne, A., Badawi, N., Aamand, J., & Smets, B.F. 2014. Fine scale spatial variability of microbial pesticide degradation in soil: scales, controlling factors, and implications. Frontiers in Microbiology, 5, 1–14.

143 Decho, A. W., & Gutierrez, Tony. 2017. Microbial Extracellular Polymeric Substances (EPSs) in Ocean Systems. Frontiers in Microbiology, 8, 922.

Della Chiesa, S., la Cecilia, D., Genova, G., Balotti, A., Thalheimer, M., Tappeiner, U., & Niedrist, G. 2019. Farmers as data sources: Cooperative framework for mapping soil properties for permanent crops in South Tyrol (Northern Italy). GEODERMA, 342, 93–105.

Dell’Oca, Aronne, Riva, Monica, & Guadagnini, Alberto. 2017. Moment-based metrics for global sensitivity analysis of hydrological systems. Hydrol Earth Syst Sc, 21, 6219–6234.

Dick, R. E., & Quinn, J. P. 1995. Glyphosate-degrading isolates from environmental samples: occurrence and pathways of degradation. appl microbiol biot, 43, 545–550.

Dijkhuizen, L., van der Werf, B., & Harder, W. 1980. Metabolic Regulation in Pseudomonas oxalaticus OX1. Diauxic Growth on Mixtures of Oxalate and Formate or Acetate. Archives of Microbiology, 124(2-3), 261–268.

Ding, J., Zhao, Q.L., Wei, L.L., Chen, Y., & Shu, X. 2010. Ammonium nitrogen removal from wastewater with a three-dimensional electrochemical oxidation system. Water Sci Technol., 68, 552–559.

Doherty, J., Muﬀels, C., Rumbaugh, J., & Tonkin, M. 2016. PEST Model-Indipendent Pa- rameter Estimation & Uncertainity Analysis. http://www.pesthomepage.org/PEST.php. accessed 14.04.2016.

Don, A., & Schulze, E.D. 2014. Eﬀect of Atrazine, Glyphosate and Quinclorac on Biochemical Parameters, Lipid Peroxidation and Survival in Bullfrog Tadpoles (Lithobates catesbeianus). Archives of Environmental Contamination and Toxicology, 66, 415–429.

Dornelles, M.F., & Oliveira, G.T. 2014. Eﬀect of Atrazine, Glyphosate and Quinclorac on Biochemical Parameters, Lipid Peroxidation and Survival in Bullfrog Tadpoles (Lithobates catesbeianus). Archives of Environmental Contamination and Toxicology, 415–429.

Duke, S.O. 1988. Glyphosate. In: Herbicides-Chemistry: Degradation and Mode of Action, Vol. 3. (P.C. Kearney and D.D. Kaufman, Eds.) Marcel Dekker, New York, Kearney. Taylor & Francis.

Duke, S.O. 2011. Glyphosate degradation in glyphosate-resistant and -susceptible crops and weeds. j agr food chem, 59, 5835–5841.

Duke, S.O., & Powles, S.B. 2008. Glyphosate: a once-in-a-century herbicide. pest manag sci, 64, 319–325.

EAWAG. 2019. http://eawag-bbd.ethz.ch/predict/. accessed 26.11.2016.

144 Eberbach, P. 1998. Applying non-steady-state compartmental analysis to investigate the simultaneous degradation of soluble and sorbed glyphosate (N-(phosphonomethyl)glycine) in four soils. Pestic sci, 52, 229–240.

EC Directive 2006/118/EC. 2006. European Parliament and Council, 2006. Directive 2006/118/EC of 12 December 2006 on the protection of groundwater against pollution and deterioration. OJ L. 372, 19–31.

EC Directive 2008/50/EC. 2008. European Parliament and Council, 2008. Directive 2008/50/EC of 21 May 2008 on ambient air quality and cleaner air for Europe. OJ L 152 11.6.2008, p. 1.

EC Regulation 1107/2009. 2009. European Parliament and Council, 2009. Regulation EC 1107/2009 of 21 October 2009 concerning the placing of plant protection products on the market and repealing Council Directives 79/117/EEC and 91/414/EEC.

EC Regulation 396/2005. 2005. European Parliament and Council, 2005. Regulation EC 396/2005 of 23 February 2005 on maximum residue levels of pesticides in or on food and feed of plant and animal origin and amending Council Directive 91/414/EEC. OJ L 70, 16.3.2005, p. 1.

EFSA. 2015. Conclusion On The Peer Review Of The Pesticide Risk Assessment Of The Active Substance Glyphosate. EFSA J., 13, 1–107.

EFSA, European Food Safety Authority, Arena, Maria, Auteri, Domenica, Barmaz, Stefania, Bellisai, Giulia, Brancato, Alba, Brocca, Daniela, Bura, Laszlo, Byers, Harry, Chiusolo, Ari- anna, Court Marques, Daniele, Crivellente, Federica, De Lentdecker, Chloe, Egsmose, Mark, Erdos, Zoltan, Fait, Gabriella, Ferreira, Lucien, Goumenou, Marina, Greco, Luna, Ippolito, Alessio, Istace, Frederique, Jarrah, Samira, Kardassi, Dimitra, Leuschner, Renata, Lythgo, Christopher, Magrans, Jose Oriol, Medina, Paula, Miron, Ileana, Molnar, Tunde, Nougadere, Alexandre, Padovani, Laura, Parra Morte, Juan Manuel, Pedersen, Ragnor, Reich, Hermine, Sacchi, Angela, Santos, Miguel, Seraﬁmova, Rositsa, Sharp, Rachel, Stanek, Alois, Streissl, Franz, Sturma, Juergen, Szentes, Csaba, Tarazona, Jose, Terron, Andrea, Theobald, Anne, Vagenende, Benedicte, Verani, Alessia, & Villamar-Bouza, Laura. 2018. Peer review of the pesticide risk assessment of the active substance thiophanate-methyl. EFSA Journal, 16, 5133.

Eguchi, S., & Hasegawa, S. 2008. Determination and characterization of preferential water ﬂow in unsaturated subsoil of andisol. Soil Sci. Soc. Am. J., 72, 320–330.

Eisler, R. 1989. Atrazine hazards to ﬁsh, wildlife, and invertebrates: a synoptic review. Contam. Hazard Rev., 2–34.

145 Ermakova, Inna T., Shushkova, Tatyana V., Sviridov, Alexey V., Zelenkova, Nina F., Vi- nokurova, Natalya G., Baskunov, Boris P., & Leontievsky, Alexey A. 2017. Organophospho- nates utilization by soil strains of Ochrobactrum anthropi and Achromobacter sp. Archives of Microbiology, 199, 665–675.

Fan, Wuoiang Q., Yanase, Toshihiko, Morinaga, Hidetaka, Gondo, Shigeki, Okabe, Tai- jiro, Nomura, Masatoshi, Komatsu, Tomoko, Morohashi, Ken Ichirou, Hayes, Tyrone B., Takayanagi, Ryoichi, & Nawata, Hajime. 2007. Atrazine-induced aromatase expression is SF-1 dependent: Implications for endocrine disruption in wildlife and reproductive cancers in humans. Environ. Health Perspect., 720–727.

FAO. 2003. International Code of Conduct on the Distribution and Use of Pesticides. Food and Agriculture Organization of the United Nations.

FAO. 2013. http://faostat3.fao.org/download/R/RP/E. accessed 26.05.2016.

FAO, IIASA, ISRIC, ISSCAS, & JRC. 2012. Harmonized World Soil Database (version 1.2). FAO, Rome, Italy and IIASA, Laxenburg, Austria.

Farrell, M., Hill, P. W., Wanniarachchi, S. D., Farrar, J., Bardgett, R. D., & Jones, D. L. 2011. Rapid peptide metabolism: A major component of soil nitrogen cycling? Global Biogeo- chemical Cycles, 25(3).

Ferretti, Federico, Saltelli, Andrea, & Tarantola, Stefano. 2016. Trends in sensitivity analysis practice in the last decade. Sci Tot Env, 568, 666–670.

Fishel, F., Ferrell, J., MacDonald, G., & Sellers, B. 2015. Herbicides: How Toxic Are They?, PI-133, University of Florida, The Institute of Food and Agricultural Sciences, pp 2.

Flemming, H.C. 1999. Microbial extracellular polymeric substances : characterization, structure and function. Springer.

FOCUS. 2000. FOCUS groundwater scenarios in the EU review of active substances. Report of the FOCUS Groundwater ScenarioWorkgroup, EC Document Reference Sanco/321/ 2000 rev.2, pp. 202.

Forlani, G., Mangiagalli, A., Nielsen, E., & Suardi, C.M. 1999. Degradation of the phosphonate herbicide glyphosate in soil: evidence for a possible involvement of unculturable microorganisms. soil biol biochem, 31, 991 – 997.

Fragoeiro, S., & Magan, N. 2005. Enzymatic activity, osmotic stress and degradation of pesticide mixtures in soil extract liquid broth inoculated with Phanerochaete chrysosporium and Trametes versicolor. Environmental Microbiology, 7, 348–355.

Friebele, E, Shimoyama, A., & Ponnamperuma, C. 1980. Adsorption of protein and non-protein amino acids on a clay mineral: a possible role of selection in chemical evolution. J Mol Evol., 16(3), 269–278.

146 Fukumoto, Y., Suzuki, K., Osada, T., Kuroda, K., Hanajima, D., Yasuda, T., & Haga, K. 2006. Reduction of nitrous oxide emission from pig manure composting by addition of nitrite- oxidizing bacteria. Environ Sci Technol, 40(21), 6787–6791.

Gastrin, B., Kallings, L. O., & Marcetic, A. 1968. The survival time for diﬀerent bacteria in various transport media. Acta Pathologica Et Microbiologica Scandinavica, 74(3), 371–380.

Geller, A. 1979. Sorption and Desorption of Atrazine by Three Bacterial Species Isolated from Aquatic Systems. Arch. Environm. Contam. Toxicol., 8, 713–720.

Gerke, HH, & Van Genuchten, MT. 1993. A dual-porosity model for simulating the preferential movement of water and solutes in structured porous media. Water Resources Research, 29, 305–319.

Gerritse, J., Schut, F., & Gotishal, J.C. 1992. Modelling of mixed chemostat cultures of an aerobic bacterium, Comamonas testosteroni, and an anaerobic bacterium, Veillonella alcalescens: comparison with experimental data. Appl Environ Microbiol, 58, 1466–1476.

Ghafoor, A., N.J., Jarvis, T., Thierfelder, & J., Stenström. 2011. Measurements and modeling of pesticide persistence in soil at the catchment scale. Sci Tot Env, 409, 1900 – 1908.

Giardi, M. T., Giardina, M. C., & Filacchioni, G. 1985. Chemical and biological degradation of primary metabolites of atrazine by a Nocardia strain. Agric. Biol. Chem., 1551–1558.

Goolsby, D., Thurman, E. M., Pomes, M., Meyer, M., & Battaglin, W. 1997. Herbicides and Their Metabolites in Rainfall: Origin, Transport, and Deposition Patterns across the Mid- western and Northeastern United States, 1990-1991. Environ. Sci. Technol., 31, 1325–1333.

Gottesbüren, B. 1991. Doctoral thesis. Konzeption, Entwicklung und Validierung des wis- senbasierten Herbizid-Beratungssystems HERBASYS.

Grain, Australian. 2008. Irrigation scheduling of wheat under centre pivots.

Grandcoin, A., Piel, S., & Baurès, E. 2017. AminoMethylPhosphonic acid (AMPA) in natural waters: Its sources, behavior and environmental fate. water res, 117, 187 – 197.

Graymore, M., Stagnitti, F., & Allinson, G. 2001. Impacts of atrazine in aquatic ecosystems. Environ. Int., 483–495.

Green, R.E., & Karickhoﬀ, S.W. 1990. sorption estimates for modeling. H.H. Cheng, ed. Pesticides in the Soil Environment: Processes, Impacts, and Modeling. Madison, WI: Soil Science Society of America.

Grimes, D.J., & Morrison, S.M. 1975. Bacterial bioconcentration of chlorinated hydrocarbon insecticides from aqueous systems. Microb Ecol, 2, 43–59.

147 Guijarro, K.H., Aparicio, V., De Gerónimo, E., Castellote, M., Figuerola, E.L., Costa, J.L., & Erijman, L. 2018. Soil microbial communities and glyphosate decay in soils with diﬀerent herbicide application history. Sci Tot Env, 634, 974 – 982.

Gulde, R, Anliker, S., Kohler, H.E., & Fenner, K. 2018. Ion Trapping of Amines in Protozoa: A Novel Removal Mechanism for Micropollutants in Activated Sludge. Environ Sci Technol, 52, 52–60.

Hansch, C., Leo, A., & Hoekman, D. 1995. Exploring QSAR: Hydrophobic, electronic, and steric constants. American Chemical Society.

Hao, O.J., Richard, M.G., Jenkins, D., & Blanch, H.W. 1983. The Half-Saturation Coeﬃcient for Dissolved Oxygen: A Dynamic Method for Its Determination and Its Eﬀect on Dud Species Competition. Biotechnol. Bioeng, 25(2), 403–416.

Hayes, Tyrone B, Collins, Atif, Lee, Melissa, Mendoza, Magdelena, Noriega, Nigel, Stuart, A Ali, & Vonk, Aaron. 2002. Hermaphroditic, demasculinized frogs after exposure to the herbicide atrazine at low ecologically relevant doses. Proc. Natl. Acad. Sci. U. S. A., 5476– 5480.

Heap, I. 2016. Internation survey of herbicides resistant weedsdatabase, accessed on 22.01.2018 at "http: // www. weedscience. org/ Summary/ MOA. aspx? MOAID= 12 ".

Helbling, D. E., Ackermann, M., Fenner, K., Kohler, H. E., & Johnson, D. R. 2012. The activity level of a microbial community function can be predicted from its metatranscriptome. isme j, 902–904.

Hippe, H., Caspari, D., Fiebig, K., & Gottschalk, G. 1979. Utilization of trimethylamine and other N-methyl compounds for growth and methane formation by Methanosarcina barkeri. Proc. Natl. Acad. Sci. U. S. A., 76, 494–498.

Hope, M. 2003. Lachlan Catchment Irrigation Proﬁle.

Hormann, K., & Andreesen, J. R. 1989. Reductive cleavage of sarcosine and betaine by Eu- bacterium acidaminophilum via enzyme systems diﬀerent from glycine reductase. Arch. Microbiol., 153, 50–59.

Hove-Jensen, B., Zechel, D. L., & Jochimsen, B. 2014. Utilization of glyphosate as phosphate source: biochemistry and genetics of bacterial carbon-phosphorus lyase. Microbiol. Mol. Biol. Rev., 78, 176–97.

Hoyle, F. 2015. https://www.agric.wa.gov.au/climate-change/ what-soil-organic-carbon. accessed 26.05.2016.

Huang, P. M., & Schnitzer, M. (eds). 1986. Inﬂuence of soil mineral colloids on metabolic processes, growth, adhesion, and ecology of microbes and viruses. In Interactions of Soil Minerals with Natural Organics and Microbes. Soil Science Society of America.

148 IPSC. 1990. IPSC International Programme on Chemical Safety: Atrazine Health and Safety Guide. Environmental Health Criteria, 449–454.

ISPRA. 2016. Rapporto nazionale pesticidi nelle acque. Dati 2013-2014. Italian Journal of Geosciences, 175, 1–82.

Iyer, B., Rajput, M.S., Jog, R., Joshi, E., Bharwad, K., & Rajkumar, S. 2016. Organic acid mediated repression of sugar utilization in rhizobia. Microbiological Research, 192(Supplement C), 211 – 220.

Jacob, G. S., Garbow, J. R., Hallas, L. E., Kimack, N. M., Kishore, G. M., & Schaefer, J. 1988. Metabolism of glyphosate in Pseudomonas sp. strain LBr. Appl. Environ. Microbiol., 54, 2953–2958.

James, T.K., Rahman, A., Trolove, M.R., & Parker, M.D. 2010. Enhanced degradation of atrazine in soils with a history of repeated use . 24–27.

Johnson, C D, & Truex, M J. 2006. RT3D Reaction Modules for Natural and Enhanced Atten- uation of Chloroethanes, Chloroethenes, Chloromethanes, and Daughter Products.

Johnson, C.W., Abraham, P.E., Linger, J.G., Khanna, P., Hettich, R.L., & Beckham, G.T. 2017. Eliminating a global regulator of carbon catabolite repression enhances the conversion of aromatic lignin monomers to muconate in Pseudomonas putida KT2440. Metabolic Engineering Communications, 5(Supplement C), 19 – 25.

Katagi, Toshiyuki. 2004. Photodegradation of Pesticides on Plant and Soil Surfaces. Springer New York. Pages 1–78.

Katz, Ilan, Green, Michal, Ruskol, Yuri, & Dosoretz, Carlos G. 2000. Characterization of atrazine degradation and nitrate reduction by Pseudomonas sp. strain ADP. Adv. Environ. Res., 211–218.

Kawai, S., Uno, B., & Tomita, M. 1991. Determination of glyphosate and its major metabolite aminomethylphosphonic acid by high-performance liquid chromatography after derivatiza- tion with p-toluenesulphonyl chloride. Journal of Chromatography A, 540, 411 – 415.

Keating, B.A., Carberry, P.S, Hammer, G.L, Probert, M.E, Robertson, M.J, Holzworth, D, Huth, N.I, Hargreaves, J.N.G, Meinke, H, Hochman, Z, McLean, G, Verburg, K, Snow, V, Dimes, J.P, Silburn, M, Wang, E, Brown, S, Bristow, K.L, Asseng, S, Chapman, S, McCown, R.L, Freebairn, D.M, & Smith, C.J. 2003. An overview of APSIM, a model designed for farming systems simulation. European Journal of Agronomy, 18, 267 – 288.

Kersante, A., Martin-Laurent, F., Soulas, G., & Binet, F. 2006. Interactions of earthworms with Atrazine-degrading bacteria in an agricultural soil. FEMS Microbiology Ecology, 57, 192–205.

149 Kertesz, M.A., Cook, A.M., & Leisinger, T. 1994. Microbial metabolism of sulfur and phosphorus-containing xenobiotics. FEMS Microbiology Reviews, 15.

Khan. S.U. 1978. Chapter 4 The Interaction of Organic Matter with Pesticides. doi: 10.1016/S0166-2481(08)70019-9.

Kindred, J. Scott, & Celia, Michael A. 1989. Contaminant transport and biodegradation: 2. Conceptual model and test simulations. Water Resour. Res., 25, 1149–1159.

Kolic,´ Nikolina Udikovic,´ Hršak, Dubravka, Begonja Kolar, Ana, Petric,´ Ines, Stipicevic,ˇ Sanja, Soulas, Guy, & Martin-Laurent, Fabrice. 2007. Combined metabolic activity within an atrazine-mineralizing community enriched from agrochemical factory soil. Int. Biodeterior. Biodegrad., 299–307.

Kookana, R. S., & Baskaran, S. N. R. 1998. Pesticide fate and behaviour in Australian soils in relation to contamination and management of soil and water: a review. Aust. J. Soil Res., 36(5), 715–764.

Kremer, R.J., & Means, N.E. 2009. Glyphosate and glyphosate-resistant crop interactions with rhizosphere microorganisms. European Journal of Agronomy, 31, 153 – 161.

Kremling, A., Geiselmann, J., Ropers, D., & de Jong, H. 2015. Understanding carbon catabolite repression in Escherichia coli using quantitative models. Trends in Microbiology, 23(2), 99 – 109.

Krutz, L Jason, Shaner, Dale L, Accinelli, Cesare, Zablotowicz, Robert M, & Henry, W Brien. 2008. Atrazine dissipation in s-triazine-adapted and nonadapted soil from Colorado and Mississippi: implications of enhanced degradation on atrazine fate and transport parameters. J. Environ. Qual., 848–857.

Krutz, L. Jason, Shaner, Dale L., & Zablotowicz, Robert M. 2010a. Enhanced Degradation and Soil Depth Eﬀects on the Fate of Atrazine and Major Metabolites in Colorado and Mississippi Soils. J. Environ. Qual., 1369–1377.

Krutz, L.J., Shaner, D.L., Weaver, M.A., Webb, R.M.T., Zablotowicz, R.M., Reddy, K.N., Huang, Y., & Thomson, S.J. 2010b. Agronomic and environmental implications of enhanced s-triazine degradation. Pest Manag Sci., 66(5), 461–481.

Krzysko-Łupicka,´ T., & Orlik, A. 1997. The use of glyphosate as the sole source of phosphorus or carbon for the selection of soil-borne fungal strains capable to degrade this herbicide. Chemosphere, 34, 2601 – 2605.

Kumar, Anup, & Singh, Neera. 2016. Atrazine and its metabolites degradation in mineral salts medium and soil using an enrichment culture. Environ. Monit. Assess., 1–12.

150 Kurenbach, B., Marjoshi, D., Amábile-Cuevas, C.F., Ferguson, G.C., Godsoe, W., Gibson, P., & Heinemann, J.A. 2015. Sublethal Exposure to Commercial Formulations of the Herbicides Dicamba, 2,4-Dichlorophenoxyacetic Acid, and Glyphosate Cause Changes in Antibiotic Susceptibility in Escherichia coli and Salmonella enterica serovar Typhimurium. mBio, 6. la Cecilia, D., & Maggi, F. 2016. Kinetics of Atrazine, Deisopropylatrazine, and Deethyla- trazine soil biodecomposers. J. Environ. Manage., 183(3), 673–686. la Cecilia, D., & Maggi, F. 2018. Analysis of glyphosate degradation in a soil microcosm. Environmental Pollution, 233, 201 – 207. la Cecilia, D., & Maggi, F. Under Review. Influential sources of uncertainty in glyphosate biochemical degradation in soil. Mathematics and Computers in Simulation. la Cecilia, D., Tang, F.H., Coleman, N.V., Conoley, C., Vervoort, R.W., & Maggi, F. 2018a. Glyphosate dispersion, degradation, and aquifer contamination in vineyards and wheat fields in the Po Valley, Italy. Water Research, 146, 37–54. la Cecilia, D, Porta, G., Riva, M., Vervoort, RW., Coleman, NV, Tang, FH, & Maggi, Fed- erico. 2018b. Propagation of ecohydrological uncertainty in a complex biogeochemical network of Glyphosate dispersion and degradation. Computational Methods in Water Resources (CMWR) XXII. Bridging gaps between data, models, and predictions. 154. la Cecilia, Daniele, & Maggi, Federico. 2017a. In-situ atrazine biodegradation dynamics in wheat (Triticum) crops under variable hydrologic regime. J. Contam Hydrol, 203, 104–121. la Cecilia, Daniele, & Maggi, Federico. 2017b. Stochastic sensitivity analysis of glyphosate biochemical degradation. In Syme, G., Hatton MacDonald, D., Fulton, B. and Piantadosi, J. (eds) MODSIM2017, 22nd International Congress on Modelling and Simulation. Modelling and Simulation Society of Australia and New Zealand, December 2017. 257 – 263. la Cecilia, Daniele, Riley, William J., & Maggi, Federico. 2018c. Biochemical modeling of microbial memory effects and catabolite repression on soil organic carbon compounds. Soil Biology and Biochemistry, 128, 1 – 12.

Laitinen, P., Sari, R., & Siimes, K. 2007. Glyphosate translocation from plants to soil – does this constitute a signiﬁcant proportion of residues in soil? Plant and Soil, 300(1), 51–60.

Lambert, G., & Kussell, E. 2014. Memory and Fitness Optimization of Bacteria under Fluctu- ating Environments. PLOS Genetics, 10(9), 1–10.

Langergraber, G., & Šim˚unek,J. 2005. Modeling Variably Saturated Water Flow and Multi- component Reactive Transport in Constructed Wetlands. vadose zone j, 4, 924.

Langmuir, I. 1918. The adsorption of gases on plane surfaces of glass, mica and platinum. J. Am. Chem. Soc., 40, 1361–1403.

151 Lapworth, D. J., & Gooddy, D. C. 2006. Source and persistence of pesticides in a semi-conﬁned chalk aquifer of southeast England. Environ. Pollut., 1031–1044.

Lefrancq, M., Jadas-Hécart, A., La Jeunesse, I., Landry, D., & Payraudeau, S. 2017. High frequency monitoring of pesticides in runoﬀ water to improve understanding of their transport and environmental impacts. sci tot env, 587-588, 75 – 86.

Lerch, R.N., Blanchard, P.E., & Thurman, E.M. 1998. Contribution of hydroxylated atrazine degradation products to the total atrazine load in midwestern streams. Environ. Sci. Technol., 32(1), 40–48.

Lerch, T.Z., Dignac, M.F., Nunan, N., Barriuso, E., & Mariotti, A. 2009. Ageing processes and soil microbial community eﬀects on the biodegradation of soil 13C-2,4-D nonextractable residues. Environmental Pollution, 157, 2985—-2993.

Levering, P. R., van Dijken, J. P., Veenhuis, M., & Harder, W. 1981. Arthrobacter P1, a fast growing versatile methylotroph with amine oxidase as a key enzyme in the metabolism of methylated amines. Arch. Microbiol., 129, 72–80.

Levering, P. R., Dijkhuizen, L., & Harder, W. 1984. Metabolic regulation in the facultative methylotroph Arthrobacter P1. Growth on primary amines as carbon and energy sources. Arch Microbiol, 139, 188–195.

Li, H., Joshi, S. R., & Jaisi, D. P. 2015. Degradation and isotope source tracking of glyphosate and aminomethylphosphonic acid. J. Agric. Food Chem., 64, 529–538.

Li, L., Maher, K., Navarre-Sitchler, A., Druhan, J., Meile, C., Lawrence, C., Moore, J., Perdrial, J., Sullivan, P., Thompson, A., Jin, L., Bolton, E.W., Brantley, S.L., Dietrich, W.E., Mayer, K.U., Steefel, C.I., Valocchi, A., Zachara, J., Kocar, B., McIntosh, J., Tutolo, B.M., Kumar, M., Sonnenthal, E., Bao, C., & Beisman, J. 2017. Expanding the role of reactive transport models in critical zone processes. Earth-Sci. Rev, 165, 280–301.

Li, Z., & Bowman, R.S. 2001. Retention of inorganic oxyanions by organo-kaolinite. Water Res, 35, 3771–3776.

Lichtner, P. C. 1985. Continuum Model for Simultaneous Chemical-Reactions and Mass- Trans- port in Hydrothermal Systems. Geochimica et Cosmochimica Acta, 49, 779–800.

Magasanik, B. 1961. Catabolite Repression. Cold Spring Harb Symp Quant Biol, 26, 249–256.

Maggi, F. 2015. BRTSim version 1; A general-purpose multiphase and multicomponent computational solver for biogeochemical reaction-advection-dispersion processes in porous and non-porous media, First edition, Research Report R954, The University of Sydney, Australia, pp 29.

Maggi, F., & Porporato, A. 2007. Coupled moisture and microbial dynamics in unsaturated soils. Water Resources Research, 43.

152 Maggi, F., Gu, C., Riley, W., Hornberger, G., Venterea, R., Xu, T., Spycher, N., Steefel, C., Miller, N., & Oldenburg, C. 2008. A mechanistic treatment of the dominant soil nitrogen cycling processes: Model development, testing, and application. J Geophys Res, 1–13.

Maillard, E., Lange, J., Schreiber, S., Dollinger, J., Herbstritt, B., Millet, M., & Imfeld, G. 2016. Dissipation of hydrological tracers and the herbicide S-metolachlor in batch and continuous- ﬂow wetlands. Chemosphere, 144, 2489 – 2496.

Mamy, L., Barriuso, E., & Gabrielle, B. 2005. Environmental fate of herbicides triﬂuralin, metazachlor, metamitron and sulcotrione compared with that of glyphosate, a substitute broad spectrum herbicide for diﬀerent glyphosate-resistant crops. pest manag sci, 61, 905–916.

Mandelbaum, R. T., Allan, Deborah l., & Wackett, Lawrence. P. 1995. Isolation and Charac- terization of a Pseudomonas sp . That Mineralizes the s -Triazine Herbicide Atrazine. Appl. Environ. Microbiol., 1451–1457.

Martin, C.D. 1977. Washoﬀ of herbicides applied to corn residue, Iowa State University, Digital Repository, Retrospective Theses and Dissertations, 16294, https://lib.dr.iastate.edu/rtd/16294.

Martinez, B., Tomkins, J., Wackett, L. P., Wing, R., & Sadowsky, M. J. 2001. Complete Nu- cleotide Sequence and Organization of the Atrazine Catabolic Plasmid pADP-1 from Pseu- domonas sp. Strain ADP. J. Bacteriol., 183(19), 5684–5697.

Mayer, K. U., Benner, S. G., Frind, E. O., Thornton, E. O., & Lerner, D. N. 2001. Reactive transport modeling of processes controlling the distribution and natural attenuation of pheno- lic compounds in a deep sandstone aquifer. Journal of Contaminant Hydrology, 53, 341–368.

Mcauliﬀe, K. S., Hallas, L. E., & Kulpa, C. F. 1990. Glyphosate degradation by Agrobacterium radiobacter isolated from activated sludge. J. Ind. Microbiol., 6, 219–221.

McCarl, Bruce A., & Apland, Jeﬀrey. 1986. Validation of linear programming models. Southern Journal of Agricultural Economics, 18, 155–164.

McGrath, J. W., Chin, J. P., & Quinn, J. P. 2013. Organophosphonates revealed: new insights into the microbial metabolism of ancient molecules. nat rev microbiol, 11, 412–419.

Meylan, W., Howard, P. H., & Boethling, R. S. 1992. Molecular topology/fragment contribution method for predicting soil sorption coeﬃcients. Environ Sci Technol, 1560–1567.

Mobbs, T.L., Peters, R.T., Davenport, J., Evans, M., & J., Wu. 2012. Eﬀects of four soil surfactants on four soil-water properties in sand and silt loam. JOURNAL OF SOIL AND WATER CONSERVATION, 67, 273–281.

Monod, J. 1949. The Growth of Bacterial Cultures. Annual Review of Microbiology, 3, 371– 394.

153 Moore, J. K., Braymer, H. D., & Larson, A. D. 1983. Isolation of a Pseudomonas sp. which utilizes the phosphonate herbicide glyphosate. Appl. Environ. Microbiol., 46, 316–320.

Morita, Richard Y. 1988. Bioavailability of energy and its relationship to growth and starvation survival in nature. Can. J. Microbiol., 436–441.

Morvan, X., Mouvet, C., Baran, N., & Gutierrez, A. 2006. Pesticides in the groundwater of a spring draining a sandy aquifer: Temporal variability of concentrations and ﬂuxes. J. Contam. Hydrol., 176–190.

Muehleisen, R.T., & Bergerson, J. 2016. Bayesian Calibration - What, Why And How. Interna- tional High Performance Buildings Conference. 3025–3031.

Mukherjee, A., & Ghosh, S. 1987. Regulation of fructose uptake and catabolism by succinate in Azospirillum brasilense. Journal of Bacteriology, 169, 4361–4367.

Munira, S., Farenhorst, A., Flaten, D., & Grant, D. 2016. Phosphate fertilizer impacts on glyphosate sorption by soil. Chemosphere, 153, 471 – 477.

Muñoz-Carpena, Rafael, Fox, Garey A., & Sabbagh, George J. 2009. Parameter Importance and Uncertainty in Predicting Runoﬀ Pesticide Reduction with Filter Strips. J Environ Qual, 39, 630–641.

Newman, M.M., Hoilett, N., Lorenz, N., Dick, R.P., Liles, M.R., Ramsier, C., & Kloepper, J.V. 2016. Glyphosate eﬀects on soil rhizosphere-associated bacterial communities. sci tot env, 543, 155 – 160.

Nguyen, D.B., Rose, M.T., Rose, T.J., Morris, S.G., & van Zwieten, L. 2016. Impact of glyphosate on soil microbial biomass and respiration: A meta-analysis. soil biol biochem, 92, 50 – 57.

Nielsen, L.N., Roager, H.M., Casas, M.E., Frandsen, H.L., Gosewinkel, U., Bester, K., Licht, T.R., Hendriksen, N.B., & Bahl, M.I. 2018. Glyphosate has limited short-term eﬀects on commensal bacterial community composition in the gut environment due to suﬃcient aromatic amino acid levels. Environmental Pollution, 233, 364 – 376.

Obojska, A., Lejczak, B., & Kubrak, M. 1999. Degradation of phosphonates by streptomycete isolates. Appl. Microbiol. Biotechnol., 51, 872–876.

Ogram, A.V., Jessup, R.E., OU, L.T., & RAO, P.S.C. 1985. Eﬀects of Sorption on Biological Degradation Rates of (2,4-Dichlorophenoxy)acetic Acid in Soils. Appl. Environ. Microbiol., 49(3), 582–587.

Pagel, H., Ingwersen, J., Poll, C., Kandeler, E., & Streck, T. 2014. Micro-scale modeling of pesticide degradation coupled to carbon turnover in the detritusphere: model description and sensitivity analysis. Biogeochemistry, 117, 185–204.

154 Panikov, N.S. 2010. Microbial Ecology. In: Handbook of Environmental Engineering, Vol- ume 10: Environmental Biotechnology (L. K. Wang, V. Ivanov, J.-H. Tay, Y.-T. Hung, Eds.) Springer Science + BusinessMedia, LLC 2010. Humana press.

Paracelsus. 1965. Die dritte Defension wegen des Schreibens der neuen Rezepte, Septem De- fensiones 1538. Werke Bd. 2, Darmstadt 1965, p. 510.

Paris, P., Citro, L., Di Carlo, E., Maschio, G., Pace, E., Ursino, S., Carﬁ, F., Cerioli, N.L., Esposito, D., & Romoli, D. 2013. Rapporto nazionale pesticidi nelle acque dati 2009-2010, Edizione 2013, Rapporti 175/2013, ISPRA, Italia, pp 82. ISBN: 978-88-448-0595-1.

Paris, P., Pace, E., Presicce, D.P., Maschio, G., Ursino, S., Bisceglie, S., Cornetta, T., Citro, L., Paciﬁco, R., Giardina, S., Esposito, D., Romoli, D., Floridi, E., & Tornato, A. 2016. Rapporto nazionale pesticidi nelle acque dati 2013-2014, Edizione 2016, Rapporti 244/2016, ISPRA, Italia, pp 121. ISBN: 9788844807702.

Park, J.H., Feng, Y., Ji, P., Voice, T.C., & Boyd, S.A. 2003. Assessment of Bioavailability of Soil-Sorbed Atrazine. Appl. Environ. Microbiol., 69(6), 3288–3289.

Parkhurst, D. L. and Appelo, C. A. J. 1999. User’s guide to PHREEQC (Version 2)—A computer program for speciation, batch reaction, 1 dimensional transport, and inverse geochemical calculation. Water-Resources Investigations Report 99-4259. U.S. Geological Survey : Earth Science Information Center, Open-File Reports Section [distributor]. pp. 312.

Paudel, P, Negusse, A., & Jaisi, Deb P. 2015. Birnessite-catalyzed degradation of glyphosate: a mechanistic study aided by kinetics batch studies and NMR spectroscopy. Soil Sci. Soc. Am. J., 79, 815–825.

Pipke, R., & Amrhein, N. 1988a. Degradation of the phosphonate herbicide glyphosate by arthrobacter atrocyaneus ATCC 13752. Appl. Environ. Microbiol., 54, 1293–1296.

Pipke, R, & Amrhein, N. 1988b. Isolation and characterization of a mutant of Arthrobacter sp Strain GLP-1 which utilizes the herbicide glyphosate as its sole source of phosphorus and nitrogen. Appl. Environ. Microbiol., 54, 2868–2870.

Pipke, R, Amrhein, N, Jacob, G S, Schaefer, J, & Kishore, G M. 1987. Metabolism of glyphosate in an Arthrobacter sp. Glp-1. Eur. J. Biochem., 165, 267–273.

Porta, Giovanni, la Cecilia, Daniele, Guadagnini, Alberto, & Maggi, Federico. 2018. Impli- cations of uncertain biogeochemical parameters on a complex reaction network of atrazine biodegradation in soil. Advances in Water Resources, 121, 494–498.

Prommer, H., Barry, D., Chiang, W., & Zheng, C. 2001. PHT3D - A MODFLOW / MT3DMS- based reactive multi-component transport model. Ground Water, 477–483.

155 Purcell, M., Leroux, G. D., & Carpentier, R. 1990. Atrazine action on the donor side of photosystem II in triazine-resistant and -susceptible weed biotypes. Pesticide Biochemistry and Physiology, 37, 83–89.

Radosevich, M., Traina, S. J., Hao, Y. L., & Tuovinen, O. H. 1995. Degradation and mineralization of atrazine by a soil bacterial isolate. Appl. Environ. Microbiol., 297–302.

Raillard, S., Krebber, A., Chen, Y., Ness, J.E., Bermudez, E., Trinidad, R., Fullem, R., Davis, C., Welch, M., Seﬀernick, J., Wackett, L.P., Stemmer, W.P.C., & Minshull, J. 2001. Novel enzyme activities and functional plasticity revealed by recombining highly homologous enzymes. chem biol, 8, 891 – 898.

Rattray, D.J., Standley, J., Silburn, D. M., Freebairn, D.M., & Spann, K.P. 2008. Atrazine degradation and transport in runoﬀ on a Black Vertosol. Aust. J. Soil Res., 45(8), 598–606.

Razavi, S., & Gupta, H.V. 2015. What do we mean by sensitivity analysis? the need for comprehensive characterization of global sensitivity in earth and environmental systems models. Water Resour Res, 51, 3070–3092.

Reay, David S, Nedwell, David B, Priddle, Julian, & Ellis-Evans, J. C. 1999. Temperature De- pendence of Inorganic Nitrogen Uptake : Reduced Aﬃnity for Nitrate at Suboptimal Tem- peratures in Both Algae and Bacteria. Appl. Environ. Microbiol., 2577–2584.

Refsgaard, Jens C., van der Sluijs, Jeroen P., Lajer Højberg, Anker, & Vanrolleghem, Peter A. 2007. Uncertainty in the environmental modelling process – A framework and guidance. Environ Modell Softw, 22, 1543 – 1556.

Riley, W. J., Maggi, F., Kleber, M., Torn, M. S., Tang, J. Y., Dwivedi, D., & Guerry, N. 2014. Long residence times of rapidly decomposable soil organic matter: application of a multiphase, multi-component, and vertically resolved model (BAMS1) to soil carbon dynamics. Geoscientiﬁc Model Development, 7(4), 1335–1355.

Rittmann, B. E., & McCarty, P. L. 2001. Environmental Biotechnology: Principles and Appli- cation. McGraw-Hill Series in Water Resources and Environmental Engineering.

Roca, C., & Olsson, L. 2001. Dynamic responses of Pseudomonas ﬂuorescens DF57 to nitrogen or carbon source addition. Journal of Biotechnology, 86(1), 39 – 50.

Rose, Michael T., Cavagnaro, Timothy R., Scanlan, Craig A., Rose, Terry J., Vancov, Tony, Kimber, Stephen, Kennedy, Ivan R., Kookana, Rai S., & Van Zwieten, Lukas. 2016. Impact of Herbicides on Soil Biology and Function. Pages 133 – 220 of: Sparks, Donald L. (ed), Advances in Agronomy. Advances in Agronomy, vol. 136. Academic Press.

Rosenbaum, K.K., Miller, G.L., Kremer, R.J., & Bradley, K.W. 2014. Interactions between Glyphosate, Fusarium Infection of Common Waterhemp (Amaranthus Rudis), and Soil Mi- crobial Abundance and Diversity in Soil Collections from Missouri. Weed Science, 62, 71–82.

156 Roux-Micholleta, D., Czarnesa, S., Adamb, B., Berryc, D., Commeauxa, C., Guillaumauda, N., Le Rouxa, X., & Clays-Josseranda, A. 2008. Eﬀects of steam disinfestation on community structure, abundance and activity of heterotrophic, denitrifying and nitrifying bacteria in an organic farming soil. Soil Biol Biochem, 40(7), 1836–1845.

Salem, S., Moussa, M.S., & van Loosdrecht, M.C.M. 2006. Determination of the decay rate of nitrifying bacteria. Biotechnology and Bioengineering, 94(2), 252–262.

Saltelli, Andrea, & Tarantola, Stefano. 2002. On the Relative Importance of Input Factors in Mathematical Models: Safety Assessment for Nuclear Waste Disposal. J Am Stat Assoc, 97, 702–709.

Santos-Beneit, F. 2015. The Pho regulon: a huge regulatory network in bacteria. Frontiers in Microbiology, 6, 402–415.

Sasal, M.C., Demonte, L., Cislaghi, A., Gabioud, E. A., Oszust, J. D., Wilson, M. G., Michlig, N., Beldoménico, H. R., & Repetti, M.R. 2015. Glyphosate loss by runoﬀ and its relationship with phosphorus fertilization. j agr food chem, 63, 4444–4448.

Schimel, J., Becerra, C.N., & Blankinship, J. 2017. Estimating decay dynamics for enzyme activities in soils from diﬀerent ecosystems. Soil Biology and Biochemistry, 114, 5 – 11.

Schimel, J.P., & Weintraub, M.N. 2003. The implications of exoenzyme activity on microbial carbon and nitrogen limitation in soil: a theoretical model. Soil Biol. Biochem., 35, 549–563.

Schnürer, Y., Persson, P., Nilsson, M., Nordgren, A., & Giesler, R. 2006. Eﬀects of surface sorption on microbial degradation of glyphosate. env sci tec, 40, 4145–4150.

Schönbrunn, E., Eschenburg, S., Shuttleworth, W. A., Schloss, J. V., Amrhein, N., Evans, J. N. S., & Kabsch, W. 2001. Interaction of the herbicide glyphosate with its target enzyme 5-enolpyruvylshikimate 3-phosphate synthase in atomic detail. PNAS, 98.

Scott, M. 2008. NSW Department of Primary Industries: SMARTtrain Chemical Notes 3.

Seguin, G. 1986. "Terroirs" and pedology of wine growing. Experientia, 42(8), 861–873.

SGSS. 2016. database accessed on 01.02.2017 at "https: // applicazioni. regione. emilia-romagna. it/ cartografia_ sgss/ user/ viewer. jsp? service= pedologia& bookmark= 1% 22 ". Regione Emilia-Romagna. Servizio Geologico, Sismico e dei Suoli.

Shaner, D. L., Henry, W. B., Krutz, L. J., & Hanson, B. 2007. Rapid Assay for Detecting Enhanced Atrazine Degradation in Soil. Weed Sci., 55(5), 528–535.

Shao, Z. Q., Seﬀens, W., Mulbry, W., & Behki, R. M. 1995. Cloning and expression of the s- triazine hydrolase gene (trzA) from Rhodococcus corallinus and development of Rhodococ- cus recombinant strains capable of dealkylating and dechlorinating the herbicide atrazine. J. Bacteriol., 5748–5755.

157 Shapir, N, & Mandelbaum, R T. 1997. Atrazine degradation in subsurface soil by indigenous and introduced microorganisms. J. Agric. Food Chem., 4481–4486.

Shi, S., Hermana, D.J., Heb, Z., Pett-Ridged, J., Wub, L., Zhoub, J., & Firestonea, M.K. 2018. Plant roots alter microbial functional genes supporting root litter decomposition. Soil Biology and Biochemistry, 127, 90 – 99.

Shimabukuro, R.H. 1967. Atrazine Metabolism and Herbicidal Selectivity. Plant Physiol., 42(9), 1269–1276.

Shushkova, T.V., Vinokurova, N.G., Baskunov, B.P., Zelenkova, N.F., Sviridov, A.V., Er- makova, I.T., & Leontievsky, A.A. 2016. Glyphosate acetylation as a speciﬁc trait of Achro- mobacter sp. Kg 16 physiology. Applied Microbiology and Biotechnology, 100, 847–855.

Sidoli, P., Baran, N., & Angulo-Jaramillo, R. 2016. Glyphosate and AMPA adsorption in soils: laboratory experiments and pedotransfer rules. Environ Sci Pollut Res, 23, 5733–5742.

Silva, V., Montanarella, L., Jones, A., Fernández-Ugalde, O., Mol, H.G.J., Ritsema, C.J., & Geissen, V. 2017. Distribution of glyphosate and aminomethylphosphonic acid (AMPA) in agricultural topsoils of the European Union. sci tot env.

Silva, V., Mol, H.G.J., Zomer, P., Tienstra, M., Ritsema, C.J., & Geissena, V. 2019. Pesticide residues in European agricultural soils – A hidden reality unfolded. Science of the Total Environment, 653, 1532–1545.

Simonsen, L., Fomsgaard, I.S., Svensmark, B., & Spliid, N.H. 2008. Fate and availability of glyphosate and AMPA in agricultural soil. J Environ Sci Heal B, 43, 365–375.

Singh, H. 2006. Mycoremediation: Fungal Degradation of Pesticides (6). John Wiley and Sons, Ltd.

Smit, E., Leeflang, P., Gommans, S., Van den Broek, J., Van Mil, S., & Wernars, K. 2001. Di- versity and seasonal fluctuations of the dominant members of the bacterial soil community in a wheat field as determined by cultivation and molecular methods. Appl. Environ. Microbiol., 67, 2284–2291.

Smith, Daniel, & Crowley, David E. 2006. Contribution of ethylamine degrading bacteria to atrazine degradation in soils. FEMS Microbiol. Ecol., 271–277.

Smith, Daniel, Alvey, Sam, & Crowley, David E. 2005. Cooperative catabolic pathways within an atrazine-degrading enrichment culture isolated from soil. FEMS Microbiol. Ecol., 265– 273.

Sobol’, I.M. 1993. Sensitivity Estimates for Nonlinear Mathematical Models. Mathematical Modeling and Computational Experiments, 1, 407–414.

158 Solomon, Robinson David Jebakumar, Kumar, Amit, & Satheeja Santhi, Velayudhan. 2013. Atrazine biodegradation eﬃciency, metabolite detection, and trzD gene expression by enrichment bacterial cultures from agricultural soil. J. Zhejiang Univ. Sci. B, 1162–1172.

Sprankle, P., Meggitt, W.F., & Penner, D. 1975. Adsorption, mobility, and microbial degradation of glyphosate in the soil. weed sci, 23, 229–234.

Stapper, M. 2007. Crop monitoring and Zadoks growth stages for wheat. Grains Research and Development Corporation (see GRDC), Research Update.

Steefel, C. I., & Lasaga, A. C. 1994. A coupled model for transport of multiple chemical species and kinetic precipitation/dissolution reactions with application to reactive ﬂow in single phase hydrothermal systems. Am. J. Sci., 294, 529–592.

Steﬀan, R J, McClay, K, Vainberg, S, Condee, C W, & Zhang, D. 1997. Biodegradation of the Gasoline Oxygenates Methyl tert-Butyl Ether, Ethyl tert-Butyl Ether, and tert-Amyl Methyl Ether by Propane-Oxidizing Bacteria. Appl. Environ. Microbiol., 63(11), 4216–4222.

Steﬀens, K., Jarvis, N., Lewan, E., Lindström, B., Kreuger, J., Kjellström, E., & Moeys, J. 2015. Direct and indirect eﬀects of climate change on herbicide leaching — A regional scale assessment in Sweden. Science of The Total Environment, 514, 239 – 249.

Stock, J.B., & Zhang, S. 2012. The Biochemistry of Memory. Current Biology, 23(17), 741–745.

Stolper, D.A., Revsbech, N.P., & Canﬁeld, D.E. 2010. Aerobic growth at nanomolar oxygen concentrations. PNAS, 107, 18755–18760.

Struthers, J. K., Jayachandran, K., & Moorman, T. B. 1998. Biodegradation of atrazine by Agrobacterium radiobacter J14a and use of this strain in bioremediation of contaminated soil. Appl. Environ. Microbiol., 3368–3375.

Suárez, . 2005. PRZM-3, A Model for Predicting Pesticide and Nitrogen Fate in the Crop Root and Unsaturated Soil Zones: Users Manual for Release 3.12.2, EPA/600/R-05/111, Ecosys- tems Research Division National Exposure Research Laboratory Athens, GA 30605-2700, pp. 426.

Suter, Glenn W. II. 2007. Ecological Risk Assessment. 2 edn. Boca Raton, Fl. : CRC Press.

Sviridov, A. V., Shushkova, T. V., Zelenkova, N. F., Vinokurova, N. G., Morgunov, I. G., E., Inna T., & Leontievsky, A. A. 2012. Distribution of glyphosate and methylphosphonate catabolism systems in soil bacteria Ochrobactrum anthropi and Achromobacter sp. Appl. Microbiol. Biotechnol., 93, 787–796.

Sviridov, A V, Shushkova, T V, Ermakova, I T, Ivanova, E V, Epiktetov, D O, & Leontievsky, A A. 2015. Microbial degradation of glyphosate herbicides (review). Appl. Biochem. Micro- biol., 51, 188–195.

159 Talbot, H. W., Johnson, L. M., & Munnecke, D.M. 1984. Glyphosate utilization by Pseu- domonas sp. and Alcaligenes sp. isolated from environmental sources. Curr. Microbiol., 10, 255–259.

Tang, F. H. M., la Cecilia, D., Vervoort, R.W., Coleman, N., C., Conoley, & Maggi, F. 2017. Integrating biological degradation potential into ecological risk assessment. In Syme, G., Hatton MacDonald, D., Fulton, B. and Piantadosi, J. (eds) MODSIM2017, 22nd International Congress on Modelling and Simulation. Modelling and Simulation Society of Australia and New Zealand, December 2017. 915 – 921.

Tang, F.H.M., Jeﬀries, T.C., Vervoort, R.W., Conoley, C., Coleman, N.V., & Maggi, F. 2019. Microcosm experiments and kinetic modeling of glyphosate biodegradation in soils and sed- iments. Science of The Total Environment, 658, 105 – 115.

Tang, J. Y., & Riley, W. J. 2013. A total quasi-steady-state formulation of substrate uptake kinetics in complex networks and an example application to microbial litter decomposition. Biogeosciences, 10(12), 8329–8351.

Tang, F. H. M., Maggi F. 2016. Breakdown, uptake and losses of human urine chemical compounds in barley (Hordeum vulgare) and soybean (Glycine max) agricultural plots. Eﬀective- ness of human urine use in agriculture. Nutr Cycl Agroecosys, 221–245.

Tappe, Wolfgang, Groeneweg, Joost, & Jantsch, Barbara. 2002. Diﬀuse atrazine pollution in German aquifers. Biodegradation, 3–10.

Tatti, E., Goyer, C., Chantigny, M., Wertz, S., Zebarth, B.J., Burton, D.L., & Filion, M. 2014. In- fluences of over winter conditions on denitrification and nitrous oxide-producing microorganism abundance and structure in an agricultural soil amended with different nitrogen sources. Agric Ecosyst Environ, 183, 47–59.

Tejada, M., Gómez, I., Franco-Andreu, L., & Benitez, C. 2016. Role of diﬀerent earthworms in a soil polluted with oxyﬂuorfen herbicide. Short-time response on soil biochemical properties. Ecological Engineering, 86, 39 – 44.

Tilman, D., Cassman, K. G., Matson, P. A., Naylor, R., & Polasky, S. 2002. Agricultural sustainability and intensive production practices. Nature,, 418, 671 – 677.

Tipping, E., Somerville, C. J., & Luster, J. 2016. The C:N:P:S stoichiometry of soil organic matter. Biogeochemistry, 130, 117–131.

Toner, Brandy, Fakra, Sirine, Villalobos, Mario, Warwick, Tony, & Sposito, Garrison. 2005. Spatially Resolved Characterization of Biogenic Manganese Oxide Production within a Bac- terial Bioﬁlm. Applied and Environmental Microbiology, 71, 1300–1310.

160 Torstensson, L. 1980. Role of microorganisms in decomposition. In: Hance, R. J. (ed.): In- teractions between herbicides and the soil, pp. 159-178, Academic Press Inc. (London) Ltd., ISBN: 0123238404.

Tu, M., Hurd, C., & Randall, J.M. 2001. Weed Control Methods Handbook: Tools and Tech- niques for Use in Natural Areas. The Nature Cinservancy.

UCDavis. 2015. Why Residual Herbicides are Important Tools in Bat- tling Hard-to-Control Weeds. Accessed at "http: // cropwatch. unl. edu/ why-residual-herbicides-are-important-tools-battling-hard-control-weeds " on 10.10.2016.

Udikovic-Koli´ c,´ Nikolina, Scott, C., & Martin-Laurent, F. 2007. Evolution of atrazine- degrading capabilities in the environment. appl microbiol biot, 1175–1189.

USEPA. 1987. Selection criteria for mathematical models used in exposure assessments? Sur- face water models. Exposure Assessment Group, Oﬃce of Health and Environmental Assess- ment.

USEPA. 2016. https://cfpub.epa.gov/ecotox/. accessed 09.09.2016.

USGS. 2017. https://water.usgs.gov/nawqa/pnsp/usage/maps/. accessed 02.08.2018. van Bodegom, P. 2007. Microbial maintenance: A critical review on its quantiﬁcation. Micro- bial Ecology, 513–523.

Van Genuchten, M. Th. 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Sci Soc Am J, 44, 892–898.

Venterea, R. T., & Rolston, D. E. 2000. Nitric and nitrous oxide emissions following fertilizer application to agricultural soil: Biotic and abiotic mechanisms and kinetics. J Geophys Res Atmos, 105, 15117–15129.

Vereecken, H. 2005. Mobility and leaching of glyphosate: a review. Pest Manag Sci, 61, 1139–1151.

Volk, E., Iden, S.C., Furman, A., Durner, W., & Rosenzweig, R. 2016. Biofilm effect on soil hydraulic properties: Experimental investigation using soil-grown real biofilm. Water Resources Research, 52, 5813–5828.

Vryzas, Z., Papadopoulou-Mourkidou, E., Soulios, G., & Prodromou, K. 2007. Kinetics and adsorption of metolachlor and atrazine and the conversion products (deethylatrazine, deisopropylatrazine, hydroxyatrazine) in the soil proﬁle of a river basin. Eur. J. Soil Sci., 58(5), 1186–1199.

161 Šim˚uneka,J., Jarvis, N.J., van Genuchten, M.T., & Gärdenas,˝ A. 2003. Review and comparison of models for describing non-equilibrium and preferential ﬂow and transport in the vadose zone. J. Hydrol, 272, 14–35.

Walker, A. 1974. A simulation model for prediction of herbicide persistence. J. Environ. Qual., 3, 396–401.

Wan, Ni, Wang, Hao, Ng, Chun Kiat, Mukherjee, Manisha, Ren, Dacheng, Cao, Bin, & Tang, Yinjie J. 2018. Bacterial Metabolism During Bioﬁlm Growth Investigated by 13C Tracing. Frontiers in Microbiology, 9, 2657.

Wang, J., Atolia, E., Hua, B., Savir, Y., Escalante-Chong, R., & Springer, M. 2015. Natu- ral Variation in Preparation for Nutrient Depletion Reveals a Cost–Beneﬁt Tradeoﬀ. PLOS Biology, 13(1), 1–31.

Wang, Qingfeng, & Xie, Shuguang. 2012. Isolation and characterization of a high-eﬃciency soil atrazine-degrading Arthrobacter sp. strain. Int. Biodeterior. Biodegrad., 61–66.

Wang, Shizong, Seiwert, Bettina, Kästner, Matthias, Miltner, Anja, Schäﬀer, Andreas, Reemtsma, Thorsten, Yang, Qi, & Nowak, Karolina M. 2016. (Bio)degradation of glyphosate in water-sediment microcosms - a stable isotope co-labeling approach. Water Res., 99, 91– 100.

Wertz, S., Goyer, C., Zebarth, B.J., Burton, D.L., Tatti, E., Chantigny, M.H., & Filion, M. 2012. Eﬀects of temperatures near the freezing point on N2O emissions, denitriﬁcation and on the abundance and structure of nitrifying and denitrifying soil communities. FEMS Microbiol Ecol, 83(1), 242–254.

Winchester, Paul D., Huskins, Jordan, & Ying, Jun. 2008. Agrichemicals in surface water and birth defects in the United States. Acta Paediatr. Int. J. Paediatr., 664–669.

Winkelmann, D A, & Klaine, S J. 1991. Degradation and bound residue formation of four atrazine metabolites, deethylatrazine, deisopropylatrazine, dealkylatrazine and hydroxyatrazine, in a Western Tennessee soil. Environ. Toxicol. Chem., 347–354.

Wolery, T. J. 1992. EQ3/6, A Software Package for Geochemical Modeling of Aqueous Sys- tems: Package Overview and Installation Guide (Version 7.0). Lawrence Livermore National Laboratory, University of California, Livermore, California.

Wu, Mae, Quirindongo, Mayra, Sass, Jennifer, & Wetzler, Andrew. 2010. Still Poisoning the Well. Atrazine Continues to Contaminate Surface Water and Drinking Water in the United States. Nat. Resour. Def. Counc. Washington, DC.

Xu, T., Spycher, N., Sonnenthal, E., Zhang, G., Zheng, L., & Pruess, K. 2011. TOUGHREACT Version 2.0: A simulator for subsurface reactive transport under non-isothermal multiphase ﬂow conditions. Ground Water, 6, 763–774.

162 Yalkowsky, S.H. 2003. Handbook of Aqueous Solubility Data. CRC Press.

Yu, C., & Zheng, C. 2010. HYDRUS: software for ﬂow and transport modeling in variably saturated media. Ground Water, 6, 787–791.

Zablotowicz, R. M., Accinelli, C., Krutz, L. J., & Reddy, K. N. 2009. Soil depth and tillage eﬀects on glyphosate degradation. j agr food chem, 57, 4867–4871.

Zablotowicz, R.M., Krutz, L.J., Weaver, M.A., Accinelli, C., & Reddy, K.N. 2008. Glufosinate and ammonium sulfate inhibit atrazine degradation in adapted soils. Biol Fertil Soils, 45(1), 19–26.

Zablotowicz, Robert M., Weaver, Mark A., & Locke, Martin A. 2006. Microbial adaptation for accelerated atrazine mineralization/ degradation in Mississippi Delta soils. Weed Sci., 538–547.

Zaranyika, M.F., & Nyandoro, M.G. 1993. Degradation of glyphosate in the aquatic environment: An enzymic kinetic model that takes into account microbial degradation of both free and colloidal (or sediment) particle adsorbed glyphosate. j agr food chem, 41, 838–842.

Zhalnina, K., Louie, K.B., Hao, Z., Mansoori, N., da Rocha, U.N., Shi, S., Cho, H., Karaoz, U., Loqué, D., Bowen, B.P., Firestone, M.K., Northen, T.R., & Brodie, E.L. 2018. Dynamic root exudate chemistry and microbial substrate preferences drive patterns in rhizosphere microbial community assembly. Nature Microbiology, 3, 470–480.

Zhao, Jiao, Scheibe, Timothy D., & Mahadevan, R. 2011. Model-based analysis of the role of biological, hydrological and geochemical factors aﬀecting uranium bioremediation. Biotech- nol Bioeng, 108, 1537–1548.

Zheng, W., Zhao, Z., Gong, O., Zhai, B., & Li, Z. 2018. Responses of fungal–bacterial community and network to organic inputs vary among diﬀerent spatial habitats in soil. Soil Biology and Biochemistry, 125, 54 – 63.

163

Parameter estimation of ATZ reaction network

Appendix A contains the graphs showing the parameter estimation outcome of the microbiological reactions belonging to the ATZ reaction network.

P1R1a

0.6 2 0.12 60 0.4 80 (1) (2) (3)

0.45 1.5 0.3 60 ) ) ) 0.08 40 −1 −1 ATZ, exp. (mM) −1 ATZ, exp. (mM) ATZ, model (mM) ATZ, exp. ATZ, model (mM) CH O, exp. (dM) CH O, exp. (dM) 0.3 2 1 ATZ, model 0.2 2 40 CH O, model (dM) B CH O, model (dM) 2 Ral 2 Chemicals Chemicals B B Pse Ral Chemical (mM) Biomass (mg L Biomass (mg L

Biomass (mg L 0.04 20 0.15 0.5 0.1 20

0 0 0 0 0 0 0 0.5 1 1.5 2 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 t (d) t (d) t (d)

0.4 40 0.2 120 0.16 80 (4) (5) (8)

0.3 30 0.15 90 0.12 60 ) ) ) −1 −1 ATZ, exp. (mM) ATZ, exp. (mM) −1 ATZ, model (mM) ATZ, model (mM) ATZ, exp. CH O, exp. (dM) CH O, exp. (dM) 0.2 2 20 0.1 2 60 0.08 ATZ, model 40 B CH O, model (dM) CH O, model (dM) Noc 2 2 Chemicals B Chemicals B Ral Pse Chemical (mM) Biomass (mg L Biomass (mg L Biomass (mg L 0.1 10 0.05 30 0.04 20

0 0 0 0 0 0 0 1 2 3 4 0 0.5 1 1.5 2 0 2 4 6 8 10 12 t (d) t (d) t (d)

0.16 80 0.16 120 0.16 80 (9) (10) (11)

0.12 60 0.12 90 0.12 60 ) ) ) −1 −1 −1

ATZ, exp. ATZ, exp. ATZ, exp. 0.08 ATZ, model 40 0.08 ATZ, model 60 0.08 ATZ, model 40 B B B Noc Noc Noc Chemical (mM) Chemical (mM) Chemical (mM) Biomass (mg L Biomass (mg L Biomass (mg L 0.04 20 0.04 30 0.04 20

0 0 0 0 0 0 0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 2 4 6 8 10 12 t (d) t (d) t (d)

0.16 100 0.16 100 0.16 40 (12) (13) (14)

0.12 75 0.12 75 0.12 30 ) ) ) −1 −1 −1

ATZ, exp. ATZ, exp. ATZ, exp. 0.08 ATZ, model 50 0.08 ATZ, model 50 0.08 ATZ, model 20 B B B Noc Noc Noc Chemical (mM) Chemical (mM) Chemical (mM) Biomass (mg L Biomass (mg L Biomass (mg L 0.04 25 0.04 25 0.04 10

0 0 0 0 0 0 0 2 4 6 8 10 12 0 2 4 6 8 10 12 0 2 4 6 8 10 12 t (d) t (d) t (d)

165 0.2 1.6 0.16 100 0.2 0.8 (15) (16) (17)

0.15 1.2 0.12 75 0.15 0.6 ) ) 1 )

− ATZ, exp. (mM) −1 −1 ATZ, exp. (mM) ATZ, added (mM) ATZ, exp. ATZ, model (mM) ATZ, model (mM) 0.08 ATZ, model 50 0.1 0.8 0.1 CH O, exp. (dM) 0.4 CH O, exp. (dM) 2 B 2 Noc CH O, model (dM) Chemicals CH O, model (dM) Chemicals 2 2 B Chemical (mM)

Biomass (mg L Noc Biomass (mg L Biomass (mg L B Noc 0.04 25 0.05 0.4 0.05 0.2

0 0 0 0 0 0 0 2 4 6 8 10 12 0 1 2 3 4 5 0 1 2 3 4 t (d) t (d) t (d)

0.2 80 (18)

0.15 60 )

ATZ, exp. (mM) −1 ATZ, model (mM) CH O, exp. (dM) 0.1 2 40 CH O, model (dM) 2 Chemicals B Noc Biomass (mg L 0.05 20

0 0 0 1 2 3 4 5 t (d)

Figure A1: Aerobic ATZ degradation to HOATZ along P1R1a. For experiment 17, one experimental guess point was added after interpretation of the degradation curve to allow mathematical determination of the MMM kinetic equations. Experimental data are: (1) from Mandelbaum et al. (1995); (2) to (4) from Radosevich et al. (1995); (5) from Katz et al. (2000); (8) to (15) from Smith et al. (2005); and (16) to (18) from Smith & Crowley (2006). BPse, BRal, BNoc refer to Pseudomonas sp. ADP, Ralstonia basilensis M91-3, and Nocardia sp. biomass concentrations, respectively.

P1R1b

0.4 200 (6) 0.35 80 (7) 0.3 70 0.3 150 ATZ, exp. (mM) ATZ, exp. (mM) ATZ, model (mM) )

ATZ, model (mM) 1 0.25 60

CH O, exp. (dM) ) − 2

CH O, exp. (dM) −1 2 CH O, model (dM) 2 CH O, model (dM) 0.2 50 0.2 2 100 NO−, exp. (cM) − 3 NO , exp. (cM) − 3 NO , model (cM) Chemicals 0.15 3 40

− Chemicals NO , model (cM) − 3 NO , exp. (cM) Biomass (mg L 2 − Biomass (mg L NO , exp. (cM) 0.1 NO−, model (cM) 30 0.1 2 50 2 NO−, model (cM) B 2 Pse B 0.05 20 Pse

0 0 0 10 0 0.5 1 1.5 2 0 0.5 1 1.5 2 t (d) t (d)

Figure A2: Anaerobic ATZ degradation to HOATZ along P1R1b. Experimental data (6) and (7) from Katz et al. (2000). BPse refers to Pseudomonas sp. ADP biomass concentration.

166 P1R2

0.05 2 0.03 3 HOATZ, exp. (mM) (19) 0.04 160 (21) (20) HOATZ, exp. (mM) 0.04 HOATZ, model (mM) 1.6

NH4, model (M) ) )

0.03 120 −1 ) 0.02 2 −1 HOATZ, exp. (mM) B −1 0.03 Community 1.2 HOATZ, exp. HOATZ, exp. (mM) HOATZ, model HOATZ, model (mM) 0.02 80 B CH2O, exp. (daM)

Community Chemicals 0.02 0.8

Chemicals CH2O, model (daM)

Chemical (mM) 0.01 1

B Biomass (mg L Biomass (mg L

Community Biomass (mg L 0.01 40 0.01 0.4

0 0 0 0 0 0 0 1 2 3 4 5 6 7 0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16 t (d) t (d) t (d)

Figure A3: Aerobic HOATZ degradation to NIPA along P1R2; (19) to (21) from Kumar & Singh (2016). BCommunity refers to the Community of bacteria biomass concentration.

P1R3

0.7 1.4 (22) 0.6 1.2

0.5 1 ) −1 0.4 NIPA, exp. 0.8 NIPA, model B 0.3 Pse 0.6 Chemical (mM) 0.2 0.4 Biomass (mg L

0.1 0.2

0 0 0 0.2 0.4 0.6 0.8 1 t (d)

Figure A4: Aerobic NIPA biodecomposition to CYA along P1R3. Experimental data (22) from Boundy-Mills et al. (1997). BPse refers to Pseudomonas sp. ADP biomass concentration.

P2R1 and P3R1

100 0.5 0.1 250 0.1 40 ATZ, exp. ATZ, exp. (mM) ATZ, exp. (mM) ATZ, model (23) ATZ, model (mM) (24) ATZ, model (mM) (25) DIA, exp. DIA, exp. (mM) DIA, exp. (mM) 80 DIA, model 0.4 0.08 DIA, model (mM) 200 DIA, model (mM) DEA, exp. DEA, exp. (mM) 0.075 DEA, exp. (mM) 30

DEA, model ) DEA, model (mM) ) DEA, model (mM) )

B −1 CH O, model (M) −1 CH O, model (M) −1 M) EClo 2 2

µ 60 0.3 0.06 150 Unknown, model (mM) B Com B Com 0.05 20

40 0.2 Chemicals 0.04 100 Chemicals Chemicals ( Biomass (mg L Biomass (mg L Biomass (mg L 0.025 10 20 0.1 0.02 50

0 0 0 0 0 0 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 t (d) t (d) t (d)

Figure A5: Aerobic ATZ degradation to DIATZ and DEATZ along P2R1 and P3R1, respectively. Metabolites DIATZ and DEATZ were biodegraded to DIHOATZ and DIDEATZ, respectively, during the same experiment. (23) from Solomon et al. (2013); (24) from Behki et al. (1993); (25) from Behki & Khan (1994). BEClo refers to Enterobacter cloacae biomass concentration. BRho refers to either Rhodococcus strain TE1 or B30 biomass concentration.

167 P2R2

0.08 2 100 0.5 0.08 0.2 ATZ, exp. (26) DIA, exp. (27) ATZ, model (23) DIA, added DIA, exp. DIA, model 80 DIA, model 0.4 CH2O, exp. 0.06 1.5 DEA, exp. 0.06 CH2O, model 0.15 ) )

) B DEA, model Rho −1

−1 DIA, exp. B −1 M) EClo

µ 60 0.3 DIA, added DIA, model 0.04 0.1 0.04 CH2O, exp. 1 CH2O, model 40 0.2 B Rho Chemicals (mM) Chemicals ( Chemicals (mM) Biomass (mg L Biomass (mg L Biomass (mg L 0.02 0.05 0.02 0.5 20 0.1

0 0 0 0 0 0 0 1 2 3 4 5 6 7 0 1 2 3 4 0 1 2 3 4 t (d) t (d) t (d)

Figure A6: Aerobic DIATZ biodecomposition to DIHOATZ along P2R2. Experimental data are: (23) from Solomon et al. (2013); (26) and (27) from Shao et al. (1995). BCom and BRho refer to the Community of bacteria and Rhodococcus biomass concentrations, respectively.

P2R2 and P3R2

100 0.5 0.1 40 ATZ, exp. ATZ, exp. (mM) ATZ, model (23) ATZ, model (mM) (25) DIA, exp. DIA, exp. (mM) 80 DIA, model 0.4 DIA, model (mM) DEA, exp. 0.075 DEA, exp. (mM) 30

DEA, model ) DEA, model (mM) )

B −1 CH O, model (M) −1 M) EClo 2

µ 60 0.3 B Com 0.05 20

40 0.2 Chemicals Chemicals ( Biomass (mg L Biomass (mg L 0.025 10 20 0.1

0 0 0 0 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 t (d) t (d)

Figure A7: Aerobic DIATZ and DEATZ degradation to DIHOATZ and DIDEATZ along P2R2 and P3R2, respectively. (23) from Solomon et al. (2013), (25) from Behki & Khan (1994). BEClo refers to Enterobacter cloacae biomass concentration. BRho refers to Rhodococcus strain B30 biomass concentration.

P4R1

3 3 (28) )

2 2 −1 CYA, exp. CYA, model B Ecoli

Chemical (mM) 1 1 Biomass (mg L

0 0 0 1 2 3 4 t (d)

Figure A8: Aerobic CYA biodecomposition to BIU and CO2 along P4R1. (28) from Martinez et al. (2001). BEcoli refers to Escherichia coli biomass concentration with the gene atzD

168 P4R2

1 0.1 (29)

0.8 0.08 ) −1 0.6 BIU, exp. 0.06 BIU, exp. B 0.4 Ecoli 0.04 Chemical (mM) Biomass (mg L 0.2 0.02

0 0 0 1 2 3 t (d)

Figure A9: Aerobic BIU biodecomposition to ALP and NH3 along P4R2. (29) from Martinez et al. (2001). BEcoli refers to Escherichia coli biomass concentration with the gene atzE

P4R3

5 2 (30)

4 1.6 ) −1 3 ALP, exp. 1.2 ALP, exp. B 2 Ecoli 0.8 Chemical (mM) Biomass (mg L 1 0.4

0 0 0 1 2 3 t (d)

Figure A10: Aerobic ALP biodecomposition to CO2 and NH3 along P4R3. (30) from Martinez et al. (2001). BEcoli refers to Escherichia coli biomass concentration with the gene atzF

25 5 (31)

20 4 ) −1 15 3

10 2 Chemical (mM) Biomass (mg L 5 ETA, exp. 1 ETA, model B Art 0 0 0 0.2 0.4 0.6 t (d)

Figure A11: Aerobic ETA biodegradation to NH3 and acetaldehyde along P5. (31) from Levering et al. (1984) BArt refers to Arthrobacter P1 biomass concentration.

169

Parameter estimation of GLP reaction network

Appendix B contains the graphs showing the parameter estimation outcome of the microbiological reactions belonging to the GLP reaction network.

−3 −3 x 10 x 10 1 4 0.02 1000 8 200 GLP, exp. GLP, model AMPA, exp. AMPA, model CH O/100, exp. 2 CH O/100, model 0.75 2 3 0.015 750 6 150 PO−, exp. 4 ) ) ) PO3−, model 4 -1 GLP, exp. -1 -1 B Fla GLP, model GLP, exp. AMPA, exp. GLP, model 0.5 2 0.01 AMPA, model 500 4 AMPA, exp. 100 CH2O/100, exp. AMPA, model CH2O/100, model B AgAc B Pse Chemicals (M) Chemicals (M) Chemicals (M) Biomass (mg L Biomass (mg L Biomass (mg L

0.25 1 0.005 250 2 50 Test 1 Test 2 Test 3 P1R1s P1R1s P1R1 P1R2s (a) (b) (c) 0 0 0 0 0 0 0 2 4 6 8 10 12 0 1 2 3 4 0 2 4 6 8 t (day) t (day) t (day)

Figure B1: Aerobic GLP biodegradation to AMPA along P1R1s; (a) from Balthazor & Hallas (1986). BFla refers 3 – 3 – to the Flavobacterium sp. biomass concentration, while PO4,Upt refers to the PO4 liberated during AMPA biodegradation and uptake by the microorganisms to support growth. (b) from Jacob et al. (1988). BPse refers to the Pseudomonas sp. LBr biomass concentration. (c) from Mcauliﬀe et al. (1990). BAgAc refers to the Agrobacterium radiobacter and Achromobacter Group V D biomass concentration.

−3 −3 x 10 x 10 1 40 1 4 0.03 150 (a) (b) (c) P2R1s P1R1s P1R3a P1R2s Test 4 Test 5 Test 6 0.75 30 0.75 3

) GLP, exp. ) ) GLP, exp. GLP, model 0.02 100 −1 −1 −1 GLP, model AMPA, exp. PO3−, exp. AMPA, model 4 CH O/100, exp. MTH, exp. 2 PO3−, model MTH, model 0.5 4 20 0.5 CH O/100, model 2 2 B CH O/1000, exp. − Art 2 PO , exp. 4 CH O/1000, model 3− 2 PO , model Chemicals (M) Chemicals (M) 4 Chemicals (M)

B Biomass (mg L B Biomass (mg L 0.01 50 Biomass (mg L Pse Fla 0.25 10 0.25 1

0 0 0 0 0 0 0 2 4 6 8 10 0 2 4 6 8 10 12 0 0.25 0.5 0.75 1 t (day) t (day) t (day)

3 – + Figure B2: (a) Aerobic GLP biodegradation to SRC, PO4 , and H along P2R1s; observations from Moore et al. (1983). BPse refers to the Pseudomonas PG2982 biomass concentration. (b) Aerobic AMPA biodegradation to 3 – + MTH, PO4 , and H along P1R2s; observations from Balthazor & Hallas (1986). BFla refers to the Flavobac- 3 – 3 – terium sp. biomass concentration, while PO4,Upt refers to the PO4 liberated during AMPA biodegradation and uptook by the microorganisms to support growth. (c) Aerobic MTH metabolization to formaldehyde and NH3; observations from Levering et al. (1984). BArt refers to the Arthrobacter P1 biomass concentration.

171 −6 −5 x 10 x 10 0.1 28 6 0.015 1.2 0.024 (a) (b) (c) P1R3b P2R2a P2R2a

Test 7 Test 8 Test 9 0.075 21

) 4 0.01 ) 0.8 0.016 ) −1 −1 −1

0.05 14 Chemicals (M) Chemicals (M) Chemicals (M)

MTH, exp. Biomass (mg L 2 0.005 Biomass (mg L 0.4 0.008 Biomass (mg L MTH, model O , exp. O , exp. NH , exp. 2 2 0.025 3 7 O , model O , model NH , model 2 2 3 CH , exp. SRC, exp. SRC, exp. 4 SRC, model SRC, model CH , model 4 FRMH, model FRMH, model B B B Met Pse Pse 0 0 0 0 0 0 0 2 4 6 8 0 0.05 0.1 0.15 0 0.05 0.1 0.15 t (day) t (day) t (day)

Figure B3: (a) Anaerobic MTH metabolization to CH4, CO2, and NH3 along P1R3b; observations from Hippe et al. (1979). BMet refers to the Methanosarcina barkeri biomass concentrations; (b) Aerobic SRC metabolization to GLY and formaldehyde along P2R2a; observations from Appleyard & Woods (1956). BPse refers to the Pseu- domonas Ovalis biomass concentrations; (c) Aerobic SRC metabolization to GLY and formaldehyde along P2R2a; observations from Appleyard & Woods (1956). BPse refers to the Pseudomonas Ovalis biomass concentrations.

−5 −5 −3 x 10 x 10 x 10 1.5 0.04 0.08 4 2 8 (b) SRC, exp. (c) O , exp. (a) 2 SRC, model P2R3a O , model P2R2a P2R2b FRM, exp. 2 FRM, model Test 12 GLY, model Test 10 Test 11 B ACT, exp. Pse ACT, model 0.06 B 3 1.5 6 Eub ) ) 1 0.03 ) −1 −1 −1 O , exp. 2 O , model 2 SRC, exp. 0.04 2 1 4 SRC, model FRMH, model B Pse Chemicals (M) Chemicals (M) Chemicals (M) Biomass (mg L Biomass (mg L 0.5 0.02 Biomass (mg L

0.02 1 0.5 2

0 0.01 0 0 0 0 0 0.05 0.1 0.15 0 0.4 0.8 1.2 0 0.1 0.2 0.3 t (day) t (day) t (day)

Figure B4: (a) Aerobic SRC metabolization to GLY and formaldehyde along P2R2a; observations from Appleyard & Woods (1956). BPse refers to the Pseudomonas Ovalis biomass concentrations; (b) Anaerobic SRC metabolization to MTH and acetate along P2R2b; observations from Hormann & Andreesen (1989). BEub refers to the Eubacterium acidaminophilum biomass concentrations; (c) Aerobic GLY metabolization to NH3 and formaldehyde along P2R3a; observations from Appleyard & Woods (1956). BPse refers to the Pseudomonas Ovalis biomass concentrations.

0.02 10 100 5 (a) (b) P2R3a P2R3b Test 13 Test 14 80 4 0.015 7.5 ) ) −1 −1 60 3 GLY*1000, exp. GLY*1000, model 0.01 5 ACT*1000, exp. ACT*1000, model B 40 Clo 2 Chemicals (M) Chemicals (M) Biomass (mg L Biomass (mg L

0.005 2.5 O *1000, exp. 20 1 2 O *1000, model 2 GLY*1000, model B *1000 Pse 0 0 0 0 0 0.1 0.2 0.3 0 0.5 1 1.5 2 t (day) t (day)

Figure B5: (a) Aerobic GLY metabolization to NH3 and formaldehyde along P2R3a; observations from Appleyard & Woods (1956). BPse refers to the Pseudomonas Ovalis biomass concentrations; (b) Anaerobic GLY metabolization to NH3 and formaldehyde along P2R3b; observations from Därre & Andreesen (1982a). BClo refers to the Clostridium purinolyticum biomass concentrations.

−3 −3 x 10 x 10 3 3 (a) (b) P2R1c P1R2c

2 2

PO3−, exp. PO3−, exp. 4 4 PO3−, model PO3−, model 4 4 GLP, model AMPA, model

Chemicals (M) 1 Chemicals (M) 1

Test 15 Test 16 0 0 0 0.5 1 1.5 2 0 10 20 30 40 t (day) t (day)

+ 3 – Figure B6: (a) GLP chemical degradation catalyzed by Mn ions present in birnessite to SRC, H , and PO4 along P2R1c; observations from Li et al. (2015). (b) AMPA chemical degradation catalyzed by Mn ions present in + 3 – birnessite to MTH, H , and PO4 along P1R2c; observations from Li et al. (2015).

172