Emergence of the Reactivity Continuum of Organic Matter from

Subscriber access provided by University of Sussex Library

Article

The emergence of the reactivity continuum of organic matter from kinetics of a multitude of individual molecular constituents Alina Mostovaya, Jeffrey Alistair Hawkes, Birgit Koehler, Thorsten Dittmar, and Lars J. Tranvik Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.7b02876 • Publication Date (Web): 15 Sep 2017 Downloaded from http://pubs.acs.org on September 19, 2017

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 35

Environmental Science & Technology

1

The emergence of the reactivity continuum of

2

organic matter from kinetics of a multitude of

3

individual molecular constituents

4

Alina Mostovaya*,1, Jeffrey A. Hawkes2, Birgit Koehler1, Thorsten Dittmar3, Lars J.

5

Tranvik1

6

1

7

Uppsala University, Norbyvägen 18 D, 75236, Uppsala, Sweden

8

2

9

Husargatan 3, 75124, Uppsala, Sweden

Department of Ecology and Genetics/Limnology, Evolutionary Biology Centre,

Department of Chemistry – BMC, Analytical Chemistry, Uppsala University,

10

3

11

for Chemistry and Biology of the Marine Environment, University of Oldenburg,

12

Carl-von-Ossietzky-Str. 9-11, 26129, Oldenburg, Germany

13

ABSTRACT: The reactivity continuum (RC) model is a powerful statistical approach

14

for describing the apparent kinetics of bulk organic matter (OM) decomposition.

15

Here, we used ultrahigh resolution mass spectrometry data to evaluate the main

16

premise of the RC model, namely that there is a continuous spectrum of reactivity

17

within bulk OM, where each individual reactive type undergoes exponential decay.

18

We performed a 120 days OM decomposition experiment on lake water, with an

Research Group for Marine Geochemistry (ICBM-MPI Bridging Group), Institute

1 ACS Paragon Plus Environment


Page 2 of 35

19

untreated control and a treatment preexposed to UV light, and described the loss of

20

bulk dissolved organic carbon (DOC) with RC modeling. The behavior of individual

21

molecular formulae was described by fitting the single exponential model to the

22

change in peak intensities over time. The range of the empirically-derived apparent

23

exponential decay coefficients (kexp) was indeed continuous. The character of the

24

corresponding distribution, however, differed from the conceptual expectations, due

25

to the effects of intrinsic averaging, overlaps in formula-specific loss and formation

26

rates, and the limitation of the RC model to include apparently accumulating

27

compounds in the analysis. Despite these limitations, both the RC model-simulated

28

and empirical (mass spectrometry-derived) distributions of kexp captured the effects of

29

preexposure to UV light. Overall, we present experimental evidence that the reactivity

30

continuum within bulk OM emerges from a range of reactivity of numerous

31

individual components. This constitutes direct empirical support for the major

32

assumption behind the RC model of natural OM decomposition.

33

1. INTRODUCTION

34

Organic matter (OM) is a major component of the carbon cycle. The complex

35

nature of OM was suggested in multiple studies demonstrating that different fractions

36

of OM decompose at differing rates.1–4 With the increasing understanding of OM

37

complexity, OM was conceptualized as a continuous spectrum of constituents with

38

differing reactivity,5–7 long before it was possible to identify the broad range of OM

39

components. Recent advances in analytical chemistry, and particularly high-resolution

40

mass spectrometry, have demonstrated that natural OM may consist of 104−106

41

compounds, belonging to different chemical classes and groups,8–11 yet the emerging

42

detailed insights into OM composition have not been explicitly connected to the

43

concept of the reactivity continuum. 2 ACS Paragon Plus Environment

Page 3 of 35


44

The reactivity continuum (RC) model approach is gaining popularity as a tool for

45

realistic description of decomposition in a variety of environments, including marine

46

sediments,6,7,12 soil,13,14 aquatic dissolved OM15–17 and litter.18,19 The RC model was

47

often shown to perform better than single- and multi-exponential models, based on

48

mathematical indicators assessing accuracy and parsimony,14,16,18,20 as well as the

49

predictive power of the model.15 A study analyzing the Long-term Intersite

50

Decomposition Experiment (LIDET data) with RC modeling showed that the

51

obtained RC model parameters were suitable to predict independent literature-derived

52

decomposition time series for the majority of litter-biome combinations.18 Because

53

the RC model can successfully track the effects of external regulators of

54

decomposition,18,20 it has been proposed to improve the mechanistic understanding of

55

decomposition at large spatial and temporal scales.18 Accordingly, Aumont et al.

56

(2017) used the RC model in a global ocean biogeochemical model to account for

57

variable reactivity of particulate organic carbon, and got a close match to empirical

58

observations.12

59

The RC model assumes that decomposition of bulk OM can be described as an

60

integral of single-pool exponential decay functions weighed with an initial probability

61

distribution of reactivity. When the chosen functional shape for the initial probability

62

distribution of reactivity allows a simple analytical solution (e.g., gamma distribution)

63

the RC model becomes a parameter-parsimonious and computationally simple

64

approach to analyze decomposition data. The main gamma RC model assumptions6

65

are:

66

1)

67

Each individual reactive type decays over time following a single-pool exponential model



68

2)

69

Page 4 of 35

The individual reactive types form a continuous spectrum of initial reactivity towards decomposition, that can be described by a probability distribution

70

3)

71

The gamma distribution is a flexible and suitable function to describe the initial probability distribution of reactivity.

72

Since the high-resolution analytical techniques that could provide detailed

73

information on the behavior of OM constituents became only recently available, these

74

RC model assumptions remain so far empirically untested. With the recent

75

development of ultrahigh-resolution electrospray ionization Fourier transform ion

76

cyclotron resonance mass spectrometry (FT-ICR-MS), which is currently one of the

77

most powerful analytical techniques for molecular characterization of natural OM,21–

78

23

79

Further, the signal intensity of each molecular mass is essentially linearly proportional

80

to the concentration of the corresponding isomeric mixture of compounds, provided

81

that the sample matrix is similar.24 This allows interpreting the changes in signal

82

intensity over time in a quantitative way. Therefore, using FT-ICR-MS, we were able

83

to experimentally investigate one of the main assumptions of the RC model that

84

individual reactive types form a continuum of decay rates within bulk OM.

thousands of molecular formulae can now be identified in environmental samples.

85

Taking decomposition of lake water dissolved organic matter (DOM) as a case, we

86

performed a 120 days long decomposition experiment. We incubated untreated lake

87

water as well as lake water subjected to UV exposure to stimulate DOM

88

decomposition. Decomposition of bulk DOM was analyzed with the gamma RC

89

model, obtaining the respective probability distribution of initial decay coefficients

90

k.20 In this study, we fitted the single-pool exponential model to each time series of

91

intensities related to individual molecular formulae, and extracted the corresponding

92

apparent exponential decay coefficients kexp. We compare the range of RC model-


Page 5 of 35


93

simulated and empirically-derived initial decay coefficients, and discuss our results

94

with respect to the assumptions of the RC model. Our results support the existence of

95

a reactivity continuum within DOM and provide an argument in favor of continuous

96

models of OM decay. In the context of potential application of our approach, further

97

discussion reveals the advantages and limitations of using ultrahigh resolution mass

98

spectrometry for evaluating the kinetics of OM decay.

99

2. METHODS

100

2.1. Experimental setup

101

Waters from Lakes Ramsjön (59°50'N, 17°13'E) and Övre Långsjön (59°52'N,

102

18°01'E), located in east-central Sweden, were sampled in October 2013. Water

103

samples were taken from the surface (0–0.5 m depth), 0.2 µm filtered (Supor 200, Pall

104

Corporation) and kept in the dark at 4°C for three weeks prior to the start of

105

decomposition experiments. These three weeks were required to expose portions of

106

the filtered water we to artificial UV light (144 hours under an UV lamp providing an

107

irradiance of 35 W m−2 between 300 and 400 nm), which resulted in photodegradation

108

of about 30% of dissolved organic carbon (DOC).20 Prior to the incubations the

109

filtered water was inoculated with a 64 µm plankton net filtered aliquot from the

110

respective lake (5% of the volume), and amended with inorganic nutrients (480 µg N

111

L−1 as KNO3 and 100 µg P L−1 as Na2HPO4). The water was incubated at 20°C in

112

sealed 40 mL glass vials capped with PTFE-lined silicone septa. The vials were filled

113

headspace free and submersed in water to minimize gas exchange. The incubations

114

were conducted in the dark and lasted for 120 days. Each of the two treatments (non-

115

manipulated and UV-manipulated) was duplicated within each lake, which yielded 8

116

time series of DOC loss. In the course of decomposition experiment individual vials

117

in each time series were sacrificed to measure DOC concentration and take 5 ACS Paragon Plus Environment


Page 6 of 35

118

subsamples for mass spectrometry analysis (Supporting Information (SI) Figure S1).

119

The DOC concentrations were measured on a Sievers 900 TOC Analyzer (General

120

Electric Analytical Instruments, Manchester, UK) at 20 and 18 time points in the non-

121

manipulated and UV-manipulated treatment, respectively.20 Samples for mass

122

spectrometry analysis were taken at 12 time points: water was transferred into 2 mL

123

glass vials with PTFE-lined silicone septa and stored at 4°C prior to analysis. Before

124

use, 40 mL glass vials were acid-washed (10% HCl), rinsed with MilliQ water and

125

pre-combusted for 4 h at 550°C. The 2 mL glass vials were pre-combusted for 4 h at

126

450°C. All plastic caps and silicone septa were pre-soaked in methanol for 24 h and

127

repeatedly rinsed with MilliQ water.

128

2.2. Mass spectrometry analysis

129

Lake water aliquot volumes were adjusted by diluting the sample with ultrapure

130

water to a concentration of 10 mg C L−1. Further, the aliquots were mixed with

131

methanol (HPLC-grade, Sigma-Aldrich) in a proportion of 2:1 (final concentration

132

6.7 mg C L−1), re-filtered (0.2 µm PTFE), and introduced into a 15-T ultrahigh-

133

resolution electrospray ionization Fourier transform ion cyclotron resonance mass

134

spectrometer (FT-ICR-MS; Bruker Daltonics) at 360 µl h−1. Electrospray ionization

135

(ESI) was performed in negative mode. Mass spectra were collected over 400 scans,

136

with an ion accumulation time of 0.5 s, and within a range of 150–2000 m/z. Each

137

sample was individually calibrated with an internal reference mass list generated from

138

North Equatorial Pacific Intermediate Water (NEqPIW)25 using the Bruker Daltonics

139

Data Analysis software package. Only peaks with a signal to noise ratio greater than

140

four were considered. Where possible, peaks were assigned a formula containing C,

141

H, O, N, S and P based on the following conditions: C ≥ O; O > (2P + S); N ≤ 4; S ≤ 4

142

and P ≤ 1. CHO compounds had corresponding peaks at +1.0034 mass units due to 6 ACS Paragon Plus Environment

Page 7 of 35


13

143

1% abundance of

144

confirming the single charge of the ion. Formulae containing S, P, and N>1 were not

145

associated with any systematic patterns in apparent exponential coefficients kexp, i.e.,

146

their presence did not affect the shape of empirical distribution of kexp (see section

147

2.3). Therefore, to minimize false positives, i.e., possible incorrect assignments, we

148

used a more conservative approach which does not consider S-, P- and N>1-

149

containing formulae. Unassigned masses were not considered, as most of them

150

represent either isotopologues (e.g.,

151

S1). We did not observe any significant shift or systematic pattern in the amount and

152

total intensity of unassigned peaks in the beginning versus the end of the incubations.

153

Also, unassigned peaks were not associated with any specific patterns in kexp. Peaks

154

detected in procedural blanks, as well as the peaks commonly identified as

155

contaminants, and peaks with unusually and non-reproducibly high intensity in

156

samples were removed from the dataset. In addition, the peaks with mass defect in the

157

range of 0.3-0.8 were disregarded. The peaks present in only one of the two replicates

158

of each of the four lake-treatment combinations (i.e., each lake with and without UV

159

pretreatment) were disregarded as well. Examples of original mass spectra from

160

different time points and graphical representations of formula assignment are

161

presented in SI figures S2 and S3, respectively. The SI Table 1 gives details on data

162

reduction procedures (total intensity of assigned versus unassigned peaks, etc.)

C, with intensities confirming the formula assignment and

13

C) or noise (SI Figures S2, S3, and SI Table

163

Here, before proceeding to further analysis we checked whether the time series of

164

total intensity followed the trends in bulk DOC concentration (SI Figure S4). One of

165

the replicate incubations did not meet this criterion and was therefore removed from

166

further analysis. Thereafter, to account for possible variable contamination between

167

individual time series, data from each lake-treatment-replicate combination (seven in



Page 8 of 35

168

total) was processed separately. A detection limit was applied based on dynamic

169

range. The dynamic range of assigned peaks was determined in each sample as a ratio

170

between maximum and minimum assigned peak intensity. Detection limit was

171

calculated by dividing maximum intensity in each sample by the dynamic range of the

172

worst sample. All intensities below the detection limit were disregarded. The

173

intensities were further normalized to the sum of signal intensities in each sample and

174

multiplied to account for dilution, i.e., to estimate the equivalent intensity in the

175

undiluted sample. This was done to correctly estimate the relative change in

176

concentration over the course of the decomposition experiment.24,26 For the purpose

177

of model fitting, only the peaks that were detected on at least eight days of the 12-

178

points time series were considered.

179

Our analysis strongly relies on the assumption of a linear relationship between

180

intensity and concentration, which was documented previously24 in a series of dilution

181

experiments where Amazon DOM was incrementally mixed with Atlantic DOM.

182

Around 70% of molecular formulae in the terrestrial and marine endmembers showed

183

a significant (p ≤ 0.05) linear response of signal intensity according to the

184

experimental mixing.24 For most of the remaining 30% of formulae the linear

185

response could not be tested because their initial intensity did not differ between the

186

Amazon and Atlantic endmembers (Thorsten Dittmar, personal communication). In

187

our samples the major peaks were inorganic and did not decay over time (SI Figure

188

S2). This likely provided a stable matrix, and ensured a reliable comparison of OM

189

peaks in each time series of our samples.

190

To evaluate the instrument-specific noise variations we used the 26 time point

191

measurements of an in-house DOM reference material (NEqPIW)25 (final


Page 9 of 35


192

concentration 15 mg C L−1). The standard was measured in the beginning and end of

193

every day, on 13 days in total.

194

2.3. Model fitting and statistical analysis

195

The gamma RC model was fitted to the time series of relative DOC loss, as

196

described in detail earlier.20 Briefly, the two model parameters, a and v, were

197

estimated based on the following relationship:

DOCt  a  = DOC0  a + t 

198

v

(1)

199

where a is a rate parameter indicating the average lifetime of the more reactive

200

compounds (days) and v (unitless) is a shape parameter characterizing the shape of the

201

distribution near k=0.6 The apparent decay coefficient kt (day−1) at a given time point

202

is calculated as:

203

kt =

v a+t

(2)

204

.27 The initial apparent first-order decay coefficient k0 (expected value of the initial

205

probability distribution of reactivity) is therefore defined as v/a.

206

To approach the assumption of the RC model from the empirical end we used mass

207

spectrometry-derived information on signal intensities of molecular formulae and its

208

change over time. For each formula, the relative change in signal intensity was

209

described with a single pool exponential model:28

210

It = I0e

− kexpt

(3)

211

where kexp is an apparent exponential decay coefficient (day−1) and It simulates the

212

change in intensity over time. Model fitting was conducted using generalized least

213

squares modeling to allow the errors to be correlated (R-function gnls). The goodness

214

of model fit was evaluated using the normalized root-mean-square error (NRMSE).



Page 10 of 35

215

We used the NRMSE to filter the datasets from the time series containing a lot of

216

noise. To establish an acceptable NRMSE we calculated the 95% quantile of the

217

NRMSE distribution acquired from a 26-point series of measurements of DOM

218

standard (NEqPIW) of stable concentration (15 mg C L−1). Based on these

219

calculations, the threshold for NRMSE was established at 35.8%, and between 89.7%

220

and 93.6% of formulae in the actual datasets passed this criterion.

221

To verify that our datasets are based on non-random time series of intensities we

222

compared the distribution of kexp in our datasets to the distribution of kexp derived from

223

an analogous but artificial dataset based on random numbers. The distribution of kexp

224

in the actual datasets was more narrow and skewed towards positive values (SI Figure

225

S5), as was also indicated by the non-parametric Mann-Whitney U test (P

Emergence of the Reactivity Continuum of Organic Matter from

Recommend Documents