Fitting a pre-established region-specific Parallel Factor Analysis (PARAFAC) model to new dissolved organic matter (DOM) samples has become a popular approach in DOM studies. A key step during the development of the pre-established model is to identify the fluorescence fingerprint, i.e. the number of fluorescent DOM (FDOM) components as well as their excitation and emission spectral features. In practice the samples to use for model development may not be measured immediately after sampling and will instead be stored for a relatively long time, thus raising the question whether the storage would change the intrinsic fingerprints. In this study, two PARAFAC models were separately developed and validated for the same set of surface water DOM samples from an estuary and its river, immediately after sampling and after 9-month storage respectively. The fingerprint did not change after storage, regardless of the change of the intensity of some components. The findings in this study highlighted that DOM samples stored using a simple protocol (i.e. filtration to 0.45 μm pore size without any preservatives and placed at 4 °C in the dark under airtight conditions) for a few months would not change fluorescence fingerprints for PARAFAC and broaden our understanding of the application of PARAFAC to DOM studies.

INTRODUCTION

Dissolved organic matter (DOM) is ubiquitous in the aquatic environment and plays an essential role in numerous biogeochemical processes (e.g. carbon cycling and metal bioavailability) important to ecosystems. Moreover, DOM is a big concern in drinking water treatment (Shutova et al. 2014). The biogeochemical effects of DOM on the natural water environment and drinking water treatment efficiency are tightly dependent on DOM quantity and quality, which are dependent on DOM sources. Extensive studies have been conducted to report the quantity and quality in different geographical regions or hydrological periods (Stedmon & Markager 2005; Chen et al. 2010; Singh et al. 2013; Hosen et al. 2014; Singh et al. 2014). To this end, a multivariate chemometric tool Parallel Factor Analysis (PARAFAC) applied to data rich fluorescence excitation-emission matrices (EEMs) of DOM samples has become one of the most popular approaches (Murphy et al. 2013). The input data for a PARAFAC fitting is a set of fluorescence excitation emission matrices (EEMs) corresponding to a set of samples. PARAFAC decomposes the EEMs into a set of underlying fluorescent DOM (FDOM) components, reporting the scores (proportional to the components concentrations or intensity in the samples) and the loadings (i.e. the features of the excitation and emission spectra independent of the samples) of each component (Murphy et al. 2013).

It is accepted that fitting a pre-established PARAFAC model to new DOM samples is justified under two circumstances (Larsen et al. 2010; Singh et al. 2013; Yamashita et al. 2013; Romera-Castillo et al. 2014): (1) the set of DOM samples used for developing the pre-established PARAFAC model and the later new set of DOM samples both come from the same geographical region under the same hydrological period or season; and (2) the number of new samples is too limited to develop a statistically robust new model. Therefore, it is useful to pre-establish a validated PARAFAC model based on DOM samples collected earlier. A key step is to identify the crucial information on the FDOM components, namely the number of the components and their spectral features in the pre-established model. Given the crucial information, the fluorescence intensity of the same set of components in the new DOM samples will be readily revealed and can be used to trace changes of FDOM component abundance.

For DOM fluorescence measurement after short-term storage (e.g. a few days), DOM samples are usually filtered through 0.45 μm pore size filters and stored in bottles sealed tight without headspace at 2–4 °C in the dark in the absence of preservative. Such storage may be practically important for several reasons: (1) filtration through 0.45 μm pore size filters is consistent with the operational definition for DOM (Zsolnay 2003); (2) DOM fluorescence will alter due to DOM complexation with heavy metal ions (e.g. Hg2+) or pH variation (Lu & Jaffe 2001; Patel-Sorrentino et al. 2002; Chen & Kenny 2007), and thus preservation of DOM samples by adding mercuric ion or by lowering pH with mineral acids (e.g. sulfuric acid) is not recommended. Moreover, precipitation and/or hydrolysis of DOM due to low pH is also a concern; and (3) freeze-and-thaw of DOM samples may result in fluorescence change difficult to interpret and is not recommended for samples subject to fluorescence measurement (Spencer et al. 2007).

In practice, EEMs of DOM samples for use in developing a pre-established PARAFAC model may not be measured shortly after sampling for numerous practical reasons, e.g. instrumental failure and schedule conflict for the use of instruments. Therefore, storage of the filtered samples for a relatively long time may be unavoidable. As a matter of fact, a significant proportion of microbes may pass through 0.45 μm pore size filters (Wang et al. 2007). Even filtration through 0.22 μm pore size filters, the most widely accepted sterile filtration approach still permits a considerable amount of filterable microbes to remain in the filtrate (Wang et al. 2007). Moreover, abiotic processes such as oxidation/reduction or association/dissociation may take place among DOM components. Therefore, it is commonly believed, but not proven, that long-term stored DOM samples should be discarded because their fluorescence fingerprints will be significantly different from the original samples or shortly stored samples.

It is worth verifying whether such long-term storage will result in complete removal of some FDOM components or production of some new components or change of the spectral features of some components. Provided the crucial information remains unchanged, the long-term stored samples would still be useful in PARAFAC modeling regardless of the change of the FDOM components intensity. This is an important practical concern for DOM studies. To our knowledge, this issue has not been explicitly addressed to date. The present study aims to test a long-term storage effect on FDOM components by comparison between two PARAFAC models for the same set of surface water DOM samples measured immediately after sampling and measured after 9-month storage respectively. The samples were anticipated to contain DOM components originating from different sources (e.g. soil-derived vs. aqueous microbe-derived) and distributed in different aqueous environments (e.g. freshwater vs. seawater), and thus the test would be of practical relevance.

SAMPLING AND EXPERIMENTAL

Daliao River is one of the biggest rivers in Northeast China with a total length of 95 km, stretching from the confluence point of two rivers in the upstream to the estuary. It has an annual freshwater discharge of 3.95 × 109 m3 (Li et al. 2013). More than 50% of the annual freshwater discharge occurs during July and August. Forty surface water samples were collected from Daliao River Estuary and its river in August 2013 (Figure 1 and Figure S1 in the Supplementary Information (available in the online version of this paper)). High density polyethylene (HDPE) bottles were rinsed three times by surface water before actual sampling took place.
Figure 1

Sampling sites in the Daliao River and its estuary.

Figure 1

Sampling sites in the Daliao River and its estuary.

See Table S1 (available in the online version of this paper) for a summary of the experimental design. The collected water samples were brought back on ice to a laboratory for immediate filtration through 0.45-μm pore size nylon membranes (Millipore). For each of the 40 samples, portions (∼50 ml) of the filtrate were immediately subject to the optical measurements and dissolved organic carbon (DOC) measurement, and the rest was stored in air-tight HDPE bottles with no headspace at 4 °C in the dark for 9 months prior to the next round of optical and DOC measurement.

DOC was determined by a total organic carbon (TOC) analyzer (Jena N/C 2100S) in which the filtrate was first acidified to pH 2 by HCl and then purged by high purity (99.999%) oxygen to remove the carbon dioxide, followed by high-temperature (850 °C) combustion of the remaining organic carbon under the catalysis of CeO2 to generate CO2, which was quantified by an infrared detector.

Optical measurements were all conducted at 22 °C. UV-visible absorption (200–800 nm) spectra were measured using a Shimadzu 1,800 spectrophotometer with a 1 cm quartz cuvette. A double-beam mode was employed with pure water as the absorption blank. Following Green & Blough (1994), absorbance values were corrected for instrument baseline drift, refractive index, and temperature variations and converted to Napierian absorption coefficients based on the formula: 
formula
where A is absorbance and L is path length (1 cm).

Fluorescence was measured on a Hitachi F-4500 fluorometer. Excitation and emission wavelengths were scanned both with 3-nm steps, from 246 to 498 nm and from 280 nm to 598 nm, respectively. The scan rate was 2400 nm min−1. The band-widths of the excitation and emission monochromators were both set to be 5 nm. The voltage of the PMT detector was set to be 700 V.

Scattering light was mathematically removed from the excitation–emission matrix (EEM) by placing NaN (not-a-number) on the scattering region (Chen & Kenny 2007; Chen et al. 2011). Fluorescence intensities in the scattering-removed EEM were corrected for inner-filter effects (IFEs) based on the approach employed in our earlier studies (Chen & Kenny 2007; Chen et al. 2011; Chen et al. 2013).

Four widely applied optical indices were calculated. Based on the Napierian absorption coefficients, the spectral slope S275–295 was calculated, as described by Helms et al. (2008). Specific absorbance at 254 nm, referred to as SUVA254, was computed by dividing absorbance at 254 nm by DOC (Weishaar et al. 2003). Humification index (HIX) was calculated by dividing fluorescence emission intensity at 435–480 nm by that at 435–480 nm and 300–345 nm at an excitation wavelength of 254 nm (Ohno 2002). Biological index (BIX) was calculated as the ratio of emission at 380 nm to that at 430 nm at an excitation wavelength of 310 nm (Huguet et al. 2009).

PARAFAC modeling was achieved using a PLS 3.0 tool package running in the software of Matlab 7.1. Normalization pre-treatment was employed to make the sum of squares of any EEM the same (i.e. equal to unity) prior to PARAFAC modeling; outlying samples as shown by their large leverages should be removed before the correct PARAFAC model is finally established (Murphy et al. 2013). The appropriateness of a model was evaluated by a combinational check of modeling residuals, the spectral features of the resolved components as well as split-half analysis (Stedmon & Bro 2008; Murphy et al. 2013). The split-half analysis was conducted by dividing the samples into four quarters (A, B, C and D). Four halves of EEMs were constructed: AB, CD, AD and BC. PARAFAC modeling of AB and AD was compared to that of CD and BC, respectively, the practice of which was referred to as the S4C4T2 validation test (Murphy et al. 2013). After the correct model was established, the normalization was reversed following Murphy et al. (2013).

The lamp intensity variation of the fluorometer was checked daily by monitoring pure water Raman scattering band intensity excited by light of 350 nm. Fluorescence intensities of the identified FDOM components at their ex/em maxima were divided by the under-peak area of the pure water Raman scattering band (Lawaetz & Stedmon 2009; Chen et al. 2013). Such normalization facilitates comparison among samples (Stedmon & Bro 2008).

Statistical analyses other than PARAFAC modeling were conducted in the software package of R (Team 2014). The similarity between two spectra (e.g. emission spectra) of two similar components was computed using the classical spectral contrast angle method in which the spectra were represented as vectors with direction (Wan et al. 2002). If two vectors of the same length (or normalized to the same length) are very similar, they would overlap significantly and thus the angle θ between the two vectors would approach zero, giving cosine (θ) approaching one (Wan et al. 2002). The paired t-test was employed to check whether a component's intensity varied significantly after the storage.

RESULTS AND DISCUSSION

The split-half diagnostic (not plotted here) indicated that six components (C1–C6) were identified for initial samples and stored samples, respectively. The excitation and emission spectra of the components are shown in Figure 2. The excitation and emission maxima wavelength for the components were: C1 (ex/em = ∼279/310 nm), C2 (ex/em = ∼303/340 nm), C3 (ex/em = ∼310/380 nm), C4 (ex/em = ∼320/425 nm), C5 (ex/em = ∼385/475 nm) and C6 (ex/em = ∼/430 nm, no obvious excitation maxima). Note that C6 did not show evident excitation maxima within the wavelength range 246–498 nm. The PARAFAC components in the two models were similar to those reported previously (Cory & McKnight 2005; Murphy et al. 2008; Fellman et al. 2009; Stubbins et al. 2014). C1 and C2 resmbled tyrosine-like and tryptophan-like FDOM, respectively. C3–C6 resembled fluorescence features of fulvic acid or humic acid.
Figure 2

The excitation (ex) and emission (em) spectra of C1–C6 immediately after water sampling (blue) and after 9-month storage (red) (the full colour version of this figure is available in the online version of this paper.

Figure 2

The excitation (ex) and emission (em) spectra of C1–C6 immediately after water sampling (blue) and after 9-month storage (red) (the full colour version of this figure is available in the online version of this paper.

After storage, the mean intensity of a component increased for C1, C3 and C4 (p < 0.05), decreased for C6 (p < 0.05) and did not change for C2 and C5 (p > 0.05). As shown in Figures S2–S7 (available in the online version of this paper), the change of a component was highly variable depending on the samples, suggesting that the micro-environments were different among the samples which were collected from different sampling sites. Increase or decrease of FDOM intensities during DOM sample storage were also observed in other DOM studies (Moran et al. 2000; Asmala et al. 2014). Note that some sampling points showed much higher intensities of C2 than other components (Figure S3) indicating a strong production of C2 at these sampling sites.

Despite the change of fluorescence intensities for some components, all the components C1–C6 in the initial samples were spectrally similar to those in the stored samples, as indicated by the high values of cosine (θ) ranging between 0.990 and 0.999. The generation of new types of FDOM components other than C1–C6 during storage was not observed. This is because any filterable microbes remaining in the solutions were largely those which had already been there in the natural water columns before water sampling. Since the same microbes had already been there in the natural water samples before water sampling, it is rational that any FDOM component generated by the same microbes during storage showed similar or the same spectra as that in the initial samples before storage. New FDOM might be produced by new microbes which were accidently introduced into the water samples during water filtration, but such contamination could be avoided by careful experimentation.

The complete removal of a FDOM component from any of the water samples during storage was also not observed. Although a complete removal of one or some FDOM components from some samples after storage would be possible, it is less likely to occur to all the samples; this statement is particularly true for samples collected from diverse aquatic environments (e.g. freshwater vs. seawater) showing a variety of FDOM component intensities. The complete removal of one or multiple FDOM components from all stored samples might occur only under certain circumstances, e.g. that all samples contain equally low concentrations of some components which are readily consumed by microbes given enough storage time and adequate nutrients for the microbes. In this study, no sample entirely lost any of the six FDOM components after storage. Provided the loss had occurred to some FDOM components in some samples, a PARAFAC modeling of the whole sample data set would still be able to identify the same set of FDOM components. This is a basic merit of PARAFAC modeling. That is to say, PARAFAC does not require that each sample contain all the FDOM components before PARAFAC can identify the FDOM components from the multiple samples.

Supplementary to the PARAFAC results, the two fluorescence-based indices, i.e. HIX and BIX did not change (p > 0.05, t-test) after storage (Figure S8, available in the online version of this paper). Therefore, these two indices were stable over the storage, and the post-storage fingerprints could still be applicable to trace back the initial humification status (based on HIX) or the initial contribution of biological processes (based on BIX) of the initial samples.

The storage decreased the mean absorbance of the 40 samples, but the decrease was not uniform across the wavelength, as illustrated by the absorbance decrease at several typical wavelengths of 254, 280, 350, 400, 450 and 498 nm (Figure S9(a)). The absolute decrease was greater with shorter wavelength, but the relative decrease showed an opposite trend (Figure S9(a)). (Figure S9 is available in the online version of this paper.) Similar preferential removal of long-wavelength-absorbing chromophores was also observed in the bioassays of Asmala et al. (2014).

The spectral slope S275–295 ranged between 0.0135 and 0.0223 nm−1 with a mean of 0.0188 nm−1 before sample storage. After storage, the spectral slope ranged between 0.0135 and 0.0233 nm−1 with a mean of 0.0194 nm−1. The spectral slope after storage (y) could be linearly correlated to that before the storage (x) in a fitting function y = 1.031x (p < 0.05). The increase of S275–295 after storage indicated a shift from higher molecular weight (MW) to lower MW of DOM, as S275–295 was found to be inversely correlated to the MW of DOM (Helms et al. 2008). The shift to lower MW was supported by the decreasing SUVA254 (Figure S9(b)) which has proven to be a reliable indicator for mean aromaticity of DOM (Weishaar et al. 2003) and was observed to increase with increasing MW of DOM (Porcal et al. 2013).

The above absorption-related observations fit well in the framework of the charge-transfer between electron donors (D) and acceptors (A) proposed for UV-Vis absorption of HS (Del Vecchio & Blough 2004; Sharpless & Blough 2014). According to that model, decrease of MW would preferentially decrease the long-wavelength (i.e. > 350 nm) absorption originating from the D-A pairs, but increase the short-wavelength absorption slopes. However, the shift from higher MW to lower MW was not accompanied by evident DOC change (p > 0.05, paired t-test).

CONCLUSION

This study observed that 9-month storage showed no difference on PARAFAC-modeled excitation or emission spectra of six FDOM components identified in surface water samples collected from Daliao River and its estuary. This suggests that for the purpose of developing a PARAFAC model to use for later DOM samples, the simple storage protocol as deployed in this study is promising. Moreover, two fluorescence-based indices, i.e. HIX and BIX were stable over the storage, indicating the possibility of tracing the original states of the samples based on the post-storage fluorescence fingerprints. The findings of this study broaden our understanding in application of PARAFAC to DOM studies. More work to verify the findings for DOM samples in other geographical regions or hydrological periods may be solicited.

ACKNOWLEDGEMENT

The work was supported by an intramural funding of the Chinese Research Academy of Environmental Sciences (project no. GYK5091302).

REFERENCES

REFERENCES
Asmala
E.
Autio
R.
Kaartokallio
H.
Stedmon
C. A.
Thomas
D.
2014
Processing of humic-rich riverine dissolved organic matter by estuarine bacteria: effects of predegradation and inorganic nutrients
.
Aquatic Sci
.
76
,
451
463
.
Chen
H.
Kenny
J. E.
2007
A study of pH effects on humic substances using chemometric analysis of excitation-emission matrices
.
Annals Environ. Sci
.
1
,
1
9
.
Del Vecchio
R.
Blough
N. V.
2004
On the origin of the optical properties of humic substances
.
Environ. Sci. Technol
.
38
(
14
),
3885
3891
.
Fellman
J. B.
Miller
M. P.
Cory
R. M.
D'Amore
D. V.
White
D.
2009
Characterizing dissolved organic matter using PARAFAC modeling of fluorescence spectroscopy: a comparison of two models
.
Environ. Sci. Technol
.
43
(
16
),
6228
6234
.
Hosen
J. D.
McDonough
O. T.
Febria
C. M.
Palmer
M. A.
2014
Dissolved organic matter quality and bioavailability changes across an urbanization gradient in headwater streams
.
Environ. Sci. Technol
.
48
(
14
),
7817
7824
.
Huguet
A.
Vacher
L.
Relexans
S.
Saubusse
S.
Froidefond
J.-M.
Parlanti
E.
2009
Properties of fluorescent dissolved organic matter in the Gironde Estuary
.
Org. Geochem.
40
(
6
),
706
719
.
Larsen
L. G.
Aiken
G. R.
Harvey
J. W.
Noe
G. B.
Crimaldi
J. P.
2010
Using fluorescence spectroscopy to trace seasonal DOM dynamics, disturbance effects, and hydrologic transport in the Florida Everglades
.
J. Geophys. Res. Biogeo.
(2005–2012)
115
,
G03001
, doi:10.1029/2009JG001140.
Murphy
K. R.
Stedmon
C. A.
Graeber
D.
Bro
R.
2013
Fluorescence spectroscopy and multi-way techniques
.
PARAFAC. Anal. Methods
5
(
23
),
6557
6566
.
Porcal
P.
Dillon
P. J.
Molot
L. A.
2013
Seasonal changes in photochemical properties of dissolved organic matter
.
Biogeosciences Discussions
10
(
3
),
5977
6006
.
Sharpless
C. M.
Blough
N. V.
2014
The importance of charge-transfer interactions in determining chromophoric dissolved organic matter (CDOM) optical and photochemical properties
.
Environ. Sci: Processes & Impacts
16
(
4
),
654
671
.
Stubbins
A.
Lapierre
J.-F.
Berggren
M.
Prairie
Y.
Dittmar
T.
Del Giorgio
P.
2014
What's in an EEM? molecular signatures associated with dissolved organic fluorescence in Boreal Canada
.
Environ. Sci. Technol
.
48
(
18
),
10598
10606
.
Team
2014
R: A Language and Environment for Statistical Computing
. R Foundation for Statistical Computing,
Vienna
,
Austria
. .
Wan
K. X.
Vidavsky
I.
Gross
M. L.
2002
Comparing similar spectra: from similarity index to spectral contrast angle
.
J. Am. Soc. Mass. Spectr
.
13
(
1
),
85
88
.
Weishaar
J. L.
Aiken
G. R.
Bergamaschi
B. A.
Fram
M. S.
Fujii
R.
Mopper
K.
2003
Evaluation of specific ultraviolet absorbance as an indicator of the chemical composition and reactivity of dissolved organic carbon
.
Environ. Sci. Technol
.
37
(
20
),
4702
4708
.

Supplementary data