Professional Documents
Culture Documents
ScienceDirect
journal homepage: www.elsevier.com/locate/watres
Environmental and Life Sciences Graduate Program, Trent University, ON, Canada
Chemistry Department, Trent University, ON K9J 7B8, Canada
article info
abstract
Article history:
Relationships between the molecular weight (MW) and fluorescence properties of dissolved
organic matter (DOM) are important considerations for studies seeking to connect these
properties to water treatment processes. Relationships between the size and fluorescence
2 October 2014
properties of nine allochthonous DOM sources (i.e. leaf leachates, grass, and headwaters)
were measured using asymmetrical flow field-flow fractionation (AF4) with on-line
Keywords:
PARAFAC-EEM
highest at mid-weights (0.5e1 kDa), and humic-like fluorescence (Peaks A C) was highest
at larger molecular weights (>1 kDa). Proportions of peaks B, T, and A C were significantly
(MWD)
correlated with MW (p < 0.001). The first principal component (PC1, 42% of variation in
fluorescence properties) was a significant predictor of sample MW (R2 0.63, p < 0.05),
while scores on PC2 (27% of total variance) traced a source-based gradient from deciduous
tionation (AF4)
Supramolecular assemblies
also correlated with MW (p < 0.005). A secondary peak in peak T fluorescence was associated with larger size fractions in aged sources, and scores on PC1 also traced a path from
the leachates of fresher leaves, through more humified leaves, to headwaters. Findings are
consistent with the hypothesis that the structure of aged DOM arises through supramolecular assembly.
2014 Elsevier Ltd. All rights reserved.
1.
Introduction
* Corresponding author. Tel.: 1 705 748 1011; fax: 1 705 748 1625.
guen).
E-mail address: celinegueguen@trentu.ca (C. Gue
http://dx.doi.org/10.1016/j.watres.2014.10.013
0043-1354/ 2014 Elsevier Ltd. All rights reserved.
488
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
2.
2.1.
DOM samples
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
2.2.
AF4-DAD-EEM
The methods and apparatus used for on-line size fractionation with UVeVisible and fluorescence detection are outlined
guen and Cuss, 2011). Two
in detail in an earlier study (Gue
millilitres of each sample was injected into the 0.3-mL sample
loop of an asymmetrical flow field-flow fractionator (AF2000
Focus; Postnova Analytics), equipped with a polyethersulfone
membrane (300-Da molecular weight cut-off; Postnova), and
Origin
Mean Mp 95%
CI (kDa)
0.42 0.03
0.50 0.04
0.48 0.03
0.43 0.03
1.05 0.04
1.12 0.04
1.48 0.13
1.71 0.06
2.06 0.12
489
490
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
their optimization. The SOM algorithm was applied to proportional PARAFAC loadings and log-transformed SFMW
using the SOM toolbox in Matlab R2010a by adapting the
tutorial of Bieroza et al. (2012b). To eliminate, source-specific
clustering caused by differences in the fluorescence compoguen, 2013), component
sition of leaf leachates (Cuss and Gue
proportions and log-SFMWs were standardized on [0e1] by
source according to the following formula:
N
2.3.
Data analysis
The number, identities, and relative concentrations of independent fluorophores in EEMs were assessed using parallel
factor analysis (PARAFAC) in Matlab R2010a (MathWorks) according to the procedure outlined in the DOMFluor toolbox
(Stedmon and Bro, 2008). A total of 81 EEMs were preprocessed by subtracting the EEM of Milli-Q water measured
on the same day, and normalizing the fluorescence intensities
to Raman units (r.u.) using the area under the Raman peak of
the Milli-Q EEM at an excitation wavelength of 350 nm
(Lawaetz and Stedmon, 2009). Excitation wavelengths
<210 nm were removed due to a low signal/noise ratio. Blank
subtraction, Raman normalization, and the removal of scatter
and wavelengths <210 nm were conducted using an in-house
modified, instrument-optimized version of the EEMCut algo guen, 2012a). To assess the relationship
rithm (Cuss and Gue
between molecular weight and total fluorescence composition, the loadings of PARAFAC components were transformed
into proportions of total fluorescence (i.e. loading of each
component divided by the sum of loadings for all components
in the same EEM).
Kohonen's self-organizing maps (SOM) (Kohonen, 2001) are
a type of artificial neural network originally developed for
pattern recognition in artificial intelligence research. Due to
their excellent visualization capabilities, SOM are increasingly
employed in situations involving complex data distributions,
structures, and interrelationships. For example, SOM have
been used to assess the optical properties of DOM removed in
water treatment (Bieroza et al., 2012a), to detect diesel
contamination by distinguishing DOM fluorescence (Carstea
et al., 2010), and in several diverse water resource applications (Kalteh et al., 2008, and references therein). Please refer
to the supplementary information for details about SOM and
X minX
maxX minX
(1)
3.
3.1.
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
Excitation
Emission
Peak
<250 (279)
<250 (326)
<250 (276)
265 (382)
<250 (315)
258 (346)
(<250) 329
315
420
340
470
430
395
460
B
A
T
AC
M
N*
C
the tails of MWDs (i.e. SWS and SFMW 7.72 kDa; SRM, 3.11 kDa;
SSM, 2.87 kDa; STL, 2.33 kDa; SRC, 16.2 and 28.5 kDa) had
loadings of zero on all PARAFAC components, indicating that
fluorescence was below the detection limit. These EEMs were
removed prior to subsequent analysis. The spectra of PARAFAC components were similar to those widely observed in
water samples (Coble et al., 1998; Stedmon and Markager,
2005a,b; Fellman et al., 2008), leaf leachates (Wickland et al.,
2007; Fellman et al., 2010; Beggs and Summers, 2011; Cuss
guen, 2012a, 2012b; 2013), and products of microbial
and Gue
activity (Li et al., 2008; Andrade-Eiroa et al., 2013). Component
C6 (Ex/Em 258(346)/395 nm) was similar to peak N (Ex/
Em 280/370 nm; Coble et al., 1998), but was blue-shifted in
excitation and red-shifted in emission wavelengths. Hence,
C6 has been named peak N* herein to emphasize these
differences.
Fluorescence compositions at Mp were distinct by litter species and age (Fig. 2), in agreement with prior studies of leaf
guen, 2013; Wong
leachates (Wickland et al., 2007; Cuss and Gue
and Williams, 2010). Senescent leachates were dominated by
components C1 and C3 excepting SRM, which was dominated by
C5. The DOM samples isolated from headwaters (CHW, DHW,
WET) were dominated by C2 and C4. Component C1 was absent
from WET and C6 was absent from CHW, while DHW contained
small amounts of both components. Leachates SRC and OSM
were also dominated by C2 and C4, but were distinguished from
headwaters by greater amounts of C1, C3, and C6.
3.2.
Self-organizing map
Relationships between fluorescence composition and molecular weight were clearly and vibrantly demonstrated by the
491
SOM (Fig. 3). Each hexagon is labelled with the sources and
size fractions that were the most similar to the corresponding
map vector (Fig. 3A), alongside the resulting distributions of
source-normalized PARAFAC component proportions and
size-fraction molecular weight (SFMW; Fig. 3B). The lowest
proportion of each component and molecular weight for each
DOM source are coloured blue, and the highest values are red.
Thus, source EEMs measured at the lowest MW are clustered
in the bottom left corner of the map (i.e. '-2'), while those
measured at the highest MW are in the upper right corner (i.e.
'-10'). The relative amount of each PARAFAC component for
each source and size fraction can be assessed by visually
overlaying the hexagon containing the EEM of interest in
Fig. 3A and the SOM components of interest in Fig. 3B. For
example, EEMs SSM-2 and SSM-9, (lower left corner and upper
right corner of Fig. 3A), respectively contained the highest and
lowest amounts of C3 in sample SSM because the lower left
corner of the map for C3 is dark red, and the upper right corner
is dark blue. In kind, visually overlaying the variable distribution plots for SFMW, C1, C3, and C4 in Fig. 3B reveals that
PARAFAC components C1 and C3 were negatively correlated
with molecular weight, while C4 was positively correlated
(p < 0.001 for all).
A shoulder in the middle of the distribution map for
component C3 was associated with distributional differences
for two groups of leachates (Fig. 3B, circled). Despite consistent decrease with increasing MW, the proportion of C3
remained relatively high at all MW in the first group (SSM, STL,
SWS; Mp < 0.50 kDa for all), so that its normalized value
remained high even at higher MW (e.g. at MW 1.1 kDa, SSM7 had normalized MW and C3 0.53 and 0.84, respectively). In
the second group of larger and more aged DOM (headwaters,
OSM, SRC; Mp > 1.0 kDa for all), the central shoulder corresponded with a secondary maximum in the proportion of C3
in the range of 1.3e2.5 kDa (e.g. at MW 2.4 kDa, SRC-7 had
normalized MW and C3 0.32 and 0.50, respectively). The
EEMs of the second group were distributed around the edges
of the map as C3 initially decreased with increasing MW,
rather than running through the centre like group 1. The
second group then re-converged towards the centre of the
map for 1e2 EEMs that were associated with the secondary
maximum of C3 (see also animations in Figure S4). Peak T
fluorescence similar to C3 has been previously associated with
aged DOM of different molecular weights in wetland,
Fig. 2 e FDOM compositions at sample Mp, described in terms of PARAFAC components C1-7 (darkest e lightest). Source
abbreviations are shown in Table 1.
492
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
Fig. 3 e Self-organizing map showing distributions of (A) size fractions, and (B) corresponding molecular weights (SFMW)
and proportions of PARAFAC components C1 e C7 normalized by source: red denotes the largest proportion in each source,
and blue the lowest. Molecular weight increases by number (i.e. lowest SFMW for each source is -2 and highest is -10).
Source abbreviations are shown in Table 1. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article).
estuarine, and marine systems using size exclusion chromatography (Maie et al., 2007; Romera-Castillo et al., 2014), cross
flow ultrafiltration (Liu et al., 2007), and AF4 (Stolpe et al.,
2014). It has been suggested that this secondary maximum
in peak T fluorescence may arise as a consequence of protein
encapsulation by humic substances in the molecular aggregation that follows the diagenetic processing of fresh DOM
(Wershaw, 2004; Sutton and Sposito, 2005; Tomaszewski et al.,
2011; Romera-Castillo et al., 2014). The differences observed in
the distribution of C3 between fresh leachates and larger,
DOM sources dominated by peaks A, C, and A C also supports this claim.
The distributions of PARAFAC components C2 and C6 were
bimodal, with maxima in the map corners (Fig. 3B). Accordingly, the SFMWs of sources dominated by C2 (CHW, DHW,
SRC, WET) increased around the map edges from the bottomleft to the upper-right corner, distinguishing them from the
peak-T-dominated sources in the middle (Fig. 3A, B). The
maxima of C2 were thus associated with size fractions of intermediate molecular weights (~1 kDa). One maximum of
component C6 (i.e. in the upper right corner of the map)
exhibited significant positive correlation with MW for SSM,
STL, and SRC (p < 0.001), and exhibited an increasing trend
with OSM, SRM, and SWS. The other maximum of C6 was
negatively correlated with MW for the wetland sample (WET;
p < 0.001), but composed only 5% of the fluorescence
signature in all WET EEMs.
Components C5 and C7 also exhibited source-dependent
trends, as suggested by their partitioned distributions
(Fig. 3B). C5 was negatively correlated with the MWs of SWS
(p < 0.05) and SRC (p < 0.005), and exhibited maxima at medial
MWs for SRM, STL, SSM, and CHW. Component C7 was positively correlated with MW for SSM and SRC, and negatively
correlated for WET (p < 0.001 for all). Component C5 has been
previously associated with the mid-ranged size fraction cenguen, 2012b).
tred at 1.6 kDa in leaf leachates (Cuss and Gue
However, this fluorophore has also been associated with a size
fraction <0.50 kDa in fresh and marine waters using ultrafiltration (Huguet et al., 2010).
In summary, the relationship between MW and the proportion of PARAFAC components was consistent across all
DOM sources for four of seven components: C1 and C3 were
significantly negatively correlated with MW and C4 was
significantly positively correlated, while C2 exhibited maxima
at medial MW. However, the significant correlation between
C3 and MW did not detect the important small secondary
maximum associated with more aged, larger DOM sources. In
addition to visualizing significant correlations and consistent
relationships with MW, the secondary maximum of C3 was
revealed when the SOM was applied to component loadings.
This illustrates the advanced pattern detection capabilities of
SOM, and its ability to visualize non-linear relationships between variables. SOM visualization also illustrated the
inconsistent relationships between MW and proportions of
C5, C6, and C7, which were significant for some DOM sources.
3.3.
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
493
Fig. 4 e Biplot and component loadings from principal component analysis of size-fraction fluorescence compositions: A
senescent red maple (SRM); - senescent tamarack leachate (STL); : senescent silver maple (SSM); deciduous headwater
(DHW); over-wintered silver maple (OSM); , wetland (WW); senescent reed canarygrass (SRC); C senescent white
spruce (SWS); B coniferous headwater (CHW). Major loadings (magnitude 0.20) are shown in bold type.
were positive for C1, C3, and C6, and negative for C2 and C4
(Fig. 4). These loadings reflect the structure and distributions
of variables and samples discussed for the SOM. In PC2 (27% of
variance), C5 and C6 had negative loadings, while those of C2,
C3, and C7 were positive. In PC3 (13% of variance), C4, C6, and
C7 were positively loaded, and were balanced by C1, C2, and
C5. Considered over all EEMs, scores on PC1 and PC3 were
negatively (p < 0.001) and positively (p < 0.005) correlated with
SFMW, respectively, while no such relationship was apparent
for PC2 (p > 0.05). Scores were clustered by source in the biplot
of PC1 and PC2, and also showed trends in MW and source,
respectively (Fig. 4).
Size-dependent trends were investigated further by producing scores for the fluorescence properties of samples at Mp
using the loadings on PC1. Scores thus produced were a significant predictor of Mp (R2 0.63, p < 0.05; Fig. 5A). A very
similar relationship has been previously observed for a
different collection of whole samples and fractionated leaf
guen, 2013), suggesting that fluoresleachates (Cuss and Gue
cence composition may be a useful estimator of DOM molecular weight. Scores on PC1 also separated aged headwater
sources from fresh leachates owing to the relatively low proportions of C1 and C3 (peaks B and T) in the former, and the
relatively low proportions of C2 and C4 (peaks A and A C) in
the latter (Fig. 5B). These sources were therefore also separated along a continuum of age/humification, since SRM was
collected from the ground beneath the tree, SRC was highly
browned when it was collected, and the other leaves/needles
Fig. 5 e Relationships for latent variables from PCA of PARAFAC loadings from EEMs of size fractions: (A) PC1 scores as
predictors of Mp for different DOM types, and (B) PC2 scores at Mp by DOM origin. A senescent red maple (SRM); - senescent
tamarack leachate (STL); : senescent silver maple (SSM); deciduous headwater (DHW); over-wintered silver maple
(OSM); , wetland (WW); senescent reed canarygrass (SRC); C senescent white spruce (SWS); B coniferous headwater
(CHW).
494
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
were either collected directly from the tree or shaken from the
branches (Table 1). Thus, as DOM aged and increased in MW,
C1 and C3 decreased while C2 and C4 increased. Similar patterns of change have been previously associated with very
similar PARAFAC components during the photo- and biotransformation of tannins and plant biomass leachates
, 2014). Additionally, Maie
(Maie et al., 2008; Chen and Jaffe
et al. (2008) found that leaf tannins exhibited progressively
less fluorescence in regions associated with proteins and
polyphenols (peaks B and T), and more fluorescence in regions
associated with humic material (peaks A and C) during photoand bio-transformation, suggesting that polyphenols were
converted into humic material as DOM ages. Although the use
of distinct fresh and aged sources in this study prevented the
tracking of fluorescence conversions, the common relationships between PARAFAC components in PC1 and the negative
correlation between PC1 scores and DOM age suggest that
these changes in fluorescence may be universally related to
DOM aging and concomitant molecular assembly. Scores on
PC2 traced a gradient from freshly leached deciduous DOM
through to the coniferous headwater (Fig. 5B). Interestingly,
scores for coniferous leachate (SWS) were the most similar to
those of coniferous headwater (CHW, and the loadings for
deciduous leachates (SRM, SSM, STL) were the most similar to
that of deciduous headwater DHW. The wetland water sample
(WW) was central in this gradient, bracketed by the grass and
overwintered leachate (SRC, OSM, respectively). Despite significant correlation between MW and scores on PC3, PC3 was
not linearly related to MW and no relationship was evident
between scores and source. Thus, the latent variables that
were primarily MW/age-related (PC1 and PC3) explained 55%
of the observed variation in optical properties, while variation
that was primarily attributable to source (PC2) accounted for
27%.
3.4.
Hur et al., 2012), and the DBPFP of larger DOM is higher than
that of smaller DOM (Hur et al., 2012). However, peak B and T
fluorescence that is present in relatively large DOM and
encapsulated in humic substances (i.e. following biological
processing) may not impact DBPFP in the same fashion as
exposed proteins, so that protein-like content alone may not
reflect DBPFP. Additionally, the source of protein-like fluorescence is not entirely clear because the Ex/Em region associated with proteins is also associated with smaller
polyphenolics such as tannins that are abundant in freshly
leached material (Aiken, 2014). Indeed, it has been demonstrated that the leachates of fresh litter and older duff from
the same species possess different DBPFPs (Chow et al., 2009;
Beggs and Summers, 2011; Hur et al., 2012); however, in these
studies the DBPFP also differed by species and leachate pretreatment (i.e. exposure to ambient levels of microbes or
sunlight vs. freshly leached). Given the uncertain source and
contribution of protein-like fluorescence, the measurement of
both size and optical properties may be necessary for an accurate assessment of DBPFP.
4.
Conclusions
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
Acknowledgements
The authors thank S. Watmough and K. Pinder for assistance
with field sampling at Dorset, and A. McDonough for assistance with the classification of plant species. This work was
funded in part by the Canada Research Chairs program and
the Canadian National Science and Engineering Research
Council (CG). CWC gratefully acknowledges the financial
support accorded by the Canadian Graduate Scholarship
(NSERC). Finally, we thank two reviewers who provided
constructive comments that helped improve our manuscript.
references
495
496
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
w a t e r r e s e a r c h 6 8 ( 2 0 1 5 ) 4 8 7 e4 9 7
497