Spatiotemporal Variations in Seston C:N:P Stoichiometry in a Large Eutrophic Floodplain Lake (Lake Taihu): Do the Sources of Seston Explain Stoichiometric Flexibility?

: Although sources of seston are much more complicated in lakes compared to oceans, the inﬂuences of di ﬀ erent sources on the spatiotemporal variations in seston stoichiometry are still underexplored, especially in large eutrophic ﬂoodplain lakes. Here, we investigated seston stoichiometric ratios across a typical large eutrophic ﬂoodplain lake (Lake Taihu, East China) over one year. In addition, we used the n- alkane proxies to examine the inﬂuence of the seston source on seston stoichiometry variation. Throughout the study, the average value of the C:N:P ratio of 143:19:1 across Lake Taihu was close to the canonical lake’s ratios (166:20:1). Similar to other eutrophic lakes, seston C:N ratios varied the least across all environments, but C:P and N:P ratios varied widely and showed a strong decreasing trend in ratios of N:P and C:P from growing season to senescence season. This seasonal change was mainly associated with the decreasing contribution from algal-derived materials in seston pools because the non-algal dominated seston exhibited signiﬁcantly lower ratios than algal-dominated seston. Furthermore, the spatial heterogeneity of stoichiometric ratios was also related to the seston source. During the senescence season, the terrestrial-dominated seston from agricultural watershed exhibited the lowest ratios in estuary sites compared with other areas. Statistically, the predictive power of environmental variables on stoichiometric ratios was strongly improved by adding n- alkanes proxies. Apart from source indicators, particulate phosphorus (PP) contents also partly explained the spatiotemporal variations in stoichiometric ratios. This study, thus, highlights the utility of multiple-combined n- alkane proxies in addition to simple C:N ratios to get more robust source information, which is essential for interpreting the spatiotemporal variations in seston stoichiometric ratios among eutrophic ﬂoodplain lakes and other freshwater ecosystems.


Introduction
The C:N:P ratio of suspended particulate matters, hereafter "seston", is closely tied to biogeochemical cycles and food webs in aquatic ecosystems [1][2][3]. Since A. C. Redfield first observed that C:N:P ratios of seston in the offshore surface ocean are similar to the ratios of dissolved nutrients in the deep ocean, the classical Redfield ratio (C:N:P = 106:16:1) is routinely used to interpret and local scale. While dissolved and total nutrients have been comprehensively studied in Lake Taihu, drivers of variation in particulate nutrients stoichiometry remain underexplored [20].
In this study, the seston C, N, and P concentrations were measured monthly across Lake Taihu over one year, where the environmental data were simultaneously documented. In addition, n-alkane proxies were applied to identify the source of seston in Lake Taihu. Our primary objective was to characterize spatiotemporal variations in seston C:N:P stoichiometry in Lake Taihu and to relate this variation to environmental factors, putting special attention to sources of seston. We hypothesized that seston sources are important factors in shaping the spatiotemporal pattern of seston stoichiometric ratios in Lake Taihu and similar eutrophic floodplain lakes.

Study Sites
Samples were collected at five sites ( Figure 1) that could be taken as representatives as four main sub-ecotypes of Lake Taihu [43]. Station DPE and ZSB are, respectively, situated on the estuary area of the Dapu river and Yicun river, which are the two greatest inflow rivers of Lake Taihu [38]. Site MLB is located in Meiliang Bay, a region where cyanobacteria bloom is serious and riverine inputs are rigorously constrained. Station LC is located in the center of the open lake, where the water is less enriched with nutrients but exposed to more frequent wind-induced sediment resuspension. Site XKB is located in Xukou Bay, a typical macrophyte zone near the outflow Xu river.
Water 2019, 11, x FOR PEER REVIEW 3 of 16 In this study, the seston C, N, and P concentrations were measured monthly across Lake Taihu over one year, where the environmental data were simultaneously documented. In addition, n-alkane proxies were applied to identify the source of seston in Lake Taihu. Our primary objective was to characterize spatiotemporal variations in seston C:N:P stoichiometry in Lake Taihu and to relate this variation to environmental factors, putting special attention to sources of seston. We hypothesized that seston sources are important factors in shaping the spatiotemporal pattern of seston stoichiometric ratios in Lake Taihu and similar eutrophic floodplain lakes.

Study Sites
Samples were collected at five sites ( Figure 1) that could be taken as representatives as four main sub-ecotypes of Lake Taihu [43]. Station DPE and ZSB are, respectively, situated on the estuary area of the Dapu river and Yicun river, which are the two greatest inflow rivers of Lake Taihu [38]. Site MLB is located in Meiliang Bay, a region where cyanobacteria bloom is serious and riverine inputs are rigorously constrained. Station LC is located in the center of the open lake, where the water is less enriched with nutrients but exposed to more frequent wind-induced sediment resuspension. Site XKB is located in Xukou Bay, a typical macrophyte zone near the outflow Xu river.

Fieldwork
Monthly sampling works were carried out between June 2016 and May 2017. Five liters of subsurface (0.5 m) water was collected from each site using an acid-cleaned plastic sampler. Due to the shallowness (e.g., mean depth is 1.9 m) and intensive wave-induced vertical mixing, the samples from the sub-surface could represent the entire water column [44]. The water temperature (T) was measured in situ using a YSI-6600 handle multiparameter meter (Yellow Springs Instruments, OH USA). All water samples were transferred in acid-washed 5 L bottles and temporally stored at 4 °C before further analysis.

Sample Processing and Analyses
Using standard methods [45], total suspended solids (TSS) and chlorophyll a (Chl a) were

Fieldwork
Monthly sampling works were carried out between June 2016 and May 2017. Five liters of sub-surface (0.5 m) water was collected from each site using an acid-cleaned plastic sampler. Due to the shallowness (e.g., mean depth is 1.9 m) and intensive wave-induced vertical mixing, the samples from the sub-surface could represent the entire water column [44]. The water temperature (T) was measured in situ using a YSI-6600 handle multiparameter meter (Yellow Springs Instruments, OH USA). All water samples were transferred in acid-washed 5 L bottles and temporally stored at 4 • C before further analysis.

Sample Processing and Analyses
Using standard methods [45], total suspended solids (TSS) and chlorophyll a (Chl a) were quantified immediately upon return to the Taihu Laboratory for Lake Ecosystem Research (TLLER, http://taihu.niglas.cas.cn/). Water samples for dissolved nutrients were filtered through pre-ashed 0.7 µm glass fiber filters (Whatman GF/F, Sigma-Aldrich, St. Louis, MO, USA), and all GF/F filters for seston stoichiometry and n-alkanes were freeze-dried and stored at −20 • C. Total dissolved phosphorus (TDP), dissolved nitrogen (TDN), ammonium (NH + 4 ), and phosphate (PO 3− 4 ) concentrations in filtrates were measured according to standard analytical methods [45]. Dissolved organic carbon (DOC) were determined on a TOC analyzer (Shimadzu, Kyoto, Japan). For analyzing particulate organic carbon (POC) and particulate nitrogen (PN) in seston, frozen filters were thawed and dried at 50 • C for 1 h, then weighed and kept at room temperature in a desiccator. In order to remove the carbonate fraction, these filters were fumed with HCl and re-dried in a clean fume hood. The carbon and nitrogen contents were subsequently determined on an NA-1500 elemental analyzer (Fisons Instruments, Glasgow, United Kingdom). Particulate phosphorus (PP) was analyzed on corresponding filters placed in 15 mL distilled water that acidified with H 2 SO 4 and digested with peroxodisulphate at 121 for 1 h [18]. The final phosphate concentration in digested solutions was measured spectrophotometrically as the same protocol as TP and TDP. POC, PN, and PP indicated the mass concentration per unit mass of TSS. ρPOC, ρPN, and ρPP indicated the mass concentration per unit volume. As described in the previous study [46], the n-alkanes were extracted by the microwave method. Concentrations of n-alkanes were quantified on an Agilent 7890 gas chromatography (Agilent Technologies, Santa Clara, CA, USA) with a flame ionized detector. The n-C 36 alkane was used as lab standards for peak identification and quantification.

Calculation of n-Alkane Proxies
Although many n-alkane proxies are designed, some of them have been only applied in specific habitats or studies, such as P wax and K1. Thus, four classical proxies, which are widely used in lake ecosystems, had been selected and calculated to estimate the origins of seston, including CPI, TAR HC , the modified ACL between C 16 and C 32 [28], and the ratio that reflects the aquatic macrophyte to terrestrial plants (Paq) [31]. The calculation formulas and diagnostic values in the literature of selected proxies were listed in Table 1.

Statistical Analysis
Before the statistical analysis, raw datasets were log-transformed. The normality and variances' homogeneity of transformed datasets were confirmed with Shapiro-Wilk's tests and Levene's test (data not shown), respectively. Based on previous studies [8,48], the four regular seasons were categorized into two periods: the growing season (from April to September) and senescence season (from October to March). The site, season, and site and season interaction effects were tested by two-way ANOVA analysis. The multiple pairwise-comparisons were performed by Tukey HSD tests. These above statistical analyses were conducted in R program (version 3.5.3) with "agricolae" (version 1.3.1), "car" (version 3.0.5), and "lsmeans" (version 2.30.0) packages. Flexible discriminate analysis based on optimum scoring (FDA) was performed to investigate the discriminate patterns of n-alkane proxies, which was conducted by the "mda" (version 0.4.10) package. Analysis of similarity (ANOSIM) was used to test whether n-alkane proxies of samples were different between seasons in overall Lake Taihu or within each site, which was conducted by the "vegan" (version 2.5.5) package. The correlation between variables and stoichiometric ratios were investigated by Pearson correlation, which was conducted by the "psych" (version 1.8.12) package.
Partial least squares regression (PLSR) was used in order to examine the relative influence (predictive power) of individual environmental variables on each stoichiometric ratio. Compared to multiple linear regression, PLSR is more powerful for confronting possibly correlating and non-normal predictors [49]. The ratios and predictors' datasets were log-transformed before modeling. The optimum number of components were found using Wold's R criterion [50]. The k-fold cross-validation (k = 10 segments) was chosen to test validation of models. The performance of each PLSR model was compared using the coefficient of determination (R 2 ), root mean square error (RMSE), and mean bias (Bias). The relative influence of each variable was quantified using variable importance in the projection (VIP) scores where variables >1.0 have strong influences, variables <0.8 have weak influence, and variables in the middle range (0.8-1.0) have moderately influence [41,51]. All these processes were conducted by "mdatools" (version 0.9.4) and "pls" (version 2.7.1) packages.

Spatiotemporal Variations in Seston C:N:P Stoichiometry
The mean value of seston C:N:P ratios is 143:19:1 across Lake Taihu ( Figure 2). Seston N:P and C:P ratios ranged widely with a coefficient of variation (CV) of 44% and 40%, respectively, whereas C:N ratios varied the least (CV = 11%) throughout this investigation. Strong effects of season on N:P and C:P ratios were shown across Lake Taihu (Figure 2b,c). Specifically, Seston C:P and N:P ratios in DPE, ZSB, and LC significantly different between season. Their mean ratios were higher in the growing season than the senescence season. Seston C:P and N:P ratios in MLB also showed similar decreasing tendencies but were not statistically different. In contrast, although mean seston C:N ratios in DPE increased significantly from senescence season to growing season, C:N ratios did not differ significantly within other sites or entire lake datasets ( Figure 2a).
Spatially, the lowest C:N ratios could be observed in MLB (mean = 7.4), the values did not differ significantly between different locations ( Figure 3a). In contrast, the mean N:P and C:P ratios differed systematically across Lake Taihu (Figure 3b,c). Both C:P and N:P ratios were highest at MLB (means = 178.5 and 27.4, respectively) and generally lower at estuary sites. The seston stoichiometric ratios were not significantly different between LC and XKB. Although the lowest C:N ratios could be observed in MLB (mean = 7.4), the values did not differ significantly between different locations. In contrast, the mean N:P and C:P ratios differed systematically across Lake Taihu. Both C:P and N:P ratios were highest at MLB (means = 178.5 and 27.4, respectively) and generally lower at estuary sites. The seston stoichiometric ratios were not significantly different between LC and XKB. Furthermore, ANOVA indicated that the interaction effect of site and season was significant on all three ratios ( Table 2). The N:P and C:P ratios in estuary sites were significantly lower than those of other areas during the senescence season ( Figure S1). In contrast, the C:N ratios in estuary sites were significantly higher than those of other sites during this period. ( Figure S1). The p values denote significant differences between different seasons within each site and n.s. indicates no significant differences. The boxes span the interquartile range (25% to 75%), and the median is marked by a vertical line inside the box. The whiskers extend to the most extreme data point, which is no more than 1.5 times the length of the box away from the box. The dots at the end of the boxplot represent outliers.
Spatially, the lowest C:N ratios could be observed in MLB (mean = 7.4), the values did not differ significantly between different locations (Figure 3a). In contrast, the mean N:P and C:P ratios differed systematically across Lake Taihu (Figure 3b,c). Both C:P and N:P ratios were highest at MLB (means The p values denote significant differences between different seasons within each site and n.s. indicates no significant differences. The boxes span the interquartile range (25% to 75%), and the median is marked by a vertical line inside the box. The whiskers extend to the most extreme data point, which is no more than 1.5 times the length of the box away from the box. The dots at the end of the boxplot represent outliers. ratios were highest at MLB (means = 178.5 and 27.4, respectively) and generally lower at estuary sites. The seston stoichiometric ratios were not significantly different between LC and XKB. Furthermore, ANOVA indicated that the interaction effect of site and season was significant on all three ratios ( Table 2). The N:P and C:P ratios in estuary sites were significantly lower than those of other areas during the senescence season ( Figure S1). In contrast, the C:N ratios in estuary sites were significantly higher than those of other sites during this period. ( Figure S1). , and C:P (c) in different sites across Lake Taihu. Dashed reference lines show the canonical lake C:N:P ratio. Letters denote significant differences between different locations, and n.s. indicates no significant differences. The boxes span the interquartile range (25% to 75%), and the median is marked by a vertical line inside the box. The whiskers extend to the most extreme data point, which is no more than 1.5 times the length of the box away from the box. The dots at the end of the boxplot represent outliers. , and C:P (c) in different sites across Lake Taihu. Dashed reference lines show the canonical lake C:N:P ratio. Letters denote significant differences between different locations, and n.s. indicates no significant differences. The boxes span the interquartile range (25% to 75%), and the median is marked by a vertical line inside the box. The whiskers extend to the most extreme data point, which is no more than 1.5 times the length of the box away from the box. The dots at the end of the boxplot represent outliers.

Discriminate Analysis of n-Alkane Proxies
FDA was conducted to investigate the spatiotemporal patterns of four n-alkane proxies and discriminate different sources. The first two canonical variances could already explain 90.1% of the total variance, and the other canonical variances were not shown ( Figure 4). Basing on FDA analysis, the sources of seston in Lake Taihu could be generally classified into three groups. ANOSIM results indicated that the n-alkane proxies significantly differ between each group (Table 3). Group 1 (n = 23) was characterized as the lowest ACL (20.71) and TAR HC (0.53), including most samples from MLB, DPE, ZSB, and LC during the growing season. Group 2 (n = 13) was characterized by the highest ACL (27.1), TAR HC (11.34), CPI (3.26), and lowest Paq values (0.27), including most samples from DPE and ZSB during the senescence season. Similar to Group 2, Group 3 (n = 24) was also characterized by high ACL (26.5), TAR HC (10.2), but its Paq value (0.44) was the highest. All samples from XKB and part of the samples in LC during the senescence season were classified in Group 3. Remarkably, the multivariate n-alkane proxies were significantly different between growing and senescence seasons in Lake Taihu (ANOSIM, R = 0.267, p < 0.001). Furthermore, the seasonal differences were statically significant in DPE, ZSB, and LC ( Figure 4). However, the n-alkane proxies did not significantly change between different seasons in MLB and XKB.
FDA was conducted to investigate the spatiotemporal patterns of four n-alkane proxie riminate different sources. The first two canonical variances could already explain 90.1% l variance, and the other canonical variances were not shown (Figure 4). Basing on FDA an sources of seston in Lake Taihu could be generally classified into three groups. ANOSIM r cated that the n-alkane proxies significantly differ between each group (Table 3). Group 1 ( characterized as the lowest ACL (20.71) and TARHC (0.53), including most samples from , ZSB, and LC during the growing season. Group 2 (n = 13) was characterized by the highes ), TARHC (11.34), CPI (3.26), and lowest Paq values (0.27), including most samples from DP during the senescence season. Similar to Group 2, Group 3 (n = 24) was also characteriz ACL (26.5), TARHC (10.2), but its Paq value (0.44) was the highest. All samples from XKB an e samples in LC during the senescence season were classified in Group 3. Remarkabl tivariate n-alkane proxies were significantly different between growing and senescence se ake Taihu (ANOSIM, R = 0.267, p < 0.001). Furthermore, the seasonal differences were sta ificant in DPE, ZSB, and LC ( Figure 4). However, the n-alkane proxies did not signifi ge between different seasons in MLB and XKB.

C:N:P Stoichiometric Ratios of Seston from Different Sources
Statistically, C:N, N:P, and C:P ratios are significantly different (Kruskal-Wallis test, p < 0.001) between different potential sources ( Figure 5). The mean C:N value (9.03) of Group 2 is significantly higher than Group 1 (7.59) and Group 3 (7.51) (Figure 5a). On the contrary, the mean values of N:P (8.61) and C:P (86.91) ratios from Group 2 are significantly lower than those ratios (N:P = 23.67, C:P = 176.97) of Group 1 and Group 3 (N:P = 18.51, C:P = 138.61), while the differences between Group 1 and Group 3 are also significant (Figure 5b,c). Furthermore, the variations of N:P and C:P ratios from Group 1 (CV = 37% and 35%, respectively) are greater than those of Group 2 (CV = 18% and 18%, respectively) and Group 3 (CV = 24% and 26%, respectively).

C:N:P Stoichiometric Ratios of Seston from Different Sources
Statistically, C:N, N:P, and C:P ratios are significantly different (Kruskal-Wallis test, p < 0.001) between different potential sources ( Figure 5). The mean C:N value (9.03) of Group 2 is significantly higher than Group 1 (7.59) and Group 3 (7.51) (Figure 5a). On the contrary, the mean values of N:P (8.61) and C:P (86.91) ratios from Group 2 are significantly lower than those ratios (N:P = 23.67, C:P = 176.97) of Group 1 and Group 3 (N:P = 18.51, C:P = 138.61), while the differences between Group 1 and Group 3 are also significant (Figure 5b,c). Furthermore, the variations of N:P and C:P ratios from Group 1 (CV = 37% and 35%, respectively) are greater than those of Group 2 (CV = 18% and 18%, respectively) and Group 3 (CV = 24% and 26%, respectively). Letters denote significant differences between different potential sources. The boxes span the interquartile range (25% to 75%), and the median is marked by a vertical line inside the box. The whiskers extend to the most extreme data point, which is no more than 1.5 times the length of the box away from the box. The dots at the end of the boxplot represent outliers.

PLSR Analysis
Three PLSR analyses were conducted using only the eight environmental variables sub-datasets, including T, DOC, Chl a, TSS, TDN, TDP, NH , and PO (Table S1) for predicting each stoichiometric ratio (Figure 6a-c). In order to demonstrate the importance of considering the seston source for predicting stoichiometric ratios, another three PLSR models were conducted using the combination of these environmental variables with the n-alkane proxies described above (Figure 6df). Table 4 shows the performance produced by these models for both calibration and cross-validation. After adding n-alkane proxies, the accuracies of PLSR models for C:N, N:P, and C:P ratios increased, although the performance of PLSR models for C:N ratios were still low (Table 4). In contrast, the  (a-c). Letters denote significant differences between different potential sources. The boxes span the interquartile range (25% to 75%), and the median is marked by a vertical line inside the box. The whiskers extend to the most extreme data point, which is no more than 1.5 times the length of the box away from the box. The dots at the end of the boxplot represent outliers.

PLSR Analysis
Three PLSR analyses were conducted using only the eight environmental variables sub-datasets, including T, DOC, Chl a, TSS, TDN, TDP, NH + 4 , and PO 3− 4 (Table S1) for predicting each stoichiometric ratio (Figure 6a-c). In order to demonstrate the importance of considering the seston source for predicting stoichiometric ratios, another three PLSR models were conducted using the combination of these environmental variables with the n-alkane proxies described above (Figure 6d-f). Table 4 shows the performance produced by these models for both calibration and cross-validation. After adding n-alkane proxies, the accuracies of PLSR models for C:N, N:P, and C:P ratios increased, although the performance of PLSR models for C:N ratios were still low (Table 4). In contrast, the PLSR models based on complete datasets showed high accuracies for N:P and C:P. According to VIP scores, TAR HC , ACL, and Paq were important predictors in PLSR models for N:P and C:P ratios ( Figure 6). Furthermore, PO 3− 4 and TDP concentrations were also important factors in these PLSR models.
PLSR models based on complete datasets showed high accuracies for N:P and C:P. According to VIP scores, TARHC, ACL, and Paq were important predictors in PLSR models for N:P and C:P ratios ( Figure 6). Furthermore, PO and TDP concentrations were also important factors in these PLSR models.

Discussion
As expected for most inland waters, the average value of the C:N:P ratio of 143:19:1 across Lake Taihu was close to the modified ratio of 166:20:1 from global lakes' datasets [7]. This ratio was also

Discussion
As expected for most inland waters, the average value of the C:N:P ratio of 143:19:1 across Lake Taihu was close to the modified ratio of 166:20:1 from global lakes' datasets [7]. This ratio was also concomitant with a previous survey [20], which suggested that the seston stoichiometric ratios were interannually stable but interannually variable in Lake Taihu. In general, the seston N:P and C:P ratios varied more widely than C:N ratios, which is a common pattern within freshwater ecosystems [7,9,41]. As we expected, n-alkane proxies indicated that the origins of seston across Lake Taihu were diverse and spatiotemporally changing. These source indicators were important factors for predicting seston stoichiometric ratios in PLSR models. Adding n-alkane proxies increased the predictive power of the PLSR models considerably. These results statistically supported our hypothesis that the seston sources are important factors in shaping the spatiotemporal pattern of the seston stoichiometric ratios in Lake Taihu.

Application of Multiple n-Alkane Proxies as Complementary Source Indicators
Among stoichiometric studies, n-alkanes are rarely used to identify the seston source. The C:N ratios are widely used to basically distinguish the terrestrial and algal sources because the C:N ratios of algal origin ranged from 5 to 9 and those of terrestrial origins are normally higher than 15 [19]. However, as we mentioned above, these thresholds could be confounded by many factors during the degradation and transportation process of seston [23][24][25]. Overall, the mean value of C:N ratios in Lake Taihu was 7.53, suggesting that algal dominated the seston pool according to the empirical thresholds. Although the C:N ratios were much higher in the estuary sites during the senescence season, the mean value (9.23) was closer to algal signatures than terrestrial signatures in the literature. The homogeneity of C:N ratios has led to an underappreciation of the influence of terrestrial sources on the stoichiometric variation in Lake Taihu and possibly other freshwaters [20].
In this study, the FDA and ANOSIM analyses of n-alkane proxies indicate there are at least three distinct seston groups in Lake Taihu (Figure 4). Group 1 exhibited low ACL and TAR HC ( Table 2) that fell within ranges reported for algal sources (Table 1). This, along with the known cyanobacteria blooms in Lake Taihu [52], suggested that most of the seston in MLB, DPE, ZSB, and LC during the growing season were dominated by algal-derived materials. The high average Chl a concentrations in these sites (MLB = 264.06 µg/L, DPE = 322.79 µg/L, ZEB = 173.10 µg/L, LC = 46.8 µg/L) during the growing season strongly supported our inference. Noticeably, the majority of the seston pool in MLB were classified into this group even during the senescence season. This result coincides with the persistent cyanobacterial blooms in MLB, even in the winter, when the temperature is below 10 • C [53]. Both Group 2 and Group 3 exhibited high ACL and TAR HC , which are strong signals of terrestrial materials. However, the relative higher Paq value in Group 3 indicated alternative sources of middle-chain alkanes, such as macrophytes or phytoplankton [32]. Furthermore, the relatively lower CPI values in Group 3 indicated strong effects of diagenesis process on seston [19]. The intensive wind-wave disturbance in LC could provide sufficient dissolved oxygen, which accelerates the diagenesis process of resuspended sediments. Thus, most of the seston in LC were classified into Group 3 during the senescence season when the algal biomass decreased. However, it is unexpected that samples of XKB were also classified into the same group, as it is known that macrophytes could stabilize the benthic sediments during the growing season. We proposed that the seston in XKB are resuspended sediments, which may be transported from the open-water area. It could also explain the lack of significant seasonal changes in this system. Hence, we suggest that the seston of Group 3 were mixtures of terrestrial, macrophytes, and few algal materials that were formed by the frequent wind-induced mixing process. Although the seston in Group 2 exhibited a low Paq value and high CPI, which indicated they are terrestrial origins, their mean CPI value (3.26) was still lower than the threshold in the literature for typical terrestrial plants (CPI > 5) [37]. This inconsistency suggests that the seston in estuary sites during the senescence season were not fresh materials from terrestrial plants. These sestons have been reworked by microbes and maybe from erosion soils [29]. Yet the precise composition of different sources in the seston pool could not be fully resolved by n-alkane proxies alone. Nevertheless, this study supports the utility of multiple-combined n-alkane proxies in addition to simple C:N ratios to get more robust source information in complicated freshwater ecosystems.

Spatiotemporal Variations in Stoichiometric Ratios Corresponded with Changes in Seston Sources
Consistent with other eutrophic lakes, there are significant seasonal changes with lower ratios of N:P and C:P during the senescence season and higher ratios in the growing season [20,41]. Apart from the estimation of the source, previous studies used to attribute the pronounced seasonal shifts in seston C:P ratios to a "growth by dilution" hypothesis in eutrophic lakes [8], or some researches explained these shifts by the changes of communities species [20] because diatoms have low cellular C:P ratios than cyanobacteria. However, as illustrated above, these hypotheses only dominate under certain circumstances in which seston is consistently dominated by algal origins. It is clear that there are decreasing trends of algal derived seston from the growing season to senescence season in most sampling sites across Lake Taihu, especially in DPE, ZSB, and LC ( Figure 4). This implies the overall seasonal change in stoichiometric ratios may be caused by the synchronous change of seston sources.
The C:P and N:P ratios of algal dominated seston are significantly higher than those of non-algal dominated seston, especially much higher than the terrestrial-dominated seston from estuary sites ( Figure 5). These results indicated that the shift from algal materials to non-algal materials likely explained the decreased particulate C:P ratios and N:P ratios from the senescence season to growing season. This hypothesis was supported by significant correlations between these ratios and source indicators (ACL and TAR HC ) ( Figure S2). Similarly, the VIP scores in PLSR models also suggested that ACL and TAR HC are important for predicting ratios. Normally, the typical terrestrial stoichiometric ratios tend to be higher than the algal biomass due to a higher content of structural carbohydrates and lignin [6]. However, the C:P ratios and N:P ratios of terrestrial sources in Lake Taihu were lower on average than those of algal sources but fell within ranges reported for urbanized and agricultural watersheds [54,55]. The similar low ratios were also observed in Lake Erie watersheds, where agricultural activities were intensive [41]. The relatively lower C:P ratios and N:P ratios in seston could result from significant inputs of inorganic P associated with erosional soils from agriculture-rich catchments surrounding Lake Taihu [38,56].
The low C:P and N:P ratios of terrestrial-dominated seston were fingerprinted in the seston of LC during the senescence season, resulting in similar seasonal pattern changes in LC to estuary sites. However, the increased contribution of macrophyte-derived materials would also increase the ratios because macrophytes mostly have higher ratios than the agricultural soil [57]. Therefore, the seston from mixed sources exhibited significantly higher ratios than seston dominated by terrestrial materials ( Figure 5). Accordingly, the increasing trends in XKB from the growing season to the senescence season indicated that macrophytes' detritus contributed more during the senescence season when some macrophytes were fading. The positive correlations between Paq with N:P and C:P ratios in the Group 3 seston pool also supported this inference. Moreover, these source-specific ratios also explain why the estuary sites exhibited significantly lower N:P and C:P ratios than other areas during the senescence season ( Figure S1).

The Role of Phosphorus in Influencing the Seston Stoichiometric Ratios
Obviously, there are considerable variations in stoichiometric ratios within each seston source ( Figure 5). Particularly, these variations in algal-dominated seston were greater than other seston sources ( Figure 5). For example, there are large variations within MLB during the growing season, even though all these sestons were dominated by algal sources ( Figure S1). These results imply that other potential factors would influence the spatiotemporal variation in stoichiometric ratios. PLSR models indicated that TDP levels were as important as source indicators for predicting C:P and N:P ratios ( Figure 6). Many studies argue that the dynamics of sestonic phosphorus are largely responsible for systematic changes in C:P and N:P ratios and influenced by dissolved phosphorus concentrations [7,18,22,41]. These arguments were supported by significant negative correlations between PP and C:P/N:P ratios ( Figure S2), as well as the significantly positive correlations between PP and TDP concentrations in this study ( Figure S3).
It is not surprising that algal seston exhibited low PP contents when ambient TDP was relative low (TDP ≈ 0.02 mg/L) ( Figure S3), which suggested that the algal growth may be limited by phosphorus [9]. Although there are high dissolved-phosphate concentrations on average in Lake Taihu, the bioavailable phosphorus could be quickly consumed and be deficient during serious algal blooming events. Previous mesocosm bioassays showed phosphorus limitation could appear in different locations in Lake Taihu, even in the hypertrophic northern region (e.g., MLB) [58]. Alternatively, the luxury P-uptake by algal cells may be another reason for the tied correlations between PP and TDP in algal seston. Algal can excessively uptake phosphorus when it is surplus, even though algal cells are not P-limited. Recently, a nutrient addition experiment illustrated significant elevations of algal cellular P contents and synchronous decreases of C:P/N:P ratios when excessive dissolved phosphorus was added into systems [59]. The excessive P-uptake mechanism could also explain the outstanding higher maximum PP concentrations and greater variations in stoichiometric ratios in algal dominated seston than other seston sources.
Interestingly, similar positive relationships were shown between TDP and PP within non-algal dominated seston. Different from algal dominated seston, the adsorption and desorption process of phosphorus on the mineral of terrestrial seston could be one reason for the coupling relationship [60]. However, the maximum adsorption quantities of P to non-algal dominated seston would be limited by the physiochemical characteristics of these particles, especially in those sestons that already contained high P contents and might have been partially P-saturated. This deduction was in accordance with the significant but weaker correlation between TDP and PP in terrestrial-dominated seston directly from agriculture watersheds ( Figure S3). Accordingly, the limited P-uptake in these sestons could be partly responsible for the smaller variations in stoichiometric ratios. Furthermore, the attached microbial communities have the capacity to use the seston organic phosphorus as well as inorganic phosphorus, which is partly regulated by the dissolved phosphorus as well [61]. More accurate data on the different fractions of dissolved and particulate phosphorus are needed to distinguish between these mechanisms. Nevertheless, this study highlights that P-uptake processes could play an important role in seston stoichiometry variation, and could, therefore, play an important role in biogeochemical C:N:P processing, which would differ substantially based upon predominant P-sources in different types of lakes.

Conclusions
To summarize, this study points out the importance of estimation of seston sources in order to make a correct explanation underlining the spatiotemporal patterns in seston C:N:P ratios. Compared with generally homogenous C:N ratios, the application of multiple-combined n-alkane proxies provide more information on the dynamics of three distinct seston sources in Lake Taihu. Remarkably, our results indicate that the spatiotemporal changes in stoichiometric ratios can also arise due to delivery or resuspension of non-algal dominated organic matter in addition to changes in algal stoichiometry reported in other lakes. In addition, within each seston source, there are distinct P-uptake mechanisms behind the dynamics of sestonic phosphorus, which is also partly responsible for the variations in C:P and N:P ratios. These results highlight, again, that the important role of seston source has been underappreciated in most lakes, which is caused by prior subjective knowledge or applications of insensitive indicators. This study also highly recommends multiple-combined n-alkane proxies as a complementary tool in addition to C:N ratios for source estimation within freshwater ecosystems.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4441/12/1/36/s1, Figure S1: Variations in seston stoichiometric ratio of C:N, N:P and C:P across different sites across Lake Taihu during the growing season (a-c) and senescence season (d-f)., Figure S2: Pearson correlation heatmap of variables and stoichiometric ratios in different seston sources. Figure S3: Liner relationship between TDP and PP in different seston sources. Table S1: Annual summary of variables collected across different sites in Lake Taihu.
Author Contributions: J.C. conceived the conceptualization and collected the experiment data; G.G., K.S. and X.T. supervised the research; J.C. and C.B. analyzed the data and draw the figures. J.C. wrote the manuscript. Y.H., X.J. and J.D. helped with the data interpretation. All authors have read and agreed to the published version of the manuscript.