Diatom

Smucker, Nathan J., Pilgrim, Erik M., Nietch, Christopher T., Darling, John A. & Johnson, Brent R., 2020, DNA metabarcoding effectively quantifies diatom responses to nutrients in streams, Ecological Applications 30 (8), pp. e 2205-e 2205 : 6-9

Diatom responses to TP and TN

Of 137 OTUs included in analyses (>1% relative abundance in>1% of samples), TITAN identified 52 as decreasers (low P diatoms) and 49 as increasers (high P diatoms) along the TP gradient (Appendix S1: Table S3). Sum z scores for low P diatoms identified an assemblage change point at 96 µg TP/L (5 th and 95th percentiles of 68 and 141 µg TP/L), and those for high P diatoms identified a change point at 152 µg TP/L (5 th and 95th percentiles of 139 and 310 µg TP/L; Fig. 3a). Cumulative frequency distributions of bootstrapped change points showed that most occurred from 60 to 135 µg TP/L for low P diatoms and from 135 to 160 and 210 to 220 µg TP/L for high P diatoms. For low P diatoms, half of the 52 OTUs had change points between 27 and 102 µg TP/L ( Fig. 3b, Appendix S1: Table S3), providing further evidence of substantial assemblage change within this range of TP. Beyond 136 µg TP/L, assemblage change was more gradual as indicated by the linear addition of 15 low P diatom OTU change points. Changes in high P diatoms were more gradual as indicated by a broad peak in sum z scores and distributions of 5th–95th percentiles of bootstrapped change points ( Fig. 3a), along with the more gradual addition of OTU change points along the TP gradient ( Fig. 3b). However, 11 of the 49 OTUs had change points between 128 and 164 µg TP/L, a range that also included the greatest distribution of bootstrapped change points ( Fig. 3a).

TITAN identified 50 OTUs as decreasers (low N diatoms) and 30 as increasers (high N diatoms) along the TN gradient (Appendix S1: Table S4 ). The sum z scores for low N diatoms identified an assemblage change point at 333 µg TN/L (5 th and 95th percentiles of 307 and 667 µg TN/L), and those for high N diatoms identified a change point at 823 µg TN/L (5 th and 95th percentiles of 574 and 851 µg TN/L; Fig. 3c). Cumulative frequency distributions of bootstrapped change points showed that most occurred from 250 to 450 and 560 to 650 µg TN/L for low N diatoms, which also coincided with two peaks in sum z scores, and from 560 to 870 µg TN/L for high N diatoms ( Fig. 3c). Assemblage response to TN was more gradual than to TP as indicated by a mostly linear addition of low N diatom OTU change points from 206 to 775 µg TN/L, though the magnitude of OTU change ( z scores) tended to be greatest between 250 and 450 µg TN/L ( Figs 3c and d). A more steady increase in sum z -scores and broader distributions of assemblage and OTU bootstrapped change points for high N diatoms further indicated gradual change along the TN gradient.

Boosted regression tree models explained 50–64% of the deviance in diatom metrics and NMDS axis 1 scores, which is how well the models explained observed data ( Table 4 , Appendix S1: Fig. S6 ). Cross-validation showed how well models predicted withheld data, and except for high N diatoms (32%), models performed well and explained 45–49% of the cross-validated deviance. TP was clearly the most important predictor in NMDS axis 1, low P diatom, and high P diatom models, whereas the importance of TP and TN in low N diatom and high N diatom models were about equal, indicating that TN provided substantial and unique contributions to these diatom responses. Partial dependence plots showed multiple points along TP and TN gradients at which diatom metrics had large responses and showed concentrations beyond which responses no longer decreased or increased ( Fig. 4 ). NMDS axis 1 scores showed gradual assemblage change as they declined linearly from 28 to 185 µg TP/L, but they also had a steep decline from 283 to 308 µg TP/L. Low P diatoms had steep and increasingly large declines beginning at 32, 74, and 118 µg TP/ L before having a gradual decline from 207 to 293 µg TP/L. High P diatoms had a small gradual increase from 25 to 80 µg TP/L, a large increase from 129 to 175 µg TP/L, and a final small increase from 283 to 290 µg TP/ L prior to a large hump in the response curve. Low N diatoms had a large decrease from 281 to 531 µg TN/L and a modest decrease from 594 to 788 µg TN/L. High N diatoms had large increases from 538 to 850 and from 1,212 to 1,344 µg TN/L.

Gradient forest analysis identified 99 diatom OTUs with R 2 values ranging from 0.002 to 0.614 (mean = 0.14) with mean R 2 values of the top 50 OTUs and bottom 49 being 0.23 and 0.06, respectively ( Fig. 5 , Appendix S1: Table S5, Fig. S7). Peaks in the standardized split density plot showed that the greatest changes in diatom assemblages occurred from 20–75, from 75 to 134, and from 250 to 367 µg TP/L, with smaller peaks indicating additional, albeit more minor, changes at 418, 513, and 722 µg TP/L ( Fig. 5a ). These peaks corresponded with portions of the TP gradient within which steep increases in cumulative importance occurred for several diatom OTUs ( Fig. 5b ) and with relatively steeper increases in the plots showing the cumulative change in assemblage composition ( Figs. 5c ), which represents average OTU change. Assemblage change occurred more gradually along the TN gradient between 153–833 µg TN/L ( Fig. 5d ), which encompassed the majority of change in the cumulative importance for most OTUs and assemblage change ( Fig. 5e–f ). The peaks at 4,362 and 4,875 µg TN/L likely were artefacts of low data density combined with relatively minor changes in OTU024.

Summary of diatom responses

The multiple statistical analyses identified strong relationships between diatoms and increasing concentrations of TP and TN. Less complex univariate analyses showed similar trends as well (Appendix S1: Fig. S8). While there was within-site variability in diatom metrics and NMDS axis 1 scores during the study (Appendix S1: Fig. S9), their relationships with nutrient concentrations were strong ( Figs. 2–5 , Appendix S1: Fig. S10). A synthesis of results from TITAN, boosted regression trees, and gradient forest analysis showed that multiple nutrient concentrations denoted important portions of the TP and TN gradients where large changes in diatom assemblages occurred ( Fig. 6 , Appendix S1: Tables S6–S7). In general, diatom OTUs and assemblages began changing from approximately 20–75 µg TP/L, as indicated by the largest peak in the split density plot, low P diatom change points, and a small yet notable decrease in the relative abundance of low P diatoms. The TITAN change point for low P diatoms (96 µg TP/ L), large decreases in relative abundances of low P diatoms, and another peak in split densities occurred from 75 to 150 µg TP/L. From 150 to 300 µg TP/L, the TITAN change point for high P diatoms occurred (152 µg TP/L), high P diatoms had large increases in relative abundances, and the final decrease in low P diatom relative abundances was observed. Above 300 µg TP/L, diatom assemblages were substantially altered and only a few additional minor changes occurred based on gradient forest analysis and OTU change points in TITAN.

In general, diatom assemblages had gradual but substantial changes from approximately 280 to 850 µg TN/ L ( Fig. 6 ). The TITAN change point for low N diatoms (333 µg TN/L) and the largest decrease in relative abundances of low N diatoms occurred from 280 to 525 µg TN/L. Mid-points of overall assemblage change occurred near 525 µg TN/L (e.g., peak in split density plot and middle of NMDS1 response in boosted regression trees). From 525 to 850 µg TN/L, high N diatoms had large increases in relative abundances, the final decrease in low N diatom relative abundances was observed, and the TITAN change point for high N diatoms occurred (823 µg TN/L). Above 850 µg TN/L, only minor changes in diatom assemblages occurred aside from a second increase in high N diatom relative abundances from 1,212 to 1,344 µg TN/L.

Figures

FIG. 6. Summaries of diatom OTU responses to (a) total phosphorus and (b) total nitrogen. In panel b, x-axis break is 1,499– 3,999 µg/L. Crossed circles are results directly from TITAN (change points [CP] with bootstrapped 5th and 95th quantiles and mid-points in the portions of cumulative frequency distributions showing the greatest increases in bootstrapped change points). Circles are boosted regression tree (BRT) results showing mid-points from portions of partial dependence plots with substantial change in response variables. Squares are peaks in standardized splits density plots in gradient forest (GF) analysis. Colors denote responses by low nutrient diatom OTUs (blue), high nutrient diatom OTUs (red), and assemblage change (gray; GF and nonmetric multidimensional scaling [NMDS]). Horizontal lines represent the range within which notable responses occurred (or 5th and 95th quantiles of TITAN change points). Diatom responses occurred at multiple points along the TP or TN gradient and these are denoted as BRT1, BRT2..., GF1, GF2..., and sumZ1, sumZ2. Vertical dashed gray lines demarcate portions of TP and TN gradients within which substantial changes in diatom assemblage occurred (see Summary of diatom responses in Results).

FIG. 4. Partial dependence plots from boosted regression trees showing responses of nonmetric multidimensional scaling (NMDS) axis 1 scores and of diatom metrics to TP or TN while controlling for the average effect of other variables. FF, fitted functions. Rug plots show deciles of predictor values.

FIG. 5. Results from gradient forest analysis showing diatom OTUs and assemblage responses to TP and TN. (a) Compositional change based on aggregating OTU responses were determined by split importance and values along TP and TN gradients (bars) and the ratios (blue lines) of split density (black lines) to data density (red lines). Peaks and regions of standardized split density plots with ratios>1 (horizontal dashed lines) delineate portions of the TN or TP gradient within which OTU compositional change is relatively greater than elsewhere along the nutrient gradient. (b) Cumulative change of individual OTUs is shown based on their splits’ importance distributions, which were scaled by R2 and standardized by data density for each OTU with the most important OTUs being labeled. (c) Relative rates of overall change in assemblage composition based on cumulative splits importance plots along the TP and TN gradients.

FIG. 2. Nonmetric multidimensional scaling (NMDS) ordination of all samples (n = 342) using relative abundances of OTU rbcL sequences. Spearman correlation overlays show relationships of axes with land cover, nutrients, and diatom metrics (gray lines). Axes are scaled to coincide with the possible range of correlations from —1 to 1. Ag, agriculture; TP, total phosphorus; TN, total nitrogen; HP, LP, HN, and LN, relative abundances of high and low phosphorus and nitrogen diatoms.