Compilation and network analyses of cambrian food webs

Abstract

A rich body of empirically grounded theory has developed about food webs--the networks of feeding relationships among species within habitats. However, detailed food-web data and analyses are lacking for ancient ecosystems, largely because of the low resolution of taxa coupled with uncertain and incomplete information about feeding interactions. These impediments appear insurmountable for most fossil assemblages; however, a few assemblages with excellent soft-body preservation across trophic levels are candidates for food-web data compilation and topological analysis. Here we present plausible, detailed food webs for the Chengjiang and Burgess Shale assemblages from the Cambrian Period. Analyses of degree distributions and other structural network properties, including sensitivity analyses of the effects of uncertainty associated with Cambrian diet designations, suggest that these early Paleozoic communities share remarkably similar topology with modern food webs. Observed regularities reflect a systematic dependence of structure on the numbers of taxa and links in a web. Most aspects of Cambrian food-web structure are well-characterized by a simple "niche model," which was developed for modern food webs and takes into account this scale dependence. However, a few aspects of topology differ between the ancient and recent webs: longer path lengths between species and more species in feeding loops in the earlier Chengjiang web, and higher variability in the number of links per species for both Cambrian webs. Our results are relatively insensitive to the exclusion of low-certainty or random links. The many similarities between Cambrian and recent food webs point toward surprisingly strong and enduring constraints on the organization of complex feeding interactions among metazoan species. The few differences could reflect a transition to more strongly integrated and constrained trophic organization within ecosystems following the rapid diversification of species, body plans, and trophic roles during the Cambrian radiation. More research is needed to explore the generality of food-web structure through deep time and across habitats, especially to investigate potential mechanisms that could give rise to similar structure, as well as any differences.

Figure 1. Original- and Trophic-Species Versions of the Chengjiang and Burgess Shale Food Webs

Spheres represent taxa, elongated cones represent feeding links. Position of the taxa vertically corresponds to their trophic level (TL), calculated using the short-weighted trophic level algorithm [45], with basal taxa (primary producers and detritus) shown at the bottom of the network in red, and highest trophic level taxa at the top in yellow. S: number of taxa (nodes) in the webs. L: number of trophic links. C: connectance; L/S². MaxTL: maximum trophic level of a species in the web. Images produced with Network3D software written by R. J. Williams; contact ricw@microsoft.com for more detail.

Figure 2. The Response of Species Richness (S), Connectance (L/S²) and Links per Species (L/S) to Link Removal in Two Cambrian Food Webs

Each data point shows the mean value across 100 webs for that level of link removal (except for 100% removal of low-certainty links and 0% removals, which show single values). Burgess Low and Chengjiang Low show data for removal of low-certainty links, and Burgess Random and Chengjiang Random show data for removal of random links.

Figure 3. The Response of 17 Structural Properties to Link Removal in Two Cambrian Food Webs

The data are shown in terms of percent change, or the difference between each new mean property value and the original property value, divided by the original property value, multiplied by 100. Mean property values are averaged across 100 webs for that level of link removal (except for 100% removal of low-certainty links and 0% removals, which show single values). Burgess Low and Chengjiang Low show data for removal of low-certainty links, and Burgess Random and Chengjiang Random show data for removal of random links.

Figure 4. Cumulative Link Distributions for Cambrian and Modern Food Webs

The data are presented in log-log format. Cambrian web data are shown with black (Chengjiang) and gray (Burgess) circles; the data for eight modern webs are shown with smaller colored squares. The link data are normalized (divided) by the average number of links per species in each web (i.e., 2L/S).

Figure 5. Distribution of Niche Model Errors for Cambrian and Modern Food Webs

Each column of stacked horizontal lines shows the MEs for 17 properties for a particular food web. The MEs for the Cambrian webs are shown in the first two columns, and the MEs for eight modern webs, in order of increasing species richness, are shown in the following columns. The dotted horizontal lines show ME = ±1. MEs within this range are considered to show a good fit of the model to the data. One very large ME, Herb = −7.00 for Skipwith, is not shown.

Figure 6. Comparison of Niche Model Errors for Cambrian and Modern Food Webs for 17 Structural Properties

For each property, individual ME values are shown for two Cambrian webs, followed by confidence intervals (CI) for eight modern webs, indicated by mean ME ±2.365 standard deviations (SD). Circles around four Cambrian data points indicate Cambrian web MEs that fall outside modern web CIs. 2.365 is the critical value from the t-distribution for seven degrees of freedom at a quantile of 97.5% for a two-tailed test. Dashed horizontal lines show ±1 ME boundaries, solid lines show ME = 0. MEs that fall within ±1 are considered to show a good fit of the model to the data.

Figure 7. Cumulative Link Distributions for Two Cambrian Food Webs Compared to Niche Model Predictions

The data are presented in semilogarithmic format. “All links” graphs show distributions of consumer plus resource links, “vulnerability” graphs show distributions of numbers of consumer species per species (i.e., number of links from consumers), and “generality” graphs show distributions of numbers of resource species per species (i.e., number of links to resources). Filled circles show empirical data. Solid lines show mean niche model simulation results for webs with the same S and C as the two Cambrian webs (n = 500), with 95% confidence intervals (±1.96 SD) shown by dashed lines.