Dating back to almost 400 mya, spiders are among the most diverse terrestrial predators [1]. However, despite considerable effort [1-9], their phylogenetic relationships and diversification dynamics remain poorly understood. Here, we use a synergistic approach to study spider evolution through phylogenomics, comparative transcriptomics, and lineage diversification analyses. Our analyses, based on ca. 2,500 genes from 159 spider species, reject a single origin of the orb web (the "ancient orb-web hypothesis") and suggest that orb webs evolved multiple times since the late Triassic-Jurassic. We find no significant association between the loss of foraging webs and increases in diversification rates, suggesting that other factors (e.g., habitat heterogeneity or biotic interactions) potentially played a key role in spider diversification. Finally, we report notable genomic differences in the main spider lineages: while araneoids (ecribellate orb-weavers and their allies) reveal an enrichment in genes related to behavior and sensory reception, the retrolateral tibial apophysis (RTA) clade-the most diverse araneomorph spider lineage-shows enrichment in genes related to immune responses and polyphenic determination. This study, one of the largest invertebrate phylogenomic analyses to date, highlights the usefulness of transcriptomic data not only to build a robust backbone for the Spider Tree of Life, but also to address the genetic basis of diversification in the spider evolutionary chronicle.

High throughput sequencing and phylogenomic analyses focusing on relationships among spiders have both reinforced and upturned long-standing hypotheses. Likewise, the evolution of spider webs-perhaps their most emblematic attribute-is being understood in new ways. With a matrix including 272 spider species and close arachnid relatives, we analyze and evaluate the relationships among these lineages using a variety of orthology assessment methods, occupancy thresholds, tree inference methods and support metrics. Our analyses include families not previously sampled in transcriptomic analyses, such as Symphytognathidae, the only araneoid family absent in such prior works. We find support for the major established spider lineages, including Mygalomorphae, Araneomorphae, Synspermiata, Palpimanoidea, Araneoidea and the Retrolateral Tibial Apophysis Clade, as well as the uloborids, deinopids, oecobiids and hersiliids Grade. Resulting trees are evaluated using bootstrapping, Shimodaira-Hasegawa approximate likelihood ratio test, local posterior probabilities and concordance factors. Using structured Markov models to assess the evolution of spider webs while accounting for hierarchically nested traits, we find multiple convergent occurrences of the orb web across the spider tree-of-life. Overall, we provide the most comprehensive spider tree-of-life to date using transcriptomic data and use new methods to explore controversial issues of web evolution, including the origins and multiple losses of the orb web.

Upon publication of this article, it was suggested that some of the entries of the two web-type matrices required revision. We have amended the web data and performed the relevant analyses in order to assess whether such changes overturn any of the web evolution inferences of the study. Seventy-nine out of a total of 1,085 entries were revised with new scores (7.28%) as follows: Twentynine of the entries were conservatively changed to ''?'' to reflect absence of published data on their webs; the remaining 50 entries correct errors or provide new web coding interpretations. State 8 in the matrix for the taxa with transcriptomes has been reworded as ''non-foraging silk-lined burrows.'' After revising the scoring for web building behavior and web architecture, we repeated the relevant comparative analyses using the same methodological approaches described in the original article to reconstruct the evolution of these traits in spiders. Analyses of both the ten-and the three-character-state datasets were carried out with R packages ape [1] and phytools [2], and both revised datasets have been deposited in the Harvard Dataverse repository (https://doi.org/10.7910/ DVN/EJOMZP). Re-analyses of the revised data do not alter our previous conclusions in any significant way. Figures 3 and S2 have been corrected in the article online, and the corrected Figures 3 and S2 are also shown below. The authors apologize to the readers for any inconvenience that this revision may have caused and thank the colleagues that kindly suggested changes.

Genome-scale data sets are converging on robust, stable phylogenetic hypotheses for many lineages; however, some nodes have shown disagreement across classes of data. We use spiders (Araneae) as a system to identify the causes of incongruence in phylogenetic signal between three classes of data: exons (as in phylotranscriptomics), non-coding regions (included in ultraconserved elements [UCE] analyses), and a combination of both (as in UCE analyses). Gene orthologs, coded as amino acids and nucleotides (with and without third codon positions), were generated by querying published transcriptomes for UCEs, recovering 1,931 UCE loci (codingUCEs). We expected that congeners represented in the codingUCE and UCEs data would form clades in the presence of phylogenetic signal. Non-coding regions derived from UCE sequences were recovered to test the stability of relationships. Phylogenetic relationships resulting from all analyses were largely congruent. All nucleotide data sets from transcriptomes, UCEs, or a combination of both recovered similar topologies in contrast with results from transcriptomes analyzed as amino acids. Most relationships inferred from low occupancy data sets, containing several hundreds of loci, were congruent across Araneae, as opposed to high occupancy data matrices with fewer loci, which showed more variation. Furthermore, we found that low occupancy data sets analyzed as nucleotides (as is typical of UCE data sets) can result in more congruent relationships than high occupancy data sets analyzed as amino acids (as in phylotranscriptomics). Thus, omitting data, through amino acid translation or via retention of only high occupancy loci, may have a deleterious effect in phylogenetic reconstruction.