The cataly tic domains of most retrieved sequences were delineated using Pfam. Sequences in Clade 6 have lower similarity to the classical PARPs used to generate the Pfam HMM, so the PARP catalytic domains for these Wortmannin DNA-PK sequences were identified using BLAST searches based on human PARP6 catalytic domain as the query and identifying the region of retrieved sequences that had similarity to this PARP signature. In addition, many sequences whose catalytic domain was incompletely identified by Pfam were completed by BLAST searches using closely related complete PARP catalytic domains from other closely related species, in order to provide as much sequence information as possible for the align ment and phylogeny inference. The identified PARP cat alytic domains were extracted using the extract.

pl tool in the Wildcat Toolbox set of Perl utilities. Sequences of less than 100 amino acids in length and many that were missing important structural elements of the PARP domain were discarded to allow better alignment and phylogenetic signal recovery. Many of these sequences were obtained from shotgun sequencing and are pre sumably incomplete. Phylogenetic analyses The collected PARP catalytic domains were aligned using the MUSCLE3. 8. 31 multiple alignment tool, using default settings. The multiple alignment was sub jected to a maximum likelihood analysis using PhyML3. 0 using the computer facilities at the Ohio Supercomputer Center. The substitution model parameters using for the PhyML analysis were the WAG substitution matrix, 8 I correc tion to model site rate heterogeneity and empirical equi librium frequencies.

These parameters were selected as the optimal substitution model based on analysis by ProtTest v2. 4. A parsimony based starting tree was used. Branch supports were computed in PhyML using an aLRT non parametric Shimodaira Hasegawa like procedure. Once a tree with all PARP domains had been generated, it was used to identify the six clades referred to in the text in combination with examination of domains outside of the PARP catalytic domain. After the six clades were defined, sequences from each clade were aligned separately using MUSCLE. These alignments were used to generate individual clade trees using PhyML with identical parameters. The phylogenetic trees were generated for figures using FigTree software figtree.

Align ment figures were generated using Cilengitide TEXshade and Jalview. Prediction of protein domains After sequences of PARP family members were retrieved and placed into clades, the sequences were checked for other domains at the Pfam website. Domains iden tified are shown in Figure 4. PfamB 30617 was identi fied in Clade 6A fungal proteins and extracted aligned as above. This domain was further analyzed using the Protein homology analogy recognition engine and renamed FPE. Subsequently, a consensus FPE sequence was used in BLAST searches to find other proteins containing this region.