Phylogenetic Analyses Of ITS2 Under The Evolutionary Constraints Of Secondary Structure

The internal transcribed spacer(ITS)region is the only nuclear region that being widely used in current plant phylogenetic studies.ITS2 is the main variation region within ITS,and thus has higher phylogenetic value for species discrimination among related species,especially as a DNA barcode.Previous phylogenetic analyses have always described ITS2 evolution by using DNA model.However,ITS2 is transcribed into rRNA molecule with secondary structure in the actural organisms and the evolutionary mode within the paired regions in the form of compensatory mutations,i.e.,co-evolution between base pairs.This mode conflicts with the assumption of independence that mostly adopted in DNA-based phylogeny.The impact of the inappropriate DNA models on ITS2 phylogeny is still open.Based on large datasets,this study attempts to evaluate the evolution of ITS2 under the constraints of secondary structure and assess the effect on phylogenetic analyses by using the RNA models.To make the results more general we sampled 4,219 ITS2 sequences from 2,203 species and 70 genera,which representing 42 families and 32 orders in seed plants.All sequences were downloaded from GenBank.ITS2 was delimitated by referring annotation from both GenBank and ITS2 Database web server.In conventional phylogenetic analyses,the Bayesian method was used based on DNA models.The optimal model was selected by using the MrModelTest program,and then the Bayesian tree was constructed using the MrBayes software.In the phylogenetic analyses based on RNA models,we first determined the ITS2 consensus secondary structure of each data set by using the LocARNA software.Then the best-fitting substitution model was determined using a Perl script implemented in the PHASE-3.0 package("model selection.pl").The mcmcphase.exe program was used to estimate phylogenetic trees,which is also implemented in package PHASE-3.0.We compared the tree topologies,Bayesian posterior probability and species resolution between the phylogenetic trees constructed by the two methods and then performed statistical analyses to show the advantages of using the RNA models.The results showed that:1)for all 70 data sets,the best-fitting models for paired positions were RNA models.They were mainly 16-state models,and also fewer 7-state models.The setting parameters of discrete gamma model(+G indicates gamma-distributed rates-across-sites)and mismatch states of 7-state models(equal frequencies or empirical frequencies)for likelihood adjustment were also considered in model test.The results indicated that the diversity ITS2 evolution pattern in paired region was inconsistent with those of DNA models,highlighting the necessary to use RNA models for analyzing ITS2 matrix.2)The G+C frequency in paired regions was always higher than that in the unpaired positions.The high frequency base pair was GC,and followed by UA,GU and MM,while the mutability of these base pairs was in reverse order.These contrastive results showed that that ITS2 tended to retain the stability of its secondary strucure.In particular,we have chosen four particular rates to summarize the the matrices rules.We defined rd as the average double transitions rates between Watson-Crick pairs among four elements,likewise,rv was defined as the average double transversions rates between Watson-Crick pairs among eight elements,rf(forward rate)was defined as the average single substitutions from Watson-Crick pairs to GU,likewise rb(backward rate)was defined as the average single substitutions ratesfrom GU to Watson-Crick pairs.The statistical results showed that rd and r,,parameters of 34 genera were not zero,there were 19 genera where rd was higher than rv,and rd/rv values of the other 15 genera were less than 1;in all cases rd were less than rf;in some taxa such as Cuscuta,the value of rb/rf was approximately equal to 1,but in most cases rb was higher than rf.These results indicated the existence of CBC and their occurrences were double transitions that proceed by two-step compensatory mutation mechanism,involving fixation of the GU intermediate.GU intermediate was usually more rapidly replaced by Watson-Crick pair than its reverse substitution.3)We compared the phylogenetic trees constructed by DNA models and mixed models(DNA models for unpaired positions and RNA models for paired positions),and found they were different in both Bayesian posterior probability and species resolution.The Bayesian posterior probability based on the mixed models decreased at different level in 50%,70%,90%and 95%intervals,respectively.We defined 95%as the strongly Bayesian posterior probability,there were 30 taxa in which the species resulotion were reduced,with the percentage of decrease rate up to 100%.These results supported the idea that when the co-evolved base pairs were treated together as a single evolution unit rather than two separate characters in RNA model,the number of information character will be decreased,and the Bayesian posterior probability and the species resolution will reduce correspondingly.These results thus further confirmed our expection that traditional DNA models may overestimate the Bayesian posterior probability and the species resolution for the inferred tree.4)However,a few results showed that the Bayesian posterior probability and the species resolution were increased when selecting the RNA models.This may be interpreted as the presence of hemi-CBC and the increase of homologous positions in sequence alignments for which consensus secondary structure were considered.Conclusions that based on large datasets and new method system suggest that there is inappropriate use of ITS2 for phylogenetic inference.This paper thus highlights that when using the ITS2 region for phylogenetic analyses,the evolution constraints of secondary structure imposed on ITS2 should be taken into account and the RNA models should be added to ITS2 phylogenetic inference.We provide new insights for a more accurate phylogenetic method that based on a better ITS2 evolution model.