mic range. uninfected) were JW 55 site pooled prior to analysis. Raw sequencing reads are available from NCBI with accession number SRP029983, and aligned BAM files are available from NasoniaBase. Mapping Reads to the Nasonia Genome We aligned our RNA-seq reads to the N. vitripennis reference genome version 1.0, available from NasoniaBase. We used a combination of TopHat2 and Stampy to map our sequencing reads to the reference scaffolds, in order to leverage the unique strengths of each program and maximize mapping success. We first mapped each library using Tophat2 with the following options: -N 3read-edit-dist 5read-realign-edit-dist 2 -i 50 -I 5000max-coverage-intron 5000 -M, and using the Nvit OGSv2 GFF file as a genome reference. Next, we realigned the unmapped reads from TopHat 
with Stampy with the following options: substitutionrate = 0.01sensitive. The goal of the dual mapping protocol is that while TopHat deals well with spliced reads, it is not as sensitive as Stampy in mapping reads with polymorphisms. Stampy, however, does not deal well with junctions, and is very slow. Thus, Stampy is only used to map reads that TopHat cannot accurately place. After mapping, we merged the Stampy and TopHat output to produce a single BAM file for each library. To check for 39 bias, we estimated read coverage across the length of the gene body based on computing reads mapping to each percentile bin of gene length using RSeQC. In general we see little evidence for 39 bias: mapped reads that overlap genes are generally relatively unbiased with respect to gene position. After this mapping procedure, we were still left with a substantial number of unmapped reads, especially in the infected sample. To capture additional spliced reads that remained unmapped, we remapped all unmapped reads against the OGSv2 predicted transcriptome using the very-sensitive-local option in bowtie2, and added the counts of reads mapped to each gene to the counts derived from the genome mapping. Detecting Differentially Expressed Genes To identify genes that are regulated by infection, we used the DESeq2 package in Bioconductor/R. We first counted the number of reads that overlap each transcript in the OGSv2 dataset with the htseq-count script included as part of the HTSeq Python package, using the following options: -a 30 -m -intersection-nonempty -s no -t exon -i gene_name. We then processed the resulting count files with 15126366 a custom perl script to merge counts across manually annotated genes that are split across scaffolds and add counts from our final transcriptome mapping round. We used the resulting count file as input to DESeq2. DESeq2 uses a negative binomial distribution for statistical inference about differential expression. Because our biological replicates were pooled prior to RNA extraction and sequencing, we ran DESeq2 with several modifications to the default procedure. We use the standard dispersion 6882442 estimation approach, except that we treat the infected and uninfected samples as replicates in order to estimate dispersion. We use the standard maximum a posteriori approach to fit genewise dispersions, but we do not use outlier detection, nor do we use filtering by Cook’s distance. We use the independent filtering approach to filter the final results on mean normalized counts across all genes, as genes with very low counts have little or no power to detect differential Materials and Methods Nasonia Materials, Infections, and Sequencing Individual female adult wasps of t