Analysis

Sequencing reads were trimmed to 50 bp to ensure uniform high quality in the FASTQ scores and then aligned with RSEM version 1.2.29 to hg19 (International Human Genome Sequencing Consortium 2001; Li and Dewey 2011). Because updates to the genome in GRCh38 focused mostly on noncoding regions or regions we otherwise excluded, realignment to GRCh38 is not anticipated to alter the main findings reported here. FastQC and the fastx_toolkit were used in preliminary quality control measures. RSEM-derived counts or TPM tables were used throughout the remainder of the analysis, along with a variety of packages available through Bioconductor in R (R Core Team 2018). All code is available as Supplemental File S2.

We removed spuriously expressed genes (<1 TPM in all samples) and also removed samples expressing fewer than 5000 genes (representing about one-third the maximum transcriptome coverage we observed in other samples) (Supplemental Fig. S1B–D). As a result, three TE biopsy samples were excluded for failing to pass quality control (embryos 7, 13, and 36). Additionally, we removed mitochondrial genes as these represented about one-tenth of the total RNA read counts. The total set of expressed genes for the data set was defined by those which reach at least 1 TPM of expression in one or more samples. This kept the gene list as permissive as possible (we want to identify aberrant expression) while excluding noise. For the transcriptome coverage analysis, we limited this list to 1 TPM in 10 or more samples per sample type (TE or WE). For WE, we identified 13,175 expressed transcripts, whereas TE samples expressed 6086 transcripts. We cannot rule out the possibility that the restricted transcriptome in TE is owing to sample prep bias against smaller input and not a restricted transcriptional program in this specific lineage. However, sample prep bias should be further evidenced by more jackpotting in TE samples, and this appears to not be the case (Supplemental Fig. S1F).