Universal human reference rna uhrr

Microarray data came from the FDA MAQC main study [23 (link)] and consisted of data from Affymetrix HG-U133_Plus_2 microarrays. The RNA-Seq data were from the FDA SEQC main study [16 (link)] using the Illumina HiSeq 2000 platform. The microarray data were generated by Affymetrix site 1 in the MAQC study, while the RNA-Seq data were generated by site BGI in the SEQC study. Both sets of data were generated from the same set of four human RNA samples, that is, Universal Human Reference RNA (UHRR, Agilent), Human Brain Reference RNA (HBRR, Life Technologies), and mixtures C and D of UHRR and HBRR samples in a ratio of 3:1 and 1:3, respectively. The HG-U133_Plus_2 arrays were normalized with MAS5 algorithm. The RNA-Seq data were generated with Illumina HiSeq 2000 using the paired-end 100 bp TruSeq v3 RNA-Seq protocol and were analyzed with the P2 pipeline [22 (link)] using UCSC human genome hg19 as reference. Gene counts were normalized into reads per million (RPM) with a global scaling approach [35 (link)]. The microarray and RNA-Seq data can be obtained from GEO database with series accession numbers GSE5350 and GSE47774, respectively.

We used six datasets representing the TruSeq, Smart-Seq and UMI-seq protocols and varying amounts of starting material from bulk RNA or single cell RNA. All analysed datasets contain the ERCCs spike-in RNAs. This is a set of 92 artifical poly-adenylated RNAs designed to match the characteristics of naturally occurring RNAs with respect to their length (273–2022 bp), their GC-content (31–53%) and concentrations of the ERCCs (0.01–30,000 attomol/μl). The recommended ERCC spike-in amounts result in 5–10⁷ ERCC RNA molecules in the cDNA synthesis reaction.
To reduce biological variation, we used the well-characterized Universal Human Reference RNA (UHRR; Agilent Technologies) for the two datasets produced for this study. We downloaded UHRR- and HBRR-TruSeq data from SEQC/MAQC-III2 (link). Finally, we also analyse the single cell data published in Wu et al.20 (link), for which the colorectal cancer cell-line HCT116 was used (Table 1). The input mostly being commercially distributed human samples, we expect all biological samples analysed in this study to have similarly high quality and complexity. All data that were generated for this project were submitted to GEO under accession GSE75823.

The miRXplore miRNA reference set was purchased from Miltenyi Biotech. The RNA was dissolved in nuclease-free water (Invitrogen), adjusted to 1 μM, and aliquoted for storage at −80 °C. Fragmented human reference RNA samples were prepared as described^{7 (link)}. 50 μl of Universal Human Reference RNA (UHRR; Agilent) at 1 μg/μl was mixed with 1 μl of ERCC ExFold Mix 1 (Thermo Fisher Scientific; denoted ERCC spike-ins) prepared according to the provided protocol. 2 μl of the resulting UHRR sample with ERCC spike-ins was ribo-depleted by using a Human/Mouse/Rat Ribo-zero rRNA removal kit (Illumina), fragmented to 70–100 nt by using an NEBNext Magnesium RNA Fragmentation Module (94 °C for 7 min; New England Biolabs), and treated with T4 polynucleotide kinase (Epicentre) to remove 3′ phosphates that impede TGIRT template-switching^{7 (link)}. After each of the above steps, the RNA was cleaned-up by using a Zymo RNA Clean & Concentrator kit, with 8 volumes of ethanol added to the input RNA to maximize the recovery of small RNAs^{7 (link)}. The fragment size range and RNA concentration were verified by using a 2100 Bioanalyzer (Agilent) with an Agilent 6000 RNA pico chip and aliquoted into 6 ng/3 μl portions for storage in −80 °C.