No supporting evidence of classification based on FFPE samples, ambiguity in classification of EGFR mutants, and inclusion of bona-fide platelet genes in discriminator sets indicate no biological basis for using RNA-seq from tumor-educated platelets as a source in "liquid biopsy"

In this detailed critique of the study proposing using RNA-seq from tumor-educated platelets (TEP) as a liquid biopsy source [1], several flawed assumptions leave little biological basis behind the statistical computations. First, there is no supporting evidence provided for the FFPE based classification of METoverexpression and EGFR mutation on tumor-tissues. Considering that raw reads of MET expression in a subset of healthy [N=21, mean=112, sd=77] and NSCLC [N=24, mean=11, sd=12] samples (typically with millions of reads) translates into over-expression in reality, providing the data for such computations is vital for future validation. A similar criticism applies for classifying samples based on EGFR mutations (the study uses only exon 20 and 21 from a wide range of possible mutations) with negligible counts [N=24, mean=3, sd=6]. While Ofner et. al, 2017 faced major problems associated with FFPE DNA, it is also true that Fassunke, et al., 2015 found concordance in 26 out of 26 samples for EGFR mutations in another FFPE-based study. However, Fassunke, et al., 2015 have been meticulous in describing the EGFR amplicons (exon 18 and 19 are missing in the TEP-study). Any error in initial classification renders downstream computations error-prone. The low counts of MET in the RNA-seq firmly establishes that inclusion of genes with such low counts in the set of 1100 discriminatory genes (Table S4) makes no sense as the \"real\" counts could vary wildly. Yet, TRAT1 is an example of one discriminator gene with counts of healthy [N=21, mean=164, sd=375] and NSCLC [N=24, mean=53, sd=176]. There are many such genes which should be excluded. Moving on to a discriminator with high counts (F13A1) in both healthy [N=21, mean=28228, sd=48581] and NSCLC [N=24, mean=98336, sd=74574] samples, a bonafide platelet gene that \"encodes the coagulation factor XIII A subunit\". Platelets do not have a nucleus, and thus the blue-print (chromosomes and related machinery) for making or regulating mRNA. They are boot-strapped with mRNA, like F13A1, during origination and then just go on keep collecting mRNA during circulation (which is the premise of their use in liquid biopsy). The assumption that these genes are differentially spliced in huge numbers is highly speculative without providing experimental proof. The discovery of spliceosomes in anucleate platelets [2] in 2005, 30 years after splicing was discovered in the nucleus by Sharp and Robert, probably indicates that spliceosomes are not dominant in platelets. Zucker, et al., 2017 have shown for another gene F11 that it is present in platelets as pre-mRNA and is spliced upon platelet activation [3]. Any study using the F13A1 gene as a discriminator ought to show the same two things, followed by differential counts in TEP. Ironically, F11 is not present in the discriminator set. Another blood coagulation related gene (TFPI) shows slight over-expression in NSCLC (moderate counts, healthy [N=21, mean=1352, sd=592] and NSCLC [N=24, mean=1854, sd=846]), agreeing with Iversen, et al., 1998 [4], but in contrast to Fei, et al., 2017 [5], demonstrating that the jury is still out on the levels of many such genes. Thus, circulating mRNA from tumor tissues are not discriminatoryif MET is degraded to such levels in platelets educated by NSCLC tumors, why not other possible mRNA that might have been picked during the same class? Furthermore, high count genes can only be bona-fide platelet genes, and have no supporting experimental proof of splicing differences (any one gene would suffice to instill some confidence). In conclusion, looking past the statistical smoke surrounding \"surrogate signatures\", one finds no biological relevance.