Cutting the edge

IN a recent editorial I addressed reproducibility and reliability as key components of scientific publication and emphasized the value of well performed studies with “negative” outcome (1). This aspect was also an important component of the Editorial to the second part of the special issue on “Computational Data Analysis” (2). Questions on reproducibility, appropriate use of statistics, and particularly a clear bias in prioritizing publication of studies with positive as compared to negative outcome are well known phenomena in clinical and clinically related studies. When screening the literature I came across a 10-year-old publication by two former good colleagues of mine (Dubben and Beck-Bornholdt) from my PhD time in Hamburg (3). They addressed publication bias in papers on publication bias, thus a meta-study. Some of the studies analyzed data back as early as the late 1980s. The authors conclude that “selective underreporting of research might be . . . more likely to have adverse consequences for patients than publication of deliberately falsified data.” Furthermore, they pinpoint the lack or insufficiency of reported experimental details (3,4) and published three very humorous text books on the appropriate and (deliberately) false use of statistics (e.g., Ref. 5). (These books are a must to PhD students and medical doctors: However, to read them they need to have some expertise in German, Hungarian, Mandarin, or Japanese; unfortunately, there is no translation into English or any common Roman language!). In this line, this issue of the journal compiles a loose set of articles that are dealing with standardization or quality control, and particularly with cut-off values for clinical research and diagnostics. The most serious bias in cell based analysis is the preanalytical procedures (6). This is of particular relevance for multicentric studies where cell stimulation and subsequent (immuno-) labeling is involved. In order to reduce this bias, specimens are cryopreserved, collected, and then bulk stimulation and labeling is performed (7). However, cryopreservation and thawing represents also a bias that can, depending on the cell type, affect phenotype and function. This important issue was addressed by the careful study of Wang, H€ uckelhoven, Hong, Jin, Mani, Chen, Schmitt, and Schmitt (this issue, page 246). The authors investigated in detail the influence of various procedure modifications on viability, phenotype, and response to ex vivo stimulation on human peripheral blood mononuclear cells and report on cell type dependent biases. They propose for certain cell types a resting process and an algorithm to decide for which cell type resting process should be taken into consideration. Standardization of cutoff values for discriminating positive and negative specimens is crucial in clinical decision making. Du, Li, Sun, Chen, Yu, and Ying (this issue, page 239) focused their interest on the accurate detection and quantitation of nuclei with fluorescence in situ hybridization (FISH) signals for experimental and clinical analysis. Signal overlap of multiple signals in a single nucleus may lead to erroneous results. One approach is to analyze the spatial distribution of signals in 3D images captured by confocal microscopy (8). In their study, Du and colleagues approached the problem by mathematically modeling of flattened nuclei. They demonstrate that for routine analyses of FISH signals, a dynamic cutoff for problem free nuclei with few signals is comparable to conventional cut-off. However, dynamic values are better suited for nuclei with multiple FISH spots and can be introduced into an automated image analysis pipeline. Also, accurate counting of extremely rare cells in the blood stream is central in existing and proposed clinical diagnosis and therapy decision. This includes circulating tumor, stem and progenitor cells as well as circulating endothelial (9)