Incorrect analysis leads to Nature’s sixth retraction in 2014

In what seems to be an example of researchers swiftly and transparently correcting the literature, and acknowledging errors, a pair of scientists have retracted a 2013 paper from Nature.

Here’s the retraction notice for “Genomic organization of human transcription initiation complexes,” by Bryan Venters and Frank Pugh:

We reported the presence of degenerate versions of four well known core promoter elements (BRE_u, TATA, BRE_d and INR) at most measured TFIIB binding locations found across the human genome. However, it was brought to our attention by Matthias Siebert and Johannes Söding in the accompanying Brief Communication Arising (Nature 511, E11–E12, http://dx.doi.org/10.1038/nature13587; 2014) that the core-promoter-element analyses that led to this conclusion were not correctly designed. Consequently, the individual core promoter elements were not statistically validated, and therefore there is no evidence of specificity for most reported core-promoter-element locations. To the best of our knowledge, the raw and processed human TFIIB, TBP and Pol II ChIP-exo data are valid, but subject to standard false discovery considerations. We therefore retract the paper. We sincerely apologize for adverse consequences that may have arisen from the error in our analyses.

The paper, which has been cited 11 times, according to Thomson Scientific’s Web of Knowledge, was published on September 18 of last year, and was corrected on October 2:

Minor changes were made to the core promoter consensus sequences.

Here’s the abstract of the Brief Communication Arising (which, unlike the retraction, is behind a paywall):

How cells locate the regions to initiate transcription is an open question, because core promoter elements (CPEs) are found in only a small fraction of core promoters^{1, 2, 3, 4}. A recent study⁵ measured 159,117 DNA binding regions of transcription factor IIB (TFIIB) by ChIP-exo (chromatin immunoprecipitation with lambda exonuclease digestion followed by high-throughput sequencing) in human cells, found four degenerate CPEs—upstream and downstream TFIIB recognition elements (BRE_u and BRE_d), TATA and initiator element (INR)—in nearly all of them, and concluded that these regions represent sites of transcription initiation marked by universal CPEs. We show that the claimed universality of CPEs is explained by the low specificities of the patterns used and that the same match frequencies are obtained with two negative controls (randomized sequences and scrambled patterns). Our analyses also cast doubt on the biological significance of most of the 150,753 non-messenger-RNA-associated ChIP-exo peaks, 72% of which lie within repetitive regions.

In a sign of our rising retraction rate times, this is Nature‘s sixth retraction so far in 2014. Two of the papers described the STAP stem cell work, and two were of papers by Pankaj Dhonukshe. Last year, the journal published five. In 2010, the editors wrote:

This year, Nature has published four retractions, an unusually large number. In 2009 we published one. Throughout the past decade, we have averaged about two per year, compared with about one per year in the 1990s, excluding the pulse of retractions of papers co-authored by [German physicist Jan Hendrick] Schön.

Nature retracted seven Schön papers in 2003, and three others the same year, so 2014’s figure is not yet a record.

For comparison, Science has published two retractions (and one expression of concern) this year, while Cell has published one.

Hat tip: Andy Jerkins