Tuesday, May 28, 2019

To lose one co-author’s papers to an ORI investigation may be regarded as a misfortune; to lose two looks like carelessness

This post was earlier cross-posted at Leonid Schneider's site, hence the unfrivolous tone. The version there is improved by Leonid's editing, frame-story and illustrations. Further explorations by Elisabeth Bik here.

A University of Kentucky cancer toxicologist with a history of retractions is back under investigation, as recently reported in his local press [RW]. I have nothing to contribute except to link the case to another of Xianglin Shi's papers, which spawned a long-standing thread at Pubpeer, where the insouciant disregard of the editors of Blood reduced "Peer 2" to a state of agitated despair.
Can someone please wake me up from this nightmare?
How is this possible? They are everywhere, I stopped after these ones but there are probably more.
I have contacted VCU several times regarding the seemingly endless issues like the ones pointed out here, but still no action.
Have I misunderstood something? Maybe it is not science after all but an advanced form of "Where is Wally" VCU scientist are publishing?
Peer 2 was referring to the duplicated image elements, which swarm and cavort through the paper like leit-motifs in a Wagner score, like rabbits in the marginalia of an illuminated manuscript, binding the separate Figures together in an intricately-woven web of synchronicity. It is difficult to know where to begin. Anyone trying to annotate all the replications runs quickly out of colors.





The last time I saw so many mascara eyelashes together in one place was at a Siouxsie and the Banshees concert.

Our real focus here is Shi's co-authors on this dubious paper: Professors Paul Dent and Steven Grant of Virginia Commonwealth University, where (along with P. B. Fisher) they form a troika of cancer-research excellence. Their careers are resplendent with publications and scientific honors. I worry, though, that between receiving those honors and editing journals and the endless cycle of grant applications, they have little time left to spend supervising their students or carefully reading all the papers they sign.
The Dasmahapatra episode, for instance, was not a shining display of thorough mentoring and oversight. Girija Dasmahapatra confessed to faking results in eleven Dent / Grant collaborations from 2005 to 2014 [RW] and was subsequently thrown under the bus... or thrown to the wolves... or thrown to the wolves under the bus. It may be that Dent and Grant have really terrible luck with their co-authors.
All of the 11 affected papers will be corrected or retracted, according to the ORI notice.[...]
ORI found that Respondent falsified and/or fabricated data by reporting the results of Western blot experiments and mouse imaging experiments that examined interactions between multiple histone deacetylase and/or proteasome inhibitors in several cancer models. Specifically, Respondent duplicated, reused, and/or relabeled Western blot panels and mouse images and claimed they represented different controls and/or experimental results...
The Dent / Grant oeuvre is of interest to us as encapsulating in microcosm the current biomedical research paradigm, with its feudal hierarchies of PIs and grad-student lab-bench serfs, and the perverse structural incentives which motivate Principal Investigators to minimize the time they spend on the investigations they are principalling. No-one really needs yet another cautionary tale of "How Laboratory Culture Goes Wrong", no more than anyone wants another exhaustive gallery of re-used or manipulated Western blots, but that is what you get.

Grant and Dent's productivity (and Fisher's) caught the attention of the vast and cool and unsympathetic intellects who comment at Pubpeer, with the list of separate threads currently numbering in the 50s. Those threads hint at a general disenthusiasm in the laboratory ethos for farfing around with loading controls in gel-electrophoresis probes (to prevent accidental or unconscious variations in aliquots from skewing the results)... with a belated realisation that reviewers do expect controls, requiring improvised gap-fillers. Thus, from 2004, "Synergistic induction of oxidative injury and apoptosis in human multiple myeloma cells by the proteasome inhibitor bortezomib and histone deacetylase inhibitors" (Pei, Dai and Grant, 2004):

The loading control is Tubulin three times and Actin once. It has six lanes on two occasions and five lanes on the other two (losing Lane 2). It controls for different conditions that result in three different PARP / CF slices, three different Mcl-1 and XIAP slices, etc.

2004 was an Annus Mirabilis. New students joined the team, and we were treated to confections like these, from "Contribution of disruption of the nuclear factor-kappaB pathway to induction of apoptosis in human leukemia cells by histone deacetylase inhibitors and flavopiridol" (Gao et al 2004):

These may look as if Mondrian hurt his head falling downstairs but continued to paint while experiencing temporal-lobe seizures, but the goal is to illustrate how generic slices of gel, each containing seven lanes, each provided loading controls for a series of four-lane blots by shifting the read frame. More crucially, more substantive non-control parts of other Figures (such as the C-Caspase-3 panels of Fig 5A) also turn out to be fabricated.


Many of these demonstrations require adjustments to the contrast and lightness of the as-published Figures. This sheds a cruel, unflattering light, especially on older illustrations. To leap ahead to "Mechanism of in vitro pancreatic cancer cell growth inhibition by melanoma differentiation-associated gene-7/interleukin-24 and perillyl alcohol" (Lebedeva et al 2008), the backgrounds of Figures 1B and 4D do not look at all realistic when treated so harshly:



Returning to earlier in the Dent / Grant corpus, "2-Methoxyestradiol-induced apoptosis in human leukemia cells proceeds through a reactive oxygen species and Akt-dependent process" (Gao et al 2005) is another rich source of improvised loading controls.


"Figures 7d, 7e have their own loading control, but by this point in the paper, the authors had stopped caring very much."


The protein blobs in other, non-control panels of these Figures display the same impressive consistency of shape. Recurring visual motifs follow a cetacean theme, with a sperm-whale in profile, a whale tail, an oil-lamp. Are there special stencils or pipette tips one can use to standardise the shapes of aliquots when adding them to the wells?

Recurrent blobs are also prominent throughout "The three-substituted indolinone cyclin-dependent kinase 2 inhibitor 3-[1-(3H-imidazol-4-yl)-meth-(Z)-ylidene]-5-methoxy-1,3-dihydro-indol-2-one (SU9516) kills human leukemia cells via down-regulation of Mcl-1 through a transcriptional mechanism" (Gao et al 2006). Each Figure is a hiding-place for Easter Eggs, while the loading controls appear to have been assembled from IKEA kitsets.




There are earlier entries in this PubPeer corpus dating back to 2000-2003, raising concerns involving pictorial similarities and discontinuities... these could be clues to duplication or splicing, but could just be artifacts (the image quality is not always ideal, with Western blots seemingly viewed through a shimmering gauzy scrim, or photocopied and rephotographed, or printed on newsprint as half-tones and then sent via fax). It is time to remind ourselves that the mere number of papers flagged at PubPeer is not decisive, for comments there can be trivial or misguided.

Some of the queries, in fact, could easily be motivated by the critics' need for completeness, and the authors could be forgiven for feeling persecuted, or hounded, as if their names were Clinton. Initially they responded to the critiques:

#4 Peer 3
I agree I cannot explain the straightness of the p41 images, they look too straight to me too. I know I have seen this issue before on some of the digital material from some postdocs and have always assumed it was simply the way the gel ran. However, as I pointed out, I was not paying much attention to p41 for "quality issues" as the data set from which scientific conclusions are drawn is p31 / p25.

Those responses hinted at a hands-off approach to supervision of the student researchers. Leonid Schneider probably did not help.

#8 Peer 5
Dear Dr. Dent, if your statement is true, your postdocs are playing terrible tricks on you. It is a pity that you as professor at the Dept. of Biochemistry & Molecular Biology apparently don't have sufficient experience in this western blotting technology to help you recognise manipulation. Please enlist yourself and your employees in courses on good scientific practice as soon as possible.
Leonid Schneider
(from "OSU-03012 stimulates PKR-like endoplasmic reticulum-dependent increases in 70-kDa heat shock protein expression, attenuating its lethal actions in transformed cells": Park et al 2008).

But evidently the experience was not rewarding. The authors came to view PubPeer as a louche, improper forum and lost their enthusiasm for engagement with its insatiable and incorrigible commentariat.

#2 Unregistered Submission
We appreciate the heads-up about this error. We will look in to it. However, why post it here in such an unprofessional manner? Have some class.
In this case - "Apogossypol derivative BI-97C1 (Sabutoclax) targeting Mcl-1 sensitizes prostate cancer cells to mda-7/IL-24-mediated toxicity" (Dash et al 2011) - concerns about the problematic Figure 4C were rendered moot by a Correction, though it didn't really explain how the original weird hybrid came to exist.


Anyway, this survey of the corpus is fortunately not intended to follow a chronological sequence. We have already imposed some structure on the sprawling literature by distinguishing a 'Dasmahapatra' period within it, much as one organises the catalog of a heavy-metal band according to replacement drummers or bassists. Similarly we might recognise a 'Rosato' period... Roberto Rosato appears as co-author in 13 PubPeer threads in the oeuvre from 2002 to 2012, accounting for two Retractions and an Erratum. A VCU spokesperson advised RetractionWatch that "The article’s first author, Roberto Rosato, left VCU in 2010."

This sub-corpus deserves to be explored at length, but we can only mention a few highlights - like the eight-lane kitset-assembled loading strip used to control for nine lanes of the protein of interest, in "Mechanism and functional role of XIAP and Mcl-1 down-regulation in flavopiridol/vorinostat antileukemic interactions" (Rosato et al 2007).

#3 Peer 2
July 19th, an impressive release of analyses of Steven Grant studies. I like to express my admiration for the quantity, clarity and quality.
Perhaps this was to compensate for Fig 6D in "Potentiation of the lethality of the histone deacetylase inhibitor LAQ824 by the cyclin-dependent kinase inhibitor roscovitine in human leukemia cells" (Rosato et al 2005), in which a ten-lane Actin loading strip controlled for nine lanes of the proteins of interest.

In the manner of kaleidoscope aesthetics, a repertoire of graphical elements served to illustrate a sequence of observations by rearranging them into new combinations.

Again, commentators were reduced to an abyss of abject despair:
#4 Peer 2
Are there any end to this tragedy?

So we turn quickly to "LBH-589 (panobinostat) potentiates fludarabine anti-leukemic activity through a JNK- and XIAP-dependent mechanism" (Rosato et al 2012):



Accidental conversion of a strip of Western Blot to 'tubulin' or 'actin' or 'EF-1α' could happen to anyone... but when images have been fabricated, or manipulated to substitute for bona fide documentation of an experiment, it undermines ones' faith in the substantive claims.

I originally set out to include Prof. Paul B. Fisher of VCU in this strip-mining survey of PubPeer threads, but space does not permit. As well as co-authorship on 22 entries of the Dent / Grant corpus, he has his own oeuvre of 15 non-overlapping entries. Perhaps there could be a sequel if the readers clamour for one. To pique your interest and whet your appetites, Condylocarpon amazonicum an anonymous Peer prepared some diagrams.




They look like a colored-arrow factory exploded but they are intended to show the reuse of Blots and cell-culture images within and between a trilogy of papers. One Figure has six panels of cells under different conditions but after taking account of all the reflections, rotations and enlargements, these prove to be overlapping details of only two source microphotographs.

"mda-9/Syntenin regulates the metastatic phenotype in human melanoma cells by activating nuclear factor-kappaB" (Boukerche et al 2007);
"mda-9/Syntenin promotes metastasis in human melanoma cells by activating c-Src" (Boukerche et al 2008);
"Src kinase activation is mandatory for MDA-9/syntenin-mediated activation of nuclear factor-kappaB" (Boukerche et al 2010).

First author on all three, Habib Boukerche, has his own PubPeer presence (and a retraction). Any sequel could also look at the accomplishments of frequent Grant / Dent / Fisher collaborator Devanand Sarkar... stepping-stones to his current eminence as Distinguished Professor, Endowed Scholar and Director of Education & Training at VCU.

#13 Peer 1
As for those 11 Dasmahapatra papers, and the ORI report three years ago that interested parties agreed that they were unsustainable, three have been retracted. Extrapolating this rate, the last one will be amended or withdrawn from the literature by 2027. Currently eight seven six five of them continue to be read and to accrue citations while other cancer researchers design their own investigations around them, for there is nothing to warn the unwary reader of their false-data nature, not even an Expression of Concern. The reasons for this editorial inaction at Blood and Molecular Cancer Therapeutics are unfathomable.
Posted by Smut Clyde at 3:18 PM

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)