NEW YORK (GenomeWeb) – A single-molecule sequencing study published in Nature Nanotechnology two years ago by researchers at the National Chiao Tung University in Taiwan was retracted earlier this month.
In its retraction notice, Nature Nanotechnology said the paper was retracted because the authors could not reproduce the data in a "reasonable timeframe." However, a number of researchers are concerned that not only is the data not reproducible, but that it was entirely fabricated.
"All the statistics correspond to data that have been produced by a computer, not an experiment," Vincent Croquette, research director of the ABCD biophysics lab at the École Normale Supérieure in Paris, told GenomeWeb. "It corresponds to what we'd expect to be a fraud."
Neither Nature Nanotechnology nor researchers at the National Chiau Tung University returned GenomeWeb's requests for comments.
When the study was initially published in 2013, experts raised concerns over many of the paper's claims. Stuart Lindsay, director of the Center for Single Molecule Biophysics at Arizona State University and several colleagues from ASU and other universities, sent a list of questions to the paper's authors. Lindsay said that that author's reply left more questions than answers, prompting the group to bring its concerns to Nature Nanotechnology.
In a note accompanying the retraction notice, the journal said that following the scientists' concerns, it had contacted the National Chiao Tung University, asking it to launch a formal investigation.
In their original study, the researchers claimed to use a protein transistor composed of immunoglobulin G and two gold nanoparticles bound to source and drain electrodes. According to the authors, the DNA polymerase phi29 is used to incorporate nucleotides, which generates a distinct electrical signature that is detected by the protein transistor.
However, researchers had concerns with the physics, enzymology, and chemistry of the paper. After Lindsay's group raised initial concerns over the study with Nature Nanotechnology, the journal conducted its own investigation, and asked a couple of researchers to review the study, including ENS' Croquette.
Croquette contacted the authors and requested to see the researchers' raw data. He said the authors told him that data from one of the figures had been deleted, but provided him the raw data from two of the figures in the original paper and one figure in the supplementary information. He said an analysis of that data revealed that it had been fabricated.
"When you look at the raw data, it's quite obvious," Croquette said.
Other researchers who have also reviewed the data agree with him. "The data bear all the signs of being computer-generated fakes," Lindsay told GenomeWeb.
Their conclusions stem from three main problems: the statistics associated with the signal and noise, the polymerase activity, and the transmission electron microscope photo that shows the setup of a single DNA molecule bound to the polymerase.
After reviewing the raw data, Croquette wrote to Nature Nanotechnology, detailing his concerns. In his letter, he focused primarily on the problems with the signal-to-noise ratio, which he said was "too regular."
Typically, noise follows the bell-shaped curve known as Gaussian distribution. In a Gaussian distribution most events congregate around the mean, but there will still be some rare events that occur far from the mean. But, Croquette said that the data in this paper had a triangular shape, which is a "completely bound distribution," so that there are no rare events far from the mean. According to Croquette, scientific experiments do not produce these types of distribution, but they can be computer generated.
"I've studied lots of noise," Croquette said. "If you want to produce this kind of shape, you use a random number generator."
The signals themselves were also too regular with no variability, Croquette said. In addition, the signal can be completely separated from the noise — something most researchers "only dream of doing in experiments."
In order to compare the authors' technique with another single-molecule sequencing technology, Croquette said he looked at data from Pacific Biosciences' RS II instrument, which also uses phi29 for nucleotide incorporation. The data is messier due to the expected stochastic nature of the signal and Gaussian distribution of noise. Croquette said from his years' experience of working on single-molecule sequencing technology, "this is what you'd expect."
Taekjip Ha, a professor of physics at the University of Illinois, told GenomeWeb that he had seen the paper before it was published in Nature Nanotechnology when the authors submitted it to a different journal. The journal asked Ha to pre-evaluate the study before deciding whether to subject it to a full peer review. "My feedback was, yes, it's exciting, but it cannot be true," Ha said.
For Ha, the smoking gun was the polymerase. The authors used the phi29 polymerase to incorporate nucleotides onto a DNA template. The protein transistor measures the conductance of the polymerase, which gives a different signal depending on which base is incorporated.
Typically, polymerases do not move along DNA in regular, stepwise fashion. "Usually, you see randomness," Ha said, but their data "looked almost like a metronome."
Meni Wanunu, a principal investigator in nanoscale biophysics at Northeastern University, agreed. "The enzyme seems to work like clockwork. … It's strange to see that and does not strike me as something real."Wanunu and Lindsay were also both suspicious that the electron microscopy images in the study showing the polymerase bound to the DNA were not real. "It appears to be doctored," Wanunu said. "I've never seen an image of DNA completely focused like that," he said.
The images "were obviously Photoshopped," Lindsay added. Given all the concerns with the paper, Wanunu said he "wonders all the time" how it was accepted for publication in the first place.
"As soon as the paper came out, there were plenty of complaints," Croquette added, "suggesting that the referee process did not go well."
The researchers are also concerned that the study built off a previous paper published by the same group in 2012, also in Nature Nanotechnology, that described the same protein transistor used in the 2013 retracted paper.
Croquette said that he has not examined that paper in as great as depth and that the electrochemistry described in the paper is not his area of expertise. Nevertheless, he said there were a few elements struck him as strange. For instance, the authors describe applying 5 volts of electricity to a nanotransistor that is in water. But, Croquette said that voltage above 1 to 1.5 volts would electrolyze water.
Wanunu said he too had seen the previous paper but had not analyzed it in as great of depth. Similar to the retracted paper, however, that one also had "beautiful data," with limited explanations of how the data was obtained.