ABOVE: Accusations of alleged image duplication and reuse in papers by a prominent researcher at Harvard Medical School are currently under investigation. ©iStock, jxfzsy
In early February, microbiologist and science integrity consultant Elisabeth Bik released reports from her latest investigation into a suspected case of scientific misconduct. In her report, she outlined 59 instances of alleged image duplication or reuse in papers published by Khalid Shah, a neuroscientist at Harvard Medical School and its teaching hospital Brigham and Women’s Hospital. Shah uses gene editing approaches to develop cell-based therapeutics for brain cancers.
According to Bik, she started her investigation following a tip from someone who was familiar with Shah’s laboratory. She looked into potential research misconduct and claimed to find image reuse or duplication problems across 29 papers, which he published in 21 different journals between 2001 and 2024. Shah is listed as the first, second, or corresponding author on 25 of the 29 papers.
Following Bik’s report, Harvard Medical School and Brigham and Women’s Hospital opened an official investigation into the claims. Shah declined an interview, but deferred to Paul Anderson, chief academic officer at Mass General Brigham, who said in an emailed statement, “We take very seriously any questions, concerns, or allegations regarding research conducted at our hospitals and undertake a robust and confidential process to assess and respond to any claims that are brought to our attention in accordance with hospital policy and federal regulations.”
To check for potential image problems, Bik employed the AI-based image analysis tool ImageTwin, which uses proprietary software to compare images to a database of images taken from publicly available scientific articles.
"[The case] covers a range of problems,” said Bik. “Most of these are duplicated panels from mice or from cells or tissues." For example, in a 2020 paper published in Scientific Reports, Bik said that two microscopy images looked identical but were used to represent two different treatment conditions.1
Many of the papers that Bik examined appeared to duplicate images from the same paper or reuse an image from one of Shah’s previous publications. However, while she was cross-referencing images in a paper published in Nature Communications in 2022, she noticed that some of the images looked like they were taken from publications from outside of Shah’s team.2 For example, Bik said that two images labeled as glioblastoma tissue looked identical to higher-resolution images in a 2010 paper published in PLoS ONE on the topic of head and neck squamous cell carcinoma.3 None of the authors of the 2022 Nature Communications paper authored the 2010 PLoS ONE paper, and Shi-Yong Sun, a cancer biologist at Emory University School of Medicine and corresponding author of the 2010 PLoS ONE paper, said that he’s not familiar with Shah’s laboratory, and he doesn’t know the study authors.
“It looks similar, but I don't know,” said Sun. “This is my first time [hearing about] this kind of case if they [did] indeed copy our image.” He added, “I was surprised.”
“The images that [Bik] highlighted as duplicated indeed appear to be duplicated,” said Mike Rossner, an independent biomedical image consultant at Image Data Integrity, who reviewed the allegations related to the 2022 Nature Communications paper. Rossner added that many of the images appeared to have been altered, most commonly cropped or contrast adjusted. He added, “At least as presented, all of those manipulations have affected the interpretation of the data.”
In an email, Chris Graf, research integrity director at Springer Nature, which publishes Nature Communications said, “We are looking into the paper carefully following an established process and in line with COPE best practice. While we endeavor to complete our investigations as quickly as possible, we do need to ensure that we follow due process.” Springer Nature is a member of the committee on publication ethics (COPE), an advisory body that provides journal editors with core practices and guidance on a variety of topics, including the handling of allegations of misconduct.
On February 2, 2024, journal editors added a note to the 2022 Nature Communications paper informing readers of an ongoing investigation into concerns raised about the reliability of the paper’s data.
Bik’s investigation did not end with ImageTwin. She decided to try a reverse image search on Google, which uses a specific image to drive search results. Although the search tool struggles with science images, which are typically presented as a compilation of images, schematics, and bar graphs, she decided to run a few images. “I had not expected anything, but sure enough, there was one that came up,” said Bik. In the end, Bik identified three images published on vendor websites. The vendors did not respond to requests to confirm whether either company licensed the images from Shah.
Rossner cautioned against drawing conclusions about the alleged duplication as well as the severity of the duplication. “We'll always start from the perspective of giving the author the benefit of the doubt that the duplication could be the result of clerical error.” However, he added that the authors should provide the correct source data. “Without that level of substantiation, then a single duplication that affects the interpretation of the data, depending on how it affects the overall conclusions of the paper, could be very serious,” he added.
The growing availability of image analysis tools will likely increase the number of post-publication image problems detected by science sleuths like Bik, sounding the alarm for academic journals to hear. “If you're a journal, and you're worried about these things, it makes sense to find them before publication, because once they're published, and somebody complains about it, it is a bit of an embarrassment for the journal, but it's also much more work,” said Bik.
Graf noted Springer Nature’s plans to integrate image analysis software into their publishing pipeline. “We take our responsibilities to maintain the scientific record extremely seriously, and we are committed to maintaining the highest level of integrity in the content we publish. As part of that commitment, in addition to current checks, we’ve been developing in-house AI image integrity software that checks submitted manuscripts for suspicious image duplications, and this is currently in testing. The software will also check for instances of plagiarized images and flag these suspect issues to our editorial and integrity staff who can then take appropriate action.”
Seven of Shah’s papers published since 2020 appeared in Bik’s report and appeared in the Springer Nature journals Nature Communications, Scientific Reports, and Oncogene; Cancer Biology and Therapy; Stem Cells Translational Medicine; and Frontiers in Immunology.
The press office of Taylor & Francis, which publishes Cancer Biology and Therapy, and the editorial office of Stem Cells Translational Medicine confirmed that they are aware of the concerns and are actively investigating the situation. The editorial office of Frontiers in Immunology did not respond to a request for comment.
- Bhere D, et al. Simultaneous downregulation of miR-21 and upregulation of miR-7 has anti-tumor efficacy. Sci Rep. 2020;10:1779.
- Bhere D, et al. Target receptor identification and subsequent treatment of resected brain tumors with encapsulated and engineered allogeneic stem cells. Nat Commun. 2022;13:2810.
- Elrod HA, et al. Analysis of death receptor 5 and caspase-8 expression in primary and metastatic head and neck squamous cell carcinoma and their prognostic impact. PLoS One. 2010;5(8):e12178.