The Portuguese cancer researcher Sonia Melo has now achieved the status of a zombie scientist. After an internal investigation which records are kept secret, she was cleared of all suspicions of scientific misconduct and re-installed as group leader at the Instituto de Investigação e Inovação em Saúde (I3S) in Porto (see my report here). This despite an impressive PubPeer record of data integrity concerns, and despite the fact that the European research society EMBO revoked Melo’s Installation Grant funding after having determined problems with her publications. EMBO nevertheless stick to their decision, but Melo’s Portuguese funders like Fundação para a Ciência e a Tecnologia (FCT) apparently see absolutely no need to reconsider their support, certainly not after the I3S whitewashing. Melo previously had to retract a paper (Melo et al, Nature Genetics, 2009) for data manipulations, her other works were however found not problematic by the I3S commission. In two papers in Cancer Cell (Melo et al 2010 and Melo et al 2014), the alleged duplications were apparently proven not to be duplications. As I learned, this was probably because while the top part of the gel images indeed did look suspiciously similar, the lower parts were clearly different. A possibility of digital image splicing was not considered, as it seems. In any case, even if the top bands are indeed the same, it doesn’t really matter. Cell editorial offices made on several occasions perfectly clear that data integrity is not one of their top concerns.
Two corrections were recommended by I3S experts, for Melo et al, PNAS 2011 and Melo et al, Nature 2015. The problem with the latter paper is however: it seems to be utterly irreproducible. This discovery of Glypican-1 protein as a specific and unique marker of early stages of pancreatic cancer to be detected in the patients’ blood was made in the lab of Melo’s former boss Raghu Kalluri at MD Anderson in Texas. It brought hope to many cancer patients, but also to biotech investors. Kalluri’s company Codiak Biosciences raised at least $80 Million in venture capital to market his and Melo’s patented discovery, with direct involvement of MD Anderson. It is reasonable to assume that the investors might have gotten second thoughts should Melo or even Kalluri be found guilty of misconduct. They might even be tempted to ask for their millions back. In this regard, one kind of understands why MD Anderson never bothered to investigate the data integrity concerns in Kalluri’s publications with Melo and also without her. It might even explain why the I3S investigation found something that different from that by EMBO, while analysing exactly the same papers by Melo.
Below Ana Pedro, a former peer of Melo, offers her post-publication peer review of that Nature 2015 publication. She shows why it is unlikely for Glypican-1 to be specifically detectable on the cancer cell exosomes (cell-membrane-derived vesicles which our cells shed into their surroundings and into blood). Pedro analysis is also supported by the fact that Melo et al discovery is primarily focused on the use of excessively high concentrations of a certain polyclonal Glypican-1 antibody which sale has been discontinued since. There is no published evidence that the central claims of the Nature 2015 paper were ever reproduced using any other Glypican-1 antibody either.
An anonymous commenter calling himself “Frank” (who soon turned out to be a senior researcher at I3S associated institution Ipatimup, where Melo is also employed) took as stand defending the investigation and Melo’s research integrity in the comment section on my site. He eventually abandoned it as his forum and apparently moved to join the mudslinging affray at PubPeer, assisted by one or more other anonymous defenders of Melo’s research. As evidence for the reproducibility of Melo’s Nature paper, “Frank” kept pointing to this publication from the nanotechnology engineering lab of Thomas Thundat at the University of Alberta, Canada: Etayash et al, Nanoscale, 2016. Another reader of my site countered:
“Regarding the publication in Nanoscale: There are no results showing specificity for the anti-GPC1 antibody they are using. Without any documentation showing antibody specificity in their cell model, the data is in my eyes worthless”.
In fact, even Thundat himself was not really defending the specificity of his assay. In his emails to me he spoke of another paper “from a different group” he saw, “about real patient samples” (which is not published yet) as well as “some very exciting results with mouse” which he was recently shown by his visitors from US, which he then admitted were “preliminary”.
To my question about the discontinued Glypican-1 antibody by ThermoFisher Thundat declared:
“The people whom I am going to collaborate with are going to send us the antibodies/receptors they developed. Our focus is developing readout technology for high sensitivity, real-time multiplexed detection. Specificity is going to come from detecting multiple biomarkers simultaneously. From experiments we have to determine the best combination of biomarkers”.
With this we probably can dismiss Thundat’s publication as proof of reproducibility of Melo et al seminal work in Nature.
What else? The Nature 2015 paper describes the comparative analysis of cancer serum samples versus those from healthy donors. The experimental methods description however does not make it entirely clear if both cancer and healthy samples were processed equally. We do learn that the cancer samples originated from the University of Heidelberg and the University Hospital of Dresden in Germany, while the paper’s last author Raghu Kalluri suggested in an prepared statement elsewhere those blood sera were up to 30 years old. I attempted to contact his and Melo’s responsible co-authors from the Dresden University, namely Juergen Weitz, Nuh Rahbari, Christoph Reissfelder, and Christian Pilarsky, with simple questions about the age, origins, preparation methodology and storage conditions of the cancer and healthy control samples. All rather reasonable questions, which should have been already addressed in their published Materials and Methods section. Especially since blood does tend to deteriorate rather quickly, which certainly calls for tightly controlled experimental conditions when comparing different sample groups. I received no reply from any of these four German clinicians. Even my reminder email was met with thundering silence. Shall we therefore assume that the healthy controls were indeed relatively fresh and the cancer samples were indeed up to 30 years old? At which temperature were they stored during that time? Were they actually any good for such a comparative exosome analysis? Does it matter to the authors that the Glypican-1 test they are now commercially developing might randomly assign false-positive pancreatic cancer diagnoses to perfectly healthy patients?
Our traditional academic culture teaches that a published paper must never be questioned, because it already was peer reviewed. Especially a paper published in the apical journal Nature, where the editorial scrutiny and peer review is supposed to be the most rigorous. As my former boss like to say to his critics: once you also published in Nature, then you can come questioning my paper there. Indeed, when doubts about published elite research arise anyway, the etiquette dictates that only elite researchers who themselves regularly publish on the same journal impact factor level are entitled to express a cautious opinion. Even then, their letters to editors often land in the trash bin.
However, to tell it back to our science elite with a Nobel-Prized quote: “the times they are a changin’”. Here is a post-publication peer review which is based on evidence, not eminence.
Is GPC1 from serum exosomes a marker to diagnose pancreatic cancer?
Ana Pedro, MPharm, PhD
Extracellular vesicles (EVs) are nanometer-sized membranous vesicles which are involved in cell-to-cell communication. EVs contain several types of functional molecules, such as proteins, mRNAs, and microRNAs (miRNAs). Increasing evidence suggests a key role for EV-mediated intercellular communication in a variety of cellular processes involved in tumor development and progression, including immune suppression, angiogenesis, and metastasis. Therefore, EVs are emerging as potential therapeutic targets in cancer therapy.
The publication by Melo et al., Nature 2015, claimed that presence of glypican 1 (GPC1) in serum exosomes unequivocally identifies patients with pancreatic ductal adenocarcinoma (PDAC) cancer. As PDAC has an extremely poor prognosis and currently no robust biomarkers for the early stages of the disease are available, the study by Melo et al. carried great promise. However, PubPeer commenters have raised serious concerns about the specificity of the polyclonal anti-GPC1 (ThermoFischer, catalogue PA5-28055) used in this publication. These concerns were further fuelled by the discontinuation of this antibody preparation, since the validity of the presented results depends entirely on the fidelity of those antibodies.
Considering the clinical impact of a valid PDAC marker and the grave concerns raised by parts of the scientific community on PubPeer, I here confront the data published by Melo et al. with GPC1 expression data available in several databases such as The Human Protein Atlas, PeptideAtlas, and Vesiclepedia, as well as in publications, and found important discrepancies.
GPC1 expression levels in healthy pancreas are only moderate.
By consulting The Human Protein Atlas, I found that in normal pancreas GPC1 transcripts are not highly expressed compared to other tissues (Fig. 1a) and, coherently GPC1 protein is not detected by immunohistochemistry (IHC) in this tissue (Fig. 1b). In agreement, mass spectrometry analyses summarized in PeptideAtlas for GPC1 presence in adult human tissues reveal only a modest amount of GPC1 in the pancreas (Fig. 2). Thus, consistently across databases, there are no indications that GPC1 expression in the normal pancreas is high.
GPC1 expression levels in PDAC pancreas are only moderately elevated.
Several authors found elevated GPC1 transcription levels in PDAC pancreas compared to healthy pancreas. However, in most cases the observed increases are only few-fold (Kayed et al. 2006; Gruetzman et al., 2005; Kleeff et al., 1998), as summarized in Table 1. According to The Human Protein Atlas these increases are too modest to make pancreatic cancers sites of high GPC1 expression compared to the rest of the body, because based on IHC analysis they consider the GPC1 expression level as only “medium” in 1 of 11 samples investigated, “low” in 2 samples, and “absent” in 8 samples (quantitative scaling like for Fig. 1b). This modest upregulation is also consistent with IHC analyses by Duan et al, 2013 (Table 1) who found 2 of 16 healthy pancreases, and only 35 of 63 PDAC pancreases, to be positive for anti-GPC1 staining (Table 1).
There are no reasons to believe that GPC1 is enriched in PDAC-derived EVs.
Finally, by consulting Vesiclepedia, I found that extracellular vesicles (EV) of cancerous as well as of non-cancerous cell types may contain GPC1 (Table 2) and that the inclusion of GPC1 into exosomes is not a hallmark of a smaller degree of differentiation as, e.g., EVs of differentiated keratinocytes contain more GPC1 than EVs of undifferentiated keratinocytes (Table 1 in Chavez-Muñoz et al, 2009). The latter observation is of relevance because PDAC tumour cells are considered to be less differentiated than their healthy precursor cells. Figure 3 shows that Melo et al. could readily detect GPC1 in lysates of various cell lines by immunoblot experiments. In contrast, this analysis of the matching exosomes from those cell lines probably suffered from severe background staining (Fig. 3), arguing against relative enrichment of GPC1 in exosomes as compared to cells. There were no arguments provided to trust that the band in this immunoblot from Melo et al. (green arrow in Fig. 3) is specific and the sole indicator of GPC1 presence. However, even so, then it should be concluded that, considering the amounts present in the cell lysates, GPC1 was most efficiently incorporated in the exosomes from the non-cancerous primary human dermal fibroblasts (HDF, Fig. 3).
There are no reasons to believe that PDAC tumors release particular high numbers of EVs.
An average healthy pancreas is about 80 gram in weight, which is very small compared to the rest of the body. This weight may not be so different in the early stage of pancreatic cancer, the stage which Melo et al. suggest to be able to detect. Furthermore, PDAC tumours are poorly vascularized, and so far no evidence has been presented that these tumour cells release more EVs than other cell types. Melo et al. compared the exosomes of several non-tumour cell lines with those of pancreatic tumour cell lines, and would have noted if there were considerable differences in EV numbers. Summarized, there is no reason to believe that a sufficiently large percentage of the exosomes in sera from patients with early stages of PDAC are derived from tumour cells.
Investigation of expression data in public repositories and the literature indicate that:
PDAC tumor cells:
- do not express particular high amounts of GPC1 compared to the rest of the body,
- do not selectively enrich GPC1 into their exosomes, while other cells do not selectively exclude GPC1 from their exosomes,
- do not release particular high numbers of EVs.
Consequentially the detection of GPC1 in serum exosomes is a very unlikely diagnostic marker for pancreatic cancer. Together with the serious technical concerns regarding the Melo et al. study (discussed on PubPeer), their promising message unfortunately has important unresolved issues.
Possible conflict of interest
I am an aspiring scientist who worked on exosome analysis for several tumour models. Against my desire my career was halted, and I feel that was in part because I could not reproduce several exciting messages like those of the Melo et al. study. With the here presented analyses of databases and literature I hope to show that specific detection in serum of tumour-derived exosomes based on GPC1 presence is not within expectations. I hope this will partly redeem my work, which I did with a lot of passion.