A Revolutionary Microbiome Paper From 4 Years Ago Has Just Been Retracted

Can unique bacterial signatures help predict if you have cancerous tumors? Maybe not.

Paper no longer found (well, it’s still there, but with a big “Retracted” notice at the top).Photo byAitoff/Pixabay

Bacteria thrive in all sorts of unusual places. We all carry around trillions of them in our intestines, for example, creating a diverse community that is known as the gut microbiome.

But what about in other places in the body — like in cancerous tumors? Could we check for cancer by seeing whether we detect specific tumor-dwelling bacteria?

A high-impact paper published in the prestigious journal Nature in 2020 suggested the answer might be yes.

But now, four years later, that paper has been retracted, as other researchers noted errors and inconsistencies in the data and the findings.

Searching for microbes in the unlikeliest of places

We’ve found bacteria in rocks mined from miles underground, in the deepest and darkest parts of the ocean, andeven in the clouds overhead in the sky. But what about in cancers, inside our own body?

One of the big advantages to next-generation sequencing — the ability to analyze huge amounts of DNA at once — is that it makes searching for bacteria much easier. We can use chemicals to extract all the DNA from every cell in a sample, and then put it all through a sequencing machine. This gives us tens of millions of small DNA fragments in digital format, and we can use computer algorithms to match those fragments up to our reference sequences.

This mass-sequencing method is a huge step up from previous methods because we can search blindly. Previous testing methods required us to know, ahead of time, which bacteria or cells we were hunting. Now, we can map our results against a database of millions of organisms to identify what we’ve got.

In 2020, this is exactly what researchers did. They looked at DNA samples from a collection called The Cancer Genome Atlas, or TCGA. These were previous sequencing runs on various tumors, sequenced to a high depth (the researchers kept the sequencing machine running on the data over and over, to make sure that they caught everything).

And, incredibly, the researchers found bacteria.

Some of the DNA sequences matched to bacteria that we’d previously found and studied elsewhere. And furthermore, the bacteria were specific to the tumor type; bacteria from healthy individuals were different from those found in lung cancer patients, which differed from those found in prostate cancer, etc.

This held huge promise. Imagine if we could run a blood test at your annual check-up, looking for cancer-indicating microbes. If the doctor found cancer-linked bacteria, they could refer you to a specialist — even if they hadn’t yet seen a noticeable tumor.

We could possibly detect cancer earlier, saving lives.

But there are issues with this sequencing approach, called false positives.

Ghost signatures

Most DNA sequencing performed these days is a method called shotgun sequencing, because we chop the DNA up into lots of little pieces and sequence each piece individually. After it’s all converted into 1s and 0s on a computer, we use powerful computer algorithms to assemble the short fragments into longer sequences by looking for overlaps.

It’s a very effective method. But it’s not perfect.

Some of these issues were called out by other researchers, eventually leading to the retraction of the 2020 tumor bacteria finding paper. They include:

• Contamination of reference data. Imagine that you’re building a reference database of bacteria DNA — but you accidentally include one human DNA sequence mixed in there.
  Oops!
  This is a huge problem, because this means any human genome would register a hit against your supposedly-all-bacteria database. It’s a false positive; it’s seeing bacteria where there aren’t any.
• Implausible matches. Researchers went back to see what those distinctive bacteria found in the tumors were.
  And, strangely, many of those bacteria hadn’t ever been found in connection with humans before this paper. They were usually associated with plants, or sometimes even with deep-sea ocean life.
  How would those bacteria get into humans to take up residence in cancer samples?
• Getting lost in the noise. If you have a sample with tens of millions of reads, a few of those sequences will, by pure random chance, match with strange creatures that aren’t in your sample. If I sequenced a 100% pure sample of your DNA, a few of those short little DNA fragments would, by pure chance, happen to match to an elephant. Or a shark. Or a starfish, an animal who doesn’t really even have a brain.

I’m sure you have a brain, and a very nice one if you’re reading this article! But it can be difficult to identify a threshold in sequencing data, when to say “this isn’t just random noise, this is an actual significant match.”

• Biological contamination. Some of the microbes that were found in the samples were species that are normally found on human skin or in the human gut, and they were found in a wide selection of the TCGA samples.
  It’s likely that these might have been accidentally introduced into the samples from the human scientists who collected the samples and performed the DNA extractions.

So what does this mean for me? What happens now?

If you’re a microbiome researcher, this is frustrating but not totally unexpected. Contamination and false positives are a persistent problem in microbiome research, and this is a sobering reminder that, if you look hard enough at a dataset, you can find some slight evidence for many incorrect conclusions.

For the researchers who published this paper, there’s no punishment, although a retracted paper can be seen as a black mark on an academic record. But there’s no evidence that the original authors knowingly misled or lied, and it shouldn’t be seen that way.

There’s still tons of useful information in blood samples, and there are still ways to check for signs of cancer using a blood sample. Lots of companies are still pursuing this, because it really would be a game changer when it comes to early cancer detection.

Blood Poison: How We Detect Cancer in Blood

When a doctor takes a blood sample to check for cancer, here’s what they’re looking for

medium.com

For most of us, this just means that we shouldn’t neglect traditional methods of cancer screening. That includes regular exams, mammograms for women, prostate exams in men, and regularly checking your own body, watching for unusual lumps or moles that may have changed color or shape.

But this is also a warning about trusting uncertain evidence because it supports something that we want to be true. It would be amazing to find that we can detect the signatures of specific bacteria to indicate the presence of cancer.

*

Science isn’t just about creating new hypotheses; it’s also about testing those theories to confirm whether they are correct, or to challenge them if they appear to be false. It can be contentious, because no scientist wants to hear that their work — their published, peer-reviewed, much-effort work — might not be true.

There’s still some ongoing drama about this retraction on Twitter (or drama when it comes to academics, which is ‘polite quibbling’ to everyone else), but for the rest of the microbiome community, this is an important reminder to be cautious about results that stray too close to background noise.

And finally, a well-deserved hat tip to Steven Salzberg and his microbiome research lab at Johns Hopkins University, who uncovered these inconsistencies and worked to correct the public record.

--

Want to keep up to date? Follow and download the app for free.