For decades, cancer has been thought of as a purely human disease – rogue cells multiplying out of control, with no room for anything else in the picture. But a growing body of research suggests that isn’t quite right. Some tumours, it turns out, come with company: communities of bacteria, viruses and fungi living on, between and even inside the cancer cells themselves.The trouble is that nobody has been entirely sure which cancers actually have this so-called microbiome, and which don’t. The field has been dogged by contradictory claims, competing methods and – in one particularly damaging case – a retraction, after results from a high-profile study could not be replicated.Since then, the field has been left without a clear way forward. Every research group has used its own methods and level of rigour, and there has been no agreed-upon benchmark to check new findings against. That matters because the stakes are high. If microbes really are helping some cancers grow, resist treatment or spread, they could become new targets for screening and drug development. But chasing signals that turn out to be false wastes time, money and precious patient samples.Our team set out to settle the question properly, using the largest collection of cancer genetic data in the world – Genomics England’s 100,000 Genomes Project, which includes DNA from more than 16,000 tumours. We built what we believe is the most rigorous analysis pipeline yet developed for this kind of work, designed to strip out every source of error we could identify, then applied it to the entire dataset.Our latest research found that most cancers – including those of the brain, breast and kidneys – lack a microbiome that is distinguishable from background. This suggests that earlier studies that had picked up microbial signals in these tumours may have been affected by contamination: stray DNA from laboratory equipment or even the scientists handling the samples.But some cancers were different. Tumours of the mouth, oesophagus, stomach and bowel showed clear, consistent evidence of microbial life. And it wasn’t just bacteria. We found viruses, fungi and archaea (organisms similar to bacteria but genetically distinct) living within these tumours. In some cases, we detected trichomonas, a single-celled protozoan parasite. The particular mix of species varied depending on where in the digestive tract the cancer was and was linked to features such as the cancer’s subtype and how many genetic mutations it carried.Telling real microbes from contaminationWorking out which of these microbial signals were genuine and which were laboratory contamination was the hardest part of the project. Sequencing a tumour means reading every strand of DNA in the sample, human and non-human alike. Most cancer researchers simply ignore the non-human portion. We did the opposite. We discarded the human DNA and matched everything left over against known microbial genomes to see what was hiding there.However, this approach can run into problems fast. There’s no single, definitive human genome to measure against – everyone’s DNA differs slightly, and even the best reference genomes have gaps. Any leftover human sequence that happens to resemble microbial DNA can be wrongly flagged as a hit. Then there are errors in the microbial reference libraries themselves – occasionally the wrong species ends up catalogued, or DNA from a lab technician’s skin ends up mixed in with a sample. And however carefully a lab operates, some contamination during tumour preparation is almost unavoidable.We tackled each of these problems in turn. We filtered aggressively against multiple versions of the human genome, stripping out anything ambiguous or repetitive. We used the most up-to-date DNA-matching software against carefully curated microbial databases. Sample contamination happens easily in a lab. Komsan Loonprom/Shutterstock.com To catch contamination, we compared which microbes turned up across different cancer types: species that appeared everywhere were almost certainly picked up in the lab, while species confined to just one or two cancer types were more likely to be real. Sure enough, several of the culprits we filtered out were common skin bacteria found in every cancer type – probably from the researchers who had handled the samples.This kind of large-scale, painstaking filtering was only possible because of the sheer size and quality of the Genomics England dataset. Smaller studies simply don’t have enough samples or resolution to distinguish a genuine biological pattern from a one-off contamination event.We’ve now made our data freely available as downloadable software, along with a list of the microbial species we’re confident are genuinely present in these tumours, so other researchers can apply the same rigorous approach to their own data. The hope is that this draws a line under years of conflicting claims. Scientists can then focus their efforts where the evidence is strongest. That means tracking how these microbial communities in mouth, throat, stomach and bowel cancers might influence how tumours develop and how well they respond to treatment. Ultimately, it could help these cancers be diagnosed and treated earlier.Henry Wood receives funding from Cancer Research UK, National Institute of Health Research.Anders Dohlman receives funding from The Damon Runyon Cancer Research Foundation and from Cancer Research UK.