Rigorous Testing Reveals Weak and Inconsistent Microbiome Signals in Lung Cancer Tumours

For more than a decade, scientists have been tantalised by the possibility that tumours carry their own distinctive microbial fingerprints, signatures that might one day be used to diagnose cancer from a simple blood draw or to guide treatment decisions. A new study of patients with lung cancer, however, delivers a sobering reality check. Researchers at Dalhousie University in Halifax, Canada, set out to characterise the bacterial communities associated with lung tumours, adjacent healthy lung tissue, blood, and saliva from patients undergoing curative surgery for non-small cell lung cancer, and their findings suggest that many of the microbial signals reported in low biomass samples may be far less robust than previously assumed.

The research team, led by Vanessa DeClercq and Morgan G. I. Langille of Dalhousie University’s Faculty of Medicine, designed their study specifically to confront one of the most persistent controversies in microbiome science: whether bacteria genuinely reside within tumours and circulate in the bloodstream of cancer patients, or whether the DNA sequences detected in such samples are artefacts of laboratory contamination and the extraordinary sensitivity of modern sequencing technology. To do this, they applied an unusually comprehensive battery of molecular techniques to every sample, including full-length 16S rRNA gene amplicon sequencing, nested short-read 16S amplicon sequencing, metagenomic shotgun sequencing, and digital PCR for absolute quantification of bacterial DNA. Crucially, every sample was processed in duplicate, providing technical replicates against which the reliability of each measurement could be judged.

The results were striking in their asymmetry. When the researchers attempted full-length 16S gene sequencing on their lung tumour, adjacent lung tissue, and blood samples, the technique failed to generate usable data for nearly all of these specimens. This was not a subtle signal-to-noise problem but a fundamental inability of the method to amplify meaningful bacterial genetic material from samples that contained, in effect, almost nothing to amplify. In contrast, saliva samples and positive control specimens, which harbour rich and genuine microbial communities, yielded abundant, high-quality sequence data with strong reproducibility between replicates, confirming that the laboratory pipeline itself was functioning correctly.

A second strategy, nested short-read 16S sequencing, did manage to produce sequence data from the low biomass samples, but what it produced was far from reassuring. The tumour, adjacent tissue, and blood specimens contained very few bacterial taxa, and even those few varied dramatically between technical replicates of the same sample. When the researchers compared the taxonomic composition of these samples with their negative controls, which track DNA introduced during sample collection and laboratory processing, the low biomass specimens proved compositionally indistinguishable from the controls. In other words, there was no reliable way to tell a genuine tumour-associated bacterial community from background contamination picked up along the way.

Digital PCR provided the quantitative backbone for these observations. Unlike standard sequencing, which describes the relative proportions of different bacteria but says little about how much DNA is actually present, digital PCR counts target molecules directly, delivering an absolute measurement of bacterial biomass. The measurements confirmed what the sequencing results implied: lung tumour, adjacent tissue, and blood samples contained extremely low quantities of bacterial DNA, comparable to the levels found in negative controls. Saliva samples, by contrast, carried orders of magnitude more bacterial DNA. This quantitative gulf between genuinely microbe-rich samples and the near-sterile tumour and blood specimens goes to the heart of the technical challenge, because at such low biomass, even trace amounts of contaminating DNA from reagents, laboratory environments, or collection procedures can dominate the apparent microbial profile.

Metagenomic shotgun sequencing, which reads DNA across the whole genome rather than targeting a single gene, told a consistent story. In the lung tumour, adjacent tissue, and blood samples, this approach detected very few taxa, and, importantly, there was minimal overlap between the taxa identified by metagenomic sequencing and those identified by 16S sequencing from the same specimens. If the detected bacteria reflected real communities resident in the tumours, one would expect at least reasonable agreement between independent methods probing the same DNA. Saliva and positive control samples, in contrast, showed substantial overlap of detected genera across methods, exactly what one would expect when a genuine, abundant microbial community is being measured. The discordance in the low biomass samples is a hallmark of noise rather than signal.

These findings matter because the idea of a tumour microbiome has moved rapidly from curiosity to potential clinical application. Several high-profile studies have claimed that distinctive bacterial profiles can be found in tumours and circulating blood, and that these profiles might serve as biomarkers for early detection, prognosis, or treatment selection in cancers including lung cancer. If bacterial signatures could be read reliably from a blood sample, the logic goes, clinicians might one day supplement or even replace invasive biopsies with a simple liquid biopsy. But the new study demonstrates that, at least for lung cancer, the technical foundations of such ambitions remain shaky. The bacterial signals in tumour and blood samples were both weak, meaning barely above background, and inconsistent, meaning they failed to replicate even within the same sample processed twice.

The study’s methodological rigour is itself a lesson for the field. Technical replicates, in which the same sample is independently extracted, prepared, and sequenced, provide a direct test of measurement precision. The researchers found that while high biomass samples like saliva produced highly concordant replicates, the low biomass tumour, tissue, and blood specimens yielded wildly divergent replicate profiles, a clear indicator that the apparent diversity was driven by stochastic contamination rather than stable biological communities. The authors argue that this kind of replicate testing, combined with alternative sequencing strategies and absolute quantification of bacterial DNA, should become standard practice before any microbiome finding from low biomass samples is accepted as biologically meaningful.

The contrast between the different sample types in the study also offers a measure of reassurance about the underlying methods. Saliva proved to be highly diverse, strongly reproducible across replicates, and concordant across sequencing platforms, while positive control samples behaved as expected throughout. This means the researchers’ negative findings cannot be dismissed as a failure of their equipment or protocols. Instead, the problem appears to be intrinsic to the samples themselves: the lung tumours, adjacent lung tissue, and blood of these patients contained so little bacterial DNA that no current methodology could reliably distinguish any true signal from the noise of collection and processing environments. Whether lung tumours truly harbour sparse bacterial communities or none at all remains an open question that this study suggests may be extraordinarily difficult to answer.

For patients and clinicians hoping for microbiome-based diagnostics in lung cancer, the message is one of tempered expectations rather than closed doors. The authors emphasise that their work provides important insights into site-specific microbiomes from lung cancer patients and into the formidable challenges of assessing the tumour microbiome, and they call on the research community to adopt more rigorous validation standards before clinical claims are built on fragile data. As the field grapples with reproducibility concerns that have shadowed tumour microbiome research in recent years, this study stands as a model of the kind of scrutiny required: multiple sequencing approaches, technical replicates, careful controls, and absolute quantification, all applied to the same specimens. Only through such disciplined methods, the researchers conclude, can the field separate genuine biology from artefact and determine whether the dream of reading cancer’s microbial signature is grounded in reality or destined to dissolve at the boundaries of detection.

Subject of Research: Microbiome profiling of lung tumour and blood samples from lung cancer patients using replicates and multiple sequencing methods

Article Title: Technical replicates and multiple sequencing approaches reveal weak and inconsistent microbiome signals in lung tumour and blood samples from patients with lung cancer

Article References: DeClercq, V., Comeau, A. M., Kwawukume, A., Murphy, R., Parmar, N. R., Quinn, D. P., Wright, R., Wallace, A., & Langille, M. G. I. (2026). Technical replicates and multiple sequencing approaches reveal weak and inconsistent microbiome signals in lung tumour and blood samples from patients with lung cancer. Microbiome. https://doi.org/10.1186/s40168-026-02529-z

Image Credits: AI Generated

DOI: 10.1186/s40168-026-02529-z

Keywords: tumour microbiome, lung cancer, 16S rRNA sequencing, metagenomics, digital PCR, low biomass samples, contamination, technical replicates, bacterial biomarkers, microbiome reproducibility, non-small cell lung cancer, liquid biopsy