September 4, 2024

Blood Microbiome Evidence: 9,770-Person Study

The concept of a resident blood microbiome — a stable community of microorganisms living in the bloodstream of healthy individuals — has generated both excitement and controversy in biomedical research. If real, it would fundamentally change how we interpret metagenomic next-generation sequencing (mNGS) results from blood samples. If false, the reported signals are artifacts that contaminate clinical decision-making. This webinar summary covers a landmark Nature Microbiology study that analyzed whole genome sequencing data from 9,770 healthy individuals in Singapore and found no evidence for a common blood microbiome — but did find something more nuanced than simple contamination.

Key Takeaways

Analysis of 9,770 healthy adults found microbial signatures in fewer than 16% of individuals, with most showing only a single species — far below what constitutes a microbiome
The most prevalent bacteria detected (Cutibacterium acnes, Lactobacillus, Bifidobacterium, Streptococcus species) are common skin, gut, and oral commensals, consistent with sporadic translocation rather than blood colonization
Batch-based decontamination analysis identified known reagent contaminants (Sphingomonas, Methylobacterium) that prior smaller studies had reported as blood microbiome members
DNA replication signatures confirmed some detected bacteria were recently viable, ruling out purely dead or artifactual signal
No ecological community structure (co-occurrence networks) was found among blood-detected species, unlike established microbiomes at other body sites

Why the Blood Microbiome Question Matters for mNGS

Blood has traditionally been considered sterile in healthy individuals, with microbial presence indicating infection. This view has been challenged over the past decade by studies using PCR, culture, and sequencing to report bacteria in the blood of apparently healthy people. The largest pre-existing study used quantitative PCR (qPCR) to argue that microbes are present in blood across a large fraction of the population, though at low resolution without species-level identification.

If a stable blood microbiome exists, every mNGS blood test must account for a “normal flora” baseline — a major bottleneck in biomedical research for clinical metagenomics. If it does not, the detected signals in prior studies are likely reagent contamination, computational artifacts, or transient bacteremia that does not constitute a microbiome. Distinguishing between these possibilities requires a study at population scale with rigorous contamination controls.

Study Design — Repurposing a Precision Medicine Cohort

Rather than collecting new samples (which introduces its own handling variables), the study repurposed whole genome sequencing data from a multi-ethnic precision medicine cohort in Singapore comprising Chinese, Indian, and Malay individuals. The data was originally generated for human genetics research (genome-wide association studies), providing shotgun sequencing at depth exceeding 10x that of previously published blood microbiome studies.

This design offered two advantages. First, sequencing depth averaged 373 million reads per sample — far beyond typical microbiome studies. Second, the cohort spanned multiple collection batches, extraction kits, and reagent lot numbers with detailed metadata, enabling systematic contamination detection.

Despite the depth, only about 6,000 reads per sample (by median) mapped to microbial genomes — a ratio of roughly 1 in 60,000. Computational filtering had to remove artifacts without discarding genuine signal.

Separating Signal from Contamination in Low-Biomass Samples

The study’s decontamination approach is its most methodologically valuable contribution, directly applicable to any laboratory running mNGS on low-biomass sample types.

Batch-Based Contaminant Detection

Reagent contaminants follow a predictable pattern: they appear in samples processed with a specific extraction kit lot or library prep batch and are absent from samples processed with different lots. The study exploited this by comparing microbial detection across all known batches. Species present in only one or a few batches — but not others — were flagged as probable contaminants.

For example, Sphingomonas species YG-1 appeared exclusively in a small number of batches and was absent from all others — a classic contamination signature. Notably, some prior smaller studies had reported Sphingomonas as a blood microbiome member, illustrating how studies without batch-level controls can mistake contaminants for biological signal.

Correlation-Based Filtering

Contaminant species introduced by the same reagent batch tend to co-vary in abundance across samples — their concentrations rise and fall together because they originate from the same source. The study used correlation analysis to identify these co-varying clusters, catching contaminants that batch analysis alone might miss. This approach identified additional known reagent contaminants including Methylobacterium and Acinetobacter.

After filtering, the proportion of contaminant species dropped from 21% to 10%, while the proportion of human-associated microbes increased from 40% to 70% — a meaningful improvement in signal-to-noise ratio.

What the Data Actually Shows — Sporadic Translocation, Not a Microbiome

After decontamination, fewer than 16% of healthy individuals showed any microbial signal in their blood. Most of those had only a single species detected. The species list read like a catalog of commensals from other body sites: Cutibacterium acnes (skin), Lactobacillus and Bifidobacterium species (gut), and Streptococcus species (oral cavity).

Three lines of evidence argued against calling this a microbiome.

No community structure. Gut and oral microbiomes show dense co-occurrence networks — species whose abundances correlate because they interact ecologically. Blood-detected species showed no such network, even in individuals with multiple species.

Distinct from infection profiles. Hospital blood cultures are dominated by Staphylococcus, E. coli, and Klebsiella — species rare in the healthy cohort’s blood metagenomes. Sporadic translocation is biologically different from infectious bacteremia.

Evidence of recent replication. DNA replication origin analysis confirmed that some detected bacteria — including Fusobacterium nucleatum — showed signatures of recent cell division, while contaminant species like Acinetobacter showed flat coverage. The non-contaminant detections represent bacteria that were recently alive, consistent with transient translocation.

No associations were found between microbial detection and host phenotypes including gender, ancestry, age, or BMI. One apparent association between Cutibacterium acnes and ancestry was inconsistent across cohorts and likely not biologically meaningful.

Implications for Clinical Metagenomics and Host Depletion

The study’s conclusions have direct practical implications for laboratories using mNGS on blood samples.

Baseline definition. The sporadic translocation signatures — primarily skin, gut, and oral commensals at very low abundance — provide a reference baseline for what “normal” looks like in blood mNGS data. Any pathogen detection pipeline should account for these background species to avoid false positive calls.

Custom depletion strategies. Different patient populations may have different baselines. Individuals with gum disease likely show more oral bacteria; those with inflammatory bowel disease may show more gut translocation. The Devin Filter from Micronbrane Medical enables efficient host cell depletion from blood samples, and understanding the expected background helps laboratories set appropriate detection thresholds.

Contamination control is non-negotiable. Without batch-level metadata and systematic decontamination, low-biomass mNGS studies will inevitably report reagent contaminants as biological findings. Laboratories should track extraction kit lot numbers, library prep batches, and include no-template controls in every run.

Deeper sequencing has diminishing returns. As detection thresholds drop, the proportion of contaminants and computational artifacts increases. Below approximately 0.1% relative abundance, even stool samples carry uncertain signals. For blood, the challenge is greater still.

About the Speakers

Niranjan Nagarajan, PhD is the senior group leader at the Laboratory of Metagenomic Technologies and Microbial Systems and associate director of Genome Architecture at the Genome Institute of Singapore. He is also associate professor at the Yong Loo Lin School of Medicine, National University of Singapore. His research focuses on antimicrobial resistance transmission, microbiome ecology, and computational metagenomics.

Mengchu Wu, PhD is the co-founder, CEO, and chairwoman of Micronbrane Medical. She led the discussion on clinical implications of the study’s findings for metagenomic diagnostics.