Metagenomic Quality Control: Contamination, Host Reads, and Negative Controls

Metagenomic quality control is the set of procedures that ensures your taxonomic and functional assignments reflect the true microbial community, not laboratory contamination, host DNA carryover, or bioinformatics artifacts. This guide provides a defensible, step by step approach for researchers analyzing shotgun metagenomic data. It is written for bench scientists and bioinformaticians who design or review metagenomic studies, particularly those working with low biomass samples, host associated microbiomes, or environmental specimens where contaminants can dominate results.

At a Glance

Decision Criteria

Before you start filtering, ask these questions to shape your QC plan.

What is your sample type? Low biomass samples (e.g., clinical swabs, soil extracts) are prone to environmental and reagent contamination. High biomass samples (e.g., stool) are less affected but still need host read removal when the host is present. For environmental samples without a host, host filtering may be unnecessary.

Do you have a matched negative control? Every metagenomic batch should include at least one negative control (DNA extraction blank). Without it, you cannot distinguish true low abundance microbes from contaminants. The control must be sequenced to the same depth and processed through the same pipeline.

What is your host genome representation? If your sample comes from a host (human, animal, plant), you must remove host reads unless your question specifically targets host DNA. Calculate the host read fraction early: if it exceeds 50%, consider enrichment or deeper sequencing.

What is your tolerance for false positives? Contaminant taxa that appear in negative controls at 1% relative abundance may be negligible in a pathogen detection study but critical in a community comparison. Set a generous threshold for removal (e.g., remove any OTU present in the negative control at any level) or a statistical threshold based on control read counts.

A Practical Workflow for Metagenomic Quality Control

Step 1. Raw read inspection

Start with FastQC to detect adapter contamination, low quality tails, and GC bias. For paired end reads, also use MultiQC to aggregate reports across samples. If adapters are present, trim them with Cutadapt or Trimmomatic. This step is standard for all sequencing data and is well described in resources such as the NCBI Bookshelf technical manuals and the EMBL EBI training materials on quality control.

Step 2. Host read removal (if applicable)

Map all reads against the host reference genome using Bowtie2 or BMTagger. For human samples, use the latest human genome build (GRCh38). For animal or plant studies, use the appropriate genome. Retain unmapped reads for downstream analysis. Record the percentage of reads removed. A common mistake is using an outdated host genome, which leaves human reads in the dataset. The Galaxy Training Network provides workflows for host removal in metagenomic pipelines.

Step 3. Quality filtering and trimming

Remove reads with average quality below Q20, trim bases with quality below Q15, and discard reads shorter than 50 bp after trimming. This reduces spurious mapping during taxonomic classification. Use tools such as Trimmomatic or fastp. Document the number of reads retained after each step.

Step 4. Taxonomic profiling of negative controls

Run your taxonomic classifier (e.g., Kraken2, MetaPhlAn, or Kaiju) on the negative control reads. If the control shows abundant sequences from common contaminants like Ralstonia, Pseudomonas, or Cutibacterium, treat those taxa as contaminants. For samples, remove any OTU or taxon that appears in the negative control above a threshold. A defensible threshold is removing all taxa present in the control at more than 0.1% relative abundance, or using a more conservative approach if the control is clean. The Bioconductor project offers packages like microbiome and phyloseq that support negative control based filtering [4].

Step 5. Contamination screening with positive controls and spike ins

Where possible, include a mock community positive control. Run it through the same pipeline and compare the observed composition to the expected one. Deviation indicates systematic bias. You can also spike a known genome (e.g., Salmonella or PstI digested DNA) to quantify absolute abundance and contamination. The NCBI Sequence Read Archive contains many such control datasets that you can use for benchmarking.

Step 6. Assess remaining contamination with cross sample comparisons

After filtering, examine the taxonomic composition across all samples in a study. If certain taxa appear uniformly in all samples (especially at low abundance) and also appear in the controls, they are likely contaminants. Consider removing them entirely or applying a prevalence filter (e.g., remove taxa present in fewer than 10% of samples).

Step 7. Final quality metrics and reporting

Report the number of reads before and after each QC step, the percentage of host reads removed, the list of contaminating taxa observed in negative controls, and the software versions used. Follow the MIMS guidelines. Many journals now require this information, providing it transparently strengthens your study.

Common Mistakes

Ignoring negative controls. The most common error. Many published metagenomes contain reagent contaminants that were never flagged because no control was sequenced. Even a single control can save a study. Always include and sequence it.

Over stringent host filtering. Using very high stringency alignment may remove microbial reads that map to conserved regions. Use moderate stringency (e.g., default Bowtie2 settings) and consider using a tool like BMTagger that removes exact host matches only.

Using the wrong database. A contaminant database for human samples is different from that for plant samples. Ensure your host genome is the correct species and that your taxonomic database is updated. Outdated databases miss emerging pathogens and misclassify environmental sequences.

Subtracting negative controls blindly. Simply subtracting read counts from the control can introduce negative values or remove true taxa that appear also in controls. Instead, filter taxa that exceed a threshold in the control, or use an approach such as decontam (from Bioconductor) that models contamination frequency.

Ignoring index hopping. On Illumina platforms, multiplexed libraries can experience index misassignment. This can cause reads from a high biomass sample to appear in a low biomass sample. Use unique dual indexing and check for unexpected cross sample contamination.

Limits and Uncertainty

No QC pipeline can guarantee a perfectly clean dataset. Here are the principal limitations.

Negative controls are not perfect. They capture reagent and environment contaminants but may not reflect cross contamination from sample to sample. They also cannot correct for contamination that enters after the control was processed.

Host read removal is incomplete. Even the best alignment tools leave some host reads (false negatives) and may unintentionally remove microbial reads with high homology to the host (false positives). The error rate depends on the host genome completeness and the read length.

Low biomass samples have special challenges. The signal to noise ratio is low. Many reads may be from reagents or the laboratory environment. In these cases, the negative control becomes as important as the samples themselves. The study by Lactiplantibacillus plantarum promotes intestinal goblet cell differentiation in pigs used careful controls to avoid misattributing gut microbiota functions to contaminants [6].

Database dependent classification. Taxonomic classifiers rely on reference databases that are biased toward well studied environments. Novel or poorly represented taxa may be missed or misclassified. This is not strictly a QC problem but amplifies the need for transparent reporting of database versions.

Quantification uncertainty. Relative abundances from metagenomic read mapping are compositional. A contaminant that appears at 1% in a negative control might be 10% in a true sample due to different total biomass. Statistical methods (e.g., ALDEx2, ANCOM BC) designed for compositional data can partly mitigate this.

Frequently Asked Questions

Q: How many negative controls should I include in a metagenomic study?

Include at least one negative control per extraction batch. For large studies (over 50 samples), include one control per each set of 20 samples. This matches the recommendation from many published protocols. If you use multiple extraction kits, include a control for each kit.

Q: Should I remove host reads before or after adapter trimming?

After adapter trimming and before taxonomic classification. The order is important: first ensure reads are clean of adapters, then align to the host genome, then discard host mapped reads. This prevents adapter sequences from interfering with alignment.

Q: Can I use the same negative control for both DNA extraction and PCR amplification?

No, it is better to include separate controls for each step. A DNA extraction blank controls for kit and handling contaminants. A PCR water control controls for amplification reagents. If you must use one, use the extraction blank and note that you did not specifically control for PCR contamination.

Q: What do I do if my negative control contains a large number of reads?

This indicates high contamination. First, check the taxonomic composition. If it is dominated by a few known contaminants, you can filter those taxa from all samples. If it contains diverse sequences, your reagents may be contaminated, and you should re extract and sequence new samples if possible. Report the contamination honestly even if you cannot redo the experiment.

References and Further Reading

1. NCBI Bookshelf provides comprehensive technical reference material on sequence analysis and quality control methods. The "Bioinformatics for Beginners" section is especially useful.

2. EMBL EBI Training offers free courses on metagenomics quality control, including a module on "Quality control and preprocessing of metagenomic data."

3. Galaxy Training Network provides hands on tutorials for host removal, taxon filtering, and negative control analysis using the Galaxy platform.

4. Bioconductor hosts the decontam and microbiome packages that implement statistical methods for separating contaminants from true microbes using controls.

5. NCBI Sequence Read Archive is the primary repository for metagenomic sequencing data, including negative controls and mock community datasets for benchmarking.

6. The study on Lactiplantibacillus plantarum and goblet cell differentiation in pigs exemplifies robust host read removal and control usage in host associated metagenomics PubMed.

7. The water research article on mixotrophic denitrification demonstrates how negative controls are critical for interpreting microbial contributions in engineered systems PubMed.

8. A synthetic microbial community study in BMC Microbiology showcases the use of defined positive controls to validate metagenomic workflows PubMed.

9. Research on glyphosate effects on soil microbiomes uses comprehensive controls to distinguish treatment effects from background contamination PubMed.

10. The multi omics study on mastitis in goats includes negative controls for both metagenomic and metabolomic data, providing a model for transparent reporting PubMed.

11. A Water Research article on antimicrobial resistance genes in anaerobic digestion uses viral metagenomic controls to assess contamination from livestock manure PubMed.

Related Articles

• RNA Sequencing Analysis: From FASTQ Files to Biological Questions
• RNA-seq Quality Control: What to Check Before Differential Expression
• How to Plan a Bulk RNA-seq Differential Expression Study
• Single-Cell RNA-seq Workflow: A Practical Analysis Roadmap
• Single-Cell RNA-seq Quality Control: Cells, Genes, and Mitochondrial Reads