Metagenomics analysis for microbial ecology investigation on historical samples: negligible effect of host DNA and optimal analysis strategies

This study demonstrates that host DNA removal is unnecessary for analyzing historical microbiome samples and proposes an optimized two-step k-mer classification strategy that effectively recovers microbial signals from fragmented DNA, thereby enabling the integration of natural history collections into long-term ecological research.

The Gist

The Big Picture: Time Travel for Microbes

Imagine you have a time machine, but instead of going back to see dinosaurs, you are looking at microscopic life (bacteria, fungi, etc.) that lived on plants and insects hundreds of years ago.

Scientists have millions of these "time capsules" in museums and herbariums (dried plant collections). They want to study how human activities (like using fertilizers or antibiotics) have changed these tiny ecosystems over the last 200 years.

The Problem:
When scientists try to read the DNA from these old samples, it's like trying to find a few specific grains of sand on a beach, but the beach is 99% made of giant boulders (the host plant or insect DNA). The old DNA is also broken into tiny, fragile pieces.

For a long time, scientists thought they had to remove all those giant boulders (host DNA) before they could study the sand (microbes). But removing them is expensive, difficult, and often impossible because we don't have the "blueprints" (reference genomes) for every single plant or bug in the world.

The Question:
Do we really need to remove the host DNA to get a good picture of the microbes? Or is the "noise" of the host DNA actually harmless?

---

The Experiment: Two Ways to Look at the Data

The researchers tested this by looking at 864 different samples (rice plants, ragweed, and bumblebees) from both modern times and historical collections. They ran the data through the computer twice:

1. The "Clean" Way: They tried to filter out all the host DNA first.
2. The "Raw" Way: They just analyzed everything as it came, host DNA and all.

The Result:
It turned out that it didn't matter. The results were almost identical. Whether they removed the host DNA or not, the list of microbes they found, the diversity of the community, and the overall story remained the same.

The Analogy: Imagine you are trying to listen to a quiet conversation in a crowded room.

• Old belief: You must build a soundproof wall around the room to block out the crowd before you can hear the conversation.
• New finding: The crowd is actually so loud and distinct that your brain naturally filters them out anyway. You can hear the conversation perfectly fine without building the wall. The "noise" of the host DNA doesn't drown out the "signal" of the microbes.

---

The Second Challenge: The Broken Puzzle Pieces

There was a second problem. DNA from old samples is like a shattered vase. Instead of long, smooth ribbons of DNA, you have tiny, jagged shards (some only 20-30 letters long).

Most computer programs used to identify microbes are like puzzle solvers that look for long, specific patterns (called k-mers) to figure out what piece belongs to the picture. If the pieces are too short, the standard programs (which look for long patterns) throw them away, thinking they are too broken to be useful.

The Solution:
The researchers realized that by changing the "size of the puzzle piece" the computer looks for, they could save more information.

They proposed a Two-Step Strategy:

1. Step 1: Look for long patterns first (to catch the clear, longer pieces).
2. Step 2: Take all the pieces that Step 1 missed and look for shorter patterns (to catch the tiny, broken shards).

The Analogy: Imagine you are trying to identify people in a crowd by their height.

• Standard method: You only look for people taller than 6 feet. You miss everyone else.
• The new method: First, you spot the tall people. Then, you go back and look for the shorter people using a different rule. Now you've identified almost everyone in the crowd, including the ones you would have missed before.

By using this two-step approach with different "search sizes," they could recover 71% of the microbial signals, compared to only 47% with the old single-step method.

---

Why This Matters

This paper is a game-changer for three reasons:

1. Saves Money and Time: Scientists don't need to spend thousands of dollars trying to remove host DNA from samples, especially for rare or extinct species where we don't even have the host's genetic blueprint.
2. Unlocks History: We can now use the vast collections in museums to study how microbes have changed over centuries. This helps us understand how human actions (like pollution or climate change) have reshaped the natural world.
3. Better Tools: By using the "two-step" search method, we can get more data out of the tiny, broken DNA fragments found in historical samples, turning "garbage" data into valuable scientific insights.

In short: You don't need to clean the window to see the view, and if the glass is cracked, you just need to look through it with a different pair of glasses. The story of the past is still there, waiting to be read.

Technical Summary

1. Problem Statement

Historical samples from museums and herbaria offer a unique opportunity to study long-term shifts in host-associated microbiomes driven by anthropogenic pressures (e.g., antibiotic use, fertilizer application). However, analyzing these samples via metagenomics faces two major technical hurdles:

• High Host DNA Content: Historical samples often contain overwhelming amounts of host DNA (sometimes >80%), which can obscure microbial signals. Conventional wisdom suggests that host DNA must be removed (either physically or in silico) to prevent misclassification and ensure accurate taxonomic profiling.
• Lack of Reference Genomes: In silico host removal requires a reference genome for the host species. For non-model organisms (e.g., diverse plants, extinct species, or specific arthropods), these references are often unavailable, making standard removal pipelines impossible.
• DNA Degradation: Historical DNA is highly fragmented (often <100bp). Standard taxonomic classification tools (like Kraken2) rely on k-mer matching; using standard k-mer sizes (e.g., k=35) may discard short, informative reads, while smaller k-mers may increase false positives.

The study addresses whether host DNA removal is strictly necessary for non-model organisms and seeks to define optimal k-mer strategies for analyzing degraded historical DNA.

2. Methodology

The authors employed a multi-faceted approach combining real-world data analysis, theoretical k-mer modeling, and extensive simulations.

• Datasets:
  • Real Data: 864 metagenomes comprising contemporary rice (330 samples), historical rice (95 samples), historical common ragweed (37 samples), and museum bumblebee (402 samples).
  • Simulation: 6,000 synthetic ancient DNA (aDNA) datasets generated using gargammel. These simulated three plant hosts (rice, maize, wheat) with varying genome complexities and 107 representative soil microbes, incorporating realistic DNA damage patterns (deamination) and fragment length distributions typical of herbarium samples.
• Workflows Compared:
  • Workflow A (With Removal): Reads were aligned to host reference genomes (where available) using Bowtie2; non-matching reads were retained for classification.
  • Workflow B (Without Removal): Reads were processed directly into taxonomic classification without host filtering.
• Taxonomic Classification:
  • Tools: Kraken2 (v2.0.9) and Bracken.
  • Databases: Custom databases built with varying k-mer sizes (18, 21, 24, 28, 31, 35) and the standard PlusPFP database (k=35).
  • Proposed Strategy: A two-step classification workflow where unclassified reads from a high-stringency database (long k-mer, e.g., k=31) are re-analyzed against a lower-stringency database (shorter k-mer, e.g., k=24).
• K-mer Analysis: Used JellyFish to calculate the Jaccard index of shared k-mers between host genomes and soil microbiome pangenomes to estimate the theoretical risk of cross-domain misclassification.
• Statistical Analysis: Alpha diversity (Chao1, Shannon) and Beta diversity (Bray-Curtis, PCoA, ANOSIM) were compared between workflows using R (vegan, phyloseq).

3. Key Contributions

1. Demonstration of Negligible Host DNA Impact: The study provides robust evidence that host DNA content does not significantly alter microbial diversity metrics or taxonomic profiles, even in samples with >80% host DNA and without host reference genomes.
2. Optimization of K-mer Strategies: The authors identified that standard k-mer sizes (k=35) are suboptimal for fragmented historical DNA. They demonstrated that shorter k-mers (k=24–28) recover more species-level assignments for short reads.
3. Proposal of a Two-Step Workflow: A novel, efficient strategy was proposed to maximize the recovery of short, informative reads without sacrificing precision, specifically designed for historical samples lacking host references.

4. Key Results

• Host DNA Removal is Unnecessary for Diversity Analysis:

  • Comparing workflows with and without host DNA removal revealed no significant differences in alpha diversity (Chao1, Shannon) or beta diversity (PCoA clustering) across all sample types (contemporary rice, herbarium rice, ragweed, bumblebee).
  • The number of Operational Taxonomic Units (OTUs) and their relative abundances remained highly consistent.
  • Misclassification of host DNA as microbial taxa was extremely low (<0.2% even without removal), primarily occurring only at very low k-mer sizes (k=11–15). At k=21 and above, cross-domain misclassification dropped to negligible levels (<0.1%).

• K-mer Size and Read Length:

  • Single-Step Limitations: Using a single database with k=35 (standard) resulted in only ~47.7% of microbial reads being classified at the species level in simulated historical data.
  • Optimal Single K-mer: Databases built with k=28 or k=24 recovered significantly more reads (~67–71% species-level resolution) but carried a slightly higher risk of misclassification if host DNA was not removed.
  • Two-Step Workflow Superiority: The proposed two-step approach (e.g., k=31 followed by k=24) achieved the highest species-level classification rate (71.7%) while maintaining high precision (0.97–0.99) and recall (0.49–0.74). This strategy effectively captures short reads (21–30 bp) that would be discarded by a single k=35 database.

• K-mer Overlap Analysis:

  • The Jaccard index of shared k-mers between host and microbes dropped sharply as k-mer size increased. At k=11, overlap was near 100%; at k=21, it dropped to <0.02. This confirms that using k-mer sizes ≥21 minimizes the chance of host DNA being misidentified as microbial.

5. Significance and Implications

• Unlocking Natural History Collections: This study removes a major barrier to utilizing museum and herbarium collections for microbiome research. Researchers can now analyze host-associated microbiomes of non-model or extinct species without needing host reference genomes or performing complex physical host DNA depletion.
• Methodological Standardization: The paper establishes that for historical samples, host DNA removal is not a prerequisite for accurate ecological inference. This saves time, cost, and prevents the loss of precious sample material.
• Optimized Bioinformatics Pipeline: The proposed two-step k-mer workflow offers a practical solution for maximizing data yield from degraded DNA. It allows researchers to utilize short, fragmented reads that were previously considered uninformative, thereby recovering a more complete picture of historical microbial diversity.
• Broader Applicability: The findings suggest that the "negligible effect" of host DNA holds true across diverse host types (plants and animals) and varying levels of host DNA contamination, making these strategies applicable to a wide range of paleogenomic and historical ecological studies.

In conclusion, the authors successfully argue that the focus for historical metagenomics should shift from "removing host DNA" to "optimizing k-mer parameters" to recover the maximum amount of microbial signal from degraded, host-rich samples.