A Resource Benchmark of Human Microbiota–Associated Mouse Models Reveals Donor-Dominated Outcomes

Updated on October 16, 2026

Contents Introduction Study design and samples Methods Results Discussion Code Availability References Abbreviations

Color key

Introduction

• Background
• HMA rodent models are widely used for causal microbiome-phenotype studies, but the ~95% reported phenotype-transfer rate overstates microbiome causality, and donor taxa frequently fail to colonize or diverge from donor profiles in recipients ^{Walter et al. (2020)}.
• MICROBENCH addresses this with a longitudinal metagenomic benchmarking resource: 10 donors × germ-free C57BL/6 recipients, profiled by long-read shotgun metagenomics for bacterial community composition across fecal and intestinal (Colon, Ileum, Cecum) samples. This report covers the metagenomics component; the companion metabolomic and immune analyses are reported separately.
• Rationale
• Donor-to-recipient transfer fidelity is rarely quantified rigorously across donors, time points, and gut compartments within a single experiment; existing HMA benchmarks are typically limited in donor number or sampling breadth.
• A longitudinal metagenomic dataset spanning multiple donors, time points, and gut compartments is therefore needed as a community-level benchmark for HMA studies and FMT trial design.
• Objective
• Deliver MICROBENCH and a donor-tailored long-read metagenomics pipeline as a calibrated reference for future HMA cohorts and FMT trial designs.




• Aims of the study
• Generate a longitudinal metagenomic benchmarking dataset of germ-free C57BL/6 mice colonized by fecal microbiota transfer from 10 healthy human donors, profiled for bacterial community composition across fecal and intestinal (Colon, Ileum, Cecum) samples.
• Quantify donor-to-recipient transfer fidelity across fecal and intestinal compartments and over the post-FMT time course (Weeks 1–8).
• Establish and release an end-to-end long-read metagenomics pipeline for HMA studies—comprising donor-tailored reference database construction, read-level compositional quality control, and genome-size-aware species-level abundance estimation—so that the dataset and pipeline together yield reference statistics for donor-to-recipient microbiota transfer that can be used to benchmark future HMA studies and support FMT trial design.

Study design and samples

1. Donors, N=10
1. Eight female (F) healthy donors
2. Two male (M) healthy donors


2. From barcodes to the analysis-ready matrix (click a step below to list the excluded barcodes or samples)

Plate	# Barcodesa	# Sequencedb	# Samplesc	# Analysis-readyd
Donors	10	10	10	10
01	96	96	96	95
02	96	96	96	95
03	96	96	96	95
04	96	96	96	95
05	96	96	82	82
06	96	96	95	95
07	96	96	96	95
08	96	96	93	89
09	96	96	92	92
10	96	96	95	95
11	96	96	92	91
12	96	96	96	95
13	96	96	94	93
14	41	20	20	20
Total	1,299	1,278	1,249	1,237

1. Barcodes sequenced on the PromethION, 1,299.
2. Twenty-one barcodes from Plate 14 yielded zero reads: 1,278 (1,299 − 21).
   Plate14_barcode01
   Plate14_barcode02
   Plate14_barcode03
   Plate14_barcode04
   Plate14_barcode57
   Plate14_barcode58
   Plate14_barcode59
   Plate14_barcode65
   Plate14_barcode66
   Plate14_barcode67
   Plate14_barcode73
   Plate14_barcode74
   Plate14_barcode75
   Plate14_barcode81
   Plate14_barcode82
   Plate14_barcode83
   Plate14_barcode88
   Plate14_barcode89
   Plate14_barcode90
   Plate14_barcode91
   Plate14_barcode96
3. Twenty-nine barcodes were excluded for poor read quality (all reads with a quality value of 0 in the Dorado summary output): 1,249 (1,278 − 29).
   Plate05_barcode40
   Plate05_barcode67
   Plate05_barcode85
   Plate05_barcode86
   Plate05_barcode87
   Plate05_barcode88
   Plate05_barcode89
   Plate05_barcode90
   Plate05_barcode91
   Plate05_barcode92
   Plate05_barcode93
   Plate05_barcode94
   Plate05_barcode95
   Plate05_barcode96
   Plate06_barcode88
   Plate08_barcode20
   Plate08_barcode39
   Plate08_barcode50
   Plate09_barcode08
   Plate09_barcode24
   Plate09_barcode32
   Plate09_barcode88
   Plate10_barcode88
   Plate11_barcode24
   Plate11_barcode48
   Plate11_barcode79
   Plate11_barcode80
   Plate13_barcode22
   Plate13_barcode88
4. Eight negative controls and four mouse samples with low read counts and/or poor read quality were excluded: 1,237 (1,249 − 12).
   Neg_1_X_Neg
   Neg_2_X_Neg
   Neg_3_X_Neg
   Neg_4_X_Neg
   Neg_5_X_Neg
   Neg_6_X_Neg
   Neg_7_X_Neg
   Neg_8_X_Neg
   F1_F5_3_C
   F7_F707_3_M
   F7_F718_8_F
   F7_M724_4_F


3. Samples (post-QC), N=1,237
1. Donor, N=10
2. Donor replicates, N=9 (F1 not replicated)
3. OMM12, N=12
4. Mouse samples, N=1,206


4. Tissue × timepoint (Mouse only)

Tissue	Weeks post-FMT	# Samples
Fecal (F)	1	152
	2	153
	4	152
	6	153
	8	146
Colon (C)	3	73
	8	75
Ileum (I)	3	74
	8	76
Cecum (M)	3	76
	8	76
Total		1,206




5. Sample naming convention
Donor sex/ID _ Mouse sex/ID _ Weeks post-FMT _ Sample type
1. Donor sex/ID: [ F0 | F1 | F2 | F3 | M4 | M5 | F6 | F7 | F8 | F9 | Ctrl (OMM12) | Neg (negative control) ]
2. Mouse sex/ID: [ M | F ] + Mouse ID; NA for donor samples
3. Weeks post-FMT: [ 0 (donor) | 1 | 2 | 3 | 4 | 6 | 8 | X (neg) ]
4. Sample type: [ C (colon) | F (fecal) | I (ileum) | M (cecum) | Ctrl | Neg | 1 (donor) | 2 (donor replicate) ]

Examples:

• F0_F20_6_F: Fecal sample from female mouse #20 at week 6 post-FMT, colonized by female donor #0
• M5_F509_3_M: Cecum sample from female mouse #509 at week 3 post-FMT, colonized by male donor #5
• F6_NA_0_1: Fecal sample from female donor #6 (1: primary; 2: replicate)
• Ctrl_Ctrl3_3_Ctrl: OMM12 positive control #3
• Neg_6_X_Neg: Negative control #6


6. Preprocessing and QC record: sequencing QC, read filtering, classifier benchmarking and BASEN development are on a separate page, Preprocessing.

Methods

1. Sequencing and read processing

   • Library preparation: ONT Native Barcoding Kit 96 V14 (SQK-NBD114.96), according to the manufacturer's instructions; sequencing on PromethION R10.4.1 flow cells with a P2 Solo device
   • Dorado (v1.1.1): basecalling from POD5 with model dna_r10.4.1_e8.2_400bps_sup@v5.2.0, then demultiplexing
   • Read QC: NanoPlot ^{De Coster et al. (2018)} (v1.46.1) for per-read length, quality and yield; FastQC for per-read quality checks; MultiQC ^{Ewels et al. (2016)} (v1.11) to aggregate the reports
   • Length filter: reads longer than 100,000 bp removed and long reads trimmed to 14,999 bp. The length-filtered reads are the input to host removal, so both the MAG catalogue and the taxonomic profiles are built from reads of at most 15,000 bp.
   • Host read removal: reads aligned with Minimap2 ^{Li (2018)} (v2.30) to the combined GENCODE mouse (GRCm39) and human (GRCh38) genomes, and unmapped reads extracted with SAMtools ^{Danecek et al. (2021)} (v1.16.1), so any read with a primary alignment to either genome was removed. Across all samples, 62,664,009 of 575,225,291 reads (10.9%) were mouse and 106,074 (0.02%) human, leaving 512,455,208 host-free reads.
2. Donor-derived MAG reference catalogue

   • Input: host-free reads of the 10 high-depth donor samples (*_NA_0_1); the assembly-to-binning workflow was run five times independently per donor, because metaFlye assembly is not deterministic
   • metaFlye ^{Kolmogorov et al. (2020)} (v2.9.6; --meta --nano-hq): long-read metagenome assembly using repeat graphs
   • Medaka (v2.1.0; model r1041_e82_400bps_sup_v5.2.0): consensus polishing of long-read assemblies
   • SemiBin2 ^{Pan et al. (2023)} (v2.2.1; single_easy_bin --environment human_gut --sequencing-type long_read --random-seed 42): self-supervised metagenomic binning, using Medaka's read alignments for contig coverage
   • CheckM2 ^{Chklovski et al. (2023)} (v1.1.0): bin completeness and contamination; high quality = completeness ≥90% and contamination <5% (MIMAG ^{Bowers et al. (2017)})
   • Pooling within donor with skani ^{Shaw and Yu (2023)} (v0.3.1): high-quality bins from the five runs linked at ANI ≥95% ^{Jain et al. (2018)} by single linkage (no aligned-fraction minimum); species groups recovered in ≥3 of 5 runs retained; representative = highest completeness − 5 × contamination, the core of the dRep genome score ^{Olm et al. (2017)}
   • GTDB-Tk ^{Chaumeil et al. (2022)} (v2.5.2, GTDB release 226): standardized taxonomy assignment for bacterial and archaeal genomes
   • Dereplication across donors with the same skani rule; groups holding more than one GTDB-Tk species were split, so that each MAG is one species
   • Output: 7,627 bins (50 runs) → 1,979 high-quality bins → 397 pooled MAGs → 191 cross-donor groups → 194 non-redundant donor-derived MAGs (5 without a species-level name) defining the feature universe for the custom Kraken2 database
   • Custom Kraken2 database: the 194 MAGs with a GTDB release 226 taxonomy, k-mer length 35, minimizer length 31, no decoy or host sequences (194 species, 3,233 contigs, 536,180,842 bp)
   • Random seeds: SemiBin2 --random-seed 42; metaFlye, Medaka, CheckM2, skani, GTDB-Tk and Kraken2 have no seed option
3. Taxonomic classifier evaluation and selection

   • Kraken2 ^{Wood et al. (2019)} (v2.1.6): k-mer-based per-read taxonomic classification — selected as the primary classifier
   • Minimap2 ^{Li (2018)} (v2.30) + CoverM ^{Aroney et al. (2025)} (v0.7.0): alignment-based read mapping and coverage estimation (comparator)
   • MetaPhlAn 4 (Metagenomic Phylogenetic Analysis) ^{Blanco-Míguez et al. (2023)} (v4.2, prebuilt marker database): marker-gene profiling (comparator)
   • Sylph ^{Shaw and Yu (2025)} (v0.9.0): ANI/sketch-based profiling (comparator)
   • Benchmark: OMM12 control samples classified against an OMM12-only database (46,004,597 bp) and a decoy-inclusive database (3,176,927,508 bp; GRCh38, selected non-target decoy taxa and the OMM12 references)
   • Selection criteria: support for fully customizable databases, per-read assignment outputs, and operational scalability
4. Taxonomic classification, read-level QC and species-level abundance

   • Classification: Kraken2 (v2.1.6; default settings, confidence 0) on the host-free reads of the 1,237 analysis-ready samples, against the custom 194-MAG database
   • Read-level compositional QC (BASEN): a read was removed if more than 75% of its k-mers were unassigned or if the Shannon diversity of its per-read TaxID assignment profile exceeded 1.0 (high-depth donor samples) or 0.75 (mouse, donor replicate and control samples). Reads shorter than the k-mer length (35 bp) cannot be classified by Kraken2 and carry no evidence.
   • BASEN (v1.0.0): Base-level Abundance estimation with Species-assigned Evidence using Nanopore — species-rank evidence aggregated from the retained reads' Kraken2 k-mer assignments, normalized by reference genome length to a coverage proxy, and rescaled within sample to relative abundance

Results

1. Correlation/Variability visualization
• Input: BASEN relative abundances of the 194 MAGs after read filtering (Preprocessing, Section 3C); mouse samples only (N=1,206); donor replicates and OMM12 controls not included; absent species set to 0.
• Left: For each factor, all pairwise Spearman correlations were calculated among donor-specific recipient mouse samples, excluding the donor sample. In the radar chart, the yellow and red dots indicate the mean and standard deviation of the correlation coefficients, respectively. The mean ranges from 0.4 to 1.0, while the standard deviation ranges from 0 to 0.20, as indicated in parentheses. For example, cecum samples correlate more strongly with each other than ileum samples in all 10 donors (mean rho 0.77–0.90 vs 0.58–0.72), with a lower standard deviation in 8 of 10 donors (all except F3 and F7).
• Right: Violin plots show Spearman correlations between the donor and individual mouse samples based on species relative abundance. The X-axis indicates the factor, the Y-axis shows the correlation coefficient, and each dot represents the correlation between the donor and a single mouse sample. Numbers in parentheses denote the number of mouse samples in each factor group.



2. Engraftment varies over time
• Input: BASEN relative abundances of the 194 MAGs after read filtering (Preprocessing, Section 3C); fecal mouse samples only (N=756; week 3 has no fecal samples), with the 10 high-depth donor samples (*_NA_0_1) as week 0; donor replicates and OMM12 controls not included.
• Enumerate a set of particularly prominent taxa
1. The mean relative abundance of each taxon was calculated weekly (e.g., weeks 1, 2, 4, 6, and 8) using fecal samples only, with the donor sample designated as week 0.
2. Taxa whose mean relative abundance exceeded 5% in at least one donor × week group (week 0 = donor) were selected.
3. Forty-one of the 194 species met this criterion in at least one week or donor group and were therefore defined as prominent taxa.

---

All 194 species (open in a new tab)

---

Prominent 41 species (open in a new tab)

---

• All 194 species: the prominent species are coloured and labelled, the others grey (they turn dark grey when highlighted).
• Prominent 41 species: a static version is available as a PDF (page 1: all donors; page 2: colour key; pages 3–12: one donor each).


• Find the most stable engraftment week
1. We calculated Spearman correlation coefficients between donor baseline (week 0) and each post-engraftment week (shown below); the PDFs also include the correlations between adjacent weeks.
2. Then, we selected the week with the highest correlation (most stable engraftment) using the prominent taxa (bold).
3. The selection rests on small differences: in 8 of 10 donors the highest and second-highest rho differ by less than 0.015 (F0: week 8, 0.6775, vs week 6, 0.6769); only F1 (0.040) and F8 (0.043) have a clear margin.

All 194 species	Prominent 41 species



3. Taxa and engraftment characteristics
• Using the donor-specific stable weeks defined above (the week with the highest Spearman correlation to the donor; the comparison uses the selection on the prominent 41 species), we compared relative abundances between each donor and the corresponding stable-week recipient profile.

Week	All 194 species	Prominent 41 species
1	F1, F2, M4, F6, F7, F8	F1, F3, F8
2	F0	F2, M5
4	F9	M4, F6, F7
6	M5	None
8	F3	F0, F9

• A donor-level threshold of 1% was applied to define engraftment status. Taxa with relative abundance >1% were considered present.
• We annotated taxa as follows:
• ⬤ (closed circle) when both donor and recipient were >1%
• ◯ (open circle) when the taxon was >1% in the donor only
• ✕ (cross) when the taxon was >1% in the recipient only
• Donors (rows) and taxa (columns) are each ordered by hierarchical clustering (average linkage) of their annotation patterns; distance = proportion of annotations that differ (across the 41 prominent species for donors, across the 10 donors for taxa).

Discussion

• Donor-dominated outcomes. Mice colonized by the same donor resemble each other far more than mice colonized by different donors (mean Spearman rho 0.79 vs 0.51 across all mouse samples; higher within donor for all 10 donors). Within each donor group, mice also resemble each other (mean rho 0.71–0.83) more than they resemble the donor sample (median 0.41–0.62): transfer is reproducible across recipients, but recipients do not reproduce the donor.
• Rapid establishment, then stability. The 41 prominent species make up 45% of the donor samples and 86% of the week-1 fecal communities (mean across donors), and remain at 84–85% through week 8. Week-to-week correlation is ≥ 0.90 from week 1 onward in 9 of 10 donors (F1 drops to 0.76–0.78 after week 4). Stable overall composition does not mean a fixed dominant species: in F6 mice, Prevotella copri_K rose from 0.4% at week 1 to 34% at week 4.
• Expansion of minor donor Bacteroidales. Bacteroidales rose from 6–30% of the donor samples to 36–80% of the week-8 fecal communities. In every donor, the most abundant species at week 8 was a minor member of the donor sample (0.01–4.7%). Parabacteroides distasonis, at 0.04–0.86% in every donor, exceeded 1% in the mice of all 10 donors (mean 11.0% at week 8).
• Loss of common donor taxa. Several taxa abundant in the donors did not persist: Agathobacter rectalis exceeded 1% in 7 donors and in none of their mice at week 8; Faecalibacterium prausnitzii and Bifidobacterium pseudocatenulatum persisted above 1% in 1 of 3 donors each. Thirteen of the 41 prominent species qualify only through the donor sample.
• Donor identity is carried by a few lineages. The donor-specific signal at week 8 rests mainly on Prevotella copri clades (_B persisting in M4, 13.5% → 16.2%, and F0, 9.0% → 7.8%; _G reaching 12.7% in F3 mice; _K reaching 36.5% in F6 mice) and on donor-specific Bacteroides.
• No single stable week. The week most similar to the donor depends on the species set: the selection on all 194 species and on the 41 prominent species agrees for only 2 of 10 donors (F1, F8), and in 8 of 10 donors the best and second-best weeks differ by less than 0.015 in rho. Donor-to-recipient comparisons should therefore not rest on a single snapshot week.

Limitations

• Donor technical replicates. Replicates (*_NA_0_2) were sequenced at much lower depth than the primary donor samples (5,059–296,580 reads vs 2.6–8.1 million), and F1 has none, so technical variance cannot be partitioned; same-donor correlations ranged from r = 0.25 (F3) to 0.96 (F2).
• Sampling design. The fecal series covers weeks 1, 2, 4, 6 and 8; Colon, Ileum and Cecum were collected at weeks 3 and 8 only, with no fecal sample at week 3.

Code Availability

• [Zenodo] Scripts Pending
• [GitHub] Griffin-Pipeline
• [GitHub] BASEN v1.0.0 (Base-level Abundance estimation with Species-assigned Evidence using Nanopore)

References

Aroney et al. CoverM: read alignment statistics for metagenomics. Bioinformatics 2025;41(4):btaf147.
Blanco-Míguez et al. Extending and improving metagenomic taxonomic profiling with uncharacterized species using MetaPhlAn 4. Nat Biotechnol 2023;41(11):1633-1644.
Bowers et al. Minimum information about a single amplified genome (MISAG) and a metagenome-assembled genome (MIMAG) of bacteria and archaea. Nat Biotechnol 2017;35(8):725-731.
Chaumeil et al. GTDB-Tk v2: memory friendly classification with the genome taxonomy database. Bioinformatics 2022;38(23):5315-5316.
Chklovski et al. CheckM2: a rapid, scalable and accurate tool for assessing microbial genome quality using machine learning. Nat Methods 2023;20(8):1203-1212.
Danecek et al. Twelve years of SAMtools and BCFtools. Gigascience 2021;10(2):giab008.
De Coster et al. NanoPack: visualizing and processing long-read sequencing data. Bioinformatics 2018;34(15):2666-2669.
Ewels et al. MultiQC: summarize analysis results for multiple tools and samples in a single report. Bioinformatics 2016;32(19):3047-8.
Jain et al. High throughput ANI analysis of 90K prokaryotic genomes reveals clear species boundaries. Nat Commun 2018;9(1):5114.
Kolmogorov et al. metaFlye: scalable long-read metagenome assembly using repeat graphs. Nat Methods 2020;17(11):1103-1110.
Li. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics 2018;34(18):3094-3100.
Olm et al. dRep: a tool for fast and accurate genomic comparisons that enables improved genome recovery from metagenomes through de-replication. ISME J 2017;11(12):2864-2868.
Pan et al. SemiBin2: self-supervised contrastive learning leads to better MAGs for short- and long-read sequencing. Bioinformatics 2023;39(Suppl 1):i21-i29.
Shaw and Yu. Fast and robust metagenomic sequence comparison through sparse chaining with skani. Nat Methods 2023;20(11):1661-1665.
Shaw and Yu. Rapid species-level metagenome profiling and containment estimation with sylph. Nat Biotechnol 2025;43(8):1348-1359.
Walter et al. Establishing or exaggerating causality for the gut microbiome: Lessons from human microbiota-associated rodents. Cell 2020;180(2):221-232.
Wood et al. Improved metagenomic analysis with Kraken 2. Genome Biol 2019;20(1):257.

Abbreviations