Evaluating Quantitative Metagenomics for Environmental Monitoring of Antibiotic Resistance and Establishing Detection Limits

Abstract

Metagenomics holds promise as a comprehensive, nontargeted tool for environmental monitoring. However, one key limitation is that the quantitative capacity of metagenomics is not well-defined. Here, we demonstrated a quantitative metagenomic technique and benchmarked the approach for wastewater-based surveillance of antibiotic resistance genes. To assess the variability of low-abundance oligonucleotide detection across sample matrices, we spiked DNA reference standards (meta sequins) into replicate wastewater DNA extracts at logarithmically decreasing mass-to-mass percentages (m/m%). Meta sequin ladders exhibited strong linearity at input concentrations as low as 2 × 10^-3 m/m% (R² > 0.95), with little to no reference length or GC bias. At a mean sequencing depth of 94 Gb, the limits of quantification (LoQ) and detection were calculated to be 1.3 × 10³ and 1 gene copy per μL DNA extract, respectively. In wastewater influent, activated sludge, and secondary effluent samples, 27.3, 47.7, and 44.3% of detected genes were ≤LoQ, respectively. Volumetric gene concentrations and log removal values were statistically equivalent between quantitative metagenomics and ddPCR for 16S rRNA, intI1, sul1, CTX-M-1, and vanA. The quantitative metagenomics benchmark here is a key step toward establishing metagenomics for high-throughput, nontargeted, and quantitative environmental monitoring.

Keywords: antibiotic resistance; environmental monitoring; internal standards; limit of detection; limit of quantification; quantitative metagenomics.

Figure 1

Performance of sequin ladders across sample matrices, sequencing depths, and spike-in percentages. (A) Linearity of meta sequin “Mixture A” spiked at 2% m/m, the manufacturer recommended spike-in concentration, in different wastewater matrices. Each point represents the mean log-transformed RPK sequin count per input proportion (see Table S1), resulting in 16-point reference curves. Error bars represent ± standard deviations of log-transformed data. (B) Ladder linearity at 2% m/m in an influent sample subsampled to discrete sequencing depths in Gb. Error bars are not shown for simplicity. Log–log models for each subset can be found in Table S5. (C) Total sequencing yield as a function of decreasing input m/m% and sample matrix.

Figure 2

The LoQ and LoD of quantitative metagenomics. (A) Coefficient of variation (CV%) of the sequencing yield of all 86 sequins across ladder spike-in m/m%. Influent, activated sludge, and secondary effluent samples at each m/m% were treated as technical replicates. The black dashed line marks the generally recommended threshold CV = 35% for determining the LoQ of qPCR experiments. (B) The calculated LoQ and (C) LoD of quantitative metagenomics as a function of sequencing depth and ladder spike-in m/m%.

Figure 3

Variable detection of unique ARGs across concentration range. ARGs across all 30 samples were quantified using the mean slope (1.08) and intercept (−3.39) of the log–log models with 2% m/m ladders at ∼100 Gb (CARD v3.0.3, ID = 80%, coverage = 80%). Each point represents a unique ARG detected across matrix replicates (n = 10) color coded by antibiotic class. Singletons are omitted.

Figure 4

Comparison of ddPCR and quantitative metagenomic ARG quantification. Dot plot of quantitative metagenomic gene concentrations were determined at 80, 90, and 99% amino acid identity (i.e., qMeta.80, qMeta.90, qMeta.99) and 80% query coverage. All ten biological replicates were quantified for ddPCR and quantitative metagenomics for each matrix. Error bars represent ± standard deviations. Blue * = compared to ddPCR, the data are statistically equivalent, but the effect size is greater than zero. Red * = compared to ddPCR, the data are not statistically equivalent, and the effect size is greater than zero. See Table S6 for wilcox_TOST test results. The horizontal dashed lines represent the LoQs for each sample matrix.

Figure 5

Correlations between ddPCR and quantitative metagenomics ARG concentrations. Quantitative metagenomic gene concentrations were determined at 80, 90, and 99% amino acid identity (i.e., qMeta.80, qMeta.90, qMeta.99) and 80% query coverage. Solid black diagonal line represents ideal slope of 1. Blue line represents results of linear regression with 95% confidence intervals.