TL;DR
Wastewater-based epidemiology is a powerful tool for tracking viral variants, but researchers have long worried that "differential shedding"—where one variant sheds more or less viral RNA than another—might bias our estimates of how fast a new variant is spreading. This paper provides mathematical and empirical proof that estimates of a variant's growth advantage are invariant to shedding differences, while impacts on the effective reproduction number ($R_e$) are only minor and temporary.
What Problem Does This Paper Address?
During the COVID-19 pandemic, scientists turned to sewage to track SARS-CoV-2. By sequencing the viral RNA found in wastewater, they could estimate the "growth advantage" of new variants (like Omicron) over old ones (like Delta).
However, a major theoretical hurdle remained: what if Omicron simply sheds less viral material into the sewer than Delta does? If one person infected with Delta "looks" like two people infected with Omicron in the wastewater data, wouldn't that lead us to underestimate Omicron's true prevalence and growth rate? This "differential shedding" could potentially decouple wastewater data from clinical reality, making it harder for public health officials to trust sewage-based metrics.
Methods
The authors tackled this problem using a three-pronged approach:
- Mathematical Derivation: They derived closed-form expressions to calculate the bias in growth advantage and $R_e$ estimates. They used logit transforms and convolutions to show how a constant shedding difference ($c$) affects the logistic growth curve.
- Stochastic Simulations: They ran simulations of variant emergence (starting with a dominant strain and introducing a new one) to test how different generation time distributions (Exponential vs. Gamma) and varying levels of undershedding affected the results.
- Empirical Validation: They analyzed real-world data from Switzerland during the Delta-to-Omicron transition (late 2021 to early 2022). They compared wastewater data (Next-Generation Sequencing and dPCR) from six different sewersheds against clinical sequencing data from the same regions.
Datasets
- Wastewater Samples: Daily 24-hour composite samples from six Swiss wastewater treatment plants (Altenrhein, Chur, Laupen, Lugano, Geneva, and Zurich) collected between November 2021 and January 2022.
- Clinical VOC Data: Sequenced PCR-positive clinical samples binned by variant, obtained via the LAPIS API of Cov-Spectrum (primarily from the Viollier lab).
- SGTF Data: qPCR S-gene target failure data from the Geneva University Hospital (HUG) as a proxy for Omicron prevalence.
Key Insights and Ideas
- The "Nuisance Parameter" Absorption: The most critical finding is that a constant difference in shedding between two variants does not change the estimated growth rate (the slope of the curve). Instead, the bias is entirely "absorbed" by the midpoint parameter ($t_0$). In plain English: differential shedding might make it look like a variant arrived a few days earlier or later than it actually did, but it doesn't change our calculation of how fast it is outcompeting the previous strain.
- $R_e$ is Self-Correcting: While differential shedding does introduce a bias into the effective reproduction number ($R_e$), this bias is transient. It only exists during the period when the two variants are competing for dominance. Once the new variant becomes the "new normal," the $R_e$ estimate recalibrates.
- Efficiency of Scale: The study reinforces that wastewater monitoring requires orders of magnitude fewer samples than clinical testing to achieve similar accuracy in tracking variant dynamics, making it a highly cost-effective surveillance tool.
Key Results
- Growth Rate Consistency: In Switzerland, the logistic growth rates for Omicron based on wastewater sequencing ranged from 0.14 to 0.23. These were remarkably similar to the rates derived from clinical sequencing, which ranged from 0.12 to 0.21.
- Midpoint Shift: As predicted by their math, the wastewater-derived midpoints of the growth curves were shifted (higher) compared to clinical data, confirming that Omicron likely shed less than Delta, but this did not break the growth advantage model.
- Robustness to Methodology: Whether using NGS or specific dPCR probes (targeting different mutations like S:HV69-70 or S:L452R), the growth advantage estimates remained consistent, proving the method is robust to different laboratory protocols.
Limitations and Future Work
- Constant Shedding Assumption: The mathematical proof relies on the assumption that the shedding difference between variants is a constant factor. If the shape of the shedding profile (e.g., shedding peaks earlier or lasts longer) changes significantly between variants, the bias might be more complex.
- Generation Time Shifts: The authors noted that changes in the mean generation time (the time between infections) could still impact $R_e$ estimates, a factor that needs further study as viruses evolve.
- Normalization: While the study shows growth rates are robust, absolute prevalence estimates still require accurate normalization (e.g., accounting for flow rates and population size), which remains a challenge in wastewater epidemiology.
Summary updated