Lineage abundance estimation for SARS-CoV-2 in wastewater using transcriptome quantification techniques

Abstract

Effectively monitoring the spread of SARS-CoV-2 mutants is essential to efforts to counter the ongoing pandemic. Predicting lineage abundance from wastewater, however, is technically challenging. We show that by sequencing SARS-CoV-2 RNA in wastewater and applying algorithms initially used for transcriptome quantification, we can estimate lineage abundance in wastewater samples. We find high variability in signal among individual samples, but the overall trends match those observed from sequencing clinical samples. Thus, while clinical sequencing remains a more sensitive technique for population surveillance, wastewater sequencing can be used to monitor trends in mutant prevalence in situations where clinical sequencing is unavailable.

Fig. 1

VLQ, a computational approach to lineage abundance estimation from wastewater sequencing data. a Computational similarity between RNA transcript quantification and lineage abundance estimation. b Key aspects of the kallisto algorithm in the context of lineage abundance estimation. c Our workflow uses multiple reference sequence per lineage to capture within-lineage variation. Applying kallisto (as in part b) results in abundance estimates per reference sequence. These abundances are filtered using a minimal abundance cutoff and subsequently summed per lineage to obtain abundance estimates per lineage. Finally, lineage abundances are reported

Fig. 2

Benchmarking results on simulated wastewater sequencing data of a representative mixture of background sequences and a variant lineage (VOC; here: B.1.1.7, B.1.351, B.1.427, B.1.429, P.1). a Estimated lineage abundance (VOC frequency) versus true abundance on whole genome sequencing data with a depth of 1000×. b Estimated lineage abundance (VOC frequency) versus true abundance on Spike-only sequencing data with a depth of 10,000×. c, d Relative prediction error per lineage for the estimated frequencies presented in panels a and b, respectively. Relative prediction errors are defined as the absolute difference between true and estimated frequency, relative to the true frequency. VOC, variant of concern

Fig. 3

a RNA levels in wastewater (copies/ml sludge, displayed on the left vertical axis) follow the same trend as COVID-19 case rates (cases per 100K people, displayed on the right vertical axis). b Percent genome with >20× coverage versus sludge Ct values. c Impact of genome coverage on predicted B.1.1.7 abundance for random subsamples of a sludge sample with full genome coverage. The horizontal dotted line indicated the predicted B.1.1.7 abundance for the full sample (99% genome coverage)

Fig. 4

Wastewater versus clinical abundance estimates for B.1.1.7 and B.1.526 in New Haven from early January 2021 to late April 2021. Dates of clinical sampling correspond to the date of specimen collection. In addition, ddPCR-based abundance estimates of lineages with the H69/V70 deletion (likely B.1.1.7) are shown for wastewater samples taken every six days. Confidence intervals are computed from ddPCR confidence intervals for measured copied of the H69/V70 deletion and measured copies of wild type present (see the “Methods” section). Vertical dashes on the x-axis indicate timepoints where wastewater sequencing data was obtained that passed the filtering criteria (Ct value < 31 or Ct < 34 with at least 0.5M reads aligned)

Fig. 5

Wastewater versus GISAID abundance estimates for B.1.1.7, B.1.427, B.1.429, and B.1.526 at 16 locations across 8 states of the US. Samples were collected between late December 2020 and late January 2021; the sampling date and location are indicated on the horizontal axis. Samples are sorted by location, with different locations separated by a dotted line and different states separated by a solid line