Seasonal and spatial variability of PMMoV in tropical urban wastewater-based surveillance: implications for accurate SARS-CoV-2 monitoring in Kampala, Uganda Authors: Andrew Nsawotebba, Valeria Nakintu, Innocent Morunyanga, Jonathan Kabazzi, Jordan Magala, Samuel Jefferson Mutyaba, Noah C. Hull, Victor Vusi Mabasa, Denis Smith Akejo, Fatim Cham, Susan Nabadda Published: 2026 ABSTRACT IntroductionWastewater-based surveillance (WBS) is a vital epidemiological tool for monitoring community-level pathogen transmission. Viral concentrations in wastewater are commonly normalized using fecal indicators such as Pepper Mild Mottle Virus (PMMoV) RNA; however, in tropical urban settings, PMMoV RNA signal may be influenced by seasonal rainfall, stormwater infiltration, and sanitation infrastructure, potentially compromising its reliability. This study aimed to characterize the seasonal and site-specific variability of PMMoV RNA in Kampala, Uganda, and to evaluate its suitability as a fecal indicator under tropical urban conditions.MethodsMoore swab-based influent wastewater samples were collected weekly between March 2023 and May 2024 at four wastewater treatment facilities in the Kampala Metropolitan Area. PMMoV RNA was quantified using RT-qPCR and expressed as log₁₀ genomic copies per 100 mL (GC/100 mL). Between-facility differences were assessed using the Kruskal-Wallis H test with Dunn’s post-hoc correction. The association between PMMoV RNA and monthly average rainfall was quantified using Pearson’s correlation. A linear mixed model (LMM) with rainfall and site as fixed effects and a random intercept for site was fitted to partition the variance into rainfall-driven and structural sources.ResultsPMMoV RNA was detected in 239 of 244 samples (97.95%) across all four facilities. Concentrations exhibited significant spatial and temporal variability (Kruskal-Wallis H = 9.83, p = 0.020), driven by sewershed structure, including infrastructure type, hydraulic configuration, and inflow susceptibility rather than population size alone (per capita analysis: H = 84.29, p < 0.0001). Naalya wastewater stabilization pond showed the highest and most stable concentrations (mean = 6.09 log₁₀ GC/100 mL; IQR = 1.03), while Bugolobi fecal sludge treatment plant showed the lowest central tendency (mean = 5.27) and greatest variability (IQR = 2.60), reflecting intermittent sludge inputs. A strong, significant negative correlation was identified between PMMoV RNA and monthly average rainfall (Pearson r = − 0.623, p = 0.013). The LMM confirmed rainfall as an independent, site-invariant predictor (β = −0.0077 log₁₀ GC/100 mL per mm, p < 0.001), with a 100 mm increase in rainfall corresponding to an approximately 6-fold reduction in PMMoV RNA concentration.ConclusionPMMoV RNA concentrations in Kampala’s wastewater are systematically and substantially modulated by rainfall and catchment infrastructure, with seasonal dilution sufficient to bias the interpretation of normalized pathogen signals if left unadjusted. These findings challenge the assumption of proportional dilution underpinning ratio-based normalization and highlight the need for seweshed characterization as a foundation for site selection and interpretation, alongside rainfall-adjusted interpretation frameworks, site-specific validation, and complementary fecal indicators in tropical urban WBS programs.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-026-27515-w. Background Wastewater-based surveillance (WBS) is a valuable tool for monitoring community-level pathogen transmission by measuring biomarkers in wastewater [1]. Intrinsic to WBS is its unique ability to monitor an entire population, irrespective of symptoms or access to clinical testing, objectively, passively, and comprehensively, thus being inclusive and representative [2]. Several studies have examined the relationship between raw SARS-CoV-2 concentrations in wastewater and corresponding clinical metrics [3]. However, a key challenge in WBS is understanding how fecal matter varies in wastewater, since viral pathogen concentrations can be influenced by confounding factors such as changes in population size, sewer system dynamics, and dilution from stormwater infiltration [4]. Industrial or commercial discharges, along with inflow and infiltration events, can pose additional challenges to WBS by causing inhibition and dilution [5]. Normalization of the viral signal to fecal indicators or flow is frequently done to address these concerns [6]. However, normalization of the viral signal is often performed without accounting for the complexity of the wastewater or sewer shed, which may introduce additional variability [4]. Factors such as stormwater infiltration, variable population inputs, and differences in sewer infrastructure can independently influence both pathogen and fecal indicator concentrations, and may not affect them proportionally [7]. This is particularly consequential in tropical urban settings, where seasonal rainfall, open-drainage infrastructure, and heterogeneous sewer networks can introduce substantial dilution effects that vary systematically over time but may be mistaken for true changes in pathogen burden if normalization is applied without adjustment. Numerous studies have utilized various wastewater parameters as indicators of fecal contamination to normalize SARS-CoV-2 concentrations, with commonly used markers including pepper mild mottle virus (PMMoV) and cross-assemblage phage (crAssphage) [8, 9]. PMMoV is abundant in human feces due to dietary intake of pepper products [10]. PMMoV RNA has been consistently detected in wastewater samples in African settings, supporting its use as a fecal indicator [11]. Despite its popularity, recent studies have raised concerns regarding the stability of PMMoV RNA across seasons and regions. Environmental factors such as rainfall, temperature, and local dietary habits can influence PMMoV RNA concentrations [12] potentially affecting its reliability as a fecal strength indicator. In regions with distinct wet and dry seasons, such as many parts of sub-Saharan Africa, including Uganda, a tropical country, stormwater ingress can dilute wastewater, reducing both viral and fecal indicator concentrations and complicating trend interpretation [13]. This study sought to evaluate the stability and variability of the fecal indicator PMMoV under real-world tropical urban conditions, rather than to assess normalization performance, building upon earlier WBS efforts in Kampala, Uganda [14]. We investigated its seasonal variability across multiple wet and dry seasons at four wastewater treatment plants (WWTPs) to determine the extent to which rainfall and sewershed characteristics influence its concentration. This assessment of PMMoV’s reliability under local conditions is a crucial first step in determining its suitability as a normalization tool in tropical, urban WBS. Materials and methods Study setting and sampling sites This study used purposive sampling at four selected WWTPs in a mixed sanitation system in the Kampala Metropolitan area. Sites were selected to represent structural heterogeneity across Kampala’s fragmented sanitation network, specifically to capture variation in: (i) treatment technology and influent feed type (conventional gravity sewer, fecal sludge treatment, and wastewater stabilization pond); (ii) sewershed population size, (iii) catchment geographic extent and hydraulic configuration; and (iv) anticipated susceptibility to stormwater infiltration and inflow. Together, these criteria were designed to enable assessment of how catchment-specific structural factors, beyond population size alone, influence PMMoV RNA concentrations. The four selected sites were: the Naalya wastewater stabilization pond (Naalya WWSP), the Bugolobi fecal sludge treatment plant (Bugolobi FSTP), and the inlets of Nakivubo New wastewater treatment plant, specifically at points 1 (Nakivubo WWTP Inlet 1) and 2 (Nakivubo WWTP Inlet 2), as shown in Fig. 1. Fig. 1Map of Uganda showing the sampling sites Map of Uganda showing the sampling sites Sewershed and catchment characteristics These four sampling sites represent different catchment types within Kampala’s mixed sewer and sanitation network. Together, they serve an estimated population of approximately 583,556, with individual catchment populations of roughly 220,000 (Bugolobi FSTP), 266,667 (Nakivubo WWTP Inlet 1), 88,889 (Nakivubo WWTP Inlet 2), and 8,000 (Naalya WWSP). Bugolobi FSTP operates outside the conventional gravity sewer network and receives fecal sludge from pit latrines and septic tanks across the city, transported by vacuum tanker trucks on an irregular, demand-driven basis. Desludging events from households and commercial facilities produce episodic, batch-style inputs of high-strength fecal sludge. Because of its non-continuous feed, PMMoV RNA levels at this site are likely to fluctuate significantly over time, influenced more by desludging schedules than by rainfall-induced dilution. Nakivubo WWTP Inlet 1 receives wastewater from Kampala’s central business district (CBD) and nearby high-density areas through a sewer system that carries sanitary wastewater. The CBD catchment, which serves about 266,667 people, has extensive, vulnerable, and partially sealed sewer infrastructure that is prone to stormwater infiltration. This large catchment includes flood-prone low-lying areas where, during rainy seasons, stormwater inflow increases hydraulic loading and dilutes wastewater constituents, including PMMoV RNA. Nakivubo WWTP Inlet 2 serves a smaller, more geographically confined sub-catchment within the Nakivubo sewershed, covering residential areas northeast of the CBD with an estimated population of approximately 88,889. This sanitation system covers a smaller geographic area, with shorter conveyance distances and fewer upstream inputs, reducing its exposure to stormwater inflow and infiltration compared to Inlet 1, resulting in more stable hydraulic conditions across seasons. This makes Inlet 2 a valuable internal reference point for evaluating how stormwater dynamics impact the signals observed at Inlet 1. Naalya WWSP serves a peri-urban residential area in the northeastern outskirts of Kampala, with an estimated catchment population of about 8,000 people. The network is a traditional gravity sewer that serves a relatively compact, clearly defined residential catchment. The confined nature of the catchment, the smaller and more stable contributing population, and the lower susceptibility to stormwater infiltration lead to more consistent hydraulic loading. Study design and sample collection Moore swab-based influent wastewater samples were collected weekly between March 2023 and May 2024 at the four wastewater treatment sites. Moore swabs were prepared by folding sterile gauze (120 cm long by 15 cm wide) into 8-ply square pads and securing them with a string long enough to allow for full submersion in the wastewater [15]. After autoclaving at 121 °C for 15 min, the swabs were allowed to cool and then sealed in sterile Ziplock bags for field deployment. Swabs were installed by attaching the string to a fixed structure, fully submerging them into the wastewater influent between 9:00 a.m. and 11:00 a.m., and leaving them in place for 24 h to ensure sufficient passive contact. After retrieval, the swabs were removed, packaged into clean Ziplock bags, and externally decontaminated with 70% ethanol. They were then placed into bio-bottles and transported in ice-cooled boxes to the Central Emergency Response and Surveillance Laboratory (CERSL) for analysis. Sample processing Upon receipt and accessioning at the laboratory, the wastewater was manually strained from each Moore swab into a sterile flask, yielding approximately 150 mL. The strained Moore swab was then immersed in 100 mL of phosphate-buffered saline (1X PBS) to extract residual microorganisms from the gauze matrix. The gauze was agitated in a rotary shaker at room temperature for 2 min. After shaking, the resulting volume of about 100 mL was combined with the manually strained wastewater from the original Moore Swab, generating approximately 250 mL of wastewater per sample for further processing. Viral capture and concentration A 10 mL aliquot of the resultant 250 mL wastewater suspension was transferred into a 15 mL Falcon tube (Corning Inc., Corning, NY, USA). A magnetic bead-based approach was used for viral concentration. Briefly, 100 µL of Nanotrap® Enhancement Reagent 2 (ER2) (SKU# 10112, Ceres Nanoscience, Inc., Manassas, VA) was added to the 10 mL wastewater aliquot and mixed by vortexing. Subsequently, 150 µL of Nanotrap® Microbiome A Virus Particles (SKU#44202, Ceres Nanosciences, Inc., Manassas, VA, USA) was added, mixed by inversion, and incubated at room temperature for 10 min. Pelleting was then done using the DynaMag™-15 magnet (Sergi Lab Supplies, Lubbock, Texas, USA), and the captured viral particles were eluted in 200 µL of Nanotrap® Buffer 2 (SKU# 10102-100, Ceres Nanosciences, Inc., Manassas, VA, USA). Nucleic acid extraction Nucleic acids were extracted from the 200 µL eluate using the QIAamp Viral RNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions. This silica-membrane-based method employs a chaotropic lysis buffer that simultaneously disrupts viral particles and inactivates nucleases, facilitating efficient RNA capture. A positive extraction control, a negative extraction control (nuclease-free water substituted for the sample), and a no-template control (NTC, nuclease-free water substituted into the RT-qPCR master mix) were included on every extraction batch and PCR plate, respectively. Extracted nucleic acids were eluted in a final volume of 60 µL and stored at − 80 °C until analysis. Detection and quantification of PMMoV RNA PMMoV RNA was quantified using the GoTaq® Enviro PMMoV Quant Kit (Promega Corporation, Madison, WI, USA) with the Quasar® 670 reporter dye system. Each 20 µL RT-qPCR reaction consisted of: 10 µL GoTaq® Enviro Master Mix (2×), 0.4 µL GoScript™ Enzyme Mix, 1 µL PMMoV Primer/Probe Mix (Quasar® 670, 20×), 3.6 µL nuclease-free water, and 5 µL extracted nucleic acid. The forward primer, reverse primer, and probe sequences used in this assay have been previously described by Kitajima et al. [16]. Reactions were performed on a CFX96™ Touch Real-Time PCR Detection System (Bio-Rad Laboratories, Hercules, CA, USA). Thermal cycling conditions were: reverse transcription at 45 °C for 15 min; initial denaturation at 95 °C for 2 min; followed by 40 cycles of denaturation at 95 °C for 15 s and combined annealing/extension at 62 °C for 60 s. Amplification efficiency was determined from the standard curve (see below) as required to fall within the acceptable MIQE range of 90–110% (R² ≥ 0.98). The Quasar® 670 channel was used to detect PMMoV RNA in the CFX96™ Plate Setup window. A sample was considered positive if the Cq value for PMMoV RNA was ≤ 40. Determination of PMMoV concentration in wastewater Genome copy (GC) quantification per reaction was determined using a linear regression standard curve with an R² value of 0.996 and 95.9% efficiency (slope − 3.65, intercept 42.15 derived from a 10-fold serial dilution of a synthetic 68-bp DNA oligonucleotide containing the PMMoV target amplicon region (range: 2.4 × 10⁷ to 2.4 × 10⁰ copies per reaction). The limit of detection (LOD) of the assay was defined as the lowest standard concentration producing a positive Cq in ≥ 95% of replicate reactions (LOD ~ 2.07 gene copies per reaction). Genomic copies per 100 mL of wastewater were calculated from genomic copies per reaction using the equation described by Flood et al. [17]. Climatic data Monthly average precipitation data for Kampala, Uganda, were obtained from nomadseason.com [18], based on information provided by the Copernicus Climate Change and Atmosphere Monitoring Services and the Uganda National Meteorological Authority. Due to the absence of concurrent daily gauge measurements at the study sites, long-term monthly average precipitation values derived from January 2017 to January 2022 were used as proxy rainfall estimates for the study period (March 2023–May 2024). The limitations of using historical averages in place of concurrent measurements are acknowledged and discussed in the Limitations section. Kampala experiences a tropical rainforest climate (Köppen–Geiger classification: Af), with a bimodal rainfall pattern. The primary rainy season occurs from March to May, with a secondary rainy season from October to November. Annual average precipitation is approximately 1,641 mm, distributed over approximately 271 rainy days. Rainfall is heaviest between March and May, averaging 26 rainy days and 192 mm per month, while the dry seasons occur from June to August and from December to January. Monthly average precipitation values were used in the correlation and regression analyses described below. Sunshine duration data were available from the same source but were not incorporated into the statistical models and are therefore not reported here. Data analysis Data preprocessing and non-detect handling Raw Cq and corresponding GC/reaction values from the 4 WWTPs (2023–2024) were cleaned and checked for completeness using Microsoft Excel 2016 for initial analysis, then imported into R Software Version 4.5.1 for statistical analysis and visualization [19]. Samples with no amplification before the maximum cycle threshold of 40 cycles were classified as non-detects (NDs). These were imputed using an LOD substitution approach, specifically LOD/2, to avoid bias from list-wise deletion and to enable log transformation [20]. We then computed log₁₀ GC/100 mL for distributional and regression analyses. Pre-analytical distributional assessment Normality of monthly mean log₁₀ GC/100 mL and average rainfall was evaluated using the Shapiro–Wilk test, histograms, and kernel density plots. Levene’s test assessed variance homogeneity across facilities, directly informing the choice between parametric and nonparametric tests. Descriptive and per capita statistics For each facility, the mean, median, SD, minimum, maximum, and IQR of log₁₀ GC/100 mL were computed. Weekly means were used for longitudinal visualization, with rainy seasons (March–May; October–November) demarcated. To determine whether between-facility differences reflected population size rather than infrastructure or hydrological factors, per capita concentrations (GC/100 mL ÷ estimated catchment population) were log₁₀-transformed and compared using the Kruskal-Wallis H test. Between-facility comparison of PMMoV RNA concentrations When Levene’s test confirmed heteroscedasticity, the Kruskal-Wallis H test assessed overall differences in concentration. Significant results (p < 0.05) were followed by Dunn’s post-hoc test with Bonferroni correction across all six pairwise facility comparisons. Correlation analysis Weekly log₁₀ GC/100 mL values were aggregated to monthly arithmetic means pooled across all four sites, then correlated with monthly average rainfall. Pearson’s r was the primary statistic; Spearman’s ρ was computed as a robustness check based on ranks. Linear mixed-effects model A linear mixed model (LMM) was fitted with centered monthly rainfall and site as fixed effects, and a random intercept for site to account for stable baseline differences between sites. Model fit was characterized by marginal and conditional R² (MuMIn package). Fixed-effect coefficients, SEs, and Satterthwaite p-values are reported alongside variance components (σ²u, σ²). Residual diagnostics (Q-Q and residuals-vs-fitted plots) confirmed linearity, normality, and homoscedasticity. A supplementary pooled simple linear regression (rainfall as sole predictor) was fitted to enable comparison with prior literature. Statistical significance was set at α = 0.05. Results Monthly mean log10 PMMoV RNA GC/100 mL (pooled across sites) and monthly average rainfall (mm) conformed to a normal distribution (PMMoV RNA: W = 0.929, p = 0.262; Rainfall: W = 0.909, p = 0.130; Supplementary file), supporting Pearson’s r as the primary correlation statistic. Levene’s test for homogeneity of variance applied to site-level weekly concentrations revealed significant heteroscedasticity across the four facilities (F = 31.71, p < 0.0001; Supplementary file), confirming that the assumption of equal variances was violated and that the non-parametric Kruskal-Wallis H test was appropriate for between-facility comparisons. Facility-level PMMoV RNA concentrations A total of 244 wastewater samples were collected from the four treatment facilities covering 61 epidemiological weeks from Week 12 of March 2023 to Week 20 of May 2024. Five samples (2.05%) produced no amplification before the maximum cycle threshold (Cq = 40 cycles) and were classified as NDs. All five originated from Bugolobi FSTP. These were assigned a value of LOD/2 = 123.94 GC/100 mL (log10 = 2.09) (Supplementary file). PMMoV RNA was detected in 239 of 244 samples (97.95%) across all four facilities. Descriptive statistics are summarized in the supplementary file, and the distribution is shown Fig. 2. Naalya WWSP exhibited the highest and most stable concentrations (mean = 6.09 log10 GC/100 mL, SD = 0.59, IQR = 1.03), consistent with its small, contained peri-urban catchment and stable hydraulic loading. Bugolobi FSTP showed the lowest central tendency (mean = 5.27) and the greatest variability (IQR = 2.60), reflecting the intermittent, user-driven nature of fecal sludge disposal rather than a continuous gravity-sewerfeed. Nakivubo WWTP Inlet 2 had intermediate but comparatively stable concentrations (mean = 5.72, IQR = 1.14), while Nakivubo WWTP Inlet 1, serving the central business district through a flood-prone combined sewer network, showed moderate mean concentrations (mean = 5.53 log10 GC/100 mL) alongside the highest variability among the piped-sewer sites (IQR = 3.16). Fig. 2Box-whisker plot of PMMoV RNA concentrations (log₁₀ GC/100 mL) across four sampling facilities in Kampala, Uganda (March 2023–May 2024) Box-whisker plot of PMMoV RNA concentrations (log₁₀ GC/100 mL) across four sampling facilities in Kampala, Uganda (March 2023–May 2024) Temporal trends in PMMoV RNA concentrations across wastewater facilities Weekly PMMoV RNA trajectories show fluctuating viral loads within and across facilities (Fig. 3). Across all four sites, concentrations declined during Kampala’s rainy seasons (March–May and October–November) and recovered during dry periods (Supplementary file), demonstrating a consistent inverse seasonal signal. Bugolobi FSTP showed the most erratic profile, with abrupt peaks (Weeks 25, 36, 50) and troughs (Weeks 31, 44–45, 19–20 of 2024) driven by the intermittent desludging schedule rather than rainfall. Nakivubo WWTP Inlet 1 displayed sustained seasonal declines attributable to stormwater infiltration through its flood-prone combined sewer network. Nakivubo WWTP Inlet 2 and Naalya WWSP maintained comparatively narrow concentration ranges throughout the study period. Fig. 3Four-panel time series of weekly mean PMMoV RNA concentrations (log₁₀ GC/100 mL) at each sampling facility in Kampala, Uganda (March 2023–May 2024). Panel (A) shows Nakivubo WWTP Inlet 1 (catchment population n ≈ 67,000; combined sewer serving the CBD), panel (B) shows Nakivubo WWTP Inlet 2 (n ≈ 89,000; geographically contained sub-catchment), panel (C) shows Naalya WWSP (n ≈ 8,000; peri-urban stabilization pond with stable hydraulic loading), and panel (D) shows Bugolobi FSTP (n ≈ 220,000; faecal sludge treatment plant receiving intermittent vacuum tanker inputs). Blue shaded regions indicate rainy seasons (March–May and October–November). Red inverted triangles (▼) denote non-detect observations imputed as LOD/2. The horizontal dotted red line indicates the LOD/2 threshold (log₁₀ = 2.09) Four-panel time series of weekly mean PMMoV RNA concentrations (log₁₀ GC/100 mL) at each sampling facility in Kampala, Uganda (March 2023–May 2024). Panel (A) shows Nakivubo WWTP Inlet 1 (catchment population n ≈ 67,000; combined sewer serving the CBD), panel (B) shows Nakivubo WWTP Inlet 2 (n ≈ 89,000; geographically contained sub-catchment), panel (C) shows Naalya WWSP (n ≈ 8,000; peri-urban stabilization pond with stable hydraulic loading), and panel (D) shows Bugolobi FSTP (n ≈ 220,000; faecal sludge treatment plant receiving intermittent vacuum tanker inputs). Blue shaded regions indicate rainy seasons (March–May and October–November). Red inverted triangles (▼) denote non-detect observations imputed as LOD/2. The horizontal dotted red line indicates the LOD/2 threshold (log₁₀ = 2.09) Between-facility comparison of PMMoV RNA concentrations The Kruskal-Wallis H test indicated significant global differences in raw log10 PMMoV RNA concentrations across the four facilities (H = 9.83, df = 3, p = 0.020; Supplementary file). Applying the same test to per capita concentrations (GC/100 mL per person served) yielded a substantially stronger result (H = 84.29, p < 0.0001), demonstrating that between-facility differences are not merely a function of catchment population size but reflect structural differences in catchment infrastructure and hydraulic regime. Dunn’s post-hoc pairwise comparisons Dunn’s post-hoc test with Bonferroni correction (Supplementary file; Fig. 4) identified only one statistically significant pairwise difference in raw PMMoV RNA concentrations: Bugolobi FSTP versus Naalya WWSP (padj = 0.012). In the per capita analysis, Naalya WWSP differed significantly from all three remaining sites (padj < 0.0001), a pattern attributable to its exceptionally small catchment population (~ 8,000), which concentrates a stable fecal load within a low-flow, small-volume wastewater matrix. This hydraulic concentration effect most plausibly accounts for Naalya WWSP’s disproportionately high per capita signal and has a direct normalization implication: per capita adjustment in small, low-flow catchments risks systematically overestimating fecal signal strength relative to larger, higher-flow sites. More broadly, the pronounced divergence between raw and per capita Kruskal-Wallis statistics (H = 9.83 vs. H = 84.29) demonstrates that the four sites do not behave as population-scaled analogues of one another, confirming that population size is an unreliable predictor of PMMoV RNA concentration in structurally heterogeneous mixed sanitation systems. Fig. 4Dunn’s post-hoc pairwise comparison matrix (Bonferroni-adjusted p-values). Panel A: raw log₁₀ GC/100 mL. Panel B: per capita log₁₀ GC/100 mL/person. Green = p < 0.001; amber = p < 0.05; red = p ≥ 0.05 Dunn’s post-hoc pairwise comparison matrix (Bonferroni-adjusted p-values). Panel A: raw log₁₀ GC/100 mL. Panel B: per capita log₁₀ GC/100 mL/person. Green = p < 0.001; amber = p < 0.05; red = p ≥ 0.05 Relationship between rainfall and PMMoV RNA concentration PMMoV RNA levels decreased during months with heavy rainfall, particularly in March-May 2023, October 2023, and April 2024, when precipitation was high as shown in Fig. 5. Conversely, higher concentrations were observed in drier months, such as January, July, and August, which aligns with lower rainfall. Fig. 5Monthly mean PMMoV RNA concentrations (log₁₀ GC/100 mL, lines) at four sampling facilities overlaid on monthly average rainfall (mm, bars, right axis). Blue shaded regions indicate rainy seasons Monthly mean PMMoV RNA concentrations (log₁₀ GC/100 mL, lines) at four sampling facilities overlaid on monthly average rainfall (mm, bars, right axis). Blue shaded regions indicate rainy seasons Correlation analysis Site-level weekly log10 GC/100 mL values were aggregated to monthly arithmetic means pooled across all four facilities, producing N = 15 monthly observations (Supplementary file). A statistically significant, strong negative correlation was identified between monthly average rainfall and mean monthly PMMoV RNA concentration (Pearson r = − 0.623, p = 0.013), as shown in Fig. 6. This indicates that higher rainfall was consistently associated with lower PMMoV RNA concentrations in wastewater. Spearman’s ρ, computed as a robustness check under a rank-based, distribution-free framework, confirmed the direction and significance of the association (ρ = −0.642, p = 0.010). Fig. 6Scatter plot of monthly average rainfall (mm) against mean monthly log₁₀ PMMoV RNA concentration (GC/100 mL) pooled across four sites (N = 15 months). Points are coloured by month of the year (blue = dry months; red = wet months) Scatter plot of monthly average rainfall (mm) against mean monthly log₁₀ PMMoV RNA concentration (GC/100 mL) pooled across four sites (N = 15 months). Points are coloured by month of the year (blue = dry months; red = wet months) Linear mixed-effects model Rainfall was a highly significant negative predictor of PMMoV RNA concentration (β = −0.0077 log10 GC/100 mL per mm, z = − 4.041, p < 0.001; Supplementary file), independent of site, as shown in Fig. 7. Across a rainfall increase of 100 mm, well within Kampala’s seasonal wet-to-dry range, this corresponds to a reduction of 0.77 log10 units, equivalent to approximately a 6-fold decrease in PMMoV RNA concentration. In contrast, none of the site-level fixed effects were statistically significant (all p > 0.75; Supplementary file), indicating that, after controlling for rainfall, the four facilities do not differ significantly in adjusted mean PMMoV RNA. This suggests that the between-site differences identified by the Kruskal-Wallis test are largely attributable to differential exposure to rainfall and hydraulic vulnerability rather than to independent infrastructure effects. The intraclass correlation (ICC) of 0.50 confirms that exactly half of the total variance in PMMoV RNA is structural (stable between-site differences captured by the random intercept), while the remaining half is within-site temporal variation, the component most directly explained by rainfall. Variance components, model fit statistics and residual diagnostic plots are provided in the Supplementary file. Fig. 7Forest plot of fixed-effect coefficients from the primary linear mixed model (N = 244; 4 sites). Each point is the coefficient estimate; horizontal lines show 95% confidence intervals Forest plot of fixed-effect coefficients from the primary linear mixed model (N = 244; 4 sites). Each point is the coefficient estimate; horizontal lines show 95% confidence intervals Reference simple linear regression model Each 1 mm increase in rainfall corresponded to a decrease of 0.0085 log10 GC/100 mL (SE = 0.003, t = − 2.87, p = 0.013) as shown in Fig. 6. The model explained 38.8% of the variance in monthly PMMoV RNA levels (R² = 0.388, adjusted R² = 0.340, F = 8.23, p = 0.013). This model does not account for between-site heterogeneity and was applied for reference only; the linear mixed model remains the main analytical model. Discussion This study provides evidence that PMMoV RNA concentrations in tropical urban wastewater are systematically and substantially influenced by rainfall dynamics and catchment-specific infrastructure, generating both seasonal and site-specific variability that poses real challenges for its application as a fecal indicator in WBS programs. Our findings, however, derived from 14 months of weekly surveillance at four structurally distinct sites in Kampala, Uganda, demonstrate that this variability is not random noise but a predictable, mechanistically interpretable signal that must be explicitly accounted for if WBS data are to be reliably interpreted in tropical urban settings. Rainfall as the principal driver of PMMoV RNA variability The most salient finding of this study is that rainfall is a strong, statistically significant, and site-independent predictor of PMMoV RNA concentration. The linear mixed model identified rainfall as a highly significant negative predictor (β = −0.0077 log₁₀ GC/100 mL per mm, z = − 4.041, p < 0.001). Across a seasonal rainfall increase of 100 mm, well within the range of Kampala’s wet-to-dry transition, this corresponds to a reduction of approximately 0.77 log₁₀ units, equivalent to a roughly 6-fold decrease in PMMoV RNA concentration. This effect was corroborated by a strong bivariate correlation between monthly average rainfall and pooled site-mean PMMoV RNA (Pearson r = − 0.623, p = 0.013; Spearman ρ = −0.642, p = 0.010), with the Spearman coefficient confirming the result under a rank-based, distribution-free framework. These findings are consistent with reports from temperate urban settings, where wet-weather inflow and infiltration have been identified as critical determinants of reduced PMMoV RNA levels across wastewater treatment plants [12, 21]. Our study extends this evidence to a tropical bimodal rainfall regime with distinct wet and dry seasons, demonstrating that such dilution effects are systematic, recurrent across multiple seasons, and of sufficient magnitude to materially affect the interpretation of WBS data [13, 22]. Site infrastructure as a modifier of PMMoV RNA dynamics While rainfall was the overarching driver of PMMoV RNA variability, the four sampling sites exhibited distinct concentration profiles shaped by sewershed structure—infrastructure type, hydraulic configuration, and susceptibility to stormwater infiltration—rather than population size alone. Naalya WWSP produced the highest and most stable PMMoV RNA levels (mean = 6.09 log₁₀ GC/100 mL; IQR = 1.03), consistent with its small, geographically bounded peri-urban catchment, conventional gravity sewer, and minimal exposure to stormwater infiltration [22]. At the opposite extreme, Bugolobi FSTP exhibited the lowest central tendency (mean = 5.27 log₁₀ GC/100 mL) and the widest concentration range (IQR = 2.60), reflecting its fundamentally different influent regime: city-wide vacuum tanker desludging rather than a connected pipe network renders its influent timing irregular and its PMMoV RNA dynamics largely decoupled from rainfall- a pattern consistent with reports from other decentralized sludge systems in tropical African cities, where temporal variability has been linked to irregular emptying schedules [23]. Nakivubo WWTP Inlet 1, serving Kampala’s high-density CBD through a flood-prone, partially unsealed combined sewer, showed the highest variability among gravity-sewer sites (IQR = 3.16) and exhibited sharp, sustained PMMoV RNA declines during each rainy season due to stormwater infiltration. Nakivubo WWTP Inlet 2, covering a smaller sub-catchment with shorter conveyance distances, maintained substantially more stable concentrations (IQR = 1.14), serving as an internal reference that confirms infrastructure integrity mediates the degree of stormwater impact. The Kruskal-Wallis’s test corroborated these site-level distinctions, with between site differences in raw concentrations (H = 9.83, p = 0.020) that became markedly more pronounced on a per capita basis (H = 84.29, p < 0.0001). This divergence reflects two compounding phenomena: hydraulic concentration at small low-flow sites such as Naalya WWSP, where a stable fecal load is suspended in a reduced wastewater volume, and infrastructure-driven deviation from population scaling at larger sites where stormwater infiltration and irregular sludge inputs independently modulate concentration. Collectively, these results confirm that population size neither predicts nor explains PMMoV RNA concentration in Kampala’s mixed sanitation system, and that structural catchment factors are the primary determinants of between-site heterogeneity. Sewershed characterization, encompassing drainage area, infrastructure type, conveyance distance, and inflow and infiltration susceptibility, should be a prerequisite for WBS program design and data interpretation in comparable settings. Standardized profiling tools such as drainage area delineation, infrastructure condition assessments, and population density mapping could improve cross-site comparability and support the development of site-specific interpretation thresholds. Rainfall mediates, rather than confounds, site-level differences A critical insight from the LMM is that, after controlling for rainfall, none of the site-level fixed effects remained statistically significant (all p > 0.75). This indicates that the between-site differences in raw PMMoV RNA concentrations identified by the Kruskal-Wallis test are largely mediated by differential rainfall exposure and hydraulic vulnerability, rather than representing additive, independent infrastructure effects on PMMoV RNA production or transport. The intraclass correlation coefficient (ICC = 0.50) quantifies this: half of the total variance in PMMoV RNA is attributable to stable between-site structural characteristics, captured by the random intercept, while the remaining half is within-site temporal variation, the component most directly driven by rainfall. Taken together, these results suggest that a single, site-calibrated rainfall correction factor could substantially improve the reliability of PMMoV RNA as a fecal indicator across diverse catchment types in Kampala, rather than requiring fundamentally different analytical frameworks for each site. Implications for PMMoV RNA as a normalization tool in LMICs While this study was designed to characterize the temporal and spatial behavior of PMMoV RNA itself under tropical urban conditions, rather than to evaluate normalization performance directly against a pathogen target, our findings have clear implications for its use in ratio-based normalization. Normalization of pathogen concentrations by PMMoV RNA implicitly assumes that both analytes experience proportional dilution under the same hydrological conditions [7, 24, 25]. Our data challenges this assumption in two respects. First, the infrastructure-specific drivers of PMMoV RNA variability, particularly the batch-input dynamics at Bugolobi FSTP and the stormwater-driven declines at Nakivubo WWTP Inlet 1, are unlikely to affect pathogen targets proportionally, because pathogen shedding is governed by community infection prevalence rather than desludging schedules or sewer hydraulics. Second, the magnitude of rainfall-driven PMMoV RNA reduction (approximately 6-fold per 100 mm) is large enough that, if pathogen concentrations are diluted to a lesser degree, normalization would systematically underestimate pathogen prevalence during wet seasons. In settings where WBS informs outbreak detection or early-warning systems, precisely the context where timely public health action is most critical, such bias could lead to false reassurance that transmission is declining when it is not [2]. These concerns are compounded in LMIC contexts, where concurrent rainfall measurements may be unavailable, mixed sanitation systems complicate hydraulic modeling, and laboratory capacity may limit the simultaneous quantification of multiple fecal indicators [13]. Toward more robust WBS in tropical urban settings Our findings point to several practical strategies for improving PMMoV RNA-based WBS in similar settings. Site-specific regression models linking PMMoV RNA concentration to historical or concurrent rainfall, analogous to the relationship quantified here (SLR β₁ = −0.00851 log₁₀ GC/100 mL per mm; R² = 0.39), could be used to generate rainfall-adjusted concentration estimates during wet seasons. Parallel measurement of a complementary fecal indicator with different hydrological behavior, such as crAssphage, would permit cross-validation and reduce dependence on any single marker [7, 9]. Tiered interpretation frameworks, analogous to those developed for wastewater surveillance programs in high-income countries [4, 26], could flag data collected during rainy periods for heightened caution or mandatory adjustment before use in public health decision-making. Finally, the integration of real-time gauge or satellite-derived rainfall data into routine WBS data pipelines would enable dynamic, event-specific corrections rather than reliance on historical monthly averages. Collectively, these adaptations represent achievable, low-cost modifications to existing WBS infrastructure that would substantially improve data reliability and the credibility of wastewater-derived public health signals in tropical urban Africa and comparable LMIC settings. Limitations The use of long-term monthly rainfall averages (2017–2022), rather than precipitation measurements concurrent with sample collection, introduces potential inaccuracy in interpreting rainfall effects. These historical data may not capture inter-annual variability or recent shifts in rainfall patterns. They may obscure the impact of short-term rainfall events, such as intense storms, that can significantly influence wastewater dilution. This limitation was due to data availability at the time of the study. Future WBS programs in Kampala should incorporate real-time rainfall measurements, such as gauge or satellite-derived data, to improve the precision of environmental interpretations. Conclusion This study demonstrates that PMMoV RNA concentrations in wastewater are strongly influenced by rainfall and catchment characteristics, resulting in significant seasonal and site-specific variability in Kampala, Uganda. While PMMoV RNA remains a valuable normalization marker, its susceptibility to hydrological and infrastructural influences may lead to misinterpretation if applied without adjustment [26]. These findings challenge the assumption of proportional dilution underpinning ratio-based normalization. In tropical, seasonally dynamic settings, robust wastewater-based surveillance requires rainfall-adjusted interpretation and integration of complementary indicators. As wastewater surveillance expands globally, particularly in low- and middle-income countries, developing robust interpretation frameworks that are climate-resilient and explicitly account for infrastructural conditions will be critical to ensure data reliability and translate wastewater signals into timely, actionable public health intelligence. Supplementary Information Supplementary Material 1: Data analysis output tables and figures. Supplementary Material 1: Data analysis output tables and figures.