Being able to quickly and efficiently diagnose COVID-19 is essential in monitoring the pandemic. Because the sampling process for saliva is noninvasive, and because it is inexpensive and minimizes the risk for transmissions to health care workers (Baghizadeh Fini, 2020), saliva sampling has excellent potential and advantages over other sampling methods from biological specimens such as the lower and upper respiratory tract (Wyllie et al., 2020; To et al., 2020). Given the significant individual heterogeneity in the saliva viral shedding (Ke et al., 2022; Hay et al., 2022), identifying biomarker(s) for viral shedding patterns will be crucial for improving public health interventions in the era of living with COVID-19.

To improve our understanding of SARS-CoV-2 infection dynamics in saliva to enable application of saliva testing in the fight against COVID-19, we quantified and stratified longitudinal virus dynamics in saliva samples from 144 mildly symptomatic individuals from the cohorts of the NFV clinical trial (Hosogaya et al., 2021) and the University of Illinois at Urbana-Champaign (Ke et al., 2022), and we uniquely analyzed the relationships among viral dynamics, clinical data, and micro-RNAs. Our mathematical modeling analysis indicates that viral dynamics in saliva may exhibit distinct patterns compared to those in the upper respiratory tract. We estimated that viral replication in saliva is characterized by a relatively rapid early growth phase, with a mean (standard deviation) doubling time of 1.44 (0.49) hours. Compared with prior studies analyzing viral dynamics in the upper respiratory tract using similar models, which reported doubling times of 2–4 hr (Ke et al., 2022; Gunawardana et al., 2022; Iyaniwura et al., 2024), our findings suggest that viral replication in saliva proceeds faster than in the upper respiratory tract. Multiple previous studies have also shown that viral loads in saliva rise more rapidly than in the nasal cavity, are detected with higher sensitivity early in infection, and reach their peak earlier (Ke et al., 2022; Migueres et al., 2022; Puhach et al., 2023; Savela et al., 2022; Smith et al., 2021).

In addition to the large heterogeneity in virus infection dynamics, we identified three groups (i.e. G1, G2, and G3) with different viral shedding patterns (Figure 2D). Immunocompromised patients have been reported to have a prolonged duration of viral RNA detection, lasting over three months, underscoring the critical role of host immune responses in controlling viral infections (Leung et al., 2022; Niyonkuru et al., 2021; Nakajima et al., 2021; Wei et al., 2021). Although oral immune responses remain poorly understood, Huang et al. recently confirmed by using single-cell RNA sequencing of the human minor salivary glands and gingiva that SARS-CoV-2 infection can trigger sustained, localized immune responses in saliva (Huang et al., 2021). In this study, we observed significant differences in the down-slopes of viral shedding in saliva among participants in different groups, with a more rapid decline in G1. This decline is likely attributed to a stronger immune response to SARS-CoV-2 in G1 participants than in participants in G2 and G3, as reflected in the death rate of infected cells due to the immune response (Figure 2D). Lower levels of viral replication have also been observed among infected participants with high baseline levels of mucosal IgA (but not IgG), as reported elsewhere (Havervall et al., 2022). Recently, we demonstrated that rapid anti-spike secretory IgA antibody responses can contribute to reducing duration of viral RNA detection and amounts in nasopharyngeal mucosa (Miyamoto et al., 2023). These findings highlight the importance of biomarkers that directly reflect an individual’s immune response, such as virus-specific antibody induction and T cell levels etc., in predicting viral shedding patterns. Therefore, quantifying the time-series pattern of mucosal IgA and its correlation with saliva viral load may provide crucial insights into the stratification of SARS-CoV-2 infection dynamics.

For the purpose of predicting viral shedding patterns during the early stage of infection, we first explored the association of 39 basic clinical variables, 8 daily symptoms, and the levels of 92 micro-RNAs with the stratified groups. However, none of the factors were significant (Table 1, Figure 3A, Figure 4B, Supplementary file 1A and Supplementary file 1F). On the other hand, it is noteworthy that all infection cases were mild and that most participants had clinical indicators within normal ranges. This lack of clinical heterogeneity within the cohort may have limited the ability to fully capture the diversity of infection dynamics. Moreover, the clinical parameters analyzed in this study are, a priori, unlikely to exhibit strong correlations with virologic outcomes. In contrast, we showed that mir-1846, which is an exogenous micro-RNA that is specifically classified as an Oryza sativa micro-RNA (osa-microRNA; Rakhmetullina et al., 2020), may exhibit a weak negative correlation. Exogenous micro-RNAs enter the human body primarily through food and can affect human metabolism by interacting and binding with human genes. mir-1846 is reported to interact with two human genes (Rakhmetullina et al., 2020) that are known to be associated with the progression of melanoma, various cancers, and leukemia. This suggests that mir-1846 levels may be linked to human immunity. Few studies have investigated the role of mir-1846 in humans, but our findings suggest the need for further investigations into the impact of this micro-RNA level on human immunity. Our research sheds light on the intricate patterns of viral shedding in saliva.

Our approach has several limitations that must be considered in our next study: First, our analysis was limited to participants with symptomatic infection and excluded those with asymptomatic infection (22 asymptomatic individuals out of a total of 182 individuals, i.e. 12% of participants) because we integrated datasets with different time scales from different cohorts. Although our data do not include participants infected with omicron variants, others have reported that the omicron variant may cause a higher proportion of asymptomatic infection (Garrett et al., 2022). Thus, evaluating the effect of asymptomatic infection will be important to update our stratification, especially for recent (or future emerging) VOCs. Second, potential viral rebound was neither prespecified nor systematically assessed. A subset of participants exhibited patterns consistent with possible rebound (e.g. S01-16 and S01-43 in Figure 2—figure supplement 1), which could affect estimates of viral RNA detection duration. Future studies should predefine an operational criterion for viral rebound and explicitly incorporate it into the modeling framework to strengthen robustness. Since both models considered in the present study cannot account for viral rebound, a more complex model would be required to capture this phenomenon. Third, micro-RNAs participate in the post-transcriptional regulation of gene expression; however, they do not provide direct insights into immune cell dynamics. Given the reported association between the duration of viral RNA detection and mucosal immunity as discussed above, it appears imperative to analyze modalities that are directly linked to the immune response in the future. Fourth, some of our results may have limited relevance to the current COVID-19 situation, as most people have now either been infected or vaccinated. Nevertheless, investigating the relationship between viral shedding patterns in saliva and various clinical and microRNA data, and developing a method to do so, remains important. Such research can offer valuable insights into the early response to emerging infectious viruses in the future.

Another potential limitation of this study is the timing of saliva specimen sampling, although we took great care to select and compare specimens from G1, G2, and G3 without bias. As a result of our clinical trial design (jRCT2071200023 Hosogaya et al., 2021; Miyazaki et al., 2023), participants were enrolled after the onset of symptoms, thereby restricting saliva specimen collection exclusively to the post-symptom phase. Unfortunately, we lack samples from the pre-infection, pre-symptomatic, and early infection phases. Consequently, the absence of individual-level baseline values for micro-RNA means that inter-participant heterogeneity in micro-RNA levels may obscure signals related to distinct viral infection dynamics in saliva.

In conclusion, our study revealed that the dynamics of SARS-CoV-2 infection in saliva can be classified into three groups based mainly on the duration of viral RNA detection. However, accurately predicting the variability in viral dynamics remains a challenging task, because it requires a more comprehensive understanding of the complex shedding patterns in saliva, as well as detailed clinical and molecular data. The identification of a sensitive, simple, and rapid biomarker for saliva viral shedding will be imperative for future COVID-19 outbreak control.