Research Article

Microbiology and Infectious Disease

Stratification of viral shedding patterns in saliva of COVID-19 patients

interdisciplinary Biology Laboratory (iBLab), Division of Natural Science, Graduate School of Science, Nagoya University, Japan
School of Biomedical Convergence Engineering, Pusan National University, Republic of Korea
Department of Science System Simulation, Pukyong National University, Republic of Korea
Department of Mathematics, Pusan National University, Republic of Korea
Lee Kong Chian School of Medicine, Nanyang Technological University, Singapore
The Tokyo Foundation for Policy Research, Japan
International Research Center for Neurointelligence, The University of Tokyo Institutes for Advanced Study, The University of Tokyo, Japan
Department of Chemotherapy and Mycoses, National Institute of Infectious Diseases, Japan
Research Center for Drug and Vaccine Development, National Institute of Infectious Diseases, Japan
Department of Microbiology, University of Illinois at Urbana-Champaign, United States
Department of Statistics, University of Illinois at Urbana-Champaign, United States
Theoretical Biology and Biophysics, Los Alamos National Laboratory, United States
Institute of Mathematics for Industry, Kyushu University, Japan
Institute for the Advanced Study of Human Biology (ASHBi), Kyoto University, Japan
Interdisciplinary Theoretical and Mathematical Sciences Program (iTHEMS), RIKEN, Japan
NEXT-Ganken Program, Japanese Foundation for Cancer Research (JFCR), Japan
Science Groove Inc., Japan
Division of Respirology, Rheumatology, Infectious Diseases, and Neurology, Department of Internal Medicine, Faculty of Medicine, University of Miyazaki, Japan

Jan 16, 2026

https://doi.org/10.7554/eLife.96032.3

Open access
Copyright information

Figures
Tables
Additional files

5 figures, 1 table and 2 additional files

Figures

Figure 1

Download asset Open asset

Characteristics of cohorts from the NFV clinical trial (jRCT2071200023 Hosogaya et al., 2021; Miyazaki et al., 2023) and the University of Illinois (Ke et al., 2022).

(A) Flowchart of cohorts from the NFV clinical trial and the University of Illinois along with the number of participants and inclusion criteria for our analysis is described. (B) Data collection schedule of viral load, blood test, and vital signs from participants in the NFV clinical trial is described. The gray box highlights the measurement dates for the observations included in the analysis. (C) and (D) show, for each participant (N=144 participants, 2191 samples), the timeline of sample collection and the captured SARS-CoV-2 viral RNA load for saliva RT-qPCR, respectively. The red and blue colors indicate samples for cohorts from the NFV clinical trial and the University of Illinois, respectively.

Figure 2 with 5 supplements

Download asset Open asset

Stratification of individual SARS-CoV-2 viral dynamics in saliva.

(A) UMAP of stratified viral RNA load based on the extracted features from the reconstructed individual-level viral dynamics is shown. (B) The reconstructed individual viral RNA load is shown. Colors for individual-level viral dynamics correspond to the colors of the dots in the UMAP described in (**A). (C**) The time-course patterns of each group highlighted by the Partial Least-Squares Discriminant Analysis (PLS-DA). (D) Distributions between groups of each feature used for stratification of viral shedding patterns are shown. The p-values of ANOVA for the difference in each feature among stratified groups are all less than 0.05. (E) Distributions of the number of individuals in each stratified group for the standard-of-care alone (left, n=97) and standard-of-care plus NFV administration (right, n=47) participants are shown. (F) Distributions of the number of individuals in each stratified group for Alpha variants (left, n=30), Delta variants (middle, n=13), and other variants (right, n=66) of SARS-CoV-2 are shown.

Figure 2—figure supplement 1

Download asset Open asset

Reconstructed viral dynamics in saliva samples for individual participants.

The individual-level model fits to saliva RT-qPCR results, based on the target-cell-limited model described in Equations 1-2, are presented for the same cohorts shown in Figure 1. Black and red closed dots represent measurements above and below the detection limit, respectively, with the black dashed line indicating the detection limit (1.08 log₁₀ copies/mL). Solid curves depict the reconstructed viral dynamics, while the shaded areas represent the corresponding 95% confidence intervals obtained using a bootstrap approach. Curves and shaded areas for individuals from the NFV clinical trial and the University of Illinois cohort are shown in pink and sky blue, respectively.

Figure 2—figure supplement 2

Download asset Open asset

Visual predictive checks (VPC) for the viral load models.

Observed viral load data (red lines and points) are compared with simulated prediction intervals from the models (blue shaded areas, 95% prediction interval) for the target-cell limited model (left) and the immune effector cell model (right). The VPC shows that both models adequately capture the central tendency and variability of the observed data over time since symptom onset. In the later phase, some deviations are present, but these are similarly observed across both models.

Figure 2—figure supplement 3

Download asset Open asset

Convergence diagnostics of the SAEM algorithm.

(**A, B**) Trajectories of population parameter estimates (with $ω$ denoting the variance components of the random effects) over SAEM iterations are shown for the target-cell limited model (A) and the immune effector cell model (B), respectively. In both models, parameter estimates stabilized after the burn-in period, indicating convergence. (C) Trajectories of the coefficient estimates in the linearized standard-error model are shown for the target-cell limited model (left) and the immune-effector cell model (right), respectively. (D) Convergence indicator values for the target-cell limited model (left) and the immune effector cell model (right) are shown, respectively. This indicator combines information on parameter stability and the Fisher information matrix. For both models, stabilization of the indicator after approximately N iterations indicates that the estimation procedure had reached convergence.

Figure 2—figure supplement 4

Download asset Open asset

Comparison of three model fits to viral load in saliva samples for individual participants.

Three different individual-level model fits to saliva RT-qPCR results using the target-cell-limited model and the immune effector model described in Equations 1-2 and Equations 3-6, respectively, are shown. Black and red closed dots represent measurements above and below the detection limit, respectively, with the black dashed line indicating the detection limit (1.08 log₁₀ copies/mL). Black solid curves indicate the reconstructed viral dynamics using the target-cell-limited model presented in Figure 2—figure supplement 1. The orange and blue curves indicate the estimated viral dynamics based on the immune effector model for all 144 individuals shown in Figure 1 and for the 54 individuals only in the University of Illinois cohort, respectively. Namely, the blue dashed curves are the model fits presented in Ke et al., 2022.

Figure 2—figure supplement 5

Download asset Open asset

Sensitivity analysis of the detection limit for viral load data in the NFV cohort.

Three model fits to saliva RT-qPCR results using the target-cell-limited model described in Equations 1-2. The black, pink, and green curves correspond to the model fits obtained when the detection limit for the NFV cohort’s viral load data was set to 1.08 log₁₀ copies/mL (same as the Illinois cohort), 2 log₁₀ copies/mL, and 0 log₁₀ copies/mL, respectively.

Figure 3

Download asset Open asset

Correlation between clinical data and viral shedding patterns.

(A) p-Values of ANOVA corrected by the FDR to compare clinical data among the three stratified groups are shown. Clinical data are listed in reverse order of p-values. (B) and (C) show ROC curves of random forest classifiers trained on predicting each group by using data for (B) clinical values described in Table 1 and (C) symptom onset data described in Supplementary file 1A, respectively. The corresponding AUC (area under curve) value of each ROC curve is shown on the top of each panel.

Figure 4 with 1 supplement

Download asset Open asset

Correlation between micro-RNA data and viral shedding patterns.

(A) The strategy of micro-RNA data collection from saliva samples in the NFV clinical trial is described. We picked a total of 30 participants by choosing 10 participants from each group. We chose two samples (the nearest to estimated peak and the most distant but above the detection limit in the late phase) from each participant for quantifying micro-RNA. (B) p-Values of Kruskal-Wallis ANOVA corrected by the FDR for each micro-RNA level are shown. Micro-RNA levels are listed by reverse order of p-values. Only the 24 micro-RNA levels with the lowest p-values are shown. (C) ROC curves of random forest classifiers trained on predicting each group by using levels of 92 micro-RNAs are shown. The corresponding AUC value of each ROC curve is presented on the top of each panel.

Figure 4—figure supplement 1

Download asset Open asset

Correlation between mir-1846 level and the features of infection dynamics.

The correlation between the rank of mir-1846 and the rank of (A) the peak viral load, (B) the duration of viral RNA detection, (C) up-slope and (D) down-slope is shown, respectively. The Spearman’s correlation coefficients and p-values are presented at the top of each panel. The black solid line and the gray shaded area indicate results of the linear regression and their 95% confidence levels, respectively.

Author response image 1

Download asset Open asset

Comparison of model fits before and after removing data suspected of rebound.

Black dots represent observed measurements, and the black and yellow curves show the fitted viral dynamics for the full dataset and the dataset with rebound data removed, respectively.

Tables

Table 1

Clinical data of the overall cohort from the NFV clinical trial and in groups stratified by longitudinal virus dynamics.

Clinical data	Unit	Group1 (N=33)	Group2(N=37)	Group3(N=20)	Overall(N=90)
Basic demographic information
Age	years	47 (13.9)*	40.1 (12.6)	43.4 (13.7)	43.7 (13.6)
Sex ^‡	-	46%	67%	70%	59%
Vital signs
Systolic blood pressure	$m m H g$	116 (16.3)	119.8 (14.4)	122.2 (20.3)	118.8 (16.6)
Diastolic blood pressure	$m m H g$	77.5 (14.3)	75.9 (10.9)	78.5 (14.8)	77.1 (13.2)
Pulse rate	$b p m$	83.5 (16.9)	86.9 (14.3)	79.1 (13.1)	83.8 (15.3)
SpO₂	%	97.5 (1)	97.7 (0.8)	97.6 (0.9)	97.6 (0.9)
Respiratory rate	$b p m$	17 (3.9)	16.5 (3.4)	16.8 (2.4)	16.8 (3.4)
Blood test results
White blood cell count	$10^{9} / L$	5.1 [30.2]^†	5.3 [31.9]	5.9 [34]	5.3 [32.8]
Neutrophil	%	62.4 (10.3)	60 (11.1)	59.8 (11.4)	61 (10.8)
Eosinophil	%	1 (1.8)	0.9 (1.5)	1 (1.2)	1 (1.6)
Basophil	%	0.3 (0.2)	0.4 (0.3)	0.4 (0.3)	0.3 (0.3)
Lymphocytes	%	26.1 (8.4)	28.1 (10)	28 (9.2)	27.3 (9.1)
Monocyte	%	10.1 (3.7)	10.6 (3.5)	10.8 (4.2)	10.4 (3.7)
Red blood cell count	$10^{12} / L$	5 [450.3]	5.5 [464.1]	5.4 [450.9]	5.3 [453]
Amount of hemoglobin	$g / d L$	14 (1.4)	15.1 (1.1)	14.8 (1.3)	14.6 (1.4)
Hematocrit	%	41.6 (3.6)	44.1 (3.2)	43.8 (3.6)	43 (3.6)
Platelet count	$10^{9} / L$	121.6 (88.7)	127.8 (98)	136.6 (102.4)	127.2 (94.4)
CRP	$m g / d L$	1.3 (2.6)	1.3 (1.9)	1.1 (1.8)	1.3 (2.2)
Protein	$g / d L$	7.4 (0.4)	7.4 (0.4)	7.4 (0.3)	7.4 (0.4)
Albumin	$g / d L$	4.2 (0.3)	4.3 (0.3)	4.4 (0.3)	4.3 (0.3)
ALT	$U / L$	20.2 (11.9)	29.9 (23.9)	31.3 (22)	26.2 (19.8)
AST	$U / L$	21.9 (7.8)	28.3 (14)	28.7 (13.4)	25.8 (12)
γ-GTP	$U / L$	30.9 (25.6)	42 (41.8)	47.8 (35)	38.7 (34.7)
ALP	$U / L$	69.9 (20.9)	72 (20.3)	72.8 (20.6)	71.3 (20.4)
LDH	$U / L$	172.1 (34.1)	184.1 (43.8)	186.6 (29.5)	179.7 (37.2)
Bilirubin	$m g / d L$	0.5 (0.2)	0.5 (0.2)	0.6 (0.2)	0.5 (0.2)
CK	$U / L$	85.2 (48.2)	126.7 (146.5)	118.2 (133)	107.7 (113.1)
Na	$m E q / L$	139.3 (2.9)	139.9 (2.2)	139.8 (2.8)	139.6 (2.6)
K	$m E q / L$	3.9 (0.3)	3.9 (0.3)	3.9 (0.2)	3.9 (0.3)
Cl	$m E q / L$	102.6 (3.3)	102.7 (3.6)	102.8 (2.7)	102.6 (3.3)
BUN	$m g / d L$	11.4 (3)	11.3 (2.5)	12.5 (3.1)	11.6 (2.9)
CRE	$m g / d L$	0.8 (0.2)	0.8 (0.2)	0.9 (0.2)	0.8 (0.2)
Blood glucose	$m g / d L$	105.9 (19.9)	102.4 (17.1)	109 (15.7)	105.3 (18)
HbA1c	%	5.6 (0.3)	5.5 (0.4)	5.6 (0.3)	5.6 (0.3)
Procalcitonin	$μ g / L$	0.1 (0.2)	0.1 (0.1)	0.1 (0.04)	0.1 (0.1)
Fibrinogen	$m g / d L$	371 (94.8)	365.3 (101.1)	2 (73)	364.1 (92.3)
PT	%	103.1 (20.5)	105.5 (17.5)	109.3 (15.5)	105.4 (18.4)
APTT	$s e c$	31.4 (3.7)	31.1 (5.3)	30.7 (3.2)	31.1 (4.2)
PT-INR	-	1 (0.1)	1 (0.1)	1 (0.1)	1 (0.1)
D-dimer	$m g / L$	0.8 (1.3)	0.5 (0.2)	0.5 (0.1)	0.6 (0.8)