Absent pleocytosis in infant viral central nervous system infections reflects lymphocyte depletion and cytokine activation Authors: Clare Mills, Holly Drummond, Helen E. Groves, Etimbuk Umana, Thomas Waterfield Published: 2026 ABSTRACT Enterovirus and parechovirus infections are common causes of fever in young infants and frequently prompt evaluation for meningitis. A notable feature of these infections is that many infants have detectable viral genetic material in the cerebrospinal fluid but no increase in white blood cells. This absence of pleocytosis has been attributed to early presentation, immature or suppressed immune responses, or incidental virus detection during systemic infection rather than true central nervous system (CNS) infection or meningitis. Here, we clarify the clinical and biological significance of absent pleocytosis in infant viral CNS infections using plasma proteomic profiling and detailed clinical data from a prospective multicentre cohort of febrile infants. We show that infants without pleocytosis display lymphopenia and marked activation of inflammatory and apoptotic pathways in the plasma, implicating lymphocyte depletion in impaired cerebrospinal fluid leucocyte recruitment and supporting true neurological infection. Introduction Enterovirus (EV) and parechovirus (PeV) central nervous system (CNS) infections are recognised as important causes of febrile illness in young infants, especially as diagnostic testing expands due to the low risk of invasive bacterial infection (IBI) in those with confirmed viral meningitis1–4. While outcomes are generally favourable, there remains a low but clinically significant risk of severe illness in this age group, and evidence on more subtle long-term outcomes is mixed5–8. Viral meningitis is typically defined by the detection of a virus in the cerebrospinal fluid (CSF) alongside compatible clinical features and CSF pleocytosis. In young infants, however, pleocytosis and neurological signs are frequently absent despite viral detection in CSF, raising uncertainty as to whether these cases represent CNS infection2,6,9. Viral CNS infection without CSF pleocytosis is often attributed to early or evolving viral meningitis, particularly in young infants who present soon after symptom onset, before pro-inflammatory cytokine signalling has resulted in recruitment of white blood cells (WBCs) into the CSF6,10. However, studies relating symptom duration to CSF pleocytosis are mixed9,11–14. Infants with viral CNS infection and CSF pleocytosis absence consistently exhibit lower peripheral WBC counts compared to those with pleocytosis. This has also been attributed to early presentation time, as well as immune system immaturity in infants, whereby a robust inflammatory response is not generated due to young age6,9,12,15. It is also feasible that both lower peripheral WBC counts and lack of CSF pleocytosis could be due to a virus-induced suppression of WBCs, which can occur through several mechanisms2,16. Alternatively, studies assessing viral load and genotype in children with viral CNS infection have indicated that there may be two distinct infection profiles, related to virus type. Namely, infections with CNS involvement or primarily systemic infections where viral RNA may have leaked into the CSF and is a coincidental finding of viremia14,17,18. Indeed, there remains no consensus on the mechanisms underlying the absence of pleocytosis in infants, nor on whether the sole detection of EV or PeV RNA in the CSF reliably indicates a true viral CNS infection. Here, we investigate the relationship between plasma neuroinflammatory and immune responses, CSF pleocytosis, and symptom duration in infants with EV and PeV CNS infection in an exploratory, hypothesis-generating analysis. We show that infants without pleocytosis exhibit lymphopenia and activation of plasma inflammatory and apoptotic pathways, whereas those with pleocytosis demonstrate a more muted systemic response. We further show that symptom duration does not account for these differences. These findings indicate that systemic lymphocyte depletion may impair WBC recruitment into the CSF and support the interpretation of viral detection in the cerebrospinal fluid as true CNS infection, regardless of pleocytosis. Results Patient characteristics and clinical comparisons A total of 603 infants ≤90 days of age were included in this study, with 197/603 (32.7%) cases of virus-positive CSF, with a prevalence of 21/603 (3.5%) PeV, and 173/603 (28.7%) EV (Supplementary Fig. 1). Viral blood PCR testing was performed in 87/603 (14%) infants. Patient demographics and clinical features are summarised in Table 1. No significant differences in age or sex were observed between virus-positive and virus-negative groups, or between EV and PeV infections. IBI was more frequent in virus-negative CSF infants (35/406, 8.6%) compared to virus-positive (1/197, 0.5%, P < 0.001), with only one IBI case (bacteremia) in an infant with EV-positive CSF. Compared to virus-negative infants, those with EV or PeV-positive CSF exhibited increased incidence of rash (P = 0.001) and fewer respiratory symptoms (P < 0.001). In the full cohort, 11/603 (1.8%) infants with complete CSF testing had signs of meningism.Table 1Patient characteristics and demographicsComplete CSF testingVirus positive CSF**Virus negative CSFAdjusted P Value (CSF virus + vs −)PeV positive CSFEV positive CSFAdjusted P Value (PeV vs EV CSF)Total, n (%)603197 (32.7)406 (67.3)21 (3.5)173 (28.7)Males, n (%)365 (60.5)117 (59.4)248 (61.1)0.81312 (57.1)104 (60.1)0.919Age (days) Median (IQR)39 (24–58)36 (25–53)41 (24–59)0.07929 (17–49)36 (25–53)0.317Temperature(°C) Median (IQR)38.1 (37.6–38.5)38.2 (37.8–38.5)38.0 (37.5–38.5)0.05138.0 (37.6–38.6)38.2 (37.9–0.919Unwell Appearance n (%)406 (67.3)141 (71.6)265 (65.3)0.18616 (76.2)124 (71.7)0.919Signs of Meningism n (%)11 (1.8)3 (1.5)8 (2.0)0.9990 (0.0)3 (1.7)0.999Rash n (%)120 (19.9)56 (28.4)64 (15.8)0.001*5 (23.8)51 (29.5)0.919GI Symptoms n (%)176 (29.2)46 (23.4)130 (32.0)0.0514 (19.0)42 (24.3)0.681Behaviour Change n (%)473 (78.4)163 (82.7)309 (76.1)0.11020 (95.2)141 (81.5)0.305Respiratory Symptoms, n (%)215 (35.7)42 (21.3)173 (42.6)< 0.001*3 (14.3)38 (22.0)0.862CSF WBC count (x 109/L) Median (IQR)n = 548 1.0 (0.0–4.95)n = 177 2.3 (0.3–38.0)n = 3711.0 (0.0–3.0)< 0.001*n = 19 0.3 (0.0–2.0)n = 155 3.4 (0.75–81.0)0.005*CSF pleocytosis n (%)76 (13.9)54 (30.5)22 (5.9)< 0.001*0 (0.0)54 (34.8)0.002*Peripheral WBC count (x 109/L) Median (IQR)n = 578 9.6 (6.8–12.9)n = 189 8.6 (6.6–11.4)n = 389 10.3 (7.1–13.9)< 0.001*n = 19 5.7 (4.8–8.1)n = 167 8.7 (6.7–11.5)0.002*Lymphocyte count (x 10*9/L) Median (IQR)n = 577 3.7 (2.4–5.4)n = 189 3.1 (2.0–4.7)n = 388 4.0 (2.5–5.7)0.001*n = 19 1.9 (1.2–3.6)n = 167 3.2 (2.1–4.7)0.005*Lymphopenic n (%)287 (49.7)113 (59.8)174 (44.8)0.012*15 (78.9)97 (58.1)0.219Neutrophil count (x 10*9/L) Median (IQR)n = 577 4.0 (2.5–6.6)n = 189 4.0 (2.8–5.9)n = 388 3.91 (2.4–7.1)0.999n = 19 2.8 (1.7–3.4)n = 167 4.19 (3.0–6.0)0.002*CRP (mg/L) Median (IQR)n = 595 10.0 (4.0–29.0)n = 195 10.0 (4.0–22.0)n = 400 10.9 (3.0–43.83)0.186n = 21 5.0 (3.0–8.5)n = 171 10.0 (5.0–22.0)0.005*LOS (days) Median (IQR)3 (2–4)3 (2–4)3 (2–4)0.1883 (3–4)3 (2–4)0.419IBI n (%)36 (6.0)1 (0.5)35 (8.6)< 0.001*0 (0.0)1 (0.6)0.999Continuous variables were compared using two-tailed Mann–Whitney U tests, and categorical variables were compared using two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant; FDR < 0.05, **Viruses in CSF: EV n = 173, PeV n = 21, Adenovirus n = 1, Human herpesvirus 6 n = 2. CRP C-reactive protein; CSF cerebrospinal fluid; EV Enterovirus; GI gastrointestinal; IBI invasive bacterial infection; IQR interquartile range; LOS length of stay; PeV Parechovirus; WBC white blood count. Table 1Patient characteristics and demographicsComplete CSF testingVirus positive CSF**Virus negative CSFAdjusted P Value (CSF virus + vs −)PeV positive CSFEV positive CSFAdjusted P Value (PeV vs EV CSF)Total, n (%)603197 (32.7)406 (67.3)21 (3.5)173 (28.7)Males, n (%)365 (60.5)117 (59.4)248 (61.1)0.81312 (57.1)104 (60.1)0.919Age (days) Median (IQR)39 (24–58)36 (25–53)41 (24–59)0.07929 (17–49)36 (25–53)0.317Temperature(°C) Median (IQR)38.1 (37.6–38.5)38.2 (37.8–38.5)38.0 (37.5–38.5)0.05138.0 (37.6–38.6)38.2 (37.9–0.919Unwell Appearance n (%)406 (67.3)141 (71.6)265 (65.3)0.18616 (76.2)124 (71.7)0.919Signs of Meningism n (%)11 (1.8)3 (1.5)8 (2.0)0.9990 (0.0)3 (1.7)0.999Rash n (%)120 (19.9)56 (28.4)64 (15.8)0.001*5 (23.8)51 (29.5)0.919GI Symptoms n (%)176 (29.2)46 (23.4)130 (32.0)0.0514 (19.0)42 (24.3)0.681Behaviour Change n (%)473 (78.4)163 (82.7)309 (76.1)0.11020 (95.2)141 (81.5)0.305Respiratory Symptoms, n (%)215 (35.7)42 (21.3)173 (42.6)< 0.001*3 (14.3)38 (22.0)0.862CSF WBC count (x 109/L) Median (IQR)n = 548 1.0 (0.0–4.95)n = 177 2.3 (0.3–38.0)n = 3711.0 (0.0–3.0)< 0.001*n = 19 0.3 (0.0–2.0)n = 155 3.4 (0.75–81.0)0.005*CSF pleocytosis n (%)76 (13.9)54 (30.5)22 (5.9)< 0.001*0 (0.0)54 (34.8)0.002*Peripheral WBC count (x 109/L) Median (IQR)n = 578 9.6 (6.8–12.9)n = 189 8.6 (6.6–11.4)n = 389 10.3 (7.1–13.9)< 0.001*n = 19 5.7 (4.8–8.1)n = 167 8.7 (6.7–11.5)0.002*Lymphocyte count (x 10*9/L) Median (IQR)n = 577 3.7 (2.4–5.4)n = 189 3.1 (2.0–4.7)n = 388 4.0 (2.5–5.7)0.001*n = 19 1.9 (1.2–3.6)n = 167 3.2 (2.1–4.7)0.005*Lymphopenic n (%)287 (49.7)113 (59.8)174 (44.8)0.012*15 (78.9)97 (58.1)0.219Neutrophil count (x 10*9/L) Median (IQR)n = 577 4.0 (2.5–6.6)n = 189 4.0 (2.8–5.9)n = 388 3.91 (2.4–7.1)0.999n = 19 2.8 (1.7–3.4)n = 167 4.19 (3.0–6.0)0.002*CRP (mg/L) Median (IQR)n = 595 10.0 (4.0–29.0)n = 195 10.0 (4.0–22.0)n = 400 10.9 (3.0–43.83)0.186n = 21 5.0 (3.0–8.5)n = 171 10.0 (5.0–22.0)0.005*LOS (days) Median (IQR)3 (2–4)3 (2–4)3 (2–4)0.1883 (3–4)3 (2–4)0.419IBI n (%)36 (6.0)1 (0.5)35 (8.6)< 0.001*0 (0.0)1 (0.6)0.999Continuous variables were compared using two-tailed Mann–Whitney U tests, and categorical variables were compared using two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant; FDR < 0.05, **Viruses in CSF: EV n = 173, PeV n = 21, Adenovirus n = 1, Human herpesvirus 6 n = 2. CRP C-reactive protein; CSF cerebrospinal fluid; EV Enterovirus; GI gastrointestinal; IBI invasive bacterial infection; IQR interquartile range; LOS length of stay; PeV Parechovirus; WBC white blood count. Patient characteristics and demographics Continuous variables were compared using two-tailed Mann–Whitney U tests, and categorical variables were compared using two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant; FDR < 0.05, **Viruses in CSF: EV n = 173, PeV n = 21, Adenovirus n = 1, Human herpesvirus 6 n = 2. CRP C-reactive protein; CSF cerebrospinal fluid; EV Enterovirus; GI gastrointestinal; IBI invasive bacterial infection; IQR interquartile range; LOS length of stay; PeV Parechovirus; WBC white blood count. Peripheral WBC counts were lower in infants with virus-positive CSF than in infants with virus-negative CSF (8.6 vs 10.3 × 109/L, P < 0.001), and lower in PeV than EV infections (5.7 vs 8.7 × 109/L, P = 0.002). The difference was attributable to reduced peripheral lymphocytes in virus-positive vs virus-negative groups (3.1 vs 4.0 × 10⁹/L, P < 0.001), with neutrophil counts comparable between these two groups (P = 0.999). Lymphopenia was more prevalent in virus-positive vs virus-negative infants (59.8% vs 44.8%, P = 0.012). CSF pleocytosis occurred in 54/155 (34.8%) EV-positive infants but was absent in all PeV-positive cases (0/19, 0%). Clinical and laboratory features grouped by pleocytosis status in EV-positive infants are detailed in Table 2. Infants without pleocytosis had reduced lymphocytes (2.7 vs 4.7 × 10⁹/L, P < 0.001), and a higher prevalence of lymphopenia (78.4% vs 24.1%, P < 0.001). No other significant clinical symptom differences were noted between groups.Table 2Patient characteristics of infants with EV detected in the CSF, with and without CSF pleocytosisEV without CSF pleocytosisEV with CSF pleocytosisAdjusted P valueTotal n (%)101 (65.2)54 (34.8)CSF WBC count (x 109/L) Median (IQR)1.0 (1.0–4.0)210 (72.8–403.0)Males n (%)60 (59.4)33 (61.1)0.999Age (days) Median (IQR)36 (25–56)38 (24–53)0.999Temperature (°C) Median (IQR)38.2 (37.9–38.5)38.2 (37.8–38.5)0.999Unwell Appearance n (%)73 (72.3)38 (70.4)0.999Signs of Meningism n (%)1 (1.0)2 (3.7)0.676Rash n (%)30 (29.7)14 (25.9)0.999GI Symptoms n (%)21 (20.8)17 (31.5)0.484Change in Behaviour n (%)86 (85.1)40 (74.1)0.439Respiratory symptoms n (%)24 (23.8)10 (18.5)0.999Peripheral WBC count (x 109/L) Median (IQR)n = 97 7.9 (6.0–10.2)n = 54 10.6 (8.2–12.2)<0.001*Lymphocyte count (x 10*9/L) Median (IQR)n = 97 2.7 (1.8–3.8)n = 54 4.7(3.6–6)<0.001*Lymphopenic n (%)76 (78.4)13 (24.1)<0.001*Neutrophil count (x 10*9/L) Median (IQR)n = 97 4.0 (2.7–6.2)n = 54 4.2 (3.1–5.5)0.999CRP (mg/L) Median (IQR)n = 101 15.3 (8.2–27.5)n = 53 5.0 (2.2–9.5)<0.001*Hospital admission n (%)101 (100.0)54 (100.0)0.999LOS, days Median (IQR)3 (2–4)3 (2–4)0.977PICU Admission n (%)0 (0.0)0 (0.0)0.999Continuous variables were compared using two-tailed Mann–Whitney U tests, and categorical variables were compared using two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant; FDR < 0.05. IQR interquartile range, GI gastrointestinal, CSF cerebrospinal fluid, WBC white blood cell, CRP C-reactive protein, LOS length of stay, EV Enterovirus. Table 2Patient characteristics of infants with EV detected in the CSF, with and without CSF pleocytosisEV without CSF pleocytosisEV with CSF pleocytosisAdjusted P valueTotal n (%)101 (65.2)54 (34.8)CSF WBC count (x 109/L) Median (IQR)1.0 (1.0–4.0)210 (72.8–403.0)Males n (%)60 (59.4)33 (61.1)0.999Age (days) Median (IQR)36 (25–56)38 (24–53)0.999Temperature (°C) Median (IQR)38.2 (37.9–38.5)38.2 (37.8–38.5)0.999Unwell Appearance n (%)73 (72.3)38 (70.4)0.999Signs of Meningism n (%)1 (1.0)2 (3.7)0.676Rash n (%)30 (29.7)14 (25.9)0.999GI Symptoms n (%)21 (20.8)17 (31.5)0.484Change in Behaviour n (%)86 (85.1)40 (74.1)0.439Respiratory symptoms n (%)24 (23.8)10 (18.5)0.999Peripheral WBC count (x 109/L) Median (IQR)n = 97 7.9 (6.0–10.2)n = 54 10.6 (8.2–12.2)<0.001*Lymphocyte count (x 10*9/L) Median (IQR)n = 97 2.7 (1.8–3.8)n = 54 4.7(3.6–6)<0.001*Lymphopenic n (%)76 (78.4)13 (24.1)<0.001*Neutrophil count (x 10*9/L) Median (IQR)n = 97 4.0 (2.7–6.2)n = 54 4.2 (3.1–5.5)0.999CRP (mg/L) Median (IQR)n = 101 15.3 (8.2–27.5)n = 53 5.0 (2.2–9.5)<0.001*Hospital admission n (%)101 (100.0)54 (100.0)0.999LOS, days Median (IQR)3 (2–4)3 (2–4)0.977PICU Admission n (%)0 (0.0)0 (0.0)0.999Continuous variables were compared using two-tailed Mann–Whitney U tests, and categorical variables were compared using two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant; FDR < 0.05. IQR interquartile range, GI gastrointestinal, CSF cerebrospinal fluid, WBC white blood cell, CRP C-reactive protein, LOS length of stay, EV Enterovirus. Patient characteristics of infants with EV detected in the CSF, with and without CSF pleocytosis Continuous variables were compared using two-tailed Mann–Whitney U tests, and categorical variables were compared using two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant; FDR < 0.05. IQR interquartile range, GI gastrointestinal, CSF cerebrospinal fluid, WBC white blood cell, CRP C-reactive protein, LOS length of stay, EV Enterovirus. Plasma response and CSF virus type Olink proteomic analysis was performed on 131 plasma samples to characterise the host response to viral CNS infection in infants (Supplementary Fig. 1). Demographic and clinical features of the Olink cohort were comparable to the full study population in terms of age (P = 0.277) and sex (P > 0.999) (Supplementary Table 1A). CSF WBC differences between groups also remained consistent between the Olink cohort and full study population, with higher counts in virus-positive vs virus-negative infants (P = 0.001) and in EV vs PeV cases (P = 0.025). Direct comparison between infants with and without a sample available for Olink analysis, within the clinical data cohort, also showed there were no significant differences between the proteomic and non-proteomic groups in key demographic or clinical characteristics, including: age, sex and clinical features at presentation (temperature, presence of rash, behavioural change, respiratory symptoms, unwell appearance) (Supplementary Table 1B). Differential expression analysis identified 32 differentially expressed proteins in the plasma of PeV-positive and 11 in the plasma of EV-positive CSF cases compared to virus-negative controls (Fig. 1A, B, Supplementary Tables 2, 3). Boxplots of significantly altered proteins are shown in Supplementary Fig. 2. Figure 1C shows hierarchical clustering of the significantly differentiated plasma proteins from both the PeV-positive vs virus-negative and EV-positive vs virus-negative analysis with a fold change >1.0. Samples from PeV-positive cases clustered tightly together and displayed a uniform upregulation of the included proteins. In contrast, EV-positive samples exhibited more heterogeneous clustering patterns, with variable expression across the same protein panel. This heterogeneity appeared to align with clinical parameters, particularly pleocytosis and lymphopenia, as shown by the annotation bars. EV cases with pleocytosis tended to cluster toward lower relative expression of the differentially abundant proteins, whereas those without pleocytosis showed relatively higher expression levels. No clear clustering was observed in relation to age, sex, or fever duration.Fig. 1Host plasma proteome response to EV and PeV CNS infection.A, B Volcano plots of differentially expressed proteins (FDR 0.05) between infant groups from the Olink cohort: A EV-positive CSF infants (n = 49) and viral negative CSF infants (n = 76), B PeV-positive CSF infants (n = 6) and viral negative CSF infants (n = 76). Statistical comparisons were performed using two-tailed Welch’s two-sample t-tests with Benjamini–Hochberg correction for multiple testing, as implemented in the OlinkAnalyze R package. Colour coding indicating fold change and statistical significance is defined within the figure key. C Hierarchical clustering displayed as a heatmap of the Olink cohort (n = 131) by significantly differentiated proteins (FDR < 0.05) with a fold change of > 1.0 between EV vs viral negative CSF and PeV vs viral negative CSF. Both proteins and samples were clustered using Euclidean distance and complete linkage. Coloured by protein expression values as defined within the figure key. All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, FDR false discovery rate. Host plasma proteome response to EV and PeV CNS infection. A, B Volcano plots of differentially expressed proteins (FDR 0.05) between infant groups from the Olink cohort: A EV-positive CSF infants (n = 49) and viral negative CSF infants (n = 76), B PeV-positive CSF infants (n = 6) and viral negative CSF infants (n = 76). Statistical comparisons were performed using two-tailed Welch’s two-sample t-tests with Benjamini–Hochberg correction for multiple testing, as implemented in the OlinkAnalyze R package. Colour coding indicating fold change and statistical significance is defined within the figure key. C Hierarchical clustering displayed as a heatmap of the Olink cohort (n = 131) by significantly differentiated proteins (FDR < 0.05) with a fold change of > 1.0 between EV vs viral negative CSF and PeV vs viral negative CSF. Both proteins and samples were clustered using Euclidean distance and complete linkage. Coloured by protein expression values as defined within the figure key. All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, FDR false discovery rate. Association between plasma proteomic changes, CSF pleocytosis and lymphopenia As plasma protein expression patterns appeared to relate to both CSF white cell and peripheral lymphocyte counts, and lymphopenia was the main variable distinguishing infants ≤90 days of age with and without pleocytosis, we next investigated how these parameters were associated with plasma protein expression. A correlation analysis of the plasma proteome with lymphocyte and CSF WBC counts was performed in infants with virus-positive CSF. A clear plasma response was associated with both increasing and decreasing lymphocyte counts (Fig. 2A, Supplementary Table 4). Scatter graphs of correlating proteins showed a linear relationship (Supplementary Fig. 3). A clear plasma response was associated with reduced CSF WBC counts, but not with increasing CSF WBC counts. Thirty plasma proteins were significantly negatively correlated with CSF WBC counts (Fig. 2B, Supplementary Table 5), while only three showed positive correlations. Scatter plots suggest that these positive correlations were driven by outliers rather than linear relationships (Supplementary Fig. 4).Fig. 2Plasma proteome response, CSF pleocytosis and peripheral lymphocyte count.A, B Volcano plot showing the correlation in CSF virus-positive samples between plasma protein expression and: A Peripheral lymphocyte count (n = 54), infants without data on peripheral lymphocyte count were excluded (n = 1), B CSF WBC count (n = 49), infants without data on CSF WBCs were excluded (n = 6). Statistical comparisons were performed using two-sided Pearson correlation tests with Benjamini–Hochberg correction for multiple testing. Colour coding indicating statistical significance is defined within the figure key. (C&D) Volcano plots of differentially expressed proteins (FDR 0.05) between: C EV-positive CSF infants with CSF pleocytosis (n = 21) and virus-negative CSF (n = 76), D EV-positive CSF infants without CSF pleocytosis (n = 23) and virus-negative CSF (n = 76). EV-positive infants without CSF WBCs were excluded (n = 5). Statistical comparisons were performed using two-tailed Welch’s two-sample t-tests with Benjamini–Hochberg correction for multiple testing, as implemented in the OlinkAnalyze R package. Colour coding indicating fold change and statistical significance is defined within the figure key. All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, WBC white blood cells, FDR false discovery rate. Plasma proteome response, CSF pleocytosis and peripheral lymphocyte count. A, B Volcano plot showing the correlation in CSF virus-positive samples between plasma protein expression and: A Peripheral lymphocyte count (n = 54), infants without data on peripheral lymphocyte count were excluded (n = 1), B CSF WBC count (n = 49), infants without data on CSF WBCs were excluded (n = 6). Statistical comparisons were performed using two-sided Pearson correlation tests with Benjamini–Hochberg correction for multiple testing. Colour coding indicating statistical significance is defined within the figure key. (C&D) Volcano plots of differentially expressed proteins (FDR 0.05) between: C EV-positive CSF infants with CSF pleocytosis (n = 21) and virus-negative CSF (n = 76), D EV-positive CSF infants without CSF pleocytosis (n = 23) and virus-negative CSF (n = 76). EV-positive infants without CSF WBCs were excluded (n = 5). Statistical comparisons were performed using two-tailed Welch’s two-sample t-tests with Benjamini–Hochberg correction for multiple testing, as implemented in the OlinkAnalyze R package. Colour coding indicating fold change and statistical significance is defined within the figure key. All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, WBC white blood cells, FDR false discovery rate. To further try to identify a unique plasma response associated with CSF pleocytosis, EV-positive CSF infants were grouped by the presence or absence of CSF pleocytosis and compared to the CSF virus-negative group. Differential expression analysis identified 35 proteins significantly altered in EV-positive infants without pleocytosis; there were no significantly differentially expressed proteins between EV with pleocytosis and virus-negative samples with FDR < 0.05 (Fig. 2C, D, Supplementary Table 6). There were, however, 19 proteins identified as significant using a non-adjusted p-value < 0.01 (Supplementary Fig. 5, Table 7). Of these 19 proteins, 8 overlapped with the non-pleocytosis differential expression analysis or lymphocyte correlation analysis. This overlap suggests that biologically relevant signals may be present among proteins not retained after FDR correction, although interpretation of these nominal associations should remain cautious, given the increased risk of false discoveries. Clustering of plasma response and enrichment analysis Given the apparent influence of CSF and peripheral immune cell responses on plasma protein expression, we next assessed how these parameters shaped plasma proteomic clustering and explored the functional enrichment of the associated protein signatures. A hierarchically clustered heatmap was used to display the expression of the 39 plasma proteins significantly correlated with either CSF WBC or peripheral lymphocyte count (FDR < 0.01) (Fig. 3A). The more stringent FDR threshold of <0.01 was used to restrict clustering to the most robust correlations and reduce potential noise in this analysis. The heatmap, generated with the dendrogram cut at k = 3, showed a clear separation of the samples into three distinct clusters of infants: Cluster 1, 2 and 3. All infants (17/17) in Cluster 1 had pleocytosis, in contrast to 10.6% (2/19) and 15.4% (2/13) in Clusters 2 and 3, respectively (Table 3). Cluster 2 infants were significantly younger than those in Clusters 1 and 3 (27.0 vs 39.0 and 57.5 days (P = 0.013). They also exhibited markedly lower peripheral lymphocyte counts, 1.7 × 10⁹/L, than in Clusters 1 (4.5 × 10⁹/L) and 3 (3.5 × 10⁹/L) (P = 0.004). No significant differences in sex, presenting symptoms, or hospital stay were observed among clusters.Fig. 3Plasma proteome clusters and functional analysis of EV/PeV CNS infection.A Hierarchical clustering displayed as a heatmap of all infants with EV or PeV positive CSF from the Olink cohort (n = 55) by proteins significantly correlated with CSF WBC count or peripheral lymphocyte count (FDR < 0.05). Proteins and samples were clustered using Euclidean distance and complete linkage, and the sample dendrogram was partitioned into three clusters. Coloured by protein expression values as defined within the figure key. The hierarchical clustering results are descriptive only and were not subjected to formal stability or resampling analyses. B–D Bar charts of enriched terms from proteins upregulated in heatmap clusters 1 (B), 2 (C), and 3 (D), generated using Metascape. Enrichment analysis was performed using a hypergeometric test with Benjamini–Hochberg correction for multiple comparisons. Enriched terms were clustered based on Kappa similarity, and the most significant term from each cluster is shown. Bars are coloured according to enrichment p-values (LogP). All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, WBC white blood cells, FDR false discovery rate.Table 3Patient characteristics of infants with virus-positive CSF in the three clusters defined by protein expression levels, as per Fig. 3BCluster 1Cluster 2Cluster 3Adjusted P valueMultiple comparisonsTotal n (%)21 (38.2)20 (36.4)14 (25.5)Males n (%)16 (76.2)11 (55)9 (64.3)0.580Age (days) Median (IQR)39.0 (29.0–63)27.0 (19.5–35.3)57.5 (25.0–70.0)0.013*1 vs 2 0.028* 1 vs 3 > 0.999 2 vs 3 0.009*Temperature (°C) Median (IQR)38.2 (37.8–38.5)38.2 (38.0–38.6)38.0 (37.5–38.6)0.702Unwell Appearance n (%)16 (76.2)17 (85.0)10 (71.4)0.702Rash n (%)8 (38.1)6 (30.0)5 (35.7)0.936GI Symptoms n (%)8 (38.1)4 (20.0)2 (14.3)0.530Change in Behaviour n (%)19 (90.5)16 (80.0)11 (78.6)0.702Respiratory symptoms n (%)5 (23.8)4 (20.0)5 (35.7)0.702Fever Duration n (%) 1: <6 h 2: 6–12 h 3: 12–24 h1: 15/19 (78.9) 2: 3/19 (15.8) 3: 1/19 (5.3)1: 13/19 (68.4) 2: 6/19 (31.6) 3: 0/19 (0.0)1: 9/13 (69.2) 2: 2/13 (15.4) 3: 2/13 (15.4)0.591CSF WBC count (x 109/L) Median (IQR)n = 17 257.5 (40.3–531)n = 19 1.0 (0.0–4.1)n = 13 1.0 (0.5–6.4)<0.001*1 vs 2 < 0.001* 1 vs 3 < 0.001* 2 vs 3 > 0.999Peripheral WBC count (x 109/L) Median (IQR)n = 20 10.3 (8.2–13.6)n = 20 6.0 (4.8–7.1)n = 14 8.3 (6.2–9.9)<0.001*1 vs 2 < 0.001* 1 vs 3 0.243 2 vs 3 0.052Lymphocyte count (x 10*9/L) Median (IQR)n = 20 4.5 (3.6–5.3)n = 20 1.7 (1.4–2.0)n = 14 3.5 (2.8–4.3)<0.001*1 vs 2 < 0.001* 1 vs 3 0.397 2 vs 3 0.001*Lymphopenic n (%)6/20 (30)19/20 (95)8/14 (57.1)<0.001*1 vs 2 < 0.001* 1 vs 3 0.163 2 vs 3 0.012*Neutrophil count (x 10*9/L) Median (IQR)n = 20 4.9 (3.2–6.3)n = 20 3.4 (2.8–4.0)n = 14 3.7 (3.0–4.9)0.297LOS, days Median (IQR)3 (2–4)3 (2–4)2(1–4)0.354Continuous variables were compared using the Kruskal–Wallis test, and if significant, Dunn’s multiple comparisons test. Categorical variables were compared using two-sided Fisher’s exact tests, and if significant, post-hoc pairwise two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant. IQR interquartile range, GI gastrointestinal, CSF cerebrospinal fluid, WBC white blood cell, LOS length of stay. Plasma proteome clusters and functional analysis of EV/PeV CNS infection. A Hierarchical clustering displayed as a heatmap of all infants with EV or PeV positive CSF from the Olink cohort (n = 55) by proteins significantly correlated with CSF WBC count or peripheral lymphocyte count (FDR < 0.05). Proteins and samples were clustered using Euclidean distance and complete linkage, and the sample dendrogram was partitioned into three clusters. Coloured by protein expression values as defined within the figure key. The hierarchical clustering results are descriptive only and were not subjected to formal stability or resampling analyses. B–D Bar charts of enriched terms from proteins upregulated in heatmap clusters 1 (B), 2 (C), and 3 (D), generated using Metascape. Enrichment analysis was performed using a hypergeometric test with Benjamini–Hochberg correction for multiple comparisons. Enriched terms were clustered based on Kappa similarity, and the most significant term from each cluster is shown. Bars are coloured according to enrichment p-values (LogP). All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, WBC white blood cells, FDR false discovery rate. Table 3Patient characteristics of infants with virus-positive CSF in the three clusters defined by protein expression levels, as per Fig. 3BCluster 1Cluster 2Cluster 3Adjusted P valueMultiple comparisonsTotal n (%)21 (38.2)20 (36.4)14 (25.5)Males n (%)16 (76.2)11 (55)9 (64.3)0.580Age (days) Median (IQR)39.0 (29.0–63)27.0 (19.5–35.3)57.5 (25.0–70.0)0.013*1 vs 2 0.028* 1 vs 3 > 0.999 2 vs 3 0.009*Temperature (°C) Median (IQR)38.2 (37.8–38.5)38.2 (38.0–38.6)38.0 (37.5–38.6)0.702Unwell Appearance n (%)16 (76.2)17 (85.0)10 (71.4)0.702Rash n (%)8 (38.1)6 (30.0)5 (35.7)0.936GI Symptoms n (%)8 (38.1)4 (20.0)2 (14.3)0.530Change in Behaviour n (%)19 (90.5)16 (80.0)11 (78.6)0.702Respiratory symptoms n (%)5 (23.8)4 (20.0)5 (35.7)0.702Fever Duration n (%) 1: <6 h 2: 6–12 h 3: 12–24 h1: 15/19 (78.9) 2: 3/19 (15.8) 3: 1/19 (5.3)1: 13/19 (68.4) 2: 6/19 (31.6) 3: 0/19 (0.0)1: 9/13 (69.2) 2: 2/13 (15.4) 3: 2/13 (15.4)0.591CSF WBC count (x 109/L) Median (IQR)n = 17 257.5 (40.3–531)n = 19 1.0 (0.0–4.1)n = 13 1.0 (0.5–6.4)<0.001*1 vs 2 < 0.001* 1 vs 3 < 0.001* 2 vs 3 > 0.999Peripheral WBC count (x 109/L) Median (IQR)n = 20 10.3 (8.2–13.6)n = 20 6.0 (4.8–7.1)n = 14 8.3 (6.2–9.9)<0.001*1 vs 2 < 0.001* 1 vs 3 0.243 2 vs 3 0.052Lymphocyte count (x 10*9/L) Median (IQR)n = 20 4.5 (3.6–5.3)n = 20 1.7 (1.4–2.0)n = 14 3.5 (2.8–4.3)<0.001*1 vs 2 < 0.001* 1 vs 3 0.397 2 vs 3 0.001*Lymphopenic n (%)6/20 (30)19/20 (95)8/14 (57.1)<0.001*1 vs 2 < 0.001* 1 vs 3 0.163 2 vs 3 0.012*Neutrophil count (x 10*9/L) Median (IQR)n = 20 4.9 (3.2–6.3)n = 20 3.4 (2.8–4.0)n = 14 3.7 (3.0–4.9)0.297LOS, days Median (IQR)3 (2–4)3 (2–4)2(1–4)0.354Continuous variables were compared using the Kruskal–Wallis test, and if significant, Dunn’s multiple comparisons test. Categorical variables were compared using two-sided Fisher’s exact tests, and if significant, post-hoc pairwise two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant. IQR interquartile range, GI gastrointestinal, CSF cerebrospinal fluid, WBC white blood cell, LOS length of stay. Patient characteristics of infants with virus-positive CSF in the three clusters defined by protein expression levels, as per Fig. 3B Continuous variables were compared using the Kruskal–Wallis test, and if significant, Dunn’s multiple comparisons test. Categorical variables were compared using two-sided Fisher’s exact tests, and if significant, post-hoc pairwise two-sided Fisher’s exact tests. P-values were adjusted by applying the Benjamini–Hochberg false discovery rate procedure to the set of comparisons within the table. *Statistically significant. IQR interquartile range, GI gastrointestinal, CSF cerebrospinal fluid, WBC white blood cell, LOS length of stay. Functional enrichment analysis of cluster-specific upregulated plasma proteins (Supplementary Table 8) revealed distinct biological signatures. Cluster 2, despite minimal pleocytosis, was enriched for pro-inflammatory and antiviral pathways, as well as CNS-associated processes including “regulation of neuron apoptotic process” and “positive regulation of microglial cell migration” (Fig. 3C). It also showed enrichment for “positive regulation of apoptotic processes”. Proteins upregulated in Cluster 1, the only group with consistent CSF pleocytosis, showed only four enriched terms, with “adaptive immune response” being the most significant (Fig. 3B). Notably, no CNS-related or apoptotic pathways were enriched in this cluster. Enriched terms in Cluster 3 contained a combination of enriched terms from both Cluster 1 and 2 (Fig. 3D), with the “regulation of NMDA receptor activity” cluster representing terms relating to neuron apoptotic processes. Therefore, CNS-related and apoptotic signalling was observed only in Clusters 2 and 3, which both lacked significant pleocytosis. Fever duration and host response While no association was observed between fever duration and the host proteomic response within the Olink cohort (Fig. 1D), to further investigate a possible association, differential expression analysis was carried out on infants with <6 h and >6 h from fever onset. No differentially expressed proteins were identified between groups at FDR < 0.05 (Fig. 4A), and only two proteins reached significance (p < 0.01) with an unadjusted p-value. In the full clinical data cohort, CSF WBC counts did not differ significantly across fever duration categories (<6 h, 6–12 h, 12–24 h, >24 h; Fig. 4B). Peripheral lymphocyte counts also showed no significant variation by fever duration (Fig. 4C).Fig. 4The effect of fever duration of plasma proteome response and WBC counts.A Volcano plot of differentially expressed proteins (FDR 0.05) between virus-positive CSF infants in the Olink cohort, presenting <6 h (n = 37) vs >6 h (n = 14) from first recorded fever. Infants without information on fever duration were excluded (n = 4). Statistical comparisons were performed using two-tailed Welch’s two-sample t-tests with Benjamini–Hochberg correction for multiple testing, as implemented in the OlinkAnalyze R package. Colour coding indicating statistical significance is defined within the figure key. B Box plot showing CSF WBC count by fever duration category for all participants in clinical data cohort with EV or PeV positive CSF (n = 166; <6 h n = 102, 6–12 h n = 31, 12–24 h n = 26, >24 h n = 7). Infants without data on fever duration or CSF WBC were excluded (n = 28). Lines represent median and interquartile ranges. Zero CSF WBC values were set to 0.1 for visualisation on the log-scale axis only. Statistical comparisons were performed using the Kruskal–Wallis test. C Box plot showing peripheral lymphocyte count by fever duration category for all participants with EV or PeV positive CSF (n = 175; <6 h n = 104, 6–12 h n = 35, 12–24 h n = 28, >24 h n = 8). Infants without data on fever duration or peripheral lymphocyte count were excluded (n = 19). Lines represent median and interquartile ranges. Statistical comparisons were performed using the Kruskal–Wallis test. All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, WBC white blood cells, FDR false discovery rate. The effect of fever duration of plasma proteome response and WBC counts. A Volcano plot of differentially expressed proteins (FDR 0.05) between virus-positive CSF infants in the Olink cohort, presenting <6 h (n = 37) vs >6 h (n = 14) from first recorded fever. Infants without information on fever duration were excluded (n = 4). Statistical comparisons were performed using two-tailed Welch’s two-sample t-tests with Benjamini–Hochberg correction for multiple testing, as implemented in the OlinkAnalyze R package. Colour coding indicating statistical significance is defined within the figure key. B Box plot showing CSF WBC count by fever duration category for all participants in clinical data cohort with EV or PeV positive CSF (n = 166; <6 h n = 102, 6–12 h n = 31, 12–24 h n = 26, >24 h n = 7). Infants without data on fever duration or CSF WBC were excluded (n = 28). Lines represent median and interquartile ranges. Zero CSF WBC values were set to 0.1 for visualisation on the log-scale axis only. Statistical comparisons were performed using the Kruskal–Wallis test. C Box plot showing peripheral lymphocyte count by fever duration category for all participants with EV or PeV positive CSF (n = 175; <6 h n = 104, 6–12 h n = 35, 12–24 h n = 28, >24 h n = 8). Infants without data on fever duration or peripheral lymphocyte count were excluded (n = 19). Lines represent median and interquartile ranges. Statistical comparisons were performed using the Kruskal–Wallis test. All data points are from distinct participants. EV enterovirus, PeV parechovirus, CSF cerebrospinal fluid, WBC white blood cells, FDR false discovery rate. Discussion This study constitutes a comprehensive characterisation of the host immune response and clinical characteristics of EV and PeV CNS infection in infants ≤90 days of age. We have identified distinct immune response patterns in infants without CSF pleocytosis, associated with lymphopenia and unrelated to the time of disease onset. Notably, infants lacking CSF pleocytosis exhibited a pronounced plasma proinflammatory cytokine response, with plasma protein signatures overlapping pathways implicated in apoptotic processes. These findings suggest that the absence of pleocytosis does not necessarily indicate a lack of CNS-directed immune activation but may instead reflect a distinct systemic immune response profile. In our prospective multicentre cohort of febrile infants ≤90 days of age, we found a viral CNS infection prevalence of 32.7%, demonstrating its significant burden. Consistent with previous studies, we highlight the difficulty in distinguishing viral CNS infection from other febrile illnesses based on clinical features in infants. Although infants with viral CNS infection were statistically more likely to present with a rash, its diagnostic utility was limited as only 28.4% exhibited this feature. Classic signs of meningism were rare, indicating their limited diagnostic value in this population. CSF pleocytosis was not observed in infants with PeV CNS infection and in only 34% of those with EV CNS infection, consistent with previous reports7. Infants without CSF pleocytosis exhibited significantly lower peripheral WBC counts, as supported by earlier studies, which was driven by a reduction in lymphocytes6,9,12,15. Lymphopenia was present in 79% (15/19) of PeV-infected infants, compared to 58.1% (97/167) of those with EV, indicating, along with the universal absence of CSF pleocytosis, a more consistent immunological phenotype in PeV infection. Proteomic profiling of over 700 plasma proteins associated with inflammation and neurology mirrored these findings. PeV infection showed a uniform plasma inflammatory signature, whereas EV responses were heterogeneous and closely linked to pleocytosis and lymphopenia. Infants with EV CNS infection but without pleocytosis exhibited plasma profiles more similar to PeV infection than EV with pleocytosis, indicating a shared mechanism underlying absent CSF pleocytosis. Given the variation in plasma host responses relative to CSF pleocytosis and lymphopenia, these relationships were further examined. Infants with CSF pleocytosis exhibited a detectable but subdued host response, with significant protein changes emerging only after removal of FDR correction. In contrast, infants without pleocytosis showed a strong systemic cytokine activation. Clustering of plasma proteins by CSF and peripheral lymphocyte counts identified three distinct groups: clusters 1 (n = 17), 2 (n = 19) and 3 (n = 13). Clusters 2 and 3, both characterised by minimal pleocytosis, demonstrated robust upregulation of plasma cytokines and chemokines, with enrichment of terms related to neuroinflammatory and apoptotic processes. Their plasma profiles, dominated by IFNγ-inducible cytokines (CXCL9, CXCL10, CXCL11), IL-6, TNF, and IL-15, resembled a cytokine storm, an exaggerated inflammatory state linked to lymphocyte depletion and immune dysregulation after viral infection19. This cytokine-driven lymphopenia may limit immune cell trafficking into the CNS, explaining the absence of pleocytosis. Although enrichment terms related to neuroinflammatory and apoptotic processes were identified in these two clusters, it is worth noting that these annotations are derived from plasma protein homology and may reflect systemic pathways with overlap to CNS immune signalling rather than direct evidence of central processes. The enrichment of apoptosis-associated proteins nonetheless supports the possibility that lymphocyte apoptosis contributes to systemic lymphopenia in this context. In terms of individual proteins, apoptosis-related signalling was supported by upregulation of TNF and TNFSF10, both members of the tumour necrosis factor superfamily with established roles in death receptor–mediated apoptotic pathways, together with increased TNFRSF6B, a soluble regulator of TNF-family death ligands that modulates apoptosis signalling20. Increased CD274, an immune checkpoint ligand associated with T-cell inhibition and, in certain contexts, activation-induced T-cell apoptosis, further supports a milieu consistent with lymphocyte depletion21. The coordinated elevation of these proteins provides biological support for enrichment of apoptosis-associated pathways in these clusters. Clusters 2 and 3 differed in age distribution and immune pathway enrichment, suggesting that immune maturation influences the variability in host responses. Cluster 3 uniquely expressed lymphocyte-associated proteins (CD6, SH2D1A, SIT1, TCL1A), and plasma pathway enrichment for adaptive singling, indicating stronger adaptive immune engagement, whereas Cluster 2 only reflected the predominantly cytokine-driven response. This aligns with the younger age and more pronounced lymphopenia observed in Cluster 2. Cluster 1, which included most infants with CSF pleocytosis, also showed enrichment for adaptive immune signalling in the plasma but lacked the cytokine-dominated inflammatory profile seen in Clusters 2 and 3. EV and PeV CNS infections are generally associated with favourable outcomes in infants, however evidence on more subtle long-term outcomes is mixed5–8. Studies of EV-A71, which can cause severe neuronal complications, show that cytokine responses correlate with disease severity, indicating that the host’s inflammatory response magnitude plays a key role in disease severity22. Indeed, other severe viral infections (e.g., COVID-19) are characterised by marked depletion of lymphocytes in the blood, correlated with cytokine dysregulation23,24. While this study found no indications of increased disease severity within the different clusters, we were limited to reporting length of stay as an outcome, with no long-term follow-up. Given the evidence of increased neurological disease severity with “cytokine storms” in EV-71 and other viral infections, and the increase in enriched functional terms relating to neuron apoptosis in our clusters of infants, it could be hypothesised that infants who lack CSF pleocytosis are at increased risk of worse outcomes19,22. Indeed, Kadambari et al. reported six infants with death or sequelae following EV CNS infection, over half of whom lacked CSF pleocytosis7. Future outcome studies should consider these distinct disease patterns. While we observed marked systemic inflammatory activation in plasma, this does not necessarily indicate an equivalent response in the CSF, as immune signalling is compartmentalised between blood and CSF. While plasma responses in viral CNS infection had not been previously extensively evaluated, CSF cytokine profiles have been reported and are generally muted in PeV, and raised when pleocytosis is present25–27. Low chemoattractant cytokine concentrations in PeV CNS infection or EV without pleocytosis could be viewed as low levels of CNS inflammation, insufficient to recruit WBCs. Indeed, Harvala et al.18 proposed that many EV/PeV detections in CSF represent systemic infections with viral RNA leakage rather than true CNS infection18. They reported higher CSF-to-plasma viral load ratios in infants without pleocytosis, which were driven mainly by plasma viral load differences. While viral load measurements were not available in this study, which represents a limitation, we instead suggest that an aberrant immune response may fail to control plasma viral load, while muted CSF cytokine responses may reflect the contribution of WBCs to cytokine production in the CSF. Importantly, CSF cytokine responses are reduced but not absent in cases with an absence of pleocytosis and are still higher than in virus-negative controls25–27. We conclude that RNA detection in infants without pleocytosis is unlikely to be incidental and that reliance on pleocytosis as a diagnostic criterion for viral meningitis may overlook true CNS infection. Taken together, these findings support a host-response model in which EV/PeV infection triggers a pronounced systemic interferon and cytokine response that promotes lymphocyte depletion, potentially through apoptosis (Supplementary Fig. 6). Reduced circulating lymphocyte numbers may limit leucocyte trafficking across the blood–brain barrier, resulting in absent cerebrospinal fluid pleocytosis despite ongoing immune activation. Diminished leucocyte entry into the CNS may also contribute to the lower CSF cytokine concentrations reported in other studies, given the role of infiltrating immune cells in local cytokine production. Virus-associated lymphopenia is a well-described feature of acute viral infection and has been linked to cytokine-driven depletion and adverse outcomes22–24,28. Interferon and TNF-family signalling can activate death-receptor pathways, including TRAIL-mediated cascades, promoting apoptotic lymphocyte loss. These mechanisms are well described in COVID-19, where T-cell counts inversely correlate with IL-6 and TNF-α levels and increased PD-1 (CD274) expression, alongside associations between CXCL10 and T-cell apoptosis29,30. In our cohort, coordinated upregulation of interferon-inducible chemokines, TNF-family components and CD274 aligns with this framework and provides biological plausibility for apoptosis-driven lymphopenia limiting leucocyte recruitment into the CSF. We acknowledge that this mechanism is inferred rather than directly demonstrated in this study. Another commonly proposed mechanism for the absence of CSF pleocytosis in infants is early presentation, before WBC migration into the CSF has occurred6,10,14. While some studies support this association, they often include older children, introducing potential age-related bias13,14. In our cohort, although fever duration was parent-reported and subject to recall bias, it was prospectively collected and clearly showed no association with CSF pleocytosis. While fever duration should be interpreted as an approximate marker of infection timing, the exact timing of viral acquisition or central nervous system invasion cannot be determined in observational human studies. Fever onset, although imperfect and potentially variable in young infants, represents the most consistently documented clinical feature and therefore serves as a pragmatic measure in prospective febrile infant cohorts31. We acknowledge that truly defining the role of time in the development of CSF pleocytosis would require longitudinal sampling with repeated clinical and immunological assessments. Importantly, however, the febrile infant presentation window is typically narrow, with most infants undergoing lumbar puncture within a relatively short timeframe after fever onset. Within this clinically relevant window, we observed neither a statistical association nor a consistent directional trend between fever duration and CSF pleocytosis that would explain the high proportion of infants without pleocytosis. These data therefore argue against the hypothesis that the absence of pleocytosis simply reflects earlier presentation or earlier lumbar puncture. A key limitation of this study is its exploratory design. While we comprehensively characterised systemic immune and neuroinflammatory responses at presentation, no prespecified primary hypothesis or long-term outcomes were defined. Findings should therefore be interpreted as hypothesis-generating, and future studies with longitudinal follow-up are warranted to determine the clinical significance of these distinct immune response profiles. The hierarchical clustering results are descriptive only, as they were derived from correlation-selected proteins and were not subjected to formal stability or resampling analyses. Further limitations include the limited proteomic cohort and small sample sizes in some groups. While the proteomic cohort was representative of the overall study population, inclusion depended on sample availability, introducing potential selection bias. The number of PeV cases was smaller than EV cases, reflecting the natural prevalence of infection within the cohort, which may have limited statistical power for virus-specific subgroup analyses. However, the inflammatory phenotype observed in PeV infection was consistent and biologically coherent across analyses, supporting the robustness of this signal despite the smaller sample size. There is robust evidence that EV/PeV detection enables antibiotic discontinuation and decreases hospital stays, given the low IBI prevalence in viral CNS infection infants1,2,4. Our findings support this, with only 1/197 (0.5%) case of IBI in infants with virus-positive CSF. Blood PCR testing has been recommended to further identify this low-risk group, yet only 14% of febrile infants in this study underwent viral blood PCR, highlighting its limited use in UK clinical practice1,2,4. The absence of routine EV/PeV blood PCR and lack of EV genotyping information are further limitations, as these prevent assessment of whether the EV immune heterogeneity reflects viral genotype differences or host-specific responses. Given the wide genetic diversity of EVs and known genotype-dependent differences in CSF/plasma viral loads, future work should also investigate the host response in patients with detected viremia along with the effects of EV genotype14,17,18,32. Viral CNS infection is a common cause of febrile illness in infants, yet its pathogenesis, particularly in cases without pleocytosis or signs of meningism, remains unclear. Our findings suggest that absent CSF pleocytosis in EV and PeV CNS infection is not due to early presentation or incidental viremia, but reflects a distinct immune phenotype marked by plasma cytokine activation and lymphocyte depletion. This response may limit leucocyte migration into the CSF despite systemic inflammation. These results challenge reliance on pleocytosis as a marker of CNS involvement, and we support a clinical definition for viral CNS infection whereby the detection of EV or PeV in the CSF truly identifies a viral neurological infection, regardless of CSF pleocytosis. Future studies on outcomes to inform follow-up selection in infants with viral CNS infection should explore these distinct immune responses. Methods Ethical approval This study was conducted in accordance with all relevant ethical regulations and the principles of the Declaration of Helsinki. Ethical approval for the FIDO study was obtained from the Office for Research Ethics Committees Northern Ireland Health and Social Care Research Ethics Committee B (Ref22:/NI/0002), the Public Benefit and Privacy Panel for Health and Social Care Scotland (Ref: 2122-0257), and the Children’s Health Ireland Research Ethics Committee (Ref: REC-082-22). Study design Study participants for this analysis are from the Febrile Infant Diagnostic Assessment and Outcome (FIDO) Study. The FIDO study was a multicentre, prospective, observational cohort study that enroled febrile infants at 35 paediatric emergency departments (EDs) between July 2022 and August 2023, within the UK and Ireland. The study’s protocol, methodology and recruitment process are summarised below and have also previously been reported, and adhere to the STROBE guidelines33,34. Infants were eligible for the FIDO study if they were ≤90 days of age and presented with a fever ≥38 °C at the ED or if a documented fever ≥38 °C was reported within 24 h of presentation, regardless of thermometer type. Exclusion criteria were limited to parental refusal/withdrawal of consent. Sex was recorded for all participants based on clinical records at enrolment. Gender identity data were not collected due to the age of the participants. Participant sex distributions are reported where applicable. A study overview and analysis workflow are shown in Supplementary Fig. 1. The FIDO study was registered at https://www.clinicaltrials.gov (trial registration: NCT05259683) on the 28th February 2022. Participants and their families did not receive financial compensation for participation in this study. Participants FIDO participants were included in this study if they underwent a lumbar puncture and complete CSF testing (bacterial culture or PCR and viral PCR) at the local hospital site as part of routine care (Supplementary Fig. 1). Participants were included in the Olink cohort if they had a sufficient volume of plasma sample available. All participants parents/guardians gave written informed consent for the storage and use of plasma samples as per the study consent procedures33,34. Participants with evidence of bacterial infection (IBI or urinary tract infection) were excluded from the Olink cohort as they would have contributed significantly to plasma proteome responses in the virus-negative CSF group (Supplementary Fig. 1). IBI and urinary tract infection were defined as per the FIDO study protocol34. Procedures Study data collection methods and Case Report Forms (CRFs) for the FIDO study have been described previously33,34. An initial CRF was prospectively completed to record data regarding clinical features at the first clinician assessment by the treating clinicians. Signs of meningism were defined as any of the following: bulging fontanelle, neck stiffness, seizures or focal neurological signs. Unwell appearance was defined as abnormal global assessment and/or abnormal vital signs (based on APLS recommendations for age). Fever duration was recorded as the number of hours from the first fever in categories of 0–6, 6–12, 12–24 and >24 h. A follow-up CRF was completed seven days after discharge to record laboratory results and length of stay. All laboratory testing (culture, PCR, WBC counts, and CRP measurements) was conducted at local hospital laboratories using commercially available, routinely used tests on samples collected at initial presentation, as part of routine care. CSF pleocytosis was defined as >15 WBCs per mm3 in the CSF9,35. If > 500/mm3 red blood cells were recorded in the CSF, a ratio of 500:1 white to red blood cells was applied. Lymphopenia was defined as a peripheral lymphocyte count lower than age-based reference ranges (0–28 days; 2.8–9.1 × 109/l, 29–90 days; 4.0–10.0 × 109/l). EDTA plasma samples were collected during routine phlebotomy on initial presentation to the ED and stored at −80 °C, where possible. Plasma concentrations of 735 proteins were measured using the Olink® platform (Uppsala, Sweden) “Explore Neurology” and “Inflammation” antibody-based Proximity Extension Assays. The terms “Inflammation” and “Neurology” refer to manufacturer-defined Olink panel designations based on thematic assay grouping and do not represent discrete or mutually exclusive functional classifications. Analysis was performed by an Olink Certified Service Provider (Randox Laboratories, Antrim, UK). Data were provided normalised and log2 transformed before analysis. Proteins were removed if they failed manufacturer-defined Olink assay quality control (8 proteins), and duplicate proteins that were measured on both panels were removed from the Neurology panel (3 proteins), 724 proteins were included for final analysis. Statistical analysis of clinical data Categorical clinical variables were reported as frequencies and percentages, with differences among groups analysed by Fisher’s exact test. Continuous variables were reported as mean and standard deviation (SD) for normally distributed data (based on the Shapiro–Wilk normality test), or median and interquartile ranges (IQR) for non-normally distributed data. Differences among groups were analysed using a two-tailed Student’s t-test or two-tailed Mann–Whitney U-test, depending on the normality of the data. Significance was based on a false discovery rate (FDR) using Benjamini–Hochberg correction of less than 0.05 to control the expected proportion of false discoveries among statistically significant results; an adjusted value < 0.05 corresponds to an expected false discovery proportion of 5%. Missing data for CSF WBC counts, peripheral blood cell counts and fever onset time were handled using a complete-case approach for the relevant analysis only, with no imputation applied. The number of samples excluded due to missing data is indicated in the corresponding figure legends. Analysis was performed using GraphPad Prism®version 10.4.1 for Windows (GraphPad Software, Boston, MA, USA). Statistical analysis of Olink data Olink data analysis was conducted in R (v4.3.3). Differentially expressed proteins between infection groups were identified using the R package “Olink® Analyze”. Significance was based on an FDR using Benjamini–Hochberg correction of less than 0.05. Pearson correlation coefficients between proteins and indicated cell counts were calculated, and two-sided Pearson correlation tests were used to assess statistical significance, using the R function “cor”. p-values from correlation analysis were adjusted based on an FDR (0.05) using the “p.adjust” function (Benjamini–Hochberg correction). To generate hierarchically clustered heatmaps, protein expression values were scaled per protein using the R function scale. Heatmaps were generated using the ComplexHeatmap package in R, with hierarchical clustering performed using Euclidean distance and complete linkage. Corresponding clinical variables were displayed as annotations alongside the heatmaps. For functional enrichment analysis, differentially expressed proteins were input into the Metascape Gene Annotation and Analysis Resource (http://metascape.org) to query Kyoto Encyclopaedia of Genes and Genomes (KEGG), Gene Ontology (GO) and Reactome Gene Sets resources. The following criteria were used for enrichment analysis: minimum count of 3, enrichment factor of >1.5, p < 0.01. Terms significantly enriched in the indicated protein sets were automatically clustered by Metascape into nonredundant groups and presented as bar charts36. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Supplementary Information Reporting Summary Transparent Peer Review file Source data Source data Source data