Identifying immune signatures of sepsis to increase diagnostic accuracy in very preterm babies

Abstract

Bacterial infections are a major cause of mortality in preterm babies, yet our understanding of early-life disease-associated immune dysregulation remains limited. Here, we combine multi-parameter flow cytometry, single-cell RNA sequencing and plasma analysis to longitudinally profile blood from very preterm babies (<32 weeks gestation) across episodes of invasive bacterial infection (sepsis). We identify a dynamically changing blood immune signature of sepsis, including lymphopenia, reduced dendritic cell frequencies and myeloid cell HLA-DR expression, which characterizes sepsis even when the common clinical marker of inflammation, C-reactive protein, is not elevated. Furthermore, single-cell RNA sequencing identifies upregulation of amphiregulin in leukocyte populations during sepsis, which we validate as a plasma analyte that correlates with clinical signs of disease, even when C-reactive protein is normal. This study provides insights into immune pathways associated with early-life sepsis and identifies immune analytes as potential diagnostic adjuncts to standard tests to guide targeted antibiotic prescribing.

Fig. 1. An immune signature of sepsis in very preterm babies.

a Overview of study design. b Each individual blood sample from a baby was assigned to one of five groups, dependent on contemporaneous clinical and microbiological metadata. Figure created with BioRender.com. c Timepoints for blood collection from each baby. Repeated samples from the same sepsis episode (marked with an X), or those collected during NEC (grey circles) without positive microbiological tests were excluded from analysis in Fig. 1d. d Volcano plot of 105 immune parameters comparing ‘Sepsis’ (n = 32 samples from 19 babies) versus ‘No-Sepsis’ (n = 104 samples from 17 babies). Statistical analysis was performed using the two-tailed Mann–Whitney test. Statistically significant parameters are marked in red (Log2fold change >0.6 or <-0.6 and Benjamini–Hochberg (BH) corrected p-value < 0.01). e Temporal changes in select immune parameters from (d), including frequencies of pDC, mDC and T cells; DC and classical monocyte median HLA-DR and numbers of CD4⁺ T cells, CD8⁺ T cells and mDC. Lines connect sepsis samples to those obtained approximately one week before (Pre) or one week after (Post) sepsis, from the same baby. Line colours denote which baby the sepsis episode belongs to; babies may have had more than one episode of sepsis (n = 32 episodes of sepsis from 19 babies). Two-sided Wilcoxon matched-pairs signed-rank test (Pre vs Sepsis and Sepsis vs Post) was used for (e). N may vary slightly between graphs due to experimental dropouts or data filtering (see “Methods”). Source data are provided as a Source Data file.

Fig. 2. Longitudinal scRNA-seq of PBMC in babies with MCS or NEC.

a Experimental design. Paired PBMC samples were obtained from two babies with NEC and two babies with MCS. Graphs show temporal changes in CRP over time and black arrows depict when PBMC were obtained for scRNA-seq from each baby in relation to the onset of the clinical event (dotted line). Figure created with BioRender.com. b UMAP projection of 14,729 cells (8 samples from 4 babies). c Violin plots show expression of canonical gene markers (columns) for each cell cluster depicted in (b) (rows). Box plots; central line denotes the median, upper and lower lines of box represent the 75th and 25th percentiles respectively and whiskers show the range (minimum to maximum value). The same colours are used for corresponding cell clusters in (b) and violin plots in (c). Source data are provided as a Source Data file.

Fig. 3. AREG is upregulated during MCS and NEC.

Volcano plots show differential gene expression within PBMC obtained a at onset (day 0) or b within 3 days of onset of MCS/NEC against the temporally closest blood sample from the same baby (see Fig. 2a for timing of sampling). Differential gene expression was determined with the differential expression algorithm in Loupe Browser (10X genomics, v5.1.0), which is based on the negative binomial exact test used in the sSeq method. The top 50 DEGs (by Benjamini-Hochberg (BH) adjusted p-values) are shown in maroon. c, Venn diagram (created in DiVenn 1.2) compares all DEGs (BH adjusted p-values < 0.05, Log2fold change >0.95 or <-0.95) between the three responses shown in (a). The number of DEGs in each cluster are annotated. d–g, Pathway analysis of all upregulated genes (BH adjusted p-values < 0.05) taken from distinct clusters of the Venn diagram in (c), as described. Bar graphs depict the top ten enriched terms from the Reactome 2022 gene set library (pink bars) or GO Biological Processes 2021 library (turquoise bars). P-values (Fisher exact test), adjusted for multiple testing using the Benjamini–Hochberg method, are annotated on each bar. Haemaglobin genes (HBB, HBA1, HBA2, HBG1, HBG2, HBD, HBE1, HBM, HBQ1, HBZ) were filtered prior to making all graphs. Source data are provided as a Source Data file.

Fig. 4. AREG is highly expressed in NK cells and HSPC.

a UMAP of AREG expression normalized by UMI count (11,756 cells from n = 6 samples from babies A5, A12 and B6). b, Violin plots depict log normalized AREG expression in cell clusters from (a). Box plots; central line denotes the median, upper and lower lines of box represent the 75th and 25th percentiles respectively and whiskers show the range (minimum to maximum value). c Forest plot summarizes cell-specific expression of AREG from meta-analysis of public gene datasets (metasignature.standford.edu). Boxes on each row are centred at the mean effect size and are scaled to standard error. Whiskers extend to the 95% confidence interval. d Volcano plots show NK cell DEGs (determined with the differential expression algorithm in Loupe Browser (10X genomics, v5.1.0), which is based on the negative binomial exact test used in the sSeq method) for three distinct clinical events; AREG is highlighted on each plot. e Venn diagram showing overlapping upregulated NK cell DEGs (Benjamini–Hochberg adjusted p-value < 0.05) between the three clinical episodes shown in (d). Source data are provided as a Source Data file.

Fig. 5. Amphiregulin is rapidly inducible from neonatal leukocytes.

Frequency of amphiregulin⁺ a HSPC. b NK cells. c Monocytes. d B cells. e CD4⁺ T cells. f CD8⁺ T cells and g γδ-T cells after mitogen activation of PBMC in samples obtained across a range of postnatal ages (n = 72 samples from 11 babies). Top panels (a–g) show a representative flow cytometry plot of amphiregulin staining for each cell subset. Bottom panels (a–g) show a linear regression line with shaded 95% confidence intervals. Source data are provided as a Source Data file.

Fig. 6. Perturbations in plasma amphiregulin during sepsis.

a Plasma amphiregulin levels in 9 babies (n = 23 samples) who did not encounter sepsis or NEC during their admission. Right: Box plots; central line denotes the median, upper and lower lines of box represent the 75th and 25th percentiles respectively and whiskers show the range (minimum to maximum value). b Timepoints for plasma collection from each baby who encountered sepsis. Repeated samples from the same sepsis episode (marked with an X), or those collected during NEC (grey circles) without positive microbiological tests were excluded from analysis in (d–e). c Longitudinal changes in amphiregulin (orange line) and CRP (grey dotted line) in babies who encountered one or more episodes of sepsis or NEC (representative plots are shown for 15/23 babies; yellow circles = ClinSep; red circles = MCS; grey bars = NEC). d Comparison of plasma amphiregulin, IL-6, CXCL8, TNF-α, IL-10 and IFN-γ levels between ‘No-Sepsis’ (n = 116 samples from 23 babies) versus ‘Sepsis’ with normal/low CRP (≤10 mg L⁻¹; n = 21 samples from 15 babies) versus ‘Sepsis’ with elevated CRP (>10 mg L⁻¹; n = 25 samples from 18 babies). P-values were generated by a Kruskal–Wallis test with Dunn’s post hoc correction. Box plots; central line denotes the median, upper and lower lines of box represent the 75th and 25th percentiles respectively and whiskers show the range (minimum to maximum value). e Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of a test based on CRP and amphiregulin together to diagnose/rule-out sepsis (Test negative = CRP ≤ 10 mg L⁻¹ AND amphiregulin ≤38.7 pg mL⁻¹ where the cutoff value was defined from the maximal Youden index. Test positive = CRP > 10 mg L⁻¹ AND/OR amphiregulin >38.7 pg mL⁻¹). Analysis restricted to samples obtained within 48 h of a suspected sepsis episode (NSC n = 28 from 17 babies; Sepsis (ClinSep + MCS) n = 33 from 20 babies). Prevalence in this setting = (ClinSep + MCS)/(ClinSep + MCS + NSC); the number of suspected sepsis samples that were subsequently classified as MCS or ClinSep. Source data are provided as a Source Data file.