Transcriptional analyses were based on two historical monocenter prospective cohorts of adults admitted to the interdisciplinary surgical ICU of a tertiary care hospital (University Medical Center Mannheim) with a recent diagnosis of SIRS or sepsis for discovery (2012–2014) (31, 32) and validation (2016–2017) (32). Samples for flow cytometric determination of CD15 blood cell counts included samples from an extended recruitment period (2018–2020) of the validation cohort. The criteria for inclusion and exclusion were detailed before (31, 32). In addition to demographic and clinical characteristics on admission (32), data on clinical phenotypes close to the time of blood sampling were retrieved from the medical records ( Supplementary Text 1 ) and are summarized in Tables 1 , 2 . Cells were selected from whole blood using CD15 MicroBeads (iltenyi Biotec, Bergisch Gladbach, Germany), preserving all stages of terminal granulocytic differentiation (32). Validation samples stored at −80°C were available for this study. Patient-level data on clinical phenotypes, characteristics of CD15⁺ cells and RNA preparations, as well as CD15 blood cell counts are available from heiDATA (https://heidata.uni-heidelberg.de/dataset.xhtml?persistentId=doi:10.11588/data/EIXOPN). Briefly, all cohorts included critically ill patients with a recent diagnosis of SIRS or sepsis according to sepsis-3 and septic shock according to sepsis-1/2. SIRS was due to surgical trauma and polytrauma, respectively, 12 and 4 times in the discovery cohort and 16 and 6 times in the validation cohort (32). In the discovery cohort, the septic focus was 11 times abdominal and 4 times pulmonary. In the validation cohort, it was 9 times each abdominal and pulmonary, twice soft tissue, and once urogenital (32). Information on the clinical phenotype of patients contributing to the flow cytometric blood count determination is included in the Results. The discovery cohort also included adult controls at their clinical examination a few days prior to elective surgery. The American Society of Anesthesiologists (ASA) physical status score of these presurgical controls was between 1 and 3 (mean, 1.7). One had a diagnosis of chronic obstructive pulmonary disease, four of diabetes, and two of rectal cancer.

The discovery CD15 transcriptomes (32) was analyzed for differential gene expression by one-way analysis of variance and subjected to pathway analysis by enrichment (33) of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways using JMP10 Genomics version 6 (SAS Institute). A false-positive rate of α ≤ 0.05 with false discovery rate correction (FDR-q value) was considered significant, and differential expression was assumed. A list of all DEGs and the results of the pathway enrichment analysis are available from heiDATA (https://heidata.uni-heidelberg.de/dataset.xhtml?persistentId=doi:10.11588/data/EIXOPN).

Differential expression was validated by QuantiGene Plex (QGP) assays (Thermo Fisher Scientific) ( Supplementary Table S1 ). Technical duplicates were averaged. AKIRIN1 served as a reference gene (32). Due to limited volume availability, validation samples were randomly split into overlapping subsets A and B ( Supplementary Table S1 ). Expression was considered confirmed in a patient group if the assay signal was above background for at least four patients in that group. Normalized QGP results are available from heiDATA (https://heidata.uni-heidelberg.de/dataset.xhtml?persistentId=doi:10.11588/data/EIXOPN).

Differential expression of validated DEGs was assessed in two neutrophil transcriptomes available from the Gene Expression Omnibus (GEO) Profiles database (www.ncbi.nlm.nih.gov/geoprofiles) (34) under identifiers 41143967 (24) and 48169967 (25).

Information on the localizations and functions of proteins encoded by validated DEGs was obtained from the GeneCards database (www.genecards.org) (35). A list of 442 lysosomal genes was retrieved from The Human Lysosome Gene Database (hLGDB) (http://lysosome.unipg.it, accessed on March 20, 2019) (36). We analyzed confirmed signature genes of granule biogenesis in the microarray data for the discovery CD15⁺ cells, referring to a previous study on bone marrow maturation of human neutrophils (37). The authors used the proteome profiles of the following human neutrophil organelles, determined before by Rørvig et al. (38), as a basis:

From this, they compiled lists of signature genes encoding proteins, for which localizations to these respective compartments were supported by additional existing literature (see Supplementary Table S6 in the online supplement to their report (37)). These lists served as a reference in our study.

Blood counts of granulocyte precursors were determined in EDTA-anticoagulated blood by flow cytometry ( Supplementary Figure S1 ; Supplementary Table S2 ). The results are available from heiDATA (https://heidata.uni-heidelberg.de/dataset.xhtml?persistentId=doi:10.11588/data/EIXOPN).

Groups and proportions were compared with the Mann–Whitney U test and Fisher’s exact test, respectively, using Prism 7 (GraphPad Software, San Diego). The absence of detectable gene expression by QGP in a given sample, i.e., a signal at or below background, was more frequent in the respective patient comparison group with the lower average expression level. We treated such values as missing rather than setting them to zero, i.e., background, thereby effectively reducing the sample size and following a rather conservative analytical approach. Additionally, Bonferroni adjustment was applied to QGP-validation data for DEGs with confirmed expression in both patient groups by individual selection strategy as well as to all selected DEGs with confirmed expression in both patient groups together. Differential expression was regarded as validated if statistical significance was reached for any of the strategies. In the GEO Profiles data sets, unadjusted p-values were calculated using t-tests or, in case the normality test failed, Mann–Whitney U tests. p-values <0.05 were considered significant.

Principal component analysis (PCA) was applied to the flow cytometric CD15 blood count and the QGP validation gene expression data each to obtain a condensed, lower-dimensional representation of the underlying respective data while preserving as much of the variation of the original data as possible. Here, missing values in the GPQ data were set to zero (i.e., the nominal background). PCAs were conducted using the untransformed data with centering and unit variance scaling applied to the CD15 subpopulations and QGP-validated DEGs, respectively, while singular value decomposition was employed to derive the principal components. Two-dimensional PCA plots, showing the first two principal components, are provided as an output. The analyses were performed using the programming language R (39) for statistical computing.

In the discovery cohort, mean values for the sequential organ failure assessment (SOFA) score as well as blood CRP and lactate on ICU admission were significantly higher in patients with sepsis than in SIRS patients (32). Blood lactate levels on admission exceeded 2 mM for eleven out of 15 sepsis patients that thus had septic shock according to sepsis-3. Both ICU groups had higher on-admission white blood cell counts (WBCs) than presurgical patients, but higher counts in sepsis than SIRS did not reach statistical significance either on admission or close to the time of sampling. Differences in additional blood parameters between patients with sepsis and SIRS in the discovery cohort close to the time of blood sampling did not reach statistical significance. The frequencies of mechanical ventilation and hospital mortality were higher in sepsis than in SIRS. Table 1 summarizes clinical patient characteristics of ICU patients in the discovery and validation cohorts. As an additional indicator of bacterial infection, blood PCT was routinely determined on admission in all 15 discovery cohort patients with sepsis (mean, 51.9 µg/L; range, 2.0–371.1 µg/L) but only eight out of 16 with SIRS (mean, 1.9 µg/L; range, 0.2–11.1 µg/L).

Globally, 6,730 (27.2%) of all genes showed differential expression. In PCA analysis based on these DEGs, CD15 transcriptomes separated well into three distinct clusters ( Figure 1A ). Figure 1B displays relations for DEGs across the three pairwise group comparisons. Of all DEGs, 44.3% were found in the sepsis vs. presurgical, 56.3% in the sepsis vs. SIRS, and 79.3% in the SIRS vs. presurgical comparison. Shared gene expression differences in sepsis compared to SIRS and sepsis compared to presurgical controls without a change in the same direction also in SIRS compared to presurgical controls were absent ( Figure 1B ). Eight of the top ten DEGs, judged by statistical significance, were shared between the comparisons of each ICU group with the presurgical group ( Figure 1C ). However, 1.79 times more genes were differentially expressed in SIRS than in sepsis compared to presurgical controls, and only 6.0% of all DEGs were unique to the sepsis vs. presurgical comparison compared to 16.2% for SIRS vs. presurgical.

Results of the enrichment analysis for canonical pathways in the microarray data for the discovery of CD15⁺ cells revealed significant associations for similar numbers of pathways with presurgical patients compared to either ICU group, 19 for sepsis, and 21 for SIRS ( Supplementary Figure S2 ). They were predominantly from the functional classes of organismal systems and human diseases and driven by high proportions of major histocompatibility complex class II genes. These pathways are not considered in more detail because they are, mainly, attributable to cell types other than granulocytes. Compared to presurgical patients, SIRS was associated with enrichment of only three and sepsis with 27 pathways, mainly metabolism. In a comparison of the two ICU groups, only sepsis was associated with pathway enrichment. There were in total 45 pathways, including 20 metabolism, 10 genetic information processing, and four cellular processes. Figure 1D charts normalized enrichment scores for the top eight pathways enriched in sepsis compared to SIRS. These included three pathways of mitochondrial metabolism (oxidative phosphorylation, citrate cycle, and fatty acid metabolism) and two of each of genetic information processing (proteasome and ribosome) and cellular processes (lysosome and cell cycle). Except for proteasome, these pathways were also enriched in a comparison of sepsis, but not of SIRS, to the presurgical group.

As with the discovery set, SOFA, CRP, and lactate values on admission were higher in sepsis than SIRS, while higher WBCs in sepsis on admission and close to the time of sampling were not statistically significant (32). In validation subsets A and B, respectively, blood lactate levels on admission were above 2 mM for eight out of 18 and nine out of 17 sepsis patients. In both subsets, serum creatinine levels were elevated in sepsis compared to SIRS in the absence of other statistically significant differences in clinical characteristics ( Table 1 ). Frequencies of anti-infective and catecholamine treatments as well as congestive heart failure and diabetes without chronic complications were more frequent in patients with sepsis than in SIRS ( Table 1 ). PCT on admission was available for all 21 patients with sepsis in both subsets (mean, 20.1 µg/L; range, 0.3–54.8 µg/L) but only five out of 22 with SIRS (mean, 0.8 µg/L; range, 0.3–2.0 µg/L).

To select DEGs for validation, two strategies were initially applied.

Strategy 1: The top 100 DEGs from the sepsis vs. SIRS comparison with the highest mean expression levels in sepsis and SIRS each were joined and ranked by ascending FDR-adjusted p-values. Forty-nine of the top 50 genes from this list (short of MMP8) were selected. Six additional genes were selected, for which signal ranges in the high expression group did not overlap with the 75th percentile of the low expression group and the range in the low expression group did not overlap with the 25th percentile of the high expression group.

Strategy 2: Fifty-four DEGs showed a >2-fold mean difference in sepsis and SIRS and, concomitantly, a <1.1-fold mean difference in SIRS and presurgical patients and were therefore selected.

Out of the nonredundant list of 100 DEGs identified by strategies 1 and 2 ( Figure 2A ), expression in both sepsis and SIRS was confirmed for 89, and differences between these groups were validated for 30 ( Supplementary Table S3 ). For 12 validated DEGs, GeneCards annotations and additional literature indicated endo-lysosomal associations: CTSA, GUSB, HEXA, RNASE2, SGSH, and TPP1 encode endo-lysosomal hydrolases; CDK5 (40) and FIG4 (41) regulators of endo-lysosomal transport; APP, CD68, and LAIR1 endo-lysosomal membrane proteins; and DIAPH2 a regulator of endosome dynamics (42). This observation prompted a third selection strategy.

Strategy 3: According to hLGDB, 14 DEGs from the joined list of the top 100 DEGs with the highest mean expression levels in sepsis and SIRS were lysosomal genes. Ten were already included in strategies 1 and 2, and four were additionally selected ( Figure 2A ; Supplementary Table S3 ).

Strategy 3 led to the confirmation of expression in sepsis and SIRS for four additional genes and validation of the endo-lysosomal membrane protein-encoding LDLR gene, already selected by strategy 1 due to a shorter gene list and thus less conservative Bonferroni adjustment with strategy 3 than 1 ( Supplementary Table S3 ). Figure 2B summarizes the results for the 31 validated DEGs and identifies the respective selection strategy. All but ADGRE3, LINC02289, and PRKDC, showed higher mean expression in sepsis than SIRS, ranging from 2.7-fold (ANLN) to 21.6-fold (SLAMF7). Figure 2C compares signal intensity distributions for the 13 validated DEGs with endo-lysosomal associations and Supplementary Figure S3 for the remaining 18 validated DEGs. Figure 2D revisits the discovery set microarray results for validated DEGs. Only sepsis patients formed a separate cluster.

Validated DEGs were assessed for differential expression in two external microarray data sets for density-gradient purified neutrophils from patients admitted to the ICU with and without sepsis ( Supplementary Figure S4 ). One featured a training and a validation cohort (24). Differences between sepsis and controls reached a statistical significance in either cohort separately and/or the pooled training and validation cohorts for a total of ten genes, including higher levels in sepsis for the endo-lysosomal genes APP, CDK5, GUSB, LAIR1, and SGSH. The second study compared Gram-positive, Gram-negative, and mixed sepsis to ICU controls (25). Any of the three and/or the pooled sepsis groups significantly differed from the ICU controls for a total of eleven genes, including higher sepsis levels for the endo-lysosomal genes DIAPH2, GUSB, HEXA, LAIR1, and SGSH.

Cessation of cell proliferation and switching from oxidative to glycolytic metabolism at the metamyelocyte stage is associated with a decline in expression of cell cycle and mitochondrial genes, respectively, and additionally, proteasomal genes (43). These specific changes conspicuously match our profile of canonical pathway enrichment in sepsis ( Figure 1D ). Notably, the biogenesis of the eponymous granules during terminal granulocytic differentiation is known to depend on transcriptional waves that restrict the formation of azurophilic granules to the promyelocyte stage and of specific granules to the myelocyte/metamyelocyte stage, known as the “targeting by timing” mechanism of granule protein sorting (37, 38). Importantly, azurophilic and specific granules are lysosomal in nature. Our pathway enrichment profile ( Figure 1D ), together with validated higher expression of endo-lysosomal genes in sepsis than SIRS ( Figure 2 ), led us to predict that signature genes of azurophilic and specific granule biogenesis also showed the highest expression in sepsis, reflecting elevated blood counts of immature granulocytes.

To test this, we assessed the profiles of known signature genes of granule biogenesis (37) among all 6,730 DEGs from the discovery of CD15⁺ cells ( Figure 3A ). Among the azurophilic granule genes, PLAC8 showed, on average, the strongest fold increase in both sepsis and SIRS compared to presurgical patients. In the second main gene cluster of azurophilic granule genes, increases were more pronounced for sepsis than SIRS, with the exception of GRN. Among the specific granule genes, only three genes, including CHI3L1, showed reduced levels in both sepsis and SIRS compared to presurgical. Otherwise, profiles mostly featured gradual increases from presurgical to SIRS and sepsis patients. Three specific granule genes (OLFM4, LTF, and LCN2) showed <2-fold differences in SIRS vs. presurgical but >10-fold higher levels in sepsis vs. both SIRS and presurgical. The relatively small-sized gelatinase-containing granule gene profile was dominated by increases in MMP9 and CD177 in both sepsis and SIRS. Notably, CD177 encodes an activation marker in chemotaxis and showed the highest increase in expression in septic compared to healthy high-density neutrophils in a previous microarray study (44). Compared to signature genes of azurophilic and specific granules as well as gelatinase-containing granules, the extent of differences between patient groups was moderate for ficolin-containing granules, secretory vesicles, and the cell membrane signature genes ( Figure 3A ).

Additionally, we assessed patient group differences by considering group averages for all individual genes within the complete set of signature genes, i.e., their global expression level, for a given compartment ( Figure 3B ). Sepsis but not SIRS was associated with a mean 1.7-fold elevation in global azurophilic granule gene expression ( Figure 3B ). Global expression of specific granule genes gradually increased by comparable margins from presurgical to SIRS and further to sepsis, while global gelatinase-containing granule gene expression was similarly elevated in both ICU groups compared to the presurgical group. There were no apparent patient group differences for the remaining compartments.

We did not formally analyze transcription factors but noted that the promyelocyte-myelocyte transition regulating CEBPE gene (45) showed 44% higher mean levels in sepsis than in SIRS. CEBP was not included in any of the selection strategies applied in this work.

Samples from ten patients with sepsis and four with SIRS from the validation cohort and, additionally, from ten with sepsis and ten with SIRS from an extension of the validation cohort ( Supplementary Table S4 ) were subjected to flow cytometric determination of CD15 blood counts for both the precursor populations and the mature granulocytes referred to as polymorphonuclear neutrophils (PMNs) ( Supplementary Figure S1 ). For the patients included in this analysis, SOFA and CRP values on admission in sepsis were higher than in SIRS, with sodium values only marginally higher. Slightly higher WBC values in SIRS than in sepsis on admission and close to the time of sampling did not reach statistical significance. Admission levels of blood lactate were above 2 mM for half of the 20 sepsis patients and one out of the 14 SIRS patients. Anti-infective treatment, mechanical ventilation, and administration of catecholamines, as well as diabetes without chronic complications were more frequent in sepsis than SIRS. The clinical characteristics of these patients are summarized in Table 2 . Additionally, PCT on admission was available for 18 of the sepsis patients (mean, 10.5 µg/L; range, 0.3–53.8 µg/L) and 5 of the SIRS patients (mean, 1.7 µg/L; range, 0.1–3.9 µg/L).

Throughout, blood counts of CD15 subpopulations were higher in sepsis than in SIRS ( Figure 4A ). Among the precursor populations, the mean difference was highest for late promyelocytes (16.9-fold), intermediate for myelocytes and band cells (10.9- and 11.6-fold, respectively), and lowest for metamyelocytes (6.5-fold). Only a 23% higher average count of PMNs in sepsis than in SIRS did not reach statistical significance (p = 0.5). PCA plots of the first two principal components suggest higher similarity among the SIRS samples in relation to the sepsis samples when based on CD15 subpopulation counts ( Figures 4B, C ) than on expression levels of validated DEGs ( Figures 4D, E ). The two patient groups appear less similar with than without taking PMN counts into account ( Figures 4B vs. 4C ). They also appear less similar when the analysis is based on the 13 DEGs from validation subset B compared to the 18 from subset A ( Figures 4E vs. 4D ).