Identification of a transcriptome profile associated with improvement of organ function in septic shock patients after early supportive therapy

This study is part of the multicenter prospective observational trial ShockOmics (ClinicalTrials.gov Identifier: NCT02141607) [11]. The ICUs involved are located in the Hopitaux Universitaires de Genève, Universite de Geneve (Geneva, Switzerland) and Hospital Erasme, Universite Libre de Bruxelles (Brussels, Belgium). The ethics committees of the two participating institutions approved the clinical protocol. Informed consent was obtained from the patients or their representatives. As previously described [11], we included consecutive adult (>18 years old) patients admitted for septic shock in the ICUs of two university tertiary hospital centers with an admission SOFA score at least 6 and an arterial lactate at least 2 mmol/L. Septic shock was defined in accordance with the current recommendations and international guidelines at that time [12]. Exclusion criteria included expected death within 24 h of ICU admission; more than 4 units of red blood cells or more than 1 fresh frozen plasma transfused; active hematological malignancy; metastatic cancer; chronic immunosuppression; pre-existing end-stage renal disease requiring renal replacement therapy; recent cardiac surgery; Child-Pugh C cirrhosis; and terminal illness.

We used SOFA score changes between time point 1 (T1), which was within 16 h of ICU admission, and time point 2 (T2), which was 48 h after study enrolment as an index of response to early supportive treatment to stratify subjects as responders (Rs) or non-responders (NRs) as previously reported [13]. A subject was classified as NR if the difference in SOFA score between T1 and T2 was less than 5 and the SOFA score at T2 was greater than 8. Patients who did not meet these criteria were classified as R.

Peripheral blood was collected in EDTA tubes at T1 and T2 and stored at −20 °C after adding 400 μL of 2X Denaturing solution (Ambion, Austin, TX, USA). Total RNA was extracted from 800 μL of blood using the MirVana Paris Kit and treated with Turbo DNA-free Kit (Ambion). RNA quality was assessed on an Agilent Bioanalyzer using the RNA 6000 Nano Kit (Agilent, Santa Clara, CA, USA), and samples were considered suitable for processing if the RNA integrity number was greater than 7.5.

Sequencing libraries were prepared using TruSeq Stranded Total RNA obtained from the Ribo-Zero Globin Kit (Illumina, San Diego, CA, USA) using 800 ng of total RNA. Final libraries were validated with the Agilent DNA1000 kit and sequenced on a HiSeq2500 platform, producing 50 × 2-base pair paired-end reads.

High-quality paired-end reads were aligned to the human reference genome (GRCh38) using STAR (version 2.5.2b) [14], and only uniquely mapping reads were used. Reads were assigned to genes with featureCounts (version 1.5.1) [15] using the gencode (version 25) primary assembly gene transfer file (GTF) as a reference annotation file for genomic feature boundaries.

Data preprocessing, exploratory data analysis, and analysis of differential gene expression were performed using DESeq2 package [16] built-in functions. We studied gene expression differences between R and NR at T1 and gene expression changes over time. These data were compared between T2 to T1 in R and NR groups, separately, using paired analysis. For both analyses, genes with padj < 0.01—Benjamini–Hochberg multiple test correction (FDR)—were considered differentially expressed (DEGs).

We used ClueGO version 2.3.4 [17] to identify enriched biological processes starting from the lists of DEGs identified in the over-time analysis in R and NR and the DEGs common to the two groups. The lists were used as input for the analysis after filtering for padj < 0.001. We performed an over-representation analysis (ORA) based on a right-sided hypergeometric test that uses Bonferroni step down as multiple testing correction. Enriched terms were grouped using kappa statistics, and only enriched clusters with a P value of less than 0.05 were considered statistically significant.

To validate the DEGs identified in the transition from ICU admission to 48 h later, we searched in Gene Expression Omnibus (GEO) and Array Express repositories for gene expression datasets of septic shock patients studied at multiple time points. The microarray expression dataset GSE57065 included 28 septic patients studied at the onset of shock (that is, within 30 min after the beginning of vasoactive treatment, H0) and 24 and 48 h after (H24 and H48) [8]. We stratified patients as Rs or NRs using SOFA score changes between H0 and H48 as described in the “Study design and participants” section. Patients missing the H48 blood sample were excluded from the validation that was performed on 17 Rs and 9 NRs. We compared gene expression between H0 and H48 in the R and NR groups, separately. We established the analysis modifying the GEO2R script to take into account paired data [18]. Genes with padj < 0.01—Benjamini–Hochberg multiple test correction (FDR)—and log2FC > |0.5| were considered DEGs.

The study included 31 patients with septic shock enrolled between November 2014 and March 2016. In total, 17 patients were classified as R and 14 were classified as NR. The clinical characteristics of Rs and NRs at T1 and T2 are summarized in Table 1.

At T1, no significant differences were noted between the two groups in terms of variables describing the severity of organ dysfunction, whereas, consistent with our definition of R and NR, the SOFA score at T2 was increased in NRs (P <0.001). No significant differences were noted between the two groups regarding the source of infection, circulating markers of inflammation, or leukocyte and lymphocyte counts. The length of the ICU stay was longer in NR patients (P <0.05). More patients died in the ICU in the NR group (4 out of 14 in NRs versus 2 out of 17 in Rs); however, this difference was not statistically significant.

Total RNA libraries were sequenced in several batches, producing 31.07 M ± 6.92 M and 29.33 M ± 7.26 M raw read pairs on average for Rs and NRs, respectively. Ribosomal depletion was effective for all samples; the rRNA rate on mapped data was negligible in both groups (0.80 ± 0.95% and 0.78 ± 0.82% for Rs and NRs, respectively). The percentages of reads mapping to exons (86.09 ± 5.83% exonic rate) and DNase efficiency (2.75 ± 2.02% intergenic rate) were satisfactory in all samples. We obtained on average 13.67 ± 3.77 and 13.41 ± 3.69 million of uniquely and unambiguously mapped fragments for Rs and NRs, respectively (Additional file 2: Table S1). Globin reduction was successful with average percentages of counts mapping on globin genes of approximately 0.13% and 0.14% for NRs and Rs, respectively. Moreover, depletion efficiency was not statistically different for any of the genes coding for globins (Additional file 3: Table S7). Although we cannot test whether globin reduction influences sequencing efficiency of non-globin genes, this would equally affect both NR and R samples and thus is likely to bias toward the null.

We explored whether the transcriptome profile could distinguish Rs from NRs at ICU admission (T1). Principal component analysis was performed on the 2000 most variable genes across samples and did not reveal any separation between the R and NR groups (Fig. 1). Indeed, we could not identify any statistically significant DEG at T1 between Rs and NRs (padj < 0.01).

We compared gene expression profiles between T2 and T1 in the R and NR groups independently. This analysis generated two lists of DEGs with 1979 DEGs (865 upregulated and 1114 downregulated genes) in Rs (Additional file 4: Table S2) and 1914 DEGs (711 upregulated and 1203 downregulated genes) in NRs (Additional file 5: Table S3). The two lists of DEGs were filtered at a padj < 10⁻³, yielding 1146 and 1120 genes in Rs and NRs, respectively, and were used as input for an ORA. In Rs, 14 clusters of significantly over-represented GO terms were identified with padj < 0.05 (Table 2; Additional file 6: Table S4). In NRs, the same analysis identified 15 clusters of over-represented GO terms (Table 3; Additional file 7: Table S5). A comparison between the GO networks identified in Rs and NRs was performed in Cytoscape, highlighting GO terms specifically enriched exclusively in Rs or NRs or enriched in both groups (Figs. 2 and 3). In Rs, we identified three robust GO clusters with highly interconnected GO terms (positive regulation of leukocyte activation, activation of immune response, and T-cell differentiation) (Fig. 2) that were not identified in NRs (Table 2), suggesting that the adaptive immune response is activated mostly in Rs at T2.

In NRs, we could not identify any GO cluster with highly interconnected GO terms with the exception of clusters of signal transduction and regulated exocytosis, which share many GO terms with Rs (Fig. 3). Among the GO clusters specifically enriched only in NRs, we highlighted the “Response to lipopolysaccharide” GO cluster, including CCL5, CXCL1, CXCL5, IRAK1, and NLRP3 genes that were exclusively downregulated in NRs. Analysis of the 752 DEGs common to Rs and NRs identified eight clusters of significantly enriched GO terms (Table 4) that are related to modulation of genes involved in the innate immune response and neutrophil function.

We identified 125 genes (P <0.01, Mann–Whitney) with significantly different trends between Rs and NRs in the transition from T1 to T2 (Fig. 4; Additional file 8: Table S6). Thirteen genes (CD247, FCER1A, ICOS, LCK, SH2D1A, CLC, IRAK3, VSIG4, GPR171, GPR18, HIST1H3A, ZNF784, and TIGIT) among 125 are involved in T-cell functions and support the results highlighted in the GO analysis that Rs modulate biological processes of the adaptive immune response.

We observed that, among the 125 genes, genes involved in granulocyte functions (IL5RA, FCER1A, CCR3, and PTGDR2) and genes involved in natural killer (NK) cell activity (GZMM, GZMH, and KLRG1) were significantly upregulated at T2 exclusively in Rs. We also identified PCSK9 and SORT1 genes, which are involved in the lipid pathway. PCSK9 is significantly downregulated in Rs at T2. In contrast, SORT1 is downregulated at T2 in both groups, but different fold changes are noted for each group. We also observed genes with a role in the maintenance of vascular tone and endothelial function, such as CPA3 (upregulated at T2 in Rs), ARG1 (downregulated at T2 in Rs), and AVPR1A (downregulated at T2 in NRs).