Public repositories of articles and data, such as PubMed and GEO, constitute a vast resource but they can be difficult to explore. Here we present a logical reductionist approach to investigate putative novel biomarkers for sepsis.

The steps consist of 1) identifying a gene of interest based on its differential expression in the pathological/physiological context of interest, 2) confirming the reproducibility of the initial observation, 3) determining the current body of literature linking the gene and topic, 4) extracting the known biological concepts concerning the gene, and 5) inferring putative novel roles for the gene with literature support.

All datasets were obtained from GEO and used to confirm the initial findings in relevant clinical settings/samples. Datasets were selected without prior knowledge of ACVR1B expression levels and consisted only of human studies in which transcriptome profiles were generated in septic patients and compared to uninfected controls ( Table 1 ). The task of identifying datasets was facilitated by using an interactive database recently created by our group and called SysInflam HuDB (sepsis.gxbsidra.org/dm3/geneBrowser/filteredSampleSets) (18). Other relevant information was retrieved from each GEO entry, such as: the geographic localization of the patient population, and the type of biological samples.

The work involving human subjects was approved by the ethical committee of Faculty of Tropical Medicine, Mahidol University (approval no. MUTM 2015-002-04, MUTM 2018-046-01 and MUTM 2018-039-02), Udon Thani Hospital (approval no.6/2561), Nakhon Phanom Hospital (approval no. IEC-NKP1-No.15/2558), Roi Et Hospital (approval no. 166/2559), Buriram Hospital (approval no. BR 0032, 102.3/57), and Surin Hospital (approval no. 21/2560). This study was conducted in accordance with the principles of the Declaration of Helsinki (2008) and the International Council for Harmonization and Good Clinical Practice guidelines. Written informed consent/assent form was obtained from all participants or their legal guardians.

Plasma was collected from 26 patients diagnosed with melioidosis and 9 healthy subjects. Participants were recruited at five hospitals in northeastern of Thailand (endemic region for melioidosis): Udon Thani Hospital, Nakhon Phanom Hospital, Roi Et Hospital, Buriram Hospital, and Surin Hospial. This study was part of a multi-center study of patients aged ≥15 years who were culture positive for B. pseudomallei from any clinical samples and admitted to the hospitals between January 2015 and December 2019. The inclusion and exclusion criteria were described previously by Kaewarpai et al. (19). Plasma was collected at the time of enrolment (within 24 hours of reporting of the culture results). Patients’ characteristics and clinical information was obtained from the medical records. Ten milliliters blood were collected from 3 healthy donors aged more than 18 years old at the Faculty of Tropical Medicine, Mahidol University, Bangkok and immediately used for cell stimulation.

Buffy coat from EDTA whole blood of three healthy individuals was isolated by centrifugation at 1500 x g for 15 min. Fresh buffy coat was diluted three-fold with RPMI 1640 medium (Gibco, Invitrogen, CA, USA) supplemented with 10% fetal bovine serum (FBS, Himedia, Mumbai, India), 100 U/ml penicillin (Gibco), 100 µg/ml streptomycin (Gibco), and 2 mM GLutaMAX (Gibco) and added to 96-well plates (Costar Corporation, Cambridge, MA, USA) in duplicate wells at 2x10⁵ cells in 150 µl per well. The cells were stimulated with 50 µl of plasma from melioidosis patients (n = 26), healthy donors from an endemic area (n = 9) and autologous donors (n = 3 replicates). Stimulated buffy coat was incubated at 37^°C in a 5% CO₂ atmosphere incubator for 4 h.

Total RNA was extracted from buffy coat of healthy donors following stimulation using the TRIzol^® reagent (Thermo Fisher Scientific, Darmstadt, Germany) according to manufacturer’s instruction. The RNA was re-purified using the RNeasy Mini Kit and treated with DNase according to the manufacturer’s instruction (Qiagen, Valencia, USA). RNA quantity was measured using a nanodrop spectrophotometer (Thermo Fisher Scientific).

cDNA was generated by the iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Hercules, USA). Transcription level of ACVR1B was analyzed using real-time PCR with iTaq Universal SYBR Green (Bio-Rad). The primers for ACVR1B were: ACVR1B, forward 5′ CAGCAGAACCTTGGCGGTTTA 3′, reverse 5′ GTTGGCAGATCCCAGAGGCTAC 3′ (15) and for a housekeeping gene HuP0 were forward 5′ GCTTCCTGGAGGGTGTCC 3′, reverse 5′ GGACTCGTTTGTACCCGTTG 3 ′ (20). The PCR reaction was conducted in the CFX96^™ Real-Time system (Bio-Rad) as follows: 1 cycle of 95°C for 30s followed by 50 cycles of 95°C for 10s and 60°C for 30s. After amplification, melting curve analysis was carried out from 65°C to 95°C. The fold change of ACVR1B expression was calculated by 2^(-ΔΔCt), normalized with ΔCt = mean Ct of ACVR1B – mean Ct of HuP0 derived from the same cDNA sample.

The average ACVR1B expression was calculated from source data for case (i.e., sepsis) and control groups in each dataset. To compare case to control groups, linear fold-changes (FC) were calculated for each dataset by dividing the average expression of the cases by the average expression of the control groups. An F-test of equality of variances was used to determine whether a homoscedastic or heteroscedastic t-test was adequate to compare population means. To assess statistically significant difference in gene expression, a t-test was applied to compute a p-value. The Mann-Whitney and Kruskal-Wallis tests were used to compare differences between ACVR1B expression of 3 healthy donors-derived buffy coat, which was exposed to septic plasma. A of P < 0.05 was considered statistically significant.

The ACVR1B gene was identified from data we previously generated (GSE49755) (21). An increase in ACRV1B transcript abundance was observed upon the exposure of neutrophils from healthy donors to plasma of septic patients. Briefly in this study, polymorphonuclear neutrophils (PMNs) were obtained from two healthy donors, isolated, and incubated for 6 hours with the plasma from 12 patients with confirmed melioid sepsis and 12 uninfected controls, respectively. PMN transcriptional profiles were generated using Illumina Bead arrays. As shown in Figure 1A , ACVR1B gene expression levels and variance between exposure to plasma from the control and septic groups were significantly different ( Figure 1B ; Additional File 1).

The current body of literature on the overlap of ACVR1B on one hand and sepsis, inflammation, or neutrophil on the other hand was assessed. Literature was retrieved using a PubMed query which comprised the official gene symbol, name and known aliases for ACVR1B, with the search restricted to the title and abstract: ACVR1B[tw] OR ACTRIB[tw] OR ALK4[tw] OR SKR2[tw] OR Actr-IB[tw] OR ALK-4[tw]. As of August of 2022, this query returned 510 results. In a Boolean search, no overlap was found between the ACVR1B literature and the search term “sepsis”. Extending the search to the literature on inflammation and neutrophil returned 13 and 2 articles, respectively. Supplementary Figure 1 summarizes the number of articles returned in the respective searches.

Datasets relevant to septic conditions (clinically or experimentally) were retrieved from GEO, made available on Gene Expression Browser (GXB) GXB, and used for independent validation of our initial finding. The characteristic of the datasets retrieved are shown in Figure 2 and Supplementary Table 1 . These datasets encompassed gene expression profiling of in-vivo, ex-vivo and in-vitro experiments, and were all derived from human samples. Various inflammatory conditions, including cell exposure to bacterial products, were evaluated in case-control designs. Strong evidence (p < 0.05) for an increased ACVR1B expression in sepsis was found in 6 out of 11 datasets. The other five datasets had a p-value between 0.1 and 0.05. Four of those datasets were from experiments performed in-vitro. These four in vitro experiments all have smaller sample size (ranging from 3 to 17) when compared to the in-vivo experiments (ranging from 13 to 127). Te magnitude of ACVR1B expression was associated with the experimental conditions, such as the cell types used and whether plasma of septic patients or synthetic bacterial components (e.g., bacterial lipopolysaccharide) was used.

As a first step to bridging the gap between the postulated upregulation of ACVR1B in sepsis and the rational for this observation, another systematic literature search was conducted in PubMed. The following query, restricted to title-words and excluding the term “homeobox” to avoid irrelevant gene target (ALK4 can also referred to Aristaless-like homeobox 4), was executed: ACVR1B[ti] OR ACTRIB[ti] OR ALK4[ti] OR ALK-4[ti] NOT (“genes, homeobox”[MeSH Terms] OR (“genes”[All Fields] AND “homeobox”[All Fields]) OR “homeobox genes”[All Fields] OR “homeobox”[All Fields]) AND “humans”[MeSH Terms].

We obtained 33 articles and scanned the titles to extract key biological processes/terms and fit them into 5 biologically relevant categories: biological process, biomolecules, pathway, disease, and cell type. A word-cloud summary of the words contained in the titles is shown Supplementary Figure 1B . Next, literature searches were performed on the intersection between ACVR1B and each of these key biological processes/terms. To maximize the output, the search queries regarding ACVR1B, and its synonyms were extended to text words ([tw]) as follows: ACVR1B[tw] OR ACTRIB[tw] OR ALK4[tw] OR SKR2[tw] OR Actr-IB[tw] OR ALK-4[tw]. A summary of the categories and concepts is shown in Supplementary Figure 1C .

To test the findings of our in silico exploration, we measured ACVR1B transcript abundance in the buffy coat from healthy individuals that has been incubated with septic plasma. Septic plasma obtained from melioidosis patients were used for ex vivo validation. The clinical characteristics of melioidosis patients are shown in Table 1 . Buffy coats were isolated from three healthy donors and stimulated for 4 hours with plasma from melioidosis patients (n=26) and healthy controls from endemic areas (n=9), and autologous plasma (n=3). ACVR1B expression was measured by RT-qPCR. Abundance of ACVR1B transcripts was significantly higher in cells from two healthy donors (HC-01 and HC-02) stimulated with plasma of melioidosis patients compared with plasma of healthy donors (p < 0.01) ( Figure 3 ).

As melioidosis is associated with many underlying diseases, the patients’ preexisting conditions/status and potential consequences (i.e., bacteremia, pneumonia) of an infection with B. pseudomallei were examined as factors of ACVR1B gene expression in buffy coat after exposure to plasma of melioidosis patients. Blood cells of 3 healthy donors were incubated with plasma from patients with melioidosis. ACVR1B expression was significantly different between survivors and non-survivors (p<0.05), but no differences were found between male and female, diabetes and non-diabetes, bacteraemia and non-bacteremia, hypertension and non-hypertension, pneumonia, and non-pneumonia ( Supplementary Figure 2 ).