Identification of potential therapeutic drug targets for sepsis by combining the human plasma proteome and genome: a Mendelian randomization study

Introduction

Sepsis is a life-threatening organ dysfunction syndrome caused by a dysregulated host immune response to infection (1). Sepsis patients may present with systemic or local excessive inflammation, or with systemic or local immunosuppression (2). Despite advances in nursing care, the in-hospital mortality rate of sepsis remains as high as 25–30% (3). Currently, the treatment of sepsis remains mainly supportive, with a focus on antibiotics, resuscitation, and organ function support, and a specific treatment for sepsis has yet to be established (4). After the onset of sepsis, early diagnosis and treatment are important for a good prognosis. Therefore, new therapeutic targets for sepsis urgently need to be identified and developed.

Recent advances in genomics, immunology, bioinformatics, and large-scale database modeling have driven significant breakthroughs in identifying therapeutic targets for sepsis (5-7). Mendelian randomization (MR) is an epidemiological method based on whole-genome sequencing data used to reveal causal relationships (8). It uses genetic variation as an instrumental variable (IV) to study the causal relationship between the exposure of interest and the outcome (9). With the continuous development of genome-wide association studies (GWASs) and the improvement of proteomic data, the integration of these two methods has led to the emergence of drug-target MR studies (10,11). MR enables rapid prioritization of candidate genes by leveraging large-scale GWAS data, while genetic evidence derived from MR helps reduce ineffective target validation in clinical trials. Many researchers use protein quantitative trait locus (pQTL) from variants for MR studies, which has aided in the discovery of new targets for various diseases (12-14).

In this study, we combined druggable genes and proteins derived from previous studies to explore the causal link between proteins and sepsis by MR analysis (14,15). As MR alone might not have been sufficient to identify reliable proteins in the sepsis causal pathway, colocalization, summary data-based Mendelian randomization (SMR), and heterogeneity in dependent instruments (HEIDI) analyses were subsequently performed. A druggability assessment was performed to explore the potential of these genes as therapeutic targets for sepsis. Finally, a single-cell type expression analysis and proposed time-series analysis were conducted to detect enriched cell types in sepsis patients. We present this article in accordance with the STROBE-MR reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-590/rc).

Methods

The overall experimental flowchart of this study is shown in Figure 1. We first selected and acquired a list of the druggable genes, and then examined the intersection of these genes with two pQTL datasets and their associations with sepsis using two whole-proteome MR analyses. Subsequently, we further examined whether there was a causal relationship between the acquired protein markers and sepsis by colocalization, SMR, and HEIDI analyses. In further studies, we performed protein-protein interaction (PPI) and druggability assessments of the identified protein markers to rank potential therapeutic targets. Finally, we performed cell-sorting aggregation and a mimetic time-series analysis via single-cell sequencing to explore the cell types enriched for proteins encoded by different target genes during sepsis development. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Figure 1 Flowchart of the study design (14,15). HEIDI, heterogeneity in dependent instruments; MR, Mendelian randomization; PPI, protein-protein interaction; pQTL, protein quantitative trait locus; SMR, summary data-based Mendelian randomization.

Selection of druggable genes

A total of 4,479 druggable genes were obtained from the druggable genome and support for target identification and validation in drug development study (14). The first group comprised 1,427 genes, including targets for small molecule and biological therapeutics, as well as clinical stage drug candidates. The second group included 682 genes encoding protein targets with ≥50% homology to approved drug targets, and these genes have known bioactive drug-like small molecule binding partners. Furthermore, there were also 2,370 genes that encode secretory or extracellular proteins, which exhibit lower similarity to approved drug targets, as well as members of crucial druggable gene families not yet included in the first or second group.

pQTL dataset

pQTL were used to study the correlation between genetic mutations and gene expression using protein expression as a trait (16). Two sources of pQTL were used for the analysis. The preliminary analysis used data from Zheng et al., which included 738 cis-single nucleotide polymorphisms (SNPs) in 734 proteins, to screen pharmaceutically viable proteins for further analysis (15). Next, we obtained the pQTL data of the filtered proteins from a total of 4,907 proteins in the deCODE database, as reported by Ferkingstad et al. (17). This pQTL dataset was used as the main pQTL dataset to identify potential drug targets for sepsis treatment. The IV cis-pQTL were selected based on the following criteria: (I) P value <5×10⁻⁸; (II) the SNP was not located within the human major histocompatibility complex region; (III) the SNP was located within 1 MB upstream and downstream; and (IV) a removal of linkage disequilibrium (r²) value <0.1. The selected datasets all comprised individuals of European ethnic background.

Outcome dataset

We obtained the GWAS ID (IEU-B-4980) of sepsis patients from the Medical Research Council (MRC) IEU OpenGWAS database, and standardized association summary statistics were obtained through the “TwoSampleMR” R package (18). A total of 11,643 sepsis samples and 474,841 control samples were collected, and the outcome data were from individuals of European ethnic background. All sepsis cases were screened based on the Sepsis-3 definition.

Two-sample MR

Using the “TwoSampleMR” package, we analyzed a druggable protein studied by Zheng et al. (15). The exposure factor was SNP, and the outcome was sepsis. The SNPs used as IVs for causal inference must adhere to the three core assumptions of MR: (I) association with the exposure; (II) no confounding with the outcome; (III) no pleiotropic effects independent of the exposure. A Wald ratio evaluation was conducted for exposures containing a single SNP, while an inverse variance weighting (IVW) evaluation was conducted for exposures containing two or more SNPs. The IVW evaluation could synthesize causal effects across multiple SNPs, mitigates confounding bias, and enhances the accuracy of causal effect estimation. The causal effect was quantified using odds ratio (OR), with statistical significance assessed via two-sided P values. To address multiple testing, the Bonferroni correction was applied to adjust P values. A heterogeneity test, pleiotropy test, and leave-one-out analysis were run using TwoSampleMR, followed by a Steiger direction test to determine causality. Heterogeneity testing was performed using Cochran’s Q test, with the I² statistic as the measure of heterogeneity, calculated as I²=(q−df)/Q×100%, where I²≤0 indicates mild heterogeneity, I² of 0–25% indicates moderate heterogeneity, and I²>50% indicates high heterogeneity.

After selecting the proteins with a significant causal relationship with sepsis, the pQTL data of the corresponding proteins were downloaded from the deCODE database as the exposure factors (with sepsis as the outcome), and a two-sample MR analysis was then performed. For exposures containing only one SNP, the Wald ratio method was used to evaluate the MR results. As part of the directionality test, the Steiger direction test was used to determine whether the direction of causality was correct using the “TwoSampleMR” package.

SMR analysis

In the SMR, GWAS summary data from GWAS and QTL studies are used to test for pleiotropic associations between complex traits of interest and base protein expression levels (19). Colocalization signals are assessed for horizontal pleiotropy using the HEIDI test (20). The null hypothesis of the HEIDI test is that there is no horizontal pleiotropy in the colocalization signal. The SMR and HEIDI methods can be used to determine whether the effect of SNPs on phenotypes is mediated by protein expression rather than other pathways. To perform the SMR analysis, we downloaded SMR Linux (version 1.3.1), using the default parameters, from the SMR website (http://yanglab.westlake.edu.cn/software/smr).

Positioning analysis

A colocalization analysis was performed using the “coloc” package. The following five hypotheses related to exclusivity were assessed using the “coloc” package through a Bayesian approach: (I) SNPs are not associated with pQTL or sepsis; (II) SNPs are associated with pQTL; (III) SNPs and sepsis are related; (IV) SNPs are associated with both pQTL and sepsis; however, the associated SNPs are independent; and (V) SNPs are common to both pQTL and sepsis. A posterior probability was calculated for each of the following tests: H0, H1, H2, H3, and H4. To assess the posterior probability of shared variation, we gathered all the SNPs positioned within 250 kb upstream and downstream of the top SNP for each chosen protein, facilitating the colocalization analysis. A PH4 value >0.8 was considered a signal of colocalization between the GWAS and pQTL.

Analysis of drug targets

A PPI network consists of individual proteins that interact with one another (21). In this study, we used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database, setting the biological species to human (21). We identified proteins interacting with druggable proteins, using a minimum correlation coefficient of greater than 0.900 as the criterion. We then constructed the PPI network and visualized the PPI network model using the “igraph” and “ggraph” R packages. After that, we searched the DrugBank (22) to obtain all drugs and their mechanisms of action corresponding to the proteins involved in the PPI network. The selected drugs could be used for the treatment of later-stage sepsis.

Transcriptome differential expression analysis and prognostic correlation analysis

We used the “GEOquery” package to download sepsis-related datasets, including GSE65682, GSE54514, and GSE154918, from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) (23-26). The datasets were all derived from Homo sapiens. The GSE65682 data platform was GPL13667, and blood samples from 760 sepsis patients and 42 healthy controls were selected for analysis. The GSE54514 data platform was GPL6947, and blood samples from 127 sepsis patients and 36 healthy controls were selected for analysis. The GSE154918 data platform was GPL20301, and blood samples from 53 sepsis patients and 40 healthy controls were selected for analysis. The “limma” package (27) was used to standardize the GSE65682, GSE54514, and GSE154918 datasets. The Wilcon-test was used to examine whether there were significant differences in the key genes between the sepsis and normal groups in the three datasets, and ggplot2 was used to construct group comparison maps. To explore the prognostic value of the key genes, the “survival” package was used to draw Kaplan-Meier (K-M) curves of the sepsis samples containing prognostic data from the GSE65682 dataset.

Downloading and cleaning of single-cell data

We also downloaded the single-cell dataset of sepsis patients, GSE175453, from the GEO database using the “GEOquery” package (28). The dataset was sequenced on two platforms, Illumina NextSeq 500 (Homo sapiens) and Illumina NovaSeq 6000 (Homo sapiens). A total of nine human samples (four from sepsis patients and from five healthy controls) were sequenced. All of them were included in the single-cell analysis.

To process the single-cell data, the “Seurat” R package (29) was used to create single-cell data as Seurat objects, and the proportion of mitochondrial genes in each cell was calculated using the “PercentageFeatureSet” function. It is generally accepted that a high proportion of mitochondrial genes in a cell may indicate apoptosis or lysis, so we filtered out cells with ≥20% mitochondrial gene content. Additionally, since low-quality cells or empty droplets typically have very few genes and doublet cells may show an abnormally high number of genes, we excluded cells with fewer than 250 features. We also filtered the unique molecular identifiers (UMIs) in cells with fewer than 300 genes, and the number of genes with each UMI detected after log10 transformation was greater than 0.8 cells. Finally, we also filtered out genes that were expressed in fewer than 10 cells. Subsequent data normalization, a search for hypervariable genes, and dimensionality reduction clustering were performed according to the default parameters and standard procedures of the “Seurat” package. We standardized and corrected the data through SCT and integrated the data of different samples through the “Harmony” R package (30).

After completing the quality control procedure, we performed linear dimension reduction on the Seurat object and calculated the principal components (PCs) using the genes with the most variable expression in the dataset (31). We then used Seurat’s “FindNeighbors” and “FindClusters” functions to group the cells into the optimal number of clusters for cell type identification. By applying uniform manifold approximation and projection (UMAP) to reduce the information captured in the selected PCs to two dimensions, we achieved graph-based visual clustering of the cells.

Cell annotation

For the Seurat single-cell data, UMAP was used to visualize the clustering results. Subsequently, we combined the automatic annotation provided by the “SingleR” R package, together with peripheral blood cell marker genes obtained from the CellMarker 2.0 online website (http://bio-bigdata.hrbmu.edu.cn/CellMarker/), and marker genes provided in the original dataset literature, to annotate the cell clusters (28). Ultimately, 12 cell types were annotated. A violin diagram was drawn to verify the rationality of marker gene expression and grouping. CD14 was used as a marker gene for monocytes. PPBP was used as a marker gene for platelets. FCGR3A and MS4A7 were used as the marker genes for CD16⁺ monocytes. CD3D was used as the marker gene for T cells. GNLY and KLRF1 were used as the marker genes for natural killer (NK) cells. LCN2 was used as the marker gene for neutrophils. CD1C and CLEC10A were used as the marker genes for myeloid dendritic cells. MS4A1 was used as the marker gene for B cells. CD38 was used as the marker gene for plasma cells. LILRA4 was used as the marker gene for plasmacytoid dendritic cells. HBB and HBA1 were used as the marker genes for erythrocytes. CD34 was used as the marker gene for hematopoietic stem cells. Finally, the differentially expressed marker genes and key genes of all cell types among disease controls in different cell types were visualized by bubble plots.

Pseudo-time series analysis

To further analyze the differentiation status among the cell populations, the classical “monocle2” package was used to perform a quasi-temporal analysis of the cell populations. The single-cell data were processed through the construction of a Monocle object, normalization, and the filtering of low-quality cells. Highly variable genes were selected for dimensionality reduction using the DDRTree method. Subsequently, pseudo-temporal analyses of different cell types and significant genes were performed based on the sorted data.

Statistical analysis

All the calculations and statistical analyses in this study were executed using R software (version 4.2.2, https://www.r-project.org/). Unless stated otherwise, a P value <0.05 was considered statistically significant.

Results

MR analysis of pharmaceutically available proteins

We initially intersected the 734 proteins identified by Zheng et al. (15) with the 4,479 proteins encoded by druggable genes, resulting in 511 druggable gene-encoded proteins. We then performed a two-sample MR analysis of the 511 proteins and sepsis-related proteins via TwoSampleMR. A P value <0.05 was set as the significant causal screening condition for the initial screening. As detailed in Table 1, 29 proteins were found to have a causal relationship with sepsis. The OR can be used to differentiate whether each of the 29 proteins is a risk factor or protective factor for sepsis. An OR >1 indicates that the protein promotes sepsis occurrence, while an OR <1 suggests it lowers the risk of sepsis.

Because each of the 29 proteins had only one SNP, a subsequent sensitivity analysis was not feasible. We downloaded the pQTL files for these 29 proteins from the deCODE database for further analysis. First, we screened these pQTL files according to the cis-pQTL criteria, identifying the corresponding cis-pQTL for the 29 proteins. We then conducted a two-sample MR analysis of the associations between these proteins and sepsis using the “TwoSampleMR” package. For the second screening, we applied the more stringent Bonferroni correction for P value adjustment, using a corrected P adjusted value <0.05 as the criterion for significant causality to identify proteins with strong causal links to sepsis. As Table 2 shows, two proteins, CD33 and LY9, were found to have a causal relationship with sepsis, and both were positively correlated with the risk of sepsis. Ultimately, we created a scatter plot illustrating the effect estimates of various MR models involving these two proteins and sepsis (Figure 2A,2B). As Figure 2 shows, the y-axis intercept of each model line on the ordinate is nearly zero, which implies that when the SNP effect on both proteins is zero, the predicted SNP effect on sepsis also approaches zero. Meanwhile, the consistently positive slpe across all analytical models demonstrates a positive causal association between both proteins and sepsis.

Figure 2 Scatter plot of effect estimates from different models in MR analysis of protein on sepsis. (A) Scatter plot of different MR model effect estimates of CD33 on sepsis; (B) scatter plot of different MR model effect estimates of LY9 on sepsis. MR, Mendelian randomization; SNP, single nucleotide polymorphism.

Sensitivity analysis of the proteins and sepsis

Initially, we conducted a heterogeneity test for the two proteins (CD33 and LY9) and sepsis (Table 3). The results indicated that all protein heterogeneity tests had Q_p values >0.05 (Table 3), which indicated that there was no significant heterogeneity among the IVs for these proteins, and provided evidence of the reliability of the results. As depicted in the funnel plot (Figure 3A,3B), the IVs of the two proteins were evenly distributed on both sides of the IVW line, with no apparent heterogeneity, aligning with the calculated outcomes.

Figure 3 Funnel plot of heterogeneity test for MR analysis of proteins for sepsis. (A) Funnel plot of CD33 and sepsis MR; (B) funnel plot of LY9 and sepsis MR. MR, Mendelian randomization; SE, standard error.

Next, we performed a pleiotropy test on the two proteins (CD33 and LY9) and sepsis (Table 4). As detailed in Table 4, all the pleiotropic protein test P values were >0.05, and the intercepts were close to zero, which showed that the causal inferences of this study were not affected by horizontal pleiotropic effects.

We then conducted a leave-one-out sensitivity analysis test. As Figure S1 clearly shows, for both proteins (CD33 and LY9), the beta values of all the results from the leave-one-out tests for each SNP were on one side of zero, and had a high degree of overlap. This indicates that individual SNPs had a minimal effect on the outcomes of MR, suggesting that the results are stable.

To ensure that proteins influence the pathogenesis of sepsis causality in the right direction, we further used the Steiger directionality test for analysis, and found that the P values for the two proteins and sepsis were much less than 0.05, indicating that the direction was correct (Table 5).

SMR analysis and colocalization analysis

The HEIDI test in the SMR analysis was used to further verify the presence of pleiotropy. As detailed in Table 6, the P_HEIDI values for LY9 and CD33 were both >0.05, indicating the absence of pleiotropy among the SNPs of these two proteins. Additionally, the P_SMR values from the SMR analysis were <0.05, which provided further confirmation of the causal relationship between these two proteins and sepsis.

The colocalization analysis (Table 7) revealed that no protein exhibited a strong colocalization with sepsis (PP.H4 >0.8), and no protein exhibited a moderate colocalization with sepsis (0.5< PP.H4 <0.8).

Drug-target analysis

We used the STRING database to expand the PPI analysis of the two druggable targets (CD33 and LY9). After retaining targets with connections to other nodes, we constructed a PPI network consisting of seven druggable-related targets (CD33, LY9, ANPEP, FCGR3B, FCGR3A, TREM2, and SH2D1A) (Figure 4). We analyzed potential drugs for target proteins using the seven druggable-related targets in the DRUGBANK database. As Table 8 shows, proteins encoded by TREM2, LY9, and SH2D1A had no corresponding drugs available, which warrants further exploration. While the protein encoded by CD33 corresponded to two drugs (DB00056 and DB06318); the protein encoded by FCGR3B corresponded to 12 drugs (DB00092, DB00002, DB00005, DB00028, DB00056, DB00075, DB00087, DB00108, DB00110, DB00111, DB11767, and DB06607); the protein encoded by FCGR3A corresponded to 13 drugs (DB00002, DB00005, DB00028, DB00056, DB00087, DB00092, DB00110, DB00111, DB11767, DB12023, DB06607, DB00112, and DB16695); and the protein encoded by ANPEP corresponded to four drugs (DB00973, DB06196, DB06773, and DB16627).

Figure 4 Analysis of PPI interaction network. The nodes in the figure represent the targets, and the color of the line from blue to purple indicates the degree of correlation between the nodes from small to large. PPI, protein-protein interaction.

Transcriptome differential expression analysis and prognostic correlation analysis

To investigate the differences in the expression of the two druggable targets (CD33 and LY9) between the sepsis group and the normal group, we plotted group comparison graphs for three datasets (GSE65682, GSE54514, and GSE154918) (Figure 5A-5C). The graphs showed that the expression trends of CD33 and LY9 were consistent, and there were significant differences in the GSE65682 and GSE154918 datasets (P<0.05). Compared to the normal group, CD33 was highly expressed in the sepsis group, while LY9 was lowly expressed. Further, we examined the prognostic significance of the two druggable targets (CD33 and LY9) in the GSE65682 dataset. K-M survival curves were plotted for CD33 and LY9 (Figure 5D,5E), and the results indicated that the expression level of LY9 had prognostic significance (P<0.05). Compared with the low-expression group, the high-expression group of LY9 had a greater survival probability and longer survival time.

Figure 5 Transcriptome differential expression analysis and prognostic KM curves. (A) Group comparison diagram of patentable targets between the sepsis and normal groups in the GSE54514 dataset; (B) druggable targets for sepsis that can be identified from GSE65682 dataset and the comparison charts between the control and the sepsis groups; (C) group comparison diagram of druggable targets between the sepsis and normal groups in the GSE154918 dataset; (D) the prognostic KM curve of druggale targets CD33 in GSE65682; (E) the prognostic KM curve of druggale targets LY9 in GSE65682. *, P<0.05; ***, P<0.001. KM, Kaplan-Meier.

Single-cell analysis

We performed UMAP dimensionality reduction clustering with the resolution set to 1.4 and obtained 36 cell clusters (Figure 6A). Subsequently, we manually annotated 12 cell types (i.e., monocytes, platelets, CD16⁺ monocytes, T cells, NK cells, neutrophils, myeloid dendritic cells, B cells, plasma cells, plasmacytoid dendritic cells, erythrocytes, and hematopoietic stem cells) via cell marker gene annotation (Figure 6B). Bubble plots were generated to display the marker genes of the 12 cell types (Figure 6C). Finally, we presented the differences in the expression of the two druggable targets (CD33 and LY9) between the sepsis group and the normal group across different cell types using bubble plots (Figure 6D). The plot showed that CD33 was mainly highly expressed in myeloid dendritic cells and monocytes in the sepsis group, while LY9 was mainly highly expressed in plasmacytoid dendritic cells, plasma cells, and T cells in the sepsis group.

Figure 6 GSE175453 single-celled data analysis. (A) The UMAP clustering figure with a resolution of 1.4, including 36 distinct cell clusters; (B) UMAP dimensionality reduction clustering map after manual annotation; (C) the expression of specific markers of 12 different cell types; the color depth represents the level of gene expression, and the size of the dot represents the proportion of the gene in the cell; (D) the expression of the drug target in different cell types in the sepsis group and normal group; the color represents the level of gene expression; and the size of the dot represents the proportion of the gene in the cell. UMAP, uniform manifold approximation and projection.

To further investigate the differentiation status among the cell clusters and the changes in the expression of druggable targets at different time points, we first conducted a pseudo-temporal analysis of all cell clusters (Figure 7A-7C); Figure 7A shows the distribution of all cells across time periods; Figure 7B presents 5 pseudo-time stages; and Figure 7C illustrates the color gradient from dark to light indicating the time progression, and showed that the cellular differentiation time sequence was from stages 1-5-2-3-4. Subsequently, we generated heatmaps to depict the changes in the expression of CD33 and LY9 across the pseudo-time stages (Figure 7D). CD33 was highly expressed in the mid pseudo-time stage, which was consistent with its high expression in myeloid dendritic cells and monocytes, which were also highly expressed in the mid pseudo-time stage. Conversely, LY9 exhibited high expression in the late pseudo-time stage, which was consistent with its high expression in plasmacytoid dendritic cells, plasma cells, and T cells, which were mainly expressed in the late pseudo-time stage.

Figure 7 Single-cell pseudo-timing analysis of GSE175453. (A) Differentiation trajectories of different cell types (different colors indicate different cell types); (B) changes in cells at different stages over time (different colors indicate different time stages); (C) temporal analysis diagram of cells (colors from dark to light indicate temporal order); (D) pseudo-temporal dynamic heatmap of the patentable targets over time (red indicates up-regulated expression; blue indicates down-regulated expression).

Discussion

In this study, we first performed a comprehensive preliminary assessment of the relationship between the proteins encoded by 511 druggable genes and sepsis, which yielded a total of 29 proteins that were causally associated with sepsis. Subsequently, we performed a more rigorous validation of these 29 proteins using the deCODE database to further identify proteins with strong causal associations with sepsis. We found strong causal positive associations between CD33 and LY9, and the risk of developing sepsis. Subsequent colocalization analyses of these two proteins did not reveal a strong colocalized relationship (PP.H4 >0.8), but both passed the SMR and HEIDI tests. Thus, our study identified two proteins (CD33 and LY9) with convincing evidence. We then verified the differential expression of these two proteins between different cells in the normal group and sepsis group and between cells at different time points. Finally, we analyzed these two proteins, and found that two drugs have been developed against the protein encoded by CD33; these two drugs are currently mainly used for the treatment of leukemia. Conversely, no drugs have been developed for the protein encoded by LY9, which may be a possible new drug development target.

We identified a causal relationship between the two proteins (CD33 and LY9) and the incidence of sepsis by analyzing previously published genomics and proteomics-related articles. CD33 (Siglec-3) is a 67 kDa glycosylated type 1 transmembrane protein belonging to the salivary acid-binding Ig-like lectin (Siglec) family (32). In the brain, CD33 is predominantly expressed on the surface of microglia and is considered one of the most relevant factors for Alzheimer’s disease (AD) risk (33). Increased CD33 expression is strongly associated with an increased susceptibility to AD (34). The results of a MR study also suggested that the elevated peripheral expression of CD33 is causally associated with the development of AD, and may be a potential target for the treatment of AD (35). In addition, CD33 plays an important role in acute myeloid leukemia (AML) (36). CD33 is expressed on the surface of more than 80% of AML cells with a high average antigenic density, and it is generally believed that CD33 is not expressed outside the hematopoietic system. The best known clinically available CD33-targeted immunotherapy is the anti-CD33 antibody gemtuzumab ozogamicin (37), which was approved for use in 2000, and is one of the most thoroughly studied targeted therapies in adults. The immunocouple against CD33, AVE9633 (38), is a potent microtubule protein inhibitor. Continued development has been discontinued in phase 1 clinical trials due to a loss of activity when the drug is used at apparently sufficient doses.

LY9, also known as CD229, is a member of the signaling lymphocyte activation family (SLAM) of immune receptors (39). LY9 acts as a co-signaling molecule, is mainly expressed on the surface of B and T cells, and regulates lymphocyte homeostasis and activation (40). Studies showed that LY9 is strongly overexpressed in multiple myeloma (MM) and most marginal zone lymphomas, and it is expected to be a biological marker and potential therapeutic target for MM (41,42). While information about effective drugs targeting LY9 is scarce, Radhakrishnan et al. have pioneered the development of CD229 chimeric antigen receptor (CAR) T cells. These cells have shown high activity both in vitro and in vivo against MM plasma cells, memory B cells, and MM propagating cells. Hence, they hold promise as a potentially effective therapeutic approach (43).

In this study, we also performed a differential transcriptome expression analysis and prognostic analysis. High or low LY9 expression was significantly associated with survival, while no significant association between survival and CD33 expression was found. Interestingly, the results of the MR analysis revealed a positive correlation between LY9 expression and the risk of developing sepsis, which contradicted the results of the prognostic analysis. According to the MR analysis results, LY9 is indeed a risk factor for sepsis, but the odds ratio value ranged between 1.05 and 1.2 in Zheng et al.’s study (15) and Deocde, which suggests that the effect of LY9 may be reversed by abnormalities in a few SNPs. MR analysis relies on the effect of the genetic variation on the relationship between the exposure and outcome. There may be a non-linear relationship between LY9 and sepsis (rather than a simple direct linear relationship), which may be due to unconsidered biological mechanisms or complex associations. The relationship may also be influenced by differences in samples. Notably, the samples selected for the MR analysis were all from European populations, while the transcriptome samples were not necessarily all from European populations. Nonetheless, LY9 is still of research value, and the question of whether it acts as a protective factor or a risk factor needs to be further clarified.

The strength of our study lies in the fact that we systematically investigated the association between plasma protein biomarkers and the risk of sepsis by using a two-stage proteome-wide MR design, which has the advantages of a large sample size, broad proteomic coverage, reverse causality, and a minimal risk of confounding. The credibility of the results was confirmed by analyzing the results of several studies (35,41). Transcriptome analyses, PPIs, and druggability assessments also provided insights into the candidate proteins involved in the pathogenesis of sepsis and further ranked the druggable proteins. Sepsis has plagued human health for centuries, and deciphering its underlying mechanisms is difficult. CD33 has paradoxical roles in sepsis (44), and targeting CD33 for regulation is not always beneficial; therefore, more research needs to be conducted to examine its role in inflammation. Despite the lack of validated drug information and sepsis-related studies on LY9, CD229 CAR T cells have shown good therapeutic effects against MM. Therefore, studies targeting sepsis are also valuable.

Our study had several limitations. First, our MR analysis was limited to a European population only; thus, further research needs to be conducted to determine whether the results are applicable to other populations. Second, the results of our MR analysis contradicted the results of the prognostic analysis, but we investigated all sepsis datasets with prognostic results in the current database. Because it is a non-tumor disease, there are fewer relevant available datasets for sepsis, which should be expanded and explored by future researchers. Third, only cis-SNPs were included in our analysis, and while cis-pQTL can directly regulate the expression level of proteins by directly affecting their transcription, translation, degradation, stability, or activity (45), trans-pQTL could help to expand the understanding of the relationships among proteins, diseases, and disease etiology (46). Finally, plasma proteins may also be influenced by factors other than genetics. Future epidemiologic studies of measured plasma protein levels and sepsis risk are needed to validate our findings.

Conclusions

We identified 29 proteins by GWAS and MR that may be causally related to sepsis and have a reasonable biological basis. By combining other analytical methods, our studies identified CD33 and LY9 as potential therapeutic targets for sepsis, provided novel insights for the treatment of sepsis. Further studies need to be conducted to provide clinical and experimental evidence to assess the utility and effectiveness of these drugs.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE-MR reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-590/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-590/prf

Funding: This work was supported by the National Natural Science Foundation of China (Nos. 82372155, 82000085 and 82100090) and sponsored by Development Fund for the Department of Anesthesiology, Shanghai Pulmonary Hospital and Project of Key Supported Disciplines by Shanghai Municipal Health Commission (No. 2023ZDFC0204).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-590/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Martin-Loeches I, Singer M, Leone M. Sepsis: key insights, future directions, and immediate goals. A review and expert opinion. Intensive Care Med 2024;50:2043-9. [Crossref] [PubMed]
2. Cajander S, Kox M, Scicluna BP, et al. Profiling the dysregulated immune response in sepsis: overcoming challenges to achieve the goal of precision medicine. Lancet Respir Med 2024;12:305-22. [Crossref] [PubMed]
3. Fleischmann C, Scherag A, Adhikari NK, et al. Assessment of Global Incidence and Mortality of Hospital-treated Sepsis. Current Estimates and Limitations. Am J Respir Crit Care Med 2016;193:259-72. [Crossref] [PubMed]
4. Slim MA, van Mourik N, Bakkerus L, et al. Towards personalized medicine: a scoping review of immunotherapy in sepsis. Crit Care 2024;28:183. [Crossref] [PubMed]
5. Wang Y, Wu J, Shao T, et al. Prognostic impact of early versus delayed loop diuretic administration in sepsis: a propensity score-matched analysis using the MIMIC-IV database. Transl Androl Urol 2025;14:779-90. [Crossref] [PubMed]
6. Yu Z, Qian YY. Aspirin use is associated with the reduced mortality risk in chronic obstructive pulmonary disease with sepsis: a retrospective study using the MIMIC-IV database. J Thorac Dis 2024;16:6688-98. [Crossref] [PubMed]
7. Zhou Y, Ren D, Chen Y, et al. Presepsin, procalcitonin, interleukin-6, and high-sensitivity C-reactive protein for predicting bacterial DNAaemia among patients with sepsis. J Thorac Dis 2025;17:991-1001. [Crossref] [PubMed]
8. Tan MCB, Isom CA, Liu Y, et al. Transcriptome-wide association study and Mendelian randomization in pancreatic cancer identifies susceptibility genes and causal relationships with type 2 diabetes and venous thromboembolism. EBioMedicine 2024;106:105233. [Crossref] [PubMed]
9. Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol 2013;37:658-65. [Crossref] [PubMed]
10. Sun J, Zhao J, Jiang F, et al. Identification of novel protein biomarkers and drug targets for colorectal cancer by integrating human plasma proteome with genome. Genome Med 2023;15:75. [Crossref] [PubMed]
11. Peng X, Li Y, Liu N, et al. Plasma Proteomic Insights for Identification of Novel Predictors and Potential Drug Targets in Atrial Fibrillation: A Prospective Cohort Study and Mendelian Randomization Analysis. Circ Arrhythm Electrophysiol 2024;17:e013037. [Crossref] [PubMed]
12. Zhao JH, Stacey D, Eriksson N, et al. Genetics of circulating inflammatory proteins identifies drivers of immune-mediated disease risk and therapeutic targets. Nat Immunol 2023;24:1540-51. [Crossref] [PubMed]
13. van Vugt M, Finan C, Chopade S, et al. Integrating metabolomics and proteomics to identify novel drug targets for heart failure and atrial fibrillation. Genome Med 2024;16:120. [Crossref] [PubMed]
14. Finan C, Gaulton A, Kruger FA, et al. The druggable genome and support for target identification and validation in drug development. Sci Transl Med 2017;9:eaag1166. [Crossref] [PubMed]
15. Zheng J, Haberland V, Baird D, et al. Phenome-wide Mendelian randomization mapping the influence of the plasma proteome on complex diseases. Nat Genet 2020;52:1122-31. [Crossref] [PubMed]
16. Wang QS, Hasegawa T, Namkoong H, et al. Statistically and functionally fine-mapped blood eQTLs and pQTLs from 1,405 humans reveal distinct regulation patterns and disease relevance. Nat Genet 2024;56:2054-67. [Crossref] [PubMed]
17. Ferkingstad E, Sulem P, Atlason BA, et al. Large-scale integration of the plasma proteome with genetics and disease. Nat Genet 2021;53:1712-21. [Crossref] [PubMed]
18. Hemani G, Zheng J, Elsworth B, et al. The MR-Base platform supports systematic causal inference across the human phenome. Elife 2018;7:e34408. [Crossref] [PubMed]
19. Zhu Z, Zhang F, Hu H, et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet 2016;48:481-7. [Crossref] [PubMed]
20. Sun X, Chen B, Qi Y, et al. Multi-omics Mendelian randomization integrating GWAS, eQTL and pQTL data revealed GSTM4 as a potential drug target for migraine. J Headache Pain 2024;25:117. [Crossref] [PubMed]
21. Szklarczyk D, Gable AL, Lyon D, et al. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res 2019;47:D607-13. [Crossref] [PubMed]
22. Wishart DS, Feunang YD, Guo AC, et al. DrugBank 5.0: a major update to the DrugBank database for 2018. Nucleic Acids Res 2018;46:D1074-82. [Crossref] [PubMed]
23. Davis S, Meltzer PS. GEOquery: a bridge between the Gene Expression Omnibus (GEO) and BioConductor. Bioinformatics 2007;23:1846-7. [Crossref] [PubMed]
24. Scicluna BP, Klein Klouwenberg PM, van Vught LA, et al. A molecular biomarker to diagnose community-acquired pneumonia on intensive care unit admission. Am J Respir Crit Care Med 2015;192:826-35. [Crossref] [PubMed]
25. Parnell GP, Tang BM, Nalos M, et al. Identifying key regulatory genes in the whole blood of septic patients to monitor underlying immune dysfunctions. Shock 2013;40:166-74. [Crossref] [PubMed]
26. Herwanto V, Tang B, Wang Y, et al. Blood transcriptome analysis of patients with uncomplicated bacterial infection and sepsis. BMC Res Notes 2021;14:76. [Crossref] [PubMed]
27. Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 2015;43:e47. [Crossref] [PubMed]
28. Darden DB, Dong X, Brusko MA, et al. A Novel Single Cell RNA-seq Analysis of Non-Myeloid Circulating Cells in Late Sepsis. Front Immunol 2021;12:696536. [Crossref] [PubMed]
29. Gribov A, Sill M, Lück S, et al. SEURAT: visual analytics for the integrated analysis of microarray data. BMC Med Genomics 2010;3:21. [Crossref] [PubMed]
30. Korsunsky I, Millard N, Fan J, et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods 2019;16:1289-96. [Crossref] [PubMed]
31. Groth D, Hartmann S, Klie S, et al. Principal components analysis. Methods Mol Biol 2013;930:527-47. [Crossref] [PubMed]
32. Kukan EN, Fabiano GL, Cobb BA. Siglecs as modulators of macrophage phenotype and function. Semin Immunol 2024;73:101887. [Crossref] [PubMed]
33. Eskandari-Sedighi G, Jung J. Macauley MS. CD33 isoforms in microglia and Alzheimer's disease: Friend and foe. Mol Aspects Med 2023;90:101111. [Crossref] [PubMed]
34. Eskandari-Sedighi G, Crichton M, Zia S, et al. Alzheimer's disease associated isoforms of human CD33 distinctively modulate microglial cell responses in 5XFAD mice. Mol Neurodegener 2024;19:42. [Crossref] [PubMed]
35. Gu X, Dou M, Cao B, et al. Peripheral level of CD33 and Alzheimer's disease: a bidirectional two-sample Mendelian randomization study. Transl Psychiatry 2022;12:427. [Crossref] [PubMed]
36. Appelbaum J, Price AE, Oda K, et al. Drug-regulated CD33-targeted CAR T cells control AML using clinically optimized rapamycin dosing. J Clin Invest 2024;134:e162593. [Crossref] [PubMed]
37. Collados-Ros A, Muro M, Legaz I. Gemtuzumab Ozogamicin in Acute Myeloid Leukemia: Efficacy, Toxicity, and Resistance Mechanisms-A Systematic Review. Biomedicines 2024;12:208. [Crossref] [PubMed]
38. Lapusan S, Vidriales MB, Thomas X, et al. Phase I studies of AVE9633, an anti-CD33 antibody-maytansinoid conjugate, in adult patients with relapsed/refractory acute myeloid leukemia. Invest New Drugs 2012;30:1121-31. [Crossref] [PubMed]
39. Farhangnia P, Ghomi SM, Mollazadehghomi S, et al. SLAM-family receptors come of age as a potential molecular target in cancer immunotherapy. Front Immunol 2023;14:1174138. [Crossref] [PubMed]
40. Margraf S, Garner LI, Wilson TJ, et al. A polymorphism in a phosphotyrosine signalling motif of CD229 (Ly9, SLAMF3) alters SH2 domain binding and T-cell activation. Immunology 2015;146:392-400. [Crossref] [PubMed]
41. Roncador G, Puñet-Ortiz J, Maestre L, et al. CD229 (Ly9) a Novel Biomarker for B-Cell Malignancies and Multiple Myeloma. Cancers (Basel) 2022;14:2154. [Crossref] [PubMed]
42. Setayesh SM, Ndacayisaba LJ, Rappard KE, et al. Targeted single-cell proteomic analysis identifies new liquid biopsy biomarkers associated with multiple myeloma. NPJ Precis Oncol 2023;7:95. [Crossref] [PubMed]
43. Radhakrishnan SV, Luetkens T, Scherer SD, et al. CD229 CAR T cells eliminate multiple myeloma and tumor propagating cells without fratricide. Nat Commun 2020;11:798. [Crossref] [PubMed]
44. Royster W, Wang P, Aziz M. The Role of Siglec-G on Immune Cells in Sepsis. Front Immunol 2021;12:621627. [Crossref] [PubMed]
45. Lin Z, Pan W. A robust cis-Mendelian randomization method with application to drug target discovery. Nat Commun 2024;15:6072. [Crossref] [PubMed]
46. Wen S, Xu S, Zong X, et al. Association Analysis of the Circulating Proteome With Sarcopenia-Related Traits Reveals Potential Drug Targets for Sarcopenia. J Cachexia Sarcopenia Muscle 2025;16:e13720. [Crossref] [PubMed]