Targeting capillary leak in experimental sepsis and endotoxemia: a systematic review of therapeutic agents, methods and models in animal studies

Abstract

Background

Sepsis-induced capillary leak is a major contributor to circulatory dysfunction and tissue edema, yet remains therapeutically unaddressed: numerous interventions showing promise in preclinical models have failed to demonstrate clinical efficacy. This systematic review aimed to characterise preclinical studies evaluating therapeutic interventions specifically targeting capillary leak and to identify candidates with the highest potential for clinical translation.

Methods

A systematic search of PubMed, Embase, and Web of Science identified in vivo studies in experimental sepsis and endotoxemia evaluating interventions targeting capillary leak. Data on animal models, sepsis induction methods, capillary leak assessment, therapeutic targets, and intervention timing were extracted. Risk of bias was assessed using the SYRCLE tool. As a complementary step, all retained interventions were cross-referenced against randomized or prospective clinical studies in sepsis published since January 2015 and appraised using a four-criterion translational quality score (reproducibility, quality of the model, clinically relevant timing, and multimodal read-out).

Results

Three hundred and three preclinical studies were included, exhibiting substantial heterogeneity in both models and outcomes. Rodent models predominated (81.2%), and lipopolysaccharide-induced endotoxemia was the most common induction method (59.7%). Dye-based assays constituted the most frequent capillary leak read-out in rodents (81.7%), whereas fluid balance and lymph-to-plasma ratios predominated in large-animal studies. The lung was the most frequently evaluated organ. Endothelial activation (12.2%), intercellular junctions (11.2%), and the nitric oxide pathway (8.9%) were the most targeted pathways. A beneficial effect of the intervention on at least one capillary leak outcome results was reported in 93.1% of studies, raising substantial concerns regarding publication bias. Of 232 distinct interventions, only a minority had been evaluated in randomized clinical trials since 2015, and no capillary leak-specific therapy has yet reached high translational priority.

Conclusions

Preclinical research on sepsis-induced capillary leak is characterized by methodological heterogeneity, a disproportionately high rate of studies reporting capillary leak reduction, and an overreliance on endotoxemia models. Bridging the translational gap will require standardized assessment of capillary leak, broader use of live-pathogen large-animal models, post-treatment dosing paradigms, and clinical trials enriched for endothelial dysfunction.

Registration

International Prospective Register of Systematic Reviews (PROSPERO) number CRD42024522372.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13054-026-06160-0.

Introduction

Sepsis is defined as a life-threatening organ dysfunction caused by a dysregulated host response to infection [1]. It is a severe and frequent condition in the intensive care unit (ICU), with a high mortality rate [2]. The pathophysiology of circulatory dysfunction in sepsis is intricate and remains poorly understood. A defining feature of circulatory failure in septic shock is capillary leak, which accounts for both the hypovolemia observed in the initial phase of septic shock and the subsequent development of fluid overload. The mechanisms implicated in capillary leak (i.e. transcapillary flow) can be elucidated using Starling’s revised equation [3]. First, we observe an increase in vascular permeability due to endothelial barrier dysfunction. This involves the disassembly of intercellular (tight and adherens) junctions, shedding of the glycocalyx layer, and activation of endothelial cells towards pro-inflammatory, pro-adhesive, or pro-thrombotic phenotypes in response to inflammatory mediators, such as damage- or pathogen-associated molecular patterns (DAMPs, PAMPs) [4, 5]. Second, the transcapillary pressure gradient is augmented by a reduction in interstitial pressure, thereby enhancing capillary flow [6]. Collectively, these phenomena, along with coagulation activation and microthrombi formation, contribute to microcirculatory dysfunction.

Although sepsis-induced capillary leak has been extensively investigated in animal studies with promising outcomes, the majority of promising interventions have failed to translate into clinical practice [7]. Our current approach remains limited to addressing hypovolemia caused by capillary leak through the administration of large volume of fluid, a strategy that compounds the very problem it is intended to solve, as infused fluid rapidly extravasates into already oedematous interstitial compartments. Numerous studies have demonstrated the detrimental effects of fluid overload and its association with adverse outcomes has been well documented [8–10]. Identifying therapies that specifically target the capillary leak mechanism itself is therefore a pressing clinical priority. To address the gap in translation between preclinical and clinical settings, we conducted a systematic review of preclinical studies evaluating therapeutic interventions aimed at mitigating capillary leak in sepsis models with three objectives: (1) to describe the types of models, targeted pathways, and methods used to assess capillary leak; (2) to appraise the methodological quality of the preclinical literature; and (3) to identify candidates with the highest potential for clinical translation.

Materials and methods

Review protocol

We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines and recommendations from the Cochrane Collaboration. The review protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO, CRD42024522372).

Literature search

The following queries were used (“sepsis” OR “septic shock” OR “endotoxemia”) AND (“vascular permeability” OR “capillary leak” OR “fluid balance” OR “vascular leak”) in three electronic bibliographic databases: PubMed, Embase, and Web of Science. The search was limited to articles published until March 15, 2024. Articles retrieved from the databases were imported into Zotero, where duplicate articles were identified and excluded. The remaining articles were then imported into the Rayyan platform (https://www.rayyan.ai/) and independently screened by two reviewers (SHL, HD). Any disagreement on article selection was resolved through discussion with a third reviewer (AD).

Inclusion and exclusion criteria

The inclusion criteria were as follows: (1) all in vivo studies conducted in animals with experimental sepsis induction; (2) any therapeutic intervention, whether pharmacological or non-pharmacological (including physical or procedural approaches); (3) studies including a control group; and (4) studies with capillary leakage assessment outcomes, such as histological analysis, plasma or local glycocalyx shedding products, or vascular permeability biomarkers, net fluid balance, wet-to-dry organ ratio, body electrical impedance, dynamic contrast-enhanced magnetic resonance imaging, etc.

The exclusion criteria were as follows: (1) non-original studies, such as reviews, conference abstracts, case reports, and editorials; and (2) studies conducted in humans, in vitro, in silico, or using organoids. (3) Non-therapeutic intervention studies, such as using genetically modified knockout animals, genome editing, or optogenetics; (4) studies without a control group; (5) studies without full-text availability; and (6) studies published in a language other than English.

Data extraction

For each included study, the extracted data were as follows: (1) basic information (authors’ and journals’ names, year of publication); (2) animal model characteristics (species, number of animals, and sepsis induction methods); (3) therapeutic intervention characteristics, such as name, type, target, and timing of administration relative to sepsis induction; (4) capillary leak assessment methods and evaluated organs or systems; and (5) experimental results compared with the control group.

Risk of bias assessment

The methodological quality and internal validity of each included study were appraised using the Systematic Review Centre for Laboratory animal Experimentation (SYRCLE) risk-of-bias tool, the only instrument specifically developed and validated for randomized in vivo animal studies [11]. The tool comprises ten signalling questions covering six bias domains: (1) selection bias, (2) performance bias, (3) detection bias, (4) attrition bias, (5) reporting bias, and (6) other sources of bias [12]. Each item was rated as ‘low’, ‘unclear’ or ‘high’ risk according to the original SYRCLE criteria, with ‘unclear’ attributed when the manuscript provided insufficient information to make a confident judgement, as recommended for preclinical literature. Two reviewers (SHL, HD) independently scored every included study, and any disagreement was resolved through discussion with a third reviewer (AD). The aggregated study results are provided in Additional Table S2.

Interventions with available human data and translational quality scoring

To screen the most promising candidates for clinical translation potential, we identified, among the 232 distinct interventions retained in Additional Table S1, those agents for which clinical data were available in human sepsis, septic shock or sepsis-associated acute respiratory distress syndrome (ARDS). We chose to focus on agents with human data to narrow the broad picture of the field and keep an objective criterion. Each intervention was screened against ClinicalTrials.gov and the PubMed/Embase databases. An agent was retained when at least one randomized controlled trial (RCT) or prospective interventional study could be identified in one of these indications since January 1 st, 2015, irrespective of publication status (published, ongoing, completed but not yet reported, or planned).

For each retained agent, the strength of the underlying translational evidence was then appraised against four pre-specified binary criteria, derived from the methodological weaknesses identified in our review and from established guidance on preclinical sepsis research (MQTiPSS) [13]: (1) Reproducibility—at least two independent positive studies, ideally from different research groups, reporting a beneficial effect on a capillary-leak outcome; (2) Translational model quality—at least one positive study performed in a large-animal model (pig, sheep or dog) and in a polymicrobial sepsis model (cecal ligation and puncture, fecal injection into the peritoneal cavity, or live-pathogen administration), as opposed to LPS endotoxemia alone; (3) Clinically relevant timing—at least one positive study using a post-treatment paradigm, in which the intervention was administered after sepsis induction; (4) Multimodal evaluation—a positive effect on capillary leak confirmed by at least two mechanistically distinct read-outs. Each criterion was rated as fulfilled or not on the basis of the studies referenced for the corresponding agent in our systematic review, with explicit citation of the supporting studies in Table 3.

Table 3.

Interventions with available human data and translational quality scoring

Molecules form Additional Table S1 were retained if clinical data were available in human sepsis, septic shock or sepsis-associated acute respiratory distress syndrome (ARDS) since January 2015, irrespective of publication status. To assess the quality of preclinical datas, four categories were assessed for each interventions: (1) Reproducibility — at least two independent positive studies, ideally from different research groups, reporting a beneficial effect on a capillary-leak outcome; (2) Model quality — at least one positive study performed in a large-animal model (pig, sheep or dog) and in a polymicrobial sepsis model; (3) Clinically relevant timing — at least one positive study using a post-treatment paradigm, in which the intervention was administered after sepsis induction; (4) Multimodal evaluation — a positive effect on capillary leak confirmed by at least two mechanistically distinct read-outs. Each criterion was rated as fulfilled or not on the basis of the studies referenced for the corresponding agent in our systematic review. A colour code was applied: green when all criteria were met, yellow when 3 out of 4 were fulfilled, and red when 2 or fewer criteria were satisfied. *Primary outcomes are underlined; outcomes related to capillary leak are in bold

AE: Adverse events; APACHE II: Acute Physiology and Chronic Health Evaluation II; ARDS: acute respiratory distress syndrome; BP: Blood Pressure; DPP3: dipeptidyl peptidase 3; EAA: endotoxin activity assay; EVLW: extravascular lung water; FD: Free days; ICU: intensive care unit; NE: norepinephrine; NS: not significant; PaO₂/FiO₂: arterial-oxygen-tension to inspired-oxygen-fraction ratio; RCT: randomised controlled trial; RRT: renal-replacement therapy; SDC-1: syndecan-1; SOFA: Sequential Organ Failure Assessment; VFD: ventilator-free days

Results

Database extraction

A total of 3,218 records were identified across the three databases. After removing duplicates, 2,176 publications were screened based on their titles, abstracts, and full texts (Fig. 1). In accordance with the predefined inclusion and exclusion criteria, data extraction was performed for 303 eligible preclinical studies that implemented therapeutic interventions targeting capillary leak and included a control group.

Fig. 1.

Article selection flow chart

Animal sepsis models

Of the 303 studies identified, the large majority (n = 246, 81.2%) employed small animal models, specifically mice (n = 127) and rats (n = 119). This was followed by studies using sheep (n = 26, 8.6%) and pigs (n = 21, 6.9%). Figure presents the distribution of animal species used in these studies and their temporal progression.

Fig. 2.

Distribution of animal species used in the studies. Others: rabbit (n = 3), guinea pig (n = 2), goat (n = 2), cat (n = 1), dog (n = 1), baboon (n = 1)

Sepsis induction methods

Table 1 delineates the methodologies employed for sepsis induction in animal models. Although 303 studies were included, ten of them employed more than one induction method, yielding 313 documented experimental protocols. Lipopolysaccharide (LPS) infusion was the most commonly used induction method across all animal models (n = 181, 59.7%), accounting for 60.1% of small-animal protocols and 51.0% of large-animal protocols. Cecal ligation and puncture was the second most frequently employed method in small animals (41.8%, n = 103/246), whereas live pathogen administration was the second most common method in large animal studies (27.7%, n = 13/47).

Table 1.

Distribution of sepsis induction methods, capillary leak assessment and evaluated organs by animal models

	Mouse n (%)	Rat n (%)	Sheep n (%)	Pig n (%)	Others* n (%)
Sepsis induction methods
LPS	69 (54.3)	81 (68)	13 (50)	11 (52.3)	7 (70)
CLP	62 (48.8)	41 (34.4)	3 (11.5)	0 (0)	0 (0)
FIP	1 (0.8)	0 (0)	3 (11.5)	5 (23.8)	0 (0)
Live pathogen	1 (0.8)	0 (0)	7 (26.9)	6 (28.6)	3 (30)
Capillary leak outcomes
Evans blue dye	95 (74.8)	50 (42)	0 (0)	0 (0)	2 (20)
Wet-to-dry weight ratio	35 (27.5)	41 (34.4)	5 (19.2)	6 (28.6)	2 (20)
FITC	20 (15.7)	36 (30.2)	0 (0)	0 (0)	1 (10)
BAL protein	16 (12.6)	4 (3.4)	0 (0)	3 (14.3)	1 (10)
L/P protein ratio	0 (0)	0 (0)	16 (61.5)	0 (0)	3 (30)
Fluid balance	0 (0)	2 (1.7)	7 (26.9)	8 (38)	0 (0)
Radioactivity**	0 (0)	14 (11.8)	0 (0)	3 (14.3)	2 (20)
Lung water (ELWI)	0 (0)	0 (0)	1 (3.8)	3 (14.3)	0 (0)
Others***	0 (0)	2 (1.7)	0 (0)	1 (4.8)	0 (0)
Evaluated organs
Lung	80 (63)	73 (61.3)	9 (34.6)	12 (57)	5 (50)
Gut	21 (16.5)	43 (36.1)	1 (3.8)	2 (9.5)	1 (10)
Kidney	20 (15.7)	11 (9.2)	1 (3.8)	2 (9.5)	0 (0)
Liver	11 (8.7)	8 (6.7)	0 (0)	2 (9.5)	0 (0)
Brain	11 (8.7)	4 (3.4)	0 (0)	0 (0)	0 (0)
Skin	9 (7.1)	3 (2.5)	0 (0)	1 (4.8)	1 (10)
Heart	4 (3.1)	8 (6.7)	0 (0)	2 (9.5)	0 (0)
Peritoneal cavity	4 (3.1)	0 (0)	0 (0)	1 (4.8)	0 (0)
Lymph	0 (0)	0 (0)	13 (50)	0 (0)	3 (30)
Whole body	0 (0)	0 (0)	6 (23)	7 (33.3)	0 (0)
Others****	5 (3.9)	5 (4.2)	0 (0)	0 (0)	0 (0)

Percentages are expressed as the number of studies concerned divided by the number of studies per category. Some studies used two or even three different methods, which explains that the sum of percentages exceeds 100%. LPS: lipopolysaccharide, CLP: cecal ligation and puncture, FIP: fecal injection into the peritoneal cavity, W/D ratio: wet-to-dry weight ratio, FITC: fluorescein isothiocyanate, BAL: bronchoalveolar lavage, L/P: lymph-to-plasma protein ratio, ELWI: extravascular lung water index. *Species: rabbit (n = 3), goat (n = 2), cat (n = 1), dog (n = 1), and guinea pig (n = 2). ** 125-Iodine labeled albumin (n = 17), Technetium 99mTc labeled albumin (n = 1), Indium 113m-chloride plasma transferrin (n = 1). ***Assessment methods: fluid effusions, hydraulic permeability, and Trypan Blue Dye. ****Evaluated organs: muscle, seminal vesicle, spleen, pancreas, diaphragm, and cremasteric venules

Capillary leak assessment outcomes

The methodologies utilized to assess capillary leak demonstrated considerable variation between studies involving small and large animals. In rodent studies, the predominant approach was the intravascular administration of dyes, such as Evans or Trypan Blue, or fluorescent compounds, including fluorescein isothiocyanate (FITC)-dextran, accounting for 81.7% (n = 201/246) of cases. Dye-based methods rely on the intravenous injection of a tracer that binds plasma albumin, or of fluorescein-conjugated dextrans of various molecular weights, quantified by fluorometry [14, 15]. Their main advantages are simplicity, low cost, broad organ accessibility, and a direct quantitative read-out of macromolecular flux that specifically reflects endothelial junctional integrity [16]. Among albumin-binding dyes, Evans Blue has been the most widely used, yet its irreversible albumin binding confers a prolonged plasma half-life that precludes repeated injections and therefore true multi-time-point kinetic assessments confining its use to single end-point measurements: either a late plasma sample for plasma volume estimation or, more commonly, terminal organ harvest to quantify albumin extravasation. This contrasts with tracers permitting serial blood sampling and full plasma activity-time curve analysis—radiolabeled albumin conjugates (¹²⁵I-albumin, ⁵¹Cr-albumin) being the archetype—from which plasma volume and transcapillary albumin escape rate can be simultaneously derived from a single experiment. Beyond this kinetic constraint, dye-based methods share additional limitations: sensitivity to extraction protocol (solvent, incubation time, residual blood, tissue perfusion), low sensitivity in barrier-tight organs, dye toxicity at high doses, and impracticability in large animals and humans [15].

The wet-to-dry (W/D) organ weight ratio quantifies the water content by measuring the weight ratio of the organ immediately after harvest and following a drying period of 72 h. It emerged as the second most prevalent method in small animal studies, accounting for 30.9% (n = 76/246) of the studies. In studies involving large animals, this method ranked third, with a utilization rate of 23.4% (n = 11/47). Furthermore, methods applicable to human contexts, such as fluid balance or indexed extravascular lung water (ELWI), were used in 15 and 4 (31.9% and 8.5% respectively) of the studies focusing on large animals. These techniques are surrogates of tissue water content and provide an integrative picture of barrier dysfunction together with hydrostatic and oncotic gradients, lymphatic drainage and resuscitation strategy. They are widely used because they require minimal equipment and, for fluid balance and ELWI, are routinely available at the bedside [17, 18]. Their key limitation is that they do not measure capillary leak per se: equivalent values may correspond to very different pathophysiological situations. Also, the W/D ratio underestimates accumulated water in permeability edema and remains a single-time-point measurement [19].

Among the 303 included studies, 19 (6.3%) used radiolabeled tracers which are considered the gold standard. Particularly, the injection of radiolabeled human serum albumin is regarded as the reference method as it directly tracks protein transudation and provides a calibration tied to albumin flux, can be repeated multiple times, and allows organ-specific resolution further when coupled with imaging (SPECT/PET) [20–22]. Its main limitations are the use of ionising radiation, the need for licensed handling facilities, a high cost, and ethical constraints that preclude repeated measurements.

Organs and biological compartments assessed

Table 1 illustrates the distribution of the assessed organs or biological compartments. In studies involving small animals, the lung (62.2%, n = 153/246), gut (26%, n = 64/246), and kidneys (12.6%, n = 31/246) were the most frequently targeted organs. Similarly, in studies involving large animals, the lung was the most frequently evaluated organ (44.6%, n = 21/47). Notably, studies focusing on sheep predominantly targeted the lymphatic system (50%, n = 13/26).

Therapeutic interventions

To facilitate analysis, we categorized therapeutic interventions according to the primary pathway they target. Of the 303 articles included, we identified 232 distinct therapeutic interventions, which we further classified into 15 subgroups. The categories were selected based on recent reviews addressing sepsis-induced endothelial dysfunction and are illustrated in Fig. 3 [23]. While most therapeutics were readily classified, such as fluids, some therapeutics exhibited pleiotropic effects and could not be confined to a single category. Consequently, we established a category labeled ‘others’ which encompasses various interventions, including natural compounds. The categorization was determined through discussions among four authors (HD, SHL, AH, and AD). Table 2 presents a comprehensive overview of the distribution of therapeutic interventions and study characteristics; individual agents are listed in Table S1. The most frequently examined pathway was endothelial activation, accounting for 12.2% (n = 37/303) of the studies, followed by intercellular junctions at 11.2% (n = 34/303), and the nitric oxide (NO) pathway at 8.9% (n = 27/303). Notably, 282 of 303 studies (93.1%) reported capillary leak reduction on at least one outcome. Regarding intervention timing, 142 studies (46.9%) administered the therapeutic agent after sepsis induction, while 110 studies (36.3%) used a pre-treatment paradigm.

Fig. 3.

Capillary leak targets and preclinical studies distribution. Figure illustrates a simplified version of the main pathways implicated in the capillary leak and the distribution of the therapeutic interventions encountered in the 303 articles included in the review. Circles display the number of studies and the species implicated while bars show the capillary leak assessment methods used: macromolecular tracer methods (Evans Blue dye, FITC-dextran, radiolabelled albumin), and edema-quantifying methods (wet-to-dry ratio, fluid balance, ELWI). αR: α adrenergic receptor, CaV: Calcium voltage dependent channel, EBD: Evans Blue Dye, eNOS: endothelial NO synthase, FITC: Fluorescein Isothiocyanate, GPCR: G protein-coupled receptor, IL: Interleukin, NO: Nitric oxide, PKA/C: Protein kinase A/C, ROS: reactive oxygen species, RS: Sarcoplasmic reticulum, VCAM: Vascular cell adhesion molecule, W/D: Wet to Dry

Table 2.

Therapeutic intervention characteristics

Therapeutic interventions	No of studies	Species	Intervention timing(sepsis induction reference)	Capillary leak assessment outcomes	Evaluated organs or biological compartments	Effectiveness*
Endothelial activation	37	Mouse (22), rat (13), pig (1), rabbit (1)	Before (20), at (1),after (13), before & after (3)	EBD (21), W/D (10), FITC (8), BAL (6), radioactivity (2)	Lung (28), gut (3), kidney (6), liver (2), heart (2), skin (4), muscle (1), peritoneal cavity (1)	Positive (36),no effect (1)
Intercellular junctions	34	Mouse (21), rat (12), guinea pig (1)	Before (16), at (2),after (14), before & after (1), not reported (1)	EBD (23), W/D (13), FITC (8), BAL (3)	Lung (24), gut (8), kidney (10), liver (4), heart (1), skin (1), spleen (1), muscle (1), brain (2)	Positive (32),no effect (2)
NO pathway	27	Mouse (6), rat (15), sheep (1)	Before (6), at (4),after (13), before & after (4)	EBD (11), W/D (10), FITC (5), FB (2), BAL (1), L/P (3), radioactivity (4)	Lung (15), gut (12), kidney (4), liver (3), heart (4), skin (3), seminal vesicles (1), lymph (3), whole body (2)	Positive (24),no effect (3)
Fluids	21	Mouse (3), rat (12), pig (6)	Before (2), at (6),after (13)	EBD (8), W/D (7), FB (6), FITC (4), BAL (1), radioactivity (3)	Lung (12), gut (2), kidney (1), liver (2), skin (1), brain (1), whole body (5)	Positive (18),no effect (3)
Immune cell activation	18	Mouse (6), rat (6), sheep (4), goat (2)	Before (9), at (1),after (2), before & after (5),not reported (1)	EBD (4), FITC (9), L/P (6), BAL (1)	Lung (5), gut (5), kidney (2), liver (2), heart (1), skin (1), brain (1), lymph (6)	Positive (16),no effect (2)
Vasoactive agents	17	Mouse (2), rat (4), pig (4), sheep (7)	Before (5), at (3),after (7), not reported (2)	EBD (3), W/D (3), FITC (1), FB (7), BAL (1), radioactivity (1), L/P (1), ELWI (2)	Lung (7), gut (2), kidney (1), liver (1), brain (1), lymph (1), whole body (6)	Positive (16),no effect (1)
Coagulation	14	Mouse (1), rat (10), pig (2), baboon (1)	Before (7), at (1),after (5), before & after (1)	EBD (3), W/D (5), FITC (4), radioactivity (4), fluid effusion (1)	Lung (8), gut (6), seminal vesicles (1), peritoneal cavity (1)	Positive (14)
Cell adhesion	14	Mouse (10), rat (3), rabbit (1)	Before (4), at (1),after (9)	EBD (8), W/D (2), FITC (3), BAL (1), radioactivity (2)	Lung (5), gut (7), kidney (1), skin (1), brain (1)	Positive (14)
Oxidative stress	11	Mouse (4), rat (3), pig (2), rabbit (2)	Before (3), at (1),after (7)	EBD (4), W/D (2), FITC (3), BAL (2), L/P (2)	Lung (6), gut (2), brain (1), lymph (2)	Positive (9),no effect (2)
Cytokines	10	Mouse (3), rat (6), rabbit (1)	Before (4), after (4), before & after (2)	EBD (3), W/D (6), FITC (2), BAL (1), L/P (2)	Lung (6), gut (2), kidney (1), liver (1), lymph (2)	Positive (9),no effect (1)
HMGB1 & DAMPs	9	Mouse (7), rat (2)	After (9)	EBD (9)	Lung (2), gut (5), peritoneal cavity (1)	Positive (9)
Glycocalyx	4	Mouse (3), pig (1)	After (4)	EBD (2), W/D (2), FB (1)	Lung (4), gut (1), kidney (1), liver (1), heart (1)	Positive (3),no effect (1)
Corticosteroids	4	Rat (1), dog (1), sheep (2)	Before (2), at (1),after (1)	EBD (1), FITC (8), L/P (2)	Lung (2), lymph (2)	Positive (3),no effect (1)
Physical treatment	3	Mouse (1), rat (1), sheep (1)	Before (2), before & after (1)	EBD (1), W/D (2)	Lung (2), gut (1)	Positive (2),no effect (1)
Others	80	Mouse (38), rat (31), pig (6), sheep (4), cat (1)	Before (30), at (6),after (41), before & after (2),not reported (1)	EBD (45), W/D (29), FITC (9), BAL (9), radioactivity (3), FB (1), L/P (3), ELWI (2), other dye (1), hydraulic permeability (1)	Lung (53), gut (12), kidney (7), liver (5), heart (5), muscle (1), brain (8), diaphragm (1), pancreas (1), cremasteric venules (1), peritoneal cavity (1), lymph (3)	Positive (77),no effect (3)

* Compared with control group. BAL: bronchoalveolar lavage protein, EBD: Evans Blue Dye, ELWI: extravascular lung water index, FITC: fluorescein isothiocyanate, L/P: lymph-to-plasma protein, W/D: wet-to-dry weight ratio.

Risk of bias

A formal assessment using the SYRCLE tool revealed significant quality concerns across the 303 included studies (Table S2). While baseline characteristics were generally well-standardized (92.1%), critical measures to reduce selection bias, such as randomization sequence generation (20.1%) and allocation concealment (3.9%), were poorly reported. Furthermore, blinding was infrequently documented, particularly for physiological assessments, occurring in less than 27% of studies.

Interventions with available human data

Of the 232 distinct interventions retained in Table S1, 19 agents could be linked to at least one randomized or prospective interventional clinical study in sepsis, septic shock or sepsis-associated ARDS since January 2015, and were retained for translational appraisal in Table 3. The four-criterion translational quality score showed that 2 agents fulfilled all four criteria, 7 agents fulfilled three criteria and 10 agents fulfilled two or fewer criteria. The criterion most frequently missed was the translational model quality parameter reflecting the dominance of small animal and LPS-based studies in our corpus. Detailed scoring, supporting preclinical references and clinical trial data (phase, design, endpoints, results, and future/ongoing studies) are provided in Table 3.

Discussion

This systematic review provides the most comprehensive analysis to date of preclinical studies evaluating therapeutic interventions targeting sepsis-induced capillary leak. Across 303 studies and 232 distinct interventions, we observed substantial heterogeneity in animal models, sepsis induction methods, and capillary leak assessment, together with a striking positive-result rate and a limited number of candidates that have effectively reached clinical evaluation. These findings have direct implications for how clinicians and trialists should interpret the existing preclinical literature and design future translational programmes.

Risk of bias and publication bias

Critical methodological safeguards (randomisation, allocation concealment, and outcome assessment blinding) were poorly reported across the literature. The 93.1% (282/303) of studies reporting capillary leak reduction is strongly suggestive of publication bias, consistent with previous observations in preclinical critical-care research. This pattern likely overstates the apparent maturity of the field and contributes to redundant experiments, unnecessary animal use, and ultimately to translational failures. Mechanistic heterogeneity across intervention categories precluded a formal quantitative assessment (Egger’s test, funnel plot).

Capillary leak assessment: choosing the right read-out

Methods used to assess capillary leak fall into two mechanistically distinct categories: those measuring macromolecular flux (Evans Blue dye, FITC-dextran, radiolabeled albumin), which specifically reflect endothelial junctional integrity, and those measuring tissue edema (W/D ratio, fluid balance, ELWI), which integrate barrier dysfunction with hydrostatic gradients, lymphatic drainage, and resuscitation strategy. Treating these endpoints as equivalent surrogates is a core limitation of any pooled analysis. In rodents, dye-based methods remain the most widely used and standardized approach, but they should not be considered a gold standard given post-mortem assessment, dye toxicity, and inter-protocol variability. In larger animals and at the bedside, direct capillary leak assessment is considerably more challenging. Cumulative fluid balance and W/D ratio are confounded by resuscitation strategy, renal function, and plasma albumin, while direct tracers (¹²⁵I- or ⁹⁹ᵐTc-albumin [24, 25] or¹³¹I-HSA with the Daxor BVA-100® [26]) remain costly and radioactive. Ideally, capillary leakage could be evaluated throughout the entire organism, with the capability to identify hotspots in specific organs (e.g., lungs or kidneys) and systems. More accessible bedside tools — the Pulmonary Vascular Permeability Index (PVPI) derived from transpulmonary thermodilution, and the recently proposed Vascular Leak Index (VLI), which estimates vascular leak from hematocrit-based blood volume variation and net fluid balance—represent promising avenues. However, the VLI remains an indirect, calculated surrogate heavily dependent on fluid management decisions and assumptions about fluid distribution, and cannot distinguish between organ compartments or capture the spatial heterogeneity of endothelial dysfunction [27]. Ultimately, a non-invasive, organism-wide, organ-resolved assessment of vascular permeability remains out of reach.

LPS endotoxemia is not sepsis

Rodent models, used in 81.2% of studies, are valuable for dissecting molecular and cellular mechanisms but differ substantially from human sepsis in their cardiovascular physiology, immune responses, and endotoxin tolerance. When the research question shifts toward therapeutic translation, large-animal models—particularly porcine live-bacteria models that allow realistic resuscitation, vasopressor titration, and antimicrobial dosing [28]— are likely to be more informative. By contrast, LPS-induced endotoxemia, used in 59.7% of studies, produces a transient cytokine surge, reproduces only the hyperdynamic phase of septic shock, and lacks ongoing host–pathogen interaction [29]. The dominance of LPS-endotoxemia models may substantially account for why interventions repeatedly effective in the preclinical setting have failed in clinical sepsis trials. These limitations have repeatedly been highlighted [30] and are codified in the Minimum Quality Threshold in Preclinical Sepsis Studies (MQTiPSS) framework [31], alongside the 3Rs principles of refinement, reduction, and replacement.

Finally, pre-treatment paradigms—used in over a third of included studies—have inherently limited clinical relevance, since the progression to sepsis is rarely predictable, even in clinically anticipated infections (e.g., neutropenia, post-operative settings). Studies administering the test intervention before or simultaneously with the septic insult therefore evaluate a pharmacological scenario that is irreproducible at the bedside, regardless of preclinical efficacy. Post-treatment paradigms should be the minimum standard for any intervention intended to reach clinical evaluation.

Promising candidates and the translational gap

Mapping the 232 distinct interventions reported in Table S1 against the clinical literature since 2015 highlights a striking translational attrition: only a minority of preclinical candidates has produced randomized clinical data, and none has achieved high translational priority as a capillary leak-specific therapy. This pattern is consistent with the broader translational crisis observed in critical care research and directly reflects the methodological concerns raised above.

Three observations emerge from this analysis. First, the only interventions currently endorsed by the 2026 Surviving Sepsis Campaign guidelines (albumin, low-dose corticosteroids, vasopressin) are not capillary-leak-specific therapies, and their effect on endothelial barrier integrity is indirect at best. None of the highly specific biological targets identified in our preclinical corpus—Slit2/Robo4 signaling, Tie2/Angpt1 agonism, sphingosine-1-phosphate receptor modulation, ROCK pathway, glycocalyx-protective agents—has yet generated a positive Phase III trial as a standalone intervention. Second, the most mechanistically sound candidates currently under confirmatory evaluation (adrecizumab, imatinib, recombinant thrombomodulin, mesenchymal stromal cells) share a key feature: they are being tested in specific populations defined either by a endothelial dysfunction or by a coagulation phenotype. This enrichment strategy contrasts sharply with the preclinical paradigm, in which interventions are typically tested in genetically and physiologically uniform animals with a controlled inoculum, and may explain part of the disconnect between preclinical success rates and clinical results. Biomarker-based phenotyping represents the most plausible strategy for enriching future trials [32]. Third, unambiguous negative or harmful pivotal trials (hydroxyethyl starch [33], activated protein C [34] etc.) were designed based upon compelling preclinical signals [24, 35], illustrating that strong preclinical effects on tracer extravasation or W/D ratio do not reliably predict clinical benefit, and reinforcing the methodological critique developed above. We therefore argue that, beyond the question of which target to prioritize, the translational priority of a candidate cannot be appraised independently of the preclinical platform that generated its supporting evidence.

Clinical implications

For the bedside clinician, this review conveys three practical messages. First, no proven capillary leak-targeting therapy currently exists: fluid stewardship remains the only available tool, and the 2026 Surviving Sepsis Campaign guidelines now recommend active fluid removal after the initial resuscitation phase to mitigate interstitial accumulation, consistent with evidence that positive fluid balance independently predicts mortality [36]. Second, endothelial biomarkers (syndecan-1, angiopoietin-2, soluble thrombomodulin) are increasingly available in clinical laboratories and should be considered when phenotyping patients for potential enrolment in endothelial-directed trials. Third, clinicians encountering promising preclinical data should ask three questions before extrapolating to patients: Was the model a live-pathogen large-animal preparation? Was the drug administered after sepsis induction? Was capillary leak confirmed by more than one mechanistically distinct read-out? An affirmative answer to all three substantially increases translational plausibility.

Limitations

This study has several limitations. First, the substantial heterogeneity across studies precluded quantitative synthesis or meta-analysis, preventing a reliable comparison of effect sizes across intervention categories. Many preclinical studies report statistically significant but biologically modest reductions in tracer leak, without demonstrating concomitant improvements in organ function or survival. Whether such reductions translate into clinically meaningful benefits remains unclear, and this disconnect likely contributes to translation failures. Second, our categorization of therapeutic interventions by targeted pathway is inherently imperfect given the pleiotropic mechanisms of many agents, and our scope was restricted to sepsis-induced leak, although capillary leak is equally relevant in cardiac arrest, trauma, and the perioperative setting. Third, capillary leak is only one component of a broader ‘endothelialiopathy syndrome’ that also encompasses glycocalyx shedding, microthrombus formation, and loss of vasomotor tone. Interventions targeting leak alone may therefore incompletely address the full spectrum of sepsis-induced vascular injury. Fourth, the magnitude of capillary leak depends heavily on infection site, resuscitation strategy, and renal function, factors that are rarely controlled for across studies. Finally, our four-criterion translational score is an interpretive tool not a predictor of clinical efficacy. Each criterion is binary and applied uniformly across interventions, which inevitably obscures gradations of evidence within each domain, and individual agents may shift category as ongoing trials report. Despite these limitations, we hope the score will help clinicians and trialists enrich future RCTs both biologically—by selecting patients with measurable endothelial dysfunction—and methodologically—by ensuring that preclinical support derives from models that recapitulate the post-treatment, live-pathogen, fluid-resuscitated context of human sepsis.

Conclusion

This systematic review depicts a preclinical landscape marked by substantial methodological heterogeneity, a 93.1% capillary-leak-reduction rate, and pervasive reporting limitations that collectively overstate the apparent maturity of the field. Rodent endotoxemia, although mechanistically informative, poorly captures the cardiovascular and host–pathogen complexity of human sepsis; large-animal live-bacteria models offer superior translational fidelity but remain underused. Mapping the 232 identified interventions against the clinical literature since 2015 shows that no capillary leak-specific therapy has yet reached high translational priority; the few agents endorsed by current guidelines exert indirect endothelial effects at best, and the most scientifically rational ongoing programmes rely on biologically enriched patient selection. For the bedside clinician, the message is sobering but actionable: fluid stewardship remains the cornerstone of capillary leak management, endothelial biomarkers are available to identify patients who may benefit from emerging endothelial-directed therapies, and enrolment in enriched trials represents the most direct path to closing the translational gap. Bridging that gap will ultimately require standardised, clinically applicable capillary leak assessment tools, greater investment in post-treatment paradigms in live-pathogen large-animal models, and randomised trials that select patients on the basis of measurable endothelial dysfunction.

Abbreviations

BAL

Bronchoalveolar lavage

CLP

Cecal ligation and puncture

DAMPs

Damage associated molecular patterns

EBD

Evans blue dye

ELWI

Extravascular lung water index

FIP

Fecal injection into the peritoneal cavity

FITC

Fluorescein isothiocyanate

ICU

Intensive care unit

L/P

Lymph-to-plasma protein ratio

LPS

Lipopolysaccharide

MQTiPSS

Minimum quality threshold in preclinical sepsis studies

PAMPs

Pathogen associated molecular patterns

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-analyses

RCT

Randomised clinical trial

W/D ratio

Wet-to-dry weight ratio

Author contributions

SHL and HD participated in study design, data extraction, articles inclusion, interventions classification, figures drawing and drafting of the manuscript. AD and AH participated in study design, articles inclusion, interventions classification and revised the manuscript. VL and BA participated in study design and revised the manuscript. All authors read and approved the manuscript.

Funding

This study was a part of the PhD thesis of the first co-author, SHL, who received a scholarship from the Vietnamese Government (Project 89, Decision No. 3063/QĐ-BGDĐT, dated October 10, 2023) and financial support from VetAgro Sup.

Data availability

The datasets used and analyzed in the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.