The Clinical View of Sepsis-Associated AKI: How Basic Science Can Help Solve This Problem

Abstract

The global health impact of sepsis is difficult to understate. As a complication of sepsis, the development of sepsis-associated acute kidney injury (SA-AKI) significantly increases the risk for mortality. Although several epidemiological risk factors for SA-AKI are known, the heterogeneity of this syndrome – across patients, pathogens, and treatment responses – has hindered therapeutic innovation and contributes to persistently poor outcomes. Precision medicine offers a promising framework to address this complexity, yet a substantial translational gap remains between mechanistic insights from preclinical models and the therapeutic strategies used in clinical practice. To bridge this gap, researchers should consider aligning preclinical models with human sepsis and embrace SA-AKI heterogeneity to identify treatable, mechanistically informed subtypes (endotypes). These efforts can enable the development of personalized therapies aimed at reducing the burden of SA-AKI.

Introduction

In 2017, the World Health Organization (WHO) established sepsis as a global health priority in response to the high incidence and mortality rate with consistently low public awareness^1,2. Clinically, one of the primary drivers of poor sepsis outcomes is the progression to multiple organ dysfunction, and the development of sepsis-associated acute kidney injury (SA-AKI) is especially associated with high rates of morbidity and mortality^3,4. In contemporary clinical care, rigorous quality improvement initiatives and advances in implementation science have paved the way for the modest improvements in sepsis-related outcomes observed over the past decade^5,6. Despite significant investments in clinical, translational, and basic science research, clinical practice otherwise remains static and novel therapeutics remain elusive⁷. To address this stagnation, researchers have begun to dissect the heterogeneity of SA-AKI by leveraging big data analytical approaches to group patients across large clinical datasets (subphenotype) as well as enrich patients by underlying mechanisms (endotype)^8,9. Although the need to address the heterogeneity of sepsis is well recognized – and biologically informed phenotyping has been pursued to that end – we still have a nascent understanding of the pathophysiological mechanisms responsible for SA-AKI development¹⁰. Therefore, there is considerable potential for the basic sciences to drive clinically meaningful innovation by expanding knowledge of molecular pathways underlying the pathophysiology of sepsis endotypes, revealing novel mechanisms driving extrarenal organ and intrarenal crosstalk, and dedicating efforts to investigate the progression of AKI to chronic kidney disease (CKD) in sepsis. The objective of this review is to assess the current state of SA-AKI translational research and to outline how the targeted alignment of preclinical animal models to human sepsis can better empower clinicians with the necessary tools to improve outcomes for SA-AKI^11,12.

The Current State: A Gap Persists from Bench-to-Bedside and Back Again

Understanding of pathophysiology

The conceptual framework for SA-AKI pathogenesis has shifted from attributing renal injury solely to systemic hypoperfusion to a more nuanced understanding that incorporates multiple injurious mechanisms and their interplay with host susceptibility and capacity for resilience^13,14. While septic shock can lead to impaired renal perfusion pressure, animal and human studies have demonstrated that renal blood flow is increased in a group of patients in whom SA-AKI occurs, strongly suggesting that hypoperfusion is not the only injury mechanism^15,16. Instead, the current pathobiological framework for SA-AKI involves various permutations of renal microcirculatory dysfunction, dysregulation of innate and adaptive immune responses, mitochondrial injury with uncoupling of the electron transport chain, and adaptive metabolic reprogramming that collectively impair glomerular filtration and solute/water transport across tubular epithelial cells¹⁷. The degree to which each of these perturbations plays a role in SA-AKI development varies by patient age, underlying genetics, environmental exposures, pathogen, location of primary infection, baseline kidney function, and severity of illness. Therefore, the aspiration to discover a “silver bullet” to treat SA-AKI is biologically inconceivable. A more prudent approach will focus on developing therapeutic targets informed by a patient’s individualized SA-AKI endotype.

Use of clinical subphenotypes to reveal candidate pathways for mechanistic exploration

Recent characterizations of novel subphenotypes and endotypes among patients with sepsis and SA-AKI may shed light on candidate pathophysiologic signatures that have thus far been underappreciated^18,19. In a study of adult patients admitted to the intensive care unit (ICU) with AKI, 70% of whom had sepsis as the primary trigger, latent class analysis of clinical data and biomarker levels revealed two distinct subphenotypes termed AKI-SP1 and AKI-SP2²⁰. Patients with the AKI-SP2 subphenotype demonstrated higher risk of renal non-recovery and increased 28-day mortality compared to patients identified as AKI-SP1. The AKI-SP2 subphenotype revealed a strong association with elevations in plasma Angiopoietin-2 / Angiopoietin-1 ratio and soluble tumor necrosis factor receptor 1 (sTNFR-1), establishing a possible connection between endothelial activation and the inflammatory response in these patients²⁰. A separate study of adults with sepsis aimed to retrospectively risk-stratify patients for death or receipt of kidney replacement therapy by combining previously characterized transcriptomic-derived endotypes with biomarker profiling targeting proteins associated with kidney function, immune response, and endothelial dysfunction²¹. They reported unique interactions between transcriptomic endotypes and biomarkers specific to kidney function, immune activation, and vascular injury. This highlights a promising opportunity for future prospective enrichment based on a patient’s underlying biology. As an example in pediatrics, the Pediatric Sepsis Biomarker Risk for Acute Kidney Injury (PERSEVERE-II AKI) prediction model was validated to identify critically ill children at risk for severe, persistent SA-AKI at day 3 of septic shock²². This model leveraged classification and regression tree analysis (CART) to risk stratify patients into low- versus high-risk terminal nodes based on a previously validated sepsis mortality probability assignment and three serum biomarkers: heat shock protein 70 kDa 1B (HSPA1B), granzyme B (GZMB), and interleukin 8 (IL8)²³. Despite the impressive SA-AKI risk stratification by clinical subphenotypes and molecular endotypes in each of these examples, wide clinical adoption will likely be delayed until easily accessible, cost-effective, and quick diagnostic platforms are created to delineate the mechanisms that drive these endotypes and develop targeted therapeutics to treat them in a clinically feasible timeframe.

Preclinical models utilized in basic and translational science

Compared to unpredictable patient variability in clinical research studies, basic science offers the advantage to test and control for key variables that are hypothesized to contribute to SA-AKI endotypic heterogeneity. Researchers commonly employ a variety of preclinical animal models to interrogate host-pathogen interactions responsible for SA-AKI development (Table 1). This review focuses on rodent models of sepsis given their lower costs, widespread usage, and more detailed characterization of host-pathogen interactions in sepsis compared to larger animal models²⁴. Compared to large animals, rodents remain a powerful model for genetic studies due to their well-characterized genome and robust transgenic tools. Rodent sepsis models range from the administration of concentrated bacterial cell membrane/wall components that mimic host recognition of damage- and pathogen-associated molecular patterns (DAMPs, PAMPs) to the intravenous inoculation of live pathogens into the bloodstream^25,26. Localized primary infections are also commonly used, primarily in the form of intraabdominal pathogenic triggers, through cecal ligation and puncture (CLP) or intraperitoneal cecal slurry (CS) injection^27,28. Less commonly used models include systemic viral infections, bacterial pneumonia, and colon ascendens stent peritonitis^29-31.

Table 1: Preclinical Rodent Models of SA-AKI.

	Model	Details	Advantages	Disadvantages	Antibiotics
Live Infections	Cecal ligation and puncture (CLP)	Surgically inflicted intraabdominal sepsis resulting in polymicrobial infection. Predominantly gram negative and anaerobic bacteria	Consistently leads to SA-AKI. Bacterial translocation into blood stream is consistent with human surgical infections	Surgical component makes model less generalizable to non-surgical sepsis. Variability in technique impacts severity	Cover gram-negative and anaerobic bacteria
	Cecal Slurry (CS) (Faecal Slurry)	Intraperitoneal administration of cecal stool contents harvested from donor mice and suspended in glycerol solution. Polymicrobial infection: predominantly gram negative and anaerobic bacteria	Able to dose consistently by grams of cecal slurry to grams of rodent body weight. Less technical variation than CLP	Batch to batch variation. Impact of donor and/or host gut microbiome on immune response	Cover gram-negative and anaerobic bacteria
	Staphylococcal bacteremia	Gram positive bacteria administered through tail vein injection	Model response to single pathogen with consistent dosing. Common cause of bacteremia in humans	Potential for variation by Staphylococcal aureus serotype	Ceftriaxone, ampicillin, or cefazolin for MSSA. Vancomycin for MRSA
	E. coli bacteremia	Gram negative bacteria administered through tail vein injection	Model response to single pathogen with consistent dosing. Common cause of bacteremia in humans	Potential for variation by E. coli serotype	Ceftriaxone, ampicillin
	Pneumonia	Administer live bacteria such as Pseudomonas aeruginosa, Streptococcus pneumoniae, or Staphylococcus aureus by intratracheal or orotracheal lavage	Most common cause of sepsis in humans. Can model lung-kidney interactions	Inconsistent rates of SA-AKI in rodents (as measured by serum creatinine) compared to other models. May require a second hit	Ampicillin or Ceftriaxone for Staph or Strep species. Cefepime for Pseudomonas
	COVID-19	Direct primary lung infection	Novel with direct clinical corollary for histopathologic al kidney injury	Stringent precautions	Antivirals such as remdesivir
PAMPs	Lipopolysaccharide (LPS)	Component of gram-negative cell membrane. Induces “endotoxemia”	Consistent dosing. Endotoxemia is well established clinical phenomenon	Mechanism of action is narrow. Does not model host-pathogen interactions of live bacteria	N/A
	Peptidoglycan (PG)	Component of gram-positive cell wall	Consistent dosing. Use would expand SA-AKI mechanistic understanding relative to gram-positive bacteria		N/A
	Lipoteichoic acid (LTA)	Component of gram-positive cell wall	Consistent dosing. Use would expand SA-AKI mechanistic understanding relative to gram-positive bacteria		N/A
	Polyinosinic-polycytidilic acid (poly(I:C))	Synthetic viral dsRNA agonist	Consistent dosing. Has multiple receptors with responses that vary by host cell type	Mechanism of action is narrow. Does not model host-pathogen interactions of live intracellular viruses	N/A
	CpG	Mimics unmethylated viral and bacterial DNA	Consistent dosing. Use would expand SA-AKI mechanistic understanding relative to viral infections.		N/A

Abbreviations: PAMPS, pathogen-associated molecular patterns; SA-AKI, sepsis-associated acute kidney injury; MSSA, methicillin-sensitive Staphylococcus aureus; MRSA, methicillin-resistant Staphylococcus aureus; dsRNA, double-stranded RNA

Despite these varied models, translating preclinical findings to patients with sepsis has been fraught with challenges, including several negative clinical trials that came from promising preclinical results³². This has naturally resulted in increased scrutiny from the scientific community on the utility of animal models to study sepsis or SA-AKI^33,34. To preserve the numerous experimental benefits preclinical animal models offer in SA-AKI mechanistic investigations, a Wiggers-Bernand Conference was held in 2017 which led to the publication of a minimum quality threshold for preclinical sepsis studies to increase model standardization and improve rigor and reproducibility³⁵. This expert consensus initiative touched on several points that will be discussed in further detail below, namely emphasizing benefits of aligning preclinical sepsis models with human sepsis epidemiology and treatment approaches. Similarly, the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) convened a workshop to generate consensus recommendations for improving translational barriers in AKI research by promoting a reverse translational approach¹¹. These initiatives highlight an opportunity for bidirectional translational research to bridge the gap between mechanistic discoveries at the bench with endotype manifestation in clinical practice to enhance the delivery of novel therapeutics to the bedside of patients with SA-AKI.

Aligning Preclinical Sepsis Models to Human Disease

How does sepsis manifest in humans?

Despite the heterogeneity of sepsis, several important epidemiologic findings warrant highlighting. First, the most common primary source of infection in both adult and pediatric sepsis is pneumonia, followed by abdominal infections and direct bloodstream infections^36,37. A meta-analysis of patients with SA-AKI indicates abdominal infections, gram negative bacterial infections, and infections with an unknown primary source are all predictive of mortality³⁸. Second, patients who are most likely to develop septic shock and subsequently die are at the extremes of ages: neonates and children less than 18 years old have a ~10-35% mortality, young adults <20% mortality, and adults older than 65 years nearly 50% mortality rate^39,40. Third, the presence of co-morbidities, particularly CKD, are associated with reduced rates of renal recovery and increased dialysis dependence following SA-AKI⁴¹. This suggests that sepsis may serve as a second hit to patients with reduced renal reserve. As the population of patients with chronic comorbidities continues to increase, this phenomenon also merits additional preclinical investigation⁴².

Connecting epidemiologic associations to the underlying biology is essential. Sepsis presents as a constellation of inflammatory derangements leading to a cascade of organ failures, all primarily driven by host-pathogen interactions⁴³. The specific immune-mediated injury pathways and organ failure patterns vary by patient, and the clinical manifestations within an individual can dynamically switch across the course of illness⁴⁴. Work by Seymour et al. characterized four clinical phenotypes among adults with sepsis that align with specific organ failures (the δ phenotype demonstrated the highest rates of AKI while the β phenotype most closely aligned with the presence of CKD), and each phenotype varied by incidence and risk of mortality^45,46. While this is just one example of a strategy for disentangling sepsis heterogeneity, it successfully highlights how age, baseline laboratory values, organ-specific needs for intensive care therapies, and risks for mortality are disparate from patient-to-patient.

Opportunities to align preclinical animal models to human epidemiology

The use of preclinical sepsis models provides numerous scientific benefits, from generating new mechanistic knowledge to providing a model for reverse translational validation. The selection of any sepsis model should be informed by the scientific question at hand as the complexity of sepsis-induced multi-organ failure, host-pathogen interactions, and the heterogeneity of treatment effect all serve as potential confounders to generating biologically relevant, reproducible results. Specific attention to the use of preclinical sepsis models for the purpose of directly advancing clinical care is provided here given the relative stagnation in therapeutic innovation that has stymied advancements in SA-AKI outcomes.

As mentioned previously, animal sepsis models have historically relied heavily on lipopolysaccharide (LPS) and CLP to recapitulate the pathogen-triggered systemic inflammatory response that is observed in humans (Table 1)²⁴. However, while CLP and CS are common preclinical models that specifically rely on live bacteria to study sepsis and SA-AKI, their generalizability to human disease remains limited given the relative incidence of abdominal infections in humans⁴⁷. These polymicrobial infections are challenging to standardize given the diversity of gut microbiota from host-to-host, and the model is typically titrated to severity of systemic illness (or mortality) rather than to degree of functional renal impairment or markers of kidney injury^28,48. In contrast, pneumonia models of SA-AKI have limited characterization in SA-AKI literature, yet human epidemiology suggests this primary source of infection is much more common in patients with sepsis who develop multiple organ dysfunction⁴⁹. Preclinical models of bloodstream infections offer an opportunity to characterize the systemic inflammatory response across a variety of pathogens commonly responsible for human bacteremia, such as Staphylococcus aureus, Streptococcus pneumoniae, Pseudomonas aeruginosa, and Escherichia coli^50,51. Each of these bacteria possess distinct mechanisms that allow their survival via immune evasion, tissue infiltration, and host cellular injury. Other pathogens implicated in AKI pathogenesis include parasites, such as malaria, and fungi^52,53. When selecting a live pathogen sepsis model, it is also important to recognize that pathogens vary in their renal tropism and how they (or their corresponding PAMPs) are recognized by a variety of kidney cell types. It is also worth highlighting that many technical challenges exist when studying dual-species tissue interrogation during host-pathogen interactions, including microbial genetic drift, adaptive evolution, and low microbial DNA/RNA yields within tissue⁵⁴. Gaining a better understanding of the pathogen-specific factors that drive SA-AKI development is a high-yield therapeutic opportunity given the rapidity by which bacterial and other infectious etiologies can be speciated in clinical practice⁵⁵.

To limit unpredictable biologic variability in mechanistic studies, administering well known PAMPs, such as LPS, offer the benefit of inducing dose-dependent and reproducible degrees of kidney injury. Although LPS, through its interaction with toll like receptor-4 (TLR4), is the most common PAMP used in preclinical sepsis modeling of gram-negative infections, it does not allow for mechanistic interrogation of kidney injury secondary to live bacteria or gram-positive pathogens⁵⁶. Peptidoglycan (PG) and lipoteichoic acid (LTA) are two toxins derived primarily from the cell wall of gram-positive bacteria that can also be used to model sepsis in rodents⁵⁷. Given the high incidence of Staphylococcal and Streptococcal bacteremia in humans, there is strong translational rationale for including these gram-positive bacterial affiliated PAMPs in SA-AKI models⁵⁷. PG and LTA elicit an overlapping immune response to the gram-negative endotoxemia produced by LPS while also initiating distinct downstream pathways mediated through different TLRs⁵⁷. A major limitation to using PAMPs as initiators of a systemic immune response is the lack of immune evasion, tissue infiltration, and direct cellular injury that persists until effective bacterial clearance has been achieved compared to sepsis models dependent on live bacteria²⁴. Use of PAMPs also incompletely models the significant synergy that is evolutionarily derived by pathogens to achieve their survival benefit by interfering with multiple host pathways⁵⁸. Despite these important limitations, LPS, PG, and LTA maintain utility in studying specific components of SA-AKI development, including PAMP filtration through the glomerulus and transport across peritubular capillaries, tubular epithelial bioenergetic and cell cycle responses to pathogen-derived stimuli, and paracrine signaling from kidney cells most responsive to TLR activation.

Both aforementioned models fail to acknowledge the importance and commonality of viral infections as precipitants of SA-AKI⁵⁹. Most recently, the SARS-CoV-2 pandemic raised scientific awareness of the overlap between bacterial sepsis and viral-mediated hyperinflammation⁶⁰. Viral etiologies elicit unique responses from the innate and adaptive immune system; histopathological sections from patients with COVID-19 who developed AKI implicated thrombotic microangiopathy as one of the key mechanisms responsible⁶¹. There are also validated preclinical models to study viral response pathways via viral PAMPs (and known associations with specific TLRs) such as polyinosinic-polycytidylic acid (poly(I:C)), a synthetic analog to viral double-stranded RNA that binds to TLR3, and CpG, an unmethylated DNA motif found in viral and bacterial genomes that binds to TLR9⁶². The antiviral pathways induced by poly(I:C) and CpG have been well characterized in immune cells, but very little research has been conducted using these viral analogs to study intrarenal cellular responses⁶³. As such, further research into how acute viral infections increase the risk for SA-AKI, either in isolation or when present as a co-infection with bacterial pathogens, could better inform clinicians of the utility of screening for and treating viral infections as well as incorporating pathogen-specific information into future endotyping endeavors⁶⁴.

In addition to better understanding the heterogeneity of renal responses and injury to a variety of pathogens, characterizing the host response across ages, biological sex, and murine strain differences could improve translation to the patients most susceptible to poor outcomes⁶⁵. In murine models of AKI, it is common to use male C57BL/6 mice with an age range of 8-12 weeks⁴². This age range best correlates biologically with healthy, young adult males, which inherently limits generalizability to children, females, and older adults. Future preclinical studies including a broader range of animal ages and both sexes would likely increase translatability⁶⁶. Yet they must address the inherent challenges associated with younger and older models. Neonatal murine sepsis models rely on CS and LPS due to the technical challenges of performing CLP or tail vein injections to induce bacteremia due to their small size⁶⁷. A bacterial pneumonia or viral upper respiratory tract infection model could increase generalizability given the significant predisposition for and mortality associated with respiratory infections in infants and older adults. Still, these models may require a temporally related second hit to generate sufficient kidney injury (not unlike the multitude of nephrotoxic exposures patients may receive in clinical care)^49,68. Older aged rodents are more amenable to other sepsis models given their increased size, but their frailty puts them at greater risk for early mortality prior to experimentally important time points⁶⁶. Nonetheless, development as a biological variable warrants increased investigational attention to determine how aging modulates the risk for SA-AKI development, progression, and resolution⁶⁹.

The impact of biological sex on the pathophysiology of SA-AKI also remains an unanswered question. While epidemiological studies in adults and children demonstrate a slight male preponderance for developing sepsis, the representation of males-to-females in preclinical studies exaggerates this risk⁷⁰. Preclinical research suggests that females possess an estrogen-dependent protective effect from sepsis, whereas males are at risk for testosterone-mediated immunosuppression⁷⁰. Proposed protective effects of estrogen include reducing leukocyte-endothelial interactions as well as preventing myocardial dysfunction and the secondary sequelae of heart failure^71,72. In males, testosterone has been shown to reduce cell surface expression of TLR4, decrease neutrophil and macrophage activation, and impair the production of T_H1 lymphocytes that are critical to a robust cytotoxic immune response to infection⁷³. Despite these sex differences, clinical studies have not identified any clear mortality difference between sexes, nor has it been established whether males and female develop SA-AKI at different rates^74,75. This suggests that a complex balance of protective and deleterious effects mediated by sex steroids and/or sex chromosomes may contribute to the heterogeneity of outcomes and treatment responses among patients with SA-AKI. The 33rd Acute Disease Quality Initiative (ADQI) on Sex and Gender in AKI identified opportunities to improve our understanding of the protective role of sex in various AKI endotypes⁶⁵. Crucial steps include incorporating biological sex as a distinct variable in analyses and recognizing the potential role of pubertal and menopausal changes across study cohorts, both in clinical and preclinical studies⁷⁶.

Finally, one of the biggest challenges to translating discoveries from rodent sepsis models to humans is the relative lack of host diversity in preclinical sepsis models juxtaposed with enormous biological heterogeneity in humans. Preclinical sepsis studies typically use mice raised in pathogen-free or pathogen-limited environments, which hinders the development of a mature immune system and inadequately captures how a “dirty” host responds to infection⁷⁷. Additionally, gut microbiome diversity within strains and across facilities is significant, and this heterogeneity has also been shown to impact outcomes^78-80. With regards to mouse strain diversity, relying primarily on C57BL/6 mice constrains our understanding of how different mouse strains, like BALB/c, may uniquely respond to infectious challenges^81,82. Humanized mice offer the opportunity to study how a reconstituted human immune system responds during sepsis using the mouse as a host⁸³. This could be a valuable tool among investigators studying the impact of infiltrating immune cells into the kidney during SA-AKI development, but it is important to note that humanized mouse models maintain murine cell backgrounds for other important cell types such as tubular epithelial, podocyte, glomerular, microvascular, and stromal cell populations⁸⁴. Other considerations for models to study the host response in sepsis include large animals such as sheep and pigs, zebrafish, induced pluripotent stem cells, and kidney organoids⁴². While the limited genetic diversity of murine models can pose challenges for translational relevance, they offer a reductionist framework to dissect disease mechanisms with fewer confounding variables⁷⁷. As such, aligning pathogen and host selection to the scientific question remains crucial.

How might standardized sepsis treatment in humans inform preclinical sepsis model protocols?

Timely administration of antibiotics to achieve pathogenic source control has been a mainstay of clinical sepsis treatment for decades⁸⁵. When treating patients early in the course of sepsis, clinicians routinely opt for broad-spectrum antimicrobials until a pathogen has been identified⁸⁶. The antibiotic cocktail typically includes broad coverage for gram-positive and gram-negative bacteria, with additional coverage targeting anaerobic bacteria in select patients with suspected intraabdominal infections as well as fungal and parasitic coverage in patients with specific immunologic susceptibilities and exposures⁸⁷. Identifying a causative pathogen allows for targeted narrowing of antibiotics based on known susceptibility patterns. Antibiotics are occasionally employed in preclinical sepsis models, but this is less standardized than treatment approaches in humans⁸⁸. Common antibiotics used in preclinical models are described in Table 2 along with the type of bacteria they intend to treat. Of note, the administration of antibiotics is known to cause bacterial cell lysis and release of circulating PAMPs resulting in hyperinflammation, vascular vasoplegia, and shock in both humans and rodents⁸⁹. In preclinical models, the decision to administer antibiotics is multifactorial, but with regards to aligning with human sepsis, it is most dependent on the SA-AKI time point under investigation. For early SA-AKI time points intending to mirror septic shock on presentation, withholding antibiotics may be appropriate to model the pre-recognition phase of clinical sepsis. For later timepoints, particularly beyond 24 hours of sepsis induction, administration of antibiotics should be strongly considered unless there are concerns regarding the impact of antibiotics confounding a specific mechanism under investigation⁹⁰. Dosing intervals depend on the antibiotic selection, but antibiotic coverage for at least 48-72 hours is common in preclinical sepsis models and mirrors clinical standard of care⁹¹.

Table 2: Treatment Strategies for Murine Sepsis Models.

Common Therapeutic Agents in Sepsis
Antibiotics
Medication	Bacterial Activity	Murine Dose		Clinical Corollary	Preclinical Considerations
Gentamicin	Mostly Gram (−)	5-10 mg/kg Every 24h		Neonatal sepsis	Nephrotoxicity risk
Ceftriaxone	Gram (−), Most Gram (+) Not Pseudomonas	100-200 mg/kg Every 12-24h		Sepsis, Community Acquired Pneumonia
Cefepime	Gram (−), Most Gram (+) Pseudomonas	50-150 mg/kg Every 6-8h		Hospital Acquired Infection
Vancomycin	Gram (+)	100-400 mg/kg Every 12-24h		Methicillin-Resistant Staphylococcus aureus (MRSA) infection	Nephrotoxicity risk. Typically used to target MRSA or Enterococcus
Metronidazole	Anaerobes	30-50 mg/kg Every 8-12h		Anaerobic, intraabdominal infections	Commonly used in combination with ceftriaxone for broad coverage
Piperacillin/tazobactam	Gram (−), Gram (+), Anaerobes	100-400 mg/kg Every 6-8h		Resistant anaerobic, intrabdominal infections	Known to affect creatinine tubular handling
Meropenem	Gram (−), Gram (+), Anaerobes	50-100 mg/kg Every 8h		Resistant anaerobic, intrabdominal infections
Imipenem	Gram (−), Gram (+), Anaerobes	30-50 mg/kg Every 6-12h		Resistant anaerobic, intrabdominal infections
Fluids
Composition	Route	Murine Volume	Frequency	Clinical Corollary	Preclinical Considerations
0.9% Sodium Chloride	IV/IP/SC	40-100 ml/kg	Every 6-12h	Associated with hyperchloremic metabolic acidosis and increased rates of AKI	Similar findings have been seen in rodent models relative to balanced crystalloids
Lactated Ringer’s (or other balanced crystalloid)	IV/IP/SC			Closer to physiologic pH, has mortality benefit over 0.9% sodium chloride	Has mortality benefit over 0.9% sodium chloride
Vasoactive Medications
Medication	Mechanism/Receptor	Murine Dose		Clinical Corollary	Preclinical Considerations
Norepinephrine	α1 > β1	1-35 μg/kg/min147		Vasodilatory shock	Can decrease eGFR by causing renal afferent arteriolar vasoconstriction in humans
Epinephrine	β1, α1 > β2	0.05 mg/kg/hr148		Cardiogenic shock	Less commonly used than norepinephrine. Limited dosing data in mice
Vasopressin	V1	0.00057 IU/kg/min149		2nd line in adult septic shock. Some adult data to show renoprotective effects compared to catecholamines.	Can increase eGFR by causing renal efferent arteriolar vasoconstriction in humans.
Angiotensin II	ATR1, ATR2	600 ng/kg/min150		Refractory vasodilatory shock	Directly impacts RAAS system. Has immunomodulatory effects
Dopamine	D1, D2 > β1, α1	1 mcg/kg/min151		Vasodilatory shock with impaired renal perfusion	Commonly used in neonatal models. Has theoretical benefit of increasing renal blood flow

Abbreviations: MRSA, Methicillin-Resistant Staphylococcus aureus; IV, intravenous; IP, intraperitoneal; SC, subcutaneous; AKI, acute kidney injury; eGFR, estimated glomerular filtration rate, ATR, angiotensin II receptor; RAAS, renin-angiotensin-aldosterone system

The presence of hypotension and/or shock during sepsis increases the risk for multi-organ failure and death⁹². As such, clinicians measure basic hemodynamic variables such as heart rate and blood pressure to ascertain circulating blood volume and vascular tone, which can be optimized to achieve targeted organ perfusion pressures⁸⁶. In preclinical sepsis models, fluid resuscitation is often administered but with variable fluid selection, timing, amount, and frequency of administration (Table 2)⁹⁰. In the acute phase of septic shock, clinicians commonly administer a crystalloid such as 0.9% sodium chloride or Lactated Ringer’s solution⁹³. Depending on hemodynamic response to fluid resuscitation, the total volume administered can reach up to 100% or more of estimated circulating blood volume (5+ liters in adult humans, 80+ mL/kg in adult aged mice), although in clinical practice there is ongoing debate as to whether clinicians should be more judicious with fluid administration⁹⁴. Clinical data suggest that the resuscitative fluid composition can significantly impact the rates of SA-AKI and mortality, and thus, are variables that should be considered in preclinical sepsis models^95,96. Fluid resuscitation practices are also dependent on the region a patient lives and the pathogens common to that area. Specifically, while fluid resuscitation has been a mainstay of early sepsis therapeutic bundles across much of the world, studies in Africa suggest that a restrictive fluid strategy confers a survival benefit in their local healthcare settings⁹⁷. Therefore, preclinical sepsis models targeting pathogens or subphenotypes endemic to specific regions should consider whether a liberal or restrictive fluid resuscitation strategy is best aligned to the intended clinical corollary.

If a patient remains in shock despite reaching a targeted volume of fluid resuscitation, more advanced hemodynamic measures are employed to ensure perfusion goals are met. Clinicians will prescribe one of several vasoactive medications to optimize organ perfusion pressure by improving cardiac function and/or enhancing vascular tone⁸⁶. Close hemodynamic monitoring of blood pressure and heart rate is essential for careful titration of therapies to maximize benefits and minimize medication side effects⁹⁸. This has several implications for preclinical models of SA-AKI. First, considerations for measuring heart rate and blood pressure should be made when validating an animal sepsis model⁹⁰. While technically sophisticated, implantable telemetry to measure continuous heart rate and/or arterial blood pressure has been successfully used in mice as small as 20 grams⁹⁹. Second, if a preclinical sepsis model is being used that has been associated with hypotension in the literature, considerations for normalizing the blood pressure via fluid resuscitation or an implantable minipump embedded with a vasoactive medication (Table 2) could address hypotension as an important experimental confounder that, if present and unaddressed, might decrease the likelihood of translatability of novel mechanistic findings¹⁰⁰. It is important to note that each vasoactive medication has known direct and indirect effects on renal function, both in terms of modulating afferent and efferent renal arteriolar blood flow and contributing to the physiology of water and sodium homeostasis¹⁰¹. Recognizing that fluid resuscitation and/or administration of vasoactive medications are common in treating patients with sepsis, significant challenges exist to using these same strategies in animal sepsis models from technical and cost standpoints¹⁰². Therefore, while challenging, it may be beneficial to the scientific community as a whole if validation of each new sepsis model incorporated hemodynamic measurements at the time of characterization. This might allow subsequent investigators to tailor their treatment protocols to mitigate the risks of hypotension in a manner aligned to both the model and available resources.

Aligning preclinical animal sepsis models to human pathogenic sources, epidemiologic patterns, and treatment strategies will increase the relevance to human disease (Figure 1). Certain gaps in translation from bedside-to-bench are logistically untenable, such as keeping mice on ventilators or performing invasive hemodynamic monitoring in all preclinical sepsis experiments. While these therapies are better suited for larger animals such as sheep and pigs, the use of large animals is often cost prohibitive¹⁰². To bridge these gaps, many sepsis models can adopt a reverse translational approach by titrating the model for severity recognizing important mechanistic insights can be learned without requiring profound increases in serum creatinine¹¹. Pairing transcutaneous glomerular filtration rate measurements in mice via FITC-labeled sinistrin with urinary kidney injury biomarkers such as kidney injury molecule-1 (KIM-1) offers a more sensitive detection for sepsis induced kidney injury and/or reduced renal function compared to serum creatinine alone – a strategy that can be leveraged when using less severe sepsis models^103,104. Similar to the heterogeneity observed in humans, accepting there will be inherent variability in results generated from preclinical models despite best efforts to control for confounding variables may offer a unique opportunity to increase our collective knowledge of the complexity of mechanisms responsible for SA-AKI development.

Figure 1:

(Left) To enhance translation from preclinical murine sepsis models to human disease, investigators should consider expanding the host variability across age, biological sex, and genetic strains; mirror clinical sepsis treatment bundles through the administration of antibiotics and fluid resuscitation; closely monitor vital signs such as heart rate, blood pressure, and core body temperature paired with mitigating therapies such as vasoactive medications and temperature control; and embrace artificial intelligence and data sharing practices to tie novel mechanistic insights to clinically relevant endotypes. (Right) A theoretical experimental timeline that incorporates many of these translationally aligned strategies to study early SA-AKI, the SA-AKI to AKD transition, and/or SA-AKI to CDK progression.

Opportunities for Innovation

Mechanistic underpinnings of organ crosstalk

It is well recognized that humans and animals with severe sepsis are at high risk for muti-organ failure, including SA-AKI, which is associated with increased mortality rates^105,106. While circulating inflammatory mediators induce aberrant pro- and anti-inflammatory pathways specific to each organ, investigations into the identification of secondary signaling pathways that contribute to organ crosstalk have been better elucidated recently^107-110. Axes of interest in SA-AKI pathobiology include kidney-heart, kidney-lung, kidney-liver, kidney-brain (Figure 2)^111-113. Although circulating mediators and signaling pathways have been identified that directly tie the injured kidney to each of these extrarenal organs, they have not translated to clinically useful biomarkers, suggesting a need for a deeper mechanistic understanding of the pathologic versus adaptive roles at play during organ crosstalk ¹¹⁴. A biomarker will have enhanced clinical utility if it allows for earlier identification of patients at risk for worsening multi-organ failure with subsequent options for altering management to intervene upon maladaptive pathways¹¹⁵. For example, if a patient with sepsis develops respiratory failure and the clinician has access to a biomarker informed by kidney-lung crosstalk that can perform risk stratification for developing SA-AKI, then the clinician might be able to tailor treatments that balance the risks and benefits to both the lungs and kidneys rather than just the lungs¹¹⁶. This strategy would also allow biologically informed therapies to be developed and validated within the framework of organ failure patterns, similar to the phenotypes characterized by Seymour et al⁴⁵. Clinical trials could then enrich patients who are most likely to benefit from a novel therapy targeting the mechanistically informed endotypes identified at the bench¹¹⁷.

Figure 2:

(Left) Extra-renal crosstalk between the kidney and the brain, lungs, heart, and liver are areas of investigation that could lead to the identification of novel SA-AKI endotypes with mechanistically informed therapeutic candidates. (Right) Within the kidney, cellular crosstalk between tubular epithelial cells, infiltrating and resident immune cells, and renal vascular endothelial cells is essential to maintain the homeostatic functions of the kidneys. Gaining clarity on how these intrarenal communication pathways differentially respond in SA-AKI may help to identify new diagnostic biomarkers and treatment approaches.

Within the kidney, expanding investigations into cellular crosstalk between tubular epithelial, immune, and endothelial cells may help to identify novel paracrine axes that can be therapeutically manipulated to provide renal protection and accelerate repair during sepsis (Figure 2). Thus far, most of the research into intrarenal crosstalk has focused on tubular-immune interactions during acute injury and early maladaptive repair¹¹⁸. Infiltrating and resident immune cell populations are dynamically responsive to damaged tubular epithelial cells, but this response is often exaggerated and deleterious¹¹⁹. Based on discoveries at the bench, attempts to treat AKI by modulation of immune cells in preclinical models are ongoing¹²⁰. While still understudied, interactions between renal endothelial cells and tubular epithelial cells are growing in characterization across the literature¹²¹. The endothelium is primed for expanded investigations into SA-AKI pathophysiology given its role in immune cell infiltration into local areas of tubular injury, its participation in the coagulation cascade, and its crucial responsibility in facilitating secretion and reabsorption of solutes in tandem with nearby tubular epithelial cells. Outside of the kidney, endothelial pathobiology is a significant research priority in sepsis, given the well-characterized endothelial activation and injury leading to pathologic extravascular fluid accumulation and thrombotic microangiopathies that are strongly associated with poor outcomes in sepsis¹²². Research into the role of renal endothelial activation in sepsis is crucial to increasing our understanding of how therapeutic agents targeting endothelial dysfunction might impact kidney function. Finally, tubular-tubular crosstalk is essential for homeostatic functions and danger signaling¹²³. Increased efforts are necessary to reveal how injury to the highly metabolic proximal tubule epithelial cells impacts downstream distal tubule reabsorption and other paracrine mediated tubular-tubular interactions. Tubular segment-specific biomarkers may expand our understanding of tubular-tubular interactions, enhance AKI risk-stratification, produce preventative and therapeutic strategies, and facilitate assessment of renal recovery versus risk for future injury¹²⁴.

Leveraging the power of big data techniques to delineate endotypes from mechanisms

One of the biggest areas of innovation in clinical research over the past decade has been the use of artificial intelligence (AI) to process large amounts of heterogeneous, population-level data¹²⁵. These analytical strategies identify powerful associations between clinical datapoints that are difficult to explain at a pathophysiological level and can often outperform clinicians in their ability to identify patients at risk for AKI development¹²⁶. From a dataset of over 700,000 adult patients, recurrent neural networks were used to continuously model a 48-hour AKI prediction probability that accurately predicted >90% of AKI events that subsequently required administration of dialysis¹²⁷. This model also predicted changes in AKI related biomarkers and offered confidence assessments on AKI prediction, theoretically providing clinicians with useful data to make informed clinical decisions that can be tested in future studies to determine if this strategy can mitigate AKI severity. On a mechanistic level, incorporating human blood RNA sequencing from septic patients with clinical data into an unsupervised machine learning algorithm led to the identification of 5 distinct inflammatory based endotypes with varying sepsis severity trajectories and a proposed framework for the deployment of precision therapies¹²⁸. However, as the use of AI becomes more ubiquitous in clinical research, efforts to reveal novel biological relationships exposed by “big data” techniques must remain focused on the identification of clinically meaningful endotypes rather than creating a patchwork quilt of siloed, AI-generated disease categories that lack broad generalizability for use by clinicians at the bedside¹²⁹. These examples demonstrate the tremendous potential of artificial intelligence to enhance clinical practice, yet even greater insight may be gained by linking SA-AKI mechanisms revealed at the bench to clinical endotypes presenting at the bedside.

Translational and basic science endeavors have also adopted artificial intelligence to rapidly advance the field of -omics, from high-dimensional spatial transcriptomic profiling to integrating transcriptomic, proteomic, epigenomic, lipidomic, metabolomic, and histopathologic data to ascertain complex biological interactions in SA-AKI¹³⁰. Artificial intelligence has also been used to disentangle the complexity of human biology that is challenging to model with traditional bench experiments. For example, the development of a general expression transformer (GET) to ascertain the transcriptional relationship between transcription factors, chromatin accessibility, and other gene regulatory mediators has uncovered previously uncharacterized transcriptional phenomena that correlate directly with disease manifestation across numerous cell types¹³¹. As an aspirational example of a translational research endeavor, the Kidney Precision Medicine Project (KPMP) was established as a multicenter collaborative with a specific goal “to define disease heterogeneity and determine the precise molecular pathways that will facilitate identification of specific drug targets and ultimately enable individualized care for people with AKI and CKD”¹³². Through integrated multi-omic profiling of biological specimens from healthy individuals and those with AKI and CKD, the KPMP has built a translational framework that spans from molecular analysis to the identification of disease subgroups. Although the KPMP currently includes limited specimens from patients with SA-AKI, its collaborative, multi-omic model could be adapted to create a comparable, publicly accessible database derived from preclinical sepsis models. Such a resource, with its robust interrogation of specific genes and pathways, could harness artificial intelligence to uncover novel mechanistic insights. By integrating high-dimensional biological data into AI-driven models of SA-AKI pathogenesis and prediction, basic and translational scientists will play a critical role in identifying molecular signatures, defining endotypes, and discovering therapeutic targets.

One common barrier to sepsis researchers is the inconsistent, highly variable spectrum of results produced by distinct sepsis models and preclinical hosts of interest¹³³. However, this variability often mirrors the clinical heterogeneity of sepsis in humans that challenges clinicians¹³⁴. In fact, there is growing acceptance of and intentionality to design preclinical sepsis experiments that incorporate more variables to recapitulate sepsis heterogeneity. This has led the scientific community to produce consensus guidelines stating this goal³⁵. Strategies include utilizing a broad range of host factors (age, biological sex, strain) as well as performing post-hoc risk stratification in murine studies^135,136. For example, classification and regression tree analysis can incorporate numerous categorical or continuous variables such as biomarker levels, histology and/or animal behavior scores, and measured GFR to stratify mice for risk of an outcome of interest in a training set that can then be validated prospectively in validation cohorts¹³⁷. Lasso regression is another strategy that identifies predictive variables for risk-stratification with potentially less risk for overfitting¹³⁶. Risk stratification has been successfully employed to identify distinct sepsis subphenotypes in mice with demonstration of heterogeneity of treatment effect based on those subphenotypic assignments¹³⁸. Murine sepsis randomized trials have also been designed to align with clinical observations, which demonstrates the translational potential for leveraging murine sepsis heterogeneity to advance our understanding of sepsis phenotypes¹³⁹.

Expanding the connection from SA-AKI to CKD

The recognition and appreciation for the burden of the AKI-to-CKD transition is harmonized across basic, translational, and clinical research. However, despite sepsis serving as the most common instigator of AKI among hospitalized patients, a large gap remains in understanding how SA-AKI leads to CKD in a subgroup of patients. In clinical practice, the term acute kidney disease (AKD) has increased in prominence and is defined as kidney dysfunction that persists between 7 and 90 days post-insult¹⁴⁰. While almost half of patients with sepsis develop SA-AKI, nearly 20% progress to AKD, and patients with relapsing or persistent AKD are at increased risk for developing de novo CKD¹⁴¹. In preclinical sepsis models, early SA-AKI time points (within 48 hours of insult) are the predominant focus, which precludes extrapolation to AKI-to-CKD mechanisms¹⁴². Other murine AKI models have implicated complex interactions between tubular epithelial cells, immune cells, and fibroblasts as primarily responsible for maladaptive repair^143,144. It is also recognized that sepsis-mediated inflammation is distinct from other inflammatory kidney insults which will likely impact the mechanisms responsible for the SA-AKI to CKD transition¹⁴⁵. Dedicated efforts are needed to study the trajectory of tubular recovery, innate and adaptive immune responses, and fibrotic remodeling following sepsis induction in mice. New knowledge regarding the AKI-to-CKD transition in sepsis could arm clinicians with valuable information on post-AKI follow up needs with considerations for long-term surveillance of kidney function¹⁴⁶.

Conclusion

Despite the clinical perception of stagnation in sepsis and SA-AKI therapeutic innovation, promising translational opportunities are on the horizon, but collaboration and data sharing amongst preclinical and clinical researchers is necessary to realize these advances. Preclinical model alignment to human disease will increase opportunities for novel discoveries at the bench to be applied in clinical practice. Embracing the heterogeneity of sepsis, both in animal models and in humans, and leveraging the power of machine learning to perform risk stratification as part of the analytical approach may shed light on which therapies are most likely to be efficacious, the time points by which they should be administered, and the patients in whom they are most likely to benefit. Identifying specific pathways that are responsible for endotypic variation and targeting those pathways in enriched patient populations will allow for precision in diagnosis and treatment. Finally, increasing efforts to discover the complex interplay of mechanisms responsible for SA-AKI to CKD transition will be critical to improving long-term outcomes in patients with SA-AKI. These endeavors require close partnerships between scientists and clinicians and the rigorous translation of mechanistic discoveries to endotype-specific therapies.