The nomenclature of TDNT resin follows the format (density)PEGA-(charge group)_#valency (hydrophobic groups)_#valency. For example, 10%PEGA-COOH₄C4₄ indicates that the free amino groups on the commercial PEGA resin (0.4 mmol/g) were downscaled to 10% (with 90% blocked by acetylation) for the conjugation of a TD structure containing four carboxylic acid moieties and four butyryl groups on the periphery of the dendritic TD scaffold.

The sepsis cohort in this study consists of patients ≥ 18 years old with surgical sepsis (gastrointestinal perforation or ischemia) who required surgery and were admitted to the SICU at Upstate Medical University Hospital in Syracuse, NY, USA (IRB No. 1321635, approved 2 November 2018). Sepsis-3 definitions of sepsis/septic shock were used to calculate the Sequential Organ Failure Assessment (SOFA) score using the following variables: PaO₂/FiO₂ ratio, Glasgow Coma Scale (GCS) score, mean arterial pressure (MAP), administration of vasopressors with type, dose, and infusion rate, serum Cr (mg/dL) and urine output, bilirubin level (mg/dL), and platelet count (3). A SOFA score ≥ 2 was used as the cutoff for inclusion in the study. After meeting the initial screening criteria, informed consent was obtained. A total of 10 mL of blood was collected from an indwelling catheter already in place (e.g., arterial line, PICC line, or central venous catheter) and immediately placed on ice to reduce cellular stress under hypoxia condition ex vivo. Blood was centrifuged, and plasma was collected within 1 h after blood collection. Plasma samples were stored at − 80 °C for TDNT resin screening. Systemic Immune-Inflammation Index (SII), a biomarker used to determine inflammation and thrombosis status and developed as a predictor of adverse outcomes in several disease processes, including sepsis (38), was calculated for each patient with available clinical data at the time of blood collection using the following equation:

The patient cohort for TDNT resin validation (Table 1) consisted of 20 patients with surgical sepsis (nine men, 11 women), with an average age of 64.2 years, an average SOFA score of 9.4, and an SII score of 2,738.6. The mortality rate in this cohort was 60%. Given the limited cohort size and the ex vivo nature of the experiments, gender was not considered a variable for analysis.

THP-1 human monocytic cells (ATCC) were cultured in RPMI medium (Cytiva Life Sciences, Marlborough, Massachusetts, USA) with 10% FBS and 1% penicillin/streptomycin in a humidified 5% CO₂ incubator at 37 °C. Cells were plated at 5 × 10⁴ cells/well in a 96-well plate and differentiated into macrophages with 10 ng/mL phorbol 12-myristate 12-acetate (PMA) (Alfa Aesar, Ward Hill, Massachusetts, USA) for 24 h. The medium was refreshed, and cells were stimulated with 50 ng/mL lipopolysaccharide from Escherichia coli (Sigma-Aldrich, Burlington, Massachusetts, USA) for 24 h. Supernatant was collected, pooled, and stored at − 80 °C for future use.

MILLIPLEX^® Human Cytokine/Chemokine/Growth Factor Panel A (Millipore, Burlington, Massachusetts, USA, HCYTA-60K-18; Granulocyte-macrophage colony-stimulating factor (GM-CSF), IL-10, IL-13, IL-17A, IL-18, IL-1RA, IL-1a, IL-1B, IL-2, IL-3, IL-4, IL-6, IL-7, IL-8, MCP-1, Macrophage Inflammatory Proteins (MIP)-1α, Macrophage Inflammatory Proteins (MIP)-1β, tumor necrosis factor alpha [TNF-α]) was performed according to manufacturer’s protocol using the MAGPIX Luminex xMAP^® system (Millipore) for cytokine analysis. Luminex xPONENT^® software version 4.3 (Millipore) was used for data acquisition. For mixed plasma library screening, 14 cytokines—five positively charged (IL-8, IL-10, IL-17A, MCP-1, and interferon-γ [IFN-γ]) and nine negatively charged (GM-CSF, IL-1α, IL-1β, IL-6, MIP-1α, MIP-1β, TNF-α, IL-27, and IL-1RA)—were included for analysis based on the detectable cytokines present in the pooled plasma. For individual sepsis patient profiling and screening, 15 cytokines—six positively charged (IL-7, IL-8, IL-10, IL-13, IL-17A, and MCP-1) and nine negatively charged (GM-CSF, IL-1α, IL-1β, IL-1RA, IL-6, IL-18, MIP-1α, MIP-1β, and TNF-α)—were included to cover more cytokines detected in individual patient plasma. Cytokines below the level of detection of the assay were classified as 0 for the purpose of calculation.

Cytokine profiles for mixed patient plasma and individual sepsis patient plasma were determined using a multiplex cytokine assay before and after TDNT resin incubation to evaluate individual cytokine adsorption and cytokine profile clearance by TDNT resins. Total cytokine represents the sum (by weight, Figure 1B) of all cytokines; positively charged cytokines represent the sum of all positively charged cytokines with an isoelectric point (pI) > 7.4 (Table 2); and negatively charged cytokines represent the sum of all negatively charged cytokines with a pI < 7.4 (Table 2), based on published predicted pI values in the PhosphoSitePlus database (39). The isoelectric point cutoff for classifying cytokine charge was determined based on the average human blood pH of 7.4. For individual sepsis patient plasma, additional analysis relating cytokine profile to mortality and clinical markers (e.g., SOFA score and SII) was performed. Patients were divided into high/low cytokine, SOFA score, and SII status. Cytokine levels at the 50th percentile (14,318 pg/mL for total cytokines, 4,055 pg/mL for positively charged cytokines, and 10,263 pg/mL for negatively charged cytokines), a SOFA score of 6.5 (40), and an SII of 1,767 (38) were used as cutoffs to determine high/low status.

Resin cytokine clearance efficacy (CE) for total, positively charged, and negatively charged, and individual cytokines from mixed or individual sepsis patient plasma was calculated for each resin using the following equation:

Additionally, ratios of clearance efficacy of positively charged/negatively charged cytokines and negatively charged/positively charged cytokines were calculated to evaluate resin charge selectivity for cytokine removal in individual sepsis plasma samples.

After 4 h of incubation on a rotator at room temperature with or without TDNT or commercial resin, the total protein concentration of the plasma was measured via UV absorbance at 280 nm using a NanoDrop. Percent plasma protein removal by resins for each patient sample was calculated using the following equation:

De-identified clinical information, e.g., SOFA score and SII score, was collected on the day of blood collection. Patient outcomes, e.g., hospital mortality due to sepsis, were followed until discharge. All de-identified patient data were managed according to the IRB protocol. Results are reported as mean ± SD. Graphs and statistical analyses were performed using GraphPad Prism Software, version 10.2.3 (GraphPad Software). Cytokine clearance efficiency data between resins were compared using one-way ANOVA followed by Dunnett’s multiple comparisons test. Results were considered significantly different when p < 0.05.

We engineered TDNT with diverse charge and hydrophobic combinations, as well as varying TD density on the PEGA resin, to screen TDNT for charge selectivity in protein binding (Figure 1A; Supplementary Figure 1). Two abundant proteins with a molecular weight < 30 kDa and a pI either higher or lower than neutral pH 7.4, chosen to mimic cytokines, were selected to screen the TDNT resin library (Supplementary Figure 2). Alpha-lactalbumin (αLA) from bovine milk (MW: 14.2 kDa, pI: 4.8, Sigma-Aldrich) served as a negatively charged model protein to mimic key proinflammatory cytokines (TNF-α, IL-1β, IL-6, etc.), while lysozyme (MW: 14 kDa, pI: 11.4, MP Biomedicals, Solon, Ohio, USA) served as a positively charged model protein to mimic anti-inflammatory cytokines (IL-10, IL-4, etc.). Through the adsorption screening studies on these model proteins, we confirmed that TDNT resins with charged and combined charged/hydrophobic moieties preferentially capture proteins with opposite charges (Supplementary Figure 3). At the same time, lower protein-capturing capacity was observed in TDNT resins containing only hydrophobic moieties, which resembles the molecular mechanisms of conventional polystyrene-based hemoperfusion devices (41). In general, the density of TD in the hydrogel resin synergizes with charge-only TDNT in capturing proteins with opposite charges. High TD density reduces protein-capturing efficiency in TDNT with both hydrophobic and charge moieties, especially for long fatty acid moieties, due to reduced swelling properties of the hydrogel resin in water. These results suggest that TDNT resins with charged and combined charged/hydrophobic moieties are more effective for charge-selective protein capture and require further optimization and validation for capturing cytokines in patient blood.

During sepsis, excessive cytokines are produced and self-perpetuated, a phenomenon known as a “cytokine storm”, which significantly contributes to immune dysregulation, organ failure, and early death in some patients (5–8). To validate the effectiveness and selectivity of TDNT resins for the removal of various cytokines in patient blood, we screened a library of TDNT formulations using sepsis patient plasma. Blood was collected, and plasma was isolated from a cohort of surgical sepsis patients at Upstate University Hospital with gastrointestinal perforation or ischemia. Given the heterogeneity of sepsis patients’ immune response, individual patients may have vastly different cytokine profiles. Plasma from nine sepsis patients was therefore mixed to ensure the presence of key cytokines involved in the innate immune response and to allow a normalized comparison between different resins. Additionally, mixed plasma was spiked (spiked-mixed plasma) with supernatant from LPS-stimulated THP-1 macrophages to ensure the presence of early-phase mediators, e.g., TNF, which may otherwise be low in later-stage sepsis patients and hinder the evaluation of their removal in TD resin screening. Using spiked-mixed plasma, we performed TD resin screening to identify the lead TDNT resins with different cytokine binding profiles (1): pan-affinitive (2), positively charged-cytokine selective, and (3) negatively charged-cytokine selective. As shown in Figure 1A, several key cytokines were detected at representative concentrations in the spiked-mixed septic plasma: TNF-α (4.14 ng/mL), IL-6 (7.25 ng/mL), IL-1β (0.48 ng/mL), and IL-10 (0.25 ng/mL). Notably, IFN-γ and IL-17A were only minimally detectable at 2–10 pg/mL, even in the spiked-mixed plasma of this specific cohort of severe patients with surgical sepsis. These cytokines are primarily secreted by activated T cells and NK cells in infections and autoimmune diseases (42, 43) and can be downregulated by lymphopenia in acute sepsis (44). Interestingly, very high chemokine concentrations were detected in the mixed plasma, e.g., MIP-1α (5,147 ng/mL; pI: 4.77) and IL-8 (104 ng/mL; pI: 9.1), as well as MIP-1β (6 ng/mL; pI: 5.13) and MCP-1 (5.5 ng/mL; pI: 9.4). In sepsis, these circulating chemokines are associated with organ damage and coagulopathy; however, they are not robust biomarkers for mortality in septic shock.

TDNT resins were incubated with spiked-mixed sepsis patient plasma for 4 h at room temperature on a rotator. Blank PEGA control resin and a commercial MG250^® resin (hydrophilic-coated macroporous crosslinked polystyrene resin) were applied as negative and positive controls for comparison. After incubation, samples were centrifuged, and the supernatant was collected for multiplex cytokine analysis (Figure 1C). Spiked plasma without resin incubation served as an untreated reference for calculating resin cytokine clearance efficacy. The multiplex cytokine panel included nine negatively charged cytokines (pI < 7.4) and five positively charged cytokines (pI > 7.4) produced during sepsis inflammation. Resin cytokine clearance efficacy was calculated for each individual cytokine (Figure 2A). Similar to the model protein binding shown in Supplementary Figure 3, TDs with only charge or hydrophobic moieties generally exhibit poor cytokine adsorption, except for TDs with higher charge density (OA₈ or Arg₄ at 100%) or longer lipid chains (C17₄ at 10%). As shown in Figure 2A, clusters of TD resins, indicated by different colored circles, can be identified for selective cytokine adsorption based on charge disparity, although this selectivity is not as exclusive as observed for model protein adsorption, i.e., αLA and lysozyme in Supplementary Figure 3. Hydrophobic interactions may play an important role in specific cytokine binding, which can compromise charge selectivity. In addition, the concentrations of individual cytokines may impact removal efficiency due to the dynamic binding between cytokines and serum proteins or TD nanotraps. However, no significant trend was observed based on cytokine concentrations. For example, the abundant and positively charged chemokine IL-8 (CXCL-8) could be effectively removed by almost all types of TD resins, whereas the more abundant negatively charged chemokine MIP-1α (CCL-3) or trace amounts of positively charged IFN-γ could only be removed by TDNT resins with a specific charge/lipid combination. Additionally, IL-1α and IL-1β, both present at moderate concentrations, were only moderately removed by all TD resins and the commercial MG250^® resin via nonspecific hydrophobic interactions.

In order to further explore cytokine profile selectivity, the capacity of resins to adsorb groups of positively charged cytokines (IL-8 + IL-10 + IL-17A + MCP-1 + IFN-γ) or negatively charged cytokines (GM-CSF + IL-1α + IL-1β + IL-6 + MIP-1α + MIP-1β + TNF-α + IL-27 + IL-1RA), as well as total cytokines, was analyzed in Figure 1B. Blank PEGA resin had very low cytokine clearance efficacy (1.27% of total cytokines), whereas the commercial resin MG250^® showed high cytokine clearance efficacy (99.80% of total cytokines) but lacked selectivity for differently charged cytokines (Figure 1B). To identify the most promising TDNT resin formulation with a pan-affinitive cytokine removal profile, total cytokine removal efficacy was compared. Several promising candidates were identified among the formulations containing a combination of charged and hydrophobic moieties that removed most cytokines from the spiked-mixed plasma. We identified 100% PEGA-OA₄C17₄ as the lead pan-affinitive TDNT resin, with a total cytokine removal efficacy of 99.87% (Figure 1B). To identify TDNT resin formulations with the most promising negatively charged cytokine-selective and positively charged cytokine-selective removal profiles, we compared the removal efficacy of all negatively charged cytokines with that of all positively charged cytokines. Thus, 100% PEGA-Arg₄C4₄ was the only TDNT formulation in the library that exhibited a negatively selective cytokine removal profile, removing 99.77% of negatively charged cytokines and only 25.65% of positively charged cytokines (Figure 2B). While several negatively charged and negatively charged/hydrophobic TDNT formulations effectively removed positively charged cytokines with minimal removal of negatively charged ones, 100% PEGA-COOH₄C4₄ displayed the most promising positively charged cytokine-selective removal profile, removing 91.14% of positively charged cytokines and only 2.89% of negatively charged cytokines (Figure 1B). The combination of charged moieties with the smaller, less bulky C4 hydrophobic group allows improved charge selectivity while maintaining efficacy through the synergistic combination of charged and hydrophobic interactions for cytokine removal. The smaller size of C4 results in fewer hydrophobic interactions than larger groups, such as C17, thereby reducing nonspecific protein adsorption and allowing a more charge-selective binding profile. Based on the screening of the TDNT resin library for cytokine removal efficacy in spiked-mixed sepsis patient plasma, we identified PEGA-OA₄C17₄ as the lead pan-affinitive ((pan)TDNT), PEGA-Arg₄C4₄ as the lead negatively charged cytokine-selective ((−)TDNT), and PEGA-COOH₄C4₄ as the lead positively charged cytokine-selective ((+)TDNT) formulation for further evaluation of efficacy in individual patient samples to promote effective clinical translation.

To effectively evaluate the therapeutic potential of these three lead TDNT resins for cytokine removal in individual sepsis patient plasma samples, it is critical to understand the cytokine profile and clinical status of the patients in the cohort. While extensive studies have evaluated the importance and prognostic value of individual cytokines such as TNF-α and IL-1β (45, 46) in the sepsis disease process, studies assessing the broader cytokine profile in sepsis patients are limited and often focus only on the combined prognostic value of a few key cytokines, e.g., IL-6 + IL-8 + MCP-1 + IL-10 (47). Here, we take a unique approach to examine a broader cytokine profile in plasma to categorize a cohort of 20 surgical sepsis patients (Table 1) based on the total cytokine level, positively charged cytokine profiles, and negatively charged cytokine profiles for subsequent validation of TDNT resin efficacy for cytokine removal (Figure 1D).

Using multiplex cytokine analysis, plasma levels of 15 cytokines (nine negatively charged [GM-CSF + IL-1α + IL-1β + IL-1RA + IL-6 + IL-18 + MIP-1α + MIP-1β + TNF-α] and six positively charged [IL-7 + IL-8 + IL-10 + IL-13 + IL-17A + MCP-1]) were measured in each of the 20 sepsis patients and one healthy control plasma sample. Despite the cohort having a common septic source (gastrointestinal perforation or ischemia), significant variability was observed among the 20 patients’ cytokine profiles. Total measured cytokines ranged from 1,656 to 46,049 pg/mL, positively charged cytokines ranged from 502 to 14,084 pg/mL, and negatively charged cytokines ranged from 688 to 35,004 pg/mL (Figure 3A). When comparing cytokine levels in survivor vs. nonsurvivor patients, no significant differences were observed in total cytokine levels or in subgroups of cytokines based on charge disparity (Figure 3B). In septic nonsurvivors, both total cytokines and proinflammatory cytokines appear to deviate to either extremely high or low concentrations, indicating severe immune dysregulation, whereas cytokines in sepsis survivors are mostly clustered within mid-range levels. In addition, positively charged anti-inflammatory cytokines are higher in sepsis nonsurvivors (Figure 3B). Although cytokine profiles need to be confirmed in additional large cohort studies, this trend aligns with the clinical observation that both hyperinflammation and immunosuppression contribute to mortality in sepsis through either multiple organ failure or secondary infections, respectively (48–51). Immune phenotyping in sepsis is a complex clinical challenge (52–55), and systematic studies correlating patient cytokine profiles with clinical phenotypes and sepsis outcomes remain lacking. Current methods of cytokine analysis are slower than the rapid clinical progression of sepsis. In addition, specific cytokines, e.g., IL-6, may only serve as biomarkers for sepsis development, as the attenuation of specific cytokines lack clinically efficacy in reducing sepsis mortality in many large clinical trials (13, 14, 48). Therefore, the characterization of cytokine profiles in sepsis patients and the potential attenuation of groups of cytokines via engineered TDNT resins, by targeting cytokine charge disparity, may provide a novel strategy to stratify patients and enable effective and precise immune modulation therapies for sepsis patients.

To precisely diagnose the status of sepsis, both ongoing humoral immunity (secreted macromolecules in body fluids) and current cellular immunity (immune cell status) need to be evaluated in combination with organ damage (results of hyperinflammation) for improved patient management. These three parameters are interwound in determining patient outcomes in sepsis, which is further complicated by the presence of presepsis comorbidities. The SOFA score is widely used to quantify organ damage in critically ill patients and to identify those at risk. It evaluates organ dysfunction across six systems: respiratory, cardiovascular, neurological, hepatic, renal, and coagulation (40, 56, 57). Recently, the SII, which uses platelet counts (P), lymphocyte counts (L), and neutrophil counts (N) as a biomarker (SII = P × N/L) to determine inflammation and thrombosis status, has been developed as a predictor of adverse outcomes in several disease processes, including sepsis (38). Both increased SOFA score and increased SII are strongly correlated with a higher risk of mortality in sepsis patients (38, 57). Given that the SOFA score and SII are predictive of mortality in sepsis, we sought to determine whether these clinical markers have a potential relationship with patient cytokine profiles, which could provide insight into how TDNT resin therapy might be implemented in the future.

In order to study the cytokine profiles in combination with clinical diagnosis, we applied thresholds for SOFA and SII identified in large retrospective cohort sepsis studies to divide the small cohort of surgical sepsis patients into high- and low-risk groups: a SOFA score of 6.5 (40) and an SII of 1,767 × 10⁹/L (38). Patients were further divided into high- or low-cytokine groups using the 50% cytokine level in the cohort. Consequently, patients were classified into four phenotypes based on cytokine level and SOFA score: low cytokine/low SOFA, low cytokine/high SOFA, high cytokine/low SOFA, and high cytokine/high SOFA (Figures 3C, E). Similarly, patients were divided into low cytokine/low SII, low cytokine/high SII, high cytokine/low SII, and high cytokine/high SII to study SII–cytokine correlations (Figures 3D, F). Furthermore, patients were divided into subgroups based on the profiles of positively or negatively charged cytokines (Figures 3F, G, I, J). The mortality rate was also calculated for each sepsis phenotype in the patient cohort (Figures 3E–J). We observed a trend of increasing SOFA score correlating with higher total (Figure 3E), positively charged (Figure 3F), and negatively charged cytokines (Figure 3G). No patient in the cohort fell into the region with low SOFA/high cytokine, highlighting the association of cytokine storm with organ damage. The high SOFA/low cytokine group had the lowest mortality rate of 45% compared with both high SOFA/high cytokine and low SOFA/low cytokine, 72% and 75%, respectively (Figures 3E–G). These observations, however, need to be validated in a large cohort of patient populations with diverse pathogenesis and comorbidities to confirm the trend.

The reference ranges of SII in healthy controls have been reported between 142 and 804 × 10⁹/L in a large retrospective cohort study of a healthy adult population (58) and can increase during infection and inflammation. Both reduced and elevated SII indicate dysregulated immune responses, with either neutropenia/thrombocytopenia or myelosuppression/lymphopenia, which are linked to poor clinical outcomes. A retrospective cohort study of ICU patients revealed a J-shaped relationship between SII and 28-day mortality risk. An SII of 774.46 × 10⁹/L was correlated with the lowest mortality risk (59). Patients with SII greater than 1,767 × 10⁹/L were at higher risk of sepsis mortality (38), which was thus applied to subdivide the surgical sepsis patient cohort into all four SII/cytokine phenotypes. Here, we observed a trend of decreasing cytokine levels with increasing SII. In contrast, both high- and low-cytokine profiles were observed in sepsis patients with low SII, indicating greater disease heterogeneity. Patients with either high- or low-cytokine profiles and low SII had a high mortality rate of 67%. Interestingly, only the high SII/low-cytokine group exhibited a lower mortality rate of 45% (Figures 3H–J), with a noticeable cluster of five survivors with low cytokine and moderate SII between 2000 and 2500. Due to the small sample size, the use of a single timepoint for analysis, and high overall mortality (60%, Table 1), our analysis lacks statistical power to draw definitive conclusions regarding mortality trends in different SOFA/cytokine and SII/cytokine phenotypes. Despite the lack of statistical power, it is noteworthy that patients with low cytokine levels combined with both high SOFA and SII exhibited the lowest mortality rates. This observation suggests a potential survival benefit from removing these cytokines to reduce their levels in these severe patients with medium/high SII and high SOFA scores. Future analysis of a large cohort of sepsis patients is warranted to confirm the validity of trends and the potential for using cytokine profiles combined with clinical markers of SII and SOFA scores to guide personalized medicine therapies.

Given the heterogeneous nature of the sepsis disease process and the high variability of cytokine profiles even among patients with similar septic sources (Figure 3), it is necessary to validate the efficacy of the lead positive-cytokine-selective (+)TDNT, negative-cytokine-selective (−)TNDT, and (pan)TDNT identified using spiked-mixed sepsis patient plasma with individual patient samples for comparison with commercial resin. Lead TDNT resins were incubated with individual plasma samples from 20 sepsis patients for 4 h at room temperature on a rotator to mimic the clinical duration of hemoperfusion therapy. Plasma from seven of those patients was incubated with the commercial MG250^® resin as a positive control. After incubation, samples were centrifuged, and the supernatant was collected for multiplex cytokine analysis (Figure 1D). Sepsis plasma without resin incubation served as a baseline reference for calculating cytokine clearance from each patient’s plasma. The multiplex cytokine panel included 15 cytokines in total: nine negatively charged cytokines (pI < 7.4) and six positively charged cytokines (pI > 7.4), as described for cytokine profiling in Figure 3.

The positive-cytokine-selective (+)TDNT removed cytokines with a positive charge (pI > 7.4) more effectively than those with a negative charge (pI < 7.4) (Figures 4A, 5A, F, G). When examining individual cytokines, (+)TDNT most effectively removed IL-8, IL-13, and IL-17A (Figure 4A; Supplementary Figure 4), while showing low efficacy for removal of IL-1α, IL-1RA, IL-6, and IL-18 (Figure 4A; Supplementary Figure 5). (+)TDNT had moderate cytokine removal efficacy for all other cytokines evaluated. Conversely, the negative-cytokine-selective (−)TDNT more effectively removed cytokines with a negative charge compared to those with a positive charge (Figure 4B; Supplementary Figures 5B, F, G). When examining individual cytokines, (−)TDNT most effectively removed IL-1RA, IL-18, and MIP-1β (Figure 4B; Supplementary Figure 5). Interestingly, (−)TDNT was the only formulation that could effectively remove IL-18 from sepsis patient plasma, even compared to the commercial MG250^® (Figure 4; Supplementary Figure 5). It also effectively removes IL-10 (positively charged) compared to (+)TDNT (Figures 4A, B; Supplementary Figure 4). (−)TDNT had moderate efficacy for the removal of TNF-α and IL-6 (Figure 4B; Supplementary Figure 5) and low efficacy for all other cytokines evaluated. Overall, both (−)TDNT and (+)TDNT showed significantly lower cytokine removal capacity than (pan)TDNT from sepsis patient plasma, which was comparable to the commercial MG250^® (Figures 4C, D, 5C–E). The oxalic acid in (pan)TDNT can interact directly with positive amino residues in cytokines, as well as with negative amino acids via bivalent cation bridges, e.g., Ca²⁺ and Mg²⁺. In addition, the long C17 hydrophobic groups in (pan)TDNT contribute to more efficient global protein interactions. Interestingly, (pan)TDNT had poor removal efficacy for IL-1 family cytokines, e.g., IL-1α, IL-1RA, and IL-18, compared to the commercial MG250^® (Figures 4C, D; Supplementary Figure 5). MG250^® completely scavenged IL-1RA but was much less effective at removing IL-18 and inconsistent for IL-1α and IL-1β across different patients. This raises concerns for the clinical application of MG250^®, since IL-1RA is an antagonistic ligand of the interleukin-1 receptor and inhibits multiple IL-1 family cytokines, including IL-1α, IL-1β, and IL-18 (60). Thus, the complete adsorption of IL-1RA by MG250^® may result in increased inflammation due to the remaining unchecked IL-1 cytokines.

(Pan)TDNT removes significantly more total cytokines than both (+)TDNT and (−)TDNT (Figures 5E–G). (+)TDNT, however, displayed significantly higher selectivity for the removal of total positively charged cytokines (Figure 5H), and (−)TDNT displayed a significantly higher selectivity for the removal of negatively charged cytokines (Figure 5I) compared to other TDNT formulations and the commercial MG250^®, a trend that was consistent across individual patient samples (Supplementary Figure 6). Efficacy for (+)TDNT removal of positively charged cytokines (Supplementary Figure 7A), (−)TDNT removal of negatively charged cytokines (Supplementary Figure 7B), and (pan)TDNT and MG250^® removal of total cytokines (Supplementary Figures 7C, D) remained consistent irrespective of total cytokine burden. While the (+)TDNT and (−)TDNT displayed strong potential promise for selective cytokine removal based on charge, their overall efficacy for cytokine clearance was low (Supplementary Figures 5E, F). The smaller C4 hydrophobic groups in (+)TDNT and (−)TDNT allow more selective cytokine removal due to decreased nonspecific hydrophobic interactions; however, this comes at the cost of overall efficacy. Although we have demonstrated that cytokines can be selectively targeted based on their charge disparity, further optimization of the TDNT structure is needed to achieve an optimal balance between charge and hydrophobic interactions, improving overall cytokine clearance while maintaining charge-selective properties.

To evaluate the potential off-target effect of plasma incubation with TDNT, total plasma protein removal was assessed for comparison with cytokine removal. Both (pan)TDNT and (+)TDNT demonstrated significantly reduced removal of total protein from plasma compared to the commercial MG250^® control, whereas (−)TDNT demonstrated high plasma protein removal like MG250^® (Figures 6A, C). Removal of abundant plasma proteins during hemoperfusion is a significant safety concern because it can worsen tissue edema (17–21). The minimal serum protein adsorption by (pan)TDNT and (+)TDNT compared to MG250 indicates better biocompatibility. To further compare the selectivity of resins for cytokine removal relative to total plasma protein removal, the ratio of cytokine removal to total protein removal was evaluated. Compared to the commercial MG250^® control resin, (pan)TDNT exhibited a significantly higher ratio of cytokine to total protein removal from sepsis patient plasma (Figures 6B, D). This indicates that (pan)TDNT is a promising resin formulation, as it has greater selectivity for cytokine removal compared to conventional MG250^® hemoperfusion therapies. Furthermore, the adsorption of model antibiotics on TDNT resins was evaluated. For example, doxycycline adsorption occurred much faster in MG250 resin, with the saturated adsorption capacity approximately 12 times higher than that in (pan)TDNT (48 µg/mg vs. 4 µg/mg) (Supplementary Figure 8).

Due to the heterogeneity among sepsis patients, we examined whether cytokine removal might reduce the mortality rate in specific populations. When comparing cytokine levels in sepsis patient plasma before (black) and after incubation with (pan)TDNT (purple) or MG250^® commercial control (pink), similar efficacy was observed across all SOFA scores (Figure 7A) and all SII scores (Figure 7B) in the cohort. These results highlight the potential clinical benefit of reducing mortality in high-risk patient groups with high cytokine and high SOFA (72% mortality) or high SII (67% mortality) by removing excessive cytokines, thereby shifting these patients into the lower-risk, low-cytokine, and high-SOFA/SII population, which has a mortality rate of 45%. (pan)TDNT displayed comparable cytokine clearance efficacy regardless of total cytokine levels or the charge of cytokines present in the patient sample prior to resin incubation, as well as patient SOFA score and SII (Supplementary Figure 9). Although (pan)TDNT was effective across the range of SOFA and SII scores in the patient population, this finding highlights the need for more robust and reliable methods to define sepsis patient phenotypes for effective implementation of precise immune modulation therapies in the future. Many studies over the past decade have attempted to establish phenotypes and/or immunotypes for sepsis patients (52–55); however, a method for categorization that can be used reliably for clinical decision-making has yet to be developed. As efforts continue to optimize the promising TDNT resin platform for effective and selective removal of cytokines for precise immune modulation in sepsis, it is also critical to define a reliable method to determine which patient types will derive clinical benefit from this therapy.