Background: Early fever after CAR-T therapy can reflect both an infection or CRS. Distinguishing both etiologies in these commonly neutropenic patients represents an unresolved clinical challenge. Accordingly, the vast majority of patients receive empiric broad-spectrum antibiotics, which can promote antibiotic resistance and negatively influence the gut microbiome, which has recently been linked to inferior CAR-T outcomes (Smith et al, Nature Medicine 2022). Here, we aimed to characterize infection- vs. CRS-specific differences in routine and exploratory serum proteomics, and to develop a clinical tool that reliably discriminates between both. Methods: Early fever events (day 0-30) were retrospectively characterized as CRS vs. infection in 62 patients treated with standard-of-care CD19 CAR-T for r/r B-NHL (23 Axi-cel, 30 Tisa-cel, 9 Brexu-cel). Severe infections were defined as requiring IV anti-infectives and/or hospitalization in the presence of microbiologic evidence or distinct clinical signs of infection; CRS was graded according to ASTCT criteria. Cumulative incidence curves were calculated as time-to-event-analysis for CRS 1°, ≥2° and severe infections (day of event = index event). Serum inflammatory markers (CRP, interleukin-6 [IL-6], procalcitonin [PCT]) were measured daily from CAR infusion until discharge and on subsequent outpatient visits. Inflammatory markers were tested using binary logistic regression for severe infection vs. CRSonlyindex events. Discriminatory thresholds were determined from receiver operating characteristic (ROC) curves by optimizing the Youden J statistic. Longitudinal plasma proteomics was performed in 52 patients across four time points (day 0, 4, 14, 28) using a 92-protein multiplex proximity extension assay (Olink Proteomics). We applied a linear mixed model (LMM) accounting for both infection vs. CRS status and time effects. Results: The cumulative 30-day incidence of CRS 1°, CRS ≥2° and severe infections was 47%, 43%, and 25%, respectively. Five patients had neither CRS nor an infection, and were subsequently excluded from further analysis. Tocilizumab was applied in the majority of CRS patients (1°: 90%, ≥2°: 100%). We did not find a significant correlation between CRS grade and severe infections (G2 = 0.13, p = n.s.). Median onset of CRS and infection was 3 and 8 days, respectively. Most events (54%) occurred in the setting of coincident severe neutropenia (ANC <500/µl). While event-day CRP was comparable between infection and CRS (median 1.8 vs. 3.4 mg/dL, n.s.), we noted increased serum IL-6 (median 2243 vs. 64 pg/mL, p=0.03) and particularly high PCT levels (median 1.6 vs. 0.3 µg/L, p<0.0001) in the infection cohort. For PCT, an optimal discriminatory threshold of 1.5 µg/L was established (AUCROC 0.78). In previous work, we demonstrated that a CAR-HEMATOTOX (HT) score ≥2 identifies patients with a high infection risk (Rejeski et al, JITC 2022). Indeed, the infection cohort displayed higher baseline HT scores (median 5 vs. 2, p=0.002). By incorporating day-of-fever PCT levels with the HT score, the discriminatory capacity for severe infection events was improved. On ROC analysis, the 'HT10 score' (addition of 10 points to the patient-specific HT score in case event-day PCT ≥1.5 µg/L) exhibited superior discrimination for severe infections (AUC 0.92, p<0.0001, sens. 80%, spec. 91%, Fig. 1A). Additionally, we identified protein-level differences between the infection (n=13) vs. CRSonly (n=39) cohorts from the Olink dataset. On day 0, we noted marked upregulation of the angiogenetic factor TNFRSF12A in the infection cohort (q=4.8E-05). Notably, we observed progressive endothelial dysfunction and immune dysregulation in the patients developing severe infections (day 28, Fig. 1B). This included an increased ANG2/1 ratio and dysregulated CD40/CD40L axis. Consistent with a response to microbial exposure, the innate immune regulators CXC3L1, CCL20, and CXCL13 were upregulated in the infection group. Conclusions: The high discriminatory capacity of the HT10 score for infections highlights the advantage of dynamic risk-assessment and emphasizes the utility of procalcitonin in routine inflammatory panels. If validated prospectively, the score will enable more patient-individual decisions on antibiotic use. Finally, the identified Olink candidate markers may further refine the discriminatory capacity of the HT10 score. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal
Read full abstract