The ratio of brain to liver glucose activity and prognosis with CAR-T in non-Hodgkin lymphoma

Dear Editor,

CD19 targeting chimeric antigen receptor T-cell (CAR-T) therapy has improved prognosis in patients with B-cell non-Hodgkin lymphoma (NHL), and a fraction are potentially cured [1–3]. However, some patients have a short response duration and those without a rapid and deep response are at higher risk of treatment failure. Positron emission tomography combined with computerized tomography (PET/CT) using ¹⁸F-fluoro-2-deoxy-D-glucose (¹⁸F-FDG) is routinely performed before CAR-T cell therapy to determine the disease burden [4]. The basis of PET/CT imaging is the higher dependence of tumor cells on glycolysis for energy generation, a phenomenon known as the Warburg effect. The Deauville response scoring system compares lesional uptake to reference organs, including mediastinal blood pool and liver [5]. Other volumetric measures of tumor burden include metabolic tumor volume (MTV) and total lesion glycolysis (TLG). The brain-to-liver glucose ratio (B2LR) is defined as the ratio of glucose uptake in the brain compared to the liver based on ¹⁸F-FDG PET/CT [6]. In multiple myeloma, the B2LR is prognostic both for progression-free survival (PFS) and overall survival (OS) after CAR-T therapy [6]. A B2LR ≤ 2.5 correlated with the presence of extramedullary disease (EMD), but remained an independent predictor even in the presence of EMD with respect to PFS and OS [6]. We hypothesized that the B2LR would provide prognostic insights on the outcomes of NHL patients undergoing CAR-T cell therapy.

After IRB approval (Supplementary Material), we evaluated 157 patients with DLBCL (79), TFL (34), FL (12), high-grade lymphoma (19), MCL (9) and a small number of patients with various types of NHL. The demographic, clinical, and laboratory characteristics of the patients are presented in Supplementary Table 1. The median number of lines of therapy before CAR-T was 3 (1–9). Our initial studies in myeloma used QREADS to estimate the B2LR [6]. QREADS is proprietary software and not widely available. In this analysis, we estimated B2LR using both QREADS and SUV from the same five regions in the brain and liver (Fig. 1A, B). The two approaches give similar results. In Supplementary Figs. 1 and 2, we provide an example of a patient with a normal B2LR as measured by the two methods. Twenty-one patients had a B2LR ≤ 2.5, while 136 had a B2LR > 2.5 based on QREADS. The two cohorts of patients were otherwise well-matched, except for a higher ECOG performance status in patients with a B2LR ≤ 2.5 (p = 0.0369). In Supplementary Table 2 we report the analysis of the B2LR using both software packages. The median B2LR (QREADS) was 2.769 (range: 0.93–3.516) while the B2LR based on SUV measurements was 3.09 (range: 1.02–7.07). The B2LR estimated using the SUV negatively correlated with age (ρ = −0.1776, p = 0.0308). Neither the average activity in the brain (ρ = −0.1161, p = 0.1161), nor in the liver (ρ = 0.043, p = 0.6002) correlated with age. The two approaches to measure the B2LR were strongly correlated (Spearman’s rho = 0.5366, p < 0.0001). The two methods to determine B2LR distributed the patients very similarly with respect to common laboratory parameters measured before the start of LD chemotherapy (Supplementary Table 2).

Fig. 1. Estimating the B2LR from PET/CT images.

Determination of the B2LR using functionality in QREADS (A) and based on SUV method (B). Pixel intensity in 5 regions of the brain and 5 regions in the liver is determined and the average pixel intensity in the brain, divided by average pixel intensity in the liver gives the B2LR. Note that the values of the B2LR based on either method are very similar.

We performed ROC analysis using disease progression at 3 or 6-months post CAR-T as the outcome. The optimal B2LR cut-off was 2.51 (AUC 0.55, p = 0.0165). 21 patients (13.4%) had B2LR ≤ 2.5 based on QREADS, and 29 patients (18.5%) had B2LR ≤ 2.5 based on SUV measurements. One patient with a B2LR ≤ 2.5 based on QREADS had B2LR > 2.5 on SUV measurement, while nine patients with B2LR ≤ 2.5 based on SUV measurements had B2LR > 2.5 by QREADS. The correlation between the two methods was quite high (χ² = 75.9, p < 0.0001) and subsequent analyses only used the SUV method to evaluate survival outcomes. Patients with B2LR ≤ 2.5 were at higher risk of Grade 3 or 4 CRS (11.1%, versus 0.74%, p = 0.0281). Moreover, every patient with B2LR ≤ 2.5 required at least one dose of tocilizumab (Supplementary Table 2).

At the 3-month follow-up, 13 of 28 (46.4%) patients, compared to 35 of 124 (28.2%), had progressed based on the B2LR (χ² = 3.4, p = 0.068). The SUV based B2LR ≤ 2.5 was prognostic for PFS (p = 0.0275) and OS (p = 0.0007). The median PFS for patients with B2LR ≤ 2.5 was 0.29 years versus 0.98 years for patients with B2LR > 2.5 (p = 0.055) (Fig. 2A). Similarly, median OS was 1.29 years with a B2LR ≤ 2.5 and 3.3 years for patients with a B2LR > 2.5 (p = 0.03) (Fig. 2B).

Fig. 2. Outcomes of patients after CAR-T as a function of the B2LR as measured by the SUV method.

A B2LR ≤ 2.5 is associated with a shorter progression free survival (A) as well as a shorter overall survival (B) with CAR-T cell therapy for NHL. Risk stratification of NHL using both the B2LR and LDH prior to CAR-T cell therapy. Patients with either B2LR ≤ 2.5 or LDH > 300 U/L (or both) have a short progression free survival (C) or overall survival (D) after CAR-T cell therapy.

We performed ROC analysis to determine an optimal cut-off for the LDH in predicting progression at 3 or 6 months. The optimal cut-offs were LDH = 299 U/L for progression at 3 months (AUC: 0.59245, p = 0.0189) and LDH = 274U/L at 6-months (AUC: 0.63815, p = 0.0005). Using a cut-off of 300 U/L, 14 of 29 (48.3%) patients with B2LR ≤ 2.5 had an LDH > 300 U/L, while 36 of 128 (28.1%) patients with B2LR > 2.5 had an LDH > 300 U/L (χ² = 4.79, p = 0.0286). We divided patients based on the B2LR and LDH using these thresholds (B2LR ≤ 2.5 or >2.5 and LDH < 300 or >300 U/L) into 2 groups with 0 or 1 of these features. Ninety-two patients had a score of 0 and 65 had a score of 1. Patients with a score of 0 had a median PFS of 2.863 years, compared to 0.35 years for patients with a score of 1 (p = 0.0127) (Fig. 2C). The same scores predicted an OS of 3.58 years versus 1.30 years, respectively (p = 0.0097) (Fig. 2D).

Patients who experience an increase in tumor metabolic activity within a 6-week window before starting LD chemotherapy have an increased risk of relapse and death after CAR-T [7]. Dean and colleagues have reported the prognostic value of MTV prior to therapy with Axi-cel [8], while Wang et al. studied the prognostic value of MTV and TLG in a cohort of 19 patients with NHL treated with CAR-T therapy. In the latter study, neither MTV nor TLG had an impact on OS [9]. The total metabolic tumor volume (TMTV) has also been shown to be highly prognostic with respect to response duration post CAR-T [10].

Here, we have further expanded on our observation that SUV based B2LR is a simple and powerful predictor of both short and long-term outcomes after CAR-T therapy [6]. The results support the use of the B2LR as a predictive tool for the identification of early progression (<3 months from CAR-T therapy) and is optimal to identify those patients who are less likely to benefit from CAR-T therapy due to a short duration of response. This index is independent of other established risk factors for progression and poor outcomes after CAR-T. Our results suggest that the single time point B2LR, shortly before initiation of LD chemotherapy, is highly prognostic with respect to the durability of response to CAR-T therapy. A B2LR ≤ 2.5 is associated with a 50% probability of disease progression within 3 months of CAR-T cell therapy. Moreover, patients with NHL who relapse early after CAR-T appear to have a very poor prognosis with a median survival of less than a year. Our results complement those of Vercellino et al. [10] and Breen et al. [7] and further simplify image-based prognostication, since the SUV determined B2LR is based on a single set of PET/CT images performed immediately before the start of LD chemotherapy. Our results show that the B2LR is independent of other proven prognostic factors such as ferritin and LDH with the advantage of simple determination and no specialized software or lengthy training are required. Our study with 157 patients is also the largest reported to date on the prognostic impact of PET/CT before CAR-T [8–13]. While the number of patients with B2LR ≤ 2.5 is small (18.4%), with the increasing use of CAR-T in lymphoma, the implications are substantial. Moreover, the range of responses in our cohort of patients is similar to what has been reported in ZUMA-1, and in real world experience with Axi-cel [1, 2], and also similar to reported outcomes with Tisa-cel and Liso-cel. Therefore, our observations are likely representative of the state of the use of CAR-T cell therapy for NHL.

Although the data were generated in a single center and requires external validation (especially given the AUC of 0.55 for the B2LR), this is somewhat mitigated by the large number of patients evaluated with outcomes including responses, toxicity, and survival that are similar to that reported in real world experience [1, 2, 14]. Our results support those of others with respect to the prognostic value of LDH, and suggest that the LDH and B2LR before the start of LD chemotherapy are complementary and predictive of both PFS and OS after CAR-T. An LDH > 300 U/L and/or a B2LR ≤ 2.5 before CAR-T in patients with NHL are associated with a poor prognosis with shorter PFS and OS. These results also support our prior observations of the B2LR in multiple myeloma [6] as well as observations by others in diffuse large B cell non-Hodgkin lymphoma, follicular lymphoma and post-transplant lymphoproliferative disorders [15]. The B2LR is a radiologic biomarker that is useful in predicting outcomes for a growing list of hematologic malignancies, with potential application to a range of additional cancer types.

Author contributions

Concept, data analysis, manuscript preparation and final approval: SD, PR, AGE, DD. Data abstraction: DD. Provided patient care, critical review of manuscript: SMA, NNB, UD, PBJ, AK, JP, JCVB, YW, LY, JRY, SMB, and DD.

Funding

This work was funded in part from a generous gift of The Centurion Fund through the Department of Development at Mayo Clinic to DD.

Data availability

The data related to this work is available from the corresponding author on reasonable request.

Competing interests

SMA: Research funding from Astra Zeneca, Affimed, ADC Therapeutics, BMS, Pfizer, Step Pharma and Takeda. NNB: Acrotech Biopharma LLC (Advisory Board), Pfizer (Steering Committee). YL: Ad Boards: Janssen, Sanofi, BMS, Regeneron, Genentech, Tessera, Legend, NexT Therapeutics. Steering Committees: Janssen, Kite/Gilead. Research: Janssen, BMS. Scientific Advisory Boards: NexImmune, Caribou. Data Safety Monitoring Board: Pfizer. DD: Consulting with personal payment: Alexion, Apellis, Argenyx, BMS, Janssen, Regeneron, Roche-Genentech, Sanofi, Sorrento and Takeda. Clinical trial support to institution: K36 Therapeutics.

Ethics approval and consent to participate

This study was approved by the Institutional Review Board at Mayo Clinic in Rochester. All patients had provided written informed consent allowing the use of their medical record for research.

Supplementary information

The online version contains supplementary material available at 10.1038/s41408-025-01425-3.