Impact of Conditioning Chemotherapy on Lymphocyte Kinetics and Outcomes in LBCL Patients Treated with CAR T-cell Therapy

Abstract

Conditioning chemotherapy (CCT) has been shown to be essential for optimal efficacy of chimeric antigen receptor (CAR) T-cell therapy. Here, we determined whether the change in absolute lymphocyte count, referred to as delta lymphocyte index (DLIx), may serve as a surrogate marker for pharmacodynamic effects of CCT and whether it associated with germline genetic variants in patients with large B-cell lymphoma (LBCL).

One-hundred and seventy-one patients were included, of which 86 (50%) received bridging therapy post-leukapheresis. Median DLIx was 0.5 × 10⁹/L (range, 0.01–2.75 X 10⁹/L) and was significantly higher in patients who achieved complete response (p=0.04). On multivariate analysis, low DLIx was associated only with use of bridging therapy (odds ratio 0.4, 95% CI 0.2–0.8, p=0.007). Low DLIx was independently associated with shorter progression-free (p=0.02) and overall survival (p=0.02). DLIx was associated with genetic variations related to drug metabolism and macrophage biology such as ABCB1, MISP and CPVL.

The impact of CCT on lymphocyte count is affected by use of bridging therapy but change in lymphocyte count is independently associated with efficacy. Studies aimed at investigating macrophage biology in this setting may suggest strategies to increase the efficacy of CCT and improve outcomes.

Introduction

Conditioning chemotherapy (CCT) has been shown to be essential for optimal efficacy of adoptive T-cell therapy including chimeric antigen receptor (CAR) T-cell therapy.^{1, 2} The favorable impact of CCT is mediated by multiple biological mechanisms, including but not limited to elimination of sinks for homeostatic cytokines (mainly interleukin [IL] 7 and IL-15) and eradication of regulatory T-cells and myeloid-derived suppressor cells.^3–5 The addition of fludarabine to cyclophosphamide has been associated with improved CAR T-cell expansion and persistence as compared to other regimens, and the combination of these agents represents the preferred CCT regimen in patients with relapsed or refractory large B-cell lymphoma (LBCL).⁵ However, the identification of patients who would benefit from more or less intensive CCT remains challenging. For instance, while patients who receive high-intensity CCT have a greater probability of achieving a favorable cytokine profile, high levels of IL-7 and IL-15 are associated with improved outcome independent of CCT intensity, prompting the question whether pharmacokinetics and pharmacogenomics of fludarabine and cyclophosphamide may play a role.⁶ Similarly, in clinical trials where the use of CCT was omitted based on a low absolute lymphocyte count (ALC) before CAR T-cell infusion, patients experienced worse outcomes, suggesting that pre-conditioning lymphocyte count alone may not be sufficient to predict the need for CCT.⁷ In this study, we analyze the impact of CCT on lymphocyte count and assess how its depth of change relates to clinical outcomes, cytokine levels, and genetic variation in patients with LBCL receiving CAR T-cell therapy.

Methods

Methods

Patient selection and CCT regimen

This is a retrospective cohort analysis of 171 consecutive patients with relapsed or refractory LBCL treated with standard of care axicabtagene ciloleucel (axi-cel) at our institution, The University of Texas MD Anderson Cancer Center (MDACC), between 01/2018 and 04/2020. Standard of care was defined as administration of commercial product outside of a clinical trial. Data cut-off was 04/2021. Patients received CCT with cyclophosphamide (500 mg/m²) and fludarabine (30 mg/m²) administered intravenously on days −5, −4 and −3, followed by axi-cel infusion (2 × 10⁶ cells/kg) on day 0. The study was approved by the Institutional Review Board of MD Anderson Cancer Center and conducted in accordance with our institutional guidelines and the principles of the Declaration of Helsinki.

DLIx definition, toxicity and response assessment

Clinical characteristics and laboratory features before CCT (day −5) and on the day of axi-cel infusion (day 0) were collected prospectively. Delta lymphocyte index (DLIx) was defined as the difference in ALC between the day of initiation of CCT and the day of axi-cel infusion. As no association between normalized DLIx (according to baseline ALC) and outcomes was observed, the latter was not used in the final analysis. DLIx was categorized in quartiles: quartile 1 (median 0.19, range 0.01–0.26), quartile 2 (median 0.41, range 0.27–0.52), quartile 3 (median 0.69, range 0.55–0.79), and quartile 4 (median 1.36, range 0.88–2.75). Their association with survival was analyzed, and quartile 1 selected as low (vs high) DLIx based on the strongest association with survival outcomes.

Cytokine release syndrome (CRS) and immune cell associated neurotoxicity syndrome (ICANS) were prospectively graded for up to 30 days after axi-cel infusion, according to the CARTOX grading system from 01/2018 to 04/2019, and according to ASTCT criteria from 05/2019 onward.^{8, 9} Performance status was defined according to the Eastern Cooperative Oncology Group (ECOG).¹⁰ The international prognostic index (IPI) was calculated as previously described.¹¹ Response status was prospectively determined by the Lugano 2014 classification.¹²

Cytokine measurements and single nucleotide polymorphism analysis

IL-7 and IL-15 were measured by multiplex assays (V-Plex, Meso Scale Diagnostics) in matched plasma samples on day −5 and day 0. Germline DNA for a subset of patients (N=132) was isolated from banked peripheral blood samples using standard procedures. Genotyping was conducted on the Illumina Genome Screening Array (GSAv2.0) with inclusion of 2% replicates for quality control and repeated for samples with call rates less than 95%. Following removal of duplicate samples, imputation was performed to HRC-r1.1 (hg19) using the Michigan Imputation Server.¹³ Variants with minor allele frequencies <1% were excluded from further analysis. A curated panel of candidate variants were selected from the PharmGKB database based on genetic variants associated with response to cyclophosphamide or fludarabine.¹⁴ A total of 144 variants were identified as having at least one significant previous association reported, of which 131 had genotyping data available in the imputed dataset (Supplementary Table 1).

Statistical methods

Association between categorical variables was evaluated using χ2 test or Fisher’s exact test. The difference in a continuous variable between patient groups was evaluated by the Mann-Whitney test. Only factors significant (p-value ≤ 0.05) on univariate analysis were included in multivariate models. Wilcoxon signed-rank test was used for cytokine level comparison. Genetic variants were analyzed under a linear additive genetic model with DLIx as the outcome. Additional covariates were included to control for the confounding effects of age, gender, and bridging therapy. Significance of each variant was assessed by one degree of freedom Wald test with the effect allele for the candidate variant analyses based on those reported in PharmGKB, and we first conducted a candidate SNP analysis with 131 total SNPs selected due to their prior associations with pharmacodynamic processes of interest. Additionally, gene-based analyses were conducted using the sequence kernel association test (SKAT), which specifies the genetic effects to be random and then performs a variance components test.¹⁵ In further exploratory investigation, a genome-wide single variant analysis was also conducted with the standard Bonferroni-corrected genome-wide association threshold of 5×10⁻⁸ used for significance.

Survival outcomes were prospectively collected. Progression-free survival (PFS) was defined as the time from the start of axi-cel infusion to progression of disease, death, or last follow-up (whichever occurred first). Overall survival (OS) was defined as the time from the start of axi-cel infusion to death or last follow-up. PFS and OS were calculated for all patients in the study and for subgroups of patients using Kaplan-Meier estimates and were compared between subgroups using the log-rank test. Multivariable Cox regression analysis was performed to assess the associations between patient characteristics and PFS or OS.

Results

Baseline characteristics and factors associated with low DLIx

One hundred and seventy-one patients were included in the analysis. Baseline characteristics collected before initiation of CCT (day −5) are shown in Table 1. Median ALC at the time of CCT initiation (day −5) was 0.6 X 10⁹/L (range, 0.04–2.8 X 10⁹/L) and below the lower limit of normal (LLN) in 131 (77%) patients. At the time of axi-cel infusion (day 0), median ALC was 0.03 X10⁹/L (range, 0–1.9 X10⁹/L), and median DLIx was 0.5 X 10⁹/L (range, 0.01–2.75 X 10⁹/L)(Figure 1A). Ten (6%) patients experienced a delay between CCT and axi-cel infusion. In these patients, the median time from CCT initiation to axi-cel infusion was 13 days (range, 7–19 days) but no association between delay and DLIx was observed (p=0.79). Eighty-six (50%) patients received bridging therapy: 52 (30%) received chemotherapy, including a platinum-based regimen in 33 patients and a hyper-fractionated cyclophosphamide-based regimen in 19; 13 (8%) received biological therapy, including polatuzumab vedotin with rituximab (no bendamustine) in 9 patients, lenalidomide in 3, ibrutinib in 1 patient; and 21 (12%) received radiation therapy (Table 1). When considering DLIx as a continuous variable, on univariate analysis, the baseline characteristics associated with low DLIx were ALC < LLN (p<0.001), platelet count < LLN (p=0.003) and use of bridging therapy (p=0.02). On multivariate analysis, only the association between low DLIx and use of bridging therapy was maintained (odds ratio [OR] 0.4, 95% confidence interval [CI] 0.2–0.8, p=0.007)(Figure 1B) whereas the association with ALC < LLN was not maintained (Table 1). When considering DLIx as a categorical variable, on univariate analysis, the baseline characteristics associated with low DLIx were IPI score of 3–5 (p=0.003), ALC < LLN (p<0.001), and use of bridging therapy (p=0.02). On multivariate analysis, low DLIx and use of bridging therapy remained independently associated (OR 0.5, 95%CI 0.2–0.9, p=0.04), whereas the association with ALC < LLN was not maintained (Table 2).

Table 1.

Baseline characteristics according to DLIx (as continuous variable)

Total (N=171)	Number (%)	Median DLIx (109/L)	p-uni	OR [95%CI]	p-multi
DLBCL/HGBCL	133 (78)	0.51	0.76	-	-
PMBCL/TFL	38 (22)	0.58
Age > 60 years	78 (46)	0.55	0.72	-	-
<= 60 years	93 (54)	0.51
Male	120 (70)	0.49	0.25	-	-
Female	51 (30)	0.62
ECOG PS 2–4	20 (12)	0.32	0.76	-	-
0–1	151 (88)	0.58
Ann Arbor Stage III-IV	140 (82)	0.51	0.91	-	-
Stage I-II	31 (18)	0.55
Extra-nodal sites > 1	109 (64)	0.47	0.20	-	-
0–1	62 (36)	0.63
IPI score 3–5	96 (56)	0.44	0.41	-	-
0–2	75 (44)	0.60
ANC low	35 (20)	0.47	0.52	-	-
normal	136 (80)	0.56
AMC low	11 (6)	0.50	0.95	-	-
normal	160 (94)	0.52
ALC low	131 (77)	0.82	<0.001	0.03	1
normal	40 (23)	1.4		[0.01–1.1]
Hemoglobin low	163 (95)	0.51	0.68	-	-
normal	8 (5)	0.90
PLT count low	88 (51)	0.40	0.003	0.7	0.27
high	83 (49)	0.69		[0.3–1.3]
CRP high	113 (66)	0.46	0.11	-	-
normal	58 (34)	0.68
Ferritin high	126 (74)	0.47	0.24	-	-
normal	45 (26)	0.61
LDH high	118 (69)	0.45	0.14	-	-
normal	53 (31)	0.61
eGFR low	31 (18)	0.55	0.91	-	-
high	140 (82)	0.52
Previous therapies > 2	135 (79)	0.48	0.29	-	-
2	36 (21)	0.61
Bridging therapy: yes	86 (50)	0.49	0.02	0.4	0.007
no	85 (50)	0.60		[0.2–0.8]
Bridging: chemotherapy	52 (30)	0.41	0.18	-	-
other/no	119 (70)	0.57
Refractory disease: yes	134 (78)	0.52	0.90	-	-
no	37 (22)	0.55
Previous auto-SCT: yes	45 (26)	0.55	0.89	-	-
no	126 (74)	0.51
Previous allo-SCT: yes	3 (2)	0.88	0.98	-	-
no	168 (98)	0.55

DLIx, delta lymphocyte index; OR, odds ratio; CI, confidence interval; uni, univariate; multi, multivariate; DLBCL, diffuse large B-cell lymphoma; HGBCL, high grade B-cell lymphoma; PMBCL, primary mediastinal B-cell lymphoma; TFL, transformed follicular lymphoma; ECOG, European Cooperative Oncology Group; PS, performance status; IPI, international prognostic index; ANC, absolute neutrophil count; AMC, absolute monocyte count; ALC, absolute lymphocyte count; PLT, platelet; CRP, C-reactive protein; LDH, lactate dehydrogenase; eGFR, estimated glomerular filtration rate; auto, autologous; allo, allogeneic; SCT, stem cell transplant

Figure 1. Factors associated with DLIx and efficacy outcomes. A) Absolute lymphocyte count on day −5 and day 0. B-D) DLIx in patients with and without bridging therapy (B) and patients with or without complete response after axi-cel therapy (C). The red horizontal bars represent medians for each group. E-F) Progression-free (E) and overall (F) survival in patients with low DLIx (Quartile 1, Q1) and high DLIx (Q2–4).

The red bar in Figure A and B and C indicates median value with 95% confidence interval

The dotted lines in Figure E and F represent 95% confidence interval

Table 2.

Baseline characteristics according to DLIx (as categorical variable)

Total (N=171)	High DLIx (N=126)	Low DLIx (N=45)	p-uni	OR [95%CI]	p-multi
DLBCL/HGBCL	101 (80)	32 (71)	0.22	-	-
Age > 60 years	59 (47)	19 (42)	0.61	-	-
Male	90 (71)	30 (67)	0.57	-	-
ECOG PS 2–4	11 (9)	9 (20)	0.06	-	-
Ann Arbor Stage III-IV	102 (81)	38 (84)	0.66	-	-
Extra-nodal sites > 1	75 (60)	34 (76)	0.07	-	-
IPI score 3–5	62 (49)	34 (76)	0.003	0.7 [0.2–1.8]	0.42
ANC low	23 (18)	12 (27)	0.28	-	-
AMC low	6 (5)	5 (11)	0.16	-	-
ALC low	86 (68)	45 (100)	<0.001	0.03 [0.01–1.1]	1
Hemoglobin low	119 (94)	44 (98)	0.68	-	-
PLT count low	59 (47)	29 (64)	0.06	-	-
CRP high	80 (63)	33 (73)	0.27	-	-
Ferritin high	91 (72)	35 (78)	0.56	-	-
LDH high	78 (62)	40 (89)	0.10	-	-
eGFR low	22 (17)	9 (20)	0.82	-	-
Previous therapies > 2	98 (78)	37 (82)	0.67	-	-
Bridging therapy: yes	56 (44)	30 (67)	0.02	0.5 [0.2–0.9]	0.04
Bridging: chemotherapy	35 (28)	17 (38)	0.26	-	-
Refractory disease: yes	98 (78)	36 (80)	0.84	-	-
Previous auto-SCT: yes	38 (30)	7 (16)	0.08	-	-
Previous allo-SCT: yes	3 (2)	0 (0)	0.57	-	-

DLIx, delta lymphocyte index; OR, odds ratio; CI, confidence interval; uni, univariate; multi, multivariate; DLBCL, diffuse large B-cell lymphoma; HGBCL, high grade B-cell lymphoma; PMBCL, primary mediastinal B-cell lymphoma; TFL, transformed follicular lymphoma; ECOG, European Cooperative Oncology Group; PS, performance status; IPI, international prognostic index; ANC, absolute neutrophil count; AMC, absolute monocyte count; ALC, absolute lymphocyte count; PLT, platelet; CRP, C-reactive protein; LDH, lactate dehydrogenase; eGFR, estimated glomerular filtration rate; auto, autologous; allo, allogeneic; SCT, stem cell transplant

Association of DLIx with toxicity and efficacy outcomes

Overall, 13 (8%) patients had grade (G) ≥3 CRS, and 61 (36%) had G≥3 ICANS. There was no association between DLIx and incidence of G≥3 CRS (p=0.28) or G≥3 ICANS (p=0.43). Of 164 patients evaluable for response, complete responses (CR) was achieved in 89 (54%) patients. DLIx was significantly higher in patients who achieved CR as compared to those who did not (p=0.04), and CR rate of 43% in patients with low DLIx and 58% in patients with high DLIx (defined as quartile 1 vs. quartile 2–4)(p=0.09)(Figure 1C–D).

After a median follow-up of 26 months (95% CI, 24–28 months), 116 (68%) patients progressed and/or died, and median progression-free survival (PFS) was 6 months (95% CI, 4–8 months). On univariate analysis, baseline characteristics associated with shorter median PFS were ECOG performance status 2–4 (3 vs 7 months, p=0.02), IPI score 3–5 (3 vs 16 months, p<0.001), use of bridging therapy (4 vs 8 months, p=0.02) and low DLIx (3 vs 9 months, p=0.02)(Figure 1E); on multivariate analysis, low DLIx remained independently associated with shorter median PFS (hazard ratio [HR] 0.6, 95%CI 0.4–0.9; p=0.04)(Table 3). Of interest, the association between low DLIx and shorter median PFS was observed also when limiting the analysis to the 85 patients who did not receive bridging therapy (2 vs 26 months, p=0.03).

Table 3.

Univariate and multivariate analysis for progression-free and overall survival

Total (N=171)	Events	Median PFS (months)	p-uni	HR [95%CI]	p-multi
DLBCL/HGBCL	75	6	0.59	-	-
PMBCL/TFL	19	7
Age > 60 years	43	6	0.91	-	-
<= 60 years	51	5
Male	68	5	0.42	-	-
Female	26	8
ECOG PS 2–4	15	3	0.02	0.8 [0.4–1.4]	0.36
0–1	79	7
IPI score 3–5	64	3	<0.001	0.5 [0.3–0.8]	0.002
0–2	30	16
Previous therapies > 2	78	5	0.11	-	-
2	16	14
Bridging therapy: yes	52	4	0.02	0.7 [0.5–1.1]	0.16
no	42	8
Refractory disease: yes	79	5	0.08	-	-
no	15	26
DLIx: low	35	3	0.02	0.6 [0.4–0.9]	0.04
high	81	9
Total (N=171)	Events	Median OS (months)	p-uni	HR [95%CI]	p-multi
DLBCL/HGBCL	45	16	0.63	-	-
PMBCL/TFL	11	18
Age > 60 years	29	16	0.30	-	-
<= 60 years	27	18
Male	39	17	0.77	-	-
Female	17	17
ECOG PS 2–4	10	8	0.03	0.8 [0.4–1.5]	0.43
0–1	46	18
IPI score 3–5	44	14	<0.001	0.3 [0.2–0.6]	0.01
0–2	12	22
Previous therapies > 2	47	17	0.19	-	-
2	9	17
Bridging therapy: yes	30	15	0.20	-	-
no	36	18
Refractory disease: yes	49	16	0.07	-	-
no	7	27
DLIx: low	29	7	0.02	0.5 [0.3–0.9]	0.02
high	65	26

DLIx, delta lymphocyte index; HR, hazard ratio; CI, confidence interval; uni, univariate; multi, multivariate; DLBCL, diffuse large B-cell lymphoma; HGBCL, high grade B-cell lymphoma; PMBCL, primary mediastinal B-cell lymphoma; TFL, transformed follicular lymphoma; ECOG, European Cooperative Oncology Group; PS, performance status; IPI, international prognostic index

At most recent follow-up, 94 (55%) patients had died and median OS was 17 months (95% CI, 10–24 months). On univariate analysis, baseline characteristics associated with shorter median OS were ECOG performance status 2–4 (8 vs 18 months, p=0.03), IPI score 3–5 (14 vs 22 months, p<0.001) and low DLIx (7 vs 26 months, p=0.02)(Figure 1F); on multivariate analysis, low DLIx remained independently associated with shorter median OS (HR 0.5, 95%CI 0.3–0.9; p=0.02)(Table 3). Of interest, the association between low DLIx and shorter median OS was observed also when limiting the analysis to the 85 patients who did not receive bridging therapy (5 months vs not reached, p<0.001).

DLIx and cytokine homeostasis

Matched plasma samples collected on day −5 and day 0 were available for 40 patients, including 8 patients with low DLIx and 32 patients with high DLIx. Mean plasma IL-7 levels were 3.65 pg/mL (range, 0.16–15.91) on day −5, and 15.47 pg/mL (range, 0.38–58.55) on day 0, with a mean increase of 11.8 pg/mL (range, 0.31–54.45). Mean plasma IL-15 levels were 5.07 pg/mL (range, 0.93–34.66) on day −5, and 25.64 pg/mL (range, 1.85–66.61) on day 0, with a mean increase of 20.57 pg/mL (range, 0.46–62.75).When comparing patients with low DLIx to those with high DLIx, there were no significant differences in mean plasma IL-7 and IL-15 levels on day −5 (p=0.28 and p=0.16, respectively), day 0 (p=0.16 and 0.11, respectively)(Figure 2A–B). No significant difference in the delta change in IL-7 and IL-15 levels between these days was observed when comparing the 2 groups (p=0.43 and p=0.17, respectively)(Figures 2C). Similarly, no significant associations were observed for IL-2 (data not shown).

Figure 2.

Change in plasma levels of IL-7 and IL-15 between day −5 and day 0 according to DLIx. A. Plasma IL-7 levels. B. Plasma IL-15 levels. C. Delta change in plasma IL-7 and IL-15 levels

Germline genetic variation and DLIx

At the conservative Bonferroni p-value threshold of 0.05 / 131 = 0.00038, none of the candidate variants previously implicated in pharmacogenetic events related to cyclophosphamide or fludarabine exposure were significantly associated with DLIx (Supplemental Table 1). However, one variant (rs8110536) in MISP was nearing Bonferroni significance with a p-value of 0.00063, and the G allele was associated with a high DLIx (β: 0.26, SE: 0.074)(Table 4). In gene-based analysis, MISP was significantly associated with DLIx (p=0.0059), suggesting that multiple genetic variants across this gene may be important in mediating the DLIx phenotype rather than the effect being driven by a single variant (Figure 3A). Of interest, within the top 12 variants with p<0.05 in the candidate variant analysis, 5 were within ABCB1, and the gene-based p-value for ABCB1 was significant (p=0.014). In the exploratory genome-wide analysis, a region on chromosome 7 contained several variants with individual p-values in the range of 10⁻⁷ (Figure 3B). While its significance was below the threshold required for genome-wide analyses, this region encompassed CPVL, which was significant in a gene-based analysis with a p-value of 3.70×10⁻⁵.

Table 4.

Candidate Pharmacogenomic Variants Associated with DLIx Phenotype

Variant	Gene	Chr.	Effect Allele	*β (SE)	p-value
rs8110536	MISP	19	G	0.26 (0.074)	0.00063
rs1128503	ABCB1	7	G	0.18 (0.057)	0.0025
rs10276036	ABCB1	7	T	0.18 (0.057)	0.0025
rs1801280	NAT2	8	C	−0.17 (0.058)	0.0045
rs2032582	ABCB1	7	A	−0.15 (0.056)	0.0099
rs1045642	ABCB1	7	G	0.13 (0.053)	0.015
rs4148737	ABCB1	7	C	0.14 (0.061)	0.021
rs2070744	NOS3	7	C	0.14 (0.061)	0.028
rs738409	PNPLA3	22	G	−0.14 (0.066)	0.035
rs246221	ABCC1	16	C	0.12 (0.057)	0.036
rs639174	DROSHA	5	T	0.14 (0.066)	0.040
rs17222723	ABCC2	10	A	0.23 (0.12)	0.045

Effect allele based on previous reports in PharmGKB (Supplemental Table 1)

^*linear additive genetic model with change in absolute lymphocyte count as the outcome with age at diagnosis, gender, and bridging therapy included in model

Figure 3. Association between germline genetic variation and DLIx. A) Association between genetic variants of MISP and DLIx. B) Association between genetic variants of CPVL and DLIx.

Dotted line represents the Bonferroni p-value threshold for the candidate gene analysis (A) and genome-wide analysis (B).

Discussion

In this study, we demonstrate that the impact of CCT on lymphocyte count is affected by use of bridging therapy but change in lymphocyte count is indipendently associated with efficacy after CAR T-cell therapy in LBCL. Bridging therapy is used in about one half of individuals in real-world practice. These patients typically have higher-risk disease and exemplify the need to identify optimal bridging strategies.^{16, 17} The pattern of utilization of bridging therapies remains very heterogeneous, and no preferable strategy has been identified to date.^{18, 19} The small number of patients treated with bridging radiation in our study did not allow us to assess its specific impact on DLIx. However, radiation therapy has been shown to be a safe and effective bridging approach for patients with LBCL, not only reducing myelosuppression compared to bridging chemotherapy, but also potentially favorably impacting the tumor immune microenvironment.^20–22 A more detailed understanding of the biological impact of radiation therapy and other chemotherapy-free bridging strategies would aid in the selection of optimal bridging therapy and could significantly impact CAR T-cell efficacy.

In our analysis, despite confirming an increase in plasma IL-7 and IL-15 levels as effect of CCT both among patients with low and high DLIx, the difference was not significantly different according to DLIx. While an increase in the levels of these cytokines has been associated with clinical response after anti-CD19 CAR T-cell therapy in previous studies,^{4, 6} our findings suggest that this observation may not be solely related to the decrease in lymphocyte count following CCT, and other biological mechanisms may be responsible for the prognostic effect of DLIx.

To this regard, we observed an association between DLIx and genetic variants involving three genes. The first gene was ABCB1, which encodes for P-glycoprotein 1 (also known as multidrug resistance protein 1), a well-known membrane efflux pump involved in chemotherapy metabolism.²³ This finding likely explains the variability in DLIx across different individuals but does not provide insight into the biological mechanisms responsible for the association of DLIx with CAR T-cell efficacy, though ABCB1 is also variably expressed in T cells.²⁴ Of note, no significant association was observed in our study with genotypic variants of DCK, coding for deoxycytidine kinase, or of CYP2C19 and CYP2B6, coding for cytochrome P450 family members, all of which are also involved in the metabolism of fludarabine and cyclophosphamide.^25–27

The other two genes whose variants were associated with DLIx included MISP, coding for mitotic spindle positioning protein, and CPVL, coding for carboxypeptidase vitellogenic like protein. Interestingly, the former is an F-actin bundling protein involved in actin-based cytoskeletal reorganizations, which are crucial for macrophage proliferation, phagocytosis and polarization.^{28, 29} The latter is a serine carboxypeptidase with exclusive expression in macrophages.^{30, 31} These findings raise the possibility that macrophages may play a role in mediating the effect of CCT. Indeed, increases in serum monocyte chemoattractant protein-1 (MCP-1) concentration in response to lymphodepletion have been associated with increased probability of achieving a CR and improved PFS in patients with LBCL treated with axi-cel.⁶ Eradication of myeloid-derived suppressor cells (MDSCs) is also thought to be an important function of CCT.^{1, 6} Of note, transcriptomic analysis performed on LBCL samples before axi-cel infusion have demonstrated that macrophage gene signatures are enriched in patients who relapse early.³² Furthermore, high levels of circulating monocytic MDSCs have been associated with lower peak CAR T-cell expansion.^{32, 33} Several agents which impact macrophage function and induce an anti-tumoral phenotype have been shown to be safe and effective for the treatment of patients with LBCL.^34–38 Investigation of these agents either as bridging therapy or in combination with CCT need to be explored to determine whether they could lead to new strategies to improve the efficacy of CAR T-cell therapy.

We acknowledge multiple limitations of this study, including its small sample size, its retrospective and single-center nature (and hence the need of a prospective validation), the change in toxicity grading systems over time, and the lack of robust power for the cytokine analysis. While these features may limit the general applicability of our findings, the DLIx is easily calculated using basic clinical data and could be readily assessed in other settings. Our observation regarding the impact of bridging therapy is also consistent with those of previous studies.¹⁶ Furthermore, even with only 132 patients in our study, both MISP and CPVL were significant in gene-based analyses and nearly reaching the stringent Bonferroni threshold for significance in individual variant analyses, suggesting that these genes are highly likely influence the response to CCT.

In conclusion, the impact of CCT on lymphocyte count is associated with efficacy after CAR T-cell therapy and is affected by use of bridging therapy. Studies aimed at investigating macrophage biology in this setting may suggest strategies to increase the efficacy of CCT and improve outcomes for patients with LBCL.

Key Points.

• DLIx after CCT is lower in patients who receive bridging therapy and is associated with clinical outcomes after CAR T-cell therapy.

• Germline genetic variants associated with macrophage biology affect DLIx and suggest that macrophages may play a role in response to CCT.

Data Availability Statement

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.