Distinct patient, tumour and chimeric antigen receptor T‐cell characteristics are associated with initiating versus sustaining responses to idecabtagene vicleucel in relapsed and refractory multiple myeloma

Summary

Idecabtagene vicleucel (ide‐cel), a B‐cell maturation antigen (BCMA)‐directed autologous chimeric antigen receptor (CAR) T‐cell therapy, was studied in relapsed and refractory multiple myeloma after >4 prior lines of therapy in KarMMa (NCT03361748). Translational analyses evaluated patient baseline, tumour and CAR T‐cell features associated with initiating and sustaining response to a single ide‐cel infusion. Baseline tumour burden, measured by soluble BCMA, and homeostatic cytokine levels released after lymphodepleting chemotherapy were among the top correlates for pharmacodynamic response. Duration of response was associated with tumour genomic features and quality of engraftment indicated by more robust CD4+ CAR T‐cell expansion, higher levels of persistent ide‐cel and sustained suppression of BCMA‐expressing cells. A working model is proposed whereby patient and microenvironment features that may modulate ide‐cel activation and expansion influence the probability of response initiation, and that tumour‐intrinsic resistance features and sustained engraftment of ide‐cel influence the durability of responses.

Translational analyses from the KarMMa trial (NCT03361748) highlight that baseline tumour burden, immune environment and chimeric antigen receptor (CAR) T‐cell dynamics influence both the initiation and durability of response to idecabtagene vicleucel (ide‐cel) in relapsed and refractory multiple myeloma. This study proposed a working model in which the probability of response is influenced by patient and microenvironment features that modulate ide‐cel activation and expansion, while the durability of responses is influenced by tumour‐intrinsic resistance features and sustained ide‐cel engraftment.

INTRODUCTION

In the phase 2 KarMMa study (NCT03361748), idecabtagene vicleucel (ide‐cel; bb2121) demonstrated an overall response rate (ORR) of 73% and a complete response rate (CRR) of 33% with a manageable safety profile, in relapsed and refractory multiple myeloma (RRMM; >4 prior lines of therapy [LoT]). Median progression‐free survival (mPFS) was 8.8 months for the study population, 12.1 months for the target dose of 450 × 10⁶ chimeric antigen receptor (CAR)+ T cells, consistent with the phase 1 study with a subset of patients experiencing ongoing responses exceeding 2 years at the data cut. ¹ , ² While responses to anticancer medicines are multifactorial, autologous cell therapies are complex given they are ‘living drugs’. Responses to autologous CAR T cells are influenced by tumour and tumour microenvironment, patient's T‐cell health, CAR T‐cell product characteristics and patient's immunologic state. ³ , ⁴ , ⁵

While CAR T‐cell responses have been described in other haematological malignancies, translational data are limited in multiple myeloma (MM). CAR T‐cell responses require activation, proliferation and induction of cytolysis of target‐expressing cells, which need the presence of antigen and an inflammatory environment enabling cell engraftment. ⁶ , ⁷ Lymphodepleting chemotherapy (LDC) creates an immune environment supportive of engraftment and releases homeostatic cytokines (e.g. interleukin [IL]‐7, IL‐15), which are associated with improved responses. ⁶ Additionally, suboptimal responses to CAR T cells have been associated with a chronic inflammation and T/CAR T‐cell dysfunction. ⁸ Tumour‐intrinsic mechanisms of resistance include proliferative subtypes of tumours or loss of CD19 antigen. ⁹ , ¹⁰ B‐cell maturation antigen (BCMA) loss has been observed in post‐CAR T‐cell RRMM, albeit at low frequencies. ¹ , ¹¹ , ¹² The association of high risk or treatment resistant (HR/R) tumour genomic features, defined in MM (e.g. biallelic inactivation of TP53 or amp1q) in the context of standard‐of‐care regimens, with CAR T responses remains unclear. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²²

Here, we evaluate associations between responses and patient, tumour and CAR T‐cell features. These analyses are the first comprehensive characterization of CAR T responses in a global study of a BCMA‐directed CAR T, demonstrating that correlates of initiating and sustaining responses were only partially overlapping and summarizes candidate biomarkers that may identify patients at high risk of imminent relapse.

METHODS

Study design, patients and endpoints

KarMMa was a single‐arm phase 2 study of ide‐cel in RRMM (n = 128). The trial design, patient demographics, efficacy and safety profiles have been described previously. ¹ , ² The current analyses are from 21 December 2020 data cut (median follow‐up: 24.8 months). Clinical responses were evaluated as per International Myeloma Working Group (IMWG) criteria. ²³ Pharmacodynamic response was defined as a reduction in soluble BCMA (sBCMA) levels by at least 50% at nadir relative to the day of infusion (see Supporting Information, Methods). PFS was evaluated per the U.S. Food and Drug Administration censoring rules.

Soluble factors, immunophenotyping and assessment of CAR T‐cell levels

Baseline and post‐infusion (days 0–28) levels of 25 immune‐related factors were assayed in plasma from peripheral blood using Luminex (see Supporting Information, Methods). Baseline levels of 92 immuno‐oncology soluble factors were also assayed using the O‐link platform (Immuno‐oncology 2 panel) in serum. sBCMA was evaluated at baseline and post‐infusion in serum as previously described. ¹ Soluble free light chains (sFLC) were evaluated in serum at the treating sites. Ide‐cel levels in peripheral blood were assessed as previously described. ⁴ CAR+ cell phenotype in ide‐cel drug product and post‐infusion were evaluated by multiparametric flow cytometry (MFC). In post‐infusion samples, CAR+ T cells were detected using a proprietary anti‐bb2121 monoclonal antibody; T‐cell memory (T_mem) phenotypes were enumerated using CCR7 and CD45RA markers; and CAR+ T‐cell activation was assessed by CD137 expression (see Supporting Information, Methods).

Ribonucleic acid (RNA) and whole genome sequencing

RNA and deoxyribonucleic acid sequencing were performed on CD138+ cells isolated from bone marrow aspirate collected pretreatment (n = 70) and at disease progression (n = 33). Whole genome sequencing was performed on matched tumour/normal (peripheral blood mononuclear cells) pairs at an average of 60×/30× depth respectively (see Supporting Information, Methods).

Statistics

Statistical tests were used to identify relationships of interest and were post hoc. Nominal p‐values, or false discovery rates (FDRs) where indicated, of p < 0.05 (two‐sided) were considered statistically significant. To evaluate correlations between biomarkers and end‐points defined by group (e.g. responders [≥partial response (PR)] vs. non‐responders [NR; <PR]), an appropriate rank‐sum test was applied (Wilcoxon or Kruskal–Wallis tests). Correlations between biomarkers and PFS were evaluated using a Cox proportional hazards model.

Inclusion and ethics

The KarMMa study protocol was approved by local or independent institutional review boards or ethics committees at participating sites. All patients provided written informed consent.

RESULTS

Clinical responses to ide‐cel were characterized by early CAR T‐cell activation, bias to T_em CAR+ cells and rapid clearance of tumour

Cytokine induction, indicating ide‐cel activation, was observed within the first day post‐infusion and to a greater degree in responding patients (Figure 1A). ¹ Rapid pharmacodynamic responses occurred with early (median, 1 month post‐infusion) nadir of sBCMA levels while the time to clinical best overall response (BOR) was variable (time to complete response [CR] range: 1–15.8 months). ¹ , ²⁴ The long half‐life of monoclonal proteins (M proteins) in some patients may explain this difference in timing. ²⁵ , ²⁶ The ide‐cel drug product had a median CD4:CD8 ratio of 4.9:1.0 and comprised cells spanning the range of differentiation phenotypes with a bias towards T central memory (T_cm) cells in the CD4 compartment and T effector memory (T_em) cells in the CD8 compartment (Figure S1A). Both CD4 and CD8 CAR+ T cells expanded post‐infusion and biased towards more CD8 CAR+ cells at peak expansion (Figure 1B). At day 11 post‐infusion (the median T_max of expansion), T_em CAR+ cells were the dominant subpopulation within both responding and non‐responding patients. Responding patients had a higher percentage of T_em cells at day 11 (CD8+: p = 0.001, CD4+: p = 0.002); higher concentrations (cells/μL) of T_cm and T_em cells expressing CD45RA (T_emra) cells were also observed in responding patients primarily due to more robust expansion of ide‐cel (Figure 1C; Figure S1B). Post‐infusion concentrations of CD8 CAR+ T cells were similar in responding patients with longer versus shorter responses (PFS ≥ vs. <9 months), while relatively more robust CD4 CAR+ T‐cell expansion was observed in longer responses (p = 0.003) (Figure 1D; Figure S1C). The proportions of memory populations at peak expansion were not different between more durable versus less durable responders (Figure S1D).

FIGURE 1.

Ide‐cel was characterized by rapid activation post‐infusion and expansion of T_em cells. (A) Concentrations of IFN‐γ and IL‐6 over time are shown as representative proinflammatory cytokine profiles in responding (IMWG BOR of PR or better) and non‐responding (<PR) patients. (B) CAR+ cell concentrations at T_max in peripheral blood for CD4+ and CD8+ cells (log₁₀ y axis) and the ratio of CD4+ to CD8+ CAR T cells (log₂ y axis) are shown in responding (PR or better) and non‐responding patients (<PR). Cell concentrations were enumerated by flow cytometry. (C) The proportion of CAR+ CD4+ and CD8+ cells by differentiation state are shown by responder (PR or better) and non‐responders (<PR). The memory phenotypes were delineated by CCR7 and CD45RA markers and enumerated by flow cytometry relative to the respective parent populations of CD4+ CAR+ or CD8+ CAR+ cells. CAR+ T cells with naive (CCR7+ CD45RA+), CM (CCR7+ CD45RA−), EM (CCR7− CD45RA−) and EMRA (CCR7− CD45RA+) phenotypes were enumerated. A higher proportion of CAR+ CD8+ (p = 0.001) and CD4+ (p = 0.002) T_em cells were observed in responders compared to non‐responders. (D) CD3+ CAR+ cell concentrations are shown over time in responders (PR or better) who remained in response at the month 9 visit (long response) versus those who progressed prior to month 9 (short response). Peak CAR+ CD4+ cell concentrations were higher in long response versus short response patients (p = 0.003) while peak CAR+ CD8+ cell concentrations were similar. For line plots, points represent the geometric mean and error bars represent standard error of the geometric mean. For box and whisker plots, the centre line represents the median, box limits represent the 25th and 75th quartiles, whiskers represent 1.5× the interquartile range or the maximum/minimum value (whichever is closer to the median), data are denoted by points and points extending beyond the upper and lower whiskers are outliers. BOR, best overall response; CAR, chimeric antigen receptor; CM, central memory; EM, effector memory; EMRA, effector memory cells re‐expressing CD45RA; ide‐cel, idecabtagene vicleucel; IFN‐γ, interferon gamma; IL, interleukin; IMWG, International Myeloma Working Group; PR, partial response.

Different baseline tumour‐associated and immune environment factors were associated with initiating than sustaining pharmacodynamic response

Analysis was focused on the subset of clinical non‐responders (NR) who lacked a clear pharmacodynamic response (<50% reduction in sBCMA; n = 18) to isolate features associated with poor CAR T‐cell activation. Higher levels of sBCMA at infusion were the strongest correlate of pharmacodynamic NRs (p = 0.00004; Figure 2A). sBCMA is a composite measure of tumour burden and was correlated with traditional measures of tumour burden in this study, although to a varying degree with each parameter (Figure S2A). ²⁷ Among pharmacodynamic responders, CAR T‐cell expansion correlated with baseline sBCMA (ρ = 0.46, p < 0.000001), while pharmacodynamic NR was associated with low expansion despite elevated sBCMA indicating dysregulation of CAR T‐cell activation (Figure S2B). Higher baseline levels of blood bilirubin (p = 0.004) and reduced sodium levels (p = 0.0004) were also correlates of pharmacodynamic NR, potentially reflecting downstream impacts of elevated tumour burden (Figure 2A). Several immune‐related soluble factors (i.e. soluble programmed death ligand 2 [sPD‐L2], sCD70, sCD137 and soluble Class I‐restricted T cell‐associated molecule [sCRTAM]) were also among the top correlates of pharmacodynamic NR at baseline (Figure S2C). The specific role of the soluble form of these factors remains to be fully elucidated but may suggest active/chronic inflammatory processes at baseline. After controlling for baseline sBCMA in a regression model of pharmacodynamic NR, lower homeostatic cytokine IL‐7 levels at infusion emerged as the strongest correlate (p = 0.005; Figure 2B); pharmacodynamic NR also correlated with lower post‐LDC levels of IL‐15 at infusion (another homeostatic cytokine; p = 0.04) and trends towards less clearance and faster recovery of T, B and NK cells after LDC (Figure S2D). These observations reinforce the importance of LDC‐induced homeostatic signalling and cytoreduction for response initiation, especially in the context of elevated sBCMA. To focus on potential mechanisms of short versus sustained responses, correlates of PFS were evaluated in pharmacodynamically responding patients. Notably, the factors described above for pharmacodynamic NR were not top correlates of PFS. Instead, elevated baseline levels of lactate dehydrogenase (LDH), a biomarker that may capture both tumour burden and high‐risk biology, and blood biomarkers associated with kidney function (i.e. blood creatinine and urea nitrogen; not shown) were associated with shorter PFS (Figure 2C). ²³ , ²⁸

FIGURE 2.

Patient baseline characteristics correlated with overall response and a BOR of CR/sCR. (A) Patient baseline soluble factors were evaluated for correlations with pharmacodynamic non‐response. sBCMA (p = 0.00004), bilirubin (p = 0.004) and sodium (p = 0.0004) concentrations on the day of infusion are shown by pharmacodynamic responding versus non‐responding patients. Pharmacodynamic response was defined as a decrease in sBCMA levels from the day of infusion by ≥50% and was analysed to determine correlates for poor CAR T‐cell activation associated with non‐response. (B) IL‐7 (p = 0.005) was the strongest secondary correlate with pharmacodynamic response in a multivariate model including sBCMA, which was the strongest correlate of pharmacodynamic non‐response. Displayed are the IL‐7 residuals (arbitrary units, AU) after regressing out the sBCMA contribution to pharmacodynamic response in the model. (C) Baseline LDH levels were associated with PFS in a Cox proportional hazard model. The PFS of patients is shown by LDH level tertile for illustrative purposes and indicated patients with baseline LDH in the highest tertile have inferior PFS. For line plots, points represent the geometric mean and error bars represent standard error of the geometric mean. For box and whisker plots, the centre line represents the median, box limits represent the 25th and 75th quartiles, whiskers represent 1.5× the interquartile range or the maximum/minimum value (whichever is closer to the median), data are denoted by points and points extending beyond the upper and lower whiskers are outliers. BOR, best overall response; CAR, chimeric antigen receptor; CR, complete response; IL, interleukin; LDH, lactate dehydrogenase; PFS, progression‐free survival; sBCMA, soluble B‐cell maturation antigen; sCR, stringent complete response.

Copy number loss on 14q or 1p was a tumour‐intrinsic genomic feature that associated with suboptimal response durability

BCMA loss, a clear tumour‐intrinsic resistance/relapse mechanism to ide‐cel, was observed at low frequency at progression and not at baseline. ¹ Point mutations in BCMA (n = 70) were not observed, and single copy number loss of the locus was observed in three patients at baseline that retained BCMA expression. Antigen loss was identified in two progression samples (one previously reported by Da Via et al.) and involved biallelic deletions at the BCMA locus. ¹¹ Previously defined MM molecular HR/R features (e.g. biallelic TP53 inactivation) were evaluated. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ One or more molecular HR/R features were present in 44% (43/97) of patients pretreatment and 48% (31/64) of progression tumour samples (Figure 3A). No individual molecular HR/R features were significantly associated with ORR or PFS, indicating ide‐cel had similar activity across these patient segments (Tables S1 and S2), and qualitatively there was no clear pattern of accumulation of specific HR/R features at progression suggestive of a dominant selective pressure (Figure 3B). Single copy number loss of a localized region at 1p31.2 and a broad region across 14q were identified through a genome wide scan of copy number gains/losses that were associated with shorter PFS. Del1p31.2 was observed in 9% (6/70) of pretreatment samples (Figure 3C). In all, 23% (16/70) of patients had del14q31.3 (i.e. reporter cytoband within large 14q region) (Figure 3D). Loss of the 1p and 14q regions overlapped in one patient. A median cancer clonal fraction (CCF) of 0.9 was observed at baseline for patients with del1p31.2 or del(14q) indicating a dominant clone. Intra‐patient clonality for these lesions numerically increased after treatment in 79% (11/14) of patients with paired pretreatment and progression samples, preliminarily suggesting enrichment under CAR T‐cell pressure. Differential gene expression analysis in patients with or without del1p31.2 or del(14q) identified 172 and 648 differentially expressed genes (DEGs; FDR <0.05), respectively. Molecular function gene ontology or Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway terms associated with extracellular matrix (ECM) interactions, amoebiasis, cell migration, protein digestion and absorption and cytokine–cytokine receptor interaction pathways were associated with downregulated DEGs (n = 579) in patients with del(14q) (Figure S3). No significant pathway enrichment was observed for upregulated DEGs (n = 69 genes) in patients with del(14q) or DEGs from del1p31.2 patients. A nominal number of baseline DEGs, with no clear interpretation, were associated with ORR or PFS for the cohort (n = 70) and potentially reflect the high degree of heterogeneity in this later‐line RRMM population. DEG analysis of paired pretreatment and progression samples was limited by small sample size and the heterogeneity of this cohort.

FIGURE 3.

Single copy number loss on 1p31.2 or across a region on 14q was associated with shorter PFS. Sankey plots illustrating the changes in frequency of any HR/R (A) and specific HR/R (B) features in evaluable pretreatment to progression (post‐treatment) samples. For patients where multiple HR/R features were present, the following classification hierarchy was used to determine to which group the patient belonged: CRBN disruption, 1q amplification, high CCF del(17p), MDMS8 gene signature and high‐risk t(4;14) breakpoint. (C) 9% (6/72) of the evaluated cohort presented with single copy number loss associated with the 1p31.2 cytoband that was associated with shorter PFS. A Kaplan–Meier plot for PFS is shown demonstrating that patients with del(1p) had suboptimal PFS relative to patients without del(1p). (D) 22% (16/72) of patients presented with single copy number loss spanning a large region of 14q, which was associated with shorter PFS. A Kaplan–Meier plot for PFS is shown demonstrating shorter PFS in patients with del(14q) relative to patients without del(14q). CCF, cancer cell fraction; CRBN, cereblon; HR/R, high risk or treatment resistance; MDMS8, molecularly defined multiple myeloma subgroup 8; PFS, progression‐free survival.

Inferred CAR T‐cell function for at least 6 months was associated with a lower risk of progression

The necessity for longer term immune surveillance in patients with RRMM remains an open question. ¹ Ide‐cel persistence has been observed in patients at ≥1 year post‐infusion, but technical challenges prevented direct assessment of CAR T‐cell function at later time points. ¹ Functional assays could not be performed on post‐infusion CAR+ cells, and CAR+ cell concentrations were insufficient to allow extensive phenotyping. sBCMA and FLCs are expressed by both myeloma and normal plasma cells (nPCs), and sensitive assays are available that detect these biomarkers, enabling longitudinal monitoring for the presence or absence of these cells. Thus, return of sBCMA or sFLCs to detectable concentrations after clearance may be a surrogate to infer loss of CAR T‐cell surveillance and cytolytic function towards plasma cells (illustrative patients; Figure S4A). Detection of these biomarkers after initial clearance does not necessarily indicate progression and can reflect repopulation of nPCs after normalization of the bone marrow. This crude indication of surveillance may offer practical utility, while acknowledging limitations to draw conclusions about functional cells per se.

Within patients who cleared sFLC (from either involved [MM] or heterotypic [nPC] light chains), those that maintained clearance (i.e. ide‐cel function) at month 3 or month 6 had superior landmark mPFS relative to those with detectable sFLC (i.e. loss of function). No clear difference in mPFS was observed at the 9‐month landmark (Figure 4A). When PFS was evaluated from the time of loss of ide‐cel function, landmark mPFS of 11 and 7 months from the time of loss of function was observed for patients with at least 6 months versus less than 6 months of ide‐cel function, respectively (Figure 4B). Similar results were observed with sBCMA (not shown), suggesting similar utility for either biomarker to identify patients at risk of progression within approximately 7–11 months. Patients with functional ide‐cel for at least 6 months versus those with shorter function had lower nadir of sBCMA at month 3, and more CD4+ CAR+ cells expressing CD137, an activation marker, at peak expansion (Figure 4C and 4D). Furthermore, patients maintaining inferred CAR T‐cell function for at least 6 months had higher levels of CAR T cells at 3 months and onward and a higher proportion of T_em CAR+ cells at 3 months (Figure 4E; Figure S4B).

FIGURE 4.

Longer suppression of MM and nPCs was associated with higher quality CAR T cell expansion and persistence and durable responses. (A) Kaplan–Meier plot of landmark PFS at 1, 3, 6 and 9 months for patients with detectable versus undetectable sFLCs (involved and/or heterotypic light chains). Detectable sFLC was determined as any concentration above the limit of detection of the assay and levels could be associated with the presence of nPCs (heterotypic light chain return or return of light chains at normal ratio) and/or MM cells (return of involved light chain at elevated levels or abnormal ratio). Landmark mPFS was longer for patients with undetectable sFLC at the 1‐, 3‐ and 6‐month landmarks, while mPFS was not different by sFLC status at the 9‐month landmark. The frequency of clinical response assessments changed from monthly to every 3 months after the month 6 time point and may explain the initial differences observed in the PFS curves for the month 9 landmark. (B) Kaplan–Meier plot of landmark PFS, whereby the landmark time point was the time of loss of inferred CAR T‐cell function. The time point for loss of inferred CAR T‐cell function was defined as the first visit where sFLCs were detectable (i.e. above limit of detection) again after initial clearance and represents the time to progression after loss of inferred function. These landmark PFS curves were stratified by each patient's maximum duration of inferred ide‐cel function binned by a maximum duration of CAR T‐cell function for M1, M3 and ≥M6, and demonstrate similar median time to progression after loss of inferred function regardless of when the function was lost. The following plots evaluate patients grouped by those patients who had sFLC clearance at 3 months (i.e. CAR T‐cell inferred function at 3 months) and maintained clearance (M6 maintained, i.e. ongoing inferred function) versus lost clearance (M6 loss, i.e. loss of inferred function) by 6 months. (C) sBCMA concentrations are plotted over the first 6 months for patients who had evaluable sBCMA at month 3. Patients with ongoing inferred CAR T‐cell function at 6 months had a lower sBCMA nadir at month 3, and lower concentrations from months 3–6 versus patients who lost inferred CAR T‐cell function at 6 months. Points represent the geometric mean and error bars are the standard error of the geometric mean. (D) Patients with longer inferred CAR T‐cell function had higher percentages of their CD4+ CAR+ cells positive for the activation marker CD137 during peak expansion, enumerated by flow cytometry. The y‐axis is the fraction of cells positive for CD137 within the CD4+ CAR+ parent population. Points represent the mean and error bars are the standard error of the mean. (E) Ide‐cel concentrations in peripheral blood were evaluated by qPCR and are plotted over time points in the first 6 months post‐infusion. Patients with ongoing inferred CAR T‐cell function at month 6 had a trend towards higher peak CAR T‐cell concentrations and better persistence of ide‐cel over the first 6 months post‐infusion than patients who lost inferred CAR T‐cell function by 6 months. Points represent the geometric mean and error bars represent the standard error of the geometric mean. CAR, chimeric antigen receptor; FLC, free light chain; ide‐cel, idecabtagene vicleucel; M, month; MM, multiple myeloma; mPFS, median progression‐free survival; nPC, normal plasma cell; qPCR, quantitative polymerase chain reaction; sBCMA, soluble B‐cell maturation antigen; sFLC, soluble free light chain.

DISCUSSION

Autologous CAR T‐cell therapies are living drugs and their effectiveness is associated with both the immunologic health of the patient and the post‐infusion profile of the cellular drug product. ³ , ⁵ Tumour burden and the patient's immune environment at baseline, which may influence the probability for activation of the CAR T cells, were associated with initiating a pharmacodynamic response to ide‐cel. Additionally, sufficient homeostatic cytokine support for ide‐cel at infusion, especially in patients with high tumour burden, was correlated with pharmacodynamic response. This suggests that mitigating baseline tumour burden and inflammation, while optimization of LDC or dependence of CAR T cells on homeostatic factors, may be areas for future study. ⁶ Potential tumour‐intrinsic mechanisms of initial response resistance (i.e. NR; <PR) to ide‐cel were not observed and implied tumour extrinsic factors as key drivers for initiating a response. Features associated with a more durable response (e.g. LDH) overlapped only partly with the above‐mentioned features (e.g. sBCMA), further suggesting that the underlying mechanisms to initiate a response may be distinct from those mechanisms influencing durability of response.

Several immune‐related soluble factors (sPD‐L2, sCD70, sCD137) are associated with resistance to ide‐cel therapy. High sPD‐L1 levels may inhibit CAR T‐cell activity by engaging inhibitory receptors, reducing anti‐tumour responses. ²⁹ Additionally, sCD137, a secreted variant of CD137, negatively regulates T‐cell activation through activation‐induced cell death, potentially explaining the limited CAR T‐cell expansion in NRs. ³⁰ While the exact role of these factors remains unclear, abnormal baseline levels suggest underlying inflammation and poor immunologic health, which hinder ide‐cel activation and survival, leading to poor treatment outcomes.

No strong correlations between existing HR/R molecular features and ORR or PFS were observed, suggesting potentially broad activity of ide‐cel across patient subsets with poor prognosis. However, this heterogeneous later‐line population of patients with RRMM may not reflect HR features at diagnosis, and the activity of ide‐cel or CAR T‐cell responses in these molecular segments in earlier LoT is an active area of study. Two potentially novel features, single copy number loss at 1p31.2 or a large region of 14q, were associated with shorter PFS. Alterations of 1p have been previously described, although at a different cytoband, and further validation of this initial observation is warranted. ³¹ , ³² The 14q region identified was broad and was not directly associated with the canonical translocations. Tumours from patients with 14q were enriched for genes involved in ECM components, cell migration and cytokine–cytokine receptor interactions. The bone marrow microenvironment is an important component influencing immune cell access and myeloma responses to therapy. ³³ , ³⁴ , ³⁵ Indeed, Evers et al. recently described mutations in, or low expression of, ECM components in myeloma cells that were associated with shorter PFS. ³⁶ Both translocations and deletions within 14q are common in MM, with many of these alterations associating with less favourable outcomes. Here, in tumours containing 14q deletions, differential expression of genes involved in the maintenance of tissue structure, cell behaviour and intercellular communication were notable. This may reflect intrinsic tumour promotion through increased capacity for proliferation or stromal independence. Furthermore, as the CCF increased from baseline to progression in those with paired biopsies, it may alternatively reflect selection for resistance to T‐cell‐mediated cytolytic mechanisms or engagement. Further mechanistic understanding of these genomic changes is warranted.

T_em CAR+ cells were the predominant phenotype proximal to T_max regardless of response status or durability, while the drug product contained a spectrum of T memory differentiation states. Trends towards an increased fraction of less differentiated cells (T_naive, T_cm) proximal to T_max in NRs and less durable responders likely indicated suboptimal ide‐cel activation and a profile more reflective of the pre‐infusion drug product. ³⁷ , ³⁸ T_em cells maintain both potent cytolytic activity and establish immune memory formation required to maintain tumour surveillance. ³⁷ Consistent with predominantly early effector cells at peak expansion, evidence of rapid pharmacodynamic activity was associated with responses and a lack of this early pharmacodynamic activity identified patients at high risk of non‐response or early progression. ¹ Extrapolating from immunologic principles of T‐helper cell function, the observations that features of CD4+ CAR+ expansion were associated with longer sFLC suppression and longer PFS suggest that T‐helper CAR+ cells may be important for response durability. ³⁷ , ³⁸ , ³⁹

Beyond the tumour‐intrinsic features associated with PFS, CAR T immune surveillance after initial tumour clearance was implicated, while acknowledging the limitations of inferring CAR T function indirectly using the methods described in this study. Immune surveillance, inferred by ongoing suppression of sBCMA or sFLCs (both involved [MM] and heterotypic [nPC] chains) through 6 months, a time point months after tumour nadir, was associated with longer PFS and features of better CAR T activation in the first months post‐infusion. Taken together, these observations suggest mid‐term immune surveillance is important for response durability and may not be completely independent from the initial activation of ide‐cel, potentially reflecting better overall engraftment. The diminished correlation between PFS and inferred immune surveillance at 9 months was surprising given the prevailing working model that long‐term immune surveillance is required for durable responses and was a feature in two cases of functional cures observed from early CD19‐directed CAR T studies. Conversely, the persistence of cilta‐cel was not correlated with response durability in the CARTITUDE‐1 RRMM study. ³⁹ , ⁴⁰ The BCMA‐ and MM‐directed CAR T space is still in its early days, and further studies are warranted to evaluate the association between long‐term immune surveillance and durable responses.

Longitudinal monitoring of sFLC or sBCMA clearance and time to return of detectable levels, especially levels associated with nPC return, may have practical utility to identify patients at risk of progression. Paiva and colleagues similarly reported that return of normal plasma cells in the context of minimal residual disease (MRD) negativity was associated with inferior PFS with ide‐cel. ⁴¹ In contrast to MRD assessments in bone marrow, however, sBCMA and sFLC are peripherally accessible biomarkers that may enable longitudinal monitoring of ide‐cel pharmacodynamics and, indirectly, engraftment dynamics associated with optimal responses.

This study suggests a working model for ide‐cel whereby (A) initiating responses were primarily dependent on early activation of the infused cells that may be influenced by the magnitude of tumour burden and resulting immune environment factors, and (B) sustaining responses were dependent on tumour‐intrinsic features and sufficient CAR T engraftment providing mid‐term immune surveillance (Figure 5).

FIGURE 5.

Working model of ide‐cel responses in RRMM from the KarMMa study. Illustrative example of qualitatively different response trajectories observed after ide‐cel infusion in the KarMMa study from the viewpoint of a pharmacodynamic biomarker like sBCMA. A small proportion of ide‐cel patients show limited evidence of pharmacodynamic response, and evidence of elevated baseline inflammation suggestive of poor patient or immunologic health, high baseline tumour burden and poor CAR T‐cell expansion were most strongly associated with this pharmacodynamic non‐response profile. Molecular features of the tumour were not associated with initiating a response. A high frequency of KarMMa patients had a pharmacodynamic response in this late line and highly refractory population, and the depth and duration of this response were correlated with the quality of CAR T‐cell engraftment (i.e. magnitude of expansion and mid‐term duration of CAR T‐cell function) and molecular features of the tumour. The analyses presented herein suggest CAR T‐cell function and immune surveillance may not be required for durable responses after approximately 6–9 months; however, these were early observations and additional studies are needed to elucidate further the relationship between long‐term immune surveillance and durability of responses in RRMM and BCMA‐directed CAR T‐cell products. BCMA, B‐cell maturation antigen; CAR, chimeric antigen receptor; ide‐cel, idecabtagene vicleucel; RRMM, relapsed and refractory multiple myeloma; sBCMA, soluble B‐cell maturation antigen.

AUTHOR CONTRIBUTIONS

Designed the research: N Martin, ET, OF, JR, EF, DBD‐M, IM, BP, N Munshi, TBC, SMK. Performed the research: N Martin, ET, JM, SMK. Analysed the data: N Martin, ET, NS, YX, OF, JR, EF, JM, DBD‐M, IM, BP, N Munshi, TBC, SMK. Drafting the manuscript and reviewing it critically: N Martin, ET, NS, YX, OF, JR, EF, JM, DBD‐M, IM, BP, N Munshi, TBC, SMK.

FUNDING INFORMATION

Financial support was provided by the Bristol Myers Squibb.

CONFLICT OF INTEREST STATEMENT

N Martin, ET, NS, TBC and SMK are employees of and hold stock in Bristol Myers Squibb. EF was an employee of and held stock in Bristol Myers Squibb. YX, JM and DBD‐M are employees of Bristol Myers Squibb. OF was an employee of and held stock in bluebird bio and 2seventy bio. JR is an employee of and holds patents with Bristol Myers Squibb and holds stock in Bristol Myers Squibb and Adaptive Biotechnologies. IM has no conflicts to declare. BP has received research funding from Amgen, Bristol Myers Squibb, BeiGene, GlaxoSmithKline, Roche and Sanofi; has received consulting fees from Amgen, Bristol Myers Squibb, GlaxoSmithKline, Janssen, Roche, Sanofi and Takeda; and has received honoraria from Adaptive Biotechnologies, Amgen, Becton Dickinson, Bristol Myers Squibb, GlaxoSmithKline, Janssen, Roche, Sanofi and Takeda. N Munshi has received research funding and consulting fees from Amgen, Bristol Myers Squibb, Janssen, Legend, Novartis, OncoPep, Pfizer and Takeda; and holds stock in OncoPep.

Supporting information

Data S1.

Martin N, Thompson E, Stong N, Xu Y, Finney O, Rytlewski J, et al. Distinct patient, tumour and chimeric antigen receptor T‐cell characteristics are associated with initiating versus sustaining responses to idecabtagene vicleucel in relapsed and refractory multiple myeloma. Br J Haematol. 2025;207(3):1019–1028. 10.1111/bjh.70012

DATA AVAILABILITY STATEMENT

The Bristol Myers Squibb policy on data sharing may be found at https://www.bms.com/researchers‐and‐partners/clinical‐trials‐and‐research/disclosure‐commitment.html.