Analysis of pre-treatment tumors reveals gatekeepers of response to CAR T cells

Main text

Chimeric antigen receptor (CAR) engineered T cells have demonstrated impressive success in the treatment of aggressive CD19+ cancers. A recent phase 3 study (ZUMA-7) of axicabtagene ciloleucel (axi-cel), an autologous CD19-targeted T cell product, demonstrated superiority compared to standard of care (SOC) chemotherapy when delivered as second-line treatment for large B cell lymphoma (LBCL).¹ This is an exciting step forward in reducing the morbidity of autologous hematopoietic cell transplantation and simultaneously improving outcomes. Despite this, many patients experience disease relapse after initial response or have no response to axi-cel. Few studies have examined the tumor-intrinsic features that associate with responses, hampering both biological insight and improved patient selection. A previous report evaluated pre-treatment lymphoma biopsies from patients enrolled in ZUMA-1, a phase 2 study of the same product used in third-line treatment of LBCL, broadly identifying that increased expression of immune-associated genes was associated with a higher likelihood of remission.² Building on these observations, Locke et al.³ evaluated pre-treatment lymphoma biopsies from patients enrolled in ZUMA-7 to identify molecular programs associated with outcomes following both axi-cel and SOC treatment, the results of which were published in a recent issue of Nature Medicine.

Using lymphoma biopsies collected either at diagnosis or after first-line therapy, the investigators analyzed gene expression using the nCounter Pan Cancer IO360 panel from NanoString. This panel amplifies transcripts from 770 tumor, immune, and microenvironment-associated genes, classified into 39 signatures, which the investigators re-classified into four clusters. Patient outcomes were measured by likelihood of ongoing response (stratified as “ongoing” or “other,” a combination of relapse and no response), event-free survival, and duration of response. For patients who received axi-cel, B cell and proliferation signatures were most closely associated with efficacy. Interestingly, although not directly compared, the beneficial impact of the B cell signature appeared to be driven by low scores in non-responders, while patients who relapsed had more similar scores to those with ongoing responses, suggesting intrinsically distinct tumor biology for non-response and relapse. No association was seen between outcomes and the B cell signature for SOC; however, the antigen presentation machinery signature was associated with improved outcomes, implicating a role of endogenous immunity in this setting. Evaluation of the re-classified clusters revealed that cluster 1, defined by expression of B cell and proliferation genes, was associated with improved outcomes following axi-cel. Cluster 2, which contained genes associated with immune suppression, was associated with worse outcomes following axi-cel. Similar to expression of the IO360 B cell signature, the benefit of cluster 1 appeared to be driven by low expression in non-responders. None of the four clusters associated with outcomes for patients receiving SOC, suggesting that expression of these tumor microenvironment (TME) genes does not impact outcomes following chemotherapy.

CD19 expression has received significant attention as a predictor of response to CAR therapy. Using both immunohistochemistry (IHC) to detect CD19 protein and quantification of CD19 transcripts, the authors found that higher CD19 expression was associated with improved outcomes for axi-cel but not for SOC. Notably, ∼85% of patients with tumors that were deemed CD19 protein negative experienced responses to axi-cel, suggesting a complex role for IHC in patient stratification. CD19^high (>median protein expression) tumors were, not surprisingly, enriched for B cell and proliferation signatures. CD19^low tumors, in contrast, were enriched for immunosuppressive signatures. It is possible that this enrichment simply results from a relative decline in B cells in these tumors, but this would not account for the specific skewing towards immunosuppressive signatures. Integration of CD19 expression with broader gene expression demonstrated that, for axi-cel-treated patients, the worst outcomes were seen in tumors that were CD19^low and cluster 2 high. Looking specifically patients who had CD19^high tumors but poor outcomes, the investigators found enrichment of glycolytic activity signatures. At first glance these findings are a bit contradictory, as the glycolytic activity signature was part of cluster 1, which was associated with improved outcomes following axi-cel. Untangling the impact of CD19 expression in this specific context may lead to better insights into the role of metabolic alterations as they relate to malignant/immune cell composition within tumors.

Perhaps the most impactful outcomes from this study came from the interrogation of how classically defined high-risk lymphoma characteristics, such as high-grade features (double/triple hit or high-grade histology), large tumor volume, and activated B cell molecular subtype, associated with response to axi-cel. These are all factors that portend worse prognosis and lower likelihood of response to standard therapies. Notably, none of these disease-intrinsic features were associated with outcomes following axi-cel, while all three associated with outcomes after SOC, consistent with historical data. Interestingly, previous analysis of data from ZUMA-1 did demonstrate an association between outcomes and tumor burden for axi-cel. Tumor burdens were overall higher in patients enrolled in ZUMA-1, perhaps because ZUMA-1 was a study of axi-cel in third line as opposed to second line. While this distinction may alter the relative contribution of tumor burden, it is curious that the association vanished when tumor volumes were, on average, only modestly lower.

In addition to evaluation of tumor biopsy specimens, the investigators evaluated axi-cel products. An enrichment of early-lineage T cell states, such as naive, stem, or stem-central memory, has been previously shown to associated with better outcomes following axi-cel.^4,5 To see if this association held true in second-line treatment, and in light of more granular molecular data, the authors evaluated expression of CCR7 and CD45A, markers of early lineage, on axi-cel products. They found that higher frequency of CCR7+CD45RA+ cells only associated with improved outcomes in tumors that were CD19^low. One might speculate that this reflects an outsized impact of early-lineage cells in TMEs that are relatively enriched for immunosuppressive cells. Analysis of the impact of early-lineage cells based on cluster 2 (immunosuppression) score did not reveal statistical differences in outcomes; however, EFS trends appeared to support this conclusion. Moving forward, it will be critical to determine if this association truly results from alterations in microenvironment or if there is a previously undescribed association between outcomes, tumor CD19 expression, and CAR T cell lineage states. In an orthogonal investigation, the authors used a distinct scoring system (IS21²) to quantify changes in T cell infiltration following successive lines of therapy. Consistent with previous reports,⁶ they found that T cells were progressively depleted over time, further rationale to collect T cells and treat patients earlier in disease course.⁷

Collectively, the authors demonstrate that predictors of response to axi-cel are significantly different than predictors of response to standard second-line chemotherapy in LBCL (Figure 1). The classical “good” tumor features did not associate with outcomes following axi-cel, broadly suggesting that prognostic indicators need to be re-defined in the era of cell therapy. By evaluating hundreds of tumor samples, this report emphatically substantiates the notion that responses to cell-based immunotherapy and SOC therapies are dictated by fundamentally different biology. The greatest limitation of this study is the use of a pre-defined panel of 770 transcripts. While more technically complicated, an unbiased approach may have enabled identification of gene programs with greater relevance in predicting outcomes, potentially leading to discovery of novel regulators of axi-cel efficacy. Further, whether these findings hold true for 41BB-based CAR therapies remains unknown. As additional correlative data emerge from large-scale clinical trials and real-world experience, integration of technologies such as single-cell RNA sequencing and spatial transcriptomics will enable increased granularity about intratumoral cellular dynamics, adding to our still-nascent understanding of how CAR therapies work.

Figure 1.

Predictors of favorable outcomes following standard-of-care chemotherapy and axicabtagene ciloleucel (axi-cel) in second-line treatment of LBCL

APM, antigen presentation machinery.