In 2017, reprogramming T lymphocytes to carry chimeric antigen receptor (CAR) targeting CD19 antigen became a novel immunotherapy commercially available for patients with refractory/relapsed B cell malignancies. Despite the unprecedented therapeutic responses achieved by CD19-CAR T-cells, the number of patients experiencing relapse stresses the need for reliable biomarkers to closely monitor clinical response and implement early consolidation strategies. Medical imaging such as [18F]-FDG PET/CT is already used for diagnosis and evaluation of hematologic malignancies and the clinical significance of several PET parameters such as Total Metabolic Tumor Volume, standardized uptake value and Total Lesion Glycolysis consumption has been extensively demonstrated. Hence, this is an area of ongoing investigation in the context of CAR T-cell therapy. This review aims to summarize recent clinical data and to emphasize the importance of further investigation of medical imaging biomarkers for CAR T-cells to optimize and personalize medical care: risk stratification, prediction of response, response assessment, and early detection of relapse.

CAR T-cell manufacturing begins with T-cell collection from patients or donors by aphaeresis. These cells are then genetically reprogrammed (e.g., using viral vectors) to express receptors for specific tumor antigens. CD19 CAR T-cell therapy uses a single chain variable fragment (scFv) derived from the variable heavy and variable light chains of an antibody against epitopes of the CD19 antigen. In second generation CARs, the scFv is connected through a transmembrane domain to a costimulatory domain (such as CD28 or 4-1BB) further linked to the CD3ζ intracellular signaling domain of the T-cell receptor (1, 2). Before CAR T-cell infusion, patients are often given a bridging therapy to control their disease during the CAR T-cell production which may take two to six weeks. Patients then receive a lymphodepleting treatment several days before infusion to create a favorable environment for CAR T-cells by removing unmodified T-cells and immunosuppressive regulatory T-cell and increase homeostatic cytokine levels (3). Finally, cells are expanded and infused back into patients to achieve tumor cell recognition and killing. The antitumor response driven by CAR T-cells is HLA independent and relies on antigen-receptor binding and on the co-stimulatory signals that enhance T-cell proliferation and/or persistence. After antigen recognition, CAR T-cells eliminate cancer cells through death receptors-, cytokines- or granzyme/perforin-induced killing.

As CAR-T-cells are still entering routine clinical practice, most of the current knowledge about imaging and outcomes has been gained from review of clinical trials. The most common cancer subtypes studied are acute lymphoblastic leukemia and non-Hodgkin lymphoma and the majority of studies used an autologous cell source (4). While the treatment of relapsed/refractory leukemia and lymphoma is increasing in clinical practice, and the first CAR T product for multiple myeloma was just recently FDA-approved (5, 6), the applications of this therapy in solid cancers remain at a nascent stage and need further investigations (7). Of note, medical imaging guides the management of lymphoma patients while it has a more limited impact for leukemia. Therefore, the focus of this review is on the contribution of medical imaging in lymphoma patients treated with CAR T-cells.

CD19 CAR T-cell therapy has a high overall response rate with complete responses in up to 90% in adult and pediatric acute lymphoblastic leukemia motivating the initiation of hundreds of CAR T-cell clinical trials worldwide and the search for more efficient designs and new antigens (8, 9).

In clinical trials of aggressive lymphoma patients, complete response rates ranged from 40% to 59% (10). In indolent lymphoma, complete response rates were even higher (11, 12). Several factors, however, can limit long-term efficacy of CAR T-cells and lead to disease relapse (13). First, the delay to start CAR T-cell therapy may allow disease progression, with tumor volume increase, which stresses the importance of rapid and reliable manufacturing and bridging therapy. CAR T-cell function may also be decreased by the poor quality and low number of cells obtained from certain patients. In addition, treatment efficacy depends on the expression of the targeted antigen by the tumor cells. Therefore, tumor heterogeneity, mutations, down-regulation or loss of tumor antigen can decrease the recognition of tumors by CAR T-cells and subsequent therapeutic response. Moreover, CAR T-cells’ viability and efficacy can be impaired by suboptimal stimulation leading to T cells exhaustion and relapse (14).

In addition to poor effector to target ratio in presence of a high tumor burden (15), antigen positive relapse can also occur as a result of tumor cells resistance to CAR T-cells (14) or immunosuppressive tumor microenvironment (16) inducing CAR T-cell dysfunction.

Biomarkers that predict short survival are critical for close monitoring during bridging therapy given that the manufacturing time may allow disease progression. Biomarkers that predict durable response at baseline may help identify patients more likely to benefit from this strategy. Biomarkers that indicate treatment failure at the 1-month milestone will identify patients who might benefit from early intervention such as therapies that reinvigorate CAR T-cells or a second infusion of CAR T-cells.

CAR T-cell is associated with a wide range of toxicities. Across studies, all patients experience at least one adverse event. On the most serious side of the spectrum are high grades Cytokine Release Syndrome (CRS) and Immune effector cell-associated Neurotoxicities Syndrome (ICANS). CRS is a systemic inflammatory response characterized among others by fever, hypotension, hypoxia and potential multiple organ failure whereas ICANS is characterized by various neurologic symptoms ranging from confusion and tremor to aphasia, dysgraphia, seizures or coma. These generally occur in the first weeks after CAR T-cell infusion as a results of high levels of cytokines not only produced by CAR T-cells but also by their activation of myeloid cells (3, 35). Pyrexia, fatigue, cytopenias and infections are also frequent after CAR T-cell therapy.

In the ZUMA-1 study, axi-cel infusion was associated with 11% incidence of grade 3 or higher CRS and 32% of ICANS grade 3 or more, with similar results in the real-world data described by Nastoupil et al. (30, 32, 36). In the JULIET study, CRS occurred in 58% patients, with grade 3 or higher CRS in 22%, while neurologic events were observed in 21% of patients, and grade 3 or higher in 12% (37). As mentioned above, several analyses showed that higher baseline tumor burden was associated with severe CRS (24, 25).

For neurotoxicity, along with abnormalities on electroencephalogram and transcranial Doppler ultrasound, dedicated brain FDG PET can contribute to diagnosis (the main observed abnormalities being cortical hypometabolism) and follow-up (38). Structural abnormalities are usually absent, but CT and MRI may identify concomitant events such as ischemic stroke or subarachnoid hemorrhage (38). However, there is a lack of prospectively collected data on the subject. Additionally, a recent study suggested that a higher FDG avidity of lymphoma, evaluated by SUVmax, was associated with more neurotoxicity (34). If confirmed in larger studies, this could also be of importance for patient management after CAR T-cell infusion. The mainstay of treatment of these toxicities is steroids, vasopressors and tocilizumab treatments.

There are several challenges to improve implementation of CAR-T-cells in routine clinical practice and the outcome of patients receiving this innovative treatment. These challenges include technical challenges (CAR-T-cell development, manufacturing), standardization of clinical trial results to facilitate the comparison (protocols, pre-conditioning of patients, CAR-T-cell formulation, quality and persistence), and identifying robust tools to optimize treatment decision (4, 7). Among these tools, imaging techniques may play a critical role. The role and use of medical imaging techniques remain to be defined but results presented above suggest that imaging will be a pivotal tool to guide treatment decisions.

Beyond Total Metabolic Tumor Volume, recent data suggest that lesion dissemination assessed on PET/CT by means of the largest distance between two lesions (normalized with the body surface area), contributes to assess the spread of the disease, and has a prognostic value, independently of Total Metabolic Tumor Volume in a cohort of first-line chemotherapy diffuse large B-cell lymphoma (39). Radiomics could further contribute to extract clinically meaningful data from medical images. Recent findings from a wide range of solid tumor types suggest that a signature combining a limited subset of pretreatment (40) or on-treatment (41, 42) imaging biomarkers are able to help to identify patients who might benefit from early intervention. Among these quantitative imaging biomarkers, several have been shown to predict responses to immunotherapies with immune checkpoint blockers such as increased tumor volume, increased tumor glucose metabolism, tumor organotropism in visceral tissues, and lower skeletal muscle index. These are all associated with unfavorable outcomes (43–47).

Hence, it is very likely that similar technologies will be applicable to CAR T-cells and that computational models will be applied to data from CT or PET/CT scans to predict outcome, while accounting for technical variability between machines and centers. These tools eliminate the bias of investigator assessment and multicenter variability, allowing their implementation in large multisite trials. Artificial intelligence could be used to combine previously cited biomarkers to build robust prognostic/predictive models. One challenge is that building robust models using artificial intelligence requires creating large datasets, hence the need to aggregate data from multiple institutions to avoid overfitting (48, 49). Eventually, deep learning could contribute to determine radiomics signature correlated with survival.

The role of medical imaging, and PET/CT in particular, in lymphoma patients treated with CAR T-cells is twofold. First, the pre-infusion Total Metabolic Tumor Volume seems promising for its prognostic value, and should probably be associated with biological parameters, such as CRP at time of lymphodepletion (26, 50). Some data also suggest that high tumor volume could be correlated with more severe cytokine release syndrome, with possible direct impact on patient management and monitoring. Second, the evaluation of response with CAR T-cell is an ongoing challenge, with more data needed, especially on the possibility of pseudoprogression, slow or late responses as well as the timing of relapses. Beyond FDG, a better knowledge and understanding of imaging data could contribute to detect and treat toxicities timely (51) and further tailor the therapeutic strategy, with the use of next-generation CAR T-cells, combination therapeutics especially in patients with high tumor burden and potentially rapid implementation of salvage therapies in case of relapse such as new CAR T-cell infusion (targeting the same or other antigens), immunomodulatory agents or radiation therapy (52).

LV, DJ, and LD designed the mini review, drafted the work, and approved the content. RB, SK, RR, and LS revised it critically for important intellectual content and approved the content. All authors contributed to the article and approved the submitted version.

LS is a paid advisory board member for Roche and Novartis, and reports receiving commercial research grants from Merck and Boehringer Ingelheim. RR has consulting or advisory role: Atara Biotherapeutics, Novartis, Magenta Therapeutics, Bristol-Myers Squibb, Gilead Sciences, and received research Funding from Atara Biotherapeutics, Incyte, Pharmacyclics, Shire, Immatics, Takeda, Gilead Sciences, Precision Biosciences, Astellas Pharma. RB received honoraria from Novartis and Gilead.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.