A paramount difference between most anti-cancer approaches and adoptive cell therapy (ACT) is the ill-defined impact of the therapeutic index (TI) in the latter due to the erratic pharmacokinetics and pharmacodynamics of cellular products. TI is defined as the ratio of a drug dose that produces toxicity over the one yielding a clinically effective response: TI=TD₅₀/ED₅₀ (where TD₅₀ is the dose that causes a toxic response in 50% of the population and ED₅₀ is the dose therapeutically effective with the same prevalence). The estimation is intuitive when the effects of a given product strictly depend on its engagement with the respective target referred to, hereafter, as ‘on-target effect’. A pathway inhibitor targeting the V600 mutation of BRAF presents with an optimal TI since targeting of the wild-type gene is minimal. By analogy, cytotoxic T lymphocytes (CTLs) directed against mutated proteins privately expressed by cancer cells will cause minimal on-target/off-tumor toxicity. Conversely, when chimeric antigen receptor (CAR) engineered T cells are aimed at non-mutated self-proteins expressed by benign and neoplastic tissues, the TI tapers according to the differential in respective surface antigen density. Similar considerations apply to tumor-infiltrating lymphocytes (TILs) recognizing non-mutated lineage-specific or tumor-specific antigens.

The differential is nullified in B-cell malignancies targeted by anti-CD19 CAR-T cells. In this case, both normal and neoplastic cells express comparable surface densities of the differentiation marker and the therapeutic window is based on acceptance of the collateral depletion of normal B cells compensated by immunoglobulin replacement.1 When the function of the corresponding benign cells is irreplaceable, as for T-cell malignancies, this approach is unworkable. Differential surface antigen density becomes critical for CAR targets overexpressed by cancer cells such as the amplification of the human epidermal growth factor receptor 2 (HER2) gene.2 Here, surface differences between neoplastic and benign tissues demarcate the 3TI

In analogy with small molecules and antibodies, the interaction between CAR-T cells or TILs and their targets can be reduced to a quasi-linear relationship when tested in controlled conditions such as in vitro cytotoxicity assays. Variability in product potency may still modulate the relationship but other factors related to the host or the tumor microenvironment (TME) can be disregarded. We observed that a tight quantitative correlation exists between the cytotoxic activity of naturally-occurring TILs and surface density of their cognate epitope.4 Assuming comparable distribution of adoptively transferred cells among tissues, this linearity could serve as a conceptual parameter to estimate the number of effectors required to kill benign cells expressing low densities of cognate antigen (toxic dose) compared with cancer cells expressing higher densities (effective dose). For ACT, however, things are not that simple and we argue that even in the hypothetical condition where the TI=1 (no difference between TD and ED), other factors sway the balance between toxicity and effectiveness (figure 1). In this commentary, we will focus pre-eminently on key factors that impact ACT’s clinical efficacy and contribute to a favorable balance in the TI calculation by improving the selection, monitoring and follow-up of patients according to their individual immune biological status.

Generic factors can modulate toxicity and effectiveness upstream of the junction that regulates differential T-cell distribution in benign versus malignant tissues (table 1). To fully exploit a favorable TI, the least dose estimated to yield a desired therapeutic objective should be administered to minimize the chances of toxicity. However, conditions orthogonal to on-target activity can affect the estimate including the ‘potency’ of the product. Potency, characterized at the time of administration, is the capacity of a product to affect a given result. For T cells, potency depends partially on the binding affinity of the T-cell receptor (TCR) or the CAR for the respective epitope. Avidity, the intrinsic capacity of the product to recognize its target, results from TCR/CAR affinity, their expression levels and the functional status of the T cells, which determine the threshold for activation. Moreover, the level of differentiation, functional status and responsiveness to modulatory or chemoattractive signals affects CTLs ‘fitness’ to proliferate, expand, persist and function after infusion.5–7 Enrichment of ACT products with self-sustaining T memory stem cells endowed with self-renewal aptitude and the capacity to reconstitute the entire spectrum of CTL subsets8 may increase the TI by decreasing the number of circulating cells required for optimal trafficking and expansion in the TME.

Immunogenicity may affect persistence because humoral or cellular responses may neutralize the therapeutic, particularly in the case of allogeneic products or those containing heterologous, non-fully humanized or heavily engineered proteins.9

On-target deployment may be skewed by dispersion of ACT products caused by trapping in the pulmonary circulation, sequestration in the reticuloendothelial system (RES), extravasation into interstitial spaces, ‘cellular sinks’ for homeostatic cytokines or the immune-suppressive effect of the tumor-bearing status.10–14 Critchley-Thorne et al 15 observed that the response of circulating lymphocytes to interferon (IFN) stimulation is frequently dampened in patients with cancer. The degree of dysfunction is patient-specific, determined by an indirect effect of cancer cell bioproducts16 and it is reversible ex vivo. It is consequential to postulate that CTL function may be quenched soon after reinfusion in such patients. To our knowledge, this has never been addressed by clinical trials.

Another generic modifier affecting the TI is the uncontrolled and unpredictable release of cytokines resulting from non-specific interactions between administered and endogenous cells leading to cytokine release syndrome (CRS) or other off-target toxicities such as neurotoxicity.17 Finally, the genetic background of the patient and other hidden circumstantial factors related to the host’s history play a subtle and unquantifiable role in sculpting CTL function in vivo.18

A dramatic example of discrepancy between administered dose of effector cells and their on-target deployment was described by Pockaj et al 13 decades ago. TILs adoptively transferred in patients with metastatic melanoma were labeled with radioactive ¹¹¹Indium to trace their body distribution. In 12 of 38 patients (32%) receiving an average dose of 1.3×10¹¹ TILs, no localization to metastatic sites was observed, while TILs were trapped in the lungs and in the RES of lymphatic organs. None experienced tumor regression. Among the remaining 26 patients who experienced tumor localization, only 10 (38%) responded. This momentous study underlines the dichotomy between systemic and on-target factors that determine the TI

An extreme example of critical problems in ACT is the death of a patient who received 1×10¹⁰ T cells engineered to express a third-generation anti-HER2 CAR construct that included (besides the CD3ζ chain) both costimulatory molecules 4-1BB and CD28.19 CAR-T cells were administered in combination with high-dose systemic interleukin (IL)-2 and lymphodepletion. Uncontrollable CRS and respiratory failure ensued. Toxicity was attributed to ‘on-target, off-tumor toxicity’ as HER2 is expressed by normal lung epithelium and ‘on-target, on-tumor toxicity’ due to killing of HER2+ cancer cells in lung metastases possibly resulting in ‘tumor lysis syndrome’. However, the administration of IL-2, which induces uncontrolled cytokine release, increases capillary leakage, activates the RES and turns normal tissues into acutely inflamed sites with chemoattractive and T-cell stimulatory properties11 might have derailed the CAR-T cells promoting leakage into normal organs and lowering their threshold for activation in the context of acute inflammation. Subsequent studies utilizing a second-generation CAR construct and no administration of IL-2 demonstrated limited on-target/off-tumor toxicity.20 21

The balance between the chemoattractive and immune-modulatory properties of benign versus neoplastic tissue regulates the deployment of ACT products. Since benign tissues are generally not inflamed and, therefore, non-chemoattractive, ACT products will barely engage with them, trafficking preferentially to inflamed tumors. We propose that this differential is a key TME-specific crossroad determining theTI in solid malignancies (figure 1). For instance, Castellarin et al 14 developed an immune-deficient mouse model expressing human HER2 in normal hepatic tissue. On-target/off-tumor toxicity was compared between two affinity-tuned HER2-specific CAR-T cells. In mice expressing high levels of human HER2, both high-affinity and low-affinity CAR-T cells caused lethal toxicity. In mice expressing low levels of human HER2, no liver toxicity was noted and, surprisingly, low-avidity CAR-T cells displayed stronger anti-tumor activity. Low-affinity CAR-T cells migrated out of the liver and infiltrated the tumor sooner, suggesting lesser compartmentalization in normal, non-inflamed tissues.

Response to immunotherapy is predicted by pre-existing inflammation.22 Transcriptional patterns such as the immunologic constant of rejection (ICR)23 or the tumor inflammation signature24 demonstrated that the activation of IFN-γ signaling, immune-effector mechanisms and the expression of CCR5 and CXCR3 ligand chemokines stand as independent predictors. Thus, based on the presence of T cells, their spatial distribution and functional orientation, human solid tumors are partitioned into three immune landscapes distributed at approximately equal frequency among cancers of different histology: ‘immune-active, infiltrated or hot’, ‘immune-silent, desert or cold’ and ‘immune-excluded’.25 Attraction of T cells is favored by an inflamed, chemoattractive TME, while cold tumors and non-inflamed benign tissues rest ignored. Immune-excluded tumors sit astride as a certain degree of chemoattraction seems sufficient to draw T cells to their periphery.

Immune-active tumors are characterized by genetic instability and high mutational rate.26 We observed that immunogenic cell death (ICD) is a hallmark of this phenotype, correlating tightly with the ICR signature.26 27 Moreover, most processes implicated in immune suppression such as the checkpoint cluster, regulatory T cells, myeloid suppressor cells and metabolic inhibitors dovetail with the immune-active phenotype.27 The congregation of immune-effector and immune-regulatory mechanisms within the same landscape suggests that an evolutionary balance is required for the survival of antigenic tumors in the immune-competent host, where compensatory immune resistance offset immunogenicity.27 Adoptively-transferred T cells may innately infiltrate immune-active tumors lured by chemoattraction but face diverse challenges within the TME, becoming dysfunctional because of overwhelming immune suppression and metabolic derangement.

Dysfunction of T cells is a hallmark of neoplasia and determines the outcome of anti-cancer immunotherapy.28 On antigen stimulation, CTLs transition to an active state, requiring intensive energy production. Following expansion, most T cells undergo apoptotic death while a smaller proportion returns to a quiescent anamnestic state. These evolutionary stages are echoed by metabolic transitions allowing T cells to adapt to regional and functional differences, respond to context-specific demands and reprogram energy production. These physiologic processes are disrupted in the TME that limits the availability of nutrients in competition with the high energy consumption of replicating cancer cells and the stromal response. Thus, the TME is characterized by heightened consumption of glucose and essential amino acids such as glutamine, tryptophan, cysteine, glycine and arginine.29 Metabolic derangement is compounded by hypoxic conditions that induce the production of immune-suppressive byproducts such as lactate, adenosine and kynurenine.30 Moreover, hypoxia-inducible factor α activates transforming growth factor-β and other soluble elements that bear powerful effects on T-cell trafficking, differentiation and function.31

The presence of T cells at the periphery of tumor nests defines the phenomenon of immune exclusion. A recent study characterized by immunohistochemistry the topography of immune infiltration in approximately 1000 tumor specimens demonstrating that immune exclusion occurs in a third of cancers across several histologies.25 Without knowledge about T-cell spatial distribution, immune-excluded cancers are functionally indistinguishable from the immune-active. By applying the ICR signature to head and neck carcinoma samples, we observed that both immune-excluded and immune-infiltrated tumors display an immune-effector Th1-like signature (Pai et al, ASCO 2018, Abstract # 6052). Since the ICR includes Th1 polarization markers activated by antigen exposure, it is likely that T cells come into contact with and recognize cancer cells at the border of the tumor nest but a functional barrier prevents further progression toward the core. Thus, trafficking of T cells to immune-excluded cancers is comparable to the immune-active since in both cases, chemoattraction recruits T cells to the tumor and some immunogenic stimulus promotes their persistence. The biology determining the two phenotypes diverges at this point: in inflamed tumors, no chemorepulsive signals hamper penetration but immune suppression dampens the function of T cells; immune-excluded tumors act like immune privileged tissue through chemorepulsion, while immune suppression plays a limited role. Functional barriers may not be present in baseline conditions but they may be induced when contact occurs between T cells and cancer cells: for instance, on encounter with tumor cells, T cells release IFN-γ which, in turn, induces the activation of dynamic barriers such as the expression of programmed death-ligand 1.32 Moreover, the expression of genes regulating physical barriers is, in some cases, inversely correlated with the ICR signature,33 suggesting an alternative mechanism of immune-exclusion when mechanical barriers prevent T-cell infiltration. Here, no direct contact occurs between cancer cells and T cells and, therefore, no activation of immune-effector gene signatures can be observed.

While inflamed and immune-excluded tumors are spontaneously chemoattractive, desert cancers may rest in peace ignored by immune cells.25 27 Indeed, immune-desert tumors are genetically stable, bear low mutational rates and lack ICD, closely approximating the biology of normal benign tissues and representing the ultimate challenge toward an optimal TI for ACT products.

In this commentary, we speculate that a fundamental determinant of the TI for ACT of solid tumors is the baseline status of inflammation of benign versus neoplastic tissues. This differential directs homing of T cells through chemoattraction while persistence is promoted by co-stimulatory signals in situ. Yet, no mapping of a given patient’s inflammatory status is considered in pre-treatment assessments, while it is done to evaluate chronic inflammation in the context of cardiometabolic and autoimmune disorders.34 35 A similar organ-specific mapping could categorize inflammation in the cancer-bearing host serving as a patient-specific predictor. Moreover, homing of infused T cells could be tracked by functional MRI, exploiting the metabolism of activated T cells.36 This strategy offers the opportunity to evaluate reasons for on-target/off-tumor effects during treatment.

Success of ACT against solid tumors will depend on a deeper and precise understanding of individual patient’s immune biology and the fine balance that governs distribution, deployment, expansion and activation of infused products. Here, we propose a logical process to define patient selection and therapeutic strategy for an optimal TI Several concepts raised here are speculative, derived from other disciplines or from pre-clinical models. The ultimate test in patients will require thoughtful, though not necessarily, over-burdensome study design to validate the usefulness of this logic. Progress in the treatment of solid cancers will stand a better chance only if systematic non-empirical approaches as the one described here are proactively sought.