The application of adoptive cellular therapies in any format may have generic consequences with constitutional symptoms from cytokines released or co-administered. For the most part, these are manageable in experienced hands and present no new challenges. What is new is that specificities can be engineered into T cells in analogous fashion to monoclonal antibodies that have been adapted to target selected tumor antigens. These antigens are typically normal cell constituents that are enriched in tumors. From a T cell perspective, CARs allow bypassing of thymic editing that prevents normal T cells from high avidity reactions against self-tumor, but that primarily protects from such reactions against self-tissue ("tolerance").

This bypassing of normal tolerance means that some antigen targets may be unsafe for designer T cells. This was recently shown in a designer T cell trial against G250, a prominent renal cell carcinoma antigen 2 . Antibody against G250 had been applied in humans without toxicity, but when this specificity was tested in designer T cell format, reaction occurred against low level G250 on biliary epithelium. This resulted in an intolerable hepatotoxicity in two of three patients with low infused doses in the range of 10⁹ cells (100-fold below typical Surgery Branch TIL doses 3 ), necessitating dose reductions and, in one case, systemic steroids for T cell suppression. When steroids were removed, the patient had no resurgence of liver attack - but also no tumor response.

This key study illustrated that designer T cells carried the potential for serious toxicity. The safety of comparable Phase I interventions against other antigens (folate binding protein 4 , Tag72 5 , CEA 6 , CD171 7 and GD2 8 ) indicate that toxicity is a function of the target - with no obvious means to predict which. The G250 toxicity also demonstrated that safety of a target with antibody is no assurance of safety with designer T cells 2 . This latter conclusion is not surprising given the indirect means of antibody toxicity 9 in comparison with the direct cytotoxic potency of T cells that also brings far greater sensitivity, killing with just a few antigen molecules per cell, far below immunohistochemical detection thresholds 10 .

This G250 agent was expertly managed via a dose escalation plan in a Phase I setting; the system worked: no one died. Instead, it is the evolution of more complex Strategies that raise the special concerns of this essay.

New Strategies evolved because several so-called 1^st generation IgTCR designer T cells (above) had been tested in the clinic without major tumor regressions. Two contributing problems were identified. Firstly, the infused designer T cells initially distributed widely through the blood and tissues, but then they quickly perished in the host that is already replete with T cells. Secondly, the few T cells that trafficked into tumor could initially exhibit killing, but they ultimately disappeared via a process of activation-induced cell death (AICD) or passed to a resting, inactive state.

These two problems prompted two corresponding hypotheses for improving tumor responses:

(1) Responses could be improved: if sufficient T cells were maintained systemically to sustain T cell percolation into tumor (although T cells survived for only a few days of tumor cell killing).

(2) Responses could be improved: if T cells were to activate and proliferate on antigen contact in tumor (although T cells in tumor were few in starting number).

To address hypothesis #1, Dudley, Rosenberg and colleagues 11 applied "conditioning" to create a "hematologic space" with high dose chemotherapy and/or whole body irradiation prior to T cell infusion in their TIL studies in melanoma. With the burst of IL7 and IL15 that accompanies the lymphopenic state 12 , the infused T cells rode the recovery with a homeostatic expansion, i.e., independent of antigen stimulation. As such, low doses of infused T cells could expand 100-fold in vivo to become a stable, "engrafted" component of the lymphoid compartment, in some instances >50% of the cells that would be the equivalent of 5 × 10¹¹ (0.5 kg!) tumor-specific T cells. This in turn led to dramatically improved tumor response rates with substantial numbers of durable remissions.

To address hypothesis #2, so called 2^nd generation "2-signal" CARs were created to improve their function 13 . To the basic TCRz signaling (Signal 1) of the IgTCR was added a co-stimulation Signal 2 via CD28 and/or other signaling domains, e.g., IgCD28TCRz. Signal 1 suffices for T cell killing, but Signal 1 + 2 engages the T cell proliferative capacity, avoiding AICD, and promotes T cell reactivation on antigen contact after passing to resting state. By this, even a few cells trafficking to tumor could activate and expand in situ to large numbers until tumor elimination, in the same way that virus-specific T cells respond to viral infections. Further, the added costimulation renders designer T cells resistant to regulatory T cell suppression 14 .

The benefits of these modifications for improving therapy were enticing, and to many their combination appeared irresistible. With engraftment of 2-signal designer T cells, there would be huge numbers of effectors, and they would never lose their capacity to respond against the tumor threat - or against normal tissues, thereby motivating this essay.

With two independent approaches, however, it is not just their combination but a 2 × 2 array of four distinct Strategies that confronts the investigator in choosing safely how to treat his first patients with a new designer T cell agent: 1^st generation or 2^nd? Infuse or engraft? The philosophy of patient exposures during new drug testing is aimed at proceeding from low risk to higher risk in a regulated fashion. To order these Strategies for risk, therefore, it is instructive to perform a "What-if?" analysis to consider the consequences if G250 designer T cells 2 had had their initial patient exposures under one of these more advanced Strategies.

"What if" G250 designer T cells were first applied via ...?

Strategy 1. 1^st generation, infused [Actual]

In the least aggressive Strategy, infusion of 1^st generation G250 designer T cells was seen to mediate significant toxicity. Steroids successfully suppressed the T cell reaction without reactivation after steroid withdrawal.

Strategy 2. 1^st generation, engrafted

If the same T cells had been engrafted, their resulting vast numbers would likely induce a more severe and possibly lethal toxicity if left unchecked. However, intervention with steroids would again suppress the auto-immune attack. Once brought to resting state and steroids removed, these Signal 1-only designer T cells would be inert (anergic) on contact with antigen positive tissues, and the patient safe from resurgence of his symptoms. Toxicity under this Strategy should be manageable. (See endnote 1.)

Strategy 3. 2^nd generation, infused

If G250 designer T cells were infused as before but in 2^nd generation format, they also would induce toxicity and then respond to steroids. But with removal of steroids, these now-resting 2-signal designer T cells can reactivate on antigen contact with renewed toxicity. Importantly, at low initial exposures in the dose escalation, these infused designer T cells begin as a tiny fraction of the body's T cell repertoire and undergo rapid systemic decline (e.g., 10⁹ cells infused vs 10¹² total T cells, or 0.1% at peak and lower thereafter). From the analogous clinical setting of donor lymphocyte infusion (DLI), we know that size (of dose) matters, and even with a fully competent allo-immune reaction, small numbers of allo-reactive T cells can be safely managed with a balance of GvH reaction and anti-tumor benefit 15 16 . Thus, toxicity under this Strategy should also be manageable.

Strategy 4. 2^nd generation, engrafted

If 2^nd generation T cells had instead been engrafted, G250-specific T cells would not only be capable of reactivation after steroids, but they would be vast in number. With up to 10% of the reconstituted T cell pool being antigen specific after the lowest injected dose (e.g., 10¹¹ cells expanding from 10⁹ injected) 17 , these cells would be virtually impossible to control, like too high a dose in DLI settings. Maximal immune suppression would be required at all times, with infectious complications and a predictably fatal outcome. Had the initial patient exposure of G250 T cells been by Strategy 4, the consequences could have been dire.

With these options, it can be seen that there are now choices, not just of dose levels as in typical Phase I drug studies, but of Strategies, with distinct consequences to each. With these Strategies available, how does one best advance the therapeutic aims while remaining faithful to principles of patient protection via an incremental exposure to risk? This brings us to the concept of Strategy Escalation. Strategy 1, simple infusion of 1^st generation, is the most conservative; Strategies 2 and 3, engraftment OR 2^nd generation, are intermediate in risk; Strategy 4, engraftment and 2^nd generation, is the most aggressive. To proceed from the untested state for a new target ("0") to its most potent implementation, one could envision a Strategy Escalation path of 0 → 1 → (2 or 3) → 4.

But do I advocate that escalations for all new agents first pass through a Strategy 1 test, infusion 1^st generation (0 → Strategy 1)? No, I do not. If the target was previously tested with a Strategy 1, it does provide more confidence of the safety or hazard for the more aggressive strategies. The G250 test by Strategy 1 showed it was unsafe as a target, from which one may forego all more advanced Strategies, thereby sparing patients from more serious injury. Ultimately, however, drugs must be tested for safety in a setting that reflects their potential utility. Sufficient evidence exists from diverse trials with infusion of 1^st generation designer T cells to infer that none will be therapeutically successful by Strategy 1, and safety in this format becomes of mainly academic interest. If we instead start with a more advanced Strategy, what rationale could be invoked?

Strategy 2 with engraftment of 1^st generation showed considerable benefit in the analogous setting of TILs where simple infusions had not yielded high response rates 12 . The promise of Strategy 3 with 2-signals to sustain an antitumor reaction in situ is an hypothesis based on encouraging preclinical data; clinical trials are just now underway. Both of these have a rationale for realistic benefit to patients where Strategy 1 no longer does. If we bypass Strategy 1 for initial human trials, there is more risk with first patient exposures via engraftment (0 → Strategy 2 test) OR 2^nd generation (0 → Strategy 3 test), but there is also a rationale for controlling toxicities should they occur, as discussed above.

I would argue, however, that proceeding with an untested target (e.g., as was G250) to the most aggressive Strategy 4 (engraftment AND 2^nd generation) is too much risk. A 0 → Strategy 4 test presumes much about the quality of our knowledge of the potential normal tissue targets and their susceptibility, and, of all Strategies, this one alone allows no exit strategy if we guess wrong. (See Appendix 1 for examples.) No one could foresee the hepatotoxicity of G250 designer T cells 2 or the cardiotoxicity of trastuzumab antibody (Herceptin^®) 18 prior to the actual human trials. The graded exposures of their respective Phase I/II studies were essential to revealing toxicities before a Grade V event (death). After a target is shown to be safe by one Strategy, one may proceed with fair confidence to more aggressive Strategies, as shown in Figure 1.

Although safer development drives the Strategy Escalation concept, the discipline of this structure can assist in finding more optimal development paths as well. For example, while a case can be made for safely escalating T cells from a prior Strategy 1 or 2 to Strategy 4, these paths are not necessarily recommended (dotted in Figure 1). Three reasons unrelated to T cell safety may be considered for all paths instead passing through a full Strategy 3 test first:

(1) Lower hazard: The NMA conditioning of Strategy 4 is routinely accompanied by infectious complications that can occasionally be fatal [ 19 12 ; see also Appendix 1: Designer T cell study deaths];

(2) Lower cost: The clinical (non-manufacturing) costs in the real-world hospital setting are in the range of $4-$8,000 for simple infusion (Strategy 3) versus $60-$100,000 for engraftment protocols (Strategy 4), per our own experience 20 21 22 ; and finally and importantly,

(3) Better science: A direct 0 → Strategy 4 test with engraftment obscures any chance to test the core driving hypothesis of current research, e.g., that additional signals, as embodied in the advanced generation designer T cells, can promote a fully competent T cell response with in situ expansion until tumor elimination.

To this latter point, T cells do this quite efficiently in virus infections without conditioning, and when we have proven ourselves capable to bypass immunization and antigen-presenting cells via this technology, I expect we will prevail similarly with designer T cells against tumor. At the moment that we succeed with the right CARs, such engraftment strategies, with their attendant costs and hazard, will predictably be retired. Hence, in my opinion, engraftment should be viewed as an intervening measure, applied only until we get better at immunology, to compensate for our still-imperfect T cell engineering.

Further, when targeting a normal self antigen, a Strategy 3 infusion may allow "tuning" of the activity against tumor versus normal tissue by judicious dose exposures and a gradation of suppressive therapies (as needed) in the manner of DLI 15 , where a Strategy 4 engraftment with its hard-to-control cell numbers may fail. That is, with each new product tested under Strategy 3, an appropriate dose escalation plan affords the best chance to define an optimal biologic dose (OBD) to establish proof-of-concept anti-tumor activity and conditions of safety to normal tissues.

At this point in time, however, the first studies with 2^nd generation designer T cells under Strategy 3 (infused) are just coming on-line, and none has yet completed a full escalation with appropriate cytokine support (e.g., IL2). Thus, it is too early to infer sufficiency or deficiency of any of the existing 2^nd generation reagents to eliminate tumors - without engraftment. But where these more advanced reagents are proven therapeutically inadequate (and safe) under Strategy 3 infusions, then engraftment via Strategy 4 with its higher cost and hazard is a justifiable next step in the Strategy Escalation.

Hence, for untested targets, it is my opinion that Strategy Escalations of 0 → 2 (1^st generation, engrafted) or 0 → 3 (2^nd generation, infused) are safe and acceptable for initial human exposures. For all targets, tested and untested, I believe for reasons of safety, science and cost that 2^nd generation engrafted should instead have a full prior test of 2^nd generation infused, i.e., a Strategy Escalation of (0 or 1 or 2) → 3 → 4. (See endnote 2.) This is represented in Figure 2.

CD28, 4-1BB, OX40 and others. I have defined all of these constructs, single or multiple, as 2^nd generation: they all make T cells more potent (quality), some more than others. The best co-stimulation combinations will make T cells quantitatively more able to mediate toxicity, possibly at lower starting cell exposures, but do not introduce qualitatively novel risks. Unrecognized toxicities against self-tissues should still be adequately covered via infusions (Strategy 3) under a dose-escalation plan with appropriately low starting doses, as in tuning donor lymphocyte infusions (DLI) 15 . Similarly, risks with engraftment (Strategy 4) are not qualitatively different among different 2^nd gen constructs once proven safe in a Strategy 3 test.

This falls into two categories: Growth factors (e.g., IL2, IL7, IL15) and Immune Modulators (e.g., IL12, IFNg). Growth factors constitutively expressed improve cell numbers (quantity) by prolonging T cell survival/expansion. Critically, none has been associated with T cell immortalization. For infusion protocols, the impact on quantity is incremental and manageable (versus the quantum changes for engraftment) and likely does not create new types of risks for 1^st or 2^nd generation when infused. (See endnote 3.) Immune modulators like IL12 make T cells more potent (quality) without affecting cell numbers. The anti-self potency can be managed by the same dose escalation as DLI protocols (above). By this Strategy discussion, it appears that there is no untoward risk by Strategy 1 or 3 infusions. Where these cytokines take on special significance, however, is in engraftment protocols. With 10¹¹ or more cells post-recovery secreting cytokine, high systemic exposures may create a risk that is off-target and potentially life-long. With this qualitatively new risk, such a study might merit designation as a Strategy 5 protocol, to be conducted post Strategy 4, if ineffective. (However, see below, On-Off gene control.)

Antigen-Fc molecules have been shown to stimulate designer T cells, 1^st or 2^nd generation, in the presence of monocytes that crosslink Ag-Fc and supply B7 for CD28 engagement and costimulation 31 . This molecule may in principle be used in vivo to reactivate and expand designer T cells in conjunction with any Strategy (1 and 3, post-infusion; 2 and 4, post-engraftment). The ability to control the dose and duration of Ag-Fc exposure allows assignment of Strategy 1+ or 4+, for example, without major risk increment.

Anti-apoptotic genes can replace growth factors (e.g., IL2) by blocking apoptosis from cytokine withdrawal, e.g., via Bcl-xl over-expression [ 32 ; Emtage & Junghans, unpublished data], impacting therapy along the cell number axis (quantity). This has the advantage of avoiding systemic cytokine exposures, whether exogenous or expressed in the T cells (above). However, the potential for transformation and immortalization with a Bcl family member 32 distinguishes this class from the expressed cytokines. This introduces a qualitatively new risk, meriting designation as a Strategy 5 protocol, to be tested (with appropriate rationale) only after failure of a prior Strategy 3 or 4.

This measure would be unnecessary for most infusion protocols, where the dose escalation and suppressive measures provide adequate protection as discussed in the main text (an exception might be with anti-apoptosis genes). The fail-safe feature of incorporated suicide genes presents a potential escape from any toxicity, however it manifests 33 . In the most relevant clinical model, herpes TK (hTK) has been employed in allo-transplant, where it has successfully combated serious GvHD 34 . In the case of 2^nd generation engraftment, a suicide gene could take a Strategy 4 down to a Strategy 4-. Yet, even here, the investigator will want to consider the rapidity and completeness of the suicide (for hTK, hours to days, depending on T cell cycling) versus the rapidity and intensity of onset of adverse effects. In the Her2 study, with a moderate (10¹⁰) dose of T cells, the patient had respiratory distress by 15 minutes post-infusion, requiring intubation, and was dead in 5 days. (See Appendix 1: Designer T cell study deaths.) A suicide gene could not have prevented the initial event but perhaps the ensuing death. Thus, the option of suicide gene control of non-hyperacute toxicities could take the designer T cells under Strategy 4 engraftment to a risk level approaching simple infusion (e.g., Strategy 3+) by reducing effector cell numbers (cell numbers being the essential difference between 3 and 4). However, it does nothing to improve safety or expense of conditioning, or to correct a muddled hypothesis test with the combined approach. The suicide gene ablation for serious toxicity in engraftment also loses the opportunity to "tune" the therapy in the manner of DLI, available to infusion protocols (e.g., Strategy 3), where a balance of anti-self and anti-tumor activity may be achieved with patient benefit 15 . Lastly, if fully tested under Strategy 3, where suicide genes are generally unneeded, a 2^nd generation designer T cell does not require a suicide gene in a subsequent Strategy 4 because safety of the target was previously established.

In analogy to suicide genes, parallel descriptions could be made for control of genes desirable for expression (e.g., of cytokine) that is time-limited without terminating the T cells, allowing for resumption of activity at a later time if needed. Thus, an engrafted 2^nd gen designer T cell with co-expressed cytokine under a Tet-On promoter 35 , potentially termed Strategy 5 because of the added risk of systemic cytokine, is downgraded to a Strategy 4+ because of the potential to shut off growth factor on Tet withdrawal, thereby avoiding need for a prior Strategy 4 trial for patient safety.