Introduction
CAR T cell technology has risen in prominence as a result of the durable, objective clinical responses reported in early phase trials testing CD19 CAR T cells against B cell leukaemia [
1‐
6]. CAR’s typically consist of scFv tumour targeting domains fused to T cell signalling receptors that, when expressed in a T cell, can effectively re-direct immune effector activity towards the cell surface target antigen specified by the scFv domain and independent of HLA restriction [
7‐
10]. Initial testing of CAR T cell therapy against solid tumours has proven to be less efficacious [
11‐
14]. These early clinical trials employed first-generation CAR technology and no patient pre-conditioning. Against this background, we conceived a trial design that questioned the role of CAR T cell dose and the relative intensity of patient pre-conditioning upon the function and clinical impact of adoptively transferred CAR T cells.
The tumour-associated antigen explored in this trial was carcinoembryonic antigen (CEA; CEACAM5; CD66e) which is expressed at high levels in a broad range of tumours including those of the gastrointestinal tract and has been extensively explored as a cancer vaccine target [
15]. A phage-selected CEACAM5-specific scFv (MFE23), shown to be well tolerated in imaging and antibody-directed pro-drug therapy strategies [
16‐
18], was fused to CD3ζ to generate a first-generation CAR termed MFEζ that was extensively characterized for structure and function in T cell lines and primary human T cells [
19‐
23]. The in vivo anti-tumour activity of anti-CEACAM5 CAR T cells [
24] further supported a strategy of targeting CEACAM5 within the context of a clinical trial.
The trial proposal involved dose-escalation of CAR T cells within 3 cohorts to reach a maximum dose of 5 × 10
10 total T cells with fludarabine pre-conditioning and systemic IL2 support. Subsequent cohorts were to receive the maximum dose of MFEζ CAR T cells combined with an increased intensity of pre-conditioning delivered by the combination of cyclophosphamide and fludarabine [
25]. The trial opened in 2007 after significant delays during the regulatory process due to issues reported in other immune-based trials [
26,
27]. Fourteen patients were recruited prior to early termination due to transient acute toxicity after completion of cohort 4. We now report the clinical and scientific observations that confirm that intensity of pre-conditioning impacts upon the relative frequency but not absolute number of systemic MFEζ CAR T cells. Furthermore, systemic cytokine data imply immune activation of first-generation MFEζ CAR T cells in vivo, whilst evidence of raised IL-6, paralleling that seen in CD19 CAR T cell trials targeting B cell leukaemia [
5], putatively implies a common mechanism of in vivo CAR T activity that is dependent upon patient pre-conditioning and, potentially, all generations of CAR design.
Materials and methods
Trial design
This was a single-centre open-label, dose-escalation Phase I study (ClinicalTrials.gov identifier NCT01212887) managed and conducted in accordance with the principles of Good Clinical Practice and UK legislative requirements (Medicines and Healthcare Regulatory Agency).
Primary objectives were to evaluate the feasibility of MFEζ CAR T cell therapy in patients with CEACAM5+ tumours, to assess toxicity and to determine dose of MFEζ CAR T cells for optimal survival in the circulation. Secondary objectives were to assess functionality of MFEζ CAR T cells isolated from the circulation, to obtain preliminary evidence of radiological response and to evaluate safety.
The trial was based on a 3 + 3 design for cohorts 1 to 3 and 4 + 3 for cohorts 4 and 5 (Table
2). If a dose-limiting toxicity was experienced, the cohort was to be expanded to six patients. All patients received pre-conditioning chemotherapy, MFEζ T cells then intravenous IL2 therapy. Patients in cohorts 1–3 received fludarabine chemotherapy (25 mg/m
2/day for 5 days) with inter-cohort escalation of MFEζ T cells. Patients in cohort 4 received maximum MFEζ T cell dose with cyclophosphamide (60 mg/kg/day for 2 days) prior to fludarabine (25 mg/m
2/day for 5 days) chemotherapy. All patients received IV IL2 (600,000 IU/Kg 15-min infusion every 8 h maximum 12 doses). IL2 was commenced 90 min after MFEζ T cells. Criteria for IL2 dose delay, reduction or discontinuation defined within the protocol resulted in administration of a variable number of IL2 doses.
Inclusion criteria for this study included patients with advanced, histologically confirmed CEACAM5+ malignancy where standard curative or palliative measures were not applicable, ≥18 years old, life expectancy over 3 months, performance status of 0 or 1, adequate renal, cardiac, haematological and biochemical function. Exclusion criteria included anti-cancer systemic treatment or radiotherapy within four weeks, on-going significant toxicity from previous therapies, brain metastases, significant non-malignant disease (including autoimmune disease), prior BMT, previous extensive radiotherapy, current other malignancies and patients taking, or likely to require systemic steroids or other immunosuppressants.
Adverse event (AE) monitoring commenced from the point of written consent. AEs were reported as per Common Terminology Criteria for Adverse Events (CTCAE) Version 3.0. The following dose-limiting toxicities were defined when they were almost certainly or probably drug related; toxicity ≥grade 3 as a result of MFEζ T cells; toxicity caused by MFEζ T cells or chemotherapy preventing commencement of IL2 within 24 h; toxicity ≥3 during IL2 therapy that did not resolve to ≤grade 2 within 48 h of stopping IL2; toxicity ≥grade 3 as a result of chemotherapy despite optimal supportive medication excluding bone marrow suppression.
Patients were treated as inpatients and discharged home when clinically appropriate. They were followed up as outpatients and underwent computerized tomography (CT) scans at 6 weeks, 3, 6 and 12 months which were reported to RECIST version 1.0.
Production of MFEζ CAR T cells
MFEζ CAR T cells were produced in compliance with Good Manufacturing Practice as previously described [
28].
Blood collection, processing and cell counts
Blood samples were collected at pre-treatment, day 0 pre-infusion, 2, 6 h, days 1, 2, 3, 4 and 5 post infusion and weeks 1, 2, 3, 4, 5, 6, 12 and then 12 weekly until off trial. Within 24 h of blood draw, plasma and PBMCs were isolated from an EDTA blood at each time point following standard procedures and stored at −80 °C and in liquid nitrogen. An additional CPT™ Vacutainer tube [Becton–Dickinson (BD), NJ, USA] was collected at each time point for mononuclear cells isolation and gDNA extraction using a Wizard® Genomic DNA Purification Kit (Promega, WI, USA) following the manufacture’s protocol. Blood counts were collected daily during hospitalization and at each visit using a certified clinical haematology service. All sample processing and subsequent assays were performed in compliance with good clinical laboratory practice guidelines and subjected to independent quality assurance control.
Laboratory assays
Real-time PCR quantification of transduced cells
A validated quantitative PCR assay (qPCR) was developed to quantify the level of MFEζ CAR T cells in patient samples. A CAR-specific qPCR amplicon (MFEζ F primer 5′-CTTATTACTGCCAGCAAAGGAGTAGTT, R primer 5′-CAAAGCTCGCTCCGTCTGTAG, probe FAM-5′-CCCACTCACGTTCGGTGCTGGC) and genomic standard qPCR amplicon (b2 M, F primer 5′-GGAATTGATTTGGGAGAGCATC, R primer 5′-CAGGTCCTGGCTCTACAATTTACTAA, probe FAM-5′-AGTGTGACTGGGCAGATCATCCACCTTC) were used to determine total genome copies (b2 M) and transduced genome copies (MFEζ) per sample.
The assay was validated using a standard curve generated from a single-cell-cloned Jurkat-MFEζ cell line (100%) diluted to 10, 1, and 0.1% with non-transduced Jurkat gDNA. Each assay included a positive control of known transduction level (4%) and a non-transduced (0%) negative control. The acceptance criteria for the qPCR assay were set as Standard Curve R
2 value ≥0.95, positive control = 4% (±2%) and lower limit of detection = 0.1% transduced cells.
IFNγ ELISA analysis
A 96-well ELISA plate was coated for 2 h at 37 °C or overnight at 4 °C with 1 µg/ml IFNγ capture antibody (MAB-285, R&D systems, MN, USA) and then washed with PBS + 0.05% Tween. IFNγ standards were then added (200–0.5 pg/ml) along with 10 and 100 µl patient plasma for each sample. Following incubation at 37 °C for 1 h, 100 µl of biotinylated IFNγ detection antibody (BAF-285, R&D systems, MN, USA) was added to each well for a further hour at 37 °C. After three washes, Streptavidin peroxidase (POD) conjugate (Roche, Basel, Switzerland) was then added and the plate incubated at 37 °C for 30 min and then POD blue substrate (Roche, Basel, Switzerland) added for 30 min. The reaction was stopped by the addition of H2SO4 and the plate read at 450 nm. The concentration of IFNγ was calculated using the standard curve.
Determination of cytokine concentrations in serum samples by Luminex bead array
Concentrations of plasma cytokines were measured using the Bio-Plex Pro™ Human Cytokine 17-plex Assay kit (Bio-Rad Laboratories Inc, California, USA). Reconstituted standards, cytokine-specific coupled beads, detection antibodies and 50 µl of thawed serum samples were combined according to manufacturer’s instructions and data acquired with the Bio-Plex™ 200 reader (Bio-Rad Laboratories Inc, California, USA). Data were analysed using Bio-Plex Manager™ software v6.0 (Bio-Rad Laboratories Inc, California, USA).
Anti-mouse scFv assay
96-well microtiter plates were coated with MFE antibody (1 µg/ml in carbonate-bicarbonate buffer, 100 µl/well) for one hour at room temperature, washed with PBS and blocked with 5% Marvel/PBS/Tween solution (150 µl per well). After washing, the wells with incubated with either positive, negative control, PBS or patient samples diluted in 1% Marvel/PBS/Tween solution (1/100 dilution, 100 µl/well) in four replicates for 1 h. The wells were washed and incubated with 100 µl per well of the appropriate secondary antibodies diluted in 1% Marvel/PBS/Tween solution. For the patient samples, rabbit anti-human IgG was added for 1 h. The wells were washed and incubated with anti-rabbit HRP antibody (100 µl/well for 1 h). After washing, 100 µl/well substrate (o-phenylenediamine in phosphate citrate buffer) was added and the reaction stopped after 5 min by adding 4 M hydrochloric acid (50 µl/well). The OD of the wells was obtained at 490 nm. The results were recorded as positive or negative according to previously set criteria for human anti-MFE antibody assay.
CEACAM5 expression in the lung
Following discontinuation of the clinical trial, we assessed whether the observed respiratory symptoms could have been attributable to CEACAM5 expression in the lungs of the patients. We accessed nine lung tissue samples from Manchester Cancer Research Centre Biobank. Eight were non-cancerous tissue from patients undergoing resection for lung cancer and the ninth was from a patient with metastatic colorectal cancer undergoing a lung resection. IHC using the Col-1 antibody [
29] and qPCR were used to explore CEACAM5
+ expression in these samples.
Discussion
The primary end point of assessing the feasibility of delivering MFEζ CAR T cell therapy was achieved with all fourteen eligible patients receiving pre-conditioning, CAR T cells and at least two doses of IL2. However, there were challenges in meeting the higher T cell dose required in cohorts 3 and 4 where the seven patients received 0.9 ± 0.4 × 10
10 T cells that was below the proposed maximum T cell dose of 5 × 10
10 T cells. The GMP compliant production methods used in this trial were established over ten years ago [
28] and current production methods including bioreactor technology [
30] now enable the routine production of high T cell numbers in manageable culture volumes.
Whilst the trial failed to identify a CAR T cell dose that resulted in long-term persistence of the MFEζ T cells, it did confirm that an increased intensity of pre-conditioning enhanced the relative frequency of CAR T cell engraftment. Furthermore, the intensity of chemotherapy was also critical for CAR T cell function as cytokines pertinent to T cell activation were only consistently detected in cohort 4. However, there was no obvious difference in the absolute numbers of systemic MFEζ T cells suggesting a lack of in vivo CAR T cell expansion that is likely to adversely impact upon the therapeutic power of the approach.
Aside from sub-optimal culture technology, current evidence suggests that second and subsequent generations of CARs can deliver increased potency of T cell signalling, persistence and anti-tumour activity [
10,
31,
32]. A recent report of hepatic artery infused second-generation CEACAM5-specific CAR T cells with systemic IL2 and no patient pre-conditioning reported tissue localization of the CAR T cells but no evidence of prolonged persistence and limited evidence of CAR T cell effector function albeit within the early stages of the overall clinical trial [
33]. Undoubtedly, CD19 CAR T cells benefit from the ready access to antigen
+ leukaemic target cells resident within the periphery that can engage the CAR and help to drive T cell persistence. CAR T cells targeting solid tumour antigens clearly require greater help which currently includes intensive pre-conditioning but will also require additional strategies to enhance localization and challenge the strongly immune-suppressive tumour microenvironment. More recent engineering strategies such as the armoured CAR approach [
34] to alter the balance of immunity within the tumour potentially enhance clinical response.
A critical issue in this trial was the transient acute respiratory toxicity observed in patients within cohort 4 which combined with the lack of prolonged high levels of CAR T cell persistence and an absence of clinical response resulted in the early termination of the trial. Our expectation before opening of the trial was the potential for bowel toxicity due to the expression of CEACAM5 within the intestine [
35] though no evidence of bowel-related toxicity was seen. We hypothesized that the respiratory symptoms observed in patients in cohort 4 may be indicative of ‘on-target off-tumour’ MFEζ T cell binding to CEACAM5 antigen present within the lung which would be consistent with the affinity (2.5 nM) of the MFE23 scFV [
36] augmented by avidity in CAR T format. There are contradictory reports in the literature with respect to CEACAM5 expression in normal lung with some documenting relatively high levels of CEACAM5 expression in normal lung tissue [
37]; however, cross-reactivity of CEACAM5-specific monoclonal antibodies has raised questions whether the detected proteins are CEACAM5 or related CEA-family members [
29]. Our assessments demonstrated CEACAM5 expression in 5 of 9 non-cancer lung resection samples supporting the hypothesis that MFEζ T cell binding to CEACAM5 antigen present within normal lung may have contributed to this toxicity (Supplementary Fig. 6). A recent study investigating non-cancer tissue sections taken from patients with lung cancer identified the clear expression of CEACAM5 as well as CEACAM1 and CEACAM6 family members in this non-cancer tissue [
38]. This report also demonstrated that IFNγ up-regulates CEACAM5 expression on normal bronchiolar epithelial cells. Thus, CAR T cell activation may drive a local feedback loop through cytokine production up-regulating CEACAM5 expression and driving local toxicity. This is consistent with the timing of the toxicity seen where no symptoms were manifest shortly after T cell infusion which would have been consistent with acute reaction with pre-existing low levels of antigen as was seen in the acute death reported with a Her2-targeted CAR trial [
39]. The delayed toxicity (peak around day 5–7 post infusion) coinciding with peak of transduced T cells in the blood and the peak IFNγ is consistent with cytokine release and/or with delayed lung recognition because of increased MFEζ CAR T cells and/or up-regulation of CEACAM5 as a result of exposure to IFNγ [
38]. The delayed nature of the respiratory toxicity would also steer away from the likelihood of it being attributable solely to the IL2 and/or the pre-conditioning chemotherapy, although the possibility that one or both of these contributed cannot be excluded.
The transient nature of the toxicity may reflect the poor persistence of the MFEζ CAR T cells. Importantly, all cohort 4 patients were managed conservatively with supportive measures only and no immune modulation such as steroids and yet fully recovered. Although the toxicity was transient, immune modulation may have reduced the severity of toxicity but also likely to reduce engraftment and efficacy, thereby impacting upon the viability of the therapy.
Aside from IFNγ, cohort 4 patients also had elevated levels of IL-6, IL-8 and MCP-1 coupled with little modulation of IL-1β that reflects observations made in B-ALL patients receiving CD19 CAR T cells [
5]. This suggests that cytokine release including IL-6 can occur with the most basic CAR design and is dependent upon intensive pre-conditioning regimes since no respiratory-related symptoms or elevated cytokine levels were seen in cohort 3 patients. In further experiments, we attempted to recapitulate this transient toxicity through the infusion of CEA transgenic mice with high doses of MFEζ CAR T cells combined with high dose IL2 (data not shown). There was no evidence of adverse toxicities observed in these animals illustrating the limitations of pre-clinical models to accurately model the human in vivo situation.
The lack of bowel toxicity in this trial contrasts with the severe colitis seen with T cells armed with TCR specific for CEACAM5 [
40]. Presumably, this reflects the marked polarized expression of cell surface CEACAM5 that is only seen in the luminal surface of the bowel [
35], thus making it relatively immune-protected from T cell cytotoxic activity as compared to the non-polarized expression profile of peptide-bound HLA molecules. The pulmonary toxicity in our trial serves to illustrate the challenges in isolating specific causes of toxicity in adoptive cell therapy where many will be multifactorial and inter-related and where there are multiple variables across different trials. These variables range from the pre-conditioning chemotherapy, IL2 regime (if used), variation in the scFv affinity for its target, dose of cells, manufacturing differences, to name but a few.
No significant treatment-related neurological toxicity was noted in this trial where all patients received a moderate dose of fludarabine (25 mg/m
2/day for 5 days) with first-generation CAR T cell therapy. However, there have been a number of reports of significant neurologic toxicities observed in CAR T cell trials. Notably a recent trial sponsored by Juno Therapeutics was halted after a number of patients died having developed cerebral oedema attributed initially to intensified lymphodepleting chemotherapy with fludarabine [
41]. The trial was allowed to resume without the intensified chemotherapy, but this was not sufficient to ameliorate the toxicity as further patient deaths subsequently occurred. Whilst the significant toxicity observed in the Juno trial is not mirrored in our trial, the details of the mechanism of the toxicity will have an important influence on the direction for the field as a whole.
To conclude, this trial underlines the importance of pre-conditioning chemotherapy for CAR T cell therapy and highlights the need to design CAR T cells to maximize the discrimination between high-level target expression within the tumour and low level within some normal tissues to avoid toxicity. The powerful avidity effect of CAR T cells means that lower affinity antibodies for CAR T cell therapy may provide better discrimination between low-level expression on normal tissue and high level on the tumour. However, where such discrimination is not possible and clinical benefits are seen, approaches such as local steroids to manage on-target, off-tumour toxicity will be essential to fully explore the therapeutic potential of this approach in patients with solid tumours. CEACAM5 remains a potentially useful target for CAR T cells with those caveats.