Review

Special Focus Issue: Adoptive cell immunotherapy for cancer

For reprint orders, please contact: reprints@futuremedicine.com

CAR T-cell therapy: toxicity and the

relevance of preclinical models

Chimeric antigen receptor (CAR) T cells form part of a broad wave of immunotherapies Milena Kalaitsidou1, Gray
that are showing promise in early phase cancer clinical trials. This clinical delivery has Kueberuwa1, Antje Schütt1
been based upon preclinical efficacy testing that confirmed the proof of principle of & David Edward Gilham*,1
1
Clinical & Experimental Immunotherapy
the therapy. However, CAR T-cell therapy does not exist alone as T cells are generally
Group, Institute of Cancer Sciences,
given in combination with patient preconditioning, most commonly in the form of Academic Healthcare Science Centre,
chemotherapy, and may also include systemic cytokine support, both of which are University of Manchester, Manchester
associated with toxicity. Consequently, complete CAR T-cell therapy includes elements Paterson Building, Wilmslow Road,
where the toxicity profile is well known, but also includes the CAR T cell itself, for Withington, Manchester, M20 4BX, UK
*Author for correspondence:
which toxicity profiles are largely unknown. With recent reports of adverse events
Tel.: +44 161 446 3236
associated with CAR T-cell therapy, there is now concern that current preclinical Fax: +44 161 446 3236
models may not be fit for purpose with respect to CAR T-cell toxicity profiling. Here, David.Gilham@ ics.manchester.ac.uk
we explore the preclinical models used to validate CAR T-cell function and examine
their potential to predict CAR T-cell driven toxicities for the future.

Keywords:  adoptive cell therapy • CAR T cells • cytokine storm • preclinical models • toxicity

Chimeric antigen receptor (CAR, also known
as chimeric immune receptor [CIR] and
T body) T-cell technology has developed rap-
idly over the last 20 years. The initial concept
featured the fusion of T-cell signaling proteins,
including the T-cell receptor β chain, CD3γ
and CD3ζ with target-binding domains.
When expressed in a T cell, these fusion pro-
teins result in T-cell effector functions being
controlled by the ligand binding specificity of
the target-binding domain [1] . Over time, this
linear arrangement of targeting domain fused
to one signaling domain has been termed ‘first-
generation’ CARs. Upon this basic structure,
a diverse range of CARs have been developed
and shown to function. The target-binding
domain most commonly used involves single-
chain antibody variable fragments (scFvs)
derived from monoclonal antibody hybrid-
omas or by phage display technology. Though
less common, intact extracellular protein
domains and receptor ligands have also been
successfully used in the CAR context [2,3] . The
explosion of interest in CARs is highlighted by
the wide number of excellent recent reviews
detailing CAR structures and issues relating
to the clinical delivery of this therapy [4–10] . In
brief here, to provide an overview of the diver-
sity of CARs, the modular nature of CARs has
encouraged the incorporation of multiple sig-
naling domains to form hybrids that increase
downstream signaling potency (Figure 1).
Hence, CARs containing two signaling
domains fused together are termed ‘second
generation,’ while those bearing three are listed
as ‘third generation.’ The combinations of sig-
naling domains vary, however, for the majority,
the cytosolic sequence from CD3ζ protein is
included to provide the main T-cell activating
signal. Recently developed ‘fourth-generation’
CARs (or TRUCKs) involve two separate
transgenes with a first, second- or third-gener-
ation CAR expressed alongside a T-cell activa-
tion responsive promoter linked to a cytokine
such as IL-12. Consequently, upon T-cell acti-
vation, high levels of IL-12 are produced that
can modulate the local microenvironment and
part of
aid CAR T-cell function [11] .

10.2217/IMT.14.123 © 2015 Future Medicine Ltd Immunotherapy (2015) 7(5), 487–497 ISSN 1750-743X 487
Review  Kalaitsidou, Kueberuwa, Schütt & Gilham

Antibody CARs TRUCKs

Fab 1st 2nd 3rd 4th

ScFv

Spacer

Transmembrane domain
CD3ζ CD28/ CD28
Signaling domain
4-1BB
4-1BB
CD3ζ OX-40 NFAT
CD3ζ IL-12
(IL-2)
Cytotoxicity (IL-8)
IL-12
Proliferation, persistence
NFAT min. prom.
Survival

Figure 1. Chimeric antigen receptors. First-generation CARs contain solely the CD3ζ chain as a single activation
domain, leading to T-cell activation and cytotoxicity based on scFv specificity. Second- and third-generation
CARs consist of one or two additional costimulatory signaling domains, respectively, such as CD28, CD27, OX-40
(CD134) and 4-1BB (CD137). Costimulation enhances the overall survival as well as proliferation and persistence of
activated T cells. TRUCKs represent the fourth generation and are currently the newest approach in adoptive T-cell
therapy. T cells are genetically modified to express CARs along with an inducible cytokine gene cassette driven
by an NFAT sensitive promoter. Consequently, immune stimulatory cytokines such as IL-12 are secreted upon CAR
engagement.
CAR: Chimeric antigen receptor; min. prom.: Minimum promoter. 

Against this background of CAR structural diver-
sity, there is no current consensus on the optimal com-
bination of signaling domains and also no obvious
rules that define the extracellular domains upon which
target binding by scFv is dependent. Consequently,
clinical studies arising from this diversity have pro-
ceeded in a largely empirical manner, with individual
research groups using locally developed CARs with
differing targeting moieties against varying diseases,
associated with different levels of patient precondition-
ing and using T cells generated using different ex vivo
systems  [12] . Despite this, objective clinical responses
have been reported in several centers, predominantly
targeting B-cell leukemia through the CD19 antigen
(see  [8] for recent review), and this is driving a major
surge of interest and activity in CAR T-cell therapy [13] .
Few, if any, therapies deliver clinical responses in
the absence of some form of toxicity and CAR T-cell
therapy is no exception to this. The clinical success of
targeting the CD19 antigen with ‘second-generation’
4-1BB.CD3ζ CARs (CART019) is associated with
cytokine-release syndrome (CRS), in which IL-6,
IL-10 and IFN-γ play a major role, that is potentially
life-threatening if not appropriately managed [14] .
Examples of main strategies used to overcome CRS
involve direct targeting of the cytokine action by
blocking access to their receptors (i.e., tocilizumab)
or indirect reduction of cytokines by removing the
transduced cells via lymphodepletion using cortico­
steroids  [14] . Similar management strategies for cyto­
kine-release syndrome are reported in acute B cell lym-
phoblastic leukemia patients receiving CD28.CD3ζ
CAR T cells [15,16] . In the very worst-case scenario, tar-
geting of normal, essential tissue through the CAR can
result in rapid autotoxicity and can lead to death. This
has occurred in a colorectal cancer patient with lung
and liver metastases who received treatment with a
Her2/neu-specific ‘third-generation’ CAR T cells [17] .
This highlights the fine line between achieving a clini-
cal anticancer response and preventing autotoxicity as
a result of CAR T-cell therapy. Antitumor activity of
CAR T cells results in the production of cytokines that
are advantageous through the direct killing of target
cells and recruiting other elements of the immune sys-
tem to elicit an enhanced antitumor response. How-
ever, ensuring that the magnitude of response does
not exceed the threshold for causing systemic toxicity
and that CAR T cells are only targeted to tumor and
not healthy tissue, remain major hurdles for clinicians
using CAR T cells. Going forward, will it be possible

488 Immunotherapy (2015) 7(5) future science group


to identify such potential toxicities in the preclinical
setting that will then allow further refinement of the
CAR strategy to enhance efficacy in the absence of
autotoxicity?
Getting CAR T cells to the clinic: preclinical
models to deliver the proof of principle
The majority of preclinical studies investigating CAR
T-cell function to date have focused upon verifying
specificity and potency of antitumor activity. Speci-
ficity has been generally confirmed by in vitro assays
establishing that the T cells endowed with a specific
CAR show a differential functional response between
the target antigen (either in purified format or on rel-
evant target cells) and nonspecific antigens. The key
advantage of the CAR is the ability to redirect T-cell
effector function in the absence of HLA-restriction.
Consequently, allogeneic T-cells can be readily used to
demonstrate this in vitro specificity without the need
for matched targets. Background activity, which is
the result of allo-recognition, is generally equivalent
between cells transduced with irrelevant CARs and
nontransduced T cells. This is typically measurable
in vitro and the benefit of CARs is seen as fold increase
above this background. Such assays prove CAR func-
tion and support translation to the animal model sys-
tem. It should be noted that background allo-recogni-
tion has not been associated with antitumor activity
in vivo in a large number of xenograft tumor studies.
For clinical translation, regulatory authorities
generally require information relating to the specific
agent to be used in the clinical trial. In the case of
CAR T cells, this means that testing in immune-com-
promised mouse models represent the only possible
route to assess the actual clinical agent. However, the
success of human T-cell engraftment in the majority
of immune-compromised mice is limited since the
residual elements of the mouse immune system still
challenge the engrafting human T cells. Nonethe-
less, several studies have shown efficacy of CAR T-cell
function using Nude, NOD/SCID or SCID/Beige
animals (for examples see [18–21]). More recently,
the availability of the highly immune-compromised
NOG/NSG mouse (NOD/SCID IL-2Rγ-/-) has
allowed efficient human T-cell engraftment. However,
engrafted T cells can drive a xeno-graft versus host
disease (xGVHD) which occurs around 50 days after
adoptive T-cell transfer at doses above 109 cells/kg [22] .
The occurrence of xGVHD is associated with trans-
duced cell persistence and this is affected by the cyto-
kine conditions used to culture the cells prior to injec-
tion into mice. Consequently, this limits the ability
to investigate the long-term effects of CAR T cells in
this model [22–24] .
future science group
CAR T-cell therapy: toxicity & the relevance of preclinical models  Review
Moreover, the majority of these studies involve CAR
T-cells targeting human antigens whose expression is
restricted to the human tumor cells used to challenge
mice in the model. Thus, targeting of the antigen which
may be expressed on normal, healthy tissue cannot be
evaluated. To examine this, immune-competent mouse
models have been used where the target antigen is the
mouse homologue of the human target antigen [25] .
While these models provide an array of additional infor-
mation about CAR T-cell function with background
antigen expression and a functional immune system,
‘mouse CAR-mouse homologue’ interactions may dif-
fer from the equivalent human scenario. In addition,
this type of model is dependent upon minimal disparity
between expression profiles of mouse and human pro-
tein homologues. Another approach is to utilize trans-
genic animals expressing the human antigen under the
control of a physiologically relevant promoter where the
CAR specific for the human antigen can be tested [26] .
These combined approaches have been used to suc-
cessfully demonstrate CAR T-cell function against
a wide diversity of target antigens, with the growing
interest in CAR T-cell approaches confirmed by the
rapidly rising number of publications specifically citing
CARs (Figure 2) .
Potential mechanisms of
CAR T-cell-mediated toxicity
While the complete CAR T-cell therapy involves
administration of agents that have well described tox-
icity profiles, toxicities arising from CAR T cells them-
selves are not as well defined. There are three poten-
tially key routes by which CAR T cells may contribute
to toxicity and must be considered (Figure 3) .
On-target, on-tumor toxicity
On target, on tumor toxicities are generally acute and
relate directly to the effects of binding of the CAR to
cognate antigen resident on the target tumor cell. The
underlying premise of the therapy is the activation of
T-cell effector functions as a result of successful CAR
engagement, including cytokine release and cyto­
toxicity. However, excessive cytokine production may
result in cytokine release syndrome typified by chills
and fevers but can also result in much more severe,
potentially life threatening reactions [27] . Moreover,
the rapid destruction of large quantities of tumor can
result in tumor lysis syndrome due to the abrupt release
of tumor cellular contents causing an array of systemic
metabolic disturbances [28] .
On-target, off-tumor toxicity
In this situation, the CAR T-cell engages the target
antigen that is expressed upon healthy, nontumor cells.
www.futuremedicine.com 489
Review  Kalaitsidou, Kueberuwa, Schütt & Gilham

B cells. Successful targeting of CD19 by CAR T cells

100
leads to ‘on-target, off-tumor’ aplasia of developing
Number of Publications 80
B cells but spares CD19-negative plasma cells and,
crucially, hematopoietic stem cells, allowing repletion
60 of the B-cell compartment post-CAR T-cell therapy.
This B-cell aplasia may be an acceptable side effect
40 due to clinical management strategies that are avail-
able. Indeed, B-cell aplasia has been used as surrogate
20 readout for CD19 CAR T-cell activity in preclinical
models  [29] . Moreover, the use of lymphotoxic steroids
0 can eradicate CAR T cells and allow B-cell repletion
1980

1990

2000

2010
Year
Figure 2. In vivo application of chimeric antigen
receptors. Timeline showing the number of publications
citing chimeric antigen receptors utilized between 1980
and 2014, highlighting an intensive increase in interest
in the past 5 years (search terms: ‘chimeric antigen
receptor’ or ‘chimeric immunoreceptor’ or ‘chimeric
T-cell receptor’ or ‘chimeric immune receptor’).
Such targeting can result in the destruction of healthy
cells expressing the specific target antigen. In the case
of B-cell malignancies, the CD19 antigen is expressed
on cells of B-cell lineage from pro-B cells to mature
if required [30] .
However, in the case of many solid tumors, the CAR
target may not be tumor specific but a tumor-associ-
ated antigen (TAA) where expression of the target anti-
gen is high on tumors but may be present at low levels
on healthy tissue. In this case, CAR T cells may target
and respond to such healthy tissue, and where this tis-
sue resides in an organ essential for life, catastrophic
consequences may occur [31] . With these toxicities in
mind, target antigen selection is probably the most crit-
ical determinant to successful CAR therapy. Antigens
with strict tumor specificity are ideal (e.g., EphA2 [32] ,
mutated EGFRvIII [33,34] and ‘unmasked’ Mucin-1
with truncated glycosylations [35]), or at least where

T cell 2 On-target off-tumor

T cell
T cell T cell Healthy tissue
T cell
T cellT cell T cell
Nontumor
T cellT cell tissue
T cell

Excessive CAR-T-cell activation

• Cytokine storm

T cell 3 Off-target off-tumor

1 On-target on-tumor

Tumor lysis syndrome Nonspecific CAR Activation of

• Tumor-derived toxicity affinity endogenous MDSCs
TCRs

Nontumor
tissue
MDSC
Tumor

Figure 3. Categorization of chimeric antigen receptors toxicities. Once T cells are activated their cytotoxic potential can lead to
undesired side effects. (1) Even in the optimal scenario of specific activation of CAR T cells via their target expressed on tumor tissue
can lead to toxicity. This is termed on-target on-tumor toxicity and may be due to excessive T-cell activation with (A) a Th1 dominated
cytokine storm or (B) due to a very high tumor load leading to massive destruction of tumor tissue, resulting in tumor lysis syndrome.
(2) In cases where the target is also expressed on normal tissue it is likely that on-target off-tumor toxicity will occur, subsequently
damaging healthy tissue. (3) Additionally, there is a possibility of off-target off-tumor toxicity. This can occur through, (C) CAR T-cell
activation by similar but distinct epitopes or the endogenous T-cell receptor engagement, or (D) Th2 cytokine secreting MDSCs
associated with long-term engraftment of CAR T cells.
CAR: Chimeric antigen receptor; MDSC: Myeloid-derived suppressor cell; TCR: T-cell receptor.

490 Immunotherapy (2015) 7(5) future science group


background antigen expression is on nonessential tis-
sues (e.g., prostate-specific antigen, NY-ESO [36] ,
MAGE  [37,38] , CD19 [39] and CD20 [40]). In practice,
however, such antigens have been difficult to identify,
particularly in the setting of solid malignancies. Given
a known background expression of a target antigen it
becomes paramount to determine whether levels are
over a threshold that can cause ‘on-target, off-tumor’
toxicity, and to determine the potential severity thereof
in humans.
Off target, off tumor toxicity
This third potential mechanism of CAR T-cell toxicity
has arisen in experimental models of CAR T-cell ther-
apy recently and relates to the response of non-CAR
T cells to the therapy [41] . Prolonged, potentially low
level, CAR T-cell activity may indirectly impact upon
the local environment resulting in skewing of normal
homeostatic cellular responses, that in turn develop
into toxicities. This mechanism of toxicity is discussed
in further detail below.
Genotoxicity
Integrating vectors based upon retroviral and lentiviral
backbones have been most commonly used to facili-
tate the stable expression in primary T cells. However,
such vectors may pose a potential risk of oncogenic
events including disruption of normal gene expression
as observed in the hematopoietic stem cell situation
with the loss of control of the LMO2 gene in X-linked
SCID [42] . However, thus far, no such toxicity has been
reported in the context of CAR therapy [43] , but this is
clearly an important consideration for the future where
CAR T cells may prevail for life time of the treated
patient.
CAR T-cell toxicity in preclinical models
Across the breadth of preclinical publications focusing
upon CAR T cells, there are very few that document
toxicity, even where toxicity has been observed in the
clinic. One example involves CAR T cells targeting the
Her2/neu (ErbB2) antigen which is overexpressed on a
range of tumors. There are a wealth of studies docu-
menting the antitumor potency of Her2/neu-specific
CAR T cells, including studies targeting this antigen
in Her2 transgenic animals combined with lympho­
depletion  [44] or anti-PD-1 antibody therapy [45] . In
each case, there has been no evidence of any pathol-
ogy associated with the Her2/neu-directed therapy,
yet a patient receiving Her2/neu-specific CAR T cells
died, potentially as a result of cytokines released due
to targeting of Her2/neu present on lung epithelial
cells [17] . The reasons for this dichotomy between pre-
clinical results and the patient situation are not clear
future science group
CAR T-cell therapy: toxicity & the relevance of preclinical models  Review
but are likely to include factors such as differences in
CAR structures, T-cell biology between the mouse
and human and differential drug metabolism capacity
between species [46] . However, while these observations
raise questions concerning the ability of current model
systems to predict toxicities, there have been recent
reports of toxicity in preclinical models systems in
three specific CAR T-cell experimental investigations
that illustrate the mechanisms of CAR T-cell toxicity.
On target off-tumor toxicity resulting from
‘atypical’ CAR targeting domains driving
toxicity through the recognition of multiple
target antigens present on healthy tissue
In contrast to the absence of toxicity observed using
CAR’s armed with scFv targeting Her2/neu in pre-
clinical model systems, human T cells armed with the
T1E28z second-generation CAR, which binds ErbB-1
homodimers and also ErbB2/3 heterodimers, dis-
played toxicity in mice [47] . This CAR transgene was
co-expressed with a chimeric protein consisting of the
IL-4α chain fused to the IL-2/IL-15 β chain permit-
ting IL-4-driven selection of gene-modified T cells [47] .
Interestingly, the route of administration of T1E28z
CAR T cells strongly influenced toxicity and antitumor
efficacy in the SCID/Beige mouse. CAR T cells admin-
istered by the intravenous route demonstrated reduced
tumor growth with no outward symptoms of adverse
effects and no evidence of pathological abnormality at
necropsy 7 days post-CAR T-cell infusion [48] . How-
ever, intraperitoneal administration of T1E28z CAR
T cells to challenge local tumor xenografts resulted in
toxicities associated with a major cytokine release syn-
drome mediated largely by human IFNγ, human IL-2
and mouse IL-6 that were readily detectable within the
circulation of mice receiving a lethal dose of T1E28z
CAR T cells [48] . Clodronate depletion of macrophages
prior to CAR T-cell transfer confirmed that these cells
are the likely source of mouse IL-6 and, interestingly,
this also reduced human IFN-γ levels. The dose and
volume of administered CAR T cells along with tumor
all appeared to correlate with toxicity. The expression of
ErbB proteins on healthy tissue and, as the authors sug-
gest, specifically upon normal mesodermal cells within
the peritoneal cavity may also have resulted in the pro-
found on-target off-tumor toxicity associated with the
T1E28z CAR T cells. Nonetheless, targeting of multiple
ErbB proteins was associated with substantial antitumor
activity underscoring the potential of the approach [48] .
Taken together, controlling the route of administra-
tion and reducing tumor burden prior to adoptive CAR
T-cell transfer would be important factors to potentially
optimize the therapy while reducing adverse side effects.
Indeed, the authors of this study have designed a Phase I
www.futuremedicine.com 491
Review  Kalaitsidou, Kueberuwa, Schütt & Gilham
trial protocol using intratumoral infusion of T1E28z
CAR T cells in head and neck cancer as a practical
means to avoid such undesired toxicity [49] .
On-target off-tumor toxicity driven by
‘standard’ CAR’s targeting rare, specialized cell
types
Whilst the direct targeting of tumor cells by CAR
T cells is the natural mechanistic choice for such a
therapeutic approach, the inherent genetic instability
of tumors raises the possibility that antigen-negative
variants may arise and allow the tumor to effectively
circumvent targeting by the CAR. Targeting of non-
neoplastic tumor associated stromal cells potentially
avoids this issue and can be achieved through targeting
antigens such as fibroblast activation protein α (FAP),
which is expressed on cancer associated fibroblasts.
T cells armed with a ‘second-generation’ CAR with
an FAP-specific scFv CAR showed effective antitumor
activity against established A549 lung cancer cells in
SCID mice [50] and FAP+HT1080 cells in the NSG
mouse model [51] . In the lung cancer model, on-target
off-tumor toxicity was considered but no evidence of
adverse pathology was observed in mice sacrificed at
short time points after CAR T-cell infusion [50] . The
human antigen restriction of the CAR used in the
mesothelioma study prevented any in depth study of
toxicity due to lack of background somatic expres-
sion  [51] . By contrast, mouse T cells engrafted with
a FAP-specific third-generation CAR (consisting of
CD28, 4-1BB and CD3ζ signaling domains) drove
significant morbidity and mortality in recipient mice,
including a drastically diminished frequency of live
bone marrow and osteogenic cells resident within
femurs and tibia [52] . Further analysis identified a pre-
viously unknown FAP+ multipotent bone marrow stem
cell population, which was depleted in FAP-specific
CAR immunotherapy resulting in the lethal bone
toxicity and cachexia [52] . Subsequent to this, stud-
ies employing a second-generation FAP-specific CAR
reported encouraging antitumor activity against a
number of tumor targets with no evidence of on-target
off-tumor toxicity [53] .
The question is ‘how such a major discrepancy is
possible between effectively similar studies targeting
the same antigen?’ The answer may lie in the total ther-
apy. In the report describing toxicity, third-generation
CAR T cells were employed along with a total body
irradiation preconditioning regimen (5Gy) and six
doses of systemic IL-2 support (2.2 × 105 IU/dose) [52] .
By contrast, the three other publications used second-
generation CAR’s in the absence of preconditioning
and only one gave supportive IL-2 albeit at a reduced
dose (1500 IU, single dose) [50,51,53] . Additionally, the
492 Immunotherapy (2015) 7(5)
different scFvs used between the studies may them-
selves endow the specific CARs with differential activi-
ties. Consequently, there are many potential individual
and combinatorial factors that could underlie these
differential observations and further study is clearly
required. However, the initial suggestion is that FAP
targeted toxicity resulted from the combination of
third-generation CAR and optimal conditions to drive
CAR T-cell engraftment. In summary, studies exam-
ining FAP targeting highlight the potential impact of
the full therapy and the potency of third-generation
CARs in targeting non-neoplastic, tumor associated
stromal cells [52] .
Chronic off-target off-tumor toxicity resulting
from CD19 targeted CAR T cells
The cytokine release syndromes observed in certain
patients receiving CD19 CAR T cells [14] have not been
exactly reproduced in mouse model systems. However,
a recent study from our group has reported short-term
Th1-biased cytokine driven toxicity in mice receiving
mouse CD19 targeted, second-generation CAR T cells
and also reported a chronic toxicity resulting from the
specific expansion of myeloid cells in response to Th2-
biased cytokine production arising from chronic CD4 +
CAR T-cell activity [41] . In this system, acute toxicity
manifested as rapid weight loss and cachexia with death
occurring 1–5 days after adoptive T-cell transfer. How-
ever, this acute toxicity primarily occurred after the
adoptive transfer of T-cells engrafted with a CAR con-
taining the full length CD28 sequence fused to CD3ζ.
T cells expressing CARs bearing a truncated CD28
extracellular domain fused to CD3ζ induced a reduced
severity of weight loss and the animals survived. How-
ever, a significant number of animals engrafted with
either CD28-based CAR developed symptoms includ-
ing weight loss and cachexia resulting in death from
approximately 2 months after adoptive transfer. Tumor
burden was not implicated as causative as animals lack-
ing tumor also suffered the same fate. Upon necropsy,
affected animals presented with enlarged spleen and
lymph nodes filled with granuloma like cells, shown
to consist of large numbers of CD11b + Gr-1+ myeloid
cells reminiscent of myeloid-derived suppressor cells
(MDSC). Cytokine analysis showed increased levels of
Th2 cytokines, specifically IL-10 and IL-13 that were
postulated to be derived from the large numbers of
CD4 + CAR T cells identified within secondary lym-
phoid structures. Interestingly, affected animals were
essentially devoid of CD8 + CAR T cells and B cells.
Together, this suggests that chronic stimulation of
resident CD4 + CAR T cells, most likely due to B cells
emerging from hematopoietic progenitor cells, caused
production of Th2 cytokines that mediated the expan-
future science group

sion of the myeloid cell compartment resulting in a
lethal granulomatous reaction [41] .
The acute and chronic toxicities observed with the
mouse CD19-specific scFv appears to be directly related
to the mouse strain used with toxicity apparent in the
BALB/c strain but largely absent in C57BL/6 or C3H
strains. This suggests that the Th2 bias of the mouse
strain is a critical element in driving this observed toxic-
ity. Nonetheless, this study demonstrates the auto-toxic
potential that exists with chronic long-term stimulation
of CAR T cells in vivo and poses the question whether
the benefits of long-lived immunotherapeutic cells that
could offer prolonged immunological memory and pro-
tection against subsequent tumor growth may be coun-
ter-balanced by the possible toxicity induced through
extended CAR T-cell stimulation.
Overcoming CAR T-cell-related issues
Each of these ‘issues’ lies at the core of the successful
antitumor activity of the CAR T cell itself. Thus, there
is a fine balancing act to be achieved between tumor
elimination and unwanted treatment-related toxicity.
To control this, many groups have developed suicide
gene systems, including HSV-TK, iCasp9 and CD20,
that once activated can selectively eliminate the CAR
T-cell population and, thereby, eliminate the symp-
toms of CAR T-cell-driven toxicity [54,55] . However,
this represents the least preferred route to manage this
issue since depletion of the CAR T cell will also mean
abrogating the therapeutic potential of the CAR T cell.
In addition, the efficacy of these approaches in manag-
ing potentially life threatening acute toxicities remains
to be evaluated.
Clearly, the targeting of truly tumor-specific targets
is the main goal, and selection systems based upon the
CAR T cell may facilitate identification of such targets
rather than adapting existing CARs [56,57] . However,
in the absence of truly tumor-specific target antigens,
recently developed systems employing dual CAR tar-
geting with antigen specificity for separate tumor anti-
gens; one CAR containing a ‘signal 1’ and the other
with a ‘signal 2’ activating domain, both signals being
required for full T-cell activation. The reduced likeli-
hood of two separate tumor antigens being simultane-
ously expressed on healthy tissue providing increased
tumor specificity, thus avoiding on-target off-tumor
toxicity  [58] . Likewise, this has been proven in prin-
ciple for TanCARs; bispecific CARs that bind two tar-
get antigens for full downstream signaling. This strat-
egy also provides a means to improve tumor specificity
of CAR T cells with potential for avoiding antigen
escape  [59] . Alternatively, inhibitory CARs (iCARs)
where the CAR is engineered to deliver control sig-
nals upon interaction with healthy tissues to regulate
future science group
CAR T-cell therapy: toxicity & the relevance of preclinical models  Review
cytokine secretion may allow for dynamic control of
CAR T cells to avoid unwanted on-target, off-tumor
toxicities [60] .
Transient expression of CARs may provide the
short-duration of redirected T-cell activity that is
required to eliminate tumor. T-cells transfected with
mRNA encoding the CAR receptor are functional
in mouse models [61] although there is some question
of the ability of mRNA transiently expressing CAR
T cells to deliver durable antitumor responses in solid
tumor models [62] . The use of nonintegrating, transient
expression vectors also avoids the potential of any pos-
sible long-term genotoxicities. Finally, the type of T cell
used for adoptive transfer may be critical, with T cells
displaying a less differentiated phenotype potentially
delivering improved therapy in vivo [63] . This has been
shown to be applicable to CARs in a recently study
by Xu and colleagues showing that CD19 transduced
T memory stem cells cultured in IL-7 and IL-15 cyto-
kines expanded more efficiently and had more potent
survival and antitumor effect in preclinical models [64] .
These examples of technological developments will
impact upon the CAR T-cell approach and assessing
the impact of each upon toxicity will be the challenge
to deliver practical information that can aid clinical
trial design.
Conclusion
CAR T-cell therapy has rapidly risen to prominence
largely as a result of highly impressive clinical results
emanating from several clinical centers employing
CD19-specific CAR technology and the subsequent
rapid influx of enthusiasm from Industry eager to
exploit this potentially powerful therapy. However,
the therapeutic approach is still very much in devel-
opment with the ability to predict toxicity resulting
from the activity of CAR T cells still in its infancy.
Consequently, clinical trials are being initiated without
significant input from models that may have predic-
tive value to aid their design. At present, preclinical
model systems are being used to determine whether the
adverse events observed in trials can be recapitulated in
mice to permit an examination of underlying mecha-
nisms driving toxicity. At the moment, for the large
part, such recapitulation of the clinical situation has
not been achieved. Toxicity induced by CD19 CAR
T cells in the BALB/c autologous model system does
not completely reflect that observed in patients, espe-
cially considering that IL-6 does not appear to be a
major driver of acute toxicity unlike that observed in
patients [41] . Moreover, Her2/neu targeted CAR T cells
have failed to show any evidence of autotoxicity in pre-
clinical models yet the death of a patient receiving such
CAR T cells [31] testifies the potency of the therapy and
www.futuremedicine.com 493
Review  Kalaitsidou, Kueberuwa, Schütt & Gilham

highlights the poor predictive nature of current model
systems.
Whilst the poor predictive nature of current models
may relate to biological differences between rodents
and humans, there is also the case that therapeutic effi-
ciency has been the underlying driver to develop CAR
T-cell therapy with toxicity taking the back seat to this
central aim. With the observations of toxicity in the
clinic, there is now a stronger drive to develop a better
understanding of the potential mechanisms of toxicity.
From the current literature examining CAR T-cell
toxicity in preclinical models, there are some general
issues that appear to correlate with toxicity:
• Where toxicity occurs, this is generally exacerbated
by higher cell doses. In most cases to date, reduc-
ing CAR T cell dose reduces the severity of the
observed toxicity;
• The relative signaling potency of CARs appears to
correlate with toxicity. First-generation CARs appear
to drive minimal levels of toxicity, while second and
third-generation CARs tend to be associated with
acute and chronic toxicity;
• Enhanced CAR T-cell engraftment resulting from
patient preconditioning and potentially, through
the use of systemic cytokine support can correlate
with increased toxicity levels;
• High levels of target antigens present on tumor
or nontumor tissue can drive cytokine release
­syndromes underpinned by CAR T-cell activity.
As described, various strategies are currently in
development to overcome toxicities observed so far
with CAR T-cell therapy. These include identification
of targets with high tumor specificity, incorporation
of suicide genes, targeting multiple antigens utilizing
co-operative CARs, TanCARs or iCARs, transient
expression of the CARs via mRNA electroporation or
using less differentiated populations of T cells grown
in variable cytokine conditions. These approaches are
likely to provide means by which to improve clinical
outcomes of CAR therapy while subverting toxicities
observed thus far.
Future perspective
Over the last decade CAR T-cell therapy has pro-
gressed from a relatively unknown field to a highly
studied one. Successful treatments of both hemato­
logical and solid malignancies have increased interest
in developing this therapy. Over the next 5–10 years,
this treatment method will likely progress to a first line
treatment for certain hematological cancers and in the
more distant future, will likely be applied to a much
broader range of neoplasms. Several novel strategies
are emerging to overcome existing hurdles to this end,
which include more consistent achievement of com-
plete remissions as well as the reduction of treatment
related toxicities. The most important probably being
identification of tumor-specific antigens, appropriate
signaling domains and optimization of each CAR on
a case by case basis.
Financial & competing interests disclosure
DE Gilham is a cofounder of Cellular Therapeutics Ltd. This
work was supported by Leukaemia and Lymphoma Research,
the Kay Kendall Leukaemia Fund, Children with Cancer and
the Medical Research Council. The authors have no other rele-
vant affiliations or financial involvement with any organization
or entity with a financial interest in or financial conflict with
the subject matter or materials discussed in the manuscript
apart from those disclosed.
No writing assistance was utilized in the production of this
manuscript.

Executive summary
• Types of toxicity mediated by chimeric antigen receptor (CAR) therapy include: ‘on-target, on-tumor’, ‘on-
target, off-tumor’ and ‘off-target, off-tumor’.
• The most serious side effects seen so far result from on-target off- tumor toxicity, therefore one of the main
hurdles of CAR therapy is the identification of antigens with high tumor specificity.
• Evidence thus far indicates the increased efficacy of second- and third-generation CARs as compared with
first-generation CARs is also associated with an increased toxicity.
• Increased input CAR T-cell dose and increased tumor burden correlate with more severe toxicity. Route of
administration and T-cell culturing conditions may also play a role.
• Current preclinical models have proven inadequate for predicting CAR toxicity in patients. However, adverse
events in clinical trials have raised the awareness of the dangers of CAR toxicity, therefore more recently there
has been a marked effort to develop improved animal models, with some success.

1 Gross G, Waks T, Eshhar Z. Expression of immunoglobulin-

References T-cell receptor chimeric molecules as functional receptors
Papers of special note have been highlighted as: with antibody-type specificity. Proc. Natl Acad. Sci.
 • of interest; •• of considerable interest USA 86(24), 10024–10028 (1989).

494 Immunotherapy (2015) 7(5) future science group


2 Kahlon KS, Brown C, Cooper LJN, Raubitschek A,
Forman SJ, Jensen MC. Specific recognition and killing
of glioblastoma multiforme by interleukin 13-zetakine
redirected cytolytic T cells. Cancer Res. 64(24), 9160–9166
(2004).
3 Ramos CA, Dotti G. Chimeric antigen receptor (CAR)-
engineered lymphocytes for cancer therapy. Expert Opin.
Biol. Ther. 11(7), 855–873 (2011).
4 Bridgeman JS, Hawkins RE, Hombach AA, Abken H,
Gilham DE. Building better chimeric antigen receptors
for adoptive T cell therapy. Curr. Gene Ther. 10(2), 77–90
(2010).
5 Dotti G, Gottschalk S, Savoldo B, Brenner MK. Design and
development of therapies using chimeric antigen receptor-
expressing T cells. Immunol. Rev. 257(1), 107–126 (2014).
6 Jensen MC, Riddell SR. Design and implementation of
adoptive therapy with chimeric antigen receptor-modified
T cells. Immunol. Rev. 257(1), 127–144 (2014).
7 June CH, Maus MV, Plesa G et al. Engineered T cells
for cancer therapy. Cancer Immunol. Immunother. 63(9),
969–975 (2014).
8 Maus MV, Grupp SA, Porter DL, June CH. Antibody-
modified T cells: CARs take the front seat for hematologic
malignancies. Blood 123(17), 2625–2635 (2014).
9 Park TS, Rosenberg SA, Morgan RA. Treating cancer with
genetically engineered T cells. Trends Biotechnol. 29(11),
550–557 (2011).
10 Sadelain M, Brentjens R, Riviere I. The basic principles
of chimeric antigen receptor design. Cancer Discov. 3(4),
388–398 (2013).
11 Chmielewski M, Hombach AA, Abken H. Of CARs
and TRUCKs: chimeric antigen receptor (CAR) T cells
engineered with an inducible cytokine to modulate the tumor
stroma. Immunol. Rev. 257(1), 83–90 (2014).
12 Lipowska-Bhalla G, Gilham DE, Hawkins RE, Rothwell
DG. Targeted immunotherapy of cancer with CAR
T cells: achievements and challenges. Cancer Immunol.
Immunother. 61(7), 953–962 (2012).
13 Vonderheide RH, June CH. Engineering T cells for cancer:
our synthetic future. Immunol. Rev. 257(1), 7–13 (2014).
14 Maude SL, Barrett D, Teachey DT, Grupp SA. Managing
cytokine release syndrome associated with novel T cell-
engaging therapies. Cancer J. 20(2), 119–122 (2014).
• Highlights the key role of IL-6 in chimeric antigen receptor
(CAR)-mediated toxicities and suggests that it may be a
good target to overcome cytokine release syndrome.
15 Davila ML, Riviere I, Wang X et al. Efficacy and toxicity
management of 19–28z CAR T cell therapy in B cell acute
lymphoblastic leukemia. Sci. Transl. Med. 6(224), 224ra225
(2014).
16 Kochenderfer JN, Dudley ME, Feldman SA et al. B-cell
depletion and remissions of malignancy along with
cytokine-associated toxicity in a clinical trial of anti-CD19
chimeric-antigen-receptor-transduced T cells. Blood 119(12),
2709–2720 (2012).
17 Kochenderfer JN, Wilson WH, Janik JE et al. Eradication
of B-lineage cells and regression of lymphoma in a patient
future science group
CAR T-cell therapy: toxicity & the relevance of preclinical models  Review
treated with autologous T cells genetically engineered to
recognize CD19. Blood 116(20), 4099–4102 (2010).
18 Cheadle EJ, Gilham DE, Hawkins RE. The combination of
cyclophosphamide and human T cells genetically engineered
to target CD19 can eradicate established B-cell lymphoma.
Br. J. Haematol. 142(1), 65–68 (2008).
19 Hillerdal V, Ramachandran M, Leja J, Essand M. Systemic
treatment with CAR-engineered T cells against PSCA delays
subcutaneous tumor growth and prolongs survival of mice.
BMC Cancer 14, 30 (2014).
20 Parente-Pereira AC, Burnet J, Ellison D et al. Trafficking
of CAR-engineered human T cells following regional or
systemic adoptive transfer in SCID beige mice. J. Clin.
Immunol. 31(4), 710–718 (2011).
21 Zhao Y, Moon E, Carpenito C et al. Multiple injections
of electroporated autologous T cells expressing a chimeric
antigen receptor mediate regression of human disseminated
tumor. Cancer Res. 70(22), 9053–9061 (2010).
22 Alcantar-Orozco EM, Gornall H, Baldan V, Hawkins RE,
Gilham DE. Potential limitations of the NSG humanized
mouse as a model system to optimize engineered human
T cell therapy for cancer. Hum. Gene Ther. Methods 24(5),
310–320 (2013).
23 Hannon M, Lechanteur C, Lucas S et al. Infusion of
clinical-grade enriched regulatory T cells delays experimental
xenogeneic graft-versus-host disease. Transfusion 54(2),
353–363 (2014).
24 Zhou Q, Facciponte J, Jin M, Shen Q, Lin Q. Humanized
NOD-SCID IL2rg-/- mice as a preclinical model for cancer
research and its potential use for individualized cancer
therapies. Cancer Lett. 344(1), 13–19 (2014).
25 Davila ML, Riviere I, Wang X et al. Efficacy and toxicity
management of 19–28z CAR T cell therapy in B cell acute
lymphoblastic leukemia. Sci. Transl. Med. 6(224), 224ra225
(2014).
26 Wilkinson RW, Ross EL, Ellison D, Zimmermann W, Snary
D, Mather SJ. Evaluation of a Transgenic Mouse Model
for Anti-Human CEA Radioimmunotherapeutics. J. Nucl.
Med. 43(10), 1368–1376 (2002).
27 Bugelski PJ, Achuthanandam R, Capocasale RJ, Treacy G,
Bouman-Thio E. Monoclonal antibody-induced cytokine-
release syndrome. Expert Rev. Clin. Immunol. 5(5), 499–521
(2009).
28 Howard SC, Jones DP, Pui CH. The tumor lysis syndrome.
N. Engl. J. Med. 364(19), 1844–1854 (2011).
29 Cheadle EJ, Hawkins RE, Batha H, O’Neill AL, Dovedi SJ,
Gilham DE. Natural expression of the CD19 antigen impacts
the long-term engraftment but not antitumor activity of
CD19-specific engineered T cells. J. Immunol. 184(4),
1885–1896 (2010).
30 Brentjens RJ, Davila ML, Riviere I et al. CD19-targeted
T cells rapidly induce molecular remissions in adults with
chemotherapy-refractory acute lymphoblastic leukemia. Sci.
Transl. Med. 5(177), 177ra38 (2013).
31 Morgan RA, Yang JC, Kitano M, Dudley ME, Laurencot
CM, Rosenberg SA. Case report of a serious adverse event
following the administration of T cells transduced with
www.futuremedicine.com 495
Review  Kalaitsidou, Kueberuwa, Schütt & Gilham

a chimeric antigen receptor recognizing ERBB2. Mol. 46 Muruganandan S, Sinal CJ. Mice as clinically relevant
Ther. 18(4), 843–851 (2010). models for the study of cytochrome P450-dependent
• Illustrates the catastrophic effects associated with CAR metabolism. Clin. Pharmacol. Ther. 83(6), 818–828 (2008).
recognizing target on nontumor cells. 47 Davies DM, Foster J, Van Der Stegen SJ et al. Flexible
targeting of ErbB dimers that drive tumorigenesis by using
32 Coffman KT, Hu M, Carles-Kinch K et al. Differential
genetically engineered T cells. Mol. Med. 18, 565–576
EphA2 epitope display on normal versus malignant cells.
(2012).
Cancer Res. 63(22), 7907–7912 (2003).
48 Van Der Stegen SJC, Davies DM, Wilkie S et al. Preclinical
33 Sampson JH, Heimberger AB, Archer GE et al. Immunologic
In vivo Modeling of cytokine release syndrome induced by
escape after prolonged progression-free survival with
ErbB-retargeted human T cells: identifying a window of
epidermal growth factor receptor variant III peptide
therapeutic opportunity? J. Immunol. 191(9), 4589–4598
vaccination in patients with newly diagnosed glioblastoma.
(2013).
J. Clin. Oncol. 28(31), 4722–4729 (2010).
34 Gan HK, Burgess AW, Clayton AHA, Scott AM. Targeting • One of the first studies to document the impact of route
of a conformationally exposed, tumor-specific epitope of of administration, tumor load and on-target off-tumor
EGFR as a strategy for cancer therapy. Cancer Res. 72(12), toxicity driven by ligand-specific CAR.
2924–2930 (2012). 49 Van Schalkwyk MC, Papa SE, Jeannon JP, Guerrero Urbano
35 Maher J, Wilkie S. CAR mechanics: driving T cells into the T, Spicer JF, Maher J. Design of a phase I clinical trial to
MUC of cancer. Cancer Res. 69(11), 4559–4562 (2009). evaluate intratumoral delivery of ErbB-targeted chimeric
antigen receptor T cells in locally advanced or recurrent
36 Gnjatic S, Nishikawa H, Jungbluth AA et al. NY-ESO-1:
head and neck cancer. Hum. Gene Ther. Clin. Dev. 24(3),
Review of an immunogenic tumor antigen. Adv. Cancer
134–142 (2013).
Res.95,1–30 (2006).
50 Kakarla S, Chow KKH, Mata M et al. Antitumor effects
37 Sang M, Lian Y, Zhou X, Shan B. MAGE-A family:
of chimeric receptor engineered human T cells directed to
attractive targets for cancer immunotherapy. Vaccine 29(47),
tumor stroma. Mol. Ther. 21(8), 1611–1620 (2013).
8496–8500 (2011).
51 Schuberth PC, Hagedorn C, Jensen SM et al. Treatment
38 Tsai J-R, Chong I-W, Chen Y-H et al. Differential expression
of malignant pleural mesothelioma by fibroblast activation
profile of MAGE family in non-small-cell lung cancer. Lung
protein-specific re-directed T cells. J. Transl. Med. 11, 187
Cancer 56(2), 185–192 (2007).
(2013).
39 Kochenderfer JN, Rosenberg SA. Treating B-cell cancer with
52 Tran E, Chinnasamy D, Yu Z et al. Immune targeting
T cells expressing anti-CD19 chimeric antigen receptors.
of fibroblast activation protein triggers recognition of
Nat. Rev. Clin. Oncol. 10(5), 267–276 (2013).
multipotent bone marrow stromal cells and cachexia. J. Exp.
40 Till BG, Jensen MC, Wang J et al. CD20-specific adoptive Med. 210(6), 1125–1135 (2013).
immunotherapy for lymphoma using a chimeric antigen
•• Report documenting toxicity resulting from the depletion
receptor with both CD28 and 4–1BB domains. pilot clinical
of a previously unknown population of osteogenic cells
trial results. Blood 119(17), 3940–3950 (2012).
directed by a third-generation FAP-specific CAR combined
41 Cheadle EJ, Sheard V, Rothwell DG et al. Differential role of
with preconditioning and systemic IL-2.
Th1 and Th2 cytokines in autotoxicity driven by CD19-
specific second-generation chimeric antigen receptor T cells 53 Wang LC, Lo A, Scholler J et al. Targeting fibroblast
in a mouse model. J. Immunol. 192(8), 3654–3665 (2014). activation protein in tumor stroma with chimeric antigen
receptor T cells can inhibit tumor growth and augment host
•• First report of acute and chronic toxicity mediated by immunity without severe toxicity. Cancer Immunol. Res. 2(2),
CD19 second-generation CAR involving the dysregulated 154–166 (2014).
response to Th1 and Th2 biased cytokines in a specific
•• A study showing no evidence of toxicity with a FAP-specific
mouse model.
second-generation CAR suggesting that factors enhancing
42 Hacein-Bey-Abina S, Von Kalle C, Schmidt M et al. LMO2- CAR T-cell engraftment and functional activity may
associated clonal T cell proliferation in two patients after
underlie potential on-target off-tumor toxicities.
gene therapy for SCID-X1. Science 302(5644), 415–419
(2003). 54 Marin V, Cribioli E, Philip B et al. Comparison of
different suicide-gene strategies for the safety improvement
43 Scholler J, Brady TL, Binder-Scholl G et al. Decade-long
of genetically manipulated T cells. Hum. Gene Ther.
safety and function of retroviral-modified chimeric antigen
Methods 23(6), 376–386 (2012).
receptor T cells. Sci. Transl. Med. 4(132), 132ra153 (2012).
55 Di Stasi A, Tey S-K, Dotti G et al. Inducible apoptosis
44 Wang LX, Westwood JA, Moeller M et al. Tumor ablation
as a safety switch for adoptive cell therapy. N. Engl.
by gene-modified T cells in the absence of autoimmunity.
J. Med. 365(18), 1673–1683 (2011).
Cancer Res 70(23), 9591–9598 (2010).
56 Alonso-Camino V, Sanchez-Martin D, Compte M et al.
45 John LB, Devaud C, Duong CP et al. Anti-PD-1 antibody
CARbodies: human antibodies against cell surface tumor
therapy potently enhances the eradication of established
antigens selected from repertoires displayed on T cell
tumors by gene-modified T cells. Clin. Cancer Res. 19(20),
chimeric antigen receptors. Mol. Ther. Nucleic Acids 2, e93
5636–5646 (2013).
(2013).

496 Immunotherapy (2015) 7(5) future science group


57 Lipowska-Bhalla G, Gilham DE, Hawkins RE, Rothwell
DG. Isolation of tumor antigen-specific single-chain variable
fragments using a chimeric antigen receptor bicistronic
retroviral vector in a Mammalian screening protocol. Hum.
Gene Ther. Methods 24(6), 381–391 (2013).
58 Kloss CC, Condomines M, Cartellieri M, Bachmann
M, Sadelain M. Combinatorial antigen recognition with
balanced signaling promotes selective tumor eradication by
engineered T cells. Nat. Biotechnol. 31(1), 71–75 (2013).
59 Grada Z, Hegde M, Byrd T et al. TanCAR: a novel bispecific
chimeric antigen receptor for cancer immunotherapy. Mol.
Ther. Nucleic Acids 2, e105 (2013).
60 Fedorov VD, Themeli M, Sadelain M. PD-1-and CTLA-
4-based inhibitory chimeric antigen receptors (iCARs)
divert off-target immunotherapy Responses. Sci. Transl.
Med. 5(215), 215ra172 (2013).
future science group
CAR T-cell therapy: toxicity & the relevance of preclinical models  Review
61 Barrett DM, Zhao Y, Liu X et al. Treatment of Advanced
leukemia in mice with mRNA engineered T cells. Hum.
Gene Ther. 22(12), 1575–1586 (2011).
62 Singh N, Liu X, Hulitt J et al. Nature of tumor control by
permanently and transiently-modified GD2 chimeric antigen
receptor T cells in xenograft models of neuroblastoma.
Cancer Immunol. Res. 2(11), 1059–1070 (2014).
63 Gattinoni L, Klebanoff CA, Restifo NP. Paths to stemness:
building the ultimate antitumour T cell. Nat. Rev.
Cancer 12(10), 671–684 (2012).
64 Xu Y, Zhang M, Ramos CA et al. Closely related
T-memory stem cells correlate with in vivo expansion of
CAR. CD19-T cells and are preserved by IL-7 and IL-15.
Blood 123(24), 3750–3759 (2014).
www.futuremedicine.com 497