Open access Original research

Combination therapy with T cell

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
engager and PD-­L1 blockade enhances
the antitumor potency of T cells as
predicted by a QSP model
Huilin Ma ‍ ‍ ,1 Hanwen Wang,1 Richard J Sové,1 Jun Wang,2 Craig Giragossian,2
Aleksander S Popel1,3

To cite: Ma H, Wang H, Sové RJ, ABSTRACT BACKGROUND

et al. Combination therapy Background T cells have been recognized as core Colorectal cancer (CRC), especially meta-
with T cell engager and PD-­ effectors for cancer immunotherapy. How to restore the
L1 blockade enhances the
static CRC (mCRC) with proficient mismatch
anti-­tumor ability of suppressed T cells or improve the repair (pMMR) or microsatellite stable
antitumor potency of T cells
as predicted by a QSP model.
lethality of cytotoxic T cells has become the main focus in (MSS) tumors, is one of the leading causes of
immunotherapy. Bispecific antibodies, especially bispecific
Journal for ImmunoTherapy cancer-­associated deaths in USA.1 In recent
of Cancer 2020;8:e001141. T cell engagers (TCEs), have shown their unique ability
to enhance the patient’s immune response to tumors by
years, immune checkpoint inhibitors (ICIs)
doi:10.1136/jitc-2020-001141
stimulating T cell activation and cytokine production in have achieved a durable clinical response in
Accepted 26 July 2020 an MHC-­independent manner. Antibodies targeting the patients with melanoma, non-­small-­cell lung
checkpoint inhibitory molecules such as programmed cell cancer (NSCLC) and other cancer types.2
death protein 1 (PD-1), PD-­ligand 1 (PD-­L1) and cytotoxic However, the results of testing these drugs in
lymphocyte activated antigen 4 are able to restore the mCRC patients with pMMR or MSS tumors
cytotoxic effect of immune suppressed T cells and have also were disappointing.3 Novel therapeutic
shown durable responses in patients with malignancies. agents or combination strategies are being
However, both types have their own limitations in treating tested and accurate biomarker-­guided patient
certain cancers. Preclinical and clinical results have
selection is needed to determine patients who
emphasized the potential of combining these two antibodies
to improve tumor response and patients’ survival. However,
will most likely benefit from the treatment.4 5
the selection and evaluation of combination partners Note that all abbreviations are described in
clinically is a costly endeavor. In addition, despite advances online supplementary table S1.
made in immunotherapy, there are subsets of patients During the past few years, the US Food
who are non-­responders, and reliable biomarkers for and Drug Administration has approved
different immunotherapies are urgently needed to improve the application of anti-­ programmed cell
the ability to prospectively predict patients’ response and death protein (PD-1) inhibitors, including
improve clinical study design. Therefore, mathematical and nivolumab (OPDIVO, Bristol-­Myers Squibb)
computational models are essential to optimize patient and pembrolizumab (KEYTRUDA, Merck)
benefit, and guide combination approaches with lower cost
© Author(s) (or their
and the anti-­ PD-­ligand 1 (L1) inhibitor
and in a faster manner.
employer(s)) 2020. Re-­use atezolizumab (TECENTRIQ, Genentech).
Method In this study, we continued to extend the
permitted under CC BY.
quantitative systems pharmacology (QSP) model we They have shown great potential for targeting
Published by BMJ. PD-1/PD-­ L1 interaction in the treatment
1
developed for a bispecific TCE to explore efficacy of
Department of Biomedical of patients with different types of cancer
combination therapy with an anti-­PD-­L1 monoclonal
Engineering, Johns Hopkins
antibody in patients with colorectal cancer. including melanoma, advanced NSCLC,
University School of Medicine,
Baltimore, Maryland, USA Results Patient-­specific response to TCE monotherapy, renal cell carcinoma (RCC), head and neck
2
Biotherapeutics Discovery anti-­PD-­L1 monotherapy and the combination therapy squamous cell cancer, triple-­negative breast
Research, Boehringer Ingelheim were predicted using this model according to each cancer and microsatellite instability-­ high
Pharmaceuticals Inc, Ridgefield, patient’s individual characteristics. mCRC (MSI-­ H mCRC).2 6 Typically, high
Connecticut, USA Conclusions Individual biomarkers for TCE monotherapy,
3
Department of Oncology and
PD-­L1 expression in cancer cells is associated
anti-­PD-­L1 monotherapy and their combination have
Sidney Kimmel Comprehensive with favorable prognosis and better disease-­
been determined based on the QSP model. Best treatment
Cancer Center, Johns Hopkins
options for specific patients could be suggested based on free survival in response to PD-1/PD-­ L1
Medicine, Baltimore, Maryland,
their own characteristics to improve clinical trial efficiency. inhibitors.7 8 However, CRC cells express less
USA
The model can be further used to assess plausible PD-­L1 and it has been reported by Valentini
Correspondence to combination strategies for different TCEs and immune et al that the expression of PD-­ L1 in MSS
Dr Huilin Ma; ​hma24@​jhmi.​edu checkpoint inhibitors in different types of cancer. CRC is mainly restricted to tumor-­infiltrating

Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141 1

Open access

immune cells.9 This could explain why MSS CRC patients and identified biomarkers for checkpoint inhibitor-­based

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
failed to respond to anti-­PD-1/PD-­L1 therapy. immunotherapy.30 Wang et al proposed a QSP model to
Despite the recent failure of anti-­PD-1/PD-­L1 therapy determined potential predictive biomarkers to improve
in MSS CRC patients, novel bispecific T cell engagers the antitumor response in HER2-­negative breast cancer.31
(TCEs) have been developed and tested in treating CRC In this work, we have extended our QSP model to
both in vitro and in vivo. Gonzalez-­Exposito et al devel- include our previously developed TCE module and newly
oped patient derived CRC organoids to explore the updated anti-­PD-­L1 module to study the efficacy of anti-­
mechanism of T cell bispecific antibody cibisatamab PD-­L1 monotherapy and the combination with TCE
(CEA-­TCB) sensitivity.10 Bacac et al reported the anti- therapy for MSS CRC patients. We studied individual
tumor activity of CEA-­TCB in 110 cell lines and the mode biomarkers for the three therapeutic approaches—TCE
of CEA-­TCB mediated CRC cell lysis in a mouse tumor monotherapy, anti-­PD-­L1 monotherapy and their combi-
model.11 Waaijer et al developed a T cell-­engaging bispe- nation. In silico virtual clinical trials (VCTs) have been
cific antibody (BsAb) to target cell surface A33 antigen conducted to compare the response to different treat-
(huA33-­BsAb), which is expressed in more than 95% of ments for the same cohort of virtual patients (VPs),
human colon cancers.12 An ongoing phase I clinical trial and to discover predictive biomarkers. Our novel QSP
of MGD007 (​Clinicaltrials.​gov, NCT02248805), a gpA33 x model enables development of biomarker-­guided patient
CD3-­BsAb, will provide more valuable information on the selection to improve clinical trial efficiency by providing
clinical safety of this approach.13 the distributions of different biomarkers, recommend
Even though there are over 500 publications listed rational therapeutic regimen and alleviate the rising
in PubMed reporting the preclinical and clinical inves- demand for personalized treatment.
tigations of BsAb,14 treating solid malignancies, which
make up 90% of all cancers, remains extremely chal-
lenging using BsAb because of their poor permeability.15 METHODS
Although the clinical outcome of BsAb is more satisfac- Model structure
tory in hematologic malignancies, some ongoing clinical The QSP model used in this study was based on our
trials have shown promising outcomes in solid tumors.16 previous models developed for NSCLC and TCE.30 32 33
A phase I study (NCT02324257, NCT02650713) led by The model structure includes central (blood), peripheral
Hoffmann-­La Roche has shown the potential of CEA-­TCB (other tissues and organs), tumor and tumor-­draining
(RO6958688; RG7802) monotherapy in treating patients lymph node compartments. The model is composed of
with MSS mCRC, and 45% of the patients showed either several individual but interconnected modules such as
partial response or stable disease. In a combination study cancer cell, T cell, immune checkpoint, antibody pharma-
of CEA-­TCB with atezolizumab, an anti-­PD-­L1 inhibitor, cokinetics (PK), antigen presentation and TCE modules
82% of the patients showed either partial response or (figure 1). The dynamics of major species in each
stable disease, which was an exciting breakthrough for a module have been described in our previous publications
BsAb in a solid tumor.17 including tumor growth, antigen processing and presen-
Although great achievements have been made by the tation, T cell activation and proliferation, T cell distribu-
combination therapy of BsAb with anti-­PD-­L1 inhibitors tion, Treg dynamics, and TCE and immune checkpoint
in solid tumors, possible disadvantages may arise such blockade PK and pharmacodynamics (PD). All governing
as difficulties of determining the source of side effects, equations for the immune checkpoint and TCE modules
drug–drug interactions, cumulative side effects and have been explained in detail in the online supplemen-
higher cost.18 19 In order to avoid possible side effects tary information provided by Jafarnejad et al30 and Ma et
and risks associated with combination therapy, it is al,33 respectively. The modular design of the model makes
important to prospectively determine whether individual it readily extensible to other therapeutic agents and their
patients will derive additional benefit from combination corresponding PK/PD or other newly discovered physio-
therapy.20 21 Patients need to be differentiated and given logical processes. 73 ordinary differential equations and
the most appropriate treatment options to improve the 105 algebraic equations were used to model all biological
therapeutic outcome. The establishment of predictive processes involved in the model. In this work, the mono-
biomarkers is, therefore, important to maximize thera- therapy of atezolizumab (MPDL3280A, RO5541267,
peutic benefit and guide selection of the best therapeutic TECENTRIQ) and the combination therapy with atezoli-
approach for oncologists.8 22 zumab and cibisatamab (RO6958688, RG7802) were
Previous studies have demonstrated the performance studied and compared. PK parameters of cibisatamab
and ability of quantitative systems pharmacology (QSP) have been reported and PK parameters of atezolizumab
modeling in determining predictive biomarkers.23–28 were fitted to experimental data. Observed and simu-
Norton et al developed a multiscale agent-­based model lated serum concentrations of atezolizumab following
of the tumor immune microenvironment, providing an intravenous dose of 1, 3, 10, 15 mg/kg and 1200 mg
information for personalized treatment for individual are provided in the supplement (online supplementary
patients.29 Jafarnejad et al built a QSP model to repre- figure S1). Dynamics of cibisatamab have been calibrated
sent the antitumor immune response in human NSCLC and described in our previous publication.33 This model

2 Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141

 Open access

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
Figure 1 Diagram of the main cellular and molecular interactions implemented in the model (modified from30 33). APC, antigen-­
presenting cell; IL2, interleukin-2; PD-­L1, programmed cell death ligand 1; TCE, T cell engager.

can be applied to most TCEs and ICIs with minor modi- supplementary information as well as governing equa-
fications. All simulations and sensitivity analyzes were tions, species, parameters, reactions, rules, events and
performed using the SimBiology platform in MATLAB descriptions related to the newly added mechanism.
R2018b (MathWorks, Natick, Massachusetts, USA).
Parameter sensitivity analysis
Functional expression of PD-L1 on cancer and immune cells Parameter sensitivity analysis (PSA) was performed to
In a previous model, Jafarnejad et al incorporated the assess the sensitivity of the QSP model to a set of parame-
dynamics of immune checkpoint blockade and demon- ters. Latin hypercube sampling (LHS) was used to assign
strated the general applicability of that module to any the values for this set of parameters with uniform trans-
anti-­PD-1/PD-­L1 inhibitor. The model was then used to formation such as tumor volume, density of Teff and
study anti-­ PD-1 therapy in NSCLC using nivolumab.30 Treg, Teff/Treg cell ratio in tumor compartment, and
A numberof PD-1, PD-­L1, PD-­L2 and other parameters CD8 +T cell clonality in blood. Partial rank correlation
involved have been carefully chosen or fitted to experi- coefficient (PRCC) analysis was performed to identify the
mental measurements. Baseline parameters were chosen most influential factors from the simulation results and
to fit a Hill function to in vitro dose-­response measure- was implemented by using the MATLAB Global Optimi-
ments of IFNγ by Jafarnejad et al and thus can be applied zation Toolbox.
for atezolizumab. However, our previous model only
considered the expression of PD-­L1 on cancer cells. The Statistical analysis
expression of PD-­L1 on antigen-­presenting cells (APCs) Statistical analysis was performed for VPs’ subcohorts.
is also an important factor leading to tumor immune Wilcoxon test was used to analyze the differences between
evasion and has been reported to have a significant effect responders and non-­responders (NRs) under the atezoli-
on the outcome of immunotherapy.34 35 zumab/cibisatamab monotherapy and combination
Upregulation of PD-­L1 on cancer cells is believed to therapy using the ggpubr package embedded in RStudio
be the major mechanism for tumor immune evasion.36 V.1.2. The impact of sensitive parameters on the overall
However, it has been reported that dendritic cells (DCs), response rate (ORR) was also studied with 95% Agresti-­
a major APCs, express cell-­surface PD-­L1 on activation Coull CI.
by toll-­like receptor ligands.37 DCs are responsible for
initiating rapid proliferation of antitumor CD8 +T cells;
however, PD-­L1 signaling induced by DCs restricts the RESULTS
proliferative capacity of CD8 +T cells during activation, A virtual cohort of 2000 patients was created by LHS
and a previous study demonstrated that DCs lacking PD-­L1 method. Each VP was generated with a random sample
expression resulted in significantly increased numbers of of parameter values based on the list of parameters in the
antigen-­specific CD8+T cells.38 We, therefore, extended PSA. The baseline number and ranges of all parameters
the current model with PD-­L1 expression in APCs, which listed in the PSA were based on clinical and experimental
limits the proliferation of Teff in TdLN compartment. evidence (online supplementary tables S2 and S3), the
Blockade of PD-­L1 signaling during the priming phase in baseline values are based on experimental measure-
the TdLN compartment will restore the normal prolifer- ments.8 30 33 39 40 Note that the ranges for parameters were
ative capacity of Teff. CRC express less PD-­L1 than some chosen to be physiologically reasonable if experimental
other types of cancer, but APCs express similar levels of measurements are unavailable. To avoid generating
PD-­L1 among different cancer types. Detailed parameters implausible patients due to uncertainty in parameter
used for ICIs expression level are provided in the online ranges, several physiological parameters were used to

Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141 3

Open access

screen VPs such as tumor diameter, T cell density in the table S4). To more closely mimic real patient populations,

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
blood, activated T cell density in the tumor and Teff to we applied these simulation results to the actual clinical
Treg ratio. A lower and upper bound of these parameters trial in NCT02324257 (31 patients, cibisatamab mono-
have been set based on clinical measurements.33 VPs who therapy) and NCT02650713 (25 patients, combination
did not develop tumors or with implausible parameter therapy) by randomly sampling 31 VPs 10 000 times in
values that were outside the normal physiological range cibisatamab monotherapy and 25 VPs 10 000 times in
were regarded as non-­patients and excluded from the combination therapy. Although there was no atezoli-
virtual trial. Plausible VPs were used for estimating ORR. zumab monotherapy in these two trials, we sampled 31
VPs 10 000 times in our simulated atezolizumab mono-
In silico VCT outcomes therapy. This allowed us to obtain a 95% percentile boot-
The ORR of atezolizumab monotherapy, cibisatamab strap confidence interval (95% CI) of the ORR for the
monotherapy and combination therapy were investigated three treatments (online supplementary table S4, figure
by simulating plausible VPs in each trial. In accordance S3) and the ORR in each interval (online supplementary
with NCT02324257 and NCT02650713 trials, MSS CRC table S5). There has been several reports demonstrating
VPs in atezolizumab monotherapy were treated with little activity of ICIs such as atezolizumab in most MSS
atezolizumab 1200 mg Q3W. The same VPs were used in CRC patients,41 42 which was reflected in the lower limit
cibisatamab monotherapy and combination therapy in of our estimated 95% CI for atezolizumab monotherapy.
order to compare their responses to different therapies.
They were treated with cibisatamab 60 mg QW for cibi- Statistical analysis for NRs and responders to determine
satamab monotherapy and cibisatamab 60 mg QW with potential biomarkers
atezolizumab 1200 mg Q3W for combination therapy. PRCC was used for performing global uncertainty and
Based on the screening rules, there were 1312, 1325 and sensitivity analysis to measure the degree of association
1299 VPs left in atezolizumab monotherapy, cibisatamab between parameters and the tumor volume. In atezoli-
monotherapy and combination therapy, respectively. The zumab monotherapy and combination therapy, tumor
simulated time-­dependent percent tumor size changes growth rate and initial tumor diameter were significantly
are shown in online supplementary figure S2 (spider positively correlated to the tumor volume. Tumor muta-
plot) following RECIST V.1.1. After 400 days, most tional burden (TMB) defined as the number of clones
patients who had PR/CR and SD reached convergence, of T cells that are activated30 33 and PD-­L1 expression
where their tumor size no longer changed. Although the in cancer cells were negatively correlated to the tumor
tumor size of some patients was still changing, the tumor volume (figure 2A). In addition, CEA expression in cancer
size was getting smaller and did not affect the calculation cells was also negatively correlated to the tumor volume
of ORR. Then we calculated ORR at this time point (400 in combination therapy (figure 2B). Waterfall plots were
days). Among the patients in atezolizumab monotherapy, used to present each individual patient's response to
107/1312 had PR/CR (8.2%), 91/1312 had SD (6.9%) atezolizumab (online supplementary figure S4) or combi-
and 1114/1750 had PD (84.9%). In cibisatamab mono- nation therapy (figure 3). Obviously, higher TMB and
therapy, 69/1325 had PR/CR (5.2%), 107/1325 had SD PD-­L1 expression in cancer cells corresponded to smaller
(8.1%) and 1149/1325 had PD (86.7%). In combination tumor volume based on RECIST criteria (figure 3).
therapy, 145/1299 had PR/CR (11.2%), 114/1299 had Distribution of parameters of interest between R and
SD (8.8%) and 1040/1299 had PD (80.0%). The ORR NR are shown in figure 4. In atezolizumab monotherapy,
of cibisatamab monotherapy (5.2%) and combination TMB, PD-­L1 on both cancer cells and APCs, Teff density
therapy (11.2%) showed agreement with NCT02324257 as well as Teff/Treg ratio in tumor were significantly
and NCT02650713 trials (6% in cibisatamab monotherapy higher in responders, whereas T cell PD-1 expression
and 12% in combination therapy) (online supplementary and atezolizumab cross-­arm binding efficiency (χ) were

Figure 2 The partial rank correlation coefficient, PRCC, for individual parameters. (A) Atezolizumab monotherapy. (B)
Combination therapy. APC, antigen-­presenting cell; PD1, programmed cell death protein 1; PD-­L1, PD-­ligand 1; PRCC, Partial
rank correlation coefficient; TCE, T cell engager.

4 Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141

 Open access

will not be repeated here. Similar conclusions were

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
observed for combination therapy. Most of the parame-
ters that showed significant differences between R and
NR in monotherapies also had significant differences
between patients in combination therapy. However, in
combination therapy, PD-­L1 expression in APCs was not
significantly different between R and NR, indicating the
addition of combination therapy with cibisatamab may
compensate for the loss of activated CD8 +T cells due to
the PD-­L1 on APCs (figure 4B).

Comparison of Atezolizumab Monotherapy/Cibisatamab

monotherapy and combination therapy
In clinical trials, it is difficult to know in advance which
patients may benefit from combination therapy, particu-
larly when only a small portion of patients are likely to
derive benefits from such treatment. Prospectively being
able predict patient response to combination thera-
Figure 3 Waterfall plots of combination therapy while pies, especially for those patients who fail to respond or
varying. (A) TMB; (B) PD-1 expression; (C) PD-­L1 expression respond poorly to monotherapy, can lead to enhanced
in cancer cells; (D) PD-­L1 expression in APCs; (E) CEA clinical trial design. VCTs make it easier to determine
expression; (F) CD3 expression in teff cells. APCs, antigen-­ different responses to different therapies using the exact
presenting cells; CEA, carcinoembryonic antigen; PD1, same cohort of VPs.
programmed cell death protein 1; PD-­L1, PD-­ligand 1; TMB, Here, we focused on those VPs who had PD in atezoli-
tumor mutational burden.
zumab or cibisatamab monotherapy but had SD (PD-­SD)
or PR/CR (PD-­PR/CR) in combination therapy or had
not significantly different between responders and NRs SD in atezolizumab or cibisatamab monotherapy but PR/
(figure 4A). Note that the tumor growth rate, although CR (SD-­PR/CR) in combination therapy. Other patients
significantly positively correlated with tumor volume, who had PD or SD in both monotherapy and combina-
cannot be used as an indicator to predict patients’ tion therapy without any improvement (PD-­PD, SD-­SD)
response. The results of cibisatamab monotherapy were were grouped as reference. Then, we compared the
completely consistent with our previous reports and differences between patients in these groups (figure 5).

Figure 4 Distributions of potential biomarkers in NR and R in (A). Atezolizumab monotherapy (B). Combination therapy. (i)
Tumor growth rate; (ii) TMB; (iii) ieff density in tumor; (iv) Teff/Treg ratio in tumor; (v) PD-­L1 expression in cancer cells; (vi). PD-­
L1 expression in APCs; (vii) PD-1 expression in teff; (viii) Cross-­arm binding efficiency χ of atezolizumab; (ix) CEA expression in
cancer cells; (x) Cross-­arm binding efficiency λ of cibisatamab; (xi) CD3 expression in teff; (xii) CD3-­cibisatamab binding affinity.
APCs, antigen-­presenting cells; CEA, carcinoembryonic antigen; NS, not significant; PD1, programmed cell death protein 1; PD-­
L1, PD-­ligand 1; TMB, tumor mutational burden. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001, **** P ≤ 0.0001.

Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141 5

Open access

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
Figure 5 Distributions of potential biomarkers in VPS subgroups (PD-­PD, PD-­SD, PD-­PR/CR, SD-­SD, SD-­PR/CR) receiving
(A). Atezolizumab monotherapy versus combination therapy. (i) TMB; (ii). teff density in tumor; (iii). Teff/Treg ratio in tumor;
(iv). CEA expression in cancer cells; (v). PD-­L1 expression in cancer cells; (vi). tumor-­specific antigen binding affinity. (B)
Cibisatamab monotherapy versus combination therapy. (i). TMB; (ii) teff density in tumor; (iii) Teff/Treg ratio in tumor; (iv)
PD-­L1 expression in cancer cells; (v) CEA expression in cancer cells; (vi). Tumor-­specific antigen binding affinity. CEA,
carcinoembryonic antigen; NS, not significant; PD1, programmed cell death protein 1; PD-­L1, PD-­ligand 1; TMB, tumor
mutational burden; VPS, virtual patients. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001, **** P ≤ 0.0001.

When comparing atezolizumab monotherapy and Biomarker-guided patient selection

combination therapy, patients who benefited from combi- In another perspective, if we can prospectively predict
nation therapy including PD-­PR/CR, PD-­SD and SD-­PR/ which treatment is most likely to be effective for patients
CR had significantly higher TMB, Teff density in tumor prior to the start of clinical trials, we would be able to
and CEA expression in cancer cells, which demonstrated select subjects for inclusion in randomized clinical trials
their important role in predicting treatment outcomes and potentially reduce the number of subjects to recruit
for combination therapy. Moreover, patients in PD-­PR/ to improve clinical trial efficiency. Based on our virtual
CR and PD-­SD groups also had higher Teff/Treg ratio trials’ results, we divided all patients into five groups
and cibisatamab cross-­arm binding efficiency (λ) than based on their responses to three treatments: responder
the PD-­PD group. Parameters such as TMB, Teff density to atezolizumab monotherapy only (ROA), responder
and Teff/Treg ratio in tumor and CEA expression in to cibisatamab monotherapy only (ROC), responder to
cancer cells of patients in the PD-­PR/CR group were also both monotherapies (ROB), responder to combination
significantly higher than those in the PD-­SD group. The therapy only (ROCMB) and NR to any therapies.
distribution of these parameters can not only estimate We computed the distribution of several potential
whether patients would benefit from receiving combi- biomarkers in these five groups, respectively (figure 6). In
nation therapy, but also to determine how much benefit terms of TMB, Teff density and Teff/Treg ratio in tumor,
they might receive. In addition, patients in SD-­PR/CR their distributions in ROB were highest compared with
group had higher PD-­L1 expression in cancer cells than any other groups. This revealed why this group of patients
the other four groups, and these patients had stable was able to respond to any treatments (figure 6A–C).
disease after receiving atezolizumab monotherapy. With Group ROA had highest PD-­ L1 expression in cancer
the treatment of combination therapy, they were more cells around 1E4–1E5 sites/cell and the range was rela-
likely to benefit and showed better treatment outcome tively narrower than other groups (figure 6D). Similarly,
(PR/CR) (figure 5A). group ROC had high values of CEA expression in cancer
For cibisatamab monotherapy and combination cells and cibisatamab cross-­ arm binding efficiency (λ)
therapy, TMB and Teff density in tumor once again (figure 6E,F). Group NR had the broadest distribution of
proved their ability to predict therapeutic effect and all parameters, and most patients in this group had rela-
potential as predictive biomarkers. PD-­L1 expression in tively low parameter values. Note that patients in group
cancer cells was higher in those three groups of patients ROCMB had ideal PD-­ L1 and CEA expression level,
with improved therapeutic effects. CEA expression was however, their TMB, Teff density and Teff/Treg ratio
also higher in patients in group SD-­PR/CR than SD-­SD were lower than patients in group ROA, ROC and ROB
and their condition was improved after receiving combi- and they were the best candidates for receiving combina-
nation therapy (figure 5B). tion therapy. Ideally, if these parameters were known in

6 Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141

 Open access

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
Figure 6 Distributions of potential biomarkers in ROA, ROB, ROC, ROCMB and NR. (A) TMB; (B) teff density in tumor; (C) Teff/
Treg ratio in tumor; (D) PD-­L1 expression in cancer cells; (E) CEA expression in cancer cells; (F) Cross-­arm binding efficiency λ
of cibisatamab. CEA, carcinoembryonic antigen; NR, non-­responder; NS, not significant; PD-­L1, programmed cell death ligand
1; ROA, responder to atezolizumab monotherapy; ROB, responder to both monotherapies; ROC, responder to cibisatamab
monotherapy; ROCMB, responder to combination therapy; TMB, tumor mutational burden. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001,
**** P ≤ 0.0001.

advance for a real patient, we would be able to predict patients were assigned into a group from the first patient,
which group the patient might belong to, based on our and the ORR of each group was calculated separately
predicted parameter distribution, and then determine (figure 7B). All parameters were normalized to between
the most suitable therapy. 0 and 1 according to their range for direct compar-
ison of the influence on the ORR. Clearly, the ORR of
Predictive performance of different biomarkers patients was very significantly affected by TMB. The
It is unlikely that a single biomarker will be sufficient ORR of patients with low TMB was close to 0% until the
to predict clinical outcomes in response to immune-­ normalized parameter range was greater than 0.8 in x
targeted therapy. The effects of multiple factors need axis, which roughly corresponds to TMB greater than 76.
to be comprehensively considered to accurately predict
outcomes. Nevertheless, a single biomarker can be used
to predict the ORR, but the predictive performance of
biomarkers is different. To determine the most predictive
biomarkers, we computed and compared the predictive
ability of each parameter and how its value affected the
ORR of subcohorts of all VPs.
We investigated the performance of preidentified
biomarkers. The results were plotted as a receiver oper-
ating characteristic (ROC) curve. TMB, Teff density
and Teff/Treg ratio in tumor had high area under the
curve (AUC) in all treatments, which indicated their
great potential as good predictive biomarkers (figure 7).
PD-­L1 and CEA expression in cancer cells had high
AUCs for atezolizumab monotherapy and cibisatamab
monotherapy, respectively. They also had intermediate
AUCs in combination therapy, however, higher AUC of Figure 7 (A) ROC analysis of potential predictive
PD-­L1 expression demonstrated its ability to be a better biomarkers in (i) Atezolizumab monotherapy; (ii) Cibisatamab
monotherapy; (iii) combination therapy. (B) Preditive ability
biomarker than CEA expression in combination therapy
of potential biomarkers in (i) Atezolizumab monotherapy;
(figure 7A). (ii) Cibisatamab monotherapy; (iii) combination therapy.
Then we examined the relationship between the ORR ORR, overall response rate; PD-1, programmed cell death
and each parameter by calculating the ORR of patient protein 1; PD-­L1, PD-­ligand 1; ROC, receiver operating
subcohorts. First, we sorted all patients in ascending characteristic; TCE, T cell engager; TMB, tumor mutational
order according to a certain parameter, then every 20 burden.

Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141 7

Open access

When x was close to 1, the ORR of patient subcohort was However, this setting caused an issue when calcu-

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
about 100% and their TMB was greater than 200. Simi- lating the distribution of PD-­L2 in responders and non-­
larly, Teff density and Teff/Treg ratio were both positively responders, responders showed significantly higher
correlated to patients’ ORR, and patients with ideal Teff PD-­L2 both in cancer and APCs (online supplementary
infiltration in tumor may have higher ORR. Although the figure S5), which was contrary to our expectations. Since
ORR increased to 20% with the increase of PD-­L1 expres- atezolizumab did not block the interactions between PD-1
sion in patients, the effect of it on ORR was less compared and PD-­L2, less PD-­L2 should be more ideal due to their
with the TMB and Teff infiltration. Many patients with inhibitory effect. However, due to the ratio we assigned
high PD-­L1 expression were still unable to respond, most for PD-­L1/PD-­L2 expression, patients with higher PD-­L1
likely due to their relatively low TMB and Teff infiltration. expression were highly likely to have higher PD-­L2 than
We repeated the analysis for all three treatments; TMB patients with lower PD-­L1 expression even though the
showed the best potential to predict the outcomes of all ratio was randomly assigned for each patient between 0
treatments followed by Teff density and Teff/Treg ratio in and 0.07, which generated misleading results. To solve
tumor. These results are consistent with the ROC curves. this issue, we introduced an additional parameter δ into
According to this analysis, we were able to determine the the Hill function of ICIs (equations 10 and 11 in online
best predictive biomarkers and therefore estimate the supplementary information). By varying the value of
ORR interval that these parameter levels may correspond δ between 0 and 1, we were able to explore how PD-­L2
to. expression affects the Hill equation and then tumor
growth. We repeated the atezolizumab monotherapy
Function and impact of PD-L2 again with δ and computed its distribution in responders
PD-­L2 has been widely reported as a second ligand for and non-­responders. The ORR was 9.3%, which was a
PD-1 and inhibitor of T cell activation. Effects of PD-­L2 significant improvement compared with previous one
were also included in the current QSP model. The (8.2%). In addition, we computed the distribution of δ in
dynamics of PD-­ L2 has been reported and described responders and non-­responders and found no significant
in the online supplementary information section 1.1.30 difference between them (p>0.05) (online supplementary
However, its role is not completely understood and figure S6). To conclude, because of the low expression of
expression of PD-­L2 on cancer cells and APCs showed PD-­L2 on cancer cells and APCs, they had relatively little
some correlation with PD-­L1 expression. Taube et al have impact on tumor growth and patient response.
reported that tumor and APCs can both express PD-­L2 in
patients with advanced, treatment-­refractory solid tumors
including NSCLC, melanoma, kidney, castration-­resistant
prostate cancer) and CRC.39 They assessed PD-­L2 expres- DISCUSSION
sion in 38 tumor specimens and found that PD-­L2 was According to recent reports, the majority of CRC patients
less frequently expressed than PD-­ L1 and was almost have MSS tumors, which accounts for 80%–85% of all
geographically associated with PD-­L1 expression. More CRC patients.44 Only CRC patients with dMMR/MSI-­H
importantly, only 1/38 tumor specimen expressed PD-­L2 CRC showed response to immunotherapy due to their
only without PD-­L1 but 14/38 specimens expressed PD-­L1 highly immunogenic nature. Previous studies revealed
without PD-­L2 and 7/38 specimens expressed both PD-­L1 that most MSS tumors were ‘cold’ tumors with much less
and PD-­L2, the remaining 16/38 specimens were PD-­L1 PD-­L1 expression in cancer cells and low TMB compared
and PD-­L2 negative. This indicated that PD-­L2 is unlikely with melanoma, NSCLC and RCC.8 39 Thus, targeting the
to be expressed alone without PD-­L1 and the expression PD-1/PD-­L1 axis was ineffective in treating MSS CRC.
level of PD-­L2 was also much lower than PD-­L1 based on Nevertheless, other studies revealed a small subset of MSS
Cheng et al.43 Thus, in the model, we assumed that PD-­L2 CRC patients who may still benefit from anti-­PD-1/PD-­L1
expression was correlated with PD-­L1 and the amount of antibodies.45 Therefore, the identification of predictive
PD-­L2 cannot exceed a certain threshold by defining a biomarkers for MSS CRC patients is essential to improve
ratio between PD-­L2 expression and PD-­L1. Based on the patient outcome. Recent clinical trials are looking at novel
aforementioned report, we set the lower limit of this ratio ICIs, combination of immunotherapeutic agents and
(r_PD-­L2) to 0, meaning that PD-­L1 can be expressed better patient selection for immunotherapy treatment to
alone and set the upper limit of this ratio (r_PD-­ L2) increase response of MSS CRC patients. However, there
to 0.07 based on the measurement of PD-­L1/PD-­L2 in is a combinatorial explosion of drug candidates and ther-
mature DCs by Cheng et al.43 When generating VPs, each apies that make clinical assessment of plausible options
patient was assigned a certain amount of PD-­L1 expres- highly expensive and less feasible. Implementations of
sion and a r_PD-­L2 between 0 and 0.07. This ensured that QSP models have become an alternative method to study
all patients would have a realistic concurrent PD-­L1 and different drug combinations and their efficacy. Our QSP
PD-­L2 expression and avoid generating a large number model incorporates dynamics of ICIs and TCEs that can
of implausible patient population with unrealistic PD-­L1/ be applied to any anti-­PD-1/PD-­L1 blockades and TCEs,
PD-­L2 expression such as PD-­L2 alone or more PD-­L2 which can be used as a tool to study drug candidates and
than PD-­L1. combination strategies in silico.

8 Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141

 Open access

Our model has successfully conducted in silico VCTs it is essential to implement novel biomarker-­ guided

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
for atezolizumab monotherapy and combination therapy. patient selection to improve the overall efficiency of clin-
The predicted ORR showed consistency with clinical ical trials design. Ideally, if all these potential biomarkers
trial results. Patients receiving monotherapy with PD-1 can be measured, the best treatment can be identified
or PD-­L1 agents typically well tolerate them based on for a specific patient based on the reference ranges
previous studies, but combination therapies are always obtained from a virtual population. However, realistically,
associated with elevated risk of immune-­related adverse in the most cases, not all parameters can be measured,
events. The identification of predictive biomarkers is crit- therefore, the performance of these biomarkers can be
ical to optimize patient benefit and reduce risk of toxici- compared with predict efficacy and a rough prediction
ties. Important parameters such as TMB, amount of Teff of ORR can be made even when there are only a few
in tumor microenvironment, and PD-­ L1/CEA expres- biomarkers available.
sion in cancer cells have shown potential to be predictive PD-­L2 is commonly considered as an inhibitor of T
biomarkers. PD-­L1 expression in APCs could potentially cell activation, but its actual role in the tumor microen-
be a biomarker for atezolizumab monotherapy, but it vironment are still being elucidated. An in vivo study of
showed no correlation with patients’ response in combina- PD-­L2 KO mice has shown a potential function of PD-­L2
tion therapy. Other parameters such as cross-­arm binding for augmenting T helper 1 and CTL responses.46 Another
efficiency λ and CD3-­cibisatamab binding affinity could report revealed that the aggregated form of PD-­L2 on
affect the efficacy of cibisatamab monotherapy, which DCs may suppress the interaction between PD-1 and
should be noted and carefully selected in the design of PD-­L1.47 Despite these reports, therapy targeting both
TCE. PD-1 ligands such as PD-1 blockade still provided clin-
We then explored if patients who failed to respond to ical benefit.48 Since the functions of PD-­L2 are still not
monotherapies can benefit from combination therapy by completely understood and the primary ligand of PD-1
studying patients in PD-­PD, PD-­SD, PD-­PR/CR, SD-­SD has been proven to be PD-­L1 with presence of PD-­L2, we,
and SD-­PR/CR groups. This analysis would help guide therefore, decided not to study the impact of PD-­L2 on
treatment recommendations by assessing if combination patients’ response in our model in depth. Fortunately,
therapy may work for specific patients. We have shown the expression of PD-­L2 in CRC is very low and its effect
how patients with different parameters responded differ- on CRC is much lower than in other types of cancer.
ently. Although patients in PD-­PR/CR and PD-­SD groups Although there is more PD-­L2 on APCs, sensitivity anal-
can both benefit from combination therapy, patients ysis has shown limited impact on patients’ response. As
in PD-­PR/CR group had higher TMB, Teff density and more evidence emerges to clarify the role of PD-­L2, we
Teff/Treg ratio in tumor, and PD-­L1 or CEA expression will add these mechanisms to the model in future work.
in cancer cells than patients in PD-­SD groups. However, if We have extended the ICIs module with PD-­L1 expres-
we only compare one parameter, the distribution of this sion in APCs. However, a higher level of PD-­L1 expression
parameter in patients of PD-­PR/CR, PD-­SD and PD-­PD in tumor-­ infiltrating immune cells especially tumor-­
groups may overlap with other groups, resulting in false associated macrophages (TAMs) and myeloid-­ derived
positive results, that is, high TMB patients in PD-­PD group suppressor cells (MDSCs) has been detected and proved
were predicted in PD-­PR/CR or PD-­SD groups. However, to be associated with patients’ survival.7 49 Since these cells
this can be avoided by comparing more parameters. are important for tumor progression, efficacy of targeting
Moreover, we computed the distribution of potential PD-­ L1 on tumor-­ infiltrating immune cells should be
biomarkers in our VPs and studied how their distribution further studied. In terms of TAMs, their polarization
affected patients’ response, by grouping all patients into toward M1 or M2 subsets in the tumor microenviron-
ROA, ROC, ROB, ROCMB and NR groups. This might ment has attracted a lot of attention and ample evidence
solve the problem of unnecessary trials and help deter- exists that TAMs appear and behave as M2 phenotype,
mine the best treatment option for patients and prospec- which is an important factor in protumorigenesis.50 It
tively assign them to the right group. As we mentioned is, therefore, necessary to include TAMs and MDSCs to
before, a small subset of MSS CRC patients can still have a better understanding of the roles of PD-1/PD-­L1
respond to anti-PD-1/PD-L1 antibodies or TCE mono- in different cells and of the function of TAMs polarization
therapy. These patients need to be identified instead of in the tumor microenvironment. This addition will then
being potentially excluded from a therapy that could be make our QSP model more complete in determining
suitable for them. In addition, some patients are unlikely biomarkers and providing guidance for future clinical
to respond to monotherapies, and should receive combi- trials.
nation therapy. There is also a subset of patients who may
respond to any treatment, for them the best treatment can
be chosen based on the clinicians’ experience and other CONCLUSION
factors such as potential toxicity and patients’ preferred In summary, we performed three in silico VCTs using
dosing regimen. Finally, patients who may not respond our QSP model with an expanded ICIs module. The
to any treatment can be identified and considered for model reproduced clinical trial outcomes and showed
other treatment options in a timely manner. Therefore, good consistency with previous publications. For each

Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141 9

 Open access

therapy, we were able to identify potential patient selec- 8 Yarchoan M, Albacker LA, Hopkins AC, et al. Pd-­L1 expression

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
and tumor mutational burden are independent biomarkers in most
tion biomarkers. By comparing the predicted outcomes cancers. JCI Insight 2019;4:e126908.
of monotherapy and combination therapy in the same set 9 Valentini AM, Di Pinto F, Cariola F, et al. Pd-­L1 expression
of VPs, the model was able to identify the best treatment in colorectal cancer defines three subsets of tumor immune
microenvironments. Oncotarget 2018;9:8584–96.
options for patients based on their individual character- 10 Gonzalez-­Exposito R, Semiannikova M, Griffiths B, et al. Cea
istics. In addition, the current model can be applied to expression heterogeneity and plasticity confer resistance to the CEA-­
targeting bispecific immunotherapy antibody cibisatamab (CEA-­TCB)
other TCEs and ICIs in different types of cancer to help in patient-­derived colorectal cancer organoids. J Immunother Cancer
assess plausible combination strategies and reduce the 2019;7:101.
effort of clinical assessment. Although the knowledge gap 11 Bacac M, Fauti T, Sam J, et al. A novel carcinoembryonic antigen T-­
cell bispecific antibody (CEA TCB) for the treatment of solid tumors.
between clinical trials and QSP modeling hinders appli- Clin Cancer Res 2016;22:3286–97.
cation of these models in many respects, including the 12 Waaijer SJH, Warnders FJ, Stienen S, et al. Molecular Imaging of
Radiolabeled Bispecific T-­Cell Engager 89Zr-­AMG211 Targeting CEA-­
optimal selection of combination therapies, predicting Positive Tumors. Clin Cancer Res 2018;24:4988–96.
toxicity, duration of response (DOR) and progression 13 Moore PA, Shah K, Yang Y, et al. Development of MGD007, a gpA33
free survival, this gap will be filled as greater emphasis is X CD3-­Bispecific dart protein for T-­cell immunotherapy of metastatic
colorectal cancer. Mol Cancer Ther 2018;17:1761–72.
placed on the collection of patient-­centric biomarkers in 14 Rader C. Bispecific antibodies in cancer immunotherapy. Curr Opin
current and future clinical trials. Biotechnol 2020;65:9–16.
15 Labrijn AF, Janmaat ML, Reichert JM, et al. Bispecific antibodies:
a mechanistic review of the pipeline. Nat Rev Drug Discov
Contributors HM, CG and ASP designed the project; ASP and CG directed the 2019;18:585–608.
project; RJS, HW and HM built and modified the model; HW contributed analysis 16 Yu L, Wang J. T cell-­redirecting bispecific antibodies in cancer
tools; HM performed all simulations, collected the data, did all analysis and drafted immunotherapy: recent advances. J Cancer Res Clin Oncol
the manuscript; ASP, CG, HW, RJS and JW revised the manuscript critically; all 2019;145:941–56.
authors have read and approved the final manuscript. 17 Tabernero J, Melero I, Ros W, et al. Phase Ia and Ib studies of
the novel carcinoembryonic antigen (CEA) T-­cell bispecific (CEA
Funding This work was supported by a grant from Boehringer Ingelheim CD3 TCB) antibody as a single agent and in combination with
Pharmaceuticals and NIH grants R01CA138264 and U01CA212007. atezolizumab: preliminary efficacy and safety in patients with
metastatic colorectal cancer (mCRC). American Society of Clinical
Competing interests None declared.
Oncology 2017.
Patient consent for publication Not required. 18 Kroschinsky F, Stölzel F, von Bonin S, et al. New drugs, new
toxicities: severe side effects of modern targeted and immunotherapy
Provenance and peer review Not commissioned; externally peer reviewed. of cancer and their management. Crit Care 2017;21:89.
Data availability statement Data are available on reasonable request. All data 19 Zarogoulidis P, Chinelis P, Athanasiadou A, et al. Possible adverse
effects of immunotherapy in non-­small cell lung cancer; treatment
relevant to the study are included in the article or uploaded as online supplementary
and follow-­up of three cases. Respir Med Case Rep 2017;22:101–5.
information. The authors confirm that the data supporting the findings of this study 20 Yarchoan M, Hopkins A, Jaffee EM. Tumor mutational burden and
and MATLAB scripts (SBML file) for model and data generation are available within response rate to PD-1 inhibition. N Engl J Med 2017;377:2500–1.
the article and the Supplementary Material. The model code and script to fully 21 Zhu J, Armstrong AJ, Friedlander TW, et al. Biomarkers of
implement and reproduce the results will be shared by the corresponding author in immunotherapy in urothelial and renal cell carcinoma: PD-­L1, tumor
the GitHub repository to any qualified researcher on request. mutational burden, and beyond. J Immunother Cancer 2018;6:4.
22 Boessen R, Heerspink HJL, De Zeeuw D, et al. Improving clinical trial
Open access This is an open access article distributed in accordance with the efficiency by biomarker-­guided patient selection. Trials 2014;15:103.
Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits 23 Milberg O, Gong C, Jafarnejad M, et al. A QSP model for predicting
others to copy, redistribute, remix, transform and build upon this work for any clinical responses to monotherapy, combination and sequential
purpose, provided the original work is properly cited, a link to the licence is given, therapy following CTLA-4, PD-1, and PD-­L1 checkpoint blockade.
and indication of whether changes were made. See https://​creativecommons.​org/​ Sci Rep 2019;9:11286.
licenses/​by/​4.0​ /. 24 Wang H, Milberg O, Bartelink IH, et al. In silico simulation of a clinical
trial with anti-­CTLA-4 and anti-­PD-­L1 immunotherapies in metastatic
ORCID iD breast cancer using a systems pharmacology model. R Soc Open
Huilin Ma http://​orcid.​org/​0000-​0003-​4238-​9475 Sci 2019;6:190366.
25 Bai JPF, Earp JC, Pillai VC. Translational quantitative systems
pharmacology in drug development: from current landscape to good
practices. Aaps J 2019;21:72.
26 Cheng Y, Thalhauser CJ, Smithline S, et al. QSP toolbox:
REFERENCES computational implementation of integrated workflow components
1 Jass JR. Classification of colorectal cancer based on correlation for deploying multi-­scale mechanistic models. Aaps J
of clinical, morphological and molecular features. Histopathology 2017;19:1002–16.
2007;50:113–30. 27 van der Graaf PH. Cpt: Pharmacometrics and systems
2 Darvin P, Toor SM, Sasidharan Nair V, et al. Immune checkpoint pharmacology. CPT: Pharmacomet Syst Pharmacol 2012;1:e8–4.
inhibitors: recent progress and potential biomarkers. Exp Mol Med 28 Sové RJ, Jafarnejad M, Zhao C, et al. QSP-­IO: a quantitative
2018;50:1–11. systems pharmacology toolbox for mechanistic multi-­scale modeling
3 Coupez D, Hulo P, Touchefeu Y, et al. Pembrolizumab for for immuno-­oncology applications. CPT Pharmacometrics Syst
the treatment of colorectal cancer. Expert Opin Biol Ther Pharmacol 2020. doi:10.1002/psp4.12546. [Epub ahead of print: 02
2020;20:219–26. Jul 2020].
4 Wu Z, Guo H-­F, Xu H, et al. Development of a tetravalent Anti-­ 29 Norton K-­A, Gong C, Jamalian S, et al. Multiscale Agent-­Based and
GPA33/Anti-­CD3 bispecific antibody for colorectal cancers. Mol hybrid modeling of the tumor immune microenvironment. Processes
Cancer Ther 2018;17:2164–75. 2019;7:37.
5 Chang C-­H, Wang Y, Li R, et al. Combination therapy with bispecific 30 Jafarnejad M, Gong C, Gabrielson E, et al. A computational model
antibodies and PD-1 blockade enhances the antitumor potency of T of neoadjuvant PD-1 inhibition in non-­small cell lung cancer. Aaps J
cells. Cancer Res 2017;77:5384–94. 2019;21:79.
6 Seidel JA, Otsuka A, Kabashima K. Anti-­Pd-1 and anti-­CTLA-4 31 Wang H, Sové RJ, Jafarnejad M, et al. Conducting a virtual clinical
therapies in cancer: mechanisms of action, efficacy, and limitations. trial in HER2-­negative breast cancer using a quantitative systems
Front Oncol 2018;8:86. pharmacology model with an epigenetic modulator and immune
7 Lin H, Wei S, Hurt EM, et al. Host expression of PD-­L1 determines checkpoint inhibitors. Front Bioeng Biotechnol 2020;8:141.
efficacy of PD-­L1 pathway blockade-­mediated tumor regression. J 32 Betts A, Haddish-­Berhane N, Shah DK, et al. A Translational
Clin Invest 2018;128:805–15. Quantitative Systems Pharmacology Model for CD3 Bispecific

10 Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141

 Open access

Molecules: Application to Quantify T Cell-­Mediated Tumor Cell Killing 42 Eng C, Kim TW, Bendell J, et al. Atezolizumab with or without

J Immunother Cancer: first published as 10.1136/jitc-2020-001141 on 27 August 2020. Downloaded from http://jitc.bmj.com/ on March 4, 2022 by guest. Protected by copyright.
by P-­Cadherin LP DART®. Aaps J 2019;21:66. cobimetinib versus regorafenib in previously treated metastatic
33 Ma H, Wang H, Sové RJ, et al. A quantitative systems pharmacology colorectal cancer (IMblaze370): a multicentre, open-­label, phase 3,
model of T cell Engager applied to solid tumor. Aaps J 2020;22:85. randomised, controlled trial. Lancet Oncol 2019;20:849–61.
34 Sznurkowski JJ, Żawrocki A, Sznurkowska K, et al. Pd-­L1 expression 43 Cheng X, Veverka V, Radhakrishnan A, et al. Structure and
on immune cells is a favorable prognostic factor for vulvar squamous interactions of the human programmed cell death 1 receptor. J Biol
cell carcinoma patients. Oncotarget 2017;8:89903–12. Chem 2013;288:11771–85.
35 Bocanegra A, Fernandez-­Hinojal G, Zuazo-­Ibarra M, et al. Pd-­L1 44 Gatalica Z, Vranic S, Xiu J, et al. High microsatellite instability (MSI-­H)
expression in systemic immune cell populations as a potential colorectal carcinoma: a brief review of predictive biomarkers in the
predictive biomarker of responses to PD-­L1/PD-1 blockade therapy era of personalized medicine. Fam Cancer 2016;15:405–12.
in lung cancer. Int J Mol Sci 2019;20:1631. 45 Kikuchi T, Mimura K, Okayama H, et al. A subset of patients with
36 Menter T, Tzankov A. Mechanisms of immune evasion and immune MSS/MSI-­low-­colorectal cancer showed increased CD8(+) TILs
modulation by lymphoma cells. Front Oncol 2018;8:54. together with up-­regulated IFN-γ. Oncol Lett 2019;18:5977–85.
37 Gibbons Johnson RM, Dong H. Functional expression of 46 Shin T, Yoshimura K, Shin T, et al. In vivo costimulatory role of B7-­DC
programmed Death-­Ligand 1 (B7-­H1) by immune cells and tumor in tuning T helper cell 1 and cytotoxic T lymphocyte responses. J
cells. Front Immunol 2017;8:961. Exp Med 2005;201:1531–41.
38 Pulko V, Liu X, Krco CJ, et al. TLR3-­stimulated dendritic cells up-­ 47 Karunarathne DS, Horne-­Debets JM, Huang JX, et al. Programmed
regulate B7-­H1 expression and influence the magnitude of CD8 T cell Death-1 Ligand 2-­Mediated Regulation of the PD-­L1 to PD-1 Axis
responses to tumor vaccination. J Immunol 2009;183:3634–41. Is Essential for Establishing CD4(+) T Cell Immunity. Immunity
39 Taube JM, Klein A, Brahmer JR, et al. Association of PD-1, PD-1 2016;45:333–45.
ligands, and other features of the tumor immune microenvironment 48 Yearley JH, Gibson C, Yu N, et al. Pd-­L2 expression in human
with response to anti-­PD-1 therapy. Clin Cancer Res tumors: relevance to anti-­PD-1 therapy in cancer. Clin Cancer Res
2014;20:5064–74. 2017;23:3158–67.
40 Gatalica Z, Snyder C, Maney T, et al. Programmed cell death 1 (PD- 49 Singhal S, Stadanlick J, Annunziata MJ, et al. Human tumor-­
1) and its ligand (PD-­L1) in common cancers and their correlation associated monocytes/macrophages and their regulation of
with molecular cancer type. Cancer Epidemiol Biomarkers Prev T cell responses in early-­stage lung cancer. Sci Transl Med
2014;23:2965–70. 2019;11:eaat1500.
41 Oliveira AF, Bretes L, Furtado I. Review of PD-1/PD-­L1 inhibitors in 50 De Palma M, Lewis CE. Macrophage regulation of tumor responses
metastatic dMMR/MSI-­H colorectal cancer. Front Oncol 2019;9:396. to anticancer therapies. Cancer Cell 2013;23:277–86.

Ma H, et al. J Immunother Cancer 2020;8:e001141. doi:10.1136/jitc-2020-001141 11