Abstract

Background

   Oncofetal splice variants of extracellular matrix (ECM) proteins
   present a unique group of target antigens for the immunotherapy of
   pediatric cancers. However, limited data is available if these splice
   variants can be targeted with T cells expressing chimeric antigen
   receptors (CARs).

Methods

   To determine the expression of the oncofetal version of tenascin C
   (TNC) encoding the C domain (C.TNC) in pediatric brain and solid
   tumors, we used quantitative reverse transcription PCR and
   immunohistochemistry. Genetically modified T cells were generated from
   human peripheral blood mononuclear cells and evaluated in vitro and in
   vivo.

Results

   We demonstrate that C.TNC is expressed on a protein level in pediatric
   tumors, including diffuse intrinsic pontine glioma, osteosarcoma,
   rhabdomyosarcoma, and Ewing sarcoma. We generate C.TNC-CAR T cells and
   establish that these recognize and kill C.TNC-positive tumor cells.
   However, their antitumor activity in vivo is limited. To improve the
   effector function of C.TNC-CAR T cells, we design a leucine
   zipper-based chimeric cytokine receptor that activates interleukin-18
   signaling pathways (Zip18R). Expression of Zip18R in C.TNC-CAR T cells
   improves their ability to secrete cytokines and expand in repeat
   stimulation assays. C.TNC-CAR.Zip18R T cells also have significantly
   greater antitumor activity in vivo compared with unmodified C.TNC-CAR T
   cells.

Conclusions

   Our study identifies the C domain of the ECM protein TNC as a promising
   CAR T-cell therapy for pediatric solid tumors and brain tumors. While
   we focus here on pediatric cancer, our work has relevance to a broad
   range of adult cancers that express C.TNC.

   Keywords: Chimeric antigen receptor - CAR, Immunotherapy, T cell
     __________________________________________________________________

WHAT IS ALREADY KNOWN ON THIS TOPIC

     * Oncofetal splice variants of extracellular matrix (ECM) proteins
       have so far been studied mainly in adult cancers, and the majority
       of ECM-targeted therapeutic approaches rely on antibody drug,
       cytokine, or radioisotope conjugates.

WHAT THIS STUDY ADDS

     * We demonstrate here that the oncofetal version of tenascin C (TNC)
       encoding the C domain (C.TNC) is expressed on a protein level in
       both pediatric brain and solid tumors and can be targeted with
       C.TNC-chimeric antigen receptor T cells.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

     * Our study highlights that oncofetal splice variants of ECM proteins
       are viable targets for cellular immunotherapy.

Introduction

   Chimeric antigen receptor (CAR) T-cell therapy has shown promise for
   pediatric hematological malignancies; however, progress in the solid
   and brain tumor space has been limited.[59]1,[60]3 Although failure of
   CAR T cells in solid tumors is likely a multifactorial issue, a
   critical first step to success is target antigen selection.[61]^1 2
   Most tumor-associated antigens targeted by CAR T cells are surface
   proteins; however, there have been preclinical successes in targeting
   secreted proteins such as oncofetal fibronectin and
   collagen.[62]4,[63]8 We recently developed a computational pipeline to
   discover cancer-specific exons as targets for the immunotherapy of
   pediatric solid and brain tumors.[64]^8 Top hits included oncofetal
   splice variants and differentially expressed proteins of the
   extracellular matrix (ECM), which have been previously described in a
   broad range of adult cancers.

   One of the identified ECM proteins was oncofetal tenascin C (TNC), a
   hexameric protein that can bind to cells via several cell surface
   proteins and integrins.[65]^9 In normal development, TNC undergoes
   splicing of nine exons between FN-III domains 5 and 6.[66]^10 Changes
   in isoform expression have been linked to changes in pH, the
   upregulation of SRSF6, and growth factors such as TGFβ1 and
   FGF.[67]11,[68]16 The long isoforms of TNC have been previously
   described in cancer and can be implicated in tumor cell proliferation
   and migration as well as impairing antitumor immune
   responses.[69]17,[70]19 Antibody-drug or cytokine conjugates targeting
   oncofetal TNC for solid and brain tumors have shown promising safety
   and antitumor activity in preclinical models, and early-phase clinical
   studies have confirmed the safety of oncofetal TNC as a target for
   immunotherapy.[71]^20

   T cells can produce cytokines to self-sustain their effector function;
   however, in the setting of chronic antigen exposure, cytokine
   production is limited.[72]^21 To engineer this component of T-cell
   effector function, investigators have genetically modified T cells to
   co-express CARs with unmodified or membrane-bound cytokines, or
   chimeric cytokine receptors.[73]^2 3 For solid tumor-specific CAR T
   cells, predominately, JAK/STAT-activating cytokines interleukin (IL)-7,
   IL-12, IL-15, IL-21, and IL-23 have been investigated.[74]22,[75]31 In
   addition, we and others have shown that MyD88 signaling, either via
   IL-18 signaling or including MyD88 as a CAR signaling domain, is
   beneficial to CAR T cells in the setting of chronic antigen
   exposure.[76]32,[77]36 Building on these findings, we decided to use
   our modular leucine zip receptor (ZipR) platform[78]^37 to design a
   constitutively active IL-18 receptor.

   In this study, we explored CAR T cells redirected to the C domain, one
   of the exons expressed in oncofetal TNC (C.TNC-CAR T cells), as an
   immunotherapeutic for pediatric brain and solid tumors. We developed a
   second-generation CAR specific for C.TNC based on the G11 monoclonal
   antibody (mAb), which recognizes the human and murine C domain.[79]^38
   We found that C.TNC-CAR T cells have antigen-specific cytolysis of
   target tumor cells but limited efficacy in xenograft models. While
   altering CAR design did not improve antitumor activity, expressing a
   constitutively active IL-18 receptor in C.TNC-CAR T cells bolstered
   their effector function, resulting in improved antitumor activity in
   vivo.

Methods

   Please see [80]online supplemental methods for details on tumor cell
   lines, immunohistochemistry (IHC), reverse transcription quantitative
   PCR (RT-qPCR), retroviral vectors, CAR T-cell generation, flow
   cytometry, co-cultures, MILLIPLEX, and statistical analysis.

Patient-derived xenograft and primary sarcoma tissue samples

   Patient-derived xenografts were obtained from the Childhood Solid Tumor
   Network (CSTN) ([81]https://cstn.stjude.cloud/search/).[82]^39 Fresh
   frozen tissues were hand homogenized in DPBS (Gibco, 14190–144) +
   1% FBS (Gibco) and filtered through polystyrene test tubes with cell
   strainer caps (Falcon, 352235) to generate a single-cell suspension.
   Formalin-fixed paraffin-embedded tissues were also obtained for IHC
   purposes. De-identified formalin-fixed paraffin-embedded tissue blocks
   from clinical patient tumor samples were cut and stained as previously
   described.[83]^8

Single-cell RNA sequencing

Sample collection

   C.TNC-CAR.IL-18 receptor-based ZipR (Zip18R) T cells were generated by
   double transduction as previously described for two biological donors.
   C.TNC-CAR.Zip18R T cells were collected at baseline or after 12, 24,
   and 48 hours stimulation with LM7.green fluorescent protein
   (GFP).firefly luciferase (ffLuc) in the presence of IL-15 (PeproTech;
   5 ng/mL). Samples were sorted for viability and hashed for multiplexing
   with TotalSeq-C anti-human Hashtag Antibodies (BioLegend 394661,
   394663, 394665). To add hashing antibodies, up to 5e6 cells were
   collected and incubated with Human TruStain FcX block (BioLegend),
   followed by a hashing antibody in a volume of 100 μL.

Library preparation

   Equal cell numbers from each hashed sample were pooled before loading
   onto a Chromium Controller to generate 10,000 single-cell gel beads in
   emulsion for single-cell RNA sequencing (scRNA-seq). Libraries were
   prepared using Chromium Next GEM Single Cell 5’ V.2 (Dual index) and
   Gel Bead Kit (10x Genomics). Complementary DNA (cDNA) was amplified (13
   cycles), after which it was used for the preparation of gene expression
   and cell surface protein libraries. The cDNA content of each sample and
   library was quality-checked using a high-sensitivity DNA chip with a
   2100 Bioanalyzer (Agilent Technologies). The libraries were sequenced
   on NovaSeq (Illumina) with paired-end reads of 26 cycles for read 1 and
   90 cycles for read 2×. Median reads per cell ranged from 117,565 to
   132,623 in the baseline cohort and 13,717–31,087 reads for the
   stimulated cohort.

Data preprocessing and sample integration

   C.TNC-CAR and Zip18R sequences were added to the human reference
   transcriptome (refdata-gex-GRCh38-2020-A) for reads mapping. The Cell
   Ranger V.7.1.0 Single-Cell software suite (10x Genomics) was used to
   process scRNA-seq FASTQ files. The “cellranger multi” and “cellranger
   bamtofastq” commands were performed for sample demultiplexing and
   extracting sample FASTQ files. The “cellranger count” command was
   performed to align FASTQ files to reference genome and summarize the
   data into matrices describing gene read counts (unique molecular
   identifier (UMI)) per cell. SingleCellExperiment[84]^40 object was
   generated and used for the following analysis. Damaged or dying cells
   were filtered based on the outlier status of mitochondrial genes
   expression using scuttle.[85]^41 Cells with a detected gene number less
   than 300 were filtered out. Read counts were normalized using scuttle
   logNormCounts function.[86]^41 Top 2,000 high-variance genes were
   selected to calculate principal components using scater.[87]^41
   Harmony[88]^42 was used to integrate all the samples for batch
   correction. To get a pure T-cell population, any cells with no CD3δ
   expression (zero count) were filtered out.

Differentially expressed gene detection and pathway enrichment analysis

   For each sample, the T-cell population was classified into four
   subpopulations (C.TNC-CAR.Zip18R, C.TNC-CAR, Zip18R, non-transduced)
   based on whether the CAR or Zip18R was expressed or not (zero count).
   Then pseudo-bulk data for each subpopulation in each sample was
   generated using scuttle.[89]^41 DESeq2 pipeline[90]^43 was used to
   generate principal component analysis (PCA) plot. Delegate (pseudo-bulk
   DESeq2 method with Wald test)[91]^44 was used to detect differentially
   expressed genes. ClusterProfiler[92]^45 was used to perform Kyoto
   Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis.
   Particularly, DESeq2 default normalized counts were used to calculate
   log2 fold change between samples in two conditions (pseudo-count as 10
   defined from counts distribution). “gseKEGG” commands were performed
   with Benjamini-Hochberg (BH) adjustment method.

Statistical analysis

   To identify differentially expressed genes from different cell
   populations on scRNA-seq data, the software Delegate[93]^44 was
   performed with pseudo-bulk DESeq2[94]^43 method and Wald test. KEGG
   pathway enrichment analysis for pseudo-bulk T-cell population was
   performed with “gseKEGG” commands in ClusterProfiler[95]^45 and p value
   was adjusted by BH method.

Data availability

   The scRNA-seq data generated in this study has been deposited in the
   European Genome-Phenome Archive (EGA) under the title “Single-cell RNA
   sequencing of human IL-18R supported CAR T cells targeting oncofetal
   Tenascin C”. The raw data can be obtained by a request to EGA. The
   scRNA-seq expression matrices are available on Gene Expression Omnibus
   (GEO) under ascension code [96]GSE282046.

Xenograft mouse models

   Animal experiments followed a protocol approved by the St. Jude
   Institutional Animal Care and Use Committee (646–100595). All
   experiments used NOD-scid IL-2Rgamma^null (NSG) mice obtained from St.
   Jude’s NSG colony. Studies were done in biological duplicate, altering
   mouse gender between experiments. For imaging, mice were injected in
   the intraperitoneal (i.p.) cavity with 150 mg/kg of D-luciferin
   (PerkinElmer) 5–10 min before imaging, anesthetized with isoflurane,
   and imaged under anesthesia with a Xenogen IVIS-200 imaging system. The
   photons emitted from the luciferase-expressing cells were quantified
   using Living Image software (Caliper Life Sciences). Total emitted
   photon flux (photons per second) was used to evaluate tumor burden. For
   the i.p. tumor model, 5–6-week-old mice were injected i.p. with 1×10^6
   LM7.GFP.ffLuc cells, and after 7 days, received a single i.p. injection
   of 1×10^6 T cells. For the intracranial (i.c.) tumor model,
   10–12-week-old mice were injected i.c. with 1×10^6 DIPG007.YFP.ffLuc
   cells in 2 µL of 80% Matrigel (Corning) mixed with PBS (Gibco), and
   after 7 days, received a single i.c. injection of 1 or 2×10^6 T cells.
   In both models, mice were imaged to confirm the presence of tumors
   prior to T-cell injection, and mice were euthanized when they reached
   (1) two consecutive flux values >1×10^10, (2) a single flux value
   >1×10^11, or (3) physical euthanasia criteria (weight loss, signs of
   distress). For the subcutaneous (s.c). tumor model, 7–9-week-old mice
   were injected in the right flank s.c. with 2×10^6 A673 cells, and after
   7 days, received a single intravenous injection of T cells. Mice were
   euthanized when they met physical euthanasia criteria or when the tumor
   size was greater than 4,000 mm^3.

Results

C.TNC is expressed in pediatric solid and brain tumors

   We first set out to establish C.TNC expression in pediatric solid and
   brain tumor samples. First, we screened pediatric diffuse intrinsic
   pontine glioma (DIPG) and pediatric sarcoma cell lines using RT-qPCR
   and confirmed that all samples contained the C.TNC exon at an RNA level
   ([97]figure 1A). We screened patient-derived xenograft samples and
   demonstrated high expression of C.TNC in osteosarcoma (OS) samples
   ([98]figure 1B, [99]online supplemental STable 1). Protein expression
   was further confirmed by IHC, using the G11 mAb to stain primary
   samples of H3K27M+DIPG, ZFTA-fusion positive ependymoma, OS,
   rhabdomyosarcoma, and Ewing sarcoma ([100]figure 1C,D). C.TNC
   expression, as determined by IHC, was localized to both the cytoplasm
   and membranes of cells in primary and xenograft tumor samples, and was
   also observed within the immediate tumor microenvironment, which was
   interpreted as tumor-associated production and secretion of C.TNC by
   the evaluated pediatric tumor types. Depending on the xenograft or
   primary patient samples, neoplasms expressed C.TNC at moderate to
   strong levels based on their H-score.

Figure 1. C.TNC is expressed in solid and brain tumors. (A,B) Reverse
transcription quantitative PCR of (A) pediatric cell lines (n=3
bioreplicates, mean+SEM) and (B) patient-derived xenografts for C.TNC. ΔC[T]
is relative to GAPDH. Acute lymphoblastic leukemia cell line CCRF-CEM had no
C.TNC detected. (C) H-scores of primary human H3K27M+DIPG, ZFTA
fusion-positive ependymoma (EPN), osteosarcoma (OS), rhabdomyosarcoma (RMS),
and Ewing sarcoma (EWS) tumors. (D) Representative immunohistochemistry
staining of primary FFPE human tumor samples and tonsils (negative control
(neg Co)) with a C.TNC-specific antibody. 20× magnifications, 100 µM scale
bar. Images were taken with an Olympus BX46 microscope and a Nikon DS-Fi3
camera at a 20× magnification and edited with Photoshop 25.5.1. C.TNC,
tenascin C encoding the C domain; DIPG, diffuse intrinsic pontine glioma.

   [101]Figure 1
   [102]Open in a new tab

C.TNC-CAR T cells recognize and kill C.TNC-positive tumor cells

   We generated a retroviral vector encoding a C.TNC-CAR consisting of a
   single chain variable fragment (scFv)-based C.TNC binding domain
   derived from the G11 mAb, a short hinge, a CD28 transmembrane, and a
   CD28.zeta endodomain, a self-cleaving T2A sequence, and truncated CD19
   (tCD19) ([103]figure 2A, [104]online supplemental SFigure 4A).
   C.TNC-CAR T cells were generated by standard retroviral transduction
   and average transduction efficacy was~70% as determined by flow
   cytometry, detecting either the CAR (scFv or linker) or tCD19
   ([105]figure 2B). We evaluated C.TNC-CAR T-cell effector function
   against C.TNC-positive (sarcoma, DIPG) and C.TNC-negative (CCRF-CEM)
   cell lines. In co-culture assays, C.TNC-CAR T cells produced
   significant amounts of interferon (IFN)-γ compared with non-transduced
   (NT) T cells only in the presence of C.TNC-positive cell lines
   ([106]figure 2C,D). C.TNC-CAR T cells also had significant cytolytic
   activity against C.TNC-positive sarcoma cell lines compared with NT T
   cells ([107]figure 2E,F), and no cytotoxicity against the
   C.TNC-negative cell line CCRF-CEM ([108]figure 2G). We determined the
   phenotype of C.TNC-CAR T cells cultured with and without stimulation
   with C.TNC-positive LM7.GFP.ffLuc cells. C.TNC-CAR expression alone in
   T cells did not alter the CD4/CD8 ratios or their phenotype
   ([109]online supplemental SFigure B, C). With antigen stimulation,
   C.TNC-CAR T cells demonstrated T-cell differentiation as judged by a
   decline in naïve-like subsets as compared with NT T cells ([110]online
   supplemental SFigure D, E). To further confirm antigen-specificity, we
   generated a non-functional C.TNC.mu-CAR that lacked the intracellular
   CD28 domain and had mutated CD3ζ ITAMS ([111]online supplemental
   SFigure 4F). C.TNC.mu-CAR T cells were successfully expressed in T
   cells ([112]online supplemental SFigure 4G) and had no cytolytic
   activity against LM7.GFP.ffLuc cells compared with NT cells
   ([113]online supplemental SFigure 4H). Next, we assessed the antitumor
   activity of C.TNC-CAR T cells in vivo in our OS LM7.GFP.ffLuc and
   DIPG007.YFP.ffLuc xenograft models ([114]figure 2H,K). In both models,
   C.TNC-CAR T cells exhibited antitumor activity at early time points as
   determined by a reduction in tumor flux values as compared with
   controls ([115]figure 2I,L; [116]online supplemental SFigures 5 and 6).
   This transient antitumor activity of C.TNC-CAR T cells translated to a
   significant survival advantage in the OS model, and although this did
   not reach significance for the mice in DIPG007, the median survival
   between the groups reached 50 days ([117]figure 2J,M). To evaluate the
   presence of tumor cells, T cells, and TNC expression post C.TNC-CAR T
   cell therapy, DIPG007-bearing mice were euthanized 2 weeks after
   intratumoral injection of NT or C.TNC-CAR T cells. NT T cell-treated
   tumors, identified by H3K27M-staining, were larger in comparison to
   tumors injected with C.TNC-CAR T cells ([118]online supplemental
   SFigure 7). This was mirrored by decreased TNC expression. Human
   CD3-positive T cells were only detected in mice after C.TNC-CAR T-cell
   therapy ([119]online supplemental SFigure 7).

Figure 2. C.TNC-CAR T cells have antitumor activity in vitro and in vivo. (A)
Schematic of C.TNC-CAR design. (B) Transduction efficiency of healthy donor T
cells determined via flow cytometry on day 7 (n=5, mean+SEM). Graph shows the
percentage of positive cells stained for each respective antibody for CAR
transgene detection. (C,D) IFN-γ production measured by ELISA after 48 hours
of co-culture with (C) sarcoma or (D) brain tumor cell lines at 2:1 effector
to target (E:T) ratio. Negative values were plotted as zero (mean+SEM, n=3–4
for NT and C.TNC), two-way ANOVA, ***p<0.001, ****p<0.0001. (E,F)
Cytotoxicity after 72 hours at a 4:1 E:T ratio determined by a
luciferase-based assay (mean+SEM, n=3), two-way ANOVA, ****p<0.0001. (G)
Cytotoxicity of C.TNC-negative cell line CCRF-CEM determined by a
luciferase-based assay (n=4–5, mean+SEM). (H) Schematic of experiment in
5–6-week-old female NSG mice. 1×10^6 LM7.green fluorescent protein.firefly
luciferase cells were injected intraperitoneally (i.p.), followed by 1×10^6 T
cells injected i.p. 7 days later. (I) Flux values from weekly IVIS images
(n=5 per cohort). (J) Overall survival of the mice, Mantel-Cox test,
**p<0.01. (K) Schematic of experiment in 10–12-week-old male NSG mice. 1×10^6
DIPG007.YFP.ffLuc cells were injected intracranially (i.c.), followed by
2×10^6 T cells injected i.c. 7 days later. (L) Flux values from weekly IVIS
images (n=5 per cohort). (M) Overall survival of the mice, Mantel-Cox test.
ANOVA, analysis of variance; CAR, chimeric antigen receptor; C.TNC, tenascin
C encoding the C domain; DIPG, diffuse intrinsic pontine glioma; IFN,
interferon; NT, non-transduced; scFv, single chain variable fragment.

   [120]Figure 2
   [121]Open in a new tab

   To explore if redesigning the CAR structure could improve C.TNC-CAR
   T-cell effector function, we designed four additional CARs with changes
   in the linker, hinge, and transmembrane domains ([122]online
   supplemental SFigure 8A). All CARs were expressed in T cells
   ([123]online supplemental SFigure 8B), and all C.TNC-CAR T-cell
   populations had antitumor activity compared with NT T cells in a
   24-hour cytotoxicity assay ([124]online supplemental SFigure 8C). Based
   on their performance in their cytolytic activity, we selected the
   C.TNC-SL-CAR and C.TNC-zeta-CAR T cells for in vivo testing
   ([125]online supplemental SFigure 8D). Neither design improved the
   antitumor activity of C.TNC-CAR T cells ([126]online supplemental
   SFigure 9), and we, therefore, continued to use the original CD28.zeta
   CAR design in future experiments.

An IL-18-based constitutively active cytokine receptor improves the effector
function of C.TNC-CAR T cells

   A key requirement for CAR T-cell effector function is the provision of
   signal 3, or proinflammatory cytokine support.[127]^46 47 We and others
   have demonstrated that MyD88 signaling improves CAR T-cell effector
   function by inducing type 2 cytokine signaling and enhancing T-cell
   expansion and proliferation.[128]32,[129]36 We first evaluated if
   C.TNC-CAR T cells expressing a granulocyte-macrophage
   colony-stimulating factor (GM-CSF)/IL-18 switch receptor (GM18), which
   requires GM-CSF binding for IL-18 signaling, could better control tumor
   growth in our LM7.GFP.ffLuc in vivo model.[130]^33 C.TNC-CAR.GM18 T
   cells had limited antitumor activity compared with NT T cells
   ([131]online supplemental SFigure 10). We recently published the
   development of modular leucine zipper-based chimeric cytokine receptors
   (ZipRs) to provide constitutive JAK/STAT signaling to transduced T
   cells.[132]^37 We decided to explore here if our modular chimeric
   cytokine receptor platform could be adapted for the IL-18 receptor
   (Zip18R) for constitutive IL-18 support. We designed two Zip18Rs with
   one (1X) or two (2X) pairs of leucine zippers ([133]figure 3A) and
   cloned these into a retroviral vector that also encoded a P2A sequence
   and mClover.

Figure 3. Zip18R improves CAR T-cell effector function. (A) Schematic of 1X
and 2X Zip18Rs, the leucine zippers (Zip) are connected with a (G[4]S)[3]
linker. The 1X and 2X Zip18Rs retroviral vectors also encode a P2A skip
sequence and mClover. TM/IC: transmembrane and intracellular. (B)
Representative transduction efficiency of Ramos-Blue and Ramos-Blue KD-MyD
NF-κB/AP-1 reporter cells determined via flow cytometry. (C) Absorbance
values of transduced Ramos-Blue NF-κB/AP-1 reporter cell supernatant mixed
with QUANTI-Blue reagent (n=3 from two separate transductions). (D)
Transduction of T cells with Zip18R (n=3, mean+SEM). (E,F) Populations after
(E) 7 days or (F) 14 days of cytokine starvation as determined by flow
cytometry (n=3, mean+SEM). Two-way analysis of variance, *p<0.05, **p<0.01,
***p<0.001. (G) Schematic of A673 model. T cells were sorted prior to
injection. 7–9-week-old male NSG mice were used. (H) Tumor caliper
measurements (n=4 for tumor and CAR only, n=5 for CAR.Zip18Rs). (I)
Kaplan-Meier survival curve, Mantel-Cox test, **p<0.01. (J) Tumor caliper
measurements. (n=3 for tumor and CAR.2XZip18R, n=4 for CAR and CAR.1XZip18R).
(K) Kaplan-Meier survival curve, Mantel-Cox test, *p<0.05. CAR, chimeric
antigen receptor; NT, non-transduced; Zip18R, interleukin-18 receptor-based
leucine zipper receptor.

   [134]Figure 3
   [135]Open in a new tab

   Both Zip18Rs were functional and their functionality was dependent on
   the expression of MyD88 as judged by their ability to activate
   NFKB/AP-1 only in a reporter cell line in which MyD88 was present
   ([136]figure 3B,C). T cells could be readily transduced with retroviral
   vectors encoding Zip18Rs with a transduction efficiency of~50% as
   judged by mClover expression ([137]figure 3D). Expression of either
   Zip18R alone enhanced the survival of T cells after 1 week of cytokine
   starvation but had similar viability when starved for 2 weeks
   ([138]figure 3E,F). In vivo, the antitumor activity of CAR T cells
   targeting EphA2, a cell surface antigen,[139]^48 was significantly
   enhanced by expressing either Zip18R ([140]figure 3G–K). Since there
   was no difference between both Zip18Rs, we selected the 1X Zip18R for
   further testing.

   C.TNC-CAR.Zip18R T cells were generated by co-transduction, and the
   resulting T-cell populations consisted of~44% double-positive cells
   ([141]figure 4A). We evaluated C.TNC-CAR and C.TNC-CAR.Zip18R T cells
   against LM7.GFP.ffLuc cells in a repeat stimulation assay, in which T
   cells are co-cultured with tumor cells at a 2:1 effector to target
   ratio in the presence of IL-15, and only restimulated if they killed
   the tumor cells and expanded ([142]figure 4B). Controls included NT,
   Zip18R, C.TNC.mu-CAR, and C.TNC.mu-CAR.Zip18R T cells. C.TNC-CAR.Zip18R
   T cell expansion was significantly greater after each stimulation
   compared with C.TNC-CAR T cells ([143]figure 4C,D). All four control
   T-cell populations did not kill LM7.GFP.ffLuc cells and did not expand
   significantly after the first stimulation, confirming antigen-specific
   C.TNC-CAR and C.TNC-CAR.Zip18R T-cell expansion and demonstrating that
   expression of Zip18R in T cells does not induce autonomous cell growth
   ([144]figure 4C). During the repeat stimulation assay, we collected
   media 48 hours post each stimulation to evaluate cytokine and chemokine
   production using a 48 Multiplex Assay. After the first stimulation,
   C.TNC-CAR and C.TNC-CAR.Zip18R T cells produced significant amounts of
   type 1 (GM-CSF, IFN-γ, IL-2, tumor necrosis factor (TNF)-α), type
   2/type 17 (IL-6, IL-10, IL-13, IL-17A), and chemokines (CXCL1, CXCL9,
   CXCL10, CCL2, CCL5, CCL7, CCL22) compared with Zip18R T cells
   ([145]figure 4E–G). In addition, C.TNC-CAR.Zip18R T cells produced
   higher amounts of IL-5, IL-17A, and IL-17F and less IL-10 compared with
   C.TNC-CAR T cells. C.TNC.mu-CAR.Zip18R, Zip18R, and NT T cells did not
   produce higher levels of cytokines or chemokines in the presence of
   tumor cells ([146]online supplemental SFigure 11). There were no
   significant differences in the cytokine/chemokine expression profile of
   NT and Zip18R T cells ([147]online supplemental SFigure 11). To
   evaluate the long-term effects of Zip18R expression on cytokine and
   chemokine production, we determined cytokine production after the
   fourth stimulation. After the fourth stimulation, C.TNC-CAR T cells
   produced significantly lower levels of type 1 cytokines, and higher
   levels of chemokines ([148]figure 4H–J, [149]online supplemental
   SFigure 12A). In contrast, there was no significant decrease in type 1
   and an increase in type 2/type 17 cytokine production of
   C.TNC-CAR.Zip18R T cells, and while chemokine production was also
   increased, this did not reach significance ([150]online supplemental
   SFigure 12B).

Figure 4. Zip18R bolsters C.TNC-CAR T-cell effector function in vitro. (A)
Transduction efficiency of primary T cells was determined via flow cytometry
on day 7 post-transduction (n=4, mean+SEM). Determined via F(ab’)[2] staining
and mClover expression. (B) Repeat stimulation assay schematic. T cells were
stimulated with LM7.green fluorescent protein.firefly luciferase at an
effector to target ratio of 2:1 in the presence of IL-15 every 4 days. (C)
Expansion of T cells in repeat stimulation assay. Each graph represents one
donor. (D) Fold change of C.TNC-CAR and C.TNC.CAR.Zip18R T cells after
stimulations 1 through 10, paired t-test, ****p<0.0001. (E–G) Quantification
of (E) type 1 and (F) type 2/type 17 cytokines, and (G) chemokines 48 hours
post first stimulation of the repeat stimulation assay (n=3, mean+SEM), data
was log-transformed before statistical analysis, two-way analysis of
variance, *p<0.05, **p<0.01, ***:p<0.001, ****p<0.0001. (H–J) Comparison of
the sum of (H) type 1 cytokines, (I) type 2/type 17 cytokines, and (J)
chemokines produced post first stimulation and fourth stimulation by
C.TNC-CAR T cells and C.TNC-CAR.Zip18R T cells from repeat stimulation assay
(n=3), unpaired t-test, *p<0.05, **p<0.01. CAR, chimeric antigen receptor;
C.TNC, tenascin C encoding the C domain; GM-CSF, granulocyte-macrophage
colony-stimulating factor; IFN, interferon; IL, interleukin; NT,
non-transduced; TNF, tumor necrosis factor; Zip18R, interleukin-18
receptor-based leucine zipper receptor.

   [151]Figure 4
   [152]Open in a new tab

   To evaluate the transcriptional changes induced by Zip18R expression,
   we performed scRNA-seq of C.TNC-CAR.Zip18R T cells. We collected
   C.TNC-CAR.Zip18R T cells from two donors at baseline and after 12, 24,
   or 48 hours of co-culture with LM7.GFP.ffLuc cells in the presence of
   IL-15 ([153]figure 5A). We were able to detect NT, Zip18R+, C.TNC-CAR+,
   and C.TNC-CAR+Zip18R+cell populations within each CAR T-cell product
   ([154]online supplemental SFigure 13). Using principal component
   analysis, we found that the T-cell populations clustered according to
   the donor ([155]online supplemental SFigure 14A), which in part was due
   to differences in their gender ([156]online supplemental SFigure 14B).
   Over the course of tumor cell stimulation, C.TNC+Zip18R+T cells
   enriched while NT T cells decreased, suggesting antigen-dependent
   expansion of C.TNC-CAR.Zip18R T cells ([157]figure 5B).

Figure 5. Zip18R signaling alters the transcriptome of C.TNC-CAR T cells. (A)
Schematic of experiment. C.TNC-CAR.Zip18R T cells were collected from culture
in IL-7/IL-15 and frozen for analysis or freshly collected from a co-culture
assay with LM7.GFP.ffLuc (LM7.GL) tumor cells in the presence of IL-15 at 12,
24, and 48 hours post tumor cell stimulation. All four populations are
present in the same culture. (B) Percentages of C.TNC-CAR+Zip18R+,
C.TNC-CAR+, Zip18R+, and NT T cells at 12, 24, and 48 hours post co-culture.
(C,D) Gene Set Enrichment Analysis comparing C.TNC-CAR.Zip18R T cells to
C.TNC-CAR T cells at (C) baseline or (D) 48 hours post stimulation with
LM7.GFP.ffLuc cells. Top 10 activated or suppressed significant
(p.adjust<0.1) KEGG pathways are shown. (E–I) 1×10^6 NT, Zip18R, C.TNC-CAR,
and C.TNC-CAR.Zip18R T cells were cultured in media or against LM7 cells at a
2:1 effector to target ratio for 48 hours and then collected for flow
cytometric analysis. (E) Transduction (TDX) percentage of NT, Zip18R,
C.TNC-CAR, and C.TNC-CAR.Zip18R T cells alone and after 48 hours of
stimulation. CD3+ expression is shown for NT samples. (n=3, mean+SEM),
two-way ANOVA, ****p<0.0001. (F–I) Expression within pure isolated
populations of T cells gated on CD3+ (NT), mClover+ (Zip18R+), (G[4]S)[3]+
(C.TNC-CAR+), and (G[4]S)[3]+ mClover+ (C.TNC-CAR+Zip18R+) for (F) CD69, (G)
CD28, (H) CD39, and (I) TIM3 (n=3, mean+SEM), two-way ANOVA, *p<0.05,
**p<0.01, ***p<0.001, ****p<0.0001. Only showing significance values for each
cohort comparing Alone versus +LM7 and C.TNC-CAR versus C.TNC-CAR.Zip18R for
both conditions. All p values are reported in [158]online supplemental
SFigure 16. ANOVA, analysis of variance; CAR, chimeric antigen receptor;
C.TNC, tenascin C encoding the C domain; ffLuc, firefly luciferase; GFP,
green fluorescent protein; GM-CSF, granulocyte-macrophage colony-stimulating
factor; IL, interleukin; KEGG, Kyoto Encylopedia of Genes and Genomes; NT,
non-transduced; TNF, tumor necrosis factor; Zip18R, interleukin-18
receptor-based leucine zipper receptor.

   [159]Figure 5
   [160]Open in a new tab

   We next used Gene Set Enrichment Analysis to uncover which pathways
   were activated by Zip18R expression in C.TNC-CAR T cells. At baseline,
   we found that both donors had IL-17 signaling and cytokine–cytokine
   receptor interaction pathways upregulated in C.TNC-CAR.Zip18R T cells
   compared with C.TNC-CAR T cells ([161]figure 5C). Those pathways were
   also upregulated after 48 hours of antigen stimulation ([162]figure
   5D). In addition, for one donor, there were several immune-related
   disease pathways upregulated in C.TNC-CAR.Zip18R T cells ([163]figure
   5C,D (Donor B)). We next evaluated genes involved in MyD88 signaling,
   T-cell activation, and T-cell exhaustion at baseline and over the time
   course of antigen stimulation ([164]online supplemental SFigure 15).
   IL-2RA gene expression was upregulated in C.TNC-CAR.Zip18R T cells at
   both baseline and over the antigen stimulation course, and IFN-γ gene
   expression was elevated at baseline for both donors. We did not see
   significant changes in gene expression for other
   MyD88-signaling-related genes. At baseline, TIGIT (Donor A) and HAVCR2
   (TIM3) (Donor B) gene expression was significantly increased in
   C.TNC-CAR.Zip18R T cells. Post antigen stimulation, this only persisted
   for TIM3 at the 48-hour time point ([165]online supplemental SFigure
   15).

   To confirm our gene expression findings, we used flow cytometry to
   evaluate three different donors after 48 hours of culture in
   cytokine-starved media or co-cultured with LM7 cells. We again found
   that C.TNC-CAR and C.TNC-CAR.Zip18R T cells populations enriched after
   antigen stimulation ([166]figure 5E). Activation markers CD69 and CD28
   were similar across all four T-cell populations at baseline, suggesting
   that Zip18R constitutive signaling does not alter baseline T-cell
   activation. After antigen stimulation, we observed a significant
   increase in CD69-positive C.TNC-CAR and C.TNC-CAR.Zip18R T cells, and
   only a significant increase in CD28-positive C.TNC-CAR T cells
   ([167]figure 5F,G). For TIM3 and CD39, markers associated with T-cell
   exhaustion, we found at baseline that C.TNC-CAR.Zip18R T cells had
   significantly greater populations of TIM3-positive and CD39-positive T
   cells compared with C.TNC-CAR T cells ([168]figure 5H,I). On co-culture
   with LM7 cells, the frequency of TIM3-positive and CD39-positive T
   cells increased in C.TNC-CAR and C.TNC-CAR.Zip18R T cells with no
   significant differences between both T-cell populations. We also
   evaluated lymphocyte activation gene 3 (LAG3), programmed cell death
   protein 1 (PD-1), and cytotoxic T-lymphocyte associated protein 4
   (CTLA-4) expression, which had much lower expression levels
   ([169]online supplemental SFigure 16). C.TNC-CAR and C.TNC-CAR.Zip18R T
   cells followed similar trends, with only C.TNC-CAR T cells having an
   increase in CTLA-4 expression after antigen stimulation, mirroring the
   CD28 expression data.

Zip18R improves the antitumor activity of C.TNC-CAR T cells in vivo

   Having established that Zip18R improves the effector function of
   C.TNC-CAR T cells in vitro, we evaluated if this translates into
   improved antitumor activity in vivo ([170]figure 6A,B). In the
   LM7.GFP.ffLuc model, C.TNC-CAR.Zip18R T-cell infusions were well
   tolerated, and mice continued to gain weight at the same rate as mice
   that received other CAR or NT T-cell populations ([171]online
   supplemental SFigure 17). C.TNC-CAR.Zip18R T cells had significantly
   greater antitumor activity compared with the C.TNC-CAR T cells
   ([172]figure 6C; [173]online supplemental SFigures 18 and 19),
   resulting in a significant survival advantage ([174]figure 6D). The
   gender of mice did not impact the observed benefit in antitumor
   activity of C.TNC-CAR.Zip18R T cells ([175]online supplemental SFigure
   20). The improvement in antitumor activity was dependent on the
   expression of a functional CAR in T cells since Zip18R expression in
   C.TNC.mu-CAR or NT T cells did not improve their antitumor activity.
   Three of the C.TNC-CAR.Zip18R T-cell treated mice ([176]figure 6B) had
   tumor flux values that returned to baseline. On day 136, we
   rechallenged these mice with another i.p. injection of LM7.GFP.ffLuc
   cells to evaluate for functional C.TNC-CAR.Zip18R T-cell persistence
   ([177]figure 6E). All three mice rejected the tumor, while in control,
   untreated mice, LM7.GFP.ffLuc tumor cells grew ([178]figure 6F).

Figure 6. C.TNC-CAR.Zip18R T cells have superior antitumor activity in vivo.
1×10^6 LM7.GFP.ffLuc cells were injected into 5–6-week-old female (Donor 1)
or male (Donor 2) NSG mice. Seven days later, mice received an i.p. injection
of 1×10^6 sorted T cells. (A,B) Tumor flux values measured by IVIS imaging.
(A) Donor 1, (B) Donor 2 (n=5 per cohort per donor). (C) Combined tumor flux
values from both donors at weeks 2 and 5 (n=10, mean+SEM), one-way analysis
of variance, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. (D) Kaplan-Meier
survival curve, Mantel-Cox test. P values listed in the table. (E) Schematic
of rechallenge experiment. 5–6-week-old male mice were used for tumor only
cohort. 1×10^6 LM7.GFP.ffLuc cells were injected i.p. into naïve mice or
C.TNC-CAR.Zip18R T-cell pretreated mice. (F) Tumor flux values measured by
IVIS imaging (n=3–5). CAR, chimeric antigen receptor; C.TNC, tenascin C
encoding the C domain; ffLuc, firefly luciferase; GFP, green fluorescent
protein; i.p., intraperitoneal; NT, non-transduced; Zip18R, interleukin-18
receptor-based leucine zipper receptor.

   [179]Figure 6
   [180]Open in a new tab

   We next evaluated if Zip18R could enhance C.TNC-CAR T-cell antitumor
   activity against DIPG007.YFP.ffLuc i.c. tumors. After 2 weeks post
   T-cell injections, both C.TNC-CAR and C.TNC-CAR.Zip18R T cells had a
   significant reduction in tumor burden as compared with NT-treated mice
   ([181]online supplemental SFigure 21). We found that Zip18R enhanced
   C.TNC-CAR activity, with a significant difference in tumor flux values
   between C.TNC-CAR and C.TNC-CAR.Zip18R T-cell treated mice emerging
   after 3 weeks. However, in this model, 6/15 C.TNC-CAR.Zip18R T-cell
   treated mice needed to be euthanized in the first week post T-cell
   injections despite having low tumor flux values.

Discussion

   Here, we show that the C domain of TNC is expressed by pediatric solid
   and brain tumor cells and is also present within the microenvironment
   of the bulky primary tumor samples. T cells expressing a CAR against
   C.TNC recognized and killed tumor cell lines in an antigen-dependent
   fashion and had limited antitumor activity in vivo. While modification
   of the C.TNC-CAR design did not further improve C.TNC-CAR T-cell
   activity, expressing a novel constitutively active IL-18 receptor
   (Zip18R) in C.TNC-CAR T cells significantly improved their effector
   function, resulting in improved antitumor activity in vivo.

   Identifying suitable antigens for solid tumor and brain tumors for CAR
   T-cell therapy remains a high priority. CAR T cells typically target
   surface proteins on tumor cells, but these antigens may be
   heterogeneously expressed or expressed at low levels.[182]^49 ECM
   proteins provide a unique class of target antigens for cell therapy
   because they are secreted and can bind back to cells within the
   microenvironment, including bystander cells.[183]^50 Additionally,
   tumor and associated stromal cells can generate cancer-specific ECMs by
   differentially expressing ECM proteins or secreting alternatively
   spliced isoforms.[184]^8 Indeed, investigators have leveraged
   tumor-specific ECM proteins as CAR T-cell therapy targets. CAR T cells
   directed to COL11A1, a collagen protein upregulated in breast cancer
   and pediatric OS, had robust antitumor activity in human xenograft
   models.[185]^8 51 The extra domains A and B of fibronectin have also
   been targeted with CAR T cells in preclinical models, demonstrating
   safety and efficacy.[186]4,[187]752 53

   The long isoforms of TNC have been described in adult cancer and
   present alternatively spliced ECM protein targets.[188]^10 38 54 mAbs
   against the A1 domain (F16), A3/A4/B domains (BC-2), A4/B domains
   (ST2485), and C/D domains (81C6) of oncofetal TNC conjugated to either
   IL-2 or radioisotopes have been evaluated in preclinical models,
   demonstrating tumor localization, safety, and antitumor
   activity.[189]^20 In clinical trials, administration of F16-IL-2,
   ^131I-BC-2, or ^131I-81C6 mAb conjugates was safe and associated with
   promising antitumor activity for metastatic breast cancer and
   glioblastoma.[190]55,[191]60 We show here that pediatric solid and
   brain tumors also express the C domain of TNC, and that it is not
   expressed at the RNA level in non-neoplastic, non-diseased tissues.

   We generated T cells expressing a C.TNC-CAR, that consisted of an
   antigen binding domain derived from the G11 mAb and included a CD28ζ
   signaling domain.[192]^38 C.TNC-CAR T cells recognized and killed
   C.TNC-expressing tumor cells. However, C.TNC-CAR T cells had limited
   antitumor activity in xenograft models, and we sought to optimize the
   CAR design, as the CAR structure is important to its function.[193]^61
   62 However, none of the alternative CAR designs improved the antitumor
   activity of C.TNC-CAR T cells. In addition to CARs, several synthetic
   T-cell receptors (TCRs) have been developed, including synthetic TCRs
   and antigen receptor and human leukocyte antigen (HLA)-independent
   T-cell receptors,[194]^63 64 which have a higher sensitivity than
   standard CARs. These receptors could be explored in future studies to
   evaluate if they improve the effector function of C.TNC-CAR T cells.
   Here, we decided to focus on improving C.TNC-CAR T-cell effector
   function via strengthening signal 3.[195]^53

   IL-18 is part of the IL-1 family of cytokines and activates MyD88
   signaling.[196]^65 Several groups have demonstrated that transgenic
   expression of IL-18 or a Fab-based constitutively active IL-18 receptor
   expression bolstered the antitumor activity of CAR T cells, translating
   into increased survival in preclinical tumor models,[197]^34 36 66 and
   early phase clinical testing is in progress ([198]NCT04684563).[199]^67
   However, safety concerns were raised due to weight loss and
   antigen-independent T-cell expansion in preclinical models.[200]^34 35
   66 To remedy this potential toxicity concern, we and other
   investigators have explored activation-dependent IL-18 expression
   either by using activation-dependent promoters or through a GM18
   chimeric cytokine receptor, which signals once GM-CSF is produced by
   activated T cells.[201]^33 35 68

   We first investigated if our GM18 receptor could bolster the antitumor
   activity of C.TNC-CAR T cells. No significant benefit was observed,
   indicating that C.TNC-CAR.GM18 T cells most likely do not produce
   enough GM-CSF to induce robust GM18 triggering. To overcome this
   limitation, we designed a constitutively active cytokine receptor for
   IL-18 based on our modular ZipR platform to create Zip18R. We created
   Zip18Rs with one (1X) or two (2X) pairs of leucine zippers, and both
   activated MyD88 signaling pathways and improved the antitumor activity
   of EphA2-CAR T cells. This is in contrast to JAK/STAT ZipRs, which
   require two pairs of leucine zippers,[202]^37 indicating the leucine
   zippers need to be tailored to their cytoplasmic signaling domain. In
   vitro, Zip18R overall did not improve or decrease the ability of
   C.TNC-CAR T cells to produce cytokines or chemokines after the first
   stimulation. However, there were two notable exceptions:
   C.TNC-CAR.Zip18R T cells produced lower amounts of IL-10 and increased
   amounts of IL-17A and IL-17F, the latter being consistent with
   Zip18R-induced MyD88 signaling.[203]3469,[204]71 After the fourth
   stimulation, type 1 cytokine production was significantly reduced by
   C.TNC-CAR T cells whereas it was maintained by C.TNC-CAR.Zip18R T
   cells. Maintenance of the effector function by Zip18R was also evident
   by a significantly greater expansion of C.TNC-CAR.Zip18R T cells
   compared with C.TNC-CAR T cells in repeat stimulation assays.

   We next evaluated the transcriptional changes that occur in C.TNC-CAR T
   cells when Zip18R is expressed. Zip18R signaling led to significant
   upregulation of IL-17 and cytokine–cytokine receptor pathways at
   baseline and with antigen stimulation, validating our findings from our
   cytokine multiplex analysis. We noted that for one donor, there were
   several immune-related disease pathways upregulated. Further evaluation
   of these disease types shows related pathways such as toll-like
   receptor signaling, Th17 cell differentiation, cytokine–cytokine
   receptor signaling, and leukocyte transendothelial migration, which are
   all hallmarks of T cell and MyD88 signaling. Importantly, we found that
   Zip18R signaling may contribute to a more activated or more exhausted
   T-cell phenotype when compared with C.TNC-CAR T cells at baseline;
   however, differences are not sustained in the setting of antigen
   stimulation. Thus, our data suggests that Zip18R signaling changes
   type1/type2/type17 differentiation, resulting in improved effector
   function.

   We found that C.TNC-CAR.Zip18R T-cell safety in vivo was
   model-dependent. Against LM7.GFP.ffLuc tumors, mice treated with
   C.TNC-CAR.Zip18R T cells did not show overt toxicity as judged by
   clinical behavior and weight measurements, and only 1/10 mice died 5
   weeks post C.TNC-CAR.Zip18R T-cell therapy without high tumor burden or
   clinical signs of graft versus host disease. With three mice showing
   tumor control after rechallenging, this data suggests that
   C.TNC-CAR.Zip18R T cells were able to persist long-term. Tumors
   harvested from mice who reached euthanasia requirement were still
   C.TNC-positive (data not shown). Thus, additional studies are warranted
   to explore if further genetic modifications of C.TNC-CAR.Zip18R T cells
   could improve their efficacy, including deleting epigenetic regulators
   such as DNMT3A or Suv39h1.[205]^72 73

   In contrast to our LM7.GFP.ffLuc studies, C.TNC-CAR.Zip18R T cells were
   toxic in a subset of animals after i.c. injection into
   DIPG007.YFP.ffLuc tumors. Based on the performed IHC analysis, which
   demonstrated TNC expression only at tumor sites within in the brain, on
   target/off cancer toxicity is unlikely. Side effects of T cell-induced
   inflammation within the brain is one likely explanation, which has been
   observed in mice as well as in humans after the locoregional delivery
   of CAR T cells.[206]^74 75 Clearly, future studies are needed to
   understand the underlying mechanism and develop genetic engineering
   approaches to control Zip18R expression in C.TNC-CAR T cells or include
   safety switches to improve the safety of C.TNC-CAR.Zip18R T cells for
   the immunotherapy of brain tumors.

   In summary, our study identifies the C domain of the ECM protein TNC as
   a promising CAR T-cell therapy for pediatric solid tumors and brain
   tumors. C.TNC-CAR T cells expressing a constitutively Zip18R had
   significant antitumor activity. Although our work focuses on developing
   CAR T-cell therapies for pediatric cancers, our work has relevance to a
   broad range of adult cancers, which express C.TNC.

supplementary material

   online supplemental table 1
   [207]jitc-12-11-s001.pdf^ (41.4KB, pdf)
   DOI: 10.1136/jitc-2024-009743
   online supplemental file 1
   [208]jitc-12-11-s002.docx^ (200.2KB, docx)
   DOI: 10.1136/jitc-2024-009743
   online supplemental figure 1
   [209]jitc-12-11-s003.pdf^ (19.5MB, pdf)
   DOI: 10.1136/jitc-2024-009743
   online supplemental figure 2
   [210]jitc-12-11-s004.pdf^ (13.2MB, pdf)
   DOI: 10.1136/jitc-2024-009743

Acknowledgements