Introduction

The use of preclinical animal models remains a pivotal step in cancer immunotherapy development and is instrumental for choosing the compounds and strategies to move forward into the clinical setting. Animal models provide important preliminary information regarding pharmacodynamics, pharmacokinetics, synergistic/additive effects or mechanisms of resistance. However, in the era of cancer immunotherapy, the shortcomings of preclinical research tools remain a major limitation for the successful translation of scientific research breakthroughs, with regulatory approval rates below 10% for anticancer drugs that enter clinical development in phase 1 clinical trials.1 It is widely admitted that syngeneic mouse models hardly recapitulate many of the very essential features of human cancer such as progressive carcinogenesis and heterogeneity. A classical approach to overcoming these limitations has been the use of human tumor cell lines and patient-derived xenografts, which remain a critical tool for tumor cell-targeted drug development.

The advent of immunotherapy has further highlighted the limitations of current preclinical models for anticancer drug development. Human xenografts require an immunodeficient host in order to avoid rejection. Therefore, if immunotherapeutic strategies are to be tested, the murine host has to be engrafted with a competent human immune compartment, to obtain the so-called humanized murine models. A variety of human immune cell types and sources can be used to humanize mice, each with its advantages and disadvantages in terms of feasibility, experiment duration, recapitulation of the distinct immune compartments and tumor cell recognition.2 Mice humanized with CD34+ hematopoietic stem and progenitor cells (HSPCs) better recapitulate the composition of the different human immune cell types2 and show a significantly delayed xenograft-versus-host disease (xGVHD)-like wasting syndrome.3 Admittedly, mice humanized with HSPCs are today the most complete model to study the human immune system in the preclinical setting. However, the absence of human leukocyte antigen (HLA) expression, necessary for the development of HLA-restricted lymphocytes, requires the implantation in the mouse of human HLA-expressing tissues (most frequently liver and thymus)4 5 or HLA transgenes.6 7 These are possibly the most complete strategies for general immunology research, but the complexity associated with obtaining HSPCs, HLA-expressing healthy tissues and a tumor sample from the same patient makes this strategy hardly scalable8 and neoantigen specificity challenging. Furthermore, additional transfer of human genes is often required to obtain myeloid immune cell engraftment.8 9 Tumor-infiltrating lymphocytes (TILs) are an additional interesting source to humanize mice. As an advantage, tumor-specific T lymphocytes are enriched in the tumor microenvironment (TME) compared with peripheral blood,10 11 although their proportions remain low even in immunogenic tumors.12–14 By definition, TILs are autologous to tumor cells and show reduced xGVHD incidence due to tumor-specificity enrichment.15–17 However, obtaining sufficient cell counts for in vivo experiments requires in/ex vivo TIL expansion procedures.15–17

Humanization by peripheral blood mononuclear cells (PBMCs) is the most straightforward and readily available method of humanization with mature cells that can be autologous to tumor cells and contain a proportion of tumor-specific T-cell clones.18–20 Nevertheless, the administration of mature immune cells to immunodeficient mice consistently leads to the development of xGVHD.21

Murine host modification strategies have been developed to overcome this limitation, mainly the silencing of murine major histocompatibility complex (MHC) molecules. Although this strategy effectively delays the development of xGVHD,21 the characterization of human antitumor immune responses and the exploration of the pharmacodynamics of immunotherapeutic strategies remain limited.22–25 In fact, the vast majority of the research published on mouse models humanized with PBMCs is conducted with conventional NSG or NOD/Shi-scid IL2Rgnull (NOG) mice.26–30

Here we have explored alternative strategies to overcome xGVHD as one of the main limitations of humanized murine models engrafted with mature human immune cells. First, we have explored graft modifications both ex vivo and in vivo, to selectively eliminate cell populations responsible for xenoreactive immune responses. Second, we have explored a previously described MHC class I and II double knock-out NSG host (named “MHC-dKO NSG” hereafter), which consistently shows a marked reduction of severe xGVHD, and tested the impact of this modification on the efficacy of clinical-grade immune checkpoint inhibitors (ICIs) and an anti-epithelial cell adhesion molecule (EpCAM)/CD3 T-cell engager (αEpCAM/CD3 bispecific antibody (BsAb)). The significant extension of the survival of mice allowed for long-term antitumor immune response characterization and opened up new opportunities to better understand the pharmacodynamics of clinical-grade and experimental cancer immunotherapies.

MethodsMice

NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ) and MHC-dKO NSG (NOD.Cg-Prkdcscid H2-K1b-tm1Bpe H2-Ab1g7-em1Mvw H2-D1b-tm1Bpe Il2rgtm1Wjl/SzJ) mice were purchased from The Jackson Laboratory (Bar Harbor, Maine, USA) and maintained under specific pathogen-free conditions.

Animal experiments were conducted in accordance with Spanish laws and approval was obtained from the animal experimentation committee of the University of Navarra (reference: 007–22). 6–8-week-old mice were used for in vivo experiments.

Tumor cell lines

The human colon cancer-derived HT29 cell line was purchased from the American Type Culture Collection (ATCC) in 2019. The human lung adenocarcinoma-derived H358 cell line was kindly provided by Professor Luis Montuenga’s group (CIMA Universidad de Navarra, Pamplona, Spain) in 2023, who originally acquired the cell line from the ATCC. A master cell bank was expanded on arrival and a new ampule was thawed every 4 months for experimentation. Cells were cultured in Roswell Park Memorial Institute (RPMI)-1640 supplemented with 10% heat-inactivated fetal bovine serum and 1% penicillin–streptomycin (10,000 U/mL-10,000 μg/mL), all from Life Technologies. The murine 4T1 breast carcinoma cell line of BALB/c origin was provided by Dr Claude Leclerc (Institute Pasteur, Paris, France). This cell line was cultured in complete media containing RPMI-1640 supplemented with 10% heat-inactivated fetal bovine serum, 1% penicillin–streptomycin (10,000 U/mL-10,000 µg/mL) and 2-mercaptoethanol (5×10–5 M), all from Life Technologies. HEK293 cells were obtained from ATCC in 2008. Cells were cultured in RPMI-1640 supplemented with 10% heat-inactivated fetal bovine serum and 1% penicillin–streptomycin (10,000 U/mL-10,000 µg/mL), all from Life Technologies.

PBMCs

Blood samples were obtained from healthy donors following informed consent according to a protocol approved by the institutional ethical committee of the Clinica Universidad de Navarra (reference: 2019–76). Blood was obtained in 10 mL EDTA tubes (BD Vacutainer) and stored at room temperature until processed. PBMCs were obtained by Ficoll gradient (GE Healthcare Biosciences).

For cell depletion experiments, CD4+ cells were depleted following negative magnetic selection according to the manufacturer’s instructions (Miltenyi). Adequate CD4+ T-cell depletion was evaluated by flow cytometry.

Antibodies and other drugs

Cyclophosphamide 1000 mg was purchased from Sigma-Aldrich. The powder was reconstituted in 0.9% saline (Braun) and aliquoted at a concentration of 15 µg/µL. The concentration used for intraperitoneal injection in in vivo experiments was 5 µg/µL.

The following antibodies were developed, produced, and quality controlled at Bristol-Myers Squibb facilities: nivolumab (Opdivo—a fully anti-human programmed death-1 (PD-1) human IgG4) and ipilimumab (Yervoy—a fully human anti-human cytotoxic T-lymphocyte antigen-4 (CTLA-4) IgG1). Control mice in experiments where nivolumab and ipilimumab were tested received drug vehicle (saline).

The anti-EpCAM-CD3 BsAb had the following structure: (IgE signal peptide-(Variable Heavy chain CD3)-(GGGGSGGGGSGGGG)-(Variable Light chain CD3)-(GGGGSGGGGSGGGG)-(Variable Light chain EpCAM)-(RATPSHNSHQVPSAGGPTANSGTSGS)-(Variable Heavy chain EpCAM). The variable region sequences targeting CD3 were derived from the anti-human CD3 monoclonal antibody 12F6.31 The variable region sequences targeting EpCAM were obtained from the MOC-31 monoclonal antibody.32 The control irrelevant vector CD3-B12 was obtained by replacing the VH and VL sequences targeting EpCAM with the corresponding sequences from the B12 antibody, which is a murine IgG1-κ neutralizing antibody against a glycoprotein of HIV-1. This glycoprotein is a non-relevant antigen in murine in vivo models.33 The additional control anti-EpCAM/B12 plasmid was obtained by combining the VH and VL sequences targeting EpCAM with the corresponding sequences of the B12 antibody. The sequences were codon-optimized for Mus musculus and synthesized and cloned by GenScript (Nanjing, China) in a pcDNA3.1 backbone. The sequence of the generated construct was verified by direct sequencing and by restriction enzyme digestion. Plasmid productions were made using EndoFree Maxi Kits from Qiagen following the manufacturer’s instructions.

In vitro experiments

Expression plasmids encoding the BsAbs were transfected and expressed in HEK293 cells with the Lipofectamine 3000 Reagent Kit (Invitrogen). HEK293 cells were seeded at a density of 750,000 cells/well the day prior to transfection. The transfection was carried out according to the manufacturer’s instructions and supernatant was collected after 48 hours of culture. Tumor cells were seeded at a density of 5×104 cells/well for HT29 and 2×104 cells/well for 4T1 cells in 96-well plates. After 24 hours, 100 µL of supernatant from transfected HEK293 cells containing the BsAbs were added to the tumor cells, and incubated for 24 hours. After 24 hours, CD3+ cells were isolated from healthy donor blood samples through Ficoll gradient and subsequent magnetic column separation (LS Columns, Miltenyi Biotec). CD3+ cells (105 per well) were added to the tumor cell culture plates and incubated for 72 hours. After 72 hours of incubation, supernatants were collected for human interferon gamma (hIFN-γ) detection by ELISA.

In vivo experiments

6–8-week-old NSG or MHC-dKO NSG mice were intravenously injected with 107 human fresh PBMCs (retro-orbital injection). HT29 cells (2×106 cells/mouse) were subcutaneously injected into the right flank of mice. H358 cells (106 cells/mouse) were subcutaneously injected into the right flank of mice, embedded in a 1:1 solution of CaCl2-free and MgCl2-free phosphate-buffered saline (PBS) (1×; Gibco) and Growth Factor Reduced Phenol-red free Matrigel Matrix (Corning, Ref. 356231). Therapeutic antibodies and cyclophosphamide were intraperitoneally injected according to each experimental design.

For hydrodynamic injections, a volume of 2 mL of saline dilution containing 10 µg of the assigned plasmid was hydrodynamically injected through the tail vein with a 27 G needle at a rate of 0.4 mL/s.

Plasma samples were obtained from submandibular vein punctures and collected in Eppendorf tubes containing 20 µL of sodium heparin. Plasma samples were stored at −80°C until analysis.

Tumor growth (digital caliper) and body weight loss were measured 3 times per week. Baseline body weight was recorded prior to tumor cell administration. Animals that developed clinical signs of xGVHD (≥20% weight loss, hunched posture, reduced mobility, fur loss, tachypnea) were sacrificed.

ELISA assays

Levels of hIFN-γ in mouse plasma samples were measured by a commercial ELISA (Human IFN-γ Elisa Set, BD OptEIA, BD Biosciences) according to the manufacturer’s instructions. All samples were measured in duplicate. The detection cut-off level of the assay is 4.6875 pg/mL.

Murine alanine aminotransferase measurements

Plasma murine alanine aminotransferase (mALT) levels were analyzed using a Roche Cobas C311 analyzer (Roche Diagnostics, Indianapolis, Indiana, USA).

Histology

Tumors and livers were formalin-fixed (3.7–4%, PanReac AppliChem, ITW Reagents) for 72 hours and maintained in 70% ethanol, dehydrated, and paraffin-embedded according to standard protocols conducted at the Morphology Core Facility at CIMA. Five-micrometer sections were stained in H&E. For CD3 immunohistochemistry, a rabbit anti-human CD3 antibody (clone SP7—Epredia RM-9107-S Thermo Fisher Scientific) was used. Image acquisition was performed on an Aperio CS2 slide scanner and processed with the ScanScope software (Leica Biosystems).

Multiplex immunofluorescence staining and analysis

Multiplex immunofluorescence staining was performed on a Bond RX autostainer (Leica Biosystems), as previously described.34 35 Briefly, 4 μm-thick formalin-fixed, paraffin-embedded (FFPE) tissue sections were deparaffinized (Bond DeWax, Leica Biosystems) and rehydrated per standard protocols. Antigen retrieval was performed with BOND Epitope Retrieval Solution 1 (ER1, Leica Biosystems) or 2 (ER2, Leica Biosystems, product number AR9640), followed by sequential cycles of staining with each cycle including a 30 min combined block and primary antibody incubation (Akoya antibody diluent/block), followed by a secondary horseradish peroxidase- (HRP)-conjugated polymer. Signal amplification was achieved with TSA-Opal fluorophores. Between cycles of staining, tissue sections underwent heat-induced epitope retrieval to remove the primary/secondary-HRP antibody complexes before staining with the next antibody. Two multiplex immunofluorescence panels were performed. Primary antibodies and corresponding fluorophores were:

Panel 1: anti-CD3 (rabbit polyclonal, IgG, ready-to-use, Agilent, product number IR503) in Opal 520; anti-CD8 (mouse monoclonal, clone C8/144B, ready-to-use, Agilent, product number GA62361-2) in Opal 480; and anti-cytokeratin (mouse monoclonal, clone AE1/AE3, ready-to-use, Leica Biosystems, product number NCL- L-AE1/AE) in Opal 780. Panel 2: anti-CD137 (TNFRSF9 or 4-1BB, mouse monoclonal, clone BBK-2, 1:80, Thermo Fisher, product number MA5-13736) in Opal 520; anti-CD8 (mouse monoclonal, clone C8/144B, ready-to-use, Agilent, product number GA62361-2) in Opal 480; and anti-cytokeratin (mouse monoclonal, clone AE1/AE3, ready-to-use, Leica Biosystems, product number NCL-L-AE1/AE) in Opal 780. Nuclei were counterstained with Spectral DAPI (Akoya Biosciences, FP1490). Stained slides were then mounted in a ProLong Diamond Antifade mounting medium (Thermo Fisher Scientific). Slides were further scanned using the PhenoImager HT Automated Quantitative Pathology Imaging System (Akoya Biosciences). After image acquisition, unmixing of the spectral libraries was performed with inForm software (Akoya Biosciences). Unmixed images were then imported into the open-source digital pathology software QuPath V.0.4.4 for stitching, cell segmentation and cell phenotyping. A supervised machine learning algorithm trained on expert-provided examples was used for classifying cells as: CD3+, CD8+, CD137+, and tumorous (AE1/AE3+). CD4+ T cells were defined as CD3+CD8–. The densities of each cell population were quantified and expressed as the number of cells per mm2.

Flow cytometry

Liver and tumor tissue were harvested for flow cytometry analysis. Tumors were minced and digested with 400 U/mL collagenase D and 50 µg/mL DNase-I (Roche) for 20 min at 37°C. Tissue digestion was stopped by adding 12.5 µL of 0.5 M EDTA (Invitrogen) to each sample. Single-cell suspensions were obtained by passing samples through 70 µm cell strainers (Falcon). For peripheral blood staining, whole blood (125–150 µL/sample, depending on availability) was centrifuged and washed in PBS twice (5,000 rcf, 5 min). Afterwards, corresponding surface-staining antibody mixes were added to each sample and incubated for 10 min at 4°C. For red cell lysis, FACS Lysing Solution 10× (BD Biosciences, Catalog #349202, diluted to 1× in distilled water) was added to each sample for 5 min. Samples were then centrifuged at 5,000 rcf for 1 min and cleaned in sorting buffer (PBS 1× with 0.5% fetal bovine serum(FBS), 0.5% of 0.5 M EDTA and 1% of penicillin/streptomycin 104 U/µg/mL (Gibco)) prior to intracellular protein staining.

Samples were treated with 10 mg/mL of beriglobin (CSL Behring, Marburg, Germany) and surface stained with suitable combinations of the following fluorochrome-labeled antibodies: mouse anti-hCD3 (Clone: UCHT1)-AF647 (BioLegend), mouse anti-hCD3 (Clone: UCHT1)-AF488 (BioLegend), mouse anti-hCD8 (Clone: SK-1)-BV510 (BioLegend), mouse anti-hCD4 (Clone: OKT-4)-BV650 (BioLegend), mouse anti-hCD4 (Clone: RPA-T4)-BV605 (BD Biosciences), mouse anti-hCD14 (Clone: M5E2)-PECy7 (BioLegend), mouse anti-hCD16 (Clone: 3G8)-BV510, (BioLegend), mouse anti-hCD19 (Clone: HIB19)-BV650 (BioLegend), mouse anti-hCD45 (Clone: HI30)-PB (BioLegend), mouse anti-hCD45RO (Clone: UCHL1)-PECy7 (BD Biosciences), mouse anti-hCD56 (Clone: HCD56)-PE/Dazzle594 (BioLegend), mouse anti-hPD-1 (Clone: EH12.2H7)-PerCP-Cy5.5, mouse anti-hCCR7 (Clone: 3D12)-AF647 (BD Biosciences), mouse anti-hPD-1 (Clone: EH12-2H7)-PECy7 (BioLegend), mouse anti-hCD25 (Clone: BC96)-BV421 (BioLegend), mouse anti-hCD137 (Clone: 4B4-1)-PECy7 (BioLegend), mouse anti-hCD137 (Clone: 4B4-1)-BV421 (BioLegend), mouse anti-human granzyme B (Clone: GB11)-PE (BD Biosciences), mouse anti-hKi67 (Clone: Ki67)-AF488 (BioLegend), mouse anti-hFoxP3 (Clone: PCH101)-PE (Invitrogen), mouse anti-hEpCAM (Clone: 9C4)-PerCP-Cy5.5 (BioLegend) and rat anti-mEpCAM (Clone: G8.8)-PE (BioLegend). Cell viability was analyzed with Zombie-NIR Dye (1:1,000 dilution, BioLegend) or Live/Dead Near IR 876 nm marker (1:1,000 dilution, Invitrogen). The CytoFLEX S and CytoFLEX LX cytometers were used for data acquisition and the CytExpert V.2.5 software was used for data analysis.

Reverse transcription-quantitative PCR

Total RNA was obtained with a Maxwell RSC simply RNA Tissue kit, following the manufacturer’s instructions. Subsequently, a specific reverse transcription (RT) reaction was performed for the generation of complementary DNA (cDNA). RNA was pretreated with DNAse I using the TURBO DNAfree Kit (Ambion, #1908) and RT was carried out from 500 ng of total RNA in a final volume of 20 µL. To carry out the reaction, a mix with 10 mM dNTPs, 250 ng/mL random primers, 40 U/µL of RNase Inhibitor, 1× Buffer and 200 U/µL of M-MLV Reverse Transcriptase (Invitrogen, #28025013) were prepared. After adding the mix, samples were incubated for 1 hour at 37°C and 1 min at 95°C. Then, each cDNA was diluted with nuclease-free water to a final concentration of 50 ng/4.6 µL.

To perform the quantitative PCR (qPCR), the GoTaq qPCR Master Mix was used with 50 ng of cDNA. qPCR was carried out using the CFX96 Real-Time Detection System (BioRad). The results were normalized to the human glyceraldehyde-3-phosphate dehydrogenase (hGAPDH) gene transcripts for each sample. The relative value for the expression level of hIFN-γ was calculated by the equation Y=2−∆Ct, where Ct is the point at which the fluorescence rises significantly above baseline, and ∆Ct is the difference between control and target products (∆Ct=Ct hINF-γ – Ct GAPDH). Forward and reverse primer sequences for hIFN-γ were 5’-CTCTGCATCGTTTTGGGTTC-3’ and 5’-GCGTTGGACATTCAAGTCAG-3’, respectively. Forward and reverse primer sequences for hGAPDH were 5’- GGTCGGAGTCAACGGATTT −3’ and 5’- CCAGCATCGCCCACTTGA-3’, respectively.

Statistical analysis

Statistical analyses were performed with the Prism V.8.0.2 version software (GraphPad). Continuous unpaired variables were analyzed with the Mann-Whitney test. Continuous paired variables were analyzed with the Wilcoxon matched-pairs signed-rank test. Correlations between continuous variables were calculated with the Spearman test and linear regression lines were represented on correlation graphs. Survival was described following the Kaplan-Meier method and analyzed between groups with the log-rank (Mantel-Cox) test. Event-free survival analyses for xGVHD included ≥20% weight loss or death from any cause (except humane sacrifice due to excessive tumor burden) preceded by a ≥10% weight loss as events. A two-tailed p<0.05 was considered as statistically significant. Tumor growth differences were analyzed by two-way analysis of variance tests. When differences are statistically significant, the significance is represented with asterisks (*): *for a p value<0.05, **for a p value<0.01, ***for a p value<0.001, ****for a p value<0.0001. Error bars indicate the SEM unless otherwise indicated. In grouped body weight loss and tumor growth graphs, curves were not further represented when ≥50% of the individuals within a group were missing to avoid misleading representations.

Patients and public involvement

Patients were not involved in the conduction of the experiments presented in this document.

ResultsCD4+ T-cell depletion from the PBMC inoculum significantly reduces xGHVD but also abrogates antitumor activity

Previous data from our group explored the influence of diverse ex vivo graft modifications on xGVHD-associated weight loss. We observed that mice engrafted with PBMCs devoid of CD4+ T cells and treated with a combination of nivolumab and urelumab did not develop significant xGVHD as compared with their counterparts humanized with total PBMCs.36 This finding has been confirmed by others.37 In keeping with these notions, we explored the effect of human total PBMCs or PBMCs devoid of CD4+ T cells once engrafted in NSG mice bearing tumors derived from a human colon cancer cell line (HT29) (figure 1A). CD4+ T-cell depletion was carried out by negative magnetic selection and verified by flow cytometry (online supplemental figure 1A). Results confirmed that xGVHD development was highly dependent on the presence of CD4+ T cells, both in terms of body weight loss (figure 1B,C) and liver toxicity (figure 1D). Proportionally, the antitumor effect mediated by PBMCs was completely lost in the absence of human CD4+ T cells (figure 1E,F). As a measure of human immune cell activation, hIFN-γ was assessed in the plasma of humanized mice. Plasmatic hIFN-γ was clearly observed in mice engrafted with total PBMCs, but was significantly lower in mice engrafted with PBMCs devoid of CD4+ T cells over time (figure 1G). These results indicate that in these humanized experimental conditions, human CD4+ T cells are critical for xGVHD, but also for antitumor activity.

Figure 1

Ex vivo graft modification by CD4+ T-cell depletion significantly attenuates xGVHD but also abrogates antitumor activity. (A) Schematic representation of the experiment. (B) Individual relative body weight loss curves by experimental groups. The dashed line at −20% represents the threshold for a humane endpoint. (C) EFS analysis by experimental groups. At day +54 all mice had reached the endpoint criteria for sacrifice due to tumor burden or body weight loss. (D) mALT plasma levels throughout the experiment. The dashed line represents the upper limit of normality (50 U/L). (E) Individual tumor growth curves. The dashed line at 15 mm represents the tumor burden limit for a humane endpoint. (F) Tumor growth curves of the experimental groups. The dashed line at 15 mm represents the tumor burden limit for a humane endpoint. (G) hIFN-γ plasma concentrations throughout the experiment. Body weight ∆, relative percentage of body weight loss; EFS, ≥20% body weight loss event-free survival; hIFN-γ, human interferon gamma; i.v., intravenous; mALT, mouse alanine aminotransferase; PBMCs∆CD4, CD4-depleted peripheral blood mononuclear cells; s.c., subcutaneous; xGVHD, xenograft-versus-host disease; Φ, diameter.

Post-transplantation cyclophosphamide does not significantly reduce xGVHD and curbs antitumor activity

Recent data have confirmed in clinical practice that high-dose cyclophosphamide administration a few days after allogeneic bone marrow transplantation significantly reduces GVHD incidence, affecting preferentially the most proliferative and alloreactive T-cell clones.38 39 Following this experience, we decided to explore if intraperitoneal cyclophosphamide administration post-PBMC engraftment was able to selectively deplete xenoreactive T-cell clones thereby limiting xGVHD and sparing the antitumor effect. First, we identified a 50 mg/kg cyclophosphamide dose as tolerable and active to modify xGVHD (data not shown). Second, we explored different schedules of intraperitoneal cyclophosphamide administration: day +3, +5 or +7 post-PBMC engraftment (figure 2A). We injected the human HT29 colorectal carcinoma cell line subcutaneously on day +8. Previously, we tested that cyclophosphamide administration 24 hours prior to tumor inoculation did not show a significant impact on tumor growth (data not shown). Results confirmed that early cyclophosphamide administration (day +3) significantly reduced xGVHD-associated body weight loss (figure 2B,C) and led to a numerically lower incidence of plasma mALT elevations over the upper limit of normality (online supplemental figure 2A). However, this schema also abrogated the antitumor effect of PBMCs (figure 2D,E). Intermediate-term cyclophosphamide (day +5) preserved antitumor efficacy but did not significantly delay xGVHD-associated body weight loss (figure 2B–E). Finally, late cyclophosphamide administration (day +7) did not show any significant effect on weight loss or tumor growth control, suggesting that the proliferation peak of all T-cell clones occurs earlier (figure 2B–E). Consistently, plasma hIFN-γ levels showed a significant drop in mice treated with early cyclophosphamide (day+3), but no significant impact at the other two time points (figure 2F). These data suggest that even though xGVHD and antitumor effects appear to be differentially affected by cyclophosphamide, the window of opportunity to explore antitumor activity is too narrow to be exploited experimentally.

Figure 2

Post-transplantation cyclophosphamide does not significantly delay xenograft-versus-host disease and curbs antitumor activity. (A) Schematic representation of the experiment. (B) Individual relative body weight loss curves. The dashed line at −20% represents the threshold for a humane endpoint. (C) Relative body weight loss of the experimental groups. The dashed line at −20% represents the threshold for a humane endpoint. Statistical significances of comparisons with the HT29 control group are not shown for simplification. (D) Individual tumor growth curves. The dashed line at 15 mm represents the tumor burden limit for a humane endpoint. (E) Tumor growth curves of the experimental groups. The dashed line at 15 mm represents the tumor burden limit for a humane endpoint. Statistical significances of comparisons with the HT29 control group are not shown for simplification. (F) hIFN-γ plasma levels throughout the experiment. Body weight ∆, relative percentage of body weight loss; Cy, cyclophosphamide; D3, day +3 of the experiment; D5, day +5 of the experiment; D7, day +7 of the experiment; hIFN-γ, human interferon gamma; i.p., intraperitoneal; i.v., intravenous; PBMC, peripheral blood mononuclear cell; s.c., subcutaneous; Φ, diameter.

Severe xGVHD is abrogated in MHC-dKO NSG mice ((KbDb)null (IA)null), while the antitumor effect is preserved

It has been previously described that xGVHD is preferentially mediated by major mismatches between transferred human T-cell receptors (TCR) and murine H-2 molecules.40 Furthermore, it has been observed that murine MHC modifications significantly influence xGVHD incidence and severity in mice engrafted with human PBMCs.21 However, the impact of these modifications on the antitumor immune properties in this model has been poorly addressed. We first evaluated the impact of silencing MHC class I and II on the engraftment of human PBMCs. Both strains of mice, NSG and MHC-dKO NSG, were co-engrafted with human HT29 carcinoma cells and human PBMCs. Human PBMCs were analyzed longitudinally over time in peripheral blood and in the tumor at the end of the experiment (online supplemental figure 3A). As previously described,21 we observed a rapid shift toward a T-cell enriched composition in peripheral blood (online supplemental figure 3B). From day +8, there was a marked peripheral blood human CD45+ cell expansion, in particular in NSG mice (online supplemental figure 3C), although no major immune cell type composition differences were observed between the two mouse strains in blood or tumor tissue (online supplemental figure 3D–G), with the exception of a higher abundance of circulating regulatory T cells in NSG mice (online supplemental figure 3H). Interestingly, and consistently with a previous report,21 we observed that circulating naive (CCR7+CD45RO–) and effector (CCR7–CD45RO–) CD8+ T cells persisted longer in MHC-dKO NSG mice as compared with NSG counterparts (online supplemental figure 3I–K). We then tested the impact of silencing MHC class I and II on both xGVHD and tumor growth control, engrafting human PBMCs in NSG and MHC-dKO NSG mice. Both groups of mice were co-engrafted with human HT29 carcinoma cells and human PBMCs (figure 3A). NSG mice showed significant weight loss (figure 3B,C) and significant plasma mALT increase compared with controls with no PBMCs as early as day+2, and more remarkably on day+22 (figure 3D), both indicators of xGVHD. In contrast, MHC-dKO NSG mice did not show weight loss (figure 3B,C), or any significant changes in plasma mALT levels on day +2 or +22 (figure 3D). Antitumor activity was observed in both groups, NSG and MHC-dKO NSG mice engrafted with PBMCs, when compared with control groups (figure 3E,F). Nevertheless, the duration and characterization of responses in NSG mice were limited due to xGVHD (figure 3E,F), while in the MHC-dKO NSG group, three out of eight mice achieved complete responses which were maintained in two of them at day +53 (figure 3E,F). Consistently, detectable plasma hIFN-γ levels were observed in both NSG and MHC-dKO NSG mice, but NSG individuals showed on average fivefold higher concentrations than MHC-dKO NSG mice (figure 3G). These data indicate that MHC class I and II gene deletions uncouple antitumor efficacy and xGVHD toxicity phenomena.

Figure 3

Xenograft-versus-host disease is significantly attenuated in MHC I/II-dKO NSG mice, while the antitumor effect of PBMCs is preserved. (A) Schematic representation of the experiment. (B) Individual relative body weight loss curves. The dashed line at −20% represents the threshold for a humane endpoint. (C) Relative body weight loss curves by experimental groups. The dashed line at −20% represents the threshold for a humane endpoint. (D) mALT plasma levels throughout the experiment. The dashed line represents the upper limit of normality (50 U/L). (E) Individual tumor growth curves. The dashed line at 15 mm represents the tumor burden limit for a humane endpoint. (F) Tumor growth curves of the experimental groups. The dashed line at 15 mm represents the tumor burden limit for a humane endpoint. (G) hIFN-γ plasma levels throughout the experiment. Both NSG and dKO mice with no PBMC administration showed negligible plasma hIFN-γ levels (mean of 5.6 and 12.1 pg/mL, respectively) and were not represented for simplification. Body weight ∆, relative percentage of body weight loss; CR, complete response rate; dKO, MHC-dKO NSG mice; hIFN-γ, human interferon gamma; mALT, mouse alanine aminotransferase; MHC, major histocompatibility complex; PBMC, peripheral blood mononuclear cells; s.c., subcutaneous; Φ, diameter; .

T-cell density and activation are attenuated in MHC-dKO NSG mice livers, while the tumor immune infiltrate remains comparable

We explored whether MHC class I and II silencing produced significant changes in terms of immune infiltration of the liver (as an xGVHD target organ) and of tumors. Following an experimental design as in figure 3, we again observed significant body weight decrease in NSG mice engrafted with human PBMCs starting on day +20 (online supplemental figure 4A). We then decided to sacrifice mice on day +25 to analyze liver and tumor immune infiltrates (figure 4A). Liver human CD3+ (figure 4B–D), CD8+ (figure 4E) and CD4+ (figure 4F) T-cell infiltration was significantly higher in NSG mice compared with MHC-dKO NSG mice. Additionally, both CD8+ (figure 4G,H) and CD4+ (figure 4I,J) T cells showed a more intense expression of the cytotoxic marker granzyme b. Accordingly, plasma hIFN-γ levels were significantly higher in NSG mice (online supplemental figure 4B). In contrast, tumor human CD3+ (figure 4B,C,K), CD8+ (figure 4L) and CD4+ (figure 4M) T-cell infiltration was comparable in NSG and MHC-dKO NSG groups. Activation markers in TILs did not show consistent differences between NSG and MHC-dKO NSG groups (online supplemental figure 4C–H), with the exception of CD137, which was significantly higher in tumor-infiltrating CD8+ T cells in NSG mice (online supplemental figure 4I,J). Further exploration of the tumor microenvironment showed a higher proportion of CD4+CD25+ cells in tumors compared with livers, in both NSG and MHC-dKO NSG mice (online supplemental figure 4K,L). Overall, these data show that MHC silencing in the NSG mouse model is associated with a significant decrease in liver inflammation, while the abundance and overall phenotypic characteristics of TILs are preserved.

Figure 4

Liver as a target organ of xenograft-versus-host disease showed significantly higher human T-cell density and activation in NSG mice as compared with MHC-dKO NSG mice, while the tumor immune infiltrate remains comparable. (A) Schematic representation of the experiment. (B) Representative CD3 immunohistochemistry and H&E microphotographs of liver and tumor sections from NSG and MHC-dKO NSG mice. The bar in low-power magnification images represents 2 mm except for the “Tumor NSG” sample, where it represents 600 µm. Dashed squares represent the area amplified in high-power magnification images. The bar in high-power magnification images represents 200 µm (C). Percentage of CD3+ T cells among parental live cell events in cell suspensions from livers and tumors from NSG and MHC-dKO NSG mice. The gate for CD3+ cells with their proportion among total live events is shown. (D) CD3+ T cells in livers (proportions of live events and counts per mg of tissue). (E) CD8+ T cells in livers (proportions of live events and counts per mg of tissue). (F) CD4+ T cells in livers (proportions of live events and counts per mg of tissue). (G) GzmB expression on CD8+ cells in livers (proportions and MFIs). (H) Representative histograms of GzmB expression on CD8+ T cells in liver samples. The vertical line in the histograms represents the positivity threshold established with the FMO control. (I) GzmB expression on CD4+ cells in livers (proportions and MFIs). (J) Representative histograms of GzmB expression on CD4+ T cells in liver samples. The vertical line in the histograms represents the positivity threshold established with the FMO control. (K) CD3+ T cells in tumors (proportions of live events and counts per mg of tissue). (L) CD8+ T cells in tumors (proportions of live events and counts per mg of tissue). (M) CD4+ T cells in tumors (proportions of live events and counts per mg of tissue). dKO, MHC-dKO NSG; FMO, fluorescence minus one; FSC, forward scatter; GzmB, granzyme B; i.v., intravenous; MFI, mean fluorescence intensity; MHC, major histocompatibility complex; PBMC, peripheral blood mononuclear cell; s.c., subcutaneous.

The humanized MHC-dKO NSG model enables the antitumor activity characterization of clinical-grade immune checkpoint inhibitors nivolumab plus ipilimumab

Despite the advent and success of the so-called ICIs in the clinic, we are largely missing useful biomarkers guided by a precise understanding of the mechanism of action of these agents.41 One important limitation is the absence of in vivo models where these agents can be tested and characterized. We decided to explore the antitumor effect of clinical-grade immunotherapy in our models testing the combination of nivolumab (anti-PD-1 monoclonal antibody (mAb)) and ipilimumab (anti-CTLA-4 mAb) (figure 5A). We observed that the combined immunotherapy significantly accelerated xGVHD in NSG mice engrafted with human PBMCs in terms of body weight loss (figure 5B). It also produced a tendency toward higher plasma mALT levels (figure 5C). This precluded the characterization of antitumor responses and any long-term follow-up in this group (online supplemental figure 5A,B). In contrast, early events of weight loss ≥20% were not observed in MHC-dKO NSG mice regardless of immunotherapy administration, enabling a longer follow-up (figure 5B). MHC-dKO NSG mice treated with ipilimumab plus nivolumab showed a non-significant plasma mALT elevation at day +15 of the experiment (figure 5C), but this hypertransaminasemia did not have any impact on body weight loss or survival in this group. Of note, mALT elevation was significantly higher in NSG mice as compared with MHC-dKO NSG mice treated with immunotherapy (figure 5C). PBMCs significantly controlled tumor growth in both MHC-dKO NSG (figure 5D,E) and NSG (online supplemental figure 5A,B) mice. This effect was significantly improved with nivolumab and ipilimumab in MHC dKO NSG mice (figure 5D,E), but was not evaluable due to xGVHD in NSG mice (online supplemental figure 5A,B). Plasma hIFN-γ levels showed significantly higher levels in mice treated with nivolumab plus ipilimumab when compared with control groups (figure 5F; online supplemental figure 5C). Similar results were observed using the human H358 lung adenocarcinoma cell line (online supplemental figure 6). Overall, MHC-dKO NSG mice enabled the observation of profound and durable antitumor immune responses to nivolumab plus ipilimumab that were not assessable in NSG mice due to accelerated xGVHD.