A Highly Clinical Relevant Graft-vs-Leukemia Model in Humanized Mice

Abstract

The prognosis for acute myeloid leukemia (AML) relapse post allogeneic hematopoietic stem cell transplantation (alloSCT) is dismal. Novel effective treatment is urgently needed. Clinical benefit of alloSCT greatly relies on the graft-vs-leukemia (GVL) effect. The mechanisms that mediate immune escape of leukemia (thus causing GVL failure) remain poorly understood. Studies of human GVL have been hindered by the lack of optimal clinically relevant models. Here, using our large, longitudinal clinical tissue bank that include AML cells and G-CSF mobilized donor hematopoietic stem cells (HSCs), we successfully established a novel GVL model in humanized mice. Donor HSCs were injected into immune-deficient NSG mice to build humanized mice. Immune reconstitution in these mice recapitulated clinical scenario in the patient who received the corresponding HSCs. Allogeneic but HLA partially matched patient-derived AML cells were successfully engrafted in these humanized mice. Importantly, we observed a significantly reduced (yet incomplete elimination of) leukemia growth in humanized mice compared to that in control NSG mice, demonstrating a functional (but defective) GVL effect. Thus, for the first time, we established a novel humanized mouse model that can be used for studying human GVL responses against human AML cells in vivo. This highly clinically relevant model provides a valuable platform for investigating the mechanisms of human GVL, opening a new avenue for development of personalized and effective leukemia treatments

Introduction

Clinical benefit of allogeneic hematopoietic stem cell transplantation (alloSCT) greatly relies on the graft-vs-leukemia (GVL) effect(1–3). Despite considerable efforts, acute myeloid leukemia (AML) relapse remains the top cause of post-transplant death(4). The mechanisms that mediate immune escape of leukemia during alloSCT (thus causing GVL failure) remain poorly understood. Studies in preclinical models are the key to define the mechanisms and discover new treatment targets. However, lack of optimal clinically relevant GVL models is a major gap that precludes the move of novel treatments toward clinical applications.

Current experimental models to study GVL are mostly built by engrafting murine AML cells, either from cultured leukemia cell lines or derived from murine bone marrow cells transduced with leukemia-driven genes, into mouse models of alloSCT. Studies using these models have yielded important discoveries and significantly improved our knowledge of alloimmunity(5–10). However, these models are limited in the translation of discoveries to clinical applications as many genetic and biologic features are different between humans and mice(11–13) . In addition, murine leukemia cell lines cause rapid progression of AML in these experimental mice and may not closely reflect AML progression in patients undergoing alloSCT. Furthermore, these models are not useful to test human-specific agents.

A highly clinical relevant GVL model requires a model system containing human donor-derived immune system and patient-derived leukemia. Immune deficient NOD.Cg-Prkdc^scidIL2rg^tm1Wjl/SzJ (NSG) mice allow engraftment of human tissues, providing a useful tool to investigate human-specific disease and patient-specific drug sensitivity (14–16). Humanized mice can be generated by injection of human hematopoietic stem cells (HSCs) into NSG mice in which a human hematopoietic system (including the immune system) is established (14, 17). We reason that if humanized mice is built by engrafting the HSCs derived from clinical donors of alloSCT, further injection of patient-derived AML cells in these mice may enable interactions between the donor immune system and patient-derived AML, therefore establish a clinical relevant GVL model.

We have developed a large, longitudinal clinical tissue bank that include AML cells and G-CSF mobilized donor HSCs. In this study, we aim to establish a novel humanized GVL model by engrafting human donor HSCs and HLA partially matched patient-derived AML cells into NSG mice.

Materials and Methods

Patients

Peripheral blood samples were collected from AML patients and patients underwent allogeneic hematopoietic stem cell transplantation (alloSCT). Patients were diagnosed and treated at the Penn State Cancer Institute, Penn State University College of Medicine, Hershey, PA. The study was approved by the Institutional Review Board of Penn State University College of Medicine. Full written informed consent was obtained from all patients.

Mice

NOD-Cg-Prkdc^scidIL2rg^tm1Wjl/SzJ (NSG) mice were obtained from The Jackson Laboratory (Sacramento, CA, USA). Humanized NSG mice (hu-NSG) were generated by transplanting CD34⁺ hematopoietic stem cells (HSCs, 5-10×10⁵ cells/mouse) that were obtained from donor G-CSF mobilized peripheral blood into 3-week old NSG mice via intravenous injection. CD34⁺ HSCs were purified by magnetic beads (STEMCELL Technologies ). Mice received 140cGy whole body sublethal irradiation before the injections. The engraftment was evaluated 4, 8 and 12 weeks post transplantation by assessing the levels of human CD45⁺ cells in the peripheral blood of these mice. NSG mice successfully engrafted with human hematopoietic cells were termed hu-NSG mice. To produce hu-NSG mice bearing AML, we transferred patient-derived AML cells into hu-NSG mice 12 weeks post HSC transplantation via intra-femoral injection (Characteristics of AML patients were shown in Supplemental Table1). The study was approved by the Institutional Animal Care and Use Committee at Penn State University College of Medicine, according to National Institutes of Health guidelines for animal use.

HLA genotyping

HLA-A, B, C, and DRB1 were typed at high resolution level by next generation sequencing (ALLType kit, One Lambda, West Hills, CA) or the combination of sequence-specific oligonucleotide probe (LabType kit, One Lambda) and Sanger sequence-based typing kit (SBTexcellerator kit, GenDx, Utrecht, Netherlands).

Immunofluorescence staining and flow cytometry analysis

For cell surface molecules staining, cells were incubated with Fixable Viability Dye (eF450 or eF506, eBioscience) to assess the viability, and then followed by staining with directly conjugated mAbs for 30 min at 4 °C. The mAbs were mCD45-APC/BV510 (clone: 30-F11), hCD3-BV786/BV605 (clone: SK7), hCD4-BV711 (clone: L200), hCD8-APC-H7 (clone: SK1), hCD56-FITC/PE-CF594 (clone: B159), hCD14-BV711 (clone: MφP9), hCD11b-AF700 (clone: ICRF44), hCD45-BV786 (clone: HI30), hCD19-FITC (clone: HIB19 and SJ25C1), hCD20-FITC (clone: 2H7 and L27), hCD45RA-AF700 (clone: HI100), hCD28-PE (clone: CD28.2), hCD226-FITC (clone: DX11), hOX40-PE (clone: ik1), hHLA-DR-Percp-cy5.5/PE-cy7 (clone: G46-6), hBTLA-PE-CF594 (clone: J-168-540), hCD38-APC/PE (clone: HIT2), Hicos-AF488 (clone: DX29), h4-1BB-APC (clone: 4B4-1), hCD16-BV421 (clone: 3G8), hCD57-PE-CF594 (clone: NK-1), hKIR-FITC (clone: HP-3E4), hCD38 (clone: HIT2; BD Bioscience), hNKG2A-AF700 (clone: S19004C; Biolegend), and hCCR7-BV421 (clone: G043H7; eBioscience). For intracellular staining, anti-IFN-γ-APC (clone: B27), TNF-α-FITC (clone: MAb11), Granzyme B-AF700 (clone: GB11), Perforin-APC (clone: δG9), and Ki-67-AF488 (clone: B56) (BD Bioscience) were used. T-distributed stochastic neighbor embedding (tSNE) analysis was performed using algorithm from FlowJo software. Major immune cell populations (T cells, B cells, NK cells, granulocytes and monocytes) were manually annotated based on their classical markers.

Wright-Giemsa staining

Bone marrow cells were collected from humanized mice baring AML at week 24 after AML cell engraftment. 4×10⁵ bone marrow cells were resuspended in 200μ1 PBS and spun to the slides using cytospin (Thermo Scientific Cytospin 4 Cytocentrifuge). After the slides were air dried, Wright-Giemsa staining buffer (VWR) was applied for 1 min and then slides were washed with DI water for 2 minutes and mounted with coverslips.

Statistical analysis

GraphPad Prism7.00 (GraphPad Software, La Jolla,CA, USA) was used for statistical calculations. The normality of each continuous variable was evaluated using the Kolmogorov–Smirnov test. For data distributed normally, the comparison of variables was performed using unpaired or paired Student’s t test. For data not distributed normally, the comparison of variables was performed with a Mann–Whitney U test or a Wilcoxon signed-rank test for unpaired and paired data, respectively. All tests are two-tailed with P values less than 0.05 considered statistically significant.

Results

Model system for generating a novel humanized GVL model.

The schema for the generation of a humanized GVL model is illustrated in Figure 1. We first established humanized mice by injecting human CD34⁺HSCs into sub-lethally irradiated NSG mice. CD34⁺ HSCs were purified from G-CSF-mobilized peripheral blood of transplant donors. After 12 weeks transplantation, when engraftment of human hematopoietic cells and reconstitution of human immune system were achieved, leukemia cells derived from AML patients were injected to generate GVL model in humanized mice. To mimic clinical alloSCT, we selected leukemia cells derived from AML patients who were at least half HLA matched with the HSC donors (Fig. 1).

Figure 1. The schema for generation of humanized GVL model.

The 3-week NSG mice received 140cGy sublethal irradiation and then were transplanted with CD34⁺ hematopoietic stem cells (HSCs) via tail intravenous injection. HSCs were isolated from donor G-CSF mobilized peripheral blood using magnetic beads. The level of human CD45⁺ cells in peripheral blood was evaluated at 4, 8 and 12 weeks post transplantation by flow cytometry. At week 12, HLA partially matched leukemia cells derived from AML patients were injected via intrafemoral injection (20μl cells per mouse, mice received 140gGy sublethal irradiation before injection) to generate humanized GVL model. Peripheral blood samples were collected and evaluated at the next 4, 8 and 12 weeks. At week 12 (week 24 post HSCs transplantation), blood, bone marrow, spleen, liver and lungs were evaluated.

G-CSF-mobilized transplant donor HSCs successfully built human immune system in NSG mice

Two pairs of transplant donor HSCs and patients-derived AML cells were used in our study. The demographic characters of the AML patients and their HLA types related to the donors are listed in Supplemental Table 1. We first generated humanized mice. CD34⁺ HSCs were purified from the donor G-CSF mobilized PBMCs before injecting into sub lethally irradiated NSG mice. Serial blood samples were collected and flow cytometry analyses was used to determine the frequency of human CD45⁺ (hCD45⁺) among all CD45⁺ cells as a measure of human donor engraftment (Fig. 2A, Supplemental Fig. 1). Significant numbers of hCD45⁺ cells were detected in the blood at 4 weeks post donor cell injection (Fig. 2B). The percentage of donor cells declined thereafter but stabilized at week 12 and was sustained at ~5% up to at least week 24. The mice were euthanized at 24 weeks and the mononuclear cells from bone marrow, spleen, liver, and lungs, in addition to blood were examined. We found that hCD45⁺ cells were more frequent in organs other than blood with the highest in bone marrow (Fig. 2C). Therefore, we successfully generated humanized NSG mice (hu-NSG) by engrafting the donor G-CSF mobilized HSCs into NSG mice.

Figure 2. Reconstitution of the human immune system in NSG mice engrafted with CD34⁺ HSCs from the donor G-CSF mobilized peripheral blood.

Two pairs of transplant donor HSCs and patients-derived AML cells were used in our study. (A) Representative flow cytometry data displaying the gating of mouse CD45⁺ versus human CD45⁺ cells. (B) The level of human CD45⁺ cells in peripheral blood was assessed at 4, 8, 12, 16 and 24 weeks post HSCs injection. In pair 1, eight NSG mice were injected with HSCs, one died of no clear etiology at week 13, 4 mice received patient-derived AML at week 12. Therefore, data of 8 mice were available at week 4, 8, and 12; data of 3 mice were available at week 16 and 24. Six mice were studied in pair 2, 3 mice received patient-derive AML at week 12, therefore data of 3 mice were available at week 16 and 24. (C) The frequency of human CD45⁺ cells in organs. Mice were euthanized at 24 weeks post HSCs injection. Mononuclear cells from bone marrow, spleen and liver were assessed by flow cytometry. Shown are the summary data (n=3 for pair 1, n=3 for pair 2). (D) Frequencies of each immune cell component in blood of humanized mice were evaluated by flow cytometry. The frequencies of NK cell, monocytes, T cells and B cells are shown (week 4, n=4; week 8, n=8; week 12, n=8; week 16, n=1-2; week 24, n=2-3. Note that data was not available for all mice at some time points due to limited blood samples). (E) Immune cell component in blood, bone marrow, spleen and liver of humanized mice at 24 weeks post HSC injection. T distributed stochastic neighbor embedding (tSNE) data are shown. (F-H) Differentiation and function of human T cells in humanized mice. (F) T cell CD4/CD8 ratio (week 12, n=3; week 16 and 24, n=2). (G) Differentiation stages of CD4 and CD8 T cells (naïve T cells (T_N, CD45RA⁺CCR7⁺), central memory T cells (T_CM, CD45RA⁻CCR7⁺), effector memory T cells (T_EM, CD45RA⁻CCR7⁻) and terminally differentiated T cells (T_EMRA, CD45RA⁺CCR7⁻)). (H) Functional studies shown by the expression of Ki-67 and Granzyme B in CD8 T cells. (I) Cytokine production by T cells in spleen. CD3⁺ T cells from pooled spleen cells (n=3) at 24 weeks post HSCs injection were stimulated with anti-CD3/CD28. The expressions of IFN-γ and TNF-α are shown.

We next evaluated the development of the human immune system in the hu-NSG mice. Frequencies of each immune component among hCD45⁺ cells in blood were assessed. We observed that innate immune cells developed early in humanized mice with relatively high frequencies of NK cells and monocytes in the blood at 4 weeks, and decreased thereafter. T cells were minimal at 4 weeks, but significantly increased to over 30% at week 16 and remained high until at least week 24. We also observed significant B cell development in humanized mice, representing 50-80% of hCD45⁺ cells in blood at almost all time points (Fig. 2D and Supplemental Fig. 2). Upon euthanization at 24 week, we assessed immune components in each organ and found adequate development of human NK, monocytes, T cells, and B cells in spleen, BM, and liver (Fig. 2E).

We further examined the differentiation and function of T cells as they are the major contributors to GVL effect (1, 18). In the kinetic study of hCD45⁺ T cells from peripheral blood, we observed that the CD4/CD8 ratio was initially reversed (<1) at week 12 and 16, but increased to normal range (above 2) at 24 weeks post HSC injection (Fig. 2F). When differentiation status was evaluated based on the expression of CD45RA vs. CCR7, we found all differentiation stages, including Naïve (T_N), central memory (T_M), effector memory (T_EM) and terminally differentiated T cells (T_EMRA), in both CD4 and CD8 T cell populations, although T_EMRA were minimal especially among CD4 T cells (Fig. 2G). These T cells were functional as a significant portion of them were positive for Ki67, indicating active proliferation. In addition, intracellular GranzymeB was detected in CD8 T cells, demonstrating cytotoxic potential (Fig. 2H). We also performed in vitro stimulation assays to evaluate intracellular cytokine release by T cells upon TCR engagement with anti-CD3/CD28. Pooled spleen mononuclear cells from humanized mice (n=3) were used to provide adequate T cells for this approach. We observed intracellular production of TNF-α and IFN-γ in 38-53% of the hCD3⁺ T cells (Fig. 2I).

In addition, the activation immune markers of hCD28, hOX40, hCD226, hHLA-DR, hBTLA, hCD38, hICOS and h4-1BB on T cells; hCD16, hCD57, hKIR and hNKG2A on NK cells; and hHLA-DR and hCD38 on monocytes were evaluated. Significant expression of majority of these markers were observed. HLA-DR and CD38 were strongly expressed on monocytes, CD38 expression were detected on 31% of T cells (Supplemental Fig.3).

These data demonstrate a successful generation of hu-NSG mice by injecting G-CSF-mobilized healthy donor HSCs into NSG mice. Importantly, we observed an adequate reconstitution of human immune system including functional T cells in the hu-NSG mice.

Immune reconstitution in hu-NSG mice recapitulated clinical scenario in the patient who received the corresponding HSCs

Donor HSCs for the pair 1 model (Supplemental Table 1) were collected from a transplant donor, whose PBMCs were applied to an AML patient who underwent alloSCT. We had followed this patient clinically and collected serial blood samples 1, 3, 12 and 24 months post transplantation. This allowed us to directly compare the same donor HSCs in the patient and NSG mice. Of note, the patient received 4ϗ10⁶/kg G-CSF-mobilized donor CD34⁺ stem cells following myeloablative conditioning (Busulfan/Cyclophosphamide). Chimerism study of blood and bone marrow at 1 month showed full donor (>95%), which remained thereafter. Thus, blood samples that we collected from the patients post-transplant are largely donor-derived.

We performed comprehensive flow cytometry analysis on the serial blood samples from the AML patient. Consistent with findings in the majority of patients undergoing alloSCT (19, 20), we observed a reconstitution of innate immune cells early post transplantation. Both NK cell and monocyte developed significantly at 1 month post transplantation, and trended down thereafter, whereas T cell frequencies trended up over time (Fig. 3A). Importantly, these patterns are in line with our observations in the hu-NSG mice (Fig. 2D). Interestingly, in contrast to our findings that high frequencies of B cells (50-85%) in blood of hu-NSG mice, recovery of B cells in the patient was minimal and remained low at 6% at 24 months (Fig. 3A). This is consistent with clinical observations that B cells are the slowest to reconstitute and can take up to 5 years for full recovery post transplantation(21, 22).

Figure 3. Immune reconstitution in the patient who received alloSCT from the corresponding donor HSCs.

Serial blood samples were collected from the patients who received the pair 1 donor HSCs in a clinical alloSCT. Multicolor flow cytometry analyses were performed. (A) Frequencies of NK cells (CD45⁺CD3⁻CD56⁺), monocytes (CD45⁺CD11b⁺CD14⁺), T cells (CD45⁺CD3⁺) and B cells (CD45⁺CD19⁺ and CD20⁺) at each time point. (B) CD4/CD8 ratio. (C) T cell differentiation based on the expression of CD45RA and CCR7. (D) CD8 T cell function. Shown are data of proliferation, killing capacity, and cytokine productions at 12 months post transplantation.

We further analyzed the phenotypic and functional reconstitution of T cells in the patient. Consistent with a previous report that reconstitution of CD4 T cells occurs later than CD8 T cells(19), we observed a lower percentage of CD4 T cells and a reversed CD4/CD8 ratio at all time points to at least 2 years post transplantation (Fig. 3B), this is in line with early phase of immune reconstitution in hu-NSG mice (Fig. 2F). In addition, we observed similar pattern of T cell differentiation status in that all differentiation stages were presented with low frequencies of T_EMRA in CD4 T cells (Fig. 3C). Furthermore, we found that both CD4 and CD8 T cells were highly functional manifested by adequate expression of Ki67, intracellular perforin and Granzyme B, and production of IFN-γ and TNF-α upon in vitro stimulation with anti-CD3/CD28 (Fig. 3D, Supplemental Fig. 4). In addition, we evaluated the T cell clonality in hu-NSG mice and the patient transplanted with the same donor PBMCs. The data showed that T cells from the patient have more clones (productive simpson clonality, T cell of hu-NSG vs. patient sample: 0.1347 vs. 0.0541; Supplemental Fig 5A). We did observe a same dominant clone between them. The sequence was CASSLAQETQYF, which ranked number 10 in patient sample and ranked number 5 in hu-NSG mice.

Taken together, these data show that the immune reconstitution of G-CSF mobilized HSCs in NSG mice largely recapitulates the clinical scenario in the patient receiving the same HSCs during alloSCT. Importantly we observed a similar pattern of T cell development, differentiation, and functional status in the hu-NSG mice compared with that in the patient.

Successful engraftment of patient-derived AML in hu-NSG mice

Patient-derived AML cells were injected intra-femoral into the hu-NSG at 12 weeks post HSC injection. Two AMLs (Supplemental Table 1) that are partially HLA matched with their paired donor G-CSF mobilized HSC (used in building the hu-NSG mice) were applied in this study. Both AML express CD33 (Supplemental Fig. 6A). Serial blood samples were collected to assess development of leukemia, which was detected as the percentage of CD33⁺ cells among hCD45 gated PBMCs. In both models, leukemia blasts were detected at 4 weeks, which increased up to 70% in the blood by 12 weeks (Fig. 4A). In contrast, minimal numbers of CD33⁺ cells (mostly progenitor cells) were detected in PBMCs from hu-NSG mice without AML injection (Fig. 4A). Mice were euthanized at 12 weeks post AML injection to evaluate leukemia infiltration into selected organs. Consistent with the clinical observation that bone marrow is the major anatomic location for leukemia, the majority of human cells in the bone marrow were CD33⁺ blasts (Fig. 4B). We also detected a significant number of blasts in spleen, liver and lungs (Fig. 4B), indicating leukemia infiltration into these organs. We consistently observed prominently larger spleens in hu-NSG bearing AML compared to hu-NSG without leukemia (Fig. 4C). Furthermore, we performed Wright-Giemsa staining on bone marrow cells from the hu-NSG bearing AML. We observed morphological features of blasts similar to that of clinical samples from the AML patient (Fig. 4D, Supplemental Fig. 6B). Taken together, our data demonstrate the successful engraftment of patient-derived AML in hu-NSG mice.

Figure 4. Successful engraftment of patient-derived AML in humanized NSG mice.

Leukemia cells from AML patients were injected intrafemorally into the humanized mice at week 12 post HSCs injection (1 × 10⁶ AML cells/mouse). Two AMLs that are partially HLA matched with their paired donor HSC were applied in this study. Humanized mice (hu-NSG) without AML injection were followed as controls. (A) The frequencies of CD33⁺ leukemia cells were tested at 4, 8 and 12 weeks post AML injection. Left data displays the gating strategy for CD33⁺ leukemia cells. Right are the summary plots showing the frequency of CD33⁺ cells among human CD45⁺ cells in both hu-NSG (n=3-5) and hu-NSG bearing AML (n=3-5). (B) The frequency of CD33⁺ cells among human CD45⁺ cells in organs at week 12 post leukemia cells injection (n=3-5). (C) Image shows the size of spleens collected from hu-NSG and hu-NSG bearing AML (pair 1). (D) Shown is the Wright Giemsa staining data of bone marrow from the hu-NSG bearing AML (pair 1). Arrows indicate the leukemia cells. P values were obtained by Mann-Whitney U test. *P<0.05, **P<0.01.

Leukemia growth was decreased in hu-NSG mice compared to that in NSG mice

To determine whether the reconstituted human immune system in the hu-NSG mice has anti-leukemia effect, we conducted a separated experiment injecting the patient-derived AML cells into hu-NSG mice (as was performed in Fig. 4) vs. NSG mice (without human immune system). A serial of blood samples at 4, 8, and 12 weeks post AML injection were collected before flow cytometry analysis for leukemia burden by assessing frequencies of hCD45⁺CD33⁺ cells. At all-time points tested, the leukemia growth trended less in hu-NSG mice compared to that in NSG mice and was statistically significant at 12 weeks. This was observed in both model pairs (Fig. 5A). The absolute number of hCD45⁺CD33⁺ cells were also evaluated and showed similar trend (Supplemental Fig. 7). Consistently, leukemia infiltrates in bone marrow and liver were significantly lower in hu-NSG mice (Fig. 5B).

Figure 5. Leukemia growth was decreased in humanized NSG mice compared to that in NSG mice.

Patient-derived AML cells were injected into humanized NSG mice vs. non-humanized NSG mice. Data of both pair models are shown. (A) Flow cytometry analysis for leukemia burden by assessing frequencies of hCD45⁺CD33⁺ cells among all PBMCs of blood (n=3-5). (B) Leukemia infiltration in bone marrow, spleen and liver (n=3-5). (C) Complete Blood Count (WBC, Hgb, Hct, and Platelet) in huNSG (n=3), huNSG+AML (n=3) and NSG+AML (n=2) mice. Blood were collected at 18 weeks post AML injection. Of note, total of 5 mice for NSG+AML group were studied, 3 died of leukemia before week 18, therefore only 2 mice were available to test for CBC. P values were obtained by Mann–Whitney U test. *P<0.05.

In addition, we evaluated complete blood count (CBC) and observed that NSG mice engrafted with AML had much higher WBC (majority were blasts in differential) and severely decreased Hemoglobin (Hgb)/Hematocrit (Hct) and platelet. Hu-NSG mice with AML showed moderately elevated WBC (due to increased blast) and significantly decreased (but less severe than NSG+AML) platelet compared with that in hu-NSG (748±97×10³/mm³ vs. 1169±52×10³/mm³, P=0.0190, Fig. 5C). Hgb was only mildly decreased in hu-NSG mice with AML likely due to lower leukemia burden in these mice.

Collective, we observed a significantly reduced leukemia growth and subsequently less level of bone marrow failure in hu-NSG mice compared to that in control NSG mice, demonstrating a functional GVL effect. We therefore term these mice as hu-GVL mice.

Hu-GVL model recapitulates clinical features for leukemia immune escape in AML relapse post alloSCT

Despite evidence of functional GVL effect, we did observe significant number of AML cells in the bone marrow and other organs of hu-GVL mice (Fig. 4). This is consistent with clinical scenario of leukemia relapse post alloSCT. It has been discovered that AML cells from patients who relapsed post alloSCT had decreased HLA expression, which may contribute to the leukemia immune escape (23–25). In addition, several studies, including ours, have demonstrated an up-regulation of inhibitory pathways including PD-1 and T-Cell Immunoglobulin and ITIM Domain (TIGIT) in AML patients who relapse post alloSCT (25–29). To evaluate whether our novel hu-GVL model recapitulates these clinical features, we first assessed the expression of PD-1 and TIGIT on CD8 T cells in hu-GVL (with AML) vs. hu-NSG (without AML) mice. Serial blood samples collected at different time post AML injection were examined by flow cytometry. We found that, at the majority of time points, the frequencies of PD-1⁺ and PD-1⁺TIGIT⁺ CD8 T cells were significantly higher in hu-GVL mice compared to that in hu-NSG mice (Fig. 6A). Interestingly, this pattern was only observed in blood, whereas other organs showed comparable expression for both molecules (Supplemental Fig. 8). It is not clear why majority of CD8 T cells in organs showed no up-regulation of PD-1 and TIGIT. The different tumor microenvironment in organs (as opposed to blood) might lead to different leukemia escape mechanisms. Further studies are warranted to dissect the specific mechanisms. We next tested the expression of PD-L1 and CD155 (TIGIT ligand) on AML cells in hu-GVL (with human immune system) vs. NSG (without human immune system) mice. Significant increase of PD-L1 and CD155 expression on AML cells were observed in hu-GVL mice (Fig. 6B). Furthermore, we detected a significant down-regulation of HLA-DR on AML cells in hu-GVL mice (Fig. 6C). Thus, consistent with clinical setting of AML relapse post alloSCT, hu-GVL mice had reduced HLA and up-regulation of inhibitory pathways on AML under immune pressure.

Figure 6. Hu-GVL model recapitulates clinical features of leukemia immune escape in AML relapse post alloSCT.

(A) The expression of PD-1 and TIGIT on CD8 T cells in peripheral blood of hu-NSG vs. hu-GVL mice. Shown are data at week 16, 20 and 24 post HSCs injection (week 4, 8 and 12 post AML cell injection). (B-C) The expression of PD-L1 and CD155 (B) and HLA-DR (C) on AML blasts in the bone marrow of NSG+AML vs. hu-GVL mice. Flow cytometry data (left) and summary plots (right) for the cell frequencies are shown. (NSG+AML, n=3-5; hu-GVL, n=3-5). P values were obtained by unpaired Student’s t test (PD-L1 data in Fig. 6B) or Mann–Whitney U test (all data except for PD-L1). * P<0.05, ** P<0.01.

Discussion

In this study, we successfully established the human hematopoietic system in immune-deficient NSG mice. Immune reconstitution in these mice largely recapitulated the clinical scenario in the patient who received the corresponding HSCs. Allogeneic but HLA partially matched patient-derived AML cells were successfully engrafted in these hu-NSG mice. Importantly, we observed a significantly reduced (yet incomplete elimination of) leukemia growth in hu-NSG mice compared to that in control NSG mice, demonstrating a functional (but defective) GVL effect. Additionally, NK cells and T cells reconstituted in the humanized model showed a strong killing to the patient-derived AML in vitro (Supplemental Fig. 9). Thus, for the first time, we established a novel humanized mouse model that can be used for studying human GVL responses against human AML cells in vivo. The hu-GVL model circumvents the major defect of traditional mouse models, in which the interpretations of results are limited by the biological difference between humans and mice. This highly clinically relevant model system forms a solid base for studies of mechanisms in human GVL, and subsequently developing novel effective therapeutic approaches for leukemia relapse post alloSCT.

Our novel hu-GVL model sets a strong platform to test immune-based therapeutic agents. T cell exhaustion has been observed in patients with AML relapse after alloSCT(26, 30, 31). In experimental studies, PD-L1 blockade enhances the capacity of TCR-transgenic T cell elimination of OVA-transduced leukemia cells in mice post alloSCT(32). In addition, alloSCT results in exhaustion of donor MataHari CTLs (specific to H-Y) that infiltrate the bone marrow (BM), lymph node or spleen from AML mice early after alloSCT(33). These observations suggest an important role of T cell exhaustion in leukemia relapse post alloSCT. However, clinical studies applying PD-1 blockade to AML patients post alloSCT showed limited benefit (34–36). It is possible that other inhibitory pathways were upregulated during PD-1 inhibition, therefore combined inhibition of PD-1 and other immune checkpoints may improve the treatment. Alternatively, therapeutic regimens derived from traditional experimental models may not be optimal for clinical settings. Furthermore, AML is a highly heterogenous disease and alloSCT is a complicated procedure with a variety of clinical conditions. Taking all these into consideration, highly clinical relevant models that reflect the diversity and complex nature of clinical alloSCT are essential for testing novel human-specific agents and developing optimal therapeutic strategies for leukemia relapse post alloSCT. Our hu-GVL model ideally fits the need as 1) it is highly clinically relevant since the HSCs and AML are derived from clinical donor and patients; 2) consistent with findings in the clinical setting of AML relapse post alloSCT, we found increased PD-1 and TIGIT pathways in the hu-GVL mice. Therefore, this model provides a valuable platform for investigating the therapeutic effect of combining blockade of TIGIT and PD-1 in AML; 3) our study shows a proof of concept by successful establishment of two hu-GVL models using two pairs of donor HSCs and patient-derived AML. Conceivably, generating multiple patient-specific models with patient-derived AML and adequately HLA matched donor HSCs is highly feasible. The unique and patient-specific system will form solid base to determine patient-specific AML pathogenesis and discover new personalized treatment.

We selected G-CSF-mobilized peripheral blood from transplant donors as the source of HSCs for building hu-NSG mice. HSCs derived from umbilical cord blood and fetal liver are commonly applied in developing humanized mice(37–39). G-CSF-mobilized peripheral blood is rarely used due to its less efficient engraftment(14, 40). However, G-CSF-mobilized peripheral blood cells are most widely used clinically for alloSCT. Thus, models built with cord blood or fetal liver may not represent clinical scenarios that involve G-CSF mobilized peripheral blood.

Additionally, the limited number of cells in each cord blood sample or fetal liver limits the study of patient-specific disease conditions. Taking advantage of our long-established large tissue bank, we were able to identify donor G-CSF mobilized peripheral blood with adequate numbers of HSCs to generate efficient and stable humanized mice models. We injected 5-10×10⁵ CD34⁺ HSCs per mouse, two to five fold higher than the numbers reported using cord blood and fetal liver, into sub lethally irradiated 3-week old NSG mice. Human hematopoietic cells were successfully engrafted and at 24weeks post-transplant, hCD45⁺ cells attained 5% in blood and 20-35% in spleen, liver and bone marrow (Fig. 2A–C). Functional human immune system was also established (Fig. 2D–I). Most importantly, our longitudinal clinical samples enabled a retrospective study of immune reconstitution in the patient who received the corresponding donor G-CSF mobilized peripheral blood during alloSCT. We observed a consistent pattern of reconstitution of NK cells, monocytes, and T cells. Interestingly, in contrast to minimal numbers of B cells that were detected in patients post alloSCT, a high frequency of B cells was maintained in humanized mice. Multiple studies have demonstrated that the majority of B cells in humanized mice are non-functional(39, 41, 42), so further analysis is needed to define the significance of this humanized mouse model for the study of B cells and humoral immunity. Notably, we performed comprehensive analyses on the differentiation, immune phenotype and function of T cells, and found a highly consistent pattern between clinical samples and that of humanized mice (Fig. 2 and Fig. 3). NSG mice were applied in our study, using models providing human cytokines and supporting more completed human immune reconstitution in future studies would improve the system(40, 43). Of note, we performed additional experiments to evaluate human cytokines in patients as well as in our humanized mice models, and found similar patterns in serum cytokine level that IL-17A, IL-7, and IL-6 are most predominant cytokines in both patients post alloSCT and our humanized mice (Supplemental Fig. 10). Therefore, our data provide direct evidence that humanized mice engrafted with HSCs derived from G-CSF-mobilized peripheral blood largely recapitulate the immune reconstitution of the HSCs in patients. This highly clinically relevant model provides a novel tool to study human immune responses, especially T cell immunity in cancer and other disease conditions.

One limitation of our model is that in the hu-NSG mice, at least a portion of the human T cells had undergone education in the host mouse thymus and were therefore expected to be H2-restricted. This likely hindered HLA-restricted interactions between the T cells and HLA-expressing AML and antigen presentation cells. Engrafting patient-derived AML into bone marrow-liver-thymus (BLT) mice(44, 45), where an implanted autologous fetal thymus enables the development of HLA-selected human T cells, may help to resolve the issue. However, a wasting GVHD-like syndrome has been frequently observed in humanized BLT mice, making the experimental window severely narrowed(46–48). Limited availability of human fetal tissues and potential ethical considerations also prevent wide use of this model. Alternatively, HLA-expressing immunodeficient mice are under active development. Engrafting HLA-matched HSCs into these mice results in development of human HLA-restricted T cell responses(49). Our study provides a proof of concept that generating hu-GVL mice by engrafting patient-derived AML in humanized mice is feasible. Integrating HLA-expressing immunodeficient mice into our hu-GVL model will significantly improve the preclinical system to unveil the mechanisms of human GVL effect.

Abbreviations

alloSCT

allogeneic hematopoietic stem cell transplantation

AML

acute myeloid leukemia

GVL

graft versus leukemia

HLA

human leukocyte antigen

HSCs

hematopoietic stem cells

NSG

NOD-Cg-Prkdc^scidIL2rg^tm1Wjl/SzJ

PBMCs

Peripheral blood mononuclear cells

PD-1

programmed cell death protein 1

TIGIT

T-cell Immunoglobulin and ITIM Domain

tSNE

T-distributed stochastic neighbor embedding