A Humanized Animal Model Predicts Clonal Evolution and Therapeutic Vulnerabilities in Myeloproliferative Neoplasms

Myeloproliferative neoplasms (MPN) are clonal hematologic diseases characterized by the aberrant proliferation of one or more myeloid lineages and progressive bone marrow (BM) fibrosis. BCR-ABL–negative MPNs are characterized by an excess of red blood cells [polycythemia vera (PV)] or platelets [essential thrombocythemia (ET)] or by the deposition of reticulin fibers in the BM [myelofibrosis (MF); ref. 1]. MF is the deadliest MPN subtype, with median survival for patients of approximately 5 years, and can occur as a de novo neoplasm or progression from preexisting ET or PV. The most commonly mutated genes in MPNs are JAK2 (2–4), MPL (5), and CALR (6, 7), which are responsible for disease initiation. The acquisition of additional mutations in epigenetic regulators, transcription factors, and signaling components can modify the course of the disease and contribute to disease progression and leukemic transformation (8, 9). JAK2, MPL, and CALR mutations share a hallmark of aberrant JAK–STAT activation. As such, the JAK1/2 inhibitor ruxolitinib is a first-line therapy (in particular for MF) and can alleviate constitutional symptoms of the disease (10) but does not eliminate the malignant clone and has minimal impact on BM fibrosis and overall survival (11), highlighting the need for more effective therapies.

As the genetic basis of MPN has been elucidated, murine models have been developed to dissect disease origins and mechanisms. Transgenic and conditional knock-in mice have shown that physiologic expression of JAK2^V617F alone can induce a PV- or ET-like MPN in vivo (12), and the hematopoietic stem cell (HSC) pool contains the disease-initiating potential (13, 14). However, current mouse models do not recapitulate the clinical heterogeneity, genetic composition, or morphologic features of MPN. Apart from retroviral overexpression of MPL^W515L (5, 15), mouse MPN models do not typically generate robust reticulin fibrosis in the BM, the most significant MF pathology. The lack of faithful preclinical MPN models presents a major barrier to developing effective therapies.

To circumvent the limitations of conventional mouse models, patient-derived xenografts (PDX) are increasingly used as tools for preclinical drug development and target validation and have been shown to recapitulate the pathology and genetics of the patients from whom they are derived in various malignancies (16). Technological developments in immunodeficient mouse strains (17, 18) have enabled successful xenotransplantation of normal human CD34⁺ hematopoietic stem and progenitor cells (HSPC), as well as patient samples from myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML) subtypes that were previously refractory to engraftment. However, efforts to generate MF PDXs have been largely unsuccessful due to poor engraftment potential of MPN disease-initiating cells (19), or require specialized techniques (20).

In a previous study, we demonstrated that genetic inhibition of the polycomb repressive 2 (PRC2) cofactor JARID2 in MPN patient samples permitted engraftment in NOD/SCID-Il2rg-null-3/GM/SF (NSGS) mice (21). This caused us to revisit methodology to determine if technological advances might facilitate a generalizable MPN PDX system. Here, we present a generalizable method for xenotransplantation of human MF CD34⁺ cells into newer-generation immunodeficient mice by X-ray guided intratibial (ITib) injection. These PDXs reproduce the hallmarks of MF, including, most importantly, reticulin fibrosis in the BM. Genetic analysis shows faithful maintenance of MF patient clonal architecture within the engrafted cell population. Enhancing the utility of the model, we show this system can predict for clonal progression to secondary AML (sAML) in patient samples obtained far prior to clinical manifestation and can be manipulated to provide a platform for genetic and pharmacologic target validation in MF. Thus, this model presents a robust and malleable platform to study MPN genetics and therapeutics.

In our previous study, genetic inhibition of the PRC2 cofactor JARID2 by lentiviral short hairpin RNA (shRNA) permitted sustained engraftment of CD34⁺ cells from patients with MF in the blood, BM, and spleens of NSGS mice accompanied by marked reticulin fibrosis (21). However, we noted evidence of engraftment and modest reticulin fibrosis in recipients of MF patient cells transduced with the negative control shRNA vector. This encouraged us to revisit the methodology to determine if techniques could be adapted for a generalizable PDX model of MF. CD34⁺ cells were freshly isolated from the peripheral blood mononuclear cell (PBMC) fraction from patients with MF (Supplementary Fig. S1). Selection criteria for patient samples included blast count <5%, with more than 80% of patient samples having no detectable blasts. MF CD34⁺ cells were cultured for 16 hours in serum-free media with hematopoietic cytokines [stem cell factor (SCF), FLT3L, and thrombopoietin (TPO)], then transplanted into sublethally (200 rads) irradiated NSGS mice via X-ray guided ITib injection (Fig. 1A). NSGS mice express humanized SCF, GM-CSF, and IL3 (17) and demonstrate improved myeloid engraftment from normal human CD34⁺ HSPCs and AML subtypes that were previously difficult to engraft (22, 23). The technical adaptation of ITib injection has facilitated PDX models from difficult-to-engraft diseases such as chronic myelomonocytic leukemia (24). Comparison of ITib versus intravenous injection showed superior engraftment for both normal and neoplastic human CD34⁺ cells (Fig. 1B). We reasoned these technological advances may be sufficient to generate reliable MF PDXs.

Although the engraftment of human cells in the peripheral blood was highly variable between patient samples (Supplementary Fig. S2A), virtually all NSGS mice that received >100,000 CD34⁺ MF patient cells showed robust engraftment in the BM (Fig. 1C) and spleen at 12 weeks posttransplant (Fig. 1D). Consistent with MPN, engrafted human cells were almost exclusively CD33⁺ myeloid cells (Fig. 1C) or immature erythroid progenitors (Fig. 1E). Except for one sample, there was no difference in BM engraftment between the injected and contralateral legs (Supplementary Fig. S2B). Engraftment levels were not correlated with the number of CD34⁺ cells transplanted (Supplementary Fig. S2C) but rather inherent properties of individual samples. Analysis revealed other pathologies characteristic of MPN, such as activated JAK–STAT signaling (Supplementary Fig. S2D) and altered blood counts (Supplementary Fig. S2E; Supplementary Table S1). Strikingly, most MF PDXs showed marked reticulin fibrosis in the BM and spleen compared with mice transplanted with cord blood CD34⁺ cells or NSGS mice that were irradiated and then monitored for 12 weeks (Fig. 1F and G). The most aggressive patient samples were able to induce a lethal MPN in NSGS mice (Supplementary Fig. S2F). Mice that succumbed to lethal disease did not show evidence of transformation to sAML such as increased blood counts or circulating blasts. Rather, lethal MPN was significantly associated with the degree of reticulin fibrosis in the BM (Supplementary Fig. S2G) and peripheral anemia (Supplementary Fig. S2H), indicative of ineffective hematopoiesis. To determine if the disease could be serially transplanted, 1 × 10⁶ hCD45⁺ cells from primary PDXs were transferred to secondary NSGS recipients using the same technical approach. Despite robust engraftment in primary transplantation (Fig. 1H), MF patient cells were unable to engraft or propagate disease pathologies to secondary recipients (Supplementary Fig. S2I). This, however, does indicate that the samples initially transplanted, although neoplastic, were not transformed and the resultant engraftment was not due to leukemic subclones. Despite the lack of serial propagation, this methodology provides a robust and reproducible model for establishing PDXs from MF patient samples.

To ensure the engrafted MF cells in NSGS mice retained the molecular properties of the patient cells, single-cell RNA sequencing (RNA-seq) was performed. PBMCs from two independent cord blood and MF patient samples were compared with the engrafted hCD45⁺ cells from the BM of NSGS mice 12 weeks posttransplant. UMAP analysis showed that the PBMCs from both cord blood samples clustered together, but the MF PBMCs clustered distinctly from the normal cells as well as from each other (Fig. 2A). Although JAK2^V617F was the driver mutation in both these patients with MF, the separation from each other was likely driven by the different cooperating mutations (e.g., MF 585953 = TET2, ASXL1; MF 784981 = SETBP1, CUX1, ZRSR2). Importantly, the separation of sample-specific cell clusters was maintained in engrafted PDX cells (Fig. 2A). Although the sequencing platform here did not permit simultaneous detection of genomic mutations in single cells, we selected MF patient samples for this experiment based on high variant allele fraction (VAF) of JAK2^V617F (MF 585953 = 79%, MF 784981 = 97%), and independent analysis of hCD45⁺ cells from the BM of NSGS mice showed the VAF was sustained in the engrafted cells (see next section for details). Thus, it is likely that most cells sequenced in this experiment expressed the oncogene. Annotation of cell clusters using marker gene expression showed that the MPN-specific cell clusters were enriched for myeloid cells, platelets, and HSPCs in both the patients and the PDXs (Fig. 2A). This informs that the distinct molecular profiles of MF patient cells are maintained in this PDX model.

Experimental models implicate HSCs as the disease-initiating cell population in MPN (13, 14). Moreover, MPN founding events such as JAK2^V617F and CALR frameshift mutations (CALR^fs) can be detected in humans with clonal hematopoiesis (CH; refs. 25–27), further suggesting an HSC origin as the mutational reservoir. Most MF therapies are likely ineffective because they fail to eradicate this mutant HSC population in patients. We sought to determine if the MPN mutant HSCs were propagated in this PDX model, which could provide a system to evaluate the impact of novel therapeutics specifically on the disease-initiating cell population. Flow cytometric analysis of the BM of recipient NSGS mice showed that human HSCs (lineage^loCD34⁺CD38⁻CD45RA⁻CD90⁺) were not maintained in animals xenografted with cord blood CD34⁺ cells (Fig. 2B). This is likely due to the inflammatory environment driving HSCs toward terminal differentiation as opposed to self-renewal (28). Conversely, analysis of NSGS mice xenografted with CD34⁺ cells from patients with MF showed that human HSCs were readily detectable in most cases (Fig. 2C), suggesting that this is a permissive niche for the propagation of MPN disease-initiating cells.

Given that MPN disease-initiating HSCs were propagated, we sought to benchmark the therapeutic relevance of the system by evaluating the effect of ruxolitinib as current standard of care for MF. NSGS mice were xenografted with MF CD34⁺ cells and then 4 weeks posttransplant were randomized for treatment with ruxolitinib (or vehicle control) based on engraftment of human cells in the peripheral blood. Engraftment and pathology of MF patient cells were evaluated after 4 weeks of therapy. Although ruxolitinib treatment did not affect the engraftment of MF cells in the peripheral blood (Fig. 2D), burden of human cells was significantly reduced in the BM (Fig. 2E), accompanied by a dramatic reduction in splenomegaly (Fig. 2F and G). But consistent with human patients, treatment with ruxolitinib did not affect the abundance of human HSCs in the BM (Fig. 2H).

The genetic hierarchy of MPN is hypothesized to influence treatment response to JAK/STAT inhibitors. Exome sequencing was performed on MF patient cells prior to transplantation and hCD45⁺ cells 12 weeks posttransplant to determine if the patient's clonal architecture was preserved in PDXs (Supplementary Table S2). The mutational landscape observed in patients with MF was maintained in the xenografted cells from NSGS mice (Supplementary Fig. S3). There was high reproducibility for the same patient sample xenografted into different NSGS mice (Fig. 3A), and the VAF of MPN driver mutations showed excellent correlation between patient and PDX cells (Fig. 3B). Other cooperating recurrently mutated gene mutations also showed consistent VAFs comparing CD34⁺ cells from the patients with MF and hCD45⁺ cells from the BM of NSGS mice (Fig. 3C).

In almost all PDXs, the engrafted human cells were virtually exclusively myeloid (CD33⁺). However, xenografted cells from patient MF 300179 generated significant lymphoid progeny in NSGS mice (Supplementary Fig. S4A), which allowed comparison of mutant allele burden in different cell lineages and analysis of the developmental trajectory of this patient. Heterozygous DNMT3A^P307L and TET2^P1962L mutations were found in all cells from the patient donor sample and the xenografted cells. Homozygous JAK2^V617F and TP53^M246V mutations were detected at a lower VAF in the bulk posttransplant hCD45⁺ cells compared with the donor sample because these variants were completely absent from the B cells (CD19⁺) in the PDX (Fig. 3D). This suggests the B cells were the progeny of a nonneoplastic CH clone with DNMT3A^P307L and TET2^P1962L mutations, as they also lacked evidence of promiscuous myeloid lineage marker expression (Supplementary Fig. S4B). For MPN evolution, this clone subsequently acquired homozygous JAK2^V617F and TP53^M246V mutations in either an HSC or downstream myeloid progenitor cell, which drove excess myeloid differentiation and clinical manifestation (Fig. 3E).

One of the most feared complications of MPN is disease progression to sAML. Up to 20% of patients with MF have sAML transformation, which is driven by the acquisition of additional cooperating mutations. Patients with post-MF sAML have a dismal prognosis, with a median survival of only 6 months (9). Earlier identification of patients with MF who are more susceptible to sAML transformation could stratify such individuals for alternative interventions. Such predictions are not possible with the limit of mutation detection for conventional genomic approaches. For the PDXs from patient MF 504293, exome sequencing detected a pathogenic EZH2^Y663H mutation in the xenografted human cells that was not identified in the input CD34⁺ cells isolated from the patient (Fig. 4A). Ectopic expression of EZH2^Y663H in EZH2-null HEL cells indicated this mutation encodes a loss-of-function protein with severely compromised H3K27me3 activity (Supplementary Fig. S5A). Review of patient history uncovered that this individual eventually underwent disease progression to sAML (Fig. 4B). Strikingly, clinical sequencing performed at the time of sAML diagnosis identified the same EZH2^Y663H mutation present in the PDX, which had never been observed in previous clinical sequencing performed throughout the patient history (Fig. 4C). Of note, the PBMC sample used for PDX input was obtained from this patient 516 days prior to sAML diagnosis. This implies that the selection pressure of xenotransplantation facilitated expansion of a very rare subclone that was responsible for leukemic transformation. As the xenografted cells were obtained from the patient in the chronic MF stage years before sAML diagnosis (Fig. 4B and C), this suggests the PDX system might serve as a prediction model for patients with rare, rising sAML subclones who are at risk for leukemic progression. To examine this, droplet digital PCR (ddPCR) was performed with an EZH2^Y663H-specific probe on an independent patient sample banked at the same time as the specimen used for xenotransplantation. ddPCR analysis (Supplementary Fig. S5B) indeed confirmed the presence of very rare EZH2^Y663H-mutant cells in the PBMCs of this patient (0.17% of cells), with enrichment in the CD34⁺ cell population (0.63% of cells), although still far below the sensitivity of detection of conventional exome sequencing (Fig. 4D).

Analysis of other exome sequencing cases identified another individual (MF 764338) in whom a pathogenic variant was detected exclusively in the human cells from the PDX but not in the input CD34⁺ cells at time of collection. Approximately 1 year after this sample was obtained, the patient was noted to have increasing circulating blasts and was under consideration for a bone marrow transplant. Clinical sequencing at that time identified the TP53^R248Q mutation in the patient's PBMCs, which was the first time in the patient history this variant had been observed (Fig. 4E). The hCD45⁺ cells from the PDX were found to harbor the TP53^R248Q mutation, one of the most common TP53 variants in de novo AML and post-MPN sAML (29), at a VAF of 15% to 20% (Fig. 4F). We returned to an independent PBMC sample from the time of collection that was used for xenotransplantation to determine if the TP53^R248Q mutant could be identified by ddPCR. In whole PBMCs, TP53^R248Q-mutant cells were at the level of sensitivity of ddPCR (Supplementary Fig. S5C). But in CD34⁺ cells enriched from this time point, a distinct population of TP53^R248Q-mutant cells (0.48% of cells) could be identified (Fig. 4G). Cumulatively, these two cases demonstrate that rare disease-transforming clones are present in patients with MF long prior to diagnosis of clinical progression. Although these clones are undetectable using current clinical sequencing methodology, they can be readily identified in this PDX model.

There was one additional patient (MF 145790) in this cohort who eventually had transformation to sAML. Exome sequencing of the sAML sample compared with PBMCs and CD34⁺ cells from the pre-sAML MF stage did not identify sharp acquisition of pathogenic variants in the sAML phase. Two potentially pathogenic sAML-specific variants were identified, ASXL1^G643X and SRSF2^P95L, but the VAF of these variants was less than 5% of the total tumor fraction, which questions the functional relevance of these variants in the transformation process. As for the two patient samples described above, both of these variants were identified in PDX-derived hCD45⁺ MF cells (Supplementary Fig. S5D). Interestingly, exome sequencing of the engrafted patient cells from three independent NSGS mice showed the SRSF2^P95L variant was identified in two mice, and the ASXL1^G643X variant was identified in only one mouse, and these mice were not overlapping (Supplementary Table S2). This result suggests the ASXL1^G643X and SRSF2^P95L variants may have been present in distinct subclones that were not part of the malignant clone or perhaps independent premalignant clones arising from normal CH processes in this individual.

We sought to establish the utility of this model for genetic and pharmacologic target validation. Proof of principle focused on PIM kinase inhibition. The PIM family (PIM1–3) of kinases regulates a number of cellular processes through phosphorylation of a variety of target substrates. The proto-oncogenic potential of PIM kinases was initially described in MYC-driven lymphomagenesis (30), and PIM inhibition (PIMi) is being explored for a variety of solid tumors as well as hematopoietic malignancies (31, 32). In JAK2-mutant MPN cell lines, PIMi has been shown to restrain cell proliferation and overcome drug resistance through destabilization of MYC and suppression of prosurvival signaling pathways (33). PIM1 kinase is also upregulated in MF CD34⁺ cells regardless of their genetic background (34). Resultantly, PIMi has been presented as a therapeutic approach in JAK2-mutant MPN in combination with JAK/STAT inhibition (35). Our system provides a unique opportunity to validate these in vitro studies with cell lines using primary patient samples in vivo.

To determine the specificity of PIMi, cord blood (normal control) or MF patient CD34⁺ cells were targeted for PIM1 deletion via CRISPR/Cas9 RNP nucleofection (36). A guide RNA (gRNA) targeting the inert AAVS1 locus was used as a CRISPR/Cas9 negative control. Analysis of targeted cells via deep sequencing 72 hours after nucleofection confirmed high-efficiency editing (Supplementary Fig. S6A). Edited CD34⁺ cells were transplanted into NSGS recipients as described, and mice were sacrificed 12 weeks posttransplant to evaluate function of targeted cells. Although loss of PIM1 had no effect on the engraftment of cord blood–derived cells, PIM1-targeted MF cells showed reduced splenomegaly (Supplementary Fig. S6B) and markedly compromised engraftment in the BM (Supplementary Fig. S6C) that was not due to absence of targeted PIM1 populations (Supplementary Fig. S6A). These data suggest that PIM1 may be selectively required for MF cell function.

To evaluate pharmacologic potential, PDXs were established from MF patient samples and then randomized for treatment with the pan-PIMi INCB053914 (37), either as a single agent or in combination with ruxolitinib. Treatment was initiated 4 weeks posttransplant, and then effects were evaluated after 4 weeks of treatment (Fig. 5A). In comparison to vehicle-treated control mice, monotherapy with either INCB053914 or ruxolitinib modestly inhibited engraftment of MF cells and splenomegaly for each of three individual patient samples to varying degrees. However, combination therapy markedly inhibited engraftment of patient cells (Fig. 5B) and reduced peripheral blood counts (Supplementary Fig. S6D). Strikingly, although treatment with either single agent had no effect on the burden of human HSCs in the BM, combination therapy was effective in eradicating the JAK2^V617F-mutant MPN-initiating cell population (Fig. 5C). Both agents also reduced BM fibrosis (Supplementary Fig. S6E), although given the timeline for this experiment, this may reflect inhibition of MF cell engraftment rather than reversal of fibrosis. To determine if MPN would relapse after cessation of treatment, a cohort of NSGS recipients was monitored for a 4-week recovery after discontinuation of treatment. Although peripheral blood counts did rebound after discontinuation of treatment, most values still remained lower than vehicle-treated control mice (Supplementary Fig. S6F). After 4 weeks of recovery, human BM engraftment (Supplementary Fig. S6G), splenomegaly, and human HSC burden recovered somewhat in mice treated with PIMi plus ruxolitinib but still remained significantly suppressed compared with vehicle-treated mice (Fig. 5D). These results show that PIMi in combination with JAK–STAT inhibition specifically targets MF, particularly JAK2^V617F-mutant cells, and present this combination as an attractive therapeutic approach for patients with MF.

Although we observed reduced BM fibrosis in MF PDXs in the above experiment with PIMi plus ruxolitinib combination therapy (Supplementary Fig. S6E), this result could be attributed to the effect of ruxolitinib on inhibiting engraftment of MF patient cells because treatment was started 4 weeks posttransplant. Because robust reticulin fibrosis was observed in the vehicle-treated PDXs from MF samples 504293 and 585953 after 4 weeks of treatment (8 weeks total posttransplant), an experimental model was established to more specifically test the effect of therapeutics in reversing fibrosis. The BET bromodomain inhibitor JQ1 has been shown to be an effective antifibrosis therapy in a mouse retroviral MPL^W515L model of MF, largely through inhibiting the function of BRD4 in promoting NFκB-driven inflammation (38). JQ1 was effective for inhibiting fibrosis even as a monotherapy but showed particular potency in combination with JAK–STAT inhibition (38). To test this combination in the humanized MF model, NSGS mice were xenografted with CD34⁺ cells from patients MF 504293 and 585953 or control cord blood CD34⁺ cells, then randomized based on 4-week peripheral blood engraftment levels. At 8 weeks posttransplant, cohorts began treatment with JQ1, ruxolitinib, combination therapy, or vehicle control (Fig. 6A). Combination therapy inhibited total BM engraftment and human HSC burden in both cord blood and MF transplant recipients, although splenomegaly was specifically reduced in mice receiving MF patient samples (Fig. 6B). Blood counts were variably reduced with some selectivity for MF patient cells (Supplementary Fig. S7). The most important outcome from this experiment was effect on reticulin fibrosis. Although monotherapy with either JQ1 or ruxolitinib was able to modestly inhibit BM fibrosis over this time course, with possible JAK2^V617F versus CALR genotype-specific sensitivities (Fig. 6C), combination therapy was able to almost completely abrogate BM fibrosis (Fig. 6D). These results confirm murine data showing BET bromodomain inhibition as a potential antifibrosis therapy and validate this system as an effective model for assessing investigational drugs targeting fibrosis in human MF.

MPNs are largely incurable hematopoietic diseases. Although symptoms can be managed in most cases, many patients remain at risk for thrombohemorrhagic complications and disease transformation. Curative therapies would alleviate a substantial burden on public health but require methods to effectively eradicate the founding MPN clone. New therapeutic approaches based on patient molecular profiles are needed, which have been hampered by the inability of genetically engineered mouse models to recapitulate key MF pathologies such as reticulin fibrosis. Here, we present a system for establishing PDXs, which accurately propagates the genotypes and phenotypes of patients with MF. The model produces robust engraftment of patient-derived cells in all major hematopoietic organs, sustains disease-initiating MPN HSCs, and disseminates the most critical MF pathology of BM fibrosis. However, one limitation of this model is that the disease could not be serially transplanted. This indicates some critical human factors required to serially propagate human MPN-initiating HSCs are still lacking. This may limit the utility of this PDX system because it necessitates procurement of fresh samples from patients with MF. However, MF as a hematologic disease is amenable to such an approach. Patients in the chronic phase have regular clinical appointments, and the fibrotic BM niche results in a high proportion of HSPCs in circulation. Sufficient CD34⁺ cells for experimentation can be collected from routine blood draws for most patients in our experience. Another potential caveat is that CD34⁺ cells from the peripheral blood of patients with MF may have different biological properties than HSPCs still residing in the BM, although studies have suggested the VAF of MPN-initiating mutations is similar between blood and BM cells (39). BM samples from patients with MF were unavailable for direct comparison in this study due to the difficulty in obtaining cells from BM biopsy of these patients (“dry tap”).

One of the most feared consequences for patients with MPN is clonal evolution to sAML. Approximately 20% of patients with MF progress to sAML, which is associated with a dismal clinical prognosis and median survival of less than 6 months (40), largely due to the lack of targeted therapies. Earlier identification of patients with MF who are at risk for sAML transformation may allow stratification for more aggressive therapies, BM transplantation, or precision medicine approaches based on their mutation profile. However, the limit of detection of ∼2% VAF for most clinical sequencing approaches, combined with the fact that many of these assays sample only a portion of the genome, prohibits this type of prediction. The PDX model presented here is able to identify rare leukemic subclones present in MF PBMC samples years prior to clinical diagnosis of sAML. Presumably, the engraftment of these subclones is favored in NSGS mice due to the selective pressure of xenotransplantation. Even in samples from patients who have not progressed to sAML, there was enrichment for activated signaling mutations associated with high-risk MF (41, 42) in the PDXs relative to the patient cells such as KIT^956W in MF 455643, KRAS^Q61R in MF 530675, and NRAS^G12S in MF 542924. The lack of such selective pressures in the chronic phase of MF means these clones may lay dormant for substantial periods of time in the patients. This is much like the case in therapy-related myeloid neoplasms, in which TP53-mutant subclones lay dormant until challenged with the selective pressure of chemotherapy (43). The engraftment of these clones in this PDX model facilitated identification of the relevant cooperating mutations responsible for disease progression in the patients, with these sAML-specific mutations far below the level of detection of exome sequencing at the time of sample collection, years prior to clinical diagnosis of sAML. Using patient MF 504293 as an example, PDXs were established from day 532 patient samples and incubated for 12 weeks (+84 days = day 616). If samples are then processed and submitted for sequencing, assuming a 6-week turnaround for genomic analysis (+42 days = day 658), this would have identified the pathogenic subclone 390 days prior to clinical diagnosis of sAML. It could be argued that further technological advances in high-resolution sequencing could identify these potentially transforming mutations at low abundance in patients with MF during the chronic phase. However, error-corrected sequencing has shown that many potentially pathogenic mutations are acquired in people during age-related CH (44) and MDS (45), most of which are not functionally relevant without the right context. The power of this PDX system is that it is able to discriminate variants with functional relevance by selection during the xenotransplantation process. Thus, not only can this system prospectively identify rising sAML subclones in patients with MF long before clinical presentation, but it can also serve as an avatar to test novel therapeutic approaches for individual patients based on their cooperating mutations. With regard to modeling precision medicine, we hypothesized the EZH2^Y663H variant from patient MF 504293 may represent a gain-of-function variant given its proximity and similarity to the EZH2^Y646 activating mutations recurrently found in lymphoma (46). This may render the patient cells vulnerable to EZH2 inhibitors such as tazemetostat, which could be tested in the PDX model. However, our analysis showed the EZH2^Y663H variant was a loss-of-function mutation with impaired H3K27me3 methyltransferase function, consistent with previous observations regarding EZH2 mutations in myeloid malignancies (47). Nevertheless, the tractability of this system, as evidenced by our combination therapy experiments, should allow it to be leveraged to quickly test precision intervention approaches for individual high-risk cases.

As JAK2^V617F, CALR^fs, and MPL^W515 mutations all converge on JAK–STAT signaling, the JAK1/2 inhibitor ruxolitinib is currently first-line therapy for MF. Most patients experience a reduction in splenomegaly and an improved quality of life (48), but it does not eliminate the malignant clone and has little to no impact on BM fibrosis and overall survival (11). Moreover, 75% of patients with MPN taking ruxolitinib discontinue treatment within 5 years due to development of dose-limiting cytopenias and/or nonhematologic toxicities (49). Therefore, there is a pressing need for new treatment modalities. However, pharmacologic discovery in MPN has been hampered by the lack of appropriate models within which to evaluate the therapeutic effect on the key MF pathologies of BM fibrosis and MPN-initiating cell burden. We demonstrate here this PDX model can be used as a platform for clinical discovery for MF, recapitulating the patient experience with ruxolitinib of reduced splenomegaly without reducing MPN-initiating cell burden. Using this system, combination therapy of ruxolitinib with pan-PIM inhibition (currently in clinical trials for solid tumors) was able to effectively diminish MF patient cell engraftment and reduce splenomegaly. Most important, the combination of these agents was able to substantially reduce the burden of human JAK2^V617F-mutant HSCs in the BM of these mice. Although these results are encouraging, unfortunately this particular combination therapy has not proven feasible in patients with MPN, as myelosuppression has been observed with pan-PIM kinase inhibitors. In terms of specific fibrosis-inhibiting agents, this system was able to validate results from murine models regarding the efficacy of JQ1 in reversing BM fibrosis and support the continued clinical development of BET inhibitors for patients with MF (phase III clinical trial NCT04603495). It should be noted that although this PDX model may be closer to the human disease, it is premature to imply it is fully predictive of human clinical outcomes with potential therapeutic agents. Nevertheless, our experiments serve as proof of principle that this system can serve as a platform for therapeutic screening. In addition to constitutive JAK–STAT activation, other oncogenic signaling pathways, including MAPK, PI3K, and NFκB pathways, are hyperactivated in patients with MF (38, 50). Combinatorial approaches with inhibitors of these pathways, many of which are approved by the FDA for the treatment of other cancers, may yield therapeutic synergy with JAK–STAT inhibitors in patients with MPN. These types of hypotheses can be readily tested using this system due to the lack of specialized techniques and materials, and this model should represent a generalizable tool that can be quickly adapted across laboratories in hematology research.

The Institutional Animal Care and Use Committee at Washington University approved all animal procedures. All mice used in this study were NOD-scid-Il2rg-null-3/GM/SF (NSGS; The Jackson Laboratory #013062). Human cells were transplanted into sublethally irradiated (200 rads) 6- to 8-week-old NSGS mice via X-ray guided (UltraFocus100; Faxitron) intratibial injections (right leg) in a volume of 30 μL with 27-gauge U-100 insulin syringes (Easy Touch #08496–2755–01). For intratibial injections, mice were anesthetized with an intraperitoneal injection of ketamine/xylazine mixture (2 mg/mouse; KetaVed; Vedco). The needle was inserted approximately 0.8 cm deep into the tibia and cells were gradually released into the BM while the needle was gently removed. Supplementary Figure S1 details the number of CD34⁺ cells transplanted for individual patient samples.

Deidentified cord blood specimens were obtained from the St. Louis Cord Blood Bank. Peripheral blood samples were obtained from patients with MPN after obtaining written informed consent in accordance with Declaration of Helsinki ethical guidelines. Human studies were approved by an institutional review board, the Washington University Human Studies Committee (protocol WU #01–1014). Deidentified PBMCs from patients with MPN and cord blood samples were isolated by Ficoll gradient extraction according to standard procedures. CD34⁺ cells were isolated using magnetic enrichment (Miltenyi Biotec #130-100-453) and cultured overnight in SFEMII media (StemCell Technologies #09605) supplemented with Pen-Strep (50 U/mL), human SCF (100 ng/mL), human TPO (100 ng/mL), and human FLT3L (100 ng/mL).

All antibody staining was performed in Hank's Balanced Salt Solution (HBSS) buffer (Corning #21021CV) containing Pen-Strep (100 U/mL; Fisher Scientific #MT30002CI), HEPES (10 μmol/L; Life Technologies #15630080), and FBS (2%; Sigma #14009C). Briefly, BM cells isolated from tibias, femurs, and iliac crests were combined for calculating total BM from each mouse. The 1.0 × 10⁸ cells/mL were suspended in complete HBSS and incubated on ice for 20 minutes with the desired antibodies. The following antibodies were used at 1:100 dilutions: anti-mouse CD45-BV605 (clone 30-F11; BioLegend #103139), anti-human CD45-APC (clone 2-D1; BioLegend #368512), anti-human CD71-APCcy7 (clone CY1G4; BioLegend #334110), anti-human CD235a-PE (clone HI264; BioLegend #349105), anti-human CD3-FITC (clone OKT3; BioLegend #317306), anti-human CD19-PECy7 (clone HIB19; BioLegend #302215), anti-human CD33-BV421 (clone WM53; BD #565949), anti-human CD34-PE (clone 561; BioLegend #343606), anti-human CD90-PECy7 (clone 5E10; BioLegend #328124), anti-human CD45RA-BV421 (clone HI100; BioLegend #304130), anti-human CD38-FITC (clone HB7; Invitrogen #11–0388–42), anti-human lineage cocktail–APC (BioLegend #348803), anti-human CD45-biotin (clone HI30; BioLegend #304004), and streptavidin-APCcy7 (BD #554063).

Protein samples (10 μg) were separated in precasted 4% to 15% gradient SDS gels (Bio-Rad #456–1084) and transferred to nitrocellulose membranes (Millipore #IPVH00010). Membranes were subsequently probed with antibodies to detect pSTAT3 (Cell Signaling Technology #9145S; RRID:AB_2491009), pSTAT5 (Cell Signaling Technology #4322S), EZH2 (Cell Signaling Technology #3147S), H3K27me3 (Cell Signaling Technology #C36B11), total histone H3 (Cell Signaling Technology #D1H2), and β-actin (Santa Cruz #SC-47778; RRID:AB_2714189). Detection was performed using horseradish peroxidase (HRP)–conjugated secondary mouse or rabbit antibody and chemiluminescence HRP substrate (Millipore #WBKLS0100).

Mouse tibias were fixed in 10% neutral buffered formalin (Fisher Scientific #SF100–4) overnight at 4°C. Immediately after overnight fixation, tibias were decalcified in 14% EDTA for 12 days. This was followed by a series of hydration steps: 20% EtOH for 1 hour, 30% EtOH for 1 hour, 50% EtOH for 1 hour, and 70% EtOH for 1 hour. Tibias were rinsed with PBS and processed for paraffin embedding and sectioned at 5 μmol/L. Reticulin staining was performed by the Washington University Musculoskeletal Histology and Morphometry Core. Images were captured using an AxioObserver D1 inverted microscope (Zeiss) equipped with an Axiocam 503 color camera. Images were acquired with Plan-Apochromat objective at 63×, 1.4 NA objective using the ZEN 2 (blue edition) software.

Cells were resuspended at 1,200 cells/μL in PBS + 0.04% BSA. cDNA was prepared after the GEM generation and barcoding, followed by the GEM-RT reaction and bead cleanup steps. Purified cDNA was amplified for 11 to 13 cycles before being cleaned up using SPRIselect beads. Samples were then run on a Bioanalyzer to determine the cDNA concentration. GEX libraries were prepared as recommended by the 10× Genomics Chromium Single Cell 3′ Reagent Kits v3 user guide with appropriate modifications to the PCR cycles based on the calculated cDNA concentration. For sample preparation on the 10× Genomics platform, the Chromium Single Cell 3′ GEM Library and Gel Bead Kit v3 (PN-1000075), Chromium Chip B Single Cell Kit (10× Genomics, PN-10000153), and Chromium Dual Index Kit TT Set A (PN-1000215) were used. The concentration of each library was determined through qPCR using the KAPA library Quantification Kit according to the manufacturer's protocol (KAPA Biosystems/Roche) to produce appropriate cluster counts with the Illumina NovaSeq6000 instrument. Normalized libraries were sequenced on a NovaSeq6000 S4 Flow Cell (Illumina) using the XP workflow and a 28 × 10 × 10 × 150 sequencing recipe. A median sequencing depth of 50,000 reads/cell was targeted for each library. Single cell RNA-seq data were demultiplexed and aligned to the Genome Reference Consortium Human genome, CRChg38. The aligned data were then annotated and UMI-collapsed using Cellranger (v3.1.0, 10× Genomics). Cells with more than 10% mitochondrial gene expression and with top 5% unique feature counts were excluded. Principal component analysis was performed to reduce data using Seurat 3.0. Clusters were identified applying the dimensional reduction techniques (tSNE and UMAP) using the same package. Expression levels of conventional cell surface markers were examined to annotate clusters. Differentially expressed genes were identified within the cluster of interest. The fgsea package was employed to determine enriched signaling pathways.

Genomic DNA was extracted using the PureLink genomic DNA extraction kit (Invitrogen #K1820–02) and submitted for exome sequencing at the McDonnell Genome Institute at Washington University at an average coverage depth of 180×. Whole-exome sequence data were aligned to reference sequence build GRCh38 using BWA-mem (Sniffles, RRID:SCR_017619) version 0.7.10 (parameters: –t 8), then merged and deduplicated using picard version 1.113 (Picard, RRID:SCR_006525). Variants, including single-nucleotide variants (SNV) and indels, were detected using VarScan(2) (RRID:SCR_006849) version 2.4.2 (parameters: –min-coverage 8 –min-var-freq 0.1 –min-reads 2). Combined SNVs and indels were annotated by DoCM (Database of Curated Mutations, parameters: –filter-docm-variants true) and further by Ensembl Variant Effect Predictor of GRCh38 v95 [parameters: –coding-only false –everything –plugs (Downstream, Wildtype)] by providing gnomAD (The Genome Aggregation Database) and ClinVar (RRID:SCR_006169) variant calling format files. Variants were filtered by removing low-quality variants (parameters: –min-base-quality 15, –min-mapping-quality 20) and removing sites that exceeded 0.1% population allele frequency in gnomAD projects. CALR frameshift mutations were manually called using BAM files in Integrative Genomics Viewer.

Genomic DNA was extracted as above. In total, 20 ng total genomic DNA was used to perform ddPCR. All primers and probes used for ddPCR were designed as per MIQE guidelines. Amplicon context sequences used to design these probes are as follows: EZH2 p.Y663H, hg19|chr7:148,507,406–148,507,528 and TP53 p.R248Q, hg19|chr17:7,577,477–7,577,599. Briefly, a PCR reaction mixture was prepared with 10 μL 2× Supermix for Probes without dUTP (Bio-Rad #1863023), 1 μL 20× target (FAM) and wild-type (HEX) primers/probe (Bio-Rad), and 9 μL ddH₂O. The PCR sample was partitioned into ∼20,000 nanoliter-sized discrete droplets using the QX-100 Droplet Generator according to the manufacturer's instruction. These droplets were then gently transferred into a 96-well plate (Bio-Rad #12001925), and ddPCR was performed using a thermocycler with the following conditions: 95°C for 10 minutes, 40 cycles of 94°C for 30 seconds, 55°C for 60 seconds, and 10 minutes at 98°C. The mutant allele frequencies were measured and calculated using Bio-Rad QX-100 droplet reader and QuantaSoft v1.7.4 software, respectively. Each plate contained at least one wild-type DNA control (Promega #G152A).

CD34⁺ cells were cultured overnight in SFEMII media (Stemcell Technologies #09605) supplemented with Pen-Strep (50 U/mL), human SCF (100 ng/mL), human TPO (100 ng/mL), and human FLT3L (100 ng/mL). Then, 12 to 24 hours postculture, CD34⁺ cells were nucleofected with Cas9/ribonucleoprotein (IDT #1074181) complexed with gRNAs as previously described (36). Synthetic gRNA sequences were purchased from Synthego as follows: gPIM1.1: GUGGCGUGCAGGUCGUUGCA, gPIM1.2: CUGGAGUCGCAGUACCAGGU, gEZH2.1: CTTCTGTGAGCTCATTGCGC, and gEZH2.2: TTATGATGGGAAAGTACACG. A gRNA targeting the AAVS1 locus was used as a negative control: GGGGCCACUAGGGACAGGAU. Forty-eight hours after nucleofection, approximately 100,000 cells were set aside to measure CRISPR/Cas9 targeting efficiency using PCR amplicon-based deep sequencing. The remaining cells were transplanted into sublethally irradiated (200 rads) NSGS mice via X-ray guided intratibial injection.

CD34⁺ cells isolated from patients with MF were transplanted into NSGS mice. Four weeks posttransplant, xenografted animals were randomized into four groups to generate cohorts with equal levels of peripheral blood engraftment for treatment as follows: vehicle (5% dimethylacetamide/95% of 0.5% methylcellulose, 2×/d), INCB053914 (30 mg/kg, 2×/d), ruxolitinib, and JQ1 (50 mg/kg, 1×/d; MedChemExpress). Ruxolitinib was administered in chow formulation (2 g ruxolitinib/1 kg chow; Incyte #INCB018424). Effects were evaluated at stated time points by flow cytometric and histologic analysis as described above.

Student t test and one-way and two-way ANOVAs were used for statistical comparisons where appropriate (see figure legends for type of statistical test used for each experiment). Survival curves were analyzed using a Mantel–Cox log-rank test. Significance is indicated using the following convention: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. All graphs represent mean ± SEM.

Primary data are available under GEO accession number GSE160927.

M.C. Stubbs reports other support from Incyte Corporation during the conduct of the study; other support from Incyte Corporation outside the submitted work. H.K. Koblish reports personal fees from Incyte Corporation during the conduct of the study; personal fees from Incyte Corporation outside the submitted work. S.T. Oh reports personal fees from Incyte, personal fees from Sierra Oncology, personal fees from CTI BioPharma, personal fees from Abbvie, personal fees from Geron, personal fees from Constellation, personal fees from PharmaEssentia, personal fees from Blueprint Medicines, personal fees from Disc Medicine, personal fees from elgene/Bristol Myers Squibb, personal fees from Kartos Therapeutics, and personal fees from Novartis outside the submitted work. G.A. Challen reports grants from Incyte, personal fees from Incyte, grants from NIH, and grants from LLS during the conduct of the study. No disclosures were reported by the other authors.

H. Celik: Conceptualization, data curation, formal analysis, funding acquisition, investigation, writing–original draft. E. Krug: Investigation. C.R. Zhang: Data curation, software. W. Han: Investigation. N. Issa: Investigation. W.K. Koh: Investigation. H. Bjeije: Investigation. O. Kukhar: Investigation. M. Allen: Resources. T. Li: Data curation, software, formal analysis. D.A.C. Fisher: Resources. J.S. Fowles: Resources. T.N. Wong: Resources. M.C. Stubbs: Resources. H.K. Koblish: Resources. S.T. Oh: Resources, data curation, writing–review and editing. G.A. Challen: Conceptualization, data curation, formal analysis, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

We thank all members of the Challen laboratory for critical discussions and support, particularly Jake Fairchild for animal husbandry and Samantha Burkart for laboratory oversight. We thank the Alvin J. Siteman Cancer Center at Washington University for use of the Siteman Flow Cytometry Core, Tissue Procurement Core, and Immunomonitoring Laboratory, supported in part by NCI Grant CA91842 and NIH WLC6313040077. The Immunomonitoring Laboratory is also supported by the Andrew M. and Jane M. Bursky Center for Human Immunology and Immunotherapy Programs. We thank the Genome Technology Access Center and McDonnell Genome Institute at Washington University for genomic analysis, partially supported by NCI Grant CA91842 and by ICTS/CTSA NIH Grant UL1TR000448. This work was supported by the NIH R01HL147978 (to G.A. Challen), R01HL134952 (to S.T. Oh), and T32HL007088 (J.S. Fowles). H. Celik was supported by an Edward P. Evans Foundation Young Investigator Award, an American Cancer Society Institutional Research Grant, the Leukemia Research Foundation, and the When Everyone Survives Foundation. G.A. Challen is a scholar of the Leukemia and Lymphoma Society.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.