Sequential in vivo CRISPR screens identify the clonal dominance of Nf1 loss in long-term hematopoiesis Authors: Shiru Yuan, Qian Wu, Tong Yin, Lianting Chen, Baixuan Zhang, Pujiao Li, Xiaowei Xie, Guohuan Sun, Tao Cheng, Jun Wei, Hui Cheng Published: 2026 ABSTRACT Systematically interrogating the genetic regulators of long-term hematopoietic stem cell (HSC) function and their roles in malignant transformation in vivo remains a central challenge. We developed and validated a sequential in vivo CRISPR screening platform characterized by high sgRNA recovery after a 5-month reconstitution period and a pioneering pooled secondary transplantation strategy. This system not only identified enriched genes but also modeled the dynamics of clonal evolution. As a robust validation of our platform’s sensitivity, the screen identified Neurofibromin 1 (Nf1), a quintessential tumor suppressor, as the dominant hit using a focused transcription factor library. The primary screen recapitulated a myeloproliferative state. Strikingly, the secondary transplantation screen amplified this phenotype and led to malignant hematopoiesis. Thus, this 2-step approach establishes a powerful paradigm for studying the genetic drivers of long-term hematopoiesis and the evolutionary dynamics of hematological malignancies. 1. INTRODUCTION The lifelong maintenance of the hematopoietic system is sustained by a rare population of hematopoietic stem cells (HSCs), whose functional integrity is essential for health. HSCs possess dual capacity for self-renewal and multilineage differentiation, which underpins lifelong blood production and post-transplantation hematopoietic recovery.1 Dysregulation of hematopoietic stem and progenitor cell (HSPC) function underlies a spectrum of pathologies, ranging from bone marrow failure to leukemia.2 Consequently, systematically deciphering the genetic regulators that govern HSPC fate is essential to understanding hematopoiesis and developing therapeutic interventions. The advent of CRISPR-Cas9 has armed researchers with a powerful tool for functional genomics. CRISPR screens, in particular, enable the parallel interrogation of gene function and have been successfully applied in multiple contexts. For instance, this approach has yielded transformative insights in immunology, including the identification of novel targets for cancer therapy.3–5 However, its application to primary HSPCs has been fraught with challenges that fundamentally reflect stem cell biology. A central challenge lies in reconciling the need for long-term (LT) reconstitution to faithfully assess HSC function with the practical difficulty of maintaining the library over extended periods.6–8 Most studies have thus been limited to short-term analyses, which predominantly reflect the functions of short-term progenitors rather than true quiescent HSCs.9,10 Furthermore, the limited abundance of HSCs has been a recurring constraint, and the optimal strategy for LT functional screening remains unresolved.11,12 To address these challenges, we developed an in vivo sequential screening platform integrating LT reconstitution with pooled secondary transplantation. Through systematic refinement of the lentiviral vector (LVA) and donor cell strategy, we developed a system that ensures near-complete sgRNA library recovery and high reproducibility. This system is designed to identify potent regulators of HSC fate and to model disease progression dynamically. To validate this platform, we screened a transcription factor library, using the established role of the tumor suppressor Nf1 as a stringent benchmark for detecting and tracking disease-driving clones from initial expansion to malignant dominance. 2. MATERIALS AND METHODS 2.1. Mice Rosa26-Cas9 knock-in transgenic mice (Jax #026179) and wild-type C57BL/6 mice were bred and maintained under specific pathogen-free (SPF) conditions with continuous positive air pressure. Mouse bedding, feeding, and drinking water were sterilized through high-temperature, high-pressure treatment. Mice aged 8 to 16 weeks were used in all experiments. All animal procedures were conducted in accordance with institutional animal care guidelines and approved by the Institutional Animal Care and Use Committee (IACUC) of the State Key Laboratory of Experimental Hematology (SKLEH), Tianjin, China. 2.2. Cell culture HEK293T and Plat-E cells were maintained in Dulbecco’s modified Eagle medium (DMEM; Thermo Fisher Scientific, 11965118, Waltham, Massachusetts) supplemented with 10% fetal bovine serum (FBS; Gibco, 10091148, Thermo Fisher Scientific, Waltham, Massachusetts). Primary HSPCs were cultured in StemSpan™ SFEM (Stemcell, 09650, STEMCELL Technologies: Vancouver, British Columbia) supplemented with mouse stem cell factor (SCF; 100 ng/mL; Peprotech, 250-03-250, Cranbury, New Jersey) and thrombopoietin (TPO; 100 ng/mL; Peprotech, 315-14). 2.3. sgRNA vector design sgRNAs were designed using the Broad Institute sgRNA design tool (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). The retroviral sgRNA vector was constructed based on the pLMPd-Amt vector, while the LVA used in this study was derived from the lentiGuide-Puro backbone, a gift from Feng Zhang (Addgene plasmid #52963).5,13 The sequence of sgControl used in the transplantation experiment is “GGTGTTGATTTCGAGCCGTC,” whereas the sequence of sgNf1 is “ACTCTAACCATTGCAAACCA.” 2.4. Construction of the CRISPR/Cas9 gRNA library The gene set for the mouse transcription factor library, targeting 349 genes (Table S1, https://links.lww.com/BS/A161), was curated using bulk RNA-seq data from our previous study.14 Differentially expressed transcription factors were identified by analyzing RNA-seq data across various HSPC subsets. Three distinct gRNAs were designed for each transcription factor gene. The library also included 100 non-targeting control sgRNAs.15 The oligo pool consisted of 80-nt oligonucleotides with flanking sequences for Gibson cloning, based on a LVA backbone. Linearization of the backbone was achieved by BsmBI digestion. Primers overlapping the linearized termini of the BsmBI-digested vector were used for oligopool polymerase chain reaction (PCR) amplification.16 The amplified DNA products were ligated into the BsmBI-digested vector using the Gibson assembly method (NEB) and transformed into stable electrocompetent cells (NEB) to obtain the library plasmids. The coverage and skew ratio of gRNA in the final library were assessed using Illumina sequencing and analyzed using MAGeCK17 (version 0.5.9.5) (Yomebio, Wuhan, China). 2.5. Virus production and infection Lentivirus was generated by co-transfecting HEK293T cells with the lentiviral gRNA plasmids psPAX2 (Addgene #12260) and pMD2G. Similarly, retroviruses were generated by co-transfecting Plat-E cells with retroviral gRNA plasmids and the helper plasmid pCL-Eco (Addgene #12371). Viral supernatants were collected 48 hours post-transfection and stored at −80°C. Primary Cas9-expressing HSPCs (3–6 × 106 cells) were isolated from individual Cas9 transgenic mice using either a Lineage Cell Depletion Kit (Miltenyi, 130-110-470, Bergisch Gladbach, Germany) or CD117 MicroBeads (Miltenyi, 130-097-146), according to the manufacturer’s instructions. Purified HSPCs were activated in vitro prior to viral transduction in StemSpan™ SFEM supplemented with SCF and TPO. Viral transduction was performed by spin infection at 800× g at 25°C for 90 minutes with 10 μg/mL polybrene (Sigma-Aldrich, TR-1003-G, St. Louis, Missouri) in RetroNectin-coated plates (Takara, T100B, Kusatsu, Shiga, Japan). Eight hours after spin infection, the medium was replaced with fresh StemSpan™ SFEM supplemented with SCF and TPO. Transduced cells expressing ametrine were saved as input (>500× cell coverage per sgRNA). 2.6. HSPC transplantation and in vivo screening After viral transduction, HSPCs from different wells were pooled, washed, and resuspended in phosphate buffered saline (PBS) for intravenous injection into lethally irradiated (9 Gy) C57BL/6 recipient mice. To ensure the reliability and reproducibility of in vivo CRISPR screening, each recipient mouse received >8 × 106 donor cells with transduction efficiencies exceeding 30%. This corresponds to the transfer of approximately 3 × 104 LT HSCs per recipient. For a library of approximately 1000 sgRNAs, this setup achieved a coverage of approximately 30 cells per sgRNA. Five months after primary transplantation, HSPCs were isolated from the bone marrow of pooled recipient mice for deep sequencing. sgRNAs conferring a proliferative advantage or promoting LT hematopoietic maintenance were expected to be enriched in the sequencing data. 2.7. Flow cytometry Single-cell suspensions obtained from bone marrow, spleen, or peripheral blood were incubated with antibodies for 30 minutes at 4°C. Following staining, cells were washed with fluorescence-activated cell sorting (FACS) buffer (PBS containing 2% FBS) and analyzed using a NovoCyte Quanteon Flow Cytometer (Agilent Technologies, Santa Clara, California) or a BD FACSAria. Data analysis was performed using FlowJo software. For cell sorting, stained cells were sorted into 2 mL DNA LoBind® Tubes (Eppendorf, 022431048, Hamburg, Germany) using a BD FACSAria Cell Sorter. 2.8. sgRNA library preparation and sequencing Genomic DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen, 69506, Venlo, The Netherlands). The sgRNA cassette was amplified using a 2-step PCR protocol. The primary PCR was performed using KOD Hot Start DNA Polymerase (Millipore, 71086, Burlington, Massachusetts) with forward primer (TCGTCGGCAGCGTCAGATGTG TATAAGAGACAGt tgtggaaaggacgaaacaccg) and reverse primer (GTCTCGTGGGCTCGGAGATG TGTATAAGAGACAGccactt tttcaagttgataacgg). Primary PCR products of the amplified gRNA cassette were purified using AMPure XP beads (Beckman Coulter, A63881, Brea, California). A second-round PCR was performed to attach sample-specific indexes, preparing libraries for sequencing on the NovaSeq™ 6000 platform. 2.9. MAGeCK analysis FASTQ files were used to generate count tables using the MAGeCK-count algorithm, a command-line-based tool. The count tables were then used as input for the MAGeCK test algorithm. The remove-zero option was set to “none,” and all other parameters were kept at default values unless otherwise specified. Gene ranking was performed using the MAGeCK RRA (Robust Rank Aggregation) algorithm. p values for screening experiments were calculated using the MAGeCK algorithm.17 2.10. Pairwise correlation analysis Pairwise correlations were calculated using the cor() function in R with the default method and visualized using the GGally or ggplot2 package. Customized scatter plots and histograms were generated based on log2-transformed count values. 2.11. Bulk RNA-seq analysis High-throughput sequencing was performed using an Illumina platform, generating 150-bp paired-end reads. Raw sequencing data were subjected to quality control using FastQC. Subsequently, the data were trimmed and filtered using Trimmomatic to remove low-quality reads and adapter sequences. Clean reads were aligned to the mouse reference genome (mm10) using STAR (v2.7.10b). Alignment files (BAM format) were generated, and gene expression was quantified using featureCounts (v2.1.1). Differential expression analysis was performed using DESeq2 (v1.34.0). Raw count matrices were normalized, and differential expression was assessed using a negative binomial model. p values were adjusted using the Benjamini–Hochberg procedure to control for the false discovery rate. Genes with an absolute log2 fold change >1 and adjusted p < 0.050 were considered differentially expressed. Gene set enrichment analysis (GSEA) was performed using the clusterProfiler package (v4.10.0), focusing on REACTOME and Gene Ontology (GO) pathways. The aligned BAM files were sorted by coordinate and indexed using SAMtools (v1.20), and loaded into the Integrative Genomics Viewer (IGV, v2.19.7). The mm10 reference genome was used as the background, and the corresponding gene regions of interest were annotated to identify genomic loci targeted for editing. All analyses were performed in R (v4.1.2) and the corresponding scripts are available on request. 2.12. Statistical analysis Data were analyzed using Prism 9.0 (GraphPad Software, Boston, Massachusetts). Statistical significance was determined using a 2-tailed unpaired Student’s t test, 2-way analysis of variance (ANOVA), or log-rank test. Statistical significance is indicated in figures, and data are presented as mean ± standard error of the mean (SEM). n.s., no significance; *p < 0.050; **p < 0.010; ***p < .001. CRISPR screen data were processed using the MAGeCKFlute pipeline for normalization and gene hit identification.17 3. RESULTS 3.1. Optimized retroviral vectors achieve high initial transduction efficiency Building on our previous study, in which an optimized retroviral system achieved robust expression in primary T cells,5,18 we adapted this system for in vivo screening of primary HSPCs (Fig. 1A). HSPCs isolated from wild-type C57BL/6 mice were transduced with a retrovirus encoding a non-targeting control sgRNA (sgControl). We enriched HSPCs using either lineage depletion or CD117 (c-Kit) selection, yielding populations enriched for Lineage- or c-Kit+ cells with >40% purity. Both methods resulted in high transduction efficiency across Lineage-, LKS- (Lin-c-Kit+Sca-1-), and LKS+ (Lin-c-Kit+Sca-1+) subpopulations compared with non-transduced controls (Fig. 1B and Fig. S1a, https://links.lww.com/BS/A155). Optimization of retroviral transduction for HSPC screening. (A) Schematic representation of the experimental design for evaluating retroviral sgRNA transduction efficiency and long-term maintenance in HSPCs. (B) Representative histograms displaying transduction efficiency (percentage of Ametrine-positive cells) within Lineage-negative (Lineage−), LKS− (Lineage−c-Kit+Sca-1−), and LKS+ (Lineage−c-Kit+Sca-1+) subpopulations, derived from input Lineage-depleted (top panel) or c-Kit-enriched (bottom panel) HSPCs. (C) Longitudinal analysis of donor cell chimerism in the PB of recipient mice transplanted with Lineage-depleted (Lin−) or c-Kit-enriched input donor HSPCs transduced with a retroviral sgControl vector. n = 5, 2-way ANOVA. (D) Lineage distribution (B cells, T cells, and myeloid cells) within the sgRNA-carrying (Ametrine-positive) donor cell population in PB at 8 wk post-transplantation. n = 5, 2-way ANOVA. ANOVA = analysis of variance, HSPC = hematopoietic stem and progenitor cell, PB = peripheral blood, WT = wildtype. 3.2. Lineage- and c-Kit+ HSPCs are comparable donor cells for in vivo screening To assess LT viability, HSPCs transduced with sgControl were transplanted into lethally irradiated recipient mice. Peripheral blood was sampled weekly for 2 months, beginning 1-month post-transplantation. Donor cells from both lineage- and c-Kit+-enriched populations demonstrated robust engraftment, with chimerism levels exceeding 90% (Fig. 1C and Fig. S1b, https://links.lww.com/BS/A155). No significant differences in engraftment efficiency or lineage differentiation potential were observed between the 2 enrichment methods (Fig. 1D and Fig. S1c, d, https://links.lww.com/BS/A155). At 8 weeks post-transplantation, analysis of spleen and bone marrow revealed comparable total cell counts, chimerism rates, and lineage distributions (Fig. 2A–C and Fig. S2a, b, https://links.lww.com/BS/A156). These results confirm that both enrichment strategies yield viable donor cells for in vivo screening. Given its streamlined protocol, c-Kit+ selection was used for all subsequent experiments. Comparison of HSPC enrichment methods and viral vectors. (A) Total cell counts in BM and SP of recipient mice 8 wk post-transplantation. n = 5, unpaired t test. (B) Donor cell chimerism in BM and SP. n = 5, unpaired t test. (C) Lineage distribution of donor-derived cells in BM and SP. n = 5, unpaired t test. (D) Longitudinal analysis showing the decline in retroviral sgRNA representation (percentage of Ametrine-positive cells) in PB over time. n = 5, 2-way ANOVA. (E–F) Representative flow cytometry plot (E) and quantification of sgRNA-carrying cells (F) in PB, SP, and BM at 8 wk post-transplantation, illustrating the loss of retrovirally transduced cells. The top panel in (E) shows results when Lin− HSPCs were used as donor cells, and the bottom panel shows results when c-Kit+ HSPCs were used as donor cells. (G) Schematic representation of the experimental design for testing the optimized lentiviral sgRNA vector in c-Kit+ HSPCs. (H) Representative histograms illustrating transduction efficiency (percentage of Ametrine-positive cells) within Lineage−, LKS−, and LKS+ subpopulations in input c-Kit–enriched HSPCs. (I) Representative histograms of transduction efficiency (percentage of Ametrine-positive cells) in different lineages (B cells, T cells, myeloid cells) in PB of recipient mice transplanted with lentivirally transduced c-Kit–enriched HSPCs after 8 wk post-transplantation. ANOVA = analysis of variance, BM = bone marrow, HSPC = hematopoietic stem and progenitor cell, PB = peripheral blood, SP = spleen, WT = wildtype. 3.3. LVAs enable sustained sgRNA expression for effective LT HSPC screening Although retroviral transduction was initially efficient, longitudinal analysis revealed a sharp decline in the percentage of sgRNA-expressing cells in peripheral blood at 4 weeks post-transplantation (Fig. 2D). This signal loss indicates that retroviral vectors, although effective for T-cell screens,5 are suboptimal for LT HSPC studies. Retroviral vectors are largely cell cycle–dependent and require mitosis for efficient nuclear entry and integration, which may limit transduction of primitive quiescent stem cells.19–21 Consequently, progeny of non-transduced quiescent HSCs may gradually outcompete their transduced counterparts. This decline was also observed in bone marrow and spleen at 8 weeks, with the most pronounced reduction in bone marrow, the primary hematopoietic niche22 (Fig. 2E, F). To overcome this limitation, we used an optimized LVA expressing a murine PGK promoter-driven Ametrine reporter.23–25 This vector achieved high initial transduction efficiency in c-Kit+ HSPCs (Fig. 2G, H) and, crucially, a high engraftment rate with stable sgRNA expression for over 8 weeks post-transplantation (Fig. 2I and Fig. S2c, https://links.lww.com/BS/A156). 3.4. High sgRNA recovery in HSPCs after LT reconstitution validates an effective in vivo screening system Following confirmation of a stable vector system, a lentiviral library targeting 349 differentially expressed transcription factors across hematopoietic subsets was constructed based on prior data.14 As a quality control experiment, this library was transduced into c-Kit+ HSPCs from wild-type mice and transplanted into 3 recipients, achieving sgRNA coverage >1000 per mouse (Fig. 3A and Fig. S3a, https://links.lww.com/BS/A157). After 5 months, all hematopoietic subsets maintained high transduction efficiency, consistent with stable LT integration (Fig. 3B and Fig. S3b, https://links.lww.com/BS/A157). Four key subpopulations, including LKS+, common myeloid progenitors (CMPs), granulocyte/monocyte progenitors (GMPs), and megakaryocyte/erythroid progenitors (MEPs), were sorted for library analysis. sgRNA amplicons were generated and sequenced (Fig. S3c, d, https://links.lww.com/BS/A157), with near-complete sgRNA recovery across all 4 subsets (Table S2, https://links.lww.com/BS/A162). Although reconstitution altered sgRNA frequency distributions relative to the input library, frequencies across the 4 recovered subsets remained highly correlated (Fig. 3C and Fig. S3e–g, https://links.lww.com/BS/A157). This high degree of correlation underscores the reliability and reproducibility of the system for LT functional studies. Validation of the lentiviral screening system and identification of Nf1 enrichment. (A) Schematic representation of the quality control screening experiment using wild-type HSPCs to assess sgRNA library recovery and frequency distribution after long-term hematopoietic reconstitution. (B) Representative histograms of Ametrine fluorescence intensity (BV510) showing stable transduction efficiency in BM HSPC subsets 5 mo post-transplantation. (C) Correlation plot of sgRNA frequencies between different sorted hematopoietic subsets, demonstrating high reproducibility of the library distribution (***p < .001). (D) Schematic representation of the primary transcription factor CRISPR screen performed using Cas9-expressing HSPCs. (E) Representative histograms displaying the transduction efficiency of the lentiviral TF library in different subsets of input c-Kit–enriched donor Cas9-expressing HSPCs. (F) Representative histograms showing the stable maintenance of lentiviral library expression within different subsets in the bone marrow 5 mo post-transplantation. (G) MAGeCK RRA score analysis highlighting significantly enriched genes. The RRA-positive score signified the statistical significance of a gene’s enrichment and a value closer to zero denoted stronger evidence. Nf1 was identified as the top hit. BM = bone marrow, CMP = common myeloid progenitor, GMP = granulocyte/monocyte progenitor, HSPC = hematopoietic stem and progenitor cell, MEP = megakaryocyte/erythroid progenitor, TF = transcription factor, WT = wildtype. 3.5. In vivo screening nominates Nf1 as the top-enriched candidate associated with myeloproliferative disease Building on our preliminary optimizations, we conducted a primary CRISPR screening experiment. We transduced the transcription factor library into HSPCs from Cas9 transgenic mice, followed by transplantation into lethally irradiated recipient mice (Fig. 3D). The lentiviral library achieved high transduction efficiency in donor HSPCs and successfully maintained stable engraftment and sgRNA expression for 5 months (Fig. 3E, F and Fig. S4a, b, https://links.lww.com/BS/A158). Lineage−c-Kit+ (LK+) HSPCs were subsequently sorted from pooled recipients for sgRNA library analysis. Amplification of sgRNA fragments was confirmed, with near-complete recovery of sgRNAs from edited cells for subsequent investigations (Fig. S4c, d, https://links.lww.com/BS/A158 and Table S3, https://links.lww.com/BS/A162). LT selective pressures in vivo resulted in substantial shifts in sgRNA representation. However, sgRNA frequencies remained highly correlated across recovered biological replicates, confirming the reproducibility of our screening platform (Fig. S4e–g, https://links.lww.com/BS/A158). As an internal validation of the platform, MAGeCK ranked Neurofibromatosis type 1 (Nf1) as the top enriched candidate in the LT in vivo screen (Fig. 3G and Fig. S4h, https://links.lww.com/BS/A158). This enrichment is consistent with prior reports linking Nf1 inactivation to dysregulated RAS signaling and myeloid neoplasia and quantitatively reflects the large clonal advantage associated with Nf1 perturbation in our competitive LT setting.26–29 Prior genetic mouse models, such as Mx1-Cre; Nf1flox/flox mice that develop myeloproliferative neoplasms (MPNs) resembling juvenile myelomonocytic leukemia (JMML),30 have established the role of Nf1 inactivation in myeloid leukemia transformation.31–33 All 3 sgRNAs targeting Nf1 were enriched >40-fold, indicating a strong selective advantage conferred by Nf1 loss (Fig. S4i, https://links.lww.com/BS/A158). The enrichment of sgNf1 under LT competitive in vivo conditions supports the ability of the platform to identify potent regulators. 3.6. Nf1 deficiency drives myeloproliferative phenotype in vivo To independently confirm the phenotype associated with Nf1 loss identified using our screening system, we performed a targeted validation experiment. Cas9-expressing HSPCs were transduced with either sgControl or sgNf1 vector and subsequently transplanted into lethally irradiated recipient mice (Fig. 4A). While all mice in the sgControl group survived the entire reconstitution period, 10 of 14 mice in the sgNf1 group succumbed to the disease, leaving 4 surviving mice for endpoint analysis (Fig. 4B). After 5 months, both groups demonstrated robust donor cell engraftment in peripheral blood (Fig. 4C, D). The surviving sgNf1 recipient mice uniformly exhibited severe splenomegaly, a pathological hallmark consistent with previous reports on Nf1−/− fetal liver transplants28,34 (Fig. 4E). Comprehensive complete blood count (CBC) and flow cytometry analyses further confirmed a profound myeloid-biased expansion in the peripheral blood of sgNf1 mice (Fig. 4F–H). To further examine the temporal dynamics of Nf1-dependent phenotypes, we analyzed donor-derived sgNf1 HSCs at early time points after transplantation. Longitudinal flow cytometry revealed that the fraction of CD11b+ myeloid cells in sgNf1 recipients was elevated relative to that in sgControl recipients as early as 4 weeks, and this difference progressively increased at 12 and 20 weeks (Fig. S5a, https://links.lww.com/BS/A159). Corresponding plots confirmed early and sustained myeloid-biased differentiation in sgNf1 HSCs (Fig. S5b, https://links.lww.com/BS/A159). These results indicate that Nf1 loss triggers progressive expansion of myeloid-biased HSCs, highlighting the temporal dynamics underlying the myeloproliferative phenotype. A lethal myeloproliferative advantage conferred by Nf1 knockout in vivo. (A) Schematic representation of the validation experiment. HSPCs transduced with either sgControl or sgNf1 were transplanted into lethally irradiated recipients. After 20 wk of long-term reconstitution, mice were sacrificed and analyzed. (B) Kaplan–Meier survival curve of recipient mice transplanted with sgControl (n = 4) or sgNf1 (n = 14) transduced HSPCs, tracked from day 0 post-transplantation to 20 wk. Log-rank test. (C, D) Quantification (C) and representative flow cytometry plot (D) of donor cell chimerism in PB from sgControl and surviving sgNf1 mice at 20 wk after long-term reconstitution. n = 4, unpaired t test. (E) Quantification of spleen weight from sgControl and surviving sgNf1 recipient mice. n = 4, unpaired t test. (F) Monocyte count and WBC count analysis from PB, measured via CBC test. n = 4, unpaired t test. (G, H) Quantification of lineage distribution (G) within the sgRNA-carrying (Ametrine-positive) donor cells and corresponding representative flow cytometry analysis (H) in PB at 20 wk post-transplantation. n = 4, unpaired t test. (I, J) Analysis of CD4+ and CD8+ T-cell proportions within the sgRNA-carrying donor cells (I) and specifically within the T-cell population (J). n = 4, unpaired t test. All data are presented as mean ± SEM. *p < 0.050; **p < 0.010; ***p < .001. CBC = complete blood count, HSPC = hematopoietic stem and progenitor cell, PB = peripheral blood, WBC = white blood cell. Consistent with prior studies, the proliferative advantage of Nf1-deficient hematopoietic cells was most pronounced in myeloid lineages and least evident in T lymphocytes.34 Accordingly, when quantified within the overall sgRNA-carrying donor compartment (Ametrine+ cells), the relative proportions of CD4+ and CD8+ T cells were not significantly altered between sgControl and sgNf1 groups (Fig. 4I). Notably, when restricting the analysis to the gated T-cell population, sgNf1 recipients showed a modest shift toward a CD8-skewed T-cell composition in the context of Nf1-driven myeloproliferation35,36 (Fig. 4J). MPNs are characterized by a chronic inflammatory milieu that can reshape peripheral immune composition and activation states, potentially provoking bystander or compensatory expansion of cytotoxic CD8+ T cells during dysregulated hematopoiesis.37–39 Importantly, within our dataset, the T-lineage effects were substantially weaker than those of the myeloid expansion phenotype, consistent with previous observations that Nf1-deficient hematopoietic cells exhibited the strongest proliferative advantage in myeloid lineages. Similarly, sgNf1 mice in the bone marrow exhibited a myeloid-biased proliferative advantage (Fig. 5A–B). Consistent with the in vivo screening results, the sgNf1 group showed a 2-fold increase in the number of LK+ cells (Fig. S5c, https://links.lww.com/BS/A159), which was primarily driven by the expansion of LKS progenitor cells (Fig. 5C and Fig. S5d, https://links.lww.com/BS/A159). Histopathological examination of liver sections revealed extramedullary infiltration of malignant cells in sgNf1 mice (Fig. 5D). Comprehensive recapitulation of the established Nf1-deficient phenotypes further strengthened the biological validity of our CRISPR screening approach. Analysis of hematopoietic compartments in Nf1-deficient HSPCs reconstituted mice. (A) Lineage distribution within the sgRNA-carrying donor cells in BM at 20 wk post-transplantation. n = 4, unpaired t test. (B) Representative flow cytometry plot showing myeloid-biased expansion in the sgRNA-carrying cells of BM. (C) Quantification of LKS+ and LKS− progenitor populations within the donor Lin− cells in the BM, demonstrating a significant expansion of LKS− cells in sgNf1 mice. n = 4, unpaired t test. (D) H&E staining of liver sections showing extramedullary infiltration of malignant cells in sgNf1 mice. Scale bars: 100 µm (10× magnification, left panel) or 30 µm (40× magnification, right panel). (E) Volcano plot showing differentially expressed genes in sgNf1 donor-derived HSCs relative to sgControl cells after transplantation. (F) GO enrichment analysis of differentially expressed genes in sgNf1 HSCs. BM = bone marrow, GO = Gene Ontology, H&E = hematoxylin and eosin, HSC = hematopoietic stem cell, HSPC = hematopoietic stem and progenitor cell. To further investigate the molecular mechanisms underlying the clonal dominance of Nf1-deficient HSCs, we performed bulk RNA sequencing on donor-derived HSCs collected post-transplantation. To verify that HSCs used for downstream transcriptomic profiling carried edited Nf1 alleles, we inspected RNA-seq reads spanning the sgRNA-targeted region. Read-level visualization revealed recurrent indels centered around the expected Cas9 cleavage region, with a 5-bp deletion as the predominant event and lower-frequency insertions (Fig. S5e–f, https://links.lww.com/BS/A159). The indel events (1-bp or 5-bp deletions and 1-bp insertions) were predicted to disrupt the reading frame and generate a frameshifted Nf1 transcript, supporting the conclusion that the sgRNA-carrying cells harbored CRISPR-edited Nf1 alleles. Differential expression analysis revealed a distinct transcriptional signature in sgNf1 HSCs relative to controls (Fig. 5E). Specifically, we observed a significant downregulation of Gfi1 (log2FC = −1.006, p_adj <0.001), a critical transcription factor for maintaining HSC quiescence and self-renewal.40 Moreover, lymphoid-associated genes, including Flt3 and multiple members of the Igkv and Ighv gene families, were broadly downregulated, suggesting suppression of lymphoid priming.41 Consistent with previous reports,26,42 Nf1 KO HSCs in our dataset showed upregulation of multiple genes associated with RAS pathway signaling, such as Rasgrf2, Rasd1, and Prkcb. GO enrichment analysis further showed that genes upregulated in sgNf1 HSCs were enriched for pathways related to WNT signaling, mitotic nuclear division, sister chromatid segregation, myeloid leukocyte migration, and positive regulation of myeloid cell differentiation, whereas downregulated genes were enriched for lymphocyte-mediated immunity and the immunoglobulin complex (Fig. 5F). Taken together, these findings indicate that sgNf1 induces a transcriptional state characterized by enhanced proliferative- and myeloid-associated programs, activation of inflammatory/RAS-related signaling outputs, and attenuation of lymphoid-associated transcriptional features, thus providing molecular support for the myeloproliferative phenotype observed in vivo. 3.7. Secondary screening amplifies the malignant phenotype resulted from Nf1 deficiency To test the hypothesis that our system could model malignant progression, we performed a secondary CRISPR screening. Bone marrow cells from primary screening recipients were pooled and transplanted into secondary recipients. In stark contrast to the primary screen, these mice began to die just 1 month post-transplantation. By week 8, a CBC test was conducted on the surviving mice. Compared with recipient mice in the primary CRISPR screen, surviving recipients showed significantly elevated white blood cell (WBC) counts (Fig. 6A). The increased WBCs were primarily composed of granulocytes, monocytes, and eosinophils, with significantly higher numbers and proportions than in primary recipient mice.43,44 Conversely, the lymphocyte counts and proportions were markedly decreased (Fig. 6B, C). Flow cytometry further confirmed substantial expansion of the myeloid lineage (Fig. 6D). Intriguingly, the gRNA transduction efficiency in donor cells after secondary transplantation increased to almost 90%, likely due to the expansion of dominant clones (Fig. 6E). Flow cytometry revealed that the bone marrow cells in secondary recipients had an immature phenotype, with reduced expression of mature lineage markers (Fig. 6F). Giemsa staining of the bone marrow and peripheral blood smears confirmed the presence of numerous immature blasts with a high nucleus-to-cytoplasm ratio, indicating aggressive disease (Fig. 6G). Secondary screening amplifies the malignant phenotype of Nf1-deficiency. (A) CBC analysis, including WBC, RBC, HGB, and PLT results, for surviving mice from the pooled secondary screen (n = 3) compared to primary recipients (n = 9). Unpaired t test. (B, C) Absolute counts (B) and proportions (C) of different WBC lineages, derived from CBC results. Unpaired t test. (D) Quantification of lineage distribution (left panel) within donor cells in PB and a representative flow cytometry plot (right panel) from screening recipients. Unpaired t test. (E) Representative histogram showing high Ametrine-positive percentage (reflecting TF library maintenance) in peripheral blood of secondary recipients. (F) Representative histograms illustrating reduced expression of lineage markers in bone marrow donor cells from secondary recipients compared to those from primary recipients. (G) Giemsa staining of peripheral blood smears (top panel) and bone marrow smears (bottom panel), revealing an abundance of immature blasts with high nucleus-to-cytoplasm ratio. Scale bars: 40 µm (40× magnification, left panel) or 10 µm (100× magnification, right panel). (H) Scatter diagram of MAGeCK analysis from the pooled secondary screen, identifying Nf1 as the enriched top hit. Dashed lines indicate log2FC = +3 (top), 0 (middle), −3 (bottom). Red and green points correspond to significantly upregulated and downregulated genes, respectively, while gray points represent genes below the predefined threshold (|log2FC| ≥ 3 and p < 0.050). CBC = complete blood count, HGB = hemoglobin, PLT = platelet, RBC = red blood cell, WBC = white blood cell. Next, the bone marrow cells from secondary recipients were enriched with c-Kit microbeads, followed by library construction for downstream analyses. The amplification of sgRNA fragments was confirmed (Fig. S6a–b, https://links.lww.com/BS/A160), with nearly all sgRNAs recovered from the edited cells for subsequent investigations (Table S4, https://links.lww.com/BS/A162). MAGeCK-based visualization revealed a strong correlation between sgRNA frequencies across the recovered c-Kit+ subsets (Fig. S6c–e, https://links.lww.com/BS/A160). Deep sequencing of cells from secondary recipients revealed that Nf1 was overwhelmingly enriched, with its 3 sgRNAs increasing in abundance compared to the initial input library (Fig. 6H and Fig. S6f–g, https://links.lww.com/BS/A160). The overwhelming dominance of Nf1-knockout clones in secondary recipients demonstrates the amplification of oncogenic potential and the process of clonal evolution toward an aggressive lethal disease. This dramatic amplification of the malignant phenotype not only confirms the role of Nf1’s but also highlights the power of serial CRISPR pool screening to enhance the functional effects of regulators, facilitating their identification. 4. DISCUSSION AND CONCLUSION The primary contribution of this study was the establishment of a powerful and validated in vivo platform that successfully integrated critical features, including high-throughput screening, LT reconstitution, and a pooled secondary transplantation strategy. This combination provides an unprecedented tool for investigating the genetic determinants of HSC function and modeling the dynamic multistep process of malignant transformation. Our system meets the competing demands for LT and high-throughput screening. Although previous studies have provided valuable insights, they have struggled with low sgRNA recovery. LaFleur et al45 transduced LSK cells with a pooled library of approximately 20,000 barcoded sgRNAs and recovered approximately 500 sgRNAs per recipient mouse from the splenic lineages. Shankar et al46 screened approximately 7000 genes by using a library of 70,000 sgRNAs. They achieved a recovery rate of <30% for sgRNAs in HSPCs. Despite the high throughput of these libraries, the low recovery rate of sgRNAs has resulted in a considerable loss of data. Subsequent studies have shifted their focus toward reducing the number of gRNAs in the library, thereby limiting the throughput. Lara-Astiaso et al47 screened fewer than 100 genes (2 sgRNAs per gene) and required 7 separate experiments. The Van Trung lab designed a smaller library targeting 15 genes, 11 of which are well established for their strong and specific phenotypes.48 Moreover, most in vivo screens use relatively short endpoints, which predominantly reflect the biology of transient multipotent progenitors rather than the bona fide quiescent HSCs responsible for lifelong blood production.49–51 Our 5-month timeframe ensured that the observed phenotypes were driven by perturbations affecting stem cell populations. Furthermore, our results directly contribute to the ongoing debate on the optimal donor cell source. To obtain a sufficient number of HSPCs as donor cells, previous studies have resorted to extensive in vitro expansion to achieve adequate initial coverage.52 However, this process inherently favors the enrichment of genes associated with stress responses, such as those that mitigate DNA damage caused by in vitro expansion, rather than genes that promote LT hematopoietic maintenance or self-renewal in vivo.53 Consequently, sgRNA frequencies in donor cells may be biased before infusion. On the other hand, directly using a relatively pure population of HSCs without prior expansion avoids this issue, but necessitates a reduction in the size of the sgRNA library due to the scarcity of donor cells. By using a readily accessible c-Kit+ population, we demonstrated that including downstream progenitors may serve a crucial physiological role, acting as a “buffer” to meet the immediate hematopoietic demands post-transplantation. This, in turn, may shield true quiescent HSCs from proliferative stress, allowing for a more accurate assessment of their LT genetic dependence. This finding challenges the paradigm that only highly purified HSCs are suitable for such screens and offers a more practical and potentially physiological alternative. Transcriptomic and single-cell transplantation studies have revealed substantial heterogeneity among HSCs.54 Even when employing techniques such as cell sorting to enrich sufficient HSCs based on flow cytometric markers, the resulting donor cell population remains highly heterogeneous. This was evident in our initial quality control experiment in which the sgRNA library was transduced into wild-type HSPCs. As the HSPCs were not edited, they only underwent the selective pressure associated with LT reconstitution. After prolonged reconstitution, sgRNA frequencies in LKS+ cells showed a bimodal distribution (Fig. 3C). A plausible explanation for this phenomenon is that sgRNAs with higher frequencies were initially transduced into quiescent HSCs with superior LT reconstitution capacity, whereas those with lower frequencies were transduced into a more active HSC pool that contributed early, but waned over time. Furthermore, even if a homogeneous population of high-potency HSCs can be hypothetically obtained as the starting input cells, another challenge arises. Upon transplantation into recipient mice, some HSCs are forced to proliferate rapidly and differentiate to meet the immediate demand for hematopoiesis, imposing a bias in sgRNA frequency. In this study, the choice of practical c-Kit+ HSPCs, which included downstream progenitor cells, allowed these progenitors to handle the initial hematopoietic demand. This intrinsic biological variability and the substantial pressure of LT reconstitution make it difficult to achieve the same high-quality data as in vivo screens of other primary cells, such as T cells. The goal of an in vivo HSPC screening system should not be to obtain perfect data, but rather to be an effective and practical tool to identify key regulators. The striking phenotype of Nf1 loss in our system provides a compelling bridge between functional genomics and cancer biology. The rediscovered tumor suppressor Nf1 validated the biological relevance and sensitivity of our platform. The fact that Nf1-deficient clones dominated the population, particularly after secondary transplantation, quantitatively demonstrates the unparalleled clonal advantage conferred by Nf1 loss in a competitive in vivo environment. This result serves as a critical benchmark, proving that our platform can reflect the genetic regulators that shape hematopoiesis and malignant evolution in vivo. Although traditional gene-knockout models can demonstrate a gene’s potential role in disease, they often fail to capture the dynamics of clonal competition and evolution, which is the process by which a single mutated cell can outcompete its healthy neighbors to initiate cancer. However, our study also highlights an inherent challenge in hematopoietic screens: the “winner’s curse,” where highly potent driver mutations like Nf1 loss can create dominant clones that expand so aggressively and may mask the more subtle effects of other genes. These findings suggest that screening strategies should be stratified accordingly. For instance, screens designed to identify regulators of self-renewal or lineage fate may benefit from excluding known potent drivers of clonal expansion and malignant hematopoiesis. In conclusion, we developed and validated a powerful LT in vivo CRISPR screening platform that includes a pioneering secondary transplantation strategy. While maintaining a throughput of more than 1000 sgRNAs, the recovered cells exhibited complete sgRNA recovery and high reproducibility over LT reconstitution. We utilized the potent biology of Nf1 to validate our system and created a dynamic model of the stepwise progression of myeloid malignancy. This study provides a validated tool for functional genomics of LT hematopoiesis and a new paradigm for studying the evolutionary dynamics of malignancy in vivo with potential applications in identifying synergistic or sequential mutations. ACKNOWLEDGMENTS This work was supported by grants from the National Natural Science Foundation of China (82525002, 2468206, 82341214, 32230055), the Non-profit Central Research Institute Fund of Chinese Academy of Medical Sciences (2021-RC310-011), the Distinguished Young Scholars of Tianjin (22JCJQJC00070), and the CAMS Innovation Fund for Medical Sciences (CIFMS) (2024-I2M-ZD-010). The authors thank the laboratory members for their assistance with the experiments. The authors are grateful to Yomebio for their help in generating the custom-designed transcription factor CRISPR pool screen library, which was essential for our study. AUTHOR CONTRIBUTIONS H.C., J.W., and T.C. conceived the study. S.Y. performed the experiments with the help of L.C., B.Z., P.L., and G.S. Q.W. performed the bioinformatics analysis with the help of S.Y. and X.X. S.Y. performed data interpretation and wrote the manuscript with the help of all the authors. Supplementary Material