Transcriptional variability accelerates preleukemia by cell diversification and perturbation of protein synthesis

Transcriptional variability facilitates stochastic cell diversification and can in turn underpin adaptation to stress or injury. We hypothesize that it may analogously facilitate progression of premalignancy to cancer. To investigate this, we initiated preleukemia in mouse cells with enhanced transcriptional variability due to conditional disruption of the histone lysine acetyltransferase gene Kat2a. By combining single-cell RNA sequencing of preleukemia with functional analysis of transformation, we show that Kat2a loss results in global variegation of cell identity and accumulation of preleukemic cells. Leukemia progression is subsequently facilitated by destabilization of ribosome biogenesis and protein synthesis, which confer a transient transformation advantage. The contribution of transcriptional variability to early cancer evolution reflects a generic role in promoting cell fate transitions, which, in the case of well-adapted malignancies, contrastingly differentiates and depletes cancer stem cells. That is, transcriptional variability confers forward momentum to cell fate systems, with differential multistage impact throughout cancer evolution.


INTRODUCTION
Tumors evolve by genetic drift and natural selection (1,2). Acquisition of new mutations confers a probability of adaptation to new environmental pressures (3), facilitates progression and transformation of premalignant lesions, promotes metastasis, and drives treatment resistance (4). In recent years, it became apparent that nongenetic instability, in particular variability in methylation epialleles, can confer adaptive advantages to tumor growth and survival irrespective of mutations and function as driver of therapy resistance and disease relapse in hematological malignancies (5,6). Hematological malignancies, and, in particular, acute myeloid leukemia (AML), are strongly dependent on epigenetic regulation, both through mutation of chromatin factors and by co-option of unmutated chromatin regulators into maintenance of leukemogenic programs (7)(8)(9). Notably, AML has lower levels of mutations than solid tumors, supporting the notion that nongenetic events may be especially important in the former (7). Akin to genetic instability, epigenetic variability is increased in leukemia initiation and relapse but low in leukemia maintenance (10,11), suggesting that reconfiguration of molecular/ transcriptional programs may perturb the identity or survival of well-adapted leukemia cells by disrupting pro-oncogenic molecular signatures. We have recently captured this phenomenon upon loss of KAT2A (lysine acetyltransferase 2A), a histone acetyltransferase that promotes gene transcription through activation of transcriptional bursting and stabilization of gene expression levels. Kat2a loss (NULL) results in enhanced cell-to-cell transcriptional variability and progressive loss of leukemia stem cells (LSCs) transformed with the KMT2A-MLLT3 (MLL-AF9) gene fusion (12). Accordingly, KAT2A is required for maintenance of AML cell lines and in vitro self-renewal of patient AML blasts (13). At a cellular level, loss of Kat2a results in perturbation of leukemia lineage trajectories, with emergence of multiple incongruent differentiation pathways that deplete LSC but fail to uniformly differentiate leukemia cells (12). A similar pattern of incongruous exit from the stem cell state was observed upon KAT2A inhibition in mouse embryonic stem (ES) cells (14). MLL-AF9 results in an aggressive leukemia, both in mice and in humans, and requires minimal cooperativity from additional mutational events (7,15). Hence, it provides a good representation of a well-adapted leukemia, with minimal genetic and epigenetic variability. However, it does not reflect what is observed with more common forms of AML such as those associated with RUNX1-RUNX1T1 (AML1-ETO), where progression in mouse models is slow and infrequent (7,16), or clonal hematopoiesis, in which the associated mutations (e.g., in IDH1/2, TET2, DNMT3A) convey a self-renewal advantage but require additional genetic events for leukemia (7,16). In these cases, we postulate that malignant progression may be facilitated by nongenetic instability, which can be promoted through loss of Kat2a. We tested this hypothesis through investigation of two preleukemia mouse models Idh1 R132H and RUNX1-RUNX1T1[RT1(9a)] (17), which together represent up to 25% of human AML disease (7,16,18). We compared the effects of the respective mutations in the presence and absence of Kat2a and integrated functional in vitro and in vivo transformation assays with single-cell RNA sequencing (scRNA-seq) analysis, to illuminate consequences on transcriptional variability and differentiation trajectories and explain differential transformation progression.

Loss of Kat2a facilitates IDH1 R132H preleukemia transformation
First, we developed a new inducible Idh1 R132H allele ( fig. S1, A to C) and crossed it into an Mx1-Cre background ( fig. S1D), to activate the mutation in hematopoietic tissues. We verified the functionality of the Idh1 R132H allele by accumulation of the oncometabolite 2-hydroxyglutarate (fig. S1, E and F). Idh1 R132H mice develop leukemia rarely, with long latency and low penetrance, with no significant effects on overall survival (fig. S1G). In contrast, combination of Idh1 R132H with other leukemo genic mutations, namely, NRas and Npm1c (triple-mutant), results in short-latency high-penetrance leukemia development ( fig. S1G), confirming the preleukemic nature of the Idh1 R132H model. Accordingly, triple-mutant bone marrow (BM) cells, but not cells with Idh1 R132H alone, have enhanced colonyforming cell (CFC) assay-replating ability, an in vitro measure of transformation (fig. S1H). Comparison of RNA-seq from triple-mutant leukemias versus triple-mutant preleukemias, or versus Idh1 R132H alone, revealed a gene signature that was specific to the leukemia state and in which down-regulated genes were enriched for KAT2A chromatin targets ( fig. S1I). This association suggests that loss of KAT2A activity may contribute to progression of preleukemia to overt AML.
To investigate this putative contribution of Kat2a loss to preleukemia progression, we crossed conditional Idh1 R132H and Kat2a Flox/Flox mice, into the Mx1-Cre background (Fig. 1A), to generate Idh1 R132H animals that were heterozygous (HET) or NULL for Kat2a ( fig. S2, A and B). We analyzed Idh1 R132H Kat2a Flox/WT (Idh mut Kat2aHET) and Idh1 R132H Kat2a Flox/Flox (Idh mut Kat2aNULL) animals 4 and 20 weeks after Cre induction, to identify early and progressed Idh1 R132H preleukemia states. Analysis of BM stem and progenitor composition revealed no differences between genotypes or time points ( fig. S2, C to G). We did not observe differences in spleen or liver preleukemia burden (fig. S2, H and I). However, Idh mut Kat2aNULL samples had a significant advantage in CFC replating in early preleukemia (4 weeks) (Fig. 1B), which was not sustained at the 20-week time point. This could be compatible with earlier selection of preleukemia cells upon Kat2a loss, which is achieved later in Idh mut Kat2aHET animals as the Idh1 mut phenotype progresses (Fig. 1C).
In an attempt to understand whether the early replating advantage in vitro could lead to accelerated leukemia development in vivo in the absence of other genetic events, we transplanted BM cells from Idh mut Kat2aHET and Idh mut Kat2aNULL mice, into irradiated CD45.1 recipients and followed them up for 1 year. Similar to single Idh1 mut animals, we could not detect signs of leukemia development in transplanted mice ( fig. S3A). Transplants showed accumulation of granulocyte-monocyte progenitor (GMP)-like (Lin-Kit + Sca1 − FcR + ) donor cells, compatible with myeloproliferation ( Fig. 1, D and E), which was identical between genotypes. Peripheral blood counts ( fig. S3, B to D) and spleen and liver weights (fig. S3, E and F) were also similar. However, we observed the infiltration of the spleen and liver in one of three Idh mut Kat2aNULL recipients, which was not present in Idh mut Kat2aHET grafts ( fig. S3G). Notably, Idh mut Kat2aNULL cells showed enhanced colony-replating potential relative to Idh mut Kat2aHET, which was comparable to that of BM from rare Idh mut leukemic animals (Fig. 1F). Idh mut Kat2aNULL cells in CFC assays were enriched in c-Kit + Mac1 − cells (Fig. 1G) compatible with hindered differentiation and/or expansion of self-renewing cells. Overall, the results suggest that loss of Kat2a imparts leukemogenic properties to Idh1 mut cells but is in itself not sufficient to drive leukemogenesis in the absence of additional cooperating genetic events.
Similarly, Kat2aNULL cells directly tested in CFC assays upon retroviral transduction displayed enhanced replating potential. (Fig. 2F). In contrast, excision of Kat2a in RT1(9a) cells after in vitro transformation by three rounds of serial replating led to a reduction in colony formation (Fig. 2G), suggesting that Kat2a loss favors leukemia development only at a preleukemia stage. These latest observations mirror our previously identified role for Kat2a in maintenance of established leukemia stem-like cells and suggest that Kat2a plays stage-specific roles during leukemogenesis, which are preserved across leukemia models.

Loss of Kat2a results in preleukemia cellular diversification
We had previously associated Kat2a function in LSC maintenance with stability of transcriptional programs (12). Using scRNA-seq, we showed that Kat2a loss resulted in diversification and branching of differentiation trajectories and associated with enhanced transcriptional noise, particularly in biosynthetic programs (e.g., ribosomal biogenesis and translation). We asked whether similar mechanisms were at play in preleukemia progression facilitated by Kat2a loss. We hypothesized that enhanced transcriptional variability leading to program diversification might increase the probability of accessing or seeding leukemia programs, resulting in the observed acceleration in leukemia progression.
We performed scRNA-seq analysis of preleukemia cells on the 10X platform, comparing transcriptional landscapes of Kat2aNULL and

of 15
Kat2aWT RT1(9a) asymptomatic animals obtained 2 and 4 months after transplantation. We sequenced a total of 1767 cells sorted as RT1(9a)/GFP + Kit + stem/progenitor and retrieved an average of 174,770 aligned reads per cell, corresponding to medians of 5939 unique molecular identifiers (UMIs) and 1575 genes per cell described as Kit + Sca1 −/low CD34 − FcgR low (17), compatible with this last cell state. Accordingly, we detected expression of RT1 gene targets (24) specifically in this compartment (Fig. 4, A and  from different subclones within that compartment. We refer to this Ly6e low/− Cd34 − state as preleukemia progenitors (PLPs). We sought to confirm the differentiation alignment of the different cell states using RNA velocity (Fig. 4, D and E) (25). The algorithm infers differentiation trajectories on the basis of relative representation of unspliced and spliced transcript variants, and the latter inferred as estimates of the future status of the cells over a relatively fast time frame. The greater distances observed in Monocle to BAP and MAP, and in opposite direction to PLP, are captured by scarcity of intermediate velocities that nevertheless recapitulate the Monocle directionality. Directionalities within the GMP-like and PLP states are better defined, particularly in Kat2aNULL cells. Significantly, Kat2aNULL cells (Fig. 4E) show increased velocity, which is unique within the PLP compartment, and may correspond to the more evenly populated trajectories within the Kat2aNULL Monocle trajectory (Fig. 3B).

Loss of Kat2a increases transcriptional variability and destabilizes ribosomal biogenesis programs
Given the previously established association between KAT2A and transcriptional noise regulation (12,14,26), we asked whether the cellular diversification and higher trajectory velocities observed upon Kat2aNULL loss were accompanied by, and putatively attributable to, enhanced variability in transcription. Pairwise distance (27) defines highly variable genes on the basis of a mean expression-corrected coefficient of variation or distance to the median (DM) (28) and inverts gene-to-gene correlations to estimate dispersion or distance in gene expression programs. Pairwise distance has been used as a measure of global transcriptional variability, or noise (27). Perhaps expectedly, given the differential diversity of cell types observed between Kat2aNULL and Kat2aWT RT1(9a) preleukemias, global pairwise distance was increased in Kat2aNULL cells (Fig. 5A), putatively capturing cellular heterogeneity. However, the same gain in pairwise distance was observed in the individual cell states (Fig. 5B), suggesting that loss of Kat2a may affect transcriptional variability. GMP-to-PLP transition is also accompanied by enhanced transcriptional variability (Fig. 5B), supporting the notion that Kat2a loss may facilitate preleukemia progression through enhanced transcriptional noise. To understand the nature of the transcriptional programs perturbed upon (i) Kat2a loss and (ii) preleukemia progression, we performed differential gene expression analysis of the scRNA-seq dataset. Comparison of Kat2aNULL to Kat2aWT cells revealed minimal changes in gene expression levels ( fig. S7A), which were of down-regulation, as previously observed upon Kat2a loss (12). Consistent with our published data (12), differentially expressed genes between genotypes predominantly associated with ribosomal assembly and translation ontologies ( fig. S7B and Supplementary File 5), a pattern particularly prominent within PLP ( Fig. 5C and Supplementary File 6). The same ontologies were specifically down-regulated in Kat2aWT RT1(9a) PLPs compared to other cell states ( fig. S7C and Supplementary File 7), capturing a reported decrease in protein synthesis in RT1 leukemia (29). Ribosomal and translation ontologies ( fig. S7, D and E, and Supplementary File 8) were also down-regulated in Idh1 R132H mice. Together, our findings suggest a specific association of attenuated ribosomal programs with preleukemia progression, which may be further facilitated by Kat2a loss. Kat2a loss increases variability of ribosomal biogenesis programs in PLPs (Fig. 5D), which are themselves more variable than GMPs for the same programs ( fig. S7F), suggesting enhanced noise at the transition (Supplementary File 9). The gene expression range in Kat2aNULL PLPs favors lower mean values (Fig. 5E), specifically at 4 months. In support of the functional impact of the transcriptional perturbation, Kat2a loss results in decreased protein synthesis ( fig. S7, G and H).

Reduced protein synthesis activity transiently facilitates preleukemia progression
We tested the contribution of reduced protein synthesis activity to preleukemia progression by treatment with the S6K1 inhibitor (S6K1inh) PF4708671 (Fig. 6A), which impairs protein synthesis activity confirmed by reduced O-propargyl-puromycin (OP-Puro) incorporation in nascent peptide chains (Fig. 6, B and C). We treated Kat2aWT RT1(9a) cells with S6K1inh and tested their leukemia transformation potential in vitro through CFC assay replating. S6K1-inhibited cells displayed enhanced colony formation upon replating (Fig. 6D), suggesting a contribution to leukemia transformation. However, the increase in colony formation was transient and eventually lost upon subsequent replating (Fig. 6D). This suggests that the effects of reduced protein synthesis on leukemia cells may vary with progression of transformation, reconciling our data with prior analysis of established MLL-AF9 cells, in which reduced OP-Puro incorporation associated with Kat2aNULL-mediated extinction of LSCs (12). We observed a similar pattern of transient increase in colony formation of Idh1 R132H preleukemia cells treated with S6K1inh (Fig. 6E). Together, the data suggest that reduced ribosomal assembly and protein synthesis facilitate preleukemia progression. Exploration of lower levels of expression of translationassociated genes as a consequence of enhanced transcriptional variability may be instrumental in the acceleration of preleukemia to AML transition upon Kat2a loss. As leukemia progresses, variability in ribosomal biosynthesis programs may become attenuated with deviation from an optimal level no longer favorable to transformation.

DISCUSSION
In this study, we have shown that Kat2a loss facilitates preleukemia progression in Idh1 R132H and RUNX1-RUNX1T1(9a) mouse models of human disease, with acceleration of frank leukemia onset in the case of RT1(9a). Loss of Kat2a resulted in enhanced variability of transcription, leading to diversification of cell fates, including accumulation of PLP cells. In the context of an early genetic event such as RT1(9a) or Idh1 R132H , which do not allow for full leukemia transformation, the cellular heterogeneity that ensues creates the opportunity for specification and expansion of transformation-prone cells, on which additional molecular events may act to progress the leukemic process (Fig. 7). RT1 progenitor cells have been variably characterized as Kit + Sca1 + cells (17) and Kit + Sca1 − FcgR + GMP-like cells (19), with the (9a) variant denoting a Kit + Sca1 − CD34 −/low FcgR −/low phenotype (17). The variability in cellular composition is notable between individual animals ( fig. S5) and likely denotes the contribution of additional mutations to the establishment of full-blown leukemia (30), which may lose dependence on the presence of RUNX1-RUNX1T1 (29) and become sensitive to its level of expression (31,32). Cellular variability may emerge as a downstream consequence of different additional genetic events or reflect differential upstream vulnerability to specific mutations. Loss of Kat2a may facilitate the latter process. By destabilizing transcription and diversifying cellular output as one consequence of moving stem and progenitor cells out of their status quo, Kat2aNULL animals may generate additional types of RT1(9a) translocation-carrying cells able to respond to downstream mutations and/or be transformed by them. It is unlikely that Kat2a loss itself contributes to the genetic load. Kat2aNULL animals are not at a risk of myeloproliferation (12), and no recurrent KAT2A mutations have been described to date in association with hematological or solid cancers. In contrast, Kat2a ablation consistently changes cellular composition (12)(13)(14)33), making it a more likely facilitator event to generate "second hit"-responsive preleukemia cells. Future Further to or concomitantly with cellular diversification and putative differential susceptibility to secondary genetic hits, the molecular programs affected by Kat2a loss can also contribute to the leukemogenic process. Ribosome biosynthetic and translation factor genes are pervasive targets of KAT2A (12,33), and our data suggest that destabilization of translation acts to facilitate transformation at least transiently and down-regulation of translation-associated genes  may accompany preleukemia to leukemia progression. Enhanced transformation may be achieved by surveying and selection of biosynthetically quiescent cell states, which evade further diversification and respond to additional mutations with disease propagation and progression. Inspection of noise-responsive programs in chronic lymphocytic leukemia has captured ribosome biogenesis and translation as a significant prognostic module (34), and loss of ribosome biosynthetic activities plays a role in T acute lymphoblastic leukemia progression (35). The latter study also implicated decreased mitochondrial metabolic activity, which we have shown to be responsive to Kat2a loss, in leukemia development. However, established leukemia cells can be dependent on active translation for their maintenance (36), and AML subtypes, namely, those with RUNX1 mutations (37), are therapeutically sensitive to inhibition of protein synthesis. Despite the contribution of reduced or perturbed translational activity to transformation, our results suggest that the effect is transient, and we had previously observed that the inhibition of protein synthesis reduced colony formation in established MLL-AF9 leukemia cells (12). Accordingly, MLL-AF9 leukemia knockout for Kat2a displayed enhanced noise specifically in translation-associated genes, which was accompanied by reduced protein synthesis and associated with depletion of leukemia stem-like cells (12). It is possible that fully transformed, well-adapted leukemia cells buffer transcriptional variability to maintain stable self-renewal signatures and optimal biosynthetic, translation rates. In this context, instability of transcriptional programs may shift biosynthetic homeostasis and perturb cellular identity and mal-adapt leukemia stem-like cells, with antileukemia effects. Thus, stage-specific tuning and untuning of transcription and translation may be used to modulate cancer progression, a principle that can be extended to other cancer state transitions such as metastasis or drug resistance with prognostic and therapeutic potential.

Preleukemia mouse models
Mice were kept in a specific pathogen-free animal facility, and all experimental work was carried out under U.K. Home Office regulations. Animal research was regulated under the Animals (Scientific Procedures) Act 1986 Amendment Regulations 2012 following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body. Peripheral blood was collected by saphenous vein, and differential blood cells counts were determined using a Vet abc automated counter (Scil Animal Care, Viernheim, Germany).

Generation of an Idh1 R132H mouse model
Targeting vector was generated as follows using methods described previously (38). The WT Idh1 locus (endogenous Idh1 sequence, including arms of homology, was captured by gap repair as previously described). The following primer pairs were used to amplify the "U" cassette containing attR1 and attR2 gateway cloning sites containing a Zeo/PheoR selection cassette with the appropriate overhangs to allow insertion of this cassette between exons 2 and 3 of the Idh1 locus by recombineering ( Table 1).
The following primers were used to amplify the "G" cassette that contains attR3 and attR4 gateway sites flanking an AmpR cassette. Appropriate overhangs were incorporated into these primers to allow "gap repair" subcloning and retrieval of arms of homology 5′ and 3′ to exons 2 and 4 (5.9 and 3.6 kb, respectively) from the Idh1 containing bacterial artificial chromosome (Table 1).
A custom gene block (GeneArt, Thermo Fisher Scientific) containing sequence encoding the mutant R132H substitution in exon 3 was cloned into the subsequent U/G-captured intermediate-replacing the WT exon3 by standard restriction enzyme cloning using Sna BI and Csp CI to generate an R132H mutant U/G vector. A custom cDNA flanked by Afl II and Asc I sites, containing AttL1-loxP-En2SA-Idh1 cDNA exons3 to 9-SV40 pA-loxP-FRT was synthesized (GeneArt, Thermo Fisher Scientific) and cloned into the PL1PL2 containing an FRT-flanked NeoR cassette. This generated the "SA-Idh1 exon 3-9 cDNA" cassette.
These vectors and the PL3L4 vector were combined in the downstream L/R clonase reaction to successfully generate the Idh1R132H-NeoRTV vector. All intermediate and final vectors were sequence-verified. The Idh1 R132H-NeoRTV targeting vector was electroporated into mouse ES cells, and genotyping was performed for on target integration at the endogenous Idh1 locus using a series of long-range polymerase chain reaction (PCR) reactions (Table 1): Genomic DNA (gDNA) extracted from heterozygous-targeted single-cell mouse ES cell clones were subjected to Southern blot to confirm the structural integrity and confirmation of site-directed recombination of FRT and LoxP recombination sites, before microinjection. FRT recombination and removal of the NeoR cassette were mediated by expression of flippase via transient transfection of the pCAG-FlpO plasmid (Addgene, #89574). LoxP recombination and deletion of the "SA-Idh1 exons 3 to 9 cDNA" cassette was mediated by expression of Cre via transient transfection of the pCAG-Cre plasmid (Addgene, #13775). gDNA extracted from single-cell clones were subjected to Southern blot to confirm the successful removal of the NeoR resistance cassette and LoxP recombination to generate the Idh1 R132H-TV and Idh1 R132H-KI alleles, respectively ( fig. S1A). The primers that were used to generate Southern blot hybridization probes for the respective 3′ internal (FLP assay) and 5′ internal (Cre assay) are mentioned in Table 1.
Positively targeted heterozygous clones were selected for microinjection. Chimeric offspring were then selected for downstream breeding and germ line transmission of the Idh1 R132H-NeoRTV -targeted allele.
The FRT-flanked neomycin-resistant cassette, used for positive enrichment of targeted mouse ES cells, was removed by breeding to FLPe mice [as previously described (39)]. F1 mice were backcrossed to WT C57Bl6 mice, and mice negative for the presence of the RosaFLPe transgene and positive for the inducible Idh1Knock-In R132H allele were selected for downstream cohort generation by subsequent crosses with the inducible Mx1-Cre transgenic mouse model ( fig. S1C). Standard PCR genotyping for the WT and mutant alleles was performed (using the primers detailed in Table 1). Subsequent crosses to homozygous Nras G12D/G12D and Npm1 cA/cA cohorts were used to generate experimental model cohorts, as described previously (40).
To generate a Mx1-Cre-inducible mouse with Idh1 mut/WT -and Kat2a conditional knockout model Kat2a fl/fl conditional knockout mice have been previously described (12).
Retroviral construct MSCV-AML1/ETO-IRES-GFP was previously described (20). Viral particle production and transduction of the genotype-specific pools were done as described previously (12). Green fluorescent protein (GFP) levels were assessed by flow cytometry. One million BM cells obtained after transduction of Kat2aWT and Kat2aNULL pools were injected per animal into >8-week-old, C57/BL6 mice, which were lethally irradiated [2 × 5.5 gray (Gy)]. Seventeen mice per group were injected for preleukemia and leukemia studies. Leukemia survival studies were performed as two independent experiments. Leukemic mice were collected on the basis of symptoms of hunched posture, inappetence, and lethargy.

CFC assays
For analysis of preleukemia samples from RT1(9a) and Idh1 R132H , 50,000 BM cells were plated in MethoCult M3434 (STEMCELL Technologies), following the manufacturer's protocols. Colonies were scored 7 to 10 days after plating. Cells were collected from plates, washed, dispersed to a single-cell suspension, and serially replated for transformation analysis.

Leukemia maintenance in vitro
Pooled BM cells collected from two to three 12-week-old Kat2a floxed Mx1-Cre −/− animals without pIpC treatment were retrovirally transduced with RT1(9a) and serially replated in MethoCult M3434 for a total of three platings (4000 cells per plating). At plate 3, cells were collected, transduced with a MIGR-Cre-OP-Puro (Cre + ) or a MIGR-OP-Puro (empty) retrovirus, and cultured for 48 hours under puromycin selection, as described (20). Transduced and antibioticselected cells were assessed for colony formation over two rounds of plating in MethoCult M3434 (4000 cells per plating per condition) in the presence of puromycin. Colonies were scored 7 to 10 days after plating.

scRNA-seq preparation and analysis
Preleukemia BM samples were collected from individual animals engrafted with RT1(9a)-transduced Kat2aWT or Kat2aNULL pooled cells, 2 and 4 months after transplantation, and stored at −150°C. Cells were thawed, recovered in R20 medium, and sorted on an Influx sorter (BD) as Hoechst 33258-negative (live), GFP + [RT1(9a) reporter], and cKit + (early progenitors) singlets. Sorted cells were immediately used for library preparation with the Chromium Next GEM Single-Cell 3′ GEM, Library, and Gel Bead Kit v2 (10X Genomics). Libraries were quality-controlled and underwent pairedend sequencing on an Illumina NextSeq 500 Sequencer. Library preparation and sequencing were performed at Cancer Research UK (CRUK) Cambridge Research Institute. Raw single-cell RNA-seq fastq reads were analyzed using Cell Ranger software (v2.2) to obtain the cell-gene count matrix (Table 3). Preprocessing analysis yielded a gene-count matrix with 1675 cells in total with a median UMI count of 5939 and 1575 median genes per cell. The count-matrix data were preprocessed with Seurat v2.4 (41) as described (12). Each cell that expressed less than 500 genes was considered to be of poor quality and was filtered out. Differential gene expression was obtained with DESeq2 (42), as per the implementation in Seurat v2.4, for pairwise comparisons between genotypes, globally or at individual time points, between two individual cell compartments, or between one individual compartment and the remaining cells, e.g., for BAP-and MAP-associated signatures. For genes with adjusted P < 0.05, differential expression calculated using log 2 fold change (FC) was deemed as significant where |log 2 FC| > 0.26 (20% fold  S5, C and D) (22,23) and manual annotation of contiguous or discrete regions on the basis of dominant combinatorial marker expression. LMPP-like space, Ly6e + CD34 + Flt3 + CD79a − CD14 − (also Gata2 + Myb + ), was observed at the origin of the trajectories. Ly6e + CD34 + Flt3 − CD79a − CD14 − FcgR + (also Cebpa + ), Ly6e + CD34 − Flt3 − CD79a + CD14 − FcgR − (also CD19 + Il7r + Ebf + ), and Ly6e + CD34 − Flt3 − CD79a − CD14 + FcgR + (also Mafb + ) were GMP-like, B cell-affiliated, and monocyte/macrophageaffiliated, respectively. Uniform manifold approximation and projection (UMAP) plot regions were consistent between genotypespecific and global pseudo-time trajectories. RNA velocity analysis was performed using velocyto.py analysis pipeline (25). Raw fastq files were used to generate .loom files that were used for velocity calculation for individual genotypes and individual time points.

Measurement of protein synthesis
Protein synthesis rates were estimated using OP-Puro (Thermo Fisher Scientific) incorporation, as described (12). In detail, 1 million RT1(9a) Kat2aWT or Idh1 R132H Kat2aHET cells treated with 10 M S6K1inh PF4708671 versus DMSO (vehicle) were collected from successive replating of colony-forming assays and cultured for 2 hours in the presence of cytokines, mSCF (20 ng/ml), mIL-3 (10 ng/ml), and mIL-6 (10 ng/ml) in R20 medium. In other assays, Idh1 R132H Kat2aHET versus Kat2aNULL BM cells collected 20 weeks after pIpC treatment for locus activation/excision were thawed and cultured overnight in the same conditions. A final concentration of 12.5 M OP-Puro was added directly to 80% of each culture for the last hour of the culture period; the remainder 20% cells were treated with phosphate-buffered saline (PBS) and processed in parallel as control. After incubation, cells were washed with ice-cold PBS without Ca 2+ or Mg 2+ (Sigma-Aldrich) and resuspended in PBS/10% FBS for cell surface staining with c-Kit-APCC7, CD11b-biotin, and Gr1-biotin (all from BioLegend; see Table 2), followed by Streptavidin Brilliant Violet 605 (also from BioLegend), both staining steps for 30 min on ice. After washing, cells were fixed in 1% paraformaldehyde in PBS for 15 min on ice protected from light, washed, and permeabilized in PBS/3% FBS/0.1% saponin (permeabilization buffer) at room temperature, in the dark, for 5 min. Cells were washed and used immediately in the azide-alkyne cyclo-addition reaction with the Click-iT Cell Reaction Buffer Kit (Thermo Fisher Scientific, C10269) and Alexa Fluor 647-Azide (Thermo Fisher Scientific, A10277) with a master reaction solution freshly prepared for immediate use, as per the manufacturer's instructions. Alexa Fluor 647-Azide was used at a final concentration of 5 M. The reaction proceeded in the dark at room temperature for 30 min; cells were washed twice in permeabilization buffer and resuspended in PBS, 5 min before flow cytometry analysis.

Statistical analysis
Experiments were performed in triplicate, with any exceptions specifically indicated in the text or figure legends. Data are plotted as mean ± SD with statistical tests described in the respective figure legends. Statistical analysis was performed using GraphPad Prism 8.0 software. R language was used for analysis of single-cell RNA-seq data.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at https://science.org/doi/10.1126/ sciadv.abn4886 View/request a protocol for this paper from Bio-protocol. Table 3. Cell-gene count matrix specification.

Total number of cells sequenced 1767
Median number of genes per cell 1575 Kat2aWT 2 months 379