IRF4 deficiency vulnerates B progeny for leukemogenesis via Jak3 mutations resembling Ph-like B-ALL in humans

How disturbances of B cell development provoke adult B acute lymphoblastic leukemia (B-ALL) remains poorly understood. Here we describe Irf4 –/– mice as prone to developing B-ALL with age. Irf4 –/– pro/preB cells exhibited impaired differentiative but enhanced proliferative potential in vitro and accumulated in spleens of healthy Irf4 –/– mice, suggesting reduced adherence to the IL-7 providing bone marrow niche. Thus selected, pro/preB cells transformed acquiring proliferative IL-7 independency through Jak3 gain-of-function mutations. Targeting JAK signaling with Ruxolitinib in vivo prolonged survival of mice bearing established Irf4 –/– leukemia. Intriguingly, organ infiltration including leukemic meningeosis was selectively reduced without affecting blood blast counts. As low IRF4 expression and JAK3 mutations also characterize a subpopulation of Ph-like B-ALL in adult humans, our results imply Irf4 –/– mice as a suitable model for investigating preleukemic conditions in adults. Using this model, we identified an unexpected effect of Ruxolitinib treatment in B-ALL.


Introduction
Two counteracting signaling pathways via the Interleukin-7 receptor (IL-7R) and the preB cell receptor (preBCR) ensure an orderly progression from proB to preB cells during B lymphopoiesis. [1][2][3] ProB cells adhere to bone marrow (BM) stromal cells (SCs) expressing CXCL12 and VCAM-1 through CXCR4 and VLA-4, respectively. 4-7 IL-7 from SCs secures proliferation and further promotes adherence of proB cells via upregulation of focal adhesion kinase (FAK). 5 The formation of the preBCR composed of Igµ protein (resulting from the productive rearrangement of the IgH locus) and the surrogate light chain CD179b, marks the entrance to the preB cell stage. Signaling via the preBCR in turn induces the transcription factor interferon regulatory factor 4 (IRF4) which is also highly important during T cell differentiation. 8,9 Upon induction in preB cells, IRF4 halts cycling, induces migration away from the IL-7 rich niche, and facilitates recombination of the Igκ or Igλ locus by the RAG1/2 recombinases. 1 Despite this central role, Irf4 -/mice still have effective, albeit lowered, development of membrane-bound (m)IgM + mature B cells, 10 likely due to a partially redundant function of IRF8. Accordingly, Irf4,8 -/-B cell progenitors are completely arrested at the preB cell stage. 11 Importantly, disruption of this developmental track can provoke progenitor B acute lymphoblastic leukemia (pB-ALL). In humans, this disease preferentially affects children (age 0-19) with a four times higher incidence when compared to adults. 12 Most deaths however occur in the adult population, with the relative survival rate decreasing progressively with the age at diagnosis. 12 Not surprisingly, leukemia biology also differs between children and adults.
Some driver mutations like the ETV6-RUX1 rearrangement are almost exclusively found in ALL of children. Furthermore, clonal expression of drivers like MLL rearrangements has even been traced back to birth in a pediatric ALL patients cohort. 13 In contrast, these typical hallmarks of ALL in children are rarely encountered in adults. Instead, cases affecting adolescents and young adults (AYA) display more acquired mutations per tumour cell and a different set of driver mutations. [14][15][16][17] A famous driver in AYA-patients is the Philadelphiachromosome (Ph), a t(9;22) translocation generating the fusion protein BCR-ABL1. [18][19][20] Phlike-(or BCR-ABL1 like) ALL also preferentially affects AYA-patients and is marked by transcriptomic changes similar to Ph + -B-ALLs without expressing the fusion protein. [21][22][23][24][25][26][27] Instead, driver mutations often target CRLF2, EPOR or JAK2, while some mutations categorized as "other JAK-STAT" are involved in IL-7 signaling. 28 Importantly, preleukemic changes leading to AYA-B-ALL are unclear and animal models scarce.
Herein, we report for the first time that adult Irf4 -/mice spontaneously develop pB-ALL with similarities to AYA B-ALL. A previous report described early onset B-ALL in Irf4/Irf8 -/codeficient mice and used Irf4 -/mice as negative controls. 29 Our findings indicate that IRF4/8 double deficiency merely speeds up a process, which can be initiated by single IRF4 deficiency. Because one preleukemic alteration might better reflect the human situation and because Irf4 -/-(but not Irf4/Irf8 -/-) mice productively develop B cells, they are better suited to examine spontaneous B-leukemogenesis. We herein delineate the steps from disturbed Irf4 -/-B lymphopoiesis to overt pB-ALL, while comparing it with human AYA-B-ALL and deriving targeted drugs for treating established disease in vivo.

Irf4 -/mice spontaneously develop preB cell leukemia
Following the serendipitous finding, that some aged Irf4 -/mice developed swellings and succumbed to a consuming disease, we systematically observed 80 Irf4 -/mice with an age > 150 days. We detected 14 rapidly progressing tumours (4 in female and 10 in male mice, incidence thus: 17.5 %), that spontaneously appeared in lymph node (LN) areas (mean age at diagnosis: 268 days, median: 238 days, Fig.1a). Tumours were neither detected in mice younger than 150 days nor in C57BL/6 wild-type (wt) mice housed in the same room of our facility. Incidence and delayed disease onset are reminiscent of other spontaneous genetic mouse tumour models such as the KC pancreatic cancer model. 30 A representative tumor in situ (asterisk) is depicted in Fig.1b. All tumors were accompanied by lymphadenopathy (arrow heads) and increased spleen size, compared to healthy Irf4 -/or wt mice (Fig.1c). Suspected lymphomatous origin was corroborated microscopically (Fig.1b), with dense infiltration of mononucleated cells also in the BM, lung, and liver (Fig.1d). Due to the known impaired maturation of Irf4 -/-preB cells, 10 we reasoned that eruption of preB-leukemia might be the explanation. In a representative spleen section (Fig.1e), infiltrating cells stained positive for B220 and Igµ, (but less than untransformed "follicle" B cells), and highly positive for the proliferation marker Ki67. By flow cytometry, tumour cells infiltrating the spleen (bottom panels, marked with asterisk) expressed no mIgM and low amounts of both B220 (confirming histology) and CD2 (Fig.1f). Further, all tumours expressed λ5 (CD179b; Fig.1f) and CD19 (not shown). In addition, Igµ, but not Igκ or Igλ were found intracellularly (not shown). Flow cytometry also demonstrated tumour cells in peripheral blood (pB), BM and LN ( Fig.1g; two exemplary tumours "T8" and "TD2"). Considering the BM infiltration and the devastating progression, these attributes characterize the disease as preB-ALL.
To prove clonality, we performed PCR on the VDJ junctions of the IgH region in three tumours, cloned the amplicons and sequenced multiple clones independently (Table 1). Almost all sequences per tumour were identical, demonstrating clonality. Few detectable alternative sequences probably originated from contaminating non-transformed B cells.
As further characteristic of malignancy, the tumours (two examples) displayed copy number variations (CNV) (Fig.1h), i.e. subchromosomal alterations. Finally, only 500 transferred tumour cells readily elicited disease in wt mice already after two weeks, as tested for tumour 8 (T8) and three other tumours.

Preleukemic alterations affecting Irf4 -/-B lymphopoiesis
The uniform appearance of preB-ALL in Irf4 -/mice suggested a defined preleukemic pro/preB cell vulnerability to immortalization, possibly related to the partial maturation defect. Most likely, a second, possibly again uniform, acquired genetic alteration was necessary for bona fide leukemia development and arose with low frequency per time, explaining the affected age and relatively low penetrance.
Thus, IRF4 deficiency skews B lymphopoiesis towards less differentiated cells, while still principally allowing maturation, because Irf4 -/-G2 cells remained detectable (Fig.2a). This disturbed, but nevertheless productive B cell maturation confirms and extends previous reports. 10 Following the Hardy classification of B cell progeny (Fig. 2b) based on B220 and mIgM expression 31,32 we found the expected decrease in recirculating mature Irf4 -/-B cells (Fig.2b, "F"), accompanied by a relative increase in pro/preB cells (Fig.2b, "A-D"). To pinpoint the exact developmental hindrance, we measured expression of IL-7Rα and of CD2, which accompanies cytosolic Igµ expression (Fig.2c). 33 Importantly, we noted an increased frequency of preproB and CD2 -/dim IL-7Rα + B220 + mIgM -proB cells in Irf4 -/mice. The respective definition of cell state was corroborated by comparing cell size measurements per subpopulation (Fig.2d). These results characterize the proB to large preB cell step as the definitive bottleneck state in Irf4 -/-B lymphopoiesis.
Cells with proliferative potential and stagnated differentiation are vulnerable for malignant transformation. To directly demonstrate selective proliferation of IL-7Rα + Irf4 -/-proB cells without further differentiation, purified BM B220 + cells from Irf4 -/and wt mice were cultured with IL-7 (Fig.2e,f). After three days, Irf4 -/cells had expanded three-fold, whereas wt cell numbers remained stable. A substantial part of the surviving wt B cells had matured under IL-7 influence, shown by a population coexpressing mIgM and B220 (Fig.2e, bottom right). Part of the wt B cells remained in an immature cell state, characterized by surface 5 expression ( Fig.2f). In contrast, all expanding Irf4 -/-B220 + cells remained mIgMand therefore more immature, also supported by a greatly increased fraction expressing 5 (Fig.2f). The observed relative increase in 5 + cells (Fig.2f) between Irf4 -/and wt after culturing with IL-7 adds to the observed absolute three-fold increase in cell numbers. Thus, IL-7 culturing further accentuated the differentiation impairment beyond the proB/large preB checkpoint (Fig.2c). High IL-7 driven Irf4 -/-B cell proliferation was corroborated by Ki67 expression (sFig.1a), cluster formation and larger size compared to wt B cells (sFig.1b). Notably, IL-7 dependent Irf4 -/pro/preB cell proliferation was fully blocked by NIBR3049, 34 a selective inhibitor of JAK3 -associated with ɣ-chain cytokine signaling downstream of the IL-7R (sFig.1a). 35 These data indicate particular significance of Irf4 -/-preB cell vicinity to IL-7 producing SCs in the BM, which in vivo is secured via interaction of CXCR4 on pro/preB cell with the SC-derived chemokine CXCL12. 6,36 Notably, while Irf4 -/and wt pro/preB cells expressed CXCR4, the intensity on Irf4 -/cells was markedly lower (Fig. 2g), suggesting reduced adherence to SCs.
Together with increased IL-7Rα + Irf4 -/pro/preB cell number, these results implied that some cells might lose access to BMSCs and drain into peripheral organs. Indeed, we detected B220 + mIgM -CD2 -/dim pro/preB cells at strongly increased frequencies within spleens of healthy Irf4 -/compared to wt mice (Fig.2h). Interestingly, the frequency of pro/preB cells in the spleen also increased with age in Irf4 -/mice.

Recurrent signaling pathway alterations in Irf4 -/and human B-ALL
Importantly, peripheral pro/preB cells should face a strong survival stress due to the distance from IL-7 producing SCs. Perhaps, this circumstance was the decisive selection pressure for bona fide leukemogenesis, whereby genetic alterations providing IL-7-independency would ascertain aggressive preB cell proliferation. To test for mutations in the IL-7R pathway, we performed whole exome sequencing (WES) of three independent tumours (T8, T10, T11) compared to sorted B220 + mIgMcells from Irf4 -/-BM. Comparing the single nucleotide polymorphyisms (SNPs) between the three samples, eleven genes were affected in all three tumour samples (Fig.3a). Among these, Jak3 is directly involved in mediating signaling via gamma-chain cytokines such as IL-7. All three tumours carried Jak3 mutations with a frequency of 0.5, probably because the mutations affected one allele in all bulk cells (Table 2).
Using Sanger sequencing of Jak3, we found two additional mutations in tumour samples TD1 and TD2 (Fig.3b, Table 2). Thus, Five out of five tested primary tumour samples carried Jak3 mutations. Interestingly, although different SNVs occurred, all targeted either the active kinase domain or the pseudokinase domain regulating JAK3 activity. Some of these SNVs are already known from other leukemia mouse models. 37 To clarify the significance of the mutations for IL-7 signaling, we cloned the JAK3 R653H and JAK3 T844M mutants into a retroviral (RV) expression plasmid, which labels transduced cells with Thy1.1. We then transduced IL-7 dependent, Eµ-Myc transgene driven murine preB-leukemia cells 38,39 with these RVs and cultured them with or without IL-7. Without IL-7, mutated Jak3 transferred a clear growth advantage forThy1.1 + compared to Thy1.1cells not seen after transduction with wt Jak3 or empty vector (EV-)RV (Fig.3c). Strikingly, this growth advantage was lost in the presence of IL-7. Thus, the Jak3 mutations are gain of function mutations (GOFs) providing IL-7 independency. Furthermore, selection of JAK3-mutated cells is supported by the lack of IL-7 in vitro and therefore potentially by distance from BMSCs in vivo.

Aicda is upregulated in Irf4 -/-preB cells by LPS and lack of IL-7
Next, we focused on the mechanistic pathways of tumourigenesis. Because five out of six Jak3 mutations were produced by cytosine to thymine exchanges (Table 2), we suspected a more specific mutagenic cause. Many DNA editing enzymes including the APOBEC family member AID can introduce single nucleotide mutations. 40 AID transforms C:T base pairs into U:T mismatches, 41,42 e.g. during somatic hypermutation in germinal center B cells. [43][44][45] Thereafter, repair mechanisms most often ultimately cause C to T conversions. 46,47 Notably, AID is induced in wt preB cells by IL-7 withdrawal and LPS stimulation and was implicated as a facilitator of human B-ALL. 48 Therefore, we compared Aicda expression in sorted Irf4 -/and wt B220 + mIgMpro/preB cells to wt mesenteric LN cells as positive and CD4 + TH1 cells as negative control.
Individual leukemia samples were also included. While mLN cells highly expressed Aicda, pro/preB cells also expressed readily detectable amounts (Fig.3d). Interestingly, expression was significantly higher in Irf4 -/than in wt cells, contrasting earlier results in stimulated mature B cells. 49 Furthermore, like their wt counterpart, 48 in vitro expanded (see Fig.2e) Irf4 -/pro/preB cells upregulated Aicda further under LPS treatment or independently with IL-7 withdrawal, while IL-7 suppressed Aicda (Fig.3e). This finding poses an attractive explanation as to how BM evasion and potential exposure to microbial pathogens might initiate mutagenic processes in vulnerable Irf4 -/-preB cells cooperatively.
To test if the observed Aicda upregulation correlated with frequent C:T/G:A transitions in Irf4 -/leukemia on a genomic level, we performed WES comparing Irf4 -/and wt pro/preB cells sorted from BM and T8 tumour cells with matched tail-tip samples (Fig.3f). We found a marked preponderance of C:T/G:A transitions in the T8 sample ( Fig.3g), both in all detectable SNPs and when filtering on the core mutations likely reflecting mono-or bi-allelic appearance in all bulk cells. This finding further supports a specific mutagenic process, potentially involving AID, explainable by infectious stimuli and IL-7 withdrawal.

IRF4 re-expression leads to cell death and expression of differentiation markers
As our model was based on Irf4 deletion as a prerequisite for leukemia, we examined the effect of forced IRF4 re-expression. We established cell lines (T8-1 and T8-2) from the representative, JAK3 mutation carrying, tumour 8 and produced RVs coding for GFP with IRF4 (IRF4-RV) or without (EV-RV). When re-introducing IRF4 into T8-1 (Fig.4a) and T8-2 cells (not shown), GFP + IRF4 expressing cells gradually disappeared over time (Fig.4a). AnnexinV/PI stainings confirmed apoptosis (not shown). EV-RV had no such effect (Fig.4a). We then compared the transcriptome of the cells 24 h after transduction, when viability was still unaltered (Fig.4b,c). Intriguingly, before entering apoptosis, cells re-expressing IRF4 exhibited transcriptomic signs of differentiation beyond the preB checkpoint, such as downregulation of Ccnd3 and Igll1 (coding for CD179b), but upregulation of Rag1 and germline Igkv transcripts ( Fig.4b). Therefore, fully transformed leukemia driven by Jak3 GOF mutations remained targetable by IRF4 re-expression.

Ptk2 is upregulated in Irf4 -/leukemia and cells depend on clustering for survival
To investigate, whether IRF4 deletion might have further oncogenic effects apart from promoting Jak3-GOFs, we compared gene expression of three tumours with that of sorted Irf4 -/-B220 + mIgMpro/preB cells. Differentially expressed genes were then reduced to only those altered by IRF4 re-expression (Fig.3b,c) and matched with a published ChIPseq of genes bound by IRF4 in B cells. 50 The resulting genes (Fig.5a) were particularly notable for Ptk2 (encoding focal adhesion kinase, FAK), which was upregulated in leukemia cells (Fig.5a "Tumour vs. BM") and suppressed by IRF4 overexpression ("IRF4-RV vs. GFP-RV"). FAK integrates signals generated through cell adhesion with relevance for cancer cell survival in solid tumors. 51,52 Interestingly, FAK also regulates the localization of pro/preB cells in their BM niche 5 and is associated with stromal adhesion in BCR-ABL + B-ALL. 53 We focused on FAK also, because cultured T8.1 cells (Fig.5b) and Irf4 -/pro/preB cells formed striking cell clusters (sFig.1b). Apparently, these clusters were needed for cell survival, as demonstrated by marked reductions in viability at lower cell densities even under supply with exogenous IL-7 (Fig.5b).
Higher FAK quantities in tumour versus untransformed mIgM -Irf4 -/pro/preB samples were confirmed on a protein level using Western Blot, while mature mIgM + B cells of Irf4 -/or wt genotype contained the lowest amounts (Fig.5c).

Small compound agents including inhibitors of FAK and JAKs kill Irf4 -/leukemia cells in vitro
To complete our analysis of signaling pathways, we screened kinase inhibitors for their capacity to kill Irf4 -/leukemia cells in vitro. We included Ruxolitinib, an inhibitor of JAK1/2 downstream of JAK3, Defactinib targeting FAK, and Dexamethasone, a cornerstone for treating lymphomatous malignancies. Furthermore, we included inhibitors of NFκB (IKK, TAK1), JNK, MEK, ERK, PP2A, GFI1 and of the Bruton tyrosine kinase (BTK) acting downstream of the BCR.

A variety of these substances potently killed tumour cells (sFig.2a). Efficacy of Ruxolitinib and
Defactinib corroborated our genetic results. The strong activity of dexamethasone was also expected. Furthermore, inhibitors of GFI1 and PP2A as well as NKκB and JNK (both also active in a diffuse-large-cell-B-cell-lymphoma model) 54 were very potent. In contrast, inhibiting BTK, MEK and ERK had no impact on cell survival. We also found clear synergistic effects between Ruxolitinib and Dexamethasone (a combination successfully tested in a mouse model of T-ALL) 55 and between Defactinib and Dexamethasone or the JNK inhibitor (sFig.2b).

Irf4 -/mouse leukemia shares resemblance with a subset of human Ph-like ALL
Next, we wanted to address the question, whether the herein described correlation between IRF4 deletion and secondary Jak3 GOFs is transferable to human disease. Progenitor B-ALL has been classified into six groups, five of which are defined by clear genetic drivers. Within the sixth group, hierarchical clustering identified patients with similar transcriptomic profiles as BCR-ABL1 carrying (Ph + ) ALL. Being BCR-ABL1 negative, this disease subset has thus been termed BCR-ABL1 like or Ph-like ALL. 22,25-28,56-58 A previous report elucidating the genetics of this subset of patients found recurrent genetic alterations in signaling kinases that partially explained the Ph-like transcriptomes. 28 Especially, JAK2 and CRLF2 were altered in many cases. A further subset of Ph-like ALL however, defined as "other JAK-STAT" (sFig.3a), harbored mutations affecting IL-7 receptor signaling, including alterations in JAK3 and JAK1.
In the 154 patients analyzed, 2.6 % of all Ph-like patients carried JAK3 mutations and none of these carried detectable gene fusions, perhaps representing a further subgroup. 28 Because of the homogeneous presence of GOF Jak3 mutations in our model and because of its clear protein kinase-dependent, but gene-fusion negative oncogenesis, we considered Ph-like-B-ALL as particularly attractive for comparison. Indeed, while B-ALL preferentially arises in children, the incidence of Ph-like-B-ALL increases from 10 % in children to above 25 % in AYA and adults (sFig.3b), 28,57 reminiscent of the older age in affected mice. Furthermore, the Ph-like ALL study had reported drastic reductions of IRF4 transcripts in the published dataset of 154 Ph-like-B-ALL cases, when compared to all other B-ALL subtypes (sFig.3c). 28 Finally, to further compare our disease model to human Ph-like ALL, we sought to apply the set of 38 genes, used to initially contrast Ph-like ALL against other forms of B-ALL in the COALL (German) and DCOG (Dutch) ALL study cohorts (referred to as "Den Boer gene set"). 24 Out of these 38 genes, transcription of only 15 genes was reported in the supplemental information to the Ph-like study we used as reference. 28 When comparing regulations of these remaining 15 genes between the two scenarios "Ph-like vs. other B-ALL" 28 and "Irf4 -/leukemia vs. Irf4 -/progenitor B cells" (our study), we found that our scenario correlated with similar expression changes compared to Ph-like ALL patients (sFig.3d); perhaps due to the homogenous presence of Jak3 GOFs. Not all 15 genes were regulated in our scenario, but at least none of the genes showed obvious differences in regulation compared to the Ph-like ALL patients.
Taken together, these findings suggest sound relevance of our data to human B-ALL. Intriguingly, our model probably resembles only a small subset of Ph-like ALL, hinting at the quite substantial notion, that human Ph-like ALL itself encompasses a still diverse cancer biology.

In vivo therapy of established Irf4 -/-B-ALL
We then attempted to apply our knowledge in vivo. In preliminary studies, i.p. rather than s.c.
transfer of T8.1 tumour cells led to homogenous appearance of leukemia. We adoptively transferred 3 x 10 5 T8.1 cells i.p. into wt mice and started treatment around day 14 thereafter, when overt leukemia (Fig.6a) and deterioration was observed. To imitate clinical protocols, we started induction therapy with Dexamethasone. Within one week, leukemic cell numbers in pB were robustly reduced, although few cells remained detectable, again with very little variability ( Fig.6a). We then started maintenance therapy with Ruxolitinib or vehicle control by oral gavage twice daily ( Fig.6a-d). In a similar experiment, Defactinib was tested (sFig.4a-c). Both substances were selected from our panel because clinical experience exists for both. [59][60][61][62] Importantly, the half-life of Ruxolitinib in mice is only 0.8 h (from the "Australian Public Assessment Report for Ruxolitinib", by the Department of Health of the Australian Government), while oral bioavailability of Defactinib is also low in mice. 63 Therefore, any effectiveness in our model likely underestimates the potential in humans. To limit stress burden, we ended gavage after 12 days but further monitored mice.
Despite maintenance therapy, leukemic cells in pB reappeared, with little difference between treatment groups (Fig.6a). However, we detected a clear survival benefit for mice treated with Ruxolitinib ( Fig.6b) and less so also with Defactinib (sFig.4a). Surprisingly, sick sham-treated mice showed a striking phenotype: within seconds after gavage, temporary limpness of tail and hind legs occurred, reminiscent to experimental autoimmune encephalomyelitis. Video supplements 1 and 2 show a mouse before and 30 s after treatment intervention. Initially, paraparesis was only triggered by our intervention and vanished shortly after, but during disease progression the symptom's duration increased, until mice presented permanent paraparesis and had to be sacrificed. We established a scoring system quantifying duration of paraparesis (from 0 to 3, see Methods for details). Most importantly, this symptom was almost abolished during Ruxolitinib treatment (Fig. 6c,d), and considerably relieved by Defactinib (sFig.4b,c). We reasoned that this symptom might be due to the increased intraabdominal pressure exerted during gavage. If plexus nerves or spinal nerve roots of the lumbar region were to have a preexistent lesion, this temporary increase in pressure might explain the transient paraparesis through mechanical stress induced reductions in conductivity.
To investigate the basis for this symptom, we performed high-resolution ultrasound. Strikingly, mice with score 3 displayed an echogenic and heterogeneous paravertebral mass in the psoas muscle region (Fig. 6e) which was significantly smaller in score 0-1 mice (Fig.6f) and appeared less echogenic (Fig.6g). By histology, score 3 correlated with severe infiltration of mononucleated blasts into muscles, but more importantly also intrathecally around the spinal cord (X in Fig.6h), extending into the lumbar spinal nerve roots (arrowhead in Fig.6h).
Infiltration was decisively reduced by Ruxolitinib treatment (score 0 in Fig.6h), resembling the picture of tumour-free mice (left panels). A quantification of differences in perimyelon infiltration (X in Fig.6g) is shown in Fig.6h. As for BM, leukemia cells abolished normal hematopoiesis almost totally in sham-treated compared to tumour-free mice (CAE staining in Fig. 6g).
Interestingly however, and in contrast to neural infiltration, BM infiltration was only partially resolved by Ruxolitinib.
These findings raise the possibility that Ruxolitinib and Defactinib preferentially target infiltrated solid organs and leukemic meningeosis rather than BM or pB. In support of this idea, Ruxolitinib treatment fully blocked liver infiltration as compared to tumour-free or sham-treated mice (sFig.4d,e). As tissue infiltration is regulated by homing receptors on invading cells, we treated T8.1 cells with Ruxolitinib in vitro and checked the expression of CD29 (integrin β1), which pairs with various integrin alpha chains involved in multiple cell interactions and tissue adhesion. 64,65 Notably, T8.1 cells expressed CD29 and Ruxolitinib reduced this expression dose-dependently (sFig.4f) while it even slightly increased expression of MHC I molecules (H2D b , H2K b ), stained as specificity control. Possibly therefore, Ruxolitinib modulates the expression of adhesion molecules needed for extravasation into parenchymatous organs. If so, the compound might hinder infiltration of tumour cells into solid tissue more efficiently than into BM, thus explaining persisting blasts in the blood.

Discussion
It has been previously demonstrated that Irf4/Irf8 -/as well as Irf4/Spi1 -/mice early develop leukemia at high rates. 29,66 In contrast to these reports, we report that single deficiency for IRF4 fully suffices to establish leukemia -albeit with less penetrance and only in older mice. We found unchanged expression and gene sequence of IRF8 (not shown) and normal amounts of Spi1 transcripts (sFig.5a) in Irf4 -/pro/preB cells. Similar tumours were also described in Pax5 +/mice, a transcription factor operating at a similar B lymphopoiesis checkpoint. 67 Importantly, Pax5 expression is even upregulated in Irf4 -/pro/preB cells (sFig.5b), thus excluding that our model simply reflects reduced PAX5 levels.
As PAX5 controls IRF4 expression in preB cells, 68 Pax5 +/leukemia may in turn depend on reductions in IRF4. If so, two scenarios exist. First, the Pax5 +/genotype might imitate an IRF4 knockout because regulatory complexes binding the Irf4 locus could depend on a certain threshold amount of PAX5. For IRF4, such a scenario exists, because some regulatory programs are fully inactive even with only lowered IRF4 expression. 50 Second, Pax5 +/could cause intermediate (not fully lacking) IRF4 expression comparable to Irf4 +/mice. Since we found only one spontaneous Irf4 +/leukemia, any directly linked tumorogenesis of the two factors would then attribute the higher incidence in Pax5 +/mice to the intensely employed microbial stimuli as a difference in the two model environments. 67 Deficiency of PAX5 or IRF4 could, however, also cause tumours independently and phenotypic similarities could stem from individual activities on a related developmental checkpoint. Taken together, our study stresses the general vulnerability of pro/preB cells to immortalization. Secondary steps for mutagenesis may include infectious agents as recently found empirically in the Pax5 +/model 69 and also supported by our finding of Aicda upregulation by LPS.
Most importantly, when reviewing published data of B-ALL patients, 28 we found that the Phlike B-ALL subtype presents with lowered IRF4 (sFig.3c) and PAX5 expression (sFig.5c) and includes mutations in JAK3 in specific subtypes. Furthermore, we showed that within the Phlike predictor gene set devised by Den Boer et al., 24 our leukemia model shared considerable resemblance with a Ph-like patient cohort. (sFig. 2d). However, Ph-like B-ALL may encompass further subtypes, only part of which our model imitates: For instance, while we report recurrent JAK3 and never JAK2 GOFs, the human disease more commonly includes CLRF2-and JAK2and less JAK3-alterations. Potentially, different pathways predispose for leukemia at differing pro/preB cell checkpoints. In our study, IRF4 deficiency predisposes for independence of IL-7 via JAK3 GOFs, because the Irf4 knock-out renders cells vulnerable to transformation at precisely that differentiation stage, where IL-7 controls cell-cycling. JAK2 does not act downstream of the IL-7Rchain complex but of CD117 (c-kit) and TSLPR (encoded by CRLF2) instead. 73,74 JAK2 GOF carrying cells could therefore represent a differing subset of progenitor B cells that depends more heavily on TSLPR or CD117 signaling. Thus, a similar selection pressure as in Irf4 -/mice but affecting a different pathway, might provoke mutations in JAK2. However, we also cannot rule out the possibility, that TSLPR signaling (which also depends on the IL-7Rα chain 74 ) differs in its functions in B cell differentiation between humans and mice. Altogether, we hypothesize, that Ph-like B-ALL comprises several leukemia forms arising from the narrow pro/preB cell window, which share homologies but remain heterogeneous in their pathophysiological (e.g. IRF4 deregulation) and genetic landscape. This hypothesis is supported by the distinct subtypes of Ph-like ALL that C. Mulligan and colleagues proposed based on mutational signatures (sFig.3a and 28 ).
As proof of its usefulness as a disease model, the Irf4 -/-pB-ALL raised several therapeutic options. Intriguingly, forced IRF4 re-expression caused differentiation and cell death, encouraging treatments to boost its expression. Moreover, our transcriptomic survey highlights a fourth oncogenic (and targetable) role of IRF4-deficiency to those described above: 4) Proximity signaling: IRF4 loss upregulated FAK expression, securing cell survival likely via integrin mediated cell contacts. Targeting FAK is currently under investigation for treating solid tumors, 59,60 but our preliminary data give logic to trials also in IRF4 deregulated B-ALL.
Nevertheless, our in vivo experiments highlight Ruxolitinib as the most suitable treatment for JAK driven leukemia. This compound represents an important therapeutic agent in myeloproliferative disease and is already considered for treatment of Ph-like-ALL. 61,62 We found a surprising and novel preferential effect of Ruxolitinib on CNS-and organ infiltration, with less pronounced activity on leukemic cells in the BM. SCs may provide a sanctuary rendering leukemic cells less sensitive to targeting by Ruxolitinib. 5 However, others showed clear anti-leukemic Ruxolitinib effects in a murine Ph + ALL model. 75 Perhaps the more sensitive effects on organ infiltration are demasked only in generally less sensitive tumour types.
Especially the effects on leptomeningeal involvement are of key translational importance because current CNS-targeted therapies for ALL remain toxic, significantly lowering quality of life for patients. We provide preliminary evidence that this Ruxolitinib effect might involve downregulation of surface integrins. Future studies should test this hypothesis mechanistically and evaluate efficacy of Ruxolitinib for leukemic organ infiltration also in other leukemia models.

Murine pro/preB cell cultures
Femur and tibia bones from 8-12 weeks old mice were explanted and cleaned from adherent tissues. Cells were extracted via centrifugation at 11 x 10 3 RPM for 10 s.

Histological analyses
Tissue samples were immediately fixed in 4 % formaldehyde in PBS solution.
Histological analysis was performed on 3 µm thick sections from paraffin embedded tissue as described previously. 82  All measurements were performed using Fijii. 83

Whole Exome Sequencing and biostatistical analysis
To determine single nucleotide variants (SNV) within leukemia samples, genomic coli. The indicated numbers of clones (Fig. 1G) for each PCR amplicon were sequenced and aligned with software from IMGT/V-quest. 86

Retroviral transduction of Jak3-mutants and IL-7 independency assay
The coding sequence of murine Jak3 was amplified from pCineo-Jak3 (a gift from Olli Silvennoinen from Tampere-university in Finland) and cloned into the pMSCV-Thy1.1 expression plasmid using BamHI and SalI resctriction digestion. Site directed mutagenesis was performed following the manufacturer's protocol using the Quick-Change II site-directed mutagenesis kit (Agilent Technologies; primers employed are listed in the supplemental materials). Viral supernatant from mutated pMSCV-Thy1.1-Jak3 constructs was produced as described previously. 78 For viral transduction, 5 x

RNA sequencing and biostatistical analysis
RNA extraction from primary tumour samples and FACS sorted B220 + mIgMpro/preB cells was performed using Trizol extraction. Quality control was performed using the Bioanalyzer RNA 6000 NanoChip (Agilent Technologies Differential expression analysis was performed with DESeq2 package, 89 setting the false discovery rate (FDR) cutoff to 0.1. Data comparing IRF4-RV and GFP-RV samples was visualized as a volcano plot using R package ggplot2: -log10 of Benjamini-Hochberg adjusted P-values were plotted against log2 fold change of transcript-reads comparing T11 cells transduced according to the protocol described above for Jak3mutants with either IRF4-EV or GFP-EV (transduction efficiency routinely above 85 %).
Biological relevance was defined as a log2 fold change of > 1 or < -1, while false positives were controlled with a false discovery rate (FDR) set to 0.1. Identification of genes involved in leukemogenesis and regulated by IRF4 was performed as a multistep process: first, genes were included into the analysis, that showed > 2-fold expression change between Irf4 -/-B220 + mIgMpro/preB cells and leukemia samples.
Second, genes were filtered to only include genes that are also differentially regulated in IRF4-RV or EV-RV transduced T11 cells. Third, resultant genes were limited to genes previously shown to be bound by IRF4 using ChIP-seq. 50

Quantitative real time (qRT-)PCR
Total RNA was extracted both from primary Irf4 -/tumors as well as FACS-sorted control B220 + mIgM -BM pro/preB cells of either Irf4 -/or wt animals using the Gdansk extractme kit (EM09.1) according to the manufacturers protocol. cDNA was prepared from whole RNA samples using the RevertAid cDNA kit from Thermo Fisher (K1621).

Western blotting
To quantify FAK protein, sample cells were lysed in 100 µL RIPA buffer (Sigma-Aldrich,

Data availability
The sequencing datasets generated during the current study have been deposited in the Gene Expression Omnibus (GEO) archive and are available under the accession number GSE157958.

Statistical analysis
Statistical analysis was performed using the GraphPad 9.0 software. Data are commonly presented as mean ± SD. Prior to significance testing, normal distribution and homogeneity of variances was confirmed by Shapiro-Wilk test and Brown-Forsythe testing. Statistical significance when comparing two normally distributed groups was evaluated using two-tailed unpaired t-tests. In case of significant differences in variances between groups, Welch's correction was applied to account for non-norminal distribution of data. When comparing multiple groups, one-way or two-way analysis of variance (ANOVA) was performed, depending on the number of variables that differed between compared groups. This was followed by a Tukey's HSD, or Dunnett's post hoc test, as indicated in figure legends. An alpha level of P < 0.05 was employed for significance testing. Thresholds of P < 0.05, 0.01, 0.001 and 0.0001 are marked by one to four asterisks respectively (also detailed in each figure legend).

Study approval
All animal experiments were approved by the local government (Regierungspräsidium Gießen, G49/2018) and conducted according to the German animal protection law.      Euclidean clustering depicted as dendrograms to the left. Two samples per group.

Primers
Epb4.1l4b In the Ruxolitinib group, 4 mice were excluded and censored due to intervention related adverse reactions or due to their use for to the analysis described in e-g).