Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Signalling input from divergent pathways subverts B cell transformation

Abstract

Malignant transformation of cells typically involves several genetic lesions, whose combined activity gives rise to cancer1. Here we analyse 1,148 patient-derived B-cell leukaemia (B-ALL) samples, and find that individual mutations do not promote leukaemogenesis unless they converge on one single oncogenic pathway that is characteristic of the differentiation stage of transformed B cells. Mutations that are not aligned with this central oncogenic driver activate divergent pathways and subvert transformation. Oncogenic lesions in B-ALL frequently mimic signalling through cytokine receptors at the pro-B-cell stage (via activation of the signal-transduction protein STAT5)2,3,4 or pre-B-cell receptors in more mature cells (via activation of the protein kinase ERK)5,6,7,8. STAT5- and ERK-activating lesions are found frequently, but occur together in only around 3% of cases (P = 2.2 × 10−16). Single-cell mutation and phospho-protein analyses reveal the segregation of oncogenic STAT5 and ERK activation to competing clones. STAT5 and ERK engage opposing biochemical and transcriptional programs that are orchestrated by the transcription factors MYC and BCL6, respectively. Genetic reactivation of the divergent (suppressed) pathway comes at the expense of the principal oncogenic driver and reverses transformation. Conversely, deletion of divergent pathway components accelerates leukaemogenesis. Thus, persistence of divergent signalling pathways represents a powerful barrier to transformation, while convergence on one principal driver defines a central event in leukaemia initiation. Pharmacological reactivation of suppressed divergent circuits synergizes strongly with inhibition of the principal oncogenic driver. Hence, reactivation of divergent pathways can be leveraged as a previously unrecognized strategy to enhance treatment responses.

Main

During early B-cell development, cytokine receptors (such as the interleukin-7 receptor (IL7R) and cytokine-receptor-like factor 2 (CRLF2)) initiate survival signals through phosphorylation of the kinase JAK2 and thereby the transcription factor STAT5 (refs. 2,3). Later, following productive rearrangement of immunoglobulin V region genes and expression of a pre-B-cell receptor (pre-BCR), survival and proliferation signals instead involve the pre-B-cell linker protein BLNK and the phosphorylation of ERK kinases9. STAT5- and ERK-mediated survival signals are frequently mimicked by transforming oncogenes in B-cell acute lymphoblastic leukaemia (B-ALL). Thus, lesions in genes encoding cytokine receptors (IL7R, CRLF2, EPOR, PDGFRA and PDGFRB), cytokine-receptor-associated JAKs (JAK1, JAK2 and JAK3) and ABL1 tyrosine kinases (BCR–ABL1) mimic the effects of constitutively active cytokine receptors by phosphorylating STAT5. Meanwhile, activating lesions of genes in the RAS signalling pathway (NRAS, KRAS, PTPN11, NF1 and BRAF)6 instead cause oncogenic ERK signalling, a functional mimic of pre-BCR signalling5,8. Malignant transformation typically involves the cooperation of numerous genetic lesions1, suggesting that adding oncogenic drivers to existing mutations would accelerate tumour progression. However, loss of the tumour suppressor PTEN (resulting in hyperactivated phosphatidylinositol-3-kinase (PI3K) signalling) is synthetic lethal in BCR–ABL1 and NRASG12D-driven B-ALL10. Likewise, functionally normal epithelial cells carrying high mutation burdens give rise to overt squamous cell carcinomas11 only after these many mutations are reduced to a small set of oncogenic drivers. Here we studied the interaction of oncogenic drivers in STAT5 and ERK pathways during normal B-cell development and malignant B-cell transformation.

STAT5 and ERK activation are segregated

Studying genetic lesions in 1,148 B-ALL cases (Supplementary Tables 1, 35), we found STAT5-activating lesions in 360 cases (31.4%) and ERK-activating lesions in 386 cases (33.6%; Supplementary Table 1). Concurrent activation of STAT5 and ERK occurs less frequently in B-ALL than expected by chance (3%; odds ratio 0.13; P = 2.2 × 10−16; Extended Data Fig. 1a). Mutual exclusivity of mutations could reflect functional redundancy rather than antagonism. Unsupervised analysis of mutational co-occurrence between all lesion pairs revealed overall significantly stronger exclusivity between interpathway lesions compared with intrapathway lesions (Extended Data Fig. 1b, c). Concurrent mutation of multiple drivers is counter-selected regardless of pathway, but the observed STAT5–ERK pathway co-occurrence is significantly less frequent (P = 0.014) and supported by significantly lower log odds ratios for interpathway versus intrapathway driver combinations (P = 0.001; Extended Data Fig. 1d, e). STAT5- and ERK-pathway lesions are common in acute myeloid leukaemia (AML). However, mutual exclusivity of interpathway pairs was not significantly stronger than that observed between intrapathway drivers in AML (P = 0.7; Extended Data Figs. 1f–j and Supplementary Table 2). Hence, while STAT5- and ERK-activating mutations are frequent in both B-ALL and AML, driver mutations in the two pathways are strongly segregated in B-ALL but not in AML.

Studying biochemical pathway activation, we confirmed an inverse relationship between the levels of STAT5 phosphorylated at tyrosine 694 (STAT5-pY694) and ERK phosphorylated at threonine 202 or tyrosine 204 (ERK-pT202/Y204) in B-ALL patient-derived xenografts (PDXs) by western blotting (Supplementary Table 5 and Fig. 1a). Intermittent treatment of Ph+ B-ALL cells with the tyrosine kinase inhibitor ponatinib to suppress STAT5 signalling resulted in gradual development of ponatinib resistance over three weeks in four of eight cases. Two of the ponatinib-resistant cases (TXL3 and BLQ5) lost STAT5 activity and acquired de novo ERK phosphorylation (Fig. 1b). Consistent with ponatinib resistance and a switch from STAT5 to ERK phosphorylation, TXL3 and BLQ5 cells acquired sensitivity to trametinib (a small-molecule inhibitor of mitogen-activated protein kinase kinase 1/2 (MAP2K1/2); Fig. 1c).

Fig. 1: Segregation of STAT5 and ERK activation in human B-ALL.
figure1

a, Correlation between ERK-pT202/Y204 and STAT5-pY694 levels in B-ALL cells (n = 23 independent biological samples; P = 0.001, two-tailed t-test; r = −0.657, Pearson’s r). b, Western blots of patient-derived B-ALL cells (n = 8) and Ph+ B-ALL cells before (n = 8) and after (n = 8) ponatinib treatment. c, Analysis of patient-derived Ph+ B-ALL cells treated with trametinib or ponatinib, showing the percentage of growth inhibition (means of three independent experiments). Trametinib (left to right): TXL3, BLQ5, SFO2, LAX2, PDX2, ICN1, BLQ1 and BLQ11. Ponatinib (left to right): TXL3, BLQ5, SFO2, ICN1, PDX2, BLQ1, BLQ11 and LAX2. d, Single-cell phosphoprotein analysis of patient-derived B-ALL samples (n = 3 independent experiments). Patient sample names are indicated in black at the top left of each panel. Green and red fonts indicate proteins that are mutated in each sample, causing activation of the STAT5 or ERK pathways, respectively. For gel source data, see Supplementary Fig. 1.

Source data

To test whether STAT5- and ERK-activating mutations in rare cases with dual pathway activation co-occurred in the same cells, we performed single-cell amplicon sequencing (Supplementary Table 4). NSG mice (n = 30) bearing nine Ph+ B-ALL (STAT5-driven) PDXs were treated with ponatinib. After initial remission, 24 mice relapsed with additional STAT5- or ERK-activating lesions that reversed ponatinib sensitivity (Supplementary Table 3). STAT5- and ERK-activating lesions were found concurrently in three of the resistant cases (Supplementary Tables 3, 4). We selected these three PDXs (1F10, 2B10 and 2G10) for single-cell amplicon sequencing. STAT5 (ABL1) and ERK (PTPN11, KRAS)-activating mutations were amplified individually from single sorted B-ALL cells. However, mutant ABL1 alleles were only co-amplified with wild-type KRAS and PTPN11 and vice versa (Supplementary Table 4), suggesting that these mutations segregated to distinct clones. We then performed single-cell phosphoprotein analyses for STAT5-pY694 and ERK-pT202/Y204. Consistent with segregation of oncogenic STAT5 and ERK signalling, STAT5 and ERK phosphorylation was mutually exclusive in most cases. In four cases, we found biclonal disease with STAT5 and ERK phosphorylation confined to competing clones (Fig. 1d and Extended Data Fig. 2). The small number (roughly 2%) of concurrent phosphorylation events was in line with the low frequency of two cells being loaded and analysed in the same well.

STAT5 and ERK define B-cell development

Activation of STAT5 (downstream of cytokine receptors in pro-B cells) and of ERK (downstream of the pre-BCR in pre-B cells) are linked to distinct stages of early B-cell development that are separated by the rearrangement of immunoglobulin heavy chain (Ig-HC) and light chain (Ig-LC) genes3,12,13,14. Although aberrant activity of the RAG1 and RAG2 recombinases in B-ALL cells15 can randomly target Ig-LC loci, we observed a significant association between STAT5-driven B-ALL and germline Ig-LC configuration on the one hand, and ERK-driven B-ALL and rearranged Ig-LC loci on the other (P = 0.008; Supplementary Table 6). These findings suggest a link between STAT5- and ERK-activating lesions in B-ALL cells and cellular origins from distinct B-cell developmental stages. Analysing B-ALL PDXs and B-ALL mouse models by flow cytometry, we corroborated the association between oncogenic STAT5 signalling and pro-B cells, while ERK-driver lesions were associated with a pre-B-cell phenotype (Extended Data Figs. 3, 4).

Developmental rewiring from STAT5 to ERK

Analysis of overall and relapse-free survival of patients with STAT5-driven (n = 26) or ERK-driven (n = 67) B-ALL (from Children’s Oncology Group (COG) trial P9906) showed that high expression levels of the immunoglobulin genes IGLV and IGHM predicted favourable outcomes in patients with STAT5-driven B-ALL, but poor clinical outcomes in patients with ERK-driven B-ALL (Extended Data Fig. 4b). Developmental rewiring at the pre-BCR checkpoint affects permissiveness to oncogenic STAT5 or ERK signalling. While pro-B cells were permissive to BCR–ABL1 (STAT5), activation of ERK by the pre-BCR mimic, the Epstein–Barr virus (EBV)-encoded oncoprotein LMP2A7, induced cellular senescence and cell death. Conversely, pre-B cells were permissive to transformation by LMP2A (ERK), while BCR–ABL1 induced cell death and senescence (Extended Data Fig. 5b–e). BCR–ABL1 is a frequent oncogenic driver in B-ALL, but is never found in B-cell malignancies past the pre-BCR checkpoint. Likewise, LMP2A mimics pre-BCR signalling7 and functions as an oncogenic driver in mature B-cell malignancies, while EBV+ B-ALL is exceedingly rare.

MYC and BCL6 downstream of STAT5 and ERK

As STAT5 and ERK lesions were associated with pro-B-cell and pre-B-cell phenotypes, respectively, we studied activation of the two pathways at the pro-B to pre-B transition. Single-cell phosphoprotein analysis in sorted mouse bone-marrow pro-B-cell and pre-B-cell/immature B-cell populations showed that STAT5 activation at pro-B-cell stages was terminated at the pre-BCR checkpoint and replaced by ERK activation in more mature B-cell subsets (Fig. 2a). While STAT5 signalling transcriptionally activates MYC and suppresses BCL6 (ref. 13), ERK activity at the pre-BCR checkpoint induces MYC downregulation with concurrent upregulation of BCL6 (ref. 14). We studied B-cell differentiation in a B-cell-specific Myc–eGFP × Bcl6–mCherry double reporter knock-in mouse model, finding a similar transition from MYC+ BCL6 pro-B to MYC BCL6+ pre-B stages (here, eGFP is enhanced green fluorescent protein). Cytokine-dependent pro-B and early pre-B cells (Hardy fractions B–C′) included double-negative and MYC-expressing cells, while BCR-dependent B cells (Hardy fractions E–F) almost exclusively expressed BCL6 (Fig. 2b). Late pre-B cells (fraction D) that have already downregulated cytokine receptors but still undergo Ig-LC gene rearrangement mark the intersection between MYC+ BCL6 and MYC Bcl6+ stages and express neither MYC nor BCL6 (Fig. 2b). Functional Ig-HC expression in pro-B cells induced the pro-B to pre-B transition, leading to Ig-LC gene rearrangement and surface expression (Fig. 2c). Activation of doxycycline-inducible NRASG12D recapitulated the pro-B to pre-B transition to induce Ig-LC surface expression (Fig. 2c). Induction of the pro-B-cell to pre-B-cell transition (Ig-HC expression or activation of oncogenic ERK signalling) resulted in a switch from the MYC+ BCL6 pro-B-cell to MYC BCL6+ pre-B-cell stage (Fig. 2d). Thus, activation of BCL6 at the expense of MYC is a determinant of the pro-B to pre-B transition.

Fig. 2: STAT5–MYC and ERK–BCL6 signalling are incompatible and define distinct stages of B-cell development.
figure2

a, Analysis of Hardy fractions B–F for MyceGFP/+ Bcl6mCherry/+ bone-marrow cells, including single-cell phosphoprotein analyses of the indicated fractions. b, Expression of eGFP and mCherry in MyceGFP/+ Bcl6mCherry/+ bone marrow cells in Hardy fractions B–F. c, Surface expression of Igκλ light chain (LC) on IL7-dependent pro-B cells expressing empty vector (EV), and upon doxycycline-inducible expression of Ig-HC or NRASG12D. Numbers in the top right corners denote the percentage of cells expressing Ig-LC. d, Expression of eGFP and mCherry in IL7-dependent MyceGFP/+ Bcl6mCherry/+ B-cell precursors induced to differentiate or transduced with NRASG12D. The numbers in the top right corners denote the percentage of cells in each population: MYC BCL6, MYC+ BCL6, MYC BCL6+ and MYC+ BCL6+. e, Western blots of Ph-like B-ALL cells treated with ruxolitinib (n = 3 independent experiments). f, Expression of eGFP and mCherry in MyceGFP/+ Bcl6mCherry/+ B-cell precursors transduced with EV, BCR–ABL1 or NRASG12D, and subsequently treated with vehicle control, imatinib (1 μM) or trametinib (10 nM). g, h, Enrichment or depletion of GFP+ BCR–ABL1 or NRASG12D B-ALL cells transduced with EV, GFP–MYC or GFP–BCL6 (means ± s.d.). ad, f, Data from three independent biological replicates; eh, n = 3 independent experiments. For gel source data, see Supplementary Fig. 1.

Source data

In a STAT5-driven B-ALL PDX, pharmacological inhibition of STAT5 induces BCL6 at the expense of MYC. Consistent with opposing roles of STAT5 and STAT3 in the regulation of MYC and BCL6 (ref. 16), STAT5 dephosphorylation was paralleled by increased STAT3 phosphorylation (Fig. 2e). While oncogenic STAT5 signalling (BCR–ABL1) activated MYC and suppressed BCL6, ERK signalling (NRASG12D) activated BCL6–mCherry but reduced MYC–eGFP activity (Fig. 2f). Trametinib largely reduced BCL6-reporter activity. Flow-cytometry analysis of MYC–eGFP BCL6–mCherry dual reporter cells showed an L-shaped pattern (Fig. 2f) comparable to STAT5–ERK single-cell phosphoprotein analyses (Fig. 1d). STAT5–MYC+ and ERK–BCL6+ cells represent non-overlapping populations, reflecting distinct stages of early B-cell differentiation. MYC increased competitive fitness in STAT-driven B-ALL (BCR–ABL1), whereas BCL6 caused rapid depletion in growth competition assays (Fig. 2g). Conversely, BCL6 increased competitive fitness of ERK-driven B-ALL cells (NRASG12D). However, ERK-driven B-ALL cells were not permissive to MYC and became depleted from cell culture (Fig. 2h). Thus, while cytokine-receptor signalling in pro-B cells or its oncogenic mimics (for example, BCR–ABL1) activate STAT5 to promote a MYC-driven transcriptional program, pre-BCR signalling in pre-B cells or its oncogenic mimics (for example, NRASG12D) activate ERK to engage BCL6-dependent transcription (Extended Data Fig. 5f).

Reciprocal suppression of STAT5 and ERK

To determine the mechanistic basis for the segregation of STAT5 and ERK activation, we examined whether inducible activation of one pathway affects biochemical activity of the other. Inducible expression of a constitutively active mutant of STAT5A in mouse IL7-dependent pro-B cells strongly increased levels of STAT5-pY694 and reduced levels of ERK-pT202/Y204 (Fig. 3a). Like ERK, JNK and p38α activity was suppressed by STAT5. However, only ERK, but not JNK and p38α, was phosphorylated upon inducible expression of NRASG12D in IL7-dependent pro-B cells. STAT1 phosphorylation was barely detectable upon RAS activation, but STAT3 was strongly phosphorylated on both Y705 and S727; this was suppressed by the constitutively active STAT5ACA (Fig. 3a and Extended Data Fig. 5a). Expression of NRASG12D increased phosphorylation of STAT3-pS727 but resulted in dephosphorylation of STAT5 (Fig. 3a). To confirm biochemical interference between STAT5 and ERK signalling, we engineered murine pro-B cells to express, first, the human IL2 receptor β (hIL2Rβ) chain, which enables phosphorylation of STAT5A-Y694 and STAT5B-Y699 in response to the binding of human IL2, and second, doxycycline-inducible NRASG12D (Fig. 3b). Activation of STAT5 (hIL2), ERK (doxycycline) or a combination of both confirmed biochemical interference between the STAT5 and ERK pathways (Fig. 3b). Colony formation and proliferation was increased by activation of either STAT5 or ERK alone, but compromised by dual pathway activation (Fig. 3c, d). These observations indicate that signal input from divergent pathways represents a barrier to malignant transformation.

Fig. 3: Concurrent oncogenic STAT5 and ERK activation subverts B-cell leukaemogenesis.
figure3

a, Western blotting following doxycycline-induced expression of Stat5aCA or NRASG12D in IL7-dependent mouse pro-B cells. b, Fluorescence-activated cell sorting (FACS) plots and western blots of hIL2Rβ–TetO–NRASG12D mouse pro-B cells following the indicated treatments (doxycycline, 24 h; hIL2, 15 min, 50 ng μl−1). c, d, Colony formation (c; P = 1.9 × 10−5, two-tailed t-test) and viable cell counts (d; mean ± s.d.) of hIL2Rβ–TetO–NRASG12D mouse pro-B cells treated with doxycycline, hIL2 (10 ng μl−1), or a combination of both. ad, Data from three independent experiments. For gel source data, see Supplementary Fig. 1.

Source data

Given the strong negative association between BCR–ABL1 and oncogenic NRAS lesions in B-ALL (Extended Data Fig. 1a), we modelled concurrent activation of both oncogenic drivers (Extended Data Fig. 6a). As the sole oncogenic driver following transduction with neutral empty vector (EV) EV–Orange or EV–GFP, both STAT5 (EV–Orange plus BCL–ABL1–GFP) and ERK (EV–GFP plus NRAS–Orange) gave rise to large double-positive populations. When, instead of empty vector, a secondary oncogene engaging a diverging pathway was transduced, frequencies of double-positive cells dropped by 22-fold for STAT5 and 15-fold for ERK (Extended Data Fig. 6a). This reduction in the number of double-positive cells was mirrored by reduced colony formation when comparing single with dual pathway activation (Extended Data Fig. 6b). Pathway interference was confirmed in patient-derived BCRABL1 (MXP2) and KRASG12V (LAX7R) B-ALL cells engineered to inducibly express NRASG12D or BCR–ABL1, respectively. Concurrent activation of a divergent pathway suppressed proliferation, cellular senescence and cell death (Extended Data Fig. 6c–e).

Inactivation of divergent pathways

We next tested whether background ‘noise’ from pathways diverging from the principal oncogenic driver represents a barrier to transformation. Consistent with biochemical cross-inhibition, inducible deletion of Mapk1 (Erk2) resulted in loss of STAT3 but increased STAT5 phosphorylation in a mouse model of B-ALL cells. Conversely, inducible Stat5 deletion increased phosphorylation of both ERK and STAT3 in B-ALL (Fig. 4a, b). In contrast with deletion of the principal oncogenic driver, acute deletion of secondary divergent pathways (Mapk1 in STAT5-driven and Stat5 in ERK-driven B-ALL) only transiently slowed cell growth (Fig. 4c, d). Compared with B-ALL cells retaining intact divergent pathway components, B-ALL cells with Cre-mediated deletions strikingly increased colony formation (Fig. 4e, f).

Fig. 4: Genetic deletion of alternative pathways triggers initiation of STAT5- and ERK-driven leukaemia.
figure4

a, Western blots of BCR–ABL1 B-ALL cells upon Cre-mediated ablation of Mapk1 (Erk2). b, Western blots of NRASG12D B-ALL cells upon Cre-mediated deletion of Stat5. c, d, Enrichment or depletion of GFP+ Stat5fl/fl or Mapk1fl/fl BCR–ABL1 (c) or NRASG12D (d) B-ALL cells transduced with ERT2–GFP or Cre–ERT2–GFP. ERT2 is a mutant ligand-binding domain of the human oestrogen receptor that binds to 4-hydroxytamoxifen (4-OHT) or tamoxifen. e, f, Colony formation by Mapk1fl/fl BCR–ABL1 B-ALL cells (e) and Stat5fl/fl NRASG12D B-ALL cells (f) upon deletion of Mapk1 or Stat5, respectively. The values above each image show colony formation as a percentage of control values. g, h, Colony formation (percentage of control) of Mb1–Cre × LSL–BCR–ABL1 (g) and Mb1–Cre × LSL–KRASG12D (h) pro-B cells primed with trametinib (g; 1 nmol l−1, 10 days) and ruxolitinib (h; 10 nmol l−1, 10 days). i, j, Kaplan–Meier analyses (Mantel–Cox log-rank test) of recipient mice (n = 4 per group) bearing BCR–ABL1 (i) or NRASG12D-driven (j) B-ALL cells with or without deletion of Mapk1 (i; 24 h) or Stat5 (j; 24 h). ah, Data shown as means ± s.d. (n = 3 independent experiments. ch, Assessed by two-tailed t-test. For gel source data, see Supplementary Fig. 1.

Source data

To examine the effects of deleting divergent pathway components on leukaemia initiation in vivo, we performed limiting dilution transplant experiments based on 100, 1,000 and 10,000 pro-B cells that carried STAT5- or ERK-driver oncogenes together with Mapk1fl/fl (ERK) and Stat5fl/fl alleles for inducible ablation of divergent pathways. Mapk1fl/fl pro-B cells with BCR–ABL1 (STAT5) and Stat5fl/fl pro-B cells with NRASG12D (ERK) carried either empty vector, to achieve retention of divergent pathway components, or tamoxifen-inducible Cre, to achieve their deletion. Five days after engraftment, NSG recipient mice were injected with tamoxifen to activate Cre or empty-vector controls. Mapk1 deletion accelerated the onset of overt STAT5-driven B-ALL. Likewise, deletion of Stat5 accelerated the initiation of leukaemia in ERK-driven B-ALL. At two lower dose levels (1,000 and 100 cells), Stat5 deletion was required for initiation of ERK-driven B-ALL. Injection of 100 or 1,000 ERK-driven B-ALL cells failed to initiate fatal disease in transplant recipients unless Stat5 was deleted (Fig. 4i, j).

As surrogates for the deletion of the divergent pathway components Mapk1 and Stat5, we used small-molecule inhibitors of MAP2K1/2 (trametinib) and JAK–STAT5 (ruxolitinib). Strikingly, trametinib triggered leukaemic colony formation from mouse pro-B cells carrying an inducible BCRABL1 knock-in allele in vitro. Likewise, initiation of leukaemic colony formation from mouse pro-B cells carrying an inducible KRASG12D knock-in allele (Supplementary Table 7) was precipitated by treatment with ruxolinitib (Fig. 4g, h). We further tested pharmacological combinations of principal drivers with divergent pathway inhibition. BCI-215 activates ERK phosphoryation17 as an allosteric inhibitor of dual-specificity phosphatase 6 (DUSP6), the central negative-feedback regulator of ERK in B-ALL18. The small molecule DPH activates STAT5 phosphorylation by relieving ABL1 and related kinases from their autoinhibitory conformation19. Interestingly, trametinib in combination with DPH increased colony formation in BCR–ABL1-driven B-ALL. Likewise, BCI-215 potentiated the effects of ruxolitinib on colony formation of NRASG12D B-ALL (Extended Data Fig. 6f, g). Taken together, these results show that ablation of divergent inputs (‘noise’) and convergence on one principal oncogenic driver represent a critical event in leukaemogenesis. A recent transposon screen suggested that truncation of the ERK-signalling activator Sos1 precipitates the onset of STAT5-driven B-ALL20. Notably, prolonged suppression of JAK2–STAT5 signalling by ruxolitinib increased the risk of B-cell transformation and could prime dormant B-cell clones to develop overt ERK-driven B-cell lymphoma21.

PTPN6 contributes to ERK signalling

Previous work showed that the inhibitory phosphatase PTPN6 (SHP1) negatively regulates STAT5 (ref. 22). We hypothesized that PTPN6-mediated negative regulation of STAT5 contributes to oncogenic ERK signalling. Activation of ERK induced PTPN6 expression and activation (Extended Data Fig. 7a). The ERK-downstream transcription factors ELK1, JUN, JUNB and CREB1 (ref. 9) bound to the PTPN6 promoter (Extended Data Fig. 7b). Consistent with the suppression of divergent pathways, ERK was inactive in patient-derived STAT5-driven B-ALL cells (PDX2). However, BCI-215 not only reactivated ERK, but also induced PTPN6 phosphorylation and dephosphorylation of its substrate STAT5 (Extended Data Fig. 7c, d). Supporting the critical function of PTPN6 in ERK signalling, Ptpn6 deletion increased STAT5 phosphorylation (Extended Data Fig. 7e) and compromised the colony formation and competitive fitness of the mouse model of NRASG12D B-ALL cells (Extended Data Fig. 7f, g).

Blnk enforces convergence on the Erk pathway

The pre-B-cell linker BLNK interacts directly with PTPN6 and mediates its phosphatase activity23, and links Ras proteins to ERK signalling24. Studying the inducible activation of pre-BCR or oncogenic ERK signalling (Extended Data Fig. 8a, b) in Blnk+/+ and Blnk−/− mouse B-cell precursors revealed that BLNK was essential for the activation of ERK downstream of the pre-BCR or oncogenic NRASG12D. In the absence of Blnk, inducible pre-BCR or NRASG12D signalling failed to activate ERK and to suppress STAT5 phosphorylation. Likewise, BLNK was required for upregulation of PTPN6 in response to oncogenic ERK signalling (Extended Data Fig. 8b). Hence, we examined whether BLNK differentially affects oncogenic STAT5 and ERK signalling. Inducible expression of STAT5ACA in Blnk+/+ B-cell precursors induced rapid cell death. By contrast, induction of STAT5ACA in Blnk−/− B-cell precursors induced malignant transformation, in agreement with previous findings that STAT5BCA cooperates with defects in Blnk to promote B-ALL4. As expected, NRASG12D induction caused malignant transformation of Blnk+/+ B-cell precursors. However, NRASG12D induction only transiently accelerated proliferation of Blnk−/− B-cell precursors, with cell death following (Extended Data Fig. 8c–f). Similarly, ablation of Blnk increased colony formation after STAT5ACA induction, while colony numbers were decreased in Blnk−/− B-cell precursors when ERK was activated downstream of NRASG12D (Extended Data Fig. 8g, h).

To validate the role of BLNK in patient-derived B-ALL cells, we studied a matched B-ALL pair from the same patient at the time of diagnosis (sample LAX7; STAT5-driven, IL7RSI246S) and after relapse (sample LAX7R; ERK-driven, KRASG12V). Genetic deletion of BLNK provided a strong competitive advantage to LAX7 cells but induced cell death in LAX7R cells (Extended Data Fig. 9a–d). Western blotting revealed that pharmacological inhibition of the principal driver in LAX7 (STAT5) and LAX7R (ERK) cells resulted in reactivation of divergent suppressed pathways (Extended Data Fig. 9e). The divergent changes between LAX7 and LAX7R samples were mirrored by contrasting drug responses: LAX7 and LAX7R samples were selectively sensitive to ruxolitinib and trametinib, respectively (Extended Data Fig. 9f). Taken together, PTPN6 and BLNK function as ERK-induced negative regulators of STAT5, hence enabling pathway convergence on oncogenic ERK signalling (Extended Data Fig. 5f).

Reactivation of suppressed pathways

The current paradigm of targeted therapy in cancer is based mainly on directly suppressing the principal oncogenic driver. As a strategy for preventing drug resistance and relapse, we explored an alternative approach: one based on reactivating suppressed pathways that operate in divergent directions relative to the principal oncogenic driver. We focused on pharmacological reactivation of suppressed ERK and STAT5 signalling by BCI-215 and DPH. BCI-215 reactivated the repressed ERK pathway at the expense of STAT5 in STAT5-driven B-ALL (samples LAX7 and JFK125R). Conversely, DPH reactivated suppressed STAT5 signalling and decreased ERK phosphorylation in ERK-driven B-ALL (LAX7R, KRASG12V; Extended Data Fig. 10). We next tested whether divergent pathway reactivation amplifies the impact of direct inhibition of the principal oncogenic driver. Treatment with BCI-215 or ruxolitinib alone achieved only minor responses, whereas combinations achieved synergistic effects in eliminating B-ALL cells (Extended Data Fig. 10). DPH or trametinib alone partially suppressed ERK phosphorylation in ERK-driven (KRASG12V) B-ALL cells and had minor effects on cell viability. Mirroring complete ERK suppression, DPH and trametinib combinations had synergistic effects on B-ALL elimination (combination index 0.28; Extended Data Fig. 10c).

To validate the efficacy and feasibility of divergent pathway reactivation in vivo, we injected NSG mice with B-ALL cells from two refractory PDXs (LAX7R and JFK125R). Both cases originate from STAT5-driven primary B-ALL that developed drug resistance and relapsed with an additional ERK-driven clone (KRAS; Extended Data Fig. 10d, j). The combination of DPH and trametinib achieved synergy (P = 0.0005 versus trametinib alone; P = 0.0004 versus DPH alone) and extended the overall survival of mice bearing LAX7R (ERK-driven, KRASG12V; Extended Data Fig. 10e). As the STAT5-driven clone in JFK125R at relapse was still dominant (Extended Data Fig. 10j), mice bearing JFK125R were treated with BCI-215 and ruxolitinib combinations, which achieved synergy (P = 0.0103 versus BCI-215; P = 0.0005 versus ruxolitinib) and prolonged the overall survival of transplant recipients (Extended Data Fig. 10k).

We compared STAT5 and ERK phosphorylation through single-cell analysis of LAX7R and JFK125R before (Extended Data Fig. 10d, j) and after (Extended Data Fig. 10g, m) treatment with combinations of DPH plus trametinib and BCI-215 plus ruxolitinib in vivo. Combination treatment eradicated the dominant clone, whereas the formerly minor clone gave rise to fatal disease driven by one single population (Extended Data Fig. 10g, m). Compared with pretreatment, fatal leukaemia post-treatment was resistant to the drug combination directed at the major clone, but highly sensitive to the drug combination directed at the divergent pathway (Extended Data Fig. 10f, l). Taken together, these observations suggest that alternating treatment regimens with first a drug combination targeting the major clone, and second a drug combination targeting the divergent minor clone may be useful in eradicating high-risk biclonal disease.

Discussion

Non-transformed B cells constantly exchange information with their environment, depending on external proliferation and survival cues that engage several divergent pathways (such as cytokine-receptor and BCR pathways). Hence a diverse spectrum of signalling inputs reflects interactions of normal cells with their environment, while convergence on one central pathway represents a hallmark of transformation. We here provide evidence that the inactivation of divergent pathways that are not aligned with the principal oncogenic driver represents a critical step during transformation. The reactivation of divergent and potentially conflicting signalling pathways represents a powerful barrier to transformation. From a treatment perspective, the targeted reactivation of divergent pathways to interfere with the principal oncogenic driver may represent a promising strategy to amplify treatment responses. Our results show that this approach is orthogonal to the direct inhibition of driver oncogenes (for example, by tyrosine kinase inhibitors) and may represent a previously unrecognized strategy for overcoming drug resistance. The activation of other pathways (for example, WNT, NF-κB or Notch pathways) may likewise either coalesce into one convergent pathway, or interfere with and disrupt signalling from the principal driver oncogene. To comprehensively identify targets for the pharmacological reactivation of divergent (suppressed) pathways, it will be important to elucidate cell-type-specific mechanisms of convergence and interference between principal oncogenic drivers and their potential detractors.

Methods

Patient-derived B-ALL samples

Patient samples (Supplementary Tables 35) were sourced ethically from patients who gave informed consent, and were in compliance with the internal review boards of the Beckman Research Institute of City of Hope and the Dana Farber Cancer Institute and Harvard Medical School. We have complied with all relevant ethical regulations. Patient samples were harvested from biopsy of bone marrow from patients with ALL at the time of diagnosis or relapse. All samples (Supplementary Table 5) were transplanted into sublethally irradiated NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ mice (NSG mice, The Jackson Laboratory) through tail-vein injection. After samples were collected, ALL xenografts were cultured on OP9 stroma cells in alpha minimum essential medium (MEMα, Gibco) with GlutaMAX supplemented with 20% fetal bovine serum (FBS), 1 mM sodium pyruvate, 100 IU mL−1 penicillin and 100 μg mL−1 streptomycin at 37 °C in a humidified incubator with 5% CO2. For PDXs listed in Supplementary Tables 3, 4, patient-derived xenograft models were initiated on treatment with ponatinib (Selleckchem S1490) at a dose of 40 mg kg−1 day−1 via oral gavage when the level of peripheral blood involvement by human CD45/CD19 double-positive cells exceeded 10%. Treatment continued daily until animals progressed, as defined by a return to peripheral blood involvement of more than 10% after an initial response, or until they required humane euthanasia for clinical signs of illness, at which point disease status was assessed by necropsy and post-mortem flow cytometric evaluation of peripheral blood, spleen and bone-marrow tissues. Before experimental usage, all xenografts were tested as mycoplasma free (MycoAlertPLUS, Lonza).

Mouse primary and leukaemia cells

Bone-marrow and splenic cells were harvested from female mice (Supplementary Table 7) at six to eight weeks of age without signs of inflammation. Bone-marrow cells were obtained by flushing cavities of femur and tibia with phosphate-buffered saline (PBS). After filtration through a 40-μm cell strainer and depletion of erythrocytes using a lysis buffer (RBC lysis buffer, BioLegend), cells were washed with PBS and were either frozen for storage or subjected to further experiments. Bone-marrow cells were cultured in Iscove’s modified Dulbecco’s medium (IMDM, Gibco) with GlutaMAX supplemented with 20% FBS, 100 IU ml−1 penicillin, 100 μg ml−1 streptomycin and 50 μmol l−1 β-mercaptoethanol (Gibco). Bone-marrow cells, for IL7-dependent pro-B-cell culture, were cultured in full IMDM supplemented with 10 ng ml−1 recombinant mouse IL7 (PeproTech). To generate murine B-lineage Ph+ ALL cells, pro-B cells were retrovirally transduced with BCR–ABL1; after transformation, ALL cells were cultured in IL7-withdrawn IMDM. NRASG12D B-ALL cells were maintained with IL7 supplemented IMDM. To examine the role of MYC and BCL6 in NRASG12D B-ALL cells (Fig. 2h), IMDM was supplemented with CXCL12 (100 ng ml−1) and a low dose of IL7 (0.2 ng ml−1). All mouse experiments were subject to institutional approval by the City of Hope Comprehensive Cancer Center (COHCCC).

Genetic mouse models

C57BL/6, NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ, Mb1–Cre, Stat5fl/fl and Ptpn6fl/fl mice were purchased from The Jackson Laboratory. Mapk1fl/fl and Blnk−/− mice were obtained from M. McMahon and H. Jumaa, respectively. MyceGFP/+ mice (Jackson Laboratory stock number 021935) were crossed with Bcl6mCherry/+ mice14 to generate MyceGFP/+; Bcl6mCherry/+ reporter mice. For animals bred in-house, littermates of the same sex were randomized to experimental groups. All mouse breeding and experiments were subject to institutional approval by the Beckman Research Institute Animal Care and Use Committee. We complied with all relevant ethical regulations. All mouse models used in the study are listed in Supplementary Table 7.

Retroviral constructs and transduction

Transfection of retroviral constructs (Supplementary Table 8) was performed using Lipofectamine 2000 (Invitrogen) with Opti-MEM media (Invitrogen). Retroviral supernatant was produced by co-transfecting HEK 293FT cells with the plasmids pHIT60 (gag-pol) and pHIT123 (ecotropic env). Lentiviral supernatant was produced by co-transfecting HEK 293FT cells with the plasmids pCDNL-BH and VSV-G. 293FT cells were cultured in high-glucose Dulbecco’s modified Eagle’s medium (DMEM, Invitrogen) with GlutaMAX containing 10% fetal bovine serum, 100 IU ml−1 penicillin, 100 μg ml−1 streptomycin, 25 mmol l−1 HEPES, 1 mmol l−1 sodium pyruvate and 0.1 mmol l−1 non-essential amino acids. Regular medium was replaced after 16 h by growth medium containing 10 mmol l−1 sodium butyrate. After incubation for 8 h, the medium was replaced with regular growth medium. Twenty-four hours later, retroviral supernatant was collected, filtered through a 0.45-μm filter and loaded by centrifugation (2,000g, 90 min at 32 °C) onto non-tissue six-well plates coated with 50 μg ml−1 RetroNectin (Takara). Transduced cells with oestrogen-receptor fusion proteins were induced with 4-hydroxytamoxifen (4-OHT or tamoxifen; 1 μM). Cells transduced with constructs carrying an antibiotic-resistance marker were selected with its respective antibiotic.

Western blotting

Cells were lysed in CelLytic buffer (Sigma-Aldrich) supplemented with 1% protease-inhibitor cocktail (Calbiochem Millipore) and 1 mM phenylmethylsulfonyl fluoride (PMSF). Then, 20 μg of protein mixture per sample were separated on mini precast gels (Biorad) and transferred onto nitrocellulose membranes. The primary antibodies used are listed in Supplementary Table 11. For protein detection, we used the WesternBreeze Immunodetection System (Invitrogen), and detected light emission either by film exposure or by using the ChemiDoc MP Imaging System (BioRad). Western blot analyses (Fig. 1a) were performed to evaluate correlation between ERK-pT202/Y204 and STAT5-pY694 levels as determined by densitometry (ImageJ) in patient-derived B-ALL cells (n = 23).

Single-cell phosphoprotein analyses

For single-cell analyses of STAT5 phosphorylation (Y694) and ERK phosphorylation (T202/Y204), 500,000 cells were resuspended in 2 ml of 1× suspension buffer (Small scWest kit for small cells, K500, ProteinSimple) and run on two duplicate scWest chips. On each scWest chip, there is a 2 × 8 array of 400-well blocks patterned into a precast polyacrylamide gel. For single-cell western analyses, optimal cell loading is achieved when 2% or fewer wells contain multiple cells. Before loading of cells, scWest chips were hydrated in 1× suspension buffer for at least 10 min at room temperature. Cells were then loaded onto each chip and allowed to settle for roughly 7–10 min. Following removal of unsettled cells, scWest chips were inserted into Milo (ProteinSimple) for lysis of cells (10 s) and size-based protein separation (60 s, 240 V) according to the manufacturer’s instructions. UV capture immobilized protein bands in the gel on the scWest chip. Chips were then probed with primary antibodies (1:15; Supplementary Table 11) overnight at 4 °C, followed by probing with fluorescent secondary antibodies (1:20; Supplementary Table 11) the following day for 1  h at room temperature. To verify that each detected peak was associated with cell occupancy in each well, we probed chips for histone H3 (CST catalogue number 4499; 1:30) and then carried out DNA staining using TOTOTM-1 iodide (514/533; ThermoFisher Scientific). Chips were scanned using a microarray scanner to measure fluorescence intensities. Scout Software (ProteinSimple) was used for peak identification and data analysis. Each data point was inspected to verify that the signal was associated with a peak located at the correct peak location (distance from the well centre) in each lane in the gel on the chip.

For single-cell western analyses, optimal cell loading is achieved when 2% or fewer wells contain multiple cells (https://www.proteinsimple.com/milo.html). As an independent validation of cell loading conditions, Ramos lymphoma cells were genetically engineered to express either SYK or ZAP70, mixed in a ratio of 1 to 1, and analysed by single-cell western blot. Our analysis revealed less than 1.5% of wells (4/291) with detectable expression of both SYK and ZAP70, resulting from two cells in the same well. This is consistent with the manufacturer’s expectations of cell loading on to scWest chips and cell occupancy (ProteinSimple). The relationship between STAT5-pY694 and ERK-pT202/Y204 was visualized by a bivariate plot with nonparametric density contours using JMP version 9.0 (SAS Institute).

CRISPR-mediated gene deletion

For gene deletion in human cells, chemically synthesized CRISPR RNAs (100 μmol l) and trans-activating CRISPR RNAs (100 μmol l−1) were annealed by incubation at 95 °C for 5 min. Recombinantly produced Cas9 (40 μmol l−1) was then added to the RNA mixture to produce RNA ribonucleoprotein (RNP) complexes. Electroporation was performed by using pulse code EH-115 on a Lonza 4D 96-well electroporation system. Predesigned Alt-R CRISPR–Cas9 guide RNAs and non-targeting control guide RNAs were purchased from IDT (Supplementary Table 9).

Single-cell mutation analysis of Ph + ALL cells

Patient-derived xenograft models (PDXs) of Ph+ ALL cases were injected via the tail vein (1.0 × 106 cells per mouse) into non-irradiated 6–8-week-old female NSG mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJl, Jackson Laboratory stock number 005557). Upon engraftment (and with peripheral blood involvement of 10% or more), mice were randomized to treatment with vehicle control, ABL001 (an allosteric BCR–ABL inhibitor, courtesy of Novartis AG; 30 mg kg−1 twice daily via oral gavage), ponatinib (Selleckchem S1490; 40 mg kg−1 day−1 via oral gavage), or both. Subsets of animals were killed at engraftment (before treatment initiation) or on day five of treatment (for pharmacodynamic assessment), and the remainder were killed when they showed clinical signs of illness (reflecting progression or toxicity). Leukaemia cells were enriched from bone marrow and/or spleen via immunomagnetic depletion of murine cells and underwent targeted sequencing with a custom Archer VariantPlex assay. Across the majority of PDX models (notable exceptions being 1F10, 2B10 and 2G10), a moderate frequency of treatment-emergent mutants in the ERK pathway was observed. In most cases, these were mutually exclusive with ABL mutations (STAT5 pathway; Supplementary Table 3). To analyse whether treatment-emergent ERK- and STAT5-pathway mutations co-occurred in the same cell, we carried out single-cell mutation analysis for 1F10, 2B10 and 2G10 (Supplementary Table 4). Single-cell sorting for PDXs was performed on a FACSAria II (BD Biosciences). Genomic DNA (gDNA) was amplified from single cells using the REPLI-g single cell kit (Qiagen) according to the manufacturer’s instructions. Genomic regions of interest were amplified by polymerase chain reaction (PCR) using Q5 hot start high-fidelity DNA polymerase (New England BioLabs) with M13 tailed primers, and were analysed for mutations in ABL1, KRAS and PTPN11 by Sanger sequencing using M13 primers (Mclab). The primer sequences are listed in Supplementary Table 9.

Flow cytometry

Approximately 106 cells per sample were resuspended in PBS with 4,6-diamidino-2-phenylindole (DAPI; 0.75 μg ml−1) as a dead-cell marker. For competitive growth assays, the percentage of GFP+ or KO1+ cells was monitored by flow cytometry. For cell surface staining, PBS-washed cells were blocked with Fc blocker for 10 min on ice and then stained with the indicated antibodies (listed in Supplementary Table 11) or with an isotype control for 25 min on ice. Cells were then washed and resuspended in chilled PBS containing 0.75 μg ml−1 of DAPI to exclude dead cells. For annexin V staining, annexin V binding buffer (BD Bioscience) was used instead of PBS, and 7-aminoactinomysin D (7-AAD; BD Bioscience) instead of DAPI. Phycoerythrin (PE)-labelled annexin V was purchased from BD Bioscience. Acquisition was performed on a LSRFortessa flow cytometer (BD Biosciences). Fluorescence-based cell sorting was performed on a FACSAria II (BD Biosciences). FACS data were analysed with FlowJo software (FlowJo).

Colony-forming assays

Methylcellulose colony-forming assays were performed with 10,000 cells. Cells were resuspended in mouse MethoCult medium (without cytokines for BCR–ABL1-transformed cells; with IL7 for NRASG12D-expressing cells) and cultured on 3-cm diameter dishes, with an extra water-supply dish to prevent evaporation. Colonies were imaged and counted after 10–14 days using GelCount (Oxford Optronix).

Bone-marrow cells harvested from Mb1–Cre; Bcr+/LSL–BCR/ABL mice, cultured in the presence of IL7, were primed with vehicle control or trametinib (1 nmol l−1) for 10 days before colony-forming assays. Bone-marrow cells harvested from Mb1–Cre; LSL–K-ras-G12D mice, cultured in the presence of IL7, were primed with vehicle control or ruxolitinib (10 nmol l−1) for 10 days before colony-forming assays. Murine wild-type BCR–ABL1 B-ALL cells were primed with vehicle control, DPH (1 μM), trametinib (1 nM) or DPH plus trametinib for 10 days before colony-forming assays. Murine wild-type NRASG12D B-ALL cells cultured in the presence of IL7 were primed with vehicle control, BCI-215 (50 nM), ruxolitinib (10 nM), or BCI-215 plus ruxolitinib for 10 days before colony-forming assays.

For colony-forming assays with hIL2Rβ–TetO–NRASG12D cells, cells were replenished with doxycycline and hIL2 every other day and were imaged after 10 days.

Cell viability assays and viable cell counts

Forty thousand patient-derived B-ALL cells were seeded in a volume of 80 μl in complete growth medium on 96-well plates. Compounds were added at the indicated concentrations, specified in the relevant figure or figure legends, in a total volume of 100 μl. After culturing for three days, Cell Titer Glo (Promega) assays were performed according to the manufacturer’s instructions. Medium without cells was used as blank. Relative viability was calculated using baseline values of untreated cells as a reference. Combination indices were calculated using CompuSyn software to determine interaction (synergistic, combination index less than 1; additive, combination index = 1; antagonistic, combination index more than 1) between the two agents. A constant ratio combination design was used. Cell viability upon the genetic loss of function of Blnk was monitored by flow cytometry using DAPI (0.75 μg ml−1) as a dead-cell marker. To determine the number of viable cells, we applied the trypan blue exclusion method, using a Countess II FL automated cell counter.

Senescence β-galactosidase staining

Patient-derived B-ALL cells (5 × 105) expressing doxycycline-inducible BCR–ABL1 or NRASG12D (with or without doxycycline treatment) were seeded onto each Cell-Tak-coated well on a 24-well cell-imaging plate. Senescence-activated β-galactosidase staining was then performed using a senescence β-galactosidase staining kit (Cell Signaling Technology) according to the manufacturer’s instructions.

Pharmacological inhibitors

Imatinib was purchased from LC Laboratories. Stock solutions were prepared in sterile water at 10 mmol l−1 and stored at −20 °C. A small-molecule inhibitor of DUSP6 (2-benzylidene-3-(cyclohexylamino)-1-lndanone hydrochloride; BCI) and the ABL1 activator DPH were purchased from Sigma-Aldrich. BCI-215 was a gift from A. Vogt. A MEK inhibitor (trametinib), JAK inhibitor (ruxolitinib) and ponatinib were purchased from LC Laboratories. A STAT5 inhibitor (pimozide) was purchased from TOCRIS. Stock solutions were prepared in DMSO at 10 mmol l−1 and stored at −20 °C. For in vivo experiments, compounds were dissolved in NMP:PEG300 (1:9). Ph-like ALL cells (Fig. 2e) were treated with ruxolitinib for 20 h and then processed for western blotting.

Analysis of ChIP–seq data

Chromatin immunoprecipitation with sequencing (ChIP–seq) tracks for anti-ELK1, anti-CREB1, anti-c-JUN and anti-JUNB antibodies in a normal human B-cell sample (GM12878; see the Encyclopedia of DNA Elements (ENCODE; https://www.encodeproject.org)) on the PTPN6 gene promoter region are shown (Extended Data Fig. 7b). The ChIP–seq peaks were called using the MACS peak caller by comparing read density in the ChIP experiment relative to the input chromatin control reads, and are shown as bars under each wiggle track.

Mutational exclusivity analysis

The samples for ALL analysis originate from six different sources, including the St Jude Children’s Research Hospital Pecan Data Portal (https://pecan.stjude.cloud/proteinpaint/heatmap/HM,BALL), published studies25,26,27, and samples analysed in the Müschen and Weinstock laboratories (Supplementary Table 1). All AML whole-genome and whole-exome variant calls were downloaded from COSMIC (n = 916; https://cancer.sanger.ac.uk/cosmic; date of access 27 July 2019), and filtered to retain only coding, non-synonymous variants (Supplementary Table 2). All statistical analysis and visualization were performed in R (version 3.6.1). Variants were summarized at the gene level to obtain a gene-alteration matrix; the level of co-occurrence/mutual exclusivity was tested by Fisher’s exact test and adjusted for multiple testing using the Benjamini–Hochberg method. Co-occurrence networks were generated using the igraph package (version 1.2.4.1), with vertex size weighted by the frequency of the gene alteration and edges weighted by the co-occurrence score (defined here as the negative log10 P-value multiplied by the log odds ratio (OR)). Overall shifts in co-occurrence between pathways were tested by two complimentary methods. First, differences in logOR were tested by Mann–Whitney U-test and Kruskal–Wallis rank sum test; mutation pairs with frequencies of less than 5% and Fisher’s false discovery rates of more than 0.5 were excluded to prevent extreme logORs at low frequencies from biasing the results. Second, observed/expected values were permuted for 10,000 iterations with shuffled gene-to-pathway assignments to obtain a pathway-independent null distribution for comparison with observed values. Codes for mutual exclusivity analyses are available on GitHub at https://github.com/muschenlab/Stat5-vs-Erk under the MIT license, which allows for reuse and distribution.

Pathway mutations and IGLK/IGLL configuration

To investigate mutations in human STAT5 and ERK pathways, we analysed 95 phase II samples with both copy number and mutation data from the Target phase II data set (‘all_phase2_target_2018_pub.tar.gz’, available from cBioportal (https://www.cbioportal.org; ref. 28). To test whether these lesions are linked to cellular origins from distinct stages of B-cell development, we assigned IGLK or IGLL rearrangement events on the basis of intersecting the coordinate ranges of segmented copy numbers (log2 mean segmented copy number values below −0.7) and loci coordinates (IGLK 22q11.22, chr22:22375022–23270762 and IGLL 2p11.2, chr2:87353672–92167812). A gene list for mutations leading to STAT5 or ERK activation (7 and 32 genes, respectively) was manually curated. FLT3 mutations were considered to be STAT5 activating unless they co-occurred with ERK-activating mutations. Next, the STAT5 and ERK mutations were assigned as occurring either with the germline or the rearranged IGLK/IGLL configuration. Three samples with both ERK- and STAT5-activating mutations were removed from the analysis, resulting in a total of 45 patients who were classified on the basis of STAT5- or ERK-activating mutations and rearrangement status. To test whether STAT5-activating mutations are associated particularly with the germline configuration, while ERK-activating mutations are associated with the rearranged IGKL/IGLL configuration, in a mutually exclusive manner, we applied Fisher’s exact test with alternative hypothesis ‘less’ in R.

In vivo transplantation experiments

The indicated numbers of leukaemia cells were injected into sublethally irradiated (200 cGy) NSG mice via the tail vain. We randomly allocated 8–10-week-old female NSG mice before injection. Mice were euthanized when they showed signs of leukaemia burden, including hunched posture, weight loss and inability to move. Bone-marrow and/or spleen cells were harvested to test for leukaemic infiltration by flow cytometry. Bioimaging of leukaemia progression in mice was performed at the indicated time points using an in vivo IVIS 100 bioluminescence/optical imaging system (Xenogen). d-Luciferin (Promega) dissolved in PBS was injected intraperitoneally at a dose of 2.5 mg per mouse 15 min before measurement of the luminescence signal. General anaesthesia was induced with 5% isoflurane and continued during the procedure with 2% isoflurane introduced through a nose cone. Kaplan–Meier survival analyses were performed using GraphPad Prism 7 (GraphPad Software) to compare overall survival. Mantel–Cox log-rank tests were used for statistical analysis with GraphPad Prism 7. The minimal number of mice in each group was calculated by using the ‘cpower’ function in the R/Hmisc package. Mice were randomly allocated to groups by investigators who did not participate in any subsequent analysis. The investigators who carried out the mouse experiments (injections, drug treatments and monitoring) were blinded to the randomization process.

Synergy for in vivo transplantation experiments

To assess the additive versus synergistic activity of drug 1 plus drug 2 in vivo, we adapted the Bliss independence model to survival analysis29. With this approach, treatments are Bliss ‘independent’ if the fraction of cells surviving combination therapy equals the product of fractions that survive the individual treatments (for example, if drug 1 treatment kills 90% and drug 2 kills 50%, Bliss independence predicts that the combination kills 95%)30. A Weibull distribution, {exp[−(t/β)α]} (ref. 31), was fitted to survival data for each condition (untreated, drug 1, drug 2, drug 1 plus drug 2), and distributions of survival benefits (treated survival time minus untreated survival time) were computed for each treatment. Survival benefits of drug 1 and drug 2 were summed and added to the untreated survival distribution to compose a ‘sum of benefits’ survival distribution. Confidence intervals were based on 10,000 such simulations, in which each treatment’s survival distribution was a likelihood-weighted sample of Weibull parameters (α, β) based on the relative likelihood of having made the survival observations O = (t1, t2,…ti) given those parameters: L(O | α,β) = Πi{exp[−(ti/β)α] × ti α−1 × α × β−α}. L is the likelihood; α and β are Weibull parameters: β is the shape parameter (Weibull slope) and α is the scale parameter. The P-value for synergy between drug 1 and drug 2 was the probability of drawing data from the sum-of-benefits model ensembles with median survival duration equal to or greater than that experimentally observed with drug 1 and drug 2.

Statistics and reproducibility

Statistical analysis was performed using GraphPad Prism 8(GraphPad Software) and results are shown as mean ± s.d. Unpaired two-tailed Student’s t-tests were used to evaluate differences between two groups and P-values were plotted (P < 0.05 is considered statistically significant) unless indicated otherwise. For in vivo transplantation experiments, the minimal number of mice in each group was calculated using the ‘cpower’ function in the R/Hmisc package. Kaplan–Meier survival analysis was used to estimate overall survival scores with GraphPad Prism 8, and log-rank tests were used to compare differences between two groups. For clinical outcome analyses, we compared overall survival and relapse-free survival for patients with higher- and lower-than-median messenger RNA levels (top versus bottom half) of the gene of interest. Log-rank tests were used to compare survival differences between patient groups. The R package ‘survival’ version 2.35-8 was used for the survival analysis, and Cox proportional hazards regression model in R package for the multivariable analysis (R Development Core Team, 2009; http://www.R-project.org; ref. 32). To determine synergism effects of BCI-215 plus ruxolitinib or DPH plus trametinib in patient-derived B-ALL cells, we calculated combination indices using CompuSyn software to determine interactions (synergistic, combination index less than 1; additive, combination index = 1; antagonistic, combination index more than 1). For in vitro experiments, investigators were not blinded to allocation during experiments and outcome assessment. Mice were randomly allocated to groups by investigators who did not participate in any subsequent analysis. The investigators who carried out mouse experiments (injections, drug treatments and monitoring) were blinded to the randomization process. Experiments were repeated to ensure reproducibility of observations. No statistical methods were used to predetermine sample size.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability

Gel scans are provided in Supplementary Fig. 1. All other data are available from M.M. upon reasonable request. The Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) accession number for the gene-expression profiles of diagnostic samples from 207 patients (COG P9906) is GSE11877 (Supplementary Table 10). ChIP–seq data from human B lymphocytes (GM12878) are from ENCODE. Further information and requests for reagents can directed to and will be fulfilled by M.M. Source data are provided with this paper.

Code availability

Codes for mutual exclusivity analyses are available on GitHub at https://github.com/muschenlab/Stat5-vs-Erkunder the MIT license, which allows for reuse and distribution.

References

  1. 1.

    Fearon, E. R., Hamilton, S. R. & Vogelstein, B. Clonal analysis of human colorectal tumors. Science 238, 193–197 (1987).

    ADS  CAS  Article  Google Scholar 

  2. 2.

    Goetz, C. A., Harmon, I. R., O’Neil, J. J., Burchill, M. A. & Farrar, M. A. STAT5 activation underlies IL7 receptor-dependent B cell development. J. Immunol. 172, 4770–4778 (2004).

    CAS  Article  Google Scholar 

  3. 3.

    Malin, S. et al. Role of STAT5 in controlling cell survival and immunoglobulin gene recombination during pro-B cell development. Nat. Immunol. 11, 171–179 (2010).

    CAS  Article  Google Scholar 

  4. 4.

    Katerndahl, C. D. S. et al. Antagonism of B cell enhancer networks by STAT5 drives leukemia and poor patient survival. Nat. Immunol. 18, 694–704 (2017).

    CAS  Article  Google Scholar 

  5. 5.

    Shaw, A. C., Swat, W., Ferrini, R., Davidson, L. & Alt, F. W. Activated Ras signals developmental progression of recombinase-activating gene (RAG)-deficient pro-B lymphocytes. J. Exp. Med. 189, 123–129 (1999).

    CAS  Article  Google Scholar 

  6. 6.

    Irving, J. et al. Ras pathway mutations are prevalent in relapsed childhood acute lymphoblastic leukemia and confer sensitivity to MEK inhibition. Blood 124, 3420–3430 (2014).

    CAS  Article  Google Scholar 

  7. 7.

    Anderson, L. J. & Longnecker, R. EBV LMP2A provides a surrogate pre-B cell receptor signal through constitutive activation of the ERK/MAPK pathway. J. Gen. Virol. 89, 1563–1568 (2008).

    CAS  Article  Google Scholar 

  8. 8.

    Feldhahn, N. et al. Mimicry of a constitutively active pre-B cell receptor in acute lymphoblastic leukemia cells. J. Exp. Med. 201, 1837–1852 (2005).

    CAS  Article  Google Scholar 

  9. 9.

    Yasuda, T. et al. Erk kinases link pre-B cell receptor signaling to transcriptional events required for early B cell expansion. Immunity 28, 499–508 (2008).

    CAS  Article  Google Scholar 

  10. 10.

    Shojaee, S. et al. PTEN opposes negative selection and enables oncogenic transformation of pre-B cells. Nat. Med. 22, 379–387 (2016).

    CAS  Article  Google Scholar 

  11. 11.

    Martincorena, I. et al. Somatic mutant clones colonize the human esophagus with age. Science 362, 911–917 (2018).

    ADS  CAS  Article  Google Scholar 

  12. 12.

    Mandal, M. et al. Ras orchestrates exit from the cell cycle and light-chain recombination during early B cell development. Nat. Immunol. 10, 1110–1117 (2009).

    CAS  Article  Google Scholar 

  13. 13.

    Duy, C. et al. BCL6 is critical for the development of a diverse primary B cell repertoire. J. Exp. Med. 207, 1209–1221 (2010).

    CAS  Article  Google Scholar 

  14. 14.

    Geng, H. et al. Self-enforcing feedback activation between BCL6 and pre-B cell receptor signaling defines a distinct subtype of acute lymphoblastic leukemia. Cancer Cell 27, 409–425 (2015).

    CAS  Article  Google Scholar 

  15. 15.

    Swaminathan, S. et al. Mechanisms of clonal evolution in childhood acute lymphoblastic leukemia. Nat. Immunol. 16, 766–774 (2015).

    CAS  Article  Google Scholar 

  16. 16.

    Walker, S. R. et al. STAT5 outcompetes STAT3 to regulate the expression of the oncogenic transcriptional modulator BCL6. Mol. Cell. Biol. 33, 2879–2890 (2013).

    CAS  Article  Google Scholar 

  17. 17.

    Korotchenko, V. N. et al. In vivo structure-activity relationship studies support allosteric targeting of a dual specificity phosphatase. ChemBioChem 15, 1436–1445 (2014).

    CAS  Article  Google Scholar 

  18. 18.

    Shojaee, S. et al. Erk negative feedback control enables pre-B cell transformation and represents a therapeutic target in acute lymphoblastic leukemia. Cancer Cell 28, 114–128 (2015).

    CAS  Article  Google Scholar 

  19. 19.

    Yang, J. et al. Discovery and characterization of a cell-permeable, small-molecule c-Abl kinase activator that binds to the myristoyl binding site. Chem. Biol. 18, 177–186 (2011).

    CAS  Article  Google Scholar 

  20. 20.

    Heltemes-Harris, L. M. et al. Sleeping Beauty transposon screen identifies signaling modules that cooperate with STAT5 activation to induce B-cell acute lymphoblastic leukemia. Oncogene 35, 3454–3464 (2016).

    CAS  Article  Google Scholar 

  21. 21.

    Porpaczy, E. et al. Aggressive B-cell lymphomas in patients with myelofibrosis receiving JAK1/2 inhibitor therapy. Blood 132, 694–706 (2018).

    CAS  Article  Google Scholar 

  22. 22.

    Xiao, W. et al. Tumor suppression by phospholipase C-beta3 via SHP-1-mediated dephosphorylation of Stat5. Cancer Cell 16, 161–171 (2009).

    CAS  Article  Google Scholar 

  23. 23.

    Fusaki, N. et al. BLNK is associated with the CD72/SHP-1/Grb2 complex in the WEHI231 cell line after membrane IgM cross-linking. Eur. J. Immunol. 30, 1326–1330 (2000).

    CAS  Article  Google Scholar 

  24. 24.

    Imamura, Y. et al. BLNK binds active H-Ras to promote B cell receptor-mediated capping and ERK activation. J. Biol. Chem. 284, 9804–9813 (2009).

    CAS  Article  Google Scholar 

  25. 25.

    Nikolaev, S. I. et al. Frequent cases of RAS-mutated Down syndrome acute lymphoblastic leukaemia lack JAK2 mutations. Nat. Commun. 5, 4654 (2014).

    ADS  CAS  Article  Google Scholar 

  26. 26.

    Herold, T. et al. Adults with Philadelphia chromosome-like acute lymphoblastic leukemia frequently have IGH-CRLF2 and JAK2 mutations, persistence of minimal residual disease and poor prognosis. Haematologica 102, 130–138 (2017).

    CAS  Article  Google Scholar 

  27. 27.

    Jerchel, I. S. et al. RAS pathway mutations as a predictive biomarker for treatment adaptation in pediatric B-cell precursor acute lymphoblastic leukemia. Leukemia 32, 931–940 (2018).

    CAS  Article  Google Scholar 

  28. 28.

    Cerami, E. et al. The cBio Cancer Genomics Portal: An opening platform for exploring multidimensional cancer genomics data. Cancer Discov. 2, 401–404 (2012).

    Article  Google Scholar 

  29. 29.

    Koch, R. et al. Biomarker-driven strategy for MCL1 inhibition in T-cell lymphomas. Blood 133, 566–575 (2019).

    CAS  Article  Google Scholar 

  30. 30.

    Bliss, C. I. The toxicity of poisons applied jointly. Ann. Appl. Biol. 26, 585–615 (1939).

    CAS  Article  Google Scholar 

  31. 31.

    Weibull, W. A statistical distribution function of wide applicability. J. Appl. Mech. 18, 293–297 (1951).

    ADS  MATH  Google Scholar 

  32. 32.

    R Development Core Team. A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna (2009).

Download references

Acknowledgements

We thank L. Klemm, F. Auer and J. Winchester for help with some of the experiments, and present and former members of the Müschen Laboratory for their support and helpful discussions. Research in the Müschen Laboratory is funded by the National Institutes of Health (NIH) through National Cancer Institute (NCI) R35CA197628, R01CA157644, R01CA213138 and P01CA233412 (to M.M.); the Howard Hughes Medical Institute (HHMI-55108547 to M.M.); the Norman and Sadie Lee Foundation; the Falk Trust through a Falk Medical Research Trust Transformational Award; the Pediatric Cancer Research Foundation (PCRF) and the V Foundation for Cancer Research, the Leukemia and Lymphoma Society through MCL-7000-18 and a Blood Cancer Discoveries Grant BCDG-20327-20 (to M.M.); and the California Institute for Regenerative Medicine (CIRM) through DISC2-10061. D.M.W. is supported by NCI grants R35CA231958 and U54CA217377. M.M. is a Howard Hughes Medical Institute (HHMI) Faculty Scholar. M.M. and S.I. are supported by the Jacki & Bruce Barron Cancer Research Scholars’ Program. M.A.M. is supported by K08 Clinical Investigator Award CA212252-01A1. T.S. is a Lymphoma Research Foundation Grantee. V.K. is supported by Career Development Fellow grant (5491-20) from the Leukemia and Lymphoma Society, and by a Young Investigator Award from Alex’s Lemonade Stand Foundation.

Author information

Affiliations

Authors

Contributions

L.N.C. designed and performed experiments and contributed to all aspects of the study, in particular western blotting, single-cell phosphoprotein analyses, single-cell mutation analyses, CRISPR-mediated gene deletion, flow-cytometry analysis, growth competition assays, cell sorting, colony-forming assays, viable cell counts, determination of drug responses, in vitro combination treatment assays, in vivo transplantation experiments, and data analysis. M.A.M. performed in vivo ponatinib treatment of patient-derived xenograft models of Ph+ ALL cases in NSG mice, and subsequent targeted sequencing of leukaemia cells harvested from sacrificed mice; M.A.M. also analysed data and provided PDXs. M.E.R. designed figures, performed mutational exclusivity analyses and biostatistical analyses. R.C. performed western blotting, flow cytometry and data analysis. T.S. performed single-cell mutation analysis, flow-cytometry analysis and data analysis and provided expertise in single-cell western blotting. J.W.L. and G.X. performed flow-cytometry analysis and analysed data. K.N.C. performed in vivo transplantation experiments, flow-cytometry analysis and data analysis. K.K. generated the B-cell-specific Myc–eGFP × Bcl6–mCherry double reporter knock-in mouse model, performed flow-cytometry analysis and generated Contour plots for visualization of data obtained from single-cell phosphoprotein analysis. V.K. performed annexin V staining and senescence β-galactosidase staining and analysed data. M.A.A. and E.A. performed in vivo transplantation experiments and data analysis. G.D. performed growth competition assays and data analysis. C. Hurtz, S.S., C. Hong and Z.C. performed in vitro experiments and analysed data. P.P. performed analysis of TARGET data. C.W.C. and J.C. provided expertise for gene-editing experiments. M.H. provided expertise in bioinformatics analysis. A.V. provided the small-molecule inhibitor BCI-215 and D.M.W. provided PDXs. M.A.N. and A.P.W. provided expertise in surface proteomics in leukaemia biology. S.I. and O.L. provided expertise in leukaemia biology. H.G. performed biostatistical analysis and modeling synergy of in vivo transplantation experiments. M.M. secured funding, developed the ‘divergent pathway’ concept, provided mentorship and wrote the manuscript with the input of all co-authors.

Corresponding author

Correspondence to Markus Müschen.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Peer review information Nature thanks Michael Reth, Oscar Rueda, Veronika Sexl and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Segregation of STAT5- and ERK-activating mutations in human ALL and AML.

ae, We studied STAT5- and ERK-pathway mutations in 1,148 patient-derived B-ALL samples (Supplementary Table 1). The null hypothesis is that STAT5- and ERK-pathway mutations occur independently of each other. The expected co-occurrence of mutations in both pathways under the null hypothesis was 121. a, The observed co-occurrence of the two mutations was 35, significantly lower than expected (odds ratio (OR) = 0.13; P = 2.2 × 10−16; Fisher’s exact test). b, To analyse gene–gene co-occurrence in a more unsupervised manner, we ran a Fisher’s test for each gene pair and plotted the results as a co-occurrence network. The pathway assignment for each gene is indicated by the node colour (white, STAT5; grey, ERK), and the mutation frequency by the node size. The direction of the Fisher’s result is indicated by the line colour (green, positive/greater than expected; red, negative/lower than expected); the line width represents the strength of association (–log10 P-value × |logOR|). c, Volcano plot of gene–gene co-occurrence results. Each point represents a gene pair, coloured by the pathway assignment for the pair (green, both STAT5; red, both ERK; grey, interpathway); selected gene pairs are labelled. The overall difference in co-occurrence between pathways was tested by two complementary methods, as follows. d, First, a Fisher’s test was run over 10,000 permutations, with random shuffling of gene-pathway assignments on each iteration, to generate a null distribution for the hypothesis that the pathway does not affect co-occurrence, which we compared against the observed value from ALL patient data. e, Second, overall shifts in the distribution of logORs between pathways for Fisher’s results on individual gene pairs were tested by Welch’s two-sample t-test (left-hand part of panel; two-sided, P = 0.001) and one-way analysis of variance (ANOVA; right-hand part of panel; P = 0.0004) with Tukey’s HSD post-test (P = 0.0003 and P = 0.16). Low-frequency, non-significant gene pairs were excluded to avoid extreme odds ratios biasing results. f–j, The same analyses were carried out on 916 patient-derived AML samples (Supplementary Table 2). The expected co-occurrence of mutations in both pathways under the null hypothesis was 34; the observed co-occurrence was 24, significantly lower than expected (f; OR = 0.59, P = 0.033, Fisher’s exact test). j, Left, Welch’s two-sample t-test (two-sided, P = 0.82); right, one-way ANOVA (P = 0.57).

Extended Data Fig. 2 Single-cell phosphoprotein analyses of patient-derived B-ALL samples reveal segregation of STAT5 and ERK phosphorylation to competing clones.

ae, For single-cell phosphoprotein analyses, scWest chips were used to capture individual cells and perform size-based protein separation. Each chip includes 16 arrays of 400-well blocks (6,400 wells) on a polyacrylamide gel. Single-cell suspensions of patient-derived B-ALL cells were loaded onto the scWest chip and inserted into Milo (ProteinSimple) for cell lysis, size-based protein separation and UV capture to immobilize protein bands. a, A representative scanned image of the scWest chip probed with histone H3 antibodies to confirm cell occupancy, followed by fluorescent secondary antibodies (n = 16 independent biological samples). b, c, Chips were probed with anti-STAT5-pY694 and anti-ERK-pT202/Y204 antibodies followed by fluorescent secondary antibodies for simultaneous detection of both phosphoproteins. At the left are representative images of signals observed for STAT5-pY694 and ERK-pT202/Y204. At the right, fluorescence intensity was plotted against distance from the well centre (peak location, in μm). In addition to histone H3, chips were stained for DNA using TOTO-1 dye to verify cell occupancy. Each signal was inspected to confirm that it was associated with a peak located at the correct distance from the well centre (n = 16 independent biological samples, each in triplicate). Chips were scanned using a microarray scanner, and peak identification was performed using Scout Software (ProteinSimple). For single-cell western analyses, optimal cell loading is achieved when 2% or fewer wells contain multiple cells (https://www.proteinsimple.com/milo.html). d, e, Single-cell phosphoprotein analyses for STAT5-pY694 and ERK-pT202/Y204 were performed for patient-derived B-ALL samples (six independent biological samples, each in triplicate) with anti-STAT5-pY694 and anti-ERK-pT202/Y204 antibodies for simultaneous detection of both STAT5 and ERK phosphorylation. scWest chips were then probed for histone H3 and TOTO-1 (DNA stain) to verify cell occupancy. Scout Software (ProteinSimple) was used for peak identification and data analysis. Each data point was inspected to confirm that the signal detected was associated with a peak located at the correct peak location (distance from the well centre). d, Heatmaps illustrating cells that express STAT5-pY694 (green) and/or ERK-pT202/Y204 (red). Sample names and genetic characteristics are shown at the top of each heatmap. e, Tables summarizing the number of cells expressing neither STAT5-pY694 nor ERK-pT202/Y204 and the number of cells expressing either STAT5-pY694 or ERK-pT202/Y204, or in rare cases, both. For single-cell western analyses, optimal cell loading is achieved when 2% or fewer wells contain multiple cells.

Extended Data Fig. 3 Phenotypic characterization of B-ALL cells driven by oncogenic STAT5 or ERK.

ag, Eight different B-ALL PDX samples (Supplementary Table 5) were analysed by flow cytometry for surface expression of CRLF2 (a), VpreB (b), CD20 (c), IgM (d), Ig-κλ (e), CD34 (f) and IL7R (g). Normal CD19+ B-cell precursors from a normal human bone-marrow (BM) sample and mature B cells from a normal human peripheral blood (PB) sample, as well a mature B-cell lymphoma cell line (DLBCL), were used as references. Shown are representative FACS plots from three independent experiments.

Extended Data Fig. 4 Phenotypic differences between B-ALL cells driven by oncogenic STAT5 or ERK reflect developmental rewiring during early B-cell development.

a, Mouse B-cell precursors, STAT5-driven B-ALL cells, BCR–ABL1-driven B-ALL cells and NRASG12D-driven B-ALL cells were analysed by flow cytometry for surface expression of CD43, IL7R, VpreB, IgM, Ig-κ, IgD and CD21. Normal CD19+ B-cell precursors from bone-marrow (BM) cells and CD19+ spleen B cells from wild-type mice were used for reference. Shown are representative FACS plots from three independent biological replicates. b, For STAT5-driven B-ALL cases (n = 26) and ERK-driven B-ALL cases (n = 67), mRNA levels of IGLV, IGHM, IKZF1 and BCL6 were analysed in microarray results and associated with clinical outcome (COG P9906) at the time of diagnosis. Patients in each group were segregated into two subgroups on the basis of higher- versus lower-than-median mRNA levels for each of these four genes. Overall survival and relapse-free survival for the Kaplan–Meier plots are shown. Mantel–Cox log-rank tests (two-sided) were used to determine statistical significance. y-axes indicate time (years). 

Extended Data Fig. 5 Concurrent activation of STAT5 and ERK signalling induces B-cell senescence and cell death.

a, Levels of NRASG12D, STAT5-pY694, STAT5, STAT3-pS727, STAT3, ERK1/2-pT202/Y204 and ERK1/2 upon doxycycline-induced expression of NRASG12D or STAT5ACA (n = 3 independent experiments). b, c, IL7-dependent mouse pro-B cells (b) or pre-B cells (c) were transduced with empty vector (EV), BCR–ABL1–GFP or LMP2A–GFP. Changes in proportions of GFP+ cells were monitored by flow cytometry (n = 3 independent experiments; means ± s.d.). d, e, FACS analysis of annexin V/7AAD (left) and senescence β-galactosidase staining (centre) were performed with pro-B cells (d) and pre-B cells (e) expressing EV, LMP2A or BCR–ABL1 as indicated. Right, levels of p16 and p21 were examined (n = 3 independent experiments). Shown are representative FACS plots and images from three independent experiments. P-values were determined by two-tailed t-test. β-Galactosidase staining for senescent cells is quantified as the mean percentage of cells positive for staining ± s.d. f, As in normal B-cell development, B-ALL subtypes can be traced to specific differentiation stages with distinct requirements for survival and proliferation signals. For instance, pro-B cells depend on cytokine-receptor signalling and activation of STAT5 but not ERK. Conversely, pre-B cells depend on pre-BCR signalling and activation of ERK but not STAT5. STAT5-driven B-ALL cells (left) depend on oncogenic mimics of cytokine-receptor signalling and resemble pro-B cells that depend on STAT5 activation downstream of cytokine receptors. Mimicking pre-BCR signalling at the pro-B-cell to pre-B-cell transition (right), oncogenic RAS signalling suppresses STAT5 signalling and induces de novo expression of BCL6. For gel source data, see Supplementary Fig. 1. Source data

Extended Data Fig. 6 Pharmacological inhibition of divergent pathways facilitates B-leukaemogenesis.

a, IL7-dependent pro-B cells were first retrovirally transduced with EV–GFP (left), BCR–ABL1–GFP (second from left), EV–Orange (second from right), or NRASG12D–Orange (right). One week later, GFP+ B-ALL cells were transduced with NRASG12D–Orange for concurrent activation of ERK, and Orange+ B-ALL cells were transduced with BCR–ABL1–GFP for concurrent activation of STAT5. The ability of oncogenic STAT5 (GFP+) and ERK (Orange+) signalling to contribute to the dominant clone was monitored by flow cytometry over time. Flow cytometry was performed to monitor the proportions of GFP+, Orange+ and double-positive cells at various time points following transductions. Shown are representative FACS plots from three independent biological replicates. BA–GFP, BCR–ABL1–GFP. b, Cells from a were sorted for double-positive (GFP+ and Orange+) populations, and 10, 000 cells were seeded in methylcellulose for colony-formation assays (10 days; n = 3 independent biological replicates; means ± s.d.). P = 0.003 (left) and P = 0.007 (right) (two-tailed t-test). ce, Patient-derived BCR–ABL1 B-ALL cells (MXP2) expressing Tet-On NRASG12D and patient-derived KRASG12V B-ALL cells (LAX7R) expressing Tet-On BCR–ABL1 were induced with doxycycline. Viable cell counts (c) were measured upon induction with doxycycline. Annexin V/7AAD and senescence β-galactosidase (d, e) staining were also performed, and levels of p16 and p21 were assessed (n = 3 independent experiments). Shown are representative FACS plots and images. P-values were determined by two-tailed t-tests. Quantification for senescence β-galactosidase staining: mean percentage of cells positive for staining ± s.d. f, Murine wild-type BCR–ABL1 B-ALL cells were primed with vehicle control, DPH (1 μM), trametinib (1 nM) or both for 10 days before colony-forming assays were carried out (n = 3 independent biological replicates; means ± s.d.) g, Mouse wild-type NRASG12D B-ALL cells cultured in the presence of IL7 were primed with vehicle control, BCI-215 (50 nM), ruxolitinib (10 nM), or both for 10 days before colony-forming assays were carried out (n = 3 independent biological replicates; means ± s.d.). P-values were determined by two-tailed t-test (f, g). For gel source data, see Supplementary Fig. 1. Source data

Extended Data Fig. 7 Central role of PTPN6 in enabling oncogenic ERK signalling.

a, Levels of NRAS, PTPN6-pY564, PTPN6, ERK1/2-pT202/Y204 and ERK1/2 were assessed by western blotting following doxycycline-induced expression of NRASG12D in mouse B-cell precursors (n = 3 independent experiments). b, ChIP–seq analyses of human B lymphocytes GM12878 (ENCODE) revealed binding of ERK-dependent transcription factors ELK1, CREB1, c-JUN and JUNB to the PTPN6 locus. c, Patient-derived B-ALL cells (PDX2) were treated with BCI-215 (1 μmol l−1) for various times. Levels of STAT5-pY694, STAT5, ERK1/2-pT202/pY204 and ERK1/2 were then measured by western blotting (n = 3 independent experiments). d, Patient-derived B-ALL cells (PDX2) were treated with BCI-215 (1 μmol l−1) for various times, and levels of STAT5-pY694, STAT5, PTPN6-pY564 and PTPN6 were measured by western blotting (n = 3 independent experiments). e, Ptpn6fl/fl B-cell precursors were transduced with 4-OHT-inducible Cre or EV and then induced with 4-OHT for various times and levels of STAT5-pY694, STAT5, ERK1/2-pT202/Y204, ERK1/2 and PTPN6 were measured (n = 3 independent experiments). f, Ptpn6fl/fl B-cell precursors expressing NRASG12D were transduced with GFP-tagged, 4-OHT-inducible Cre or EV. Following induction with 4-OHT, enrichment or depletion of GFP+ cells was monitored by flow cytometry. Shown are average relative changes (means ± s.d.) of GFP+ cells following induction (n = 6 independent biological replicates). g, Quantification (n = 3 independent experiments; means ± s.d.) and representative images from serial replating assays of pre-B cells transformed with NRASG12D following Cre-mediated deletion of Ptpn6. We seeded 10, 000 cells in semisolid methylcellulose and monitored colony formation for 14 days. P-values were determined by two tailed t-test. For gel source data, see Supplementary Fig. 1. Source data

Extended Data Fig. 8 BLNK enables oncogenic ERK signalling at the expense of STAT5–MYC.

a, Blnk+/+ and Blnk−/− mouse B-cell precursors expressing doxycycline-inducible Ig-HC were treated with doxycycline for 48 h. Levels of ERK1/2-pT202/pY204, ERK1/2, PTPN6-pY564, PTPN6, STAT5-pY694, STAT5, MYC, BCL6 and BLNK were measured by western blotting (n = 3 independent experiments). b, Blnk+/+ and Blnk−/− B-cell precursors expressing doxycycline-inducible NRASG12D were treated with doxycycline for 48 h. Levels of NRAS, ERK1/2-pT202/pY204, ERK1/2, PTPN6-pY564, PTPN6, STAT5-pY694, STAT5, MYC, BCL6 and BLNK were measured by western blotting (n = 3 independent experiments). cf, Viable cell counts (c, d) and viability changes (e, f) were measured upon doxycycline-induced expression of STAT5ACA or of NRASG12D in Blnk+/+ and Blnk−/− B-cell precursors at various time points (n = 3 independent experiments; means ± s.d.). g, h, The colony-forming ability of Blnk+/+ and Blnk−/− B-cell precursors was assessed upon doxycycline-induced expression of STAT5ACA (g) or NRASG12D (h). We plated 10,000 cells; shown are mean ± s.d. values of three independent experiments and representative images. P-values were determined by two tailed t-test. For gel source data, see Supplementary Fig. 1. Source data

Extended Data Fig. 9 Divergent drug responses in a STAT5- and ERK-driven pair of primary and relapse B-ALLs.

a, cd, LAX7 (at diagnosis; STAT5-driven, IL7RSI246S) and LAX7R (relapsed; ERK-driven, KRASG12V) cells were transfected with Cas9/RNPs carrying NT or BLNK guide RNAs and then mixed with GFP+ LAX7 and GFP+ LAX7R competitor cells, respectively. a, d, Enrichment or depletion of GFP+ cells was monitored by flow cytometry (n = 3 independent experiments). Data are means ± s.d. (note that the direction of the y-axis in a is reversed, starting from a fold change of 2.5 at the bottom). b, Levels of ERK1/2-pT202/Y204, ERK1/2, STAT5-pY694 and STAT5 in patient-derived B-ALL cells (IL7RSI246S LAX7, at diagnosis; KRASG12V LAX7R, relapsed) were examined by western blotting (n = 3 independent experiments, shown are two of the technical replicates from an independent experiment). c, Efficiency of CRISPR/Cas9-mediated deletion of BLNK was assessed by western blot (crRNAs, CRISPR RNAs). e, Western blotting analyses were performed to assess levels of STAT5-pY694, STAT5, STAT3-pS727, STAT3, ERK1/2-pT202/Y204 and ERK1/2 upon overnight treatment of LAX7 cells with ruxolitinib (left) or of LAX7R cells with trametinib (right); n = 3 independent experiments. f, Patient-derived B-ALL cells (LAX7 and LAX7R) were treated with increasing concentrations of ruxolitinib or trametinib for 72 h. Relative viability (n = 3; mean ± s.d.) was measured. The x-axes indicate the concentrations of ruxolitinib or trametinib used. For gel source data, see Supplementary Fig. 1. Source data

Extended Data Fig. 10 Pharmacological reactivation of suppressed divergent pathways as a therapeutic strategy in B-ALL.

a, h, Patient-derived B-ALL cells (STAT5-driven) LAX7 (a) and JFK125R (h) were treated overnight with vehicle control (DMSO), BCI-215 (1 μM), ruxolitinib (500 nM), or a combination of both. Western blotting was performed to measure levels of STAT5-pY694, STAT5, ERK1/2-pT202/pY204 and ERK1/2 (n = 3 independent experiments). b, Patient-derived B-ALL cells (ERK-driven, LAX7R) were treated overnight with vehicle control, DPH (1 μM), trametinib (500 nM), or a combination of both. Levels of STAT5-pY694, STAT5, ERK1/2-pT202/Y204 and ERK1/2 were assessed (n = 3 independent experiment). c, i, Patient-derived B-ALL cells (LAX7 and LAX7R (c) and JFK125R (i)) were treated with increasing concentrations of BCI-215, ruxolitinib or both, or DPH, trametinib, or both for three days. Percentage growth inhibition at each concentration is shown as heatmaps (n = 3 independent experiments). Combination indices (Cis) were calculated to determine synergy for treatment combinations. DPH and trametinib concentrations used for LAX7, LAX7R and JFK125R (in μM): DPH, 0, 0.16, 0.31, 0.63,1.3, 2.5, 5; trametinib, 0, 0.032, 0.063, 0.13, 0.25, 0.5, 1. Ruxolitinib and BCI-215 concentrations used for LAX7 (in μM): ruxolitinib, 0, 0.125, 0.25, 0.5, 1, 2; BCI-215, 0, 0.04, 0.08, 0.16, 0.31, 0.63. Ruxolitinib and BCI-215 concentrations used for LAX7R and JFK125R (in μM): ruxolitinib, 0, 0.063, 0.125, 0.25, 0.5, 1, 2; BCI-215, 0, 0.02, 0.04, 0.08, 0.16, 0.31, 0.63. d, j, Single-cell phosphoprotein analyses for STAT5-pY694 and ERK-pT202/Y204 were performed for patient-derived B-ALL cells LAX7 and LAX7R (d) and JFK125R (j) before in vivo treatment (n = 3 independent experiments). e, k, Patient-derived LAX7R (e) or JFK125R (k) B-ALL cells were injected into sublethally irradiated (2 Gy) NSG mice. Recipient mice injected with LAX7R were treated six times a week for four weeks with 2 mg kg−1 DPH, 0.5 mg kg−1 trametinib or both (e). Recipient mice injected with JFK125R were treated five times a week for four weeks with 2 mg kg−1 BCI-215, 30 mg kg−1 ruxolitinib or both (k). Mice were killed when they showed signs of overt leukaemia (hunched back, weight loss and inability to move). The y-axes indicate overall survival (%).Survival curves are shown (n = 6 per group). To assess additive versus synergistic activity of single versus combination treatments in vivo, we adapted the Bliss independence model for survival analysis. With this approach, treatments are Bliss ‘independent’ if the fraction of cells surviving combination therapy equals the product of fractions that survive the individual treatments. A Weibull distribution, {exp[−(t/β)α]}, is fitted to survival data for each condition, and distributions of survival benefits (treated survival time minus untreated survival time) are computed for each treatment. Survival benefits of drugs 1 and 2 are summed and added to the untreated survival distribution to compose a ‘sum of benefits’ survival distribution (see Methods). f, g, l, m, Although combination treatments prolonged overall survival, transplant-recipient mice ultimately developed overt leukaemia. B-ALLs that developed in mice bearing LAX7R (after treatment with DPH plus trametinib; e) and JFK125R (after treatment with BCI-215 plus ruxolitinib; k) were isolated from bone marrow, suspended in cell culture medium and treated with increasing concentrations of BCI-215, ruxolitinib or both, or DPH, trametinib, or both. f, l, Percentage growth inhibition at each concentration is shown as heatmaps (n = 3 independent experiments). Combination indices were calculated to determine synergy of treatment combinations. g, m, To elucidate the clonal composition of LAX7R B-ALL post-treatment (e) and that of JFK125R post-treatment (k), we performed single-cell phosphoprotein analyses for STAT5-pY694 and ERK-pT202/Y204 (n = 3 independent experiments). For gel source data, see Supplementary Fig. 1. Source data

Supplementary information

Supplementary Figure 1

The original source images for all data obtained by electrophoretic separation with molecular weight markers and controls. Blots are labeled according to the corresponding Figure panel within the main or Extended Data Figures.

Reporting Summary

Supplementary Table 1| Identification of STAT5- and ERK-activating lesions in 1148 B-ALL patient samples.

1148 B-ALL samples were studied for STAT5-activating (CRLF2, IL3RA, IL7R, PDGFRA, PDGFRB, EPOR, ABL1, JAK1, JAK2, JAK3, TYK2 and STAT5A) and ERK-activating (NRAS, KRAS, PTPN11, NF1, RAF1 and FLT3) mutations. The samples originate from six different sources, including the St. Jude Children's Research Hospital Pecan Data Portal (https://pecan.stjude.cloud/proteinpaint/heatmap/HM,BALL), published studies (Nikolaev et al., 2014, Herold et al., 2017, Jerchel et al., 2018) as well as samples analyzed in the Müschen and Weinstock laboratories. Null hypothesis for this analysis was that STAT5- and ERK-pathway mutations occur independently of each other. The expected co-occurrence of the two mutations under the null hypothesis was 121 in 1148 B-ALL samples. The observed co-occurrence of mutations in the two pathways was 35, significantly lower than the expected (odds ratio 0.13, P=2.2e-16, Fisher’s exact test, two-sided).

Supplementary Table 2| Identification of STAT5- and ERK-activating lesions in 916 AML patient samples.

916 AML samples were studied for STAT5-activating (CALR, CBL, JAK1, JAK2, JAK3, KIT, MPL, PDGFRA, PDGFRB and STAT5A) and ERK-activating (RAF1, FLT3, KRAS, NF1, NRAS and PTPN11) mutations. The samples originate from the Catalogue of Somatic Mutations in Cancer (COSMIC; date of access: 2019-07-27). Null hypothesis for this analysis was that STAT5- and ERK-pathway mutations occur independently of each other. The expected co-occurrence of the two mutations under the null hypothesis was 34 in 916 AML samples. The observed co-occurrence of mutations in the two pathways was 24, significantly lower than the expected (odds ratio is 0.59, P=0.033, Fisher’s exact test, two-sided).

Supplementary Tables 3-11

This file contains a description of Supplementary Tables 1 and 2 (provided as separate Excel files), and Supplementary Tables 3-11.

Source data

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Chan, L.N., Murakami, M.A., Robinson, M.E. et al. Signalling input from divergent pathways subverts B cell transformation. Nature 583, 845–851 (2020). https://doi.org/10.1038/s41586-020-2513-4

Download citation

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing