Diagnostic peripheral blood (PB) and BM samples were from patients who went on to receive intensive chemotherapy within the AML14 and AML15 NCRI trials ( Supplementary Table 1 ). Cord blood was obtained from the University Hospital of Wales or NHS Blood and Transplant (NHSBT). All sample collections were in accordance with the 1964 Declaration of Helsinki. AML cell lines were purchased from ATCC and ECACC and were maintained as recommended. Cell line identity was confirmed using Cell Line Authentication Service (February 2021; Eurofins Scientific, Luxembourg). CD34⁺ hematopoietic stem/progenitor cells (HSPC) were isolated, cultured, and transduced with lentivirus as previously described (29). PKCϵ overexpression in HSPC and AML cell lines, employed a lentiviral expression construct, PKCϵ_GFP-Puro, expressing the PRKCE gene (NM0-05400.3) under the control of the EF1A promoter and EGFP and puro ^R genes linked by a T2A sequence driven by a CMV promoter. Control cells were transduced with vector lacking the PRKCE gene. Knockdown of PRKCE was achieved with shRNA lentiviral vectors (detailed in Supplemental Methods ). Control cells were transduced with a non-targeted construct. All vectors were purchased from VectorBuilder Inc (Chicago, USA). Transduced AML cell lines were selected with puromycin at 1µg/ml.

Myeloid colony forming assays were performed by limiting dilution in U bottomed 96-well plates (0.3 cells/well) using transduced (GFP⁺) HSPC enriched for cells highly expressing CD34 (CD34⁺) by FACS on day 3 of culture. Cells were cultured in Iscove’s Modified Dulbecco’s Medium (IMDM; Fisher Scientific, Loughborough, UK) supplemented with 5 ng/mL IL-3, SCF, G-CSF and GM-CSF and incubated at 37°C with 5% CO₂. Following 7 days of growth, individual colonies (> 50 cells) and clusters (> 5, < 50 cells) were scored.

Western blot analysis was conducted as previously described (30) using the antibodies described in Supplementary Table 2 . Prior to western blot and colony analysis of transduced HSPC (GFP⁺) cells were purified by FACS using AriaII (BD Biosciences). Band intensity on documented membranes were measured by densitometric analysis using ImageJ (Fiji; v. 2.0.0.71), unless otherwise stated. To do this, a region of interest (ROI) was constructed around a specific band. From this, a histogram of peak intensity was generated, and a baseline of background intensity was set from the area surrounding the band within the ROI. The area under the curve was then calculated to give an arbitrary intensity value. The band intensities of the protein for each sample were normalized to the band intensity of the loading control (GAPDH expression). PKCϵ expression in AML patient samples were subsequently converted into fg/1000 cells by comparing the relative band intensity to a recombinant PKCϵ standard of a known concentration.

PKCϵ (PRKCE) mRNA expression data from human (MicroArray; GSE42519 (31);) and murine (GSE14833 (32); and GSE6506 (33, 34)) HSPC were obtained from Bloodspot (35). For AML patient samples, PKCϵ mRNA expression was assessed using the TCGA, NEJM 2013 dataset (36) which was accessed using cBioPortal (37, 38). For survival analysis, only patients that received intensive cytarabine (Ara-C) and daunorubicin (DNR) treatment regimens were included. Patients were subsequently stratified according to PKCϵ expression using the upper (high PKCϵ) and lower (low PKCϵ) quartiles. PKCϵ (PRKCE) mRNA expression was also determined in AML14 and AML15 NCRI UK clinical trial samples by Affymetrix DNA microarray. The raw data from the Hu133A GeneChip^® was normalized using MAS5 or Robust Multi Array analysis. Two probe-sets were used (236459_at and 226101_at) as these had the highest correlation and the highest association with MAS5 present calls (20/22). The DEPMAP portal was used to analyze transcriptional correlations in 44 AML cell lines using the Expression 22Q1 Public RNASeq data set.

Flow cytometric data were acquired using an Accuri C6 cytometer (BD Biosciences, USA) and analyzed using FCS Express v6 (DeNovo, California, USA). For cell growth, viability, and drug sensitivity assays, cells were seeded at 2x10⁵/mL. After 48 hours, harvested cells were stained with a TOPRO-3 staining solution (RPMI supplemented with 100 mM HEPES and 50nM TOPRO-3 (T3605; Invitrogen, California, USA). For flow cytometric analysis, debris acquired during sample acquisition, defined as TOPRO-3 negative events with a FSC of <5x10⁴, were excluded and viability and viable cell counts were determined from within the viable cell population of TOPRO-3 negative events. Absolute counts/mL were determined from a 10 µL fixed volume acquisition. For the drug sensitivity assays, cell growth and viability were normalized to cells treated with the vehicle control. Drug sensitivity was compared by calculating the drug concentration which gave a half-maximal response (EC50). Master stocks of each agent were generated according to the manufacturer’s instructions and are outlined in Supplementary Table 3 . Once generated, all stocks were stored at -20°C in 100 µL aliquots for long-term storage unless otherwise stated. DNR accumulation was determined by measuring the arithmetic mean fluorescence (585/40nm) for cells treated with 100nM DNR, compared to vehicle controls. Immunophenotypic analysis was performed as previously described (39). P-glycoprotein (P-GP) expression on transduced AML cell lines was determined using the CD243-APC antibody (UIC2; BioLegend; Supplementary Table 2 ), while expression in patient samples has been described previously (40). Full details of the all the antibodies used for immunophenotypic studies are outlined in Supplementary Table 2 .

For cell cycle analysis, transduced AML cell lines were seeded at 2x10⁵ cells/mL and expanded for 48 hours before being harvested to 1 mL flow tubes and washed with 1 mL PBS. The washed cells were subsequently resuspended in PBS and fixed for 30 min on ice by adding 700 µL absolute ethanol. After fixation, the tubes were stored at -20°C overnight. The next day, the cells were centrifuged at 270xg and washed with 1 mL PBS before being resuspended in 50 µL staining buffer (PBS+0.5% (w/v) BSA+ 0.02% (w/v) sodium azide). The cells were then stained with 25 µL staining solution containing 40 µg/mL propidium iodide, and 0.1 mg/mL RNase (Sigma-Aldrich) diluted in PBS, for 30 min at 37°C. Samples were acquired within 20 min of this incubation using the Accuri™ C6 Plus flow cytometer. For analysis, debris and doublets were excluded, before the cell cycle status was resolved using the fluorescence intensity of propidium iodide (Sigma-Aldrich). Cell cycle analysis was performed using the Multicycle AV DNA analysis tool plug-in for FCS Express (DeNovo, Pasadena, USA).

Statistical significance was determined by the tests outlined in figure legends and were performed using GraphPad Prism v 9.0.

PKCϵ has demonstrated oncogenic properties and in solid cancers misexpression is associated with poor clinical outcomes and aggressive disease phenotypes (3–5, 41). To elucidate the frequency and pathophysiological attributes associated with PKCϵ upregulation in AML, we assessed PKCϵ expression in two independent patient sample cohorts. Initial data showed PKCϵ protein to be broadly expressed in AML ( Figure 1A ). We next carried out quantitative assessment in 70 AML patient samples. Compared to normal CD34⁺ BM blasts, PKCϵ protein was significantly overexpressed in 37% (26/70) of samples analyzed ( Figure 1B ). Patients overexpressing PKCϵ protein had a significantly lower complete remission (CR) rate compared to those with low PKCϵ protein expression (65% vs 84%; Figure 1C ) with no significant association between any other prognostically significant patient characteristic. A possible explanation for this is that PKCϵ overexpression is associated with chemotherapy resistance. To validate these findings, we used a clinically annotated mRNA dataset. We first established a significant positive correlation between PKCϵ mRNA and protein expression in AML patient samples ( Figure 1D ) which indicated that mRNA is a reasonable predictor of PKCϵ protein expression in this context but found no relationship to FAB subtype or LSC phenotype (data not shown). We subsequently evaluated the impact of PKCϵ mRNA expression in AML patients using the TCGA dataset (36). High PKCϵ mRNA expression was not significantly associated with any AML subset or known prognostic factors ( Supplementary Figure 1 ). While the impact on overall survival (OS) was not significant ( Figure 1E ; p=0.052), high PKCϵ expression was significantly associated with reduced disease-free survival (DFS; Figure 1F ). Although CR status was not available, reduced DFS is consistent with a treatment resistant phenotype. This data therefore supports the protein analysis in suggesting that high PKCϵ expression is associated with poor outcomes in AML. Together these data show that PKCϵ is frequently upregulated in AML and indicate that high PKCϵ expression is associated with poor patient outcomes, particularly in terms of response to initial chemotherapy.

Having determined that PKCϵ is frequently upregulated in AML, the mechanistic contribution of PKCϵ misexpression was evaluated. Initially we focused on whether overexpression contributes to the pathogenesis of AML by disrupting myeloid cell growth and differentiation. We first examined PKCϵ mRNA expression throughout normal myeloid development. In human and murine myeloid progenitors, PKCϵ was highly expressed in HSC and MPP, however expression was observed throughout hematopoiesis ( Supplementary Figure 2 ), indicating that PKCϵ may influence myeloid lineage commitment and differentiation. To support this, western blot analysis of human HSPC confirmed PKCϵ expression in all myeloid progenitor subsets ( Figure 2A ). The migratory differences observed are attributable to different phosphorylation states of PKCϵ (42). Further, when HSPC were cultured in conditions primarily supporting monocytic and granulocytic differentiation, PKCϵ expression increased with maturation ( Figure 2B ). Together these data indicate a potential developmental role for PKCϵ in myelopoiesis; however, we were unable to show any developmental consequences arising from PKCϵ knockdown in this context ( Supplementary Figures 3-6 ) though we did observe a significant reduction in the growth of granulocytic cells ( Supplementary Figure 4 ).

We next examined the functional consequences of PKCϵ overexpression. CD34⁺ HSPC stably overexpressing PKCϵ ( Supplementary Figures 7B-C ) were analyzed for changes in growth and differentiation by flow cytometry. Briefly, the myeloid lineages were resolved using the lineage discrimination markers CD13 and CD36, into monocytic (CD13^highCD36^high) and granulocytic (CD13^lowCD36^low) progenitor populations ( Supplementary Figure 8 ). Overall PKCϵ overexpression was associated with a 2.6-fold reduction in HSPC fold expansion over 10 days of culture ( Figure 2C ) and had a similar impact on colony formation ( Supplementary Figure 9 ). Subset analysis revealed that PKCϵ overexpression significantly reduced the growth of monocytic progenitors (1.6-fold), compared to the control cultures ( Figure 2D ). This was associated with significant upregulation of the maturation markers CD11b ( Figure 2E ) and CD14 ( Figure 2F ). In contrast, we observed no significant impact on granulocytic proliferation ( Supplementary Figure 10A ) or differentiation ( Supplementary Figures 10B, C ). Together, these data indicate that increased PKCϵ expression promotes monocyte differentiation and does not support an oncogenic role for PKCϵ upregulation in the pathogenesis of AML, which is characterized by a block in terminal differentiation.

Given that the data above are not supportive of an oncogenic role for PKCϵ overexpression in AML, we next examined the capacity of PKCϵ to promote chemoresistance in AML cells as this is a known consequence of PKCϵ overexpression in solid cancers (10, 41). To model this, we overexpressed PKCϵ in AML cell lines (U937 and HEL) which exhibited low and undetectable levels of endogenous PKCϵ protein, respectively. Ectopic PKCϵ expression was validated by western blot analysis ( Figure 3A ) and the resulting impact on cell proliferation and viability were assessed by flow cytometry. PKCϵ overexpression resulted in a 1.5-fold (U937) and 2.8-fold (HEL) reduction in growth compared to the control lines ( Figure 3B ). This was accompanied by a modest increase in the proportion of cells in the G2 phase of the cell cycle ( Supplementary Figure 11 ), suggesting that the reduced proliferation rate could be due to perturbed cell cycle progression. Although PKCϵ overexpression also resulted in a 10% reduction in cell viability ( Figure 3C ), the small effect size suggests this is unlikely to be biologically significant and may have arisen from the relative accumulation of non-viable cells due to the reduced proliferation rate of the PKCϵ cultures.

We next established the effect of PKCϵ overexpression on Ara-C and DNR chemosensitivity; two of the central agents used in AML treatment. Despite the reduced growth of these cell lines, PKCϵ overexpression significantly sensitized U937 and HEL cells to Ara-C (2.9-fold and 1.9-fold respectively; Figure 4A ), therefore inconsistent with the hypothesis that PKCϵ overexpression confers chemoresistance to this nucleotide analogue. In contrast however, PKCϵ overexpression increased the EC50 of DNR from 60nM to >100nM in both cell lines investigated ( Figure 4B ), demonstrating that PKCϵ overexpression can selectively confer resistance to DNR.

As PKCϵ overexpression in AML cell lines promoted opposing phenotypes to Ara-C and DNR, it was initially hypothesized that the mechanism of PKCϵ-mediated chemoresistance was specific to the mode of action of DNR. The cytotoxicity of DNR is, in part, mediated by oxidative stress (43). Since PKCϵ has been reported to promote antioxidant capacity in AML (14) we assessed whether the PKCϵ overexpression lines showed resistance to other agents which induce oxidative stress (glucose oxidase, antimycin A, and arsenic trioxide). We found no evidence of resistance to any of these agents (rather sensitization was observed; Supplementary Figure 12 ) indicating that enhanced antioxidant capacity is unlikely to explain DNR resistance.

We next investigated whether PKCϵ influenced intracellular DNR concentrations. DNR has intrinsic fluorescence enabling direct measurement of intracellular drug accumulation by flow cytometry. Following 2 hours of DNR treatment, the PKCϵ overexpression cell lines showed a 1.8-fold reduction in intracellular DNR compared to the control lines ( Figure 5A ) with similar data observed at the timepoint used for cytotoxicity assessment (2 days of treatment; Supplementary Figure 13A ). Reduced DNR accumulation can arise from decreased drug uptake or increased expulsion via efflux pumps. DNR enters the cell by passive diffusion and so is unlikely to explain the observed reduction in intracellular DNR. Instead, it was hypothesized that the reduced DNR accumulation was due to increased drug efflux. P-GP (aka ABCB1/MDR1/CD243) is the best characterized efflux pump and is associated with reduced OS and CR induction in AML (44, 45). P-GP has several substrates including DNR (46), but does not transport Ara-C (47), potentially explaining the differential phenotypes exhibited in response to these chemotherapeutic agents.

To determine whether DNR resistance in the PKCϵ overexpression cell lines could be explained by P-GP-mediated drug efflux, we first evaluated P-GP expression by flow cytometry ( Figures 5B, C ). PKCϵ overexpression led to a >3.0-fold increase in P-GP expression compared to the control cell lines ( Figure 5D ), suggesting that PKCϵ-mediated DNR resistance is associated with P-GP upregulation. To establish a causal relationship between PKCϵ and P-GP, inhibition assays were conducted using the selective P-GP inhibitor Zosuquidar hydrochloride [ZSQ (48)]. In the PKCϵ overexpression cell lines, ZSQ increased DNR accumulation to levels equivalent to the control cell lines ( Figure 6A ), whilst there was no significant effect on DNR accumulation in the control lines. Furthermore, ZSQ treatment overcame PKCϵ-mediated resistance, restoring DNR sensitivity to that of the control lines ( Figure 6B ). Importantly, ZSQ treatment alone did not significantly impact the growth, viability, or autofluorescence of any of the cell lines investigated ( Supplementary Figures 13B, C ). Together these data demonstrate that P-GP drug efflux fully accounts for the PKCϵ-mediated DNR resistance observed.

Overexpression systems tend to give rise to much higher levels of protein expression than is observed in physiological or diseased states. Therefore, the relationship between high PKCϵ expression and P-GP expression was also assessed in the TCGA 2013 dataset of AML patient samples. No significant relationship between P-GP and PKCϵ mRNA was observed ( Supplementary Figure 14 ), suggesting that a transcriptional mechanism of P-GP regulation by PKCϵ may be unlikely to fully account for our observations. A poor correlation between P-GP mRNA and protein expression has previously been reported in myeloid leukemia cells (49) and interrogation of the DEPMAP portal for all 44 AML cell lines also showed no positive correlation (Pearson = -0.303). This prompted us to examine the relationship between PKCϵ and P-GP protein expression in AML14/15 patient samples. Here we found the proportion of patients with above median P-GP expression was 2.8-fold higher in patients overexpressing PKCϵ than in those with low PKCϵ protein levels ( Figure 6C ). Although a larger patient cohort is required to determine the statistical significance of this observation, this does support the cell line analysis in indicating a post-translational link between PKCϵ overexpression and P-GP expression in AML.

P-GP activity has also been reported to be post-translationally regulated. P-GP contains several PKC consensus phosphorylation sites and despite conflicting evidence, phosphorylation of P-GP by PKC isoforms has been associated with P-GP activation and drug efflux (21, 23). To determine whether PKCϵ overexpression might confer DNR resistance through promoting P-GP activity, the impact of PMA, a potent PKC agonist, was assessed. PMA treatment of KG-1 cells (which endogenously express P-GP) significantly reduced DNR accumulation compared to DNR treatment alone ( Supplementary Figure 15A ). Furthermore, this reduction was antagonized by the P-GP inhibitor, ZSQ ( Supplementary Figure 15A ). In contrast, PMA treatment showed no impact on DNR accumulation in the control or PKCϵ overexpression cell lines ( Supplementary Figures 15B–E ) suggesting that, in these lines, activation of endogenous levels of P-GP are insufficient to account for increased pump activity which is more likely attributable to the increased expression of P-GP in these cells ( Figures 5B-D ), though the fact that PMA did not promote pump activity in the overexpressing lines suggests that PKCϵ is also optimally stimulating P-GP in these cells.