Sensitivity to PI3K and AKT inhibitors is mediated by divergent molecular mechanisms in subtypes of DLBCL

Diffuse large B-cell lymphoma (DLBCL) is the most common lymphoma subtype, representing roughly 30% to 40% of all cases.¹  DLBCL is a heterogeneous tumor entity,²  but according to gene expression profiling (GEP), 2 major molecular subtypes termed activated B-cell-like (ABC) and germinal center B-cell-like (GCB) DLBCL can be distinguished.³  ABC and GCB DLBCLs are addicted to different signaling pathways, rendering them sensitive or resistant to specific pathway inhibitors.⁴  Especially as relapsed DLBCL patients experience adverse survival, novel therapeutic strategies are warranted.⁵ 

Previous work has shown activation of the oncogenic phosphatidylinositol-3-kinase (PI3K)/protein kinase B (AKT) pathway in a significant number of primary DLBCL samples using immunohistochemistry.^6,7  This dependency is detectable in different molecular DLBCL subtypes.^8-10  Activated PI3K isoforms phosphorylate phosphatidylinositol-4,-5-bisphosphate to phosphatidyl-inositol-3,-4,-5-trisphosphate (PIP³), which activates AKT, whereas the tumor suppressor PTEN as major negative regulator of PI3K/AKT dephosphorylates PIP³.^11,12  In GCB DLBCLs, PTEN loss seems to be the predominant mechanism of PI3K/AKT activation, whereas in other DLBCL subtypes, activation is associated with constitutive B-cell receptor signaling.^8-10  Prior studies suggested that PI3K inhibition alone or in combination with other compounds might be a promising therapeutic strategy for DLBCL patients. However, in these analyses, rather unspecific compounds were used.^9,13,14  The clinical usefulness of PI3K inhibition was called into question based on a small clinical study investigating the efficacy of the PI3K δ (PI3Kδ) inhibitor idelalisib as none of the treated DLBCL patients responded.¹⁵  Using a pharmacologic screen, we were able to unravel, in this study, that different molecular subtypes are highly sensitive to specific PI3K/AKT inhibitors, indicating that this pathway represents a promising target for DLBCL patients.

The experiments were performed as described.^16-18  Protocols are available in the supplemental Materials and methods, available on the Blood Web site. Sequences of short hairpin RNAs (shRNAs) are summarized in supplemental Table 1.

Protocols are available in the supplemental Materials and methods. Sensitivity of a cell line to inhibitor treatment was defined as a 50% inhibitory concentration (IC₅₀) < 1 µM.

Cell viability following compound treatment was determined after 5 days using the Cell Titer Glo assay (Promega, Madison, WI) as previously described.^19,20  Previous studies investigating the selectivity profile of AZD8835 and AZD5363 have indicated that AZD8835 is a selective PI3K α/δ (PI3Kα/δ) and AZD5363 is a selective AKT inhibitor.^21-23 

Protocols are available in the supplemental Materials and methods.

GEP was performed after AZD8835²³  or AZD5363²²  treatment as previously described.¹⁸  Protocols are available in the supplemental Materials and methods.

Quantitative PCR was performed as described using predesigned assays (Applied Biosystems, Carlsbad, CA).¹⁷ 

Western blotting was performed as described previously.²⁴  The antibodies used are summarized in supplemental Table 2.

Protocols are available in the supplemental Materials and methods.

Cells were treated for 6 and 24 hours with AZD8835 or dimethyl sulfoxide (DMSO). Cytosolic and nuclear fractions were prepared using the Subcellular Protein Fractionation Kit for Cultured Cells (Thermo Fisher Scientific, Waltham, MA).

To identify novel targets for the treatment of DLBCL, we performed a pharmacologic screen utilizing compounds inhibiting pathways involved in proliferation, survival, and apoptosis (Figure 1A). To also identify compounds that are effective only in specific DLBCL subtypes, we screened 4 GCB (K422, HT, BJAB, and OCI-Ly2) and 4 ABC DLBCL (OCI-Ly3, U2932, TMD8, and HBL-1) models and determined viability after inhibitor treatment. As positive controls, we used the chemotherapeutic agent doxorubicin that was toxic to all 8 cell lines and the Bruton’s tyrosine kinase (BTK) inhibitor ibrutinib that induced cytotoxicity in ABC DLBCLs without CARD11 mutations, as previously described.²⁵  The vast majority of compounds had only little or no effect on the cell lines, suggesting that the affected pathways might not be crucial for survival. In contrast, ABC DLBCL models were highly sensitive to the PI3Kα/δ inhibitor AZD8835, whereas viability of GCB DLBCL lines was significantly reduced following AKT inhibition using AZD5363 (Figure 1A).^22,23  In contrast, only TMD8 cells were highly sensitive (IC₅₀ = 0.25 µM), and HT cells (IC₅₀ = 0.86 µM) were moderately sensitive to the PI3Kδ inhibitor idelalisib.

To determine if AZD8835 and AZD5363 indeed represent promising compounds for the treatment of molecular DLBCL subtypes, we expanded our analyses to 15 DLBCL models. Molecular characteristics for these 15 cell lines are summarized in Table 1. First, we determined the expression status of the PI3Kα and PI3Kδ isoforms, the phosphorylation status of AKT as marker of pathway activation, as well as expression of PTEN. PI3Kα and PI3Kδ expression was found virtually in all models, although expression levels varied (Figure 1B; supplemental Figure 1A). Activation of AKT was detectable in most of the cell lines, whereas PTEN was lost in the majority of GCB DLBCLs (Figure 1C; supplemental Figure 1B).

Next, we treated these lines with AZD8835 and AZD5363. Four of 5 ABC DLBCL, but only 1 GCB DLBCL line, were sensitive to AZD8835 (P = .017; Figure 1D). In contrast, AZD5363 reduced viability in 1 ABC DLBCL model, but in 8 of 10 GCB DLBCLs (P = .047; Figure 1D). As mainly GCB DLBCLs were sensitive to AZD5363, sensitivity to AKT inhibition was correlated to the PTEN expression status. Seven of the 9 sensitive lines were PTEN-deficient, whereas all of the insensitive lines expressed PTEN, suggesting that predominantly PTEN-deficient DLBCLs respond to AKT inhibition (P = .0056; Figure 1E). Interestingly, 1 of the PTEN-deficient models was the ABC DLBCL line U2932, suggesting that the PTEN expression status overrides the molecular subtype with respect to AKT inhibitor sensitivity. In summary, these results indicate that different molecular DLBCL subtypes are addicted to different branches of the PI3K/AKT pathway.

To gain insights into the mechanisms underlying the AZD5363-induced toxicity, cell proliferation and the rate of apoptosis were measured in 2 sensitive (K422 and BJAB) and 1 resistant (HBL-1) DLBCL model. AZD5363 induced apoptosis measured by Annexin-V/propidium iodide (PI) staining in BJAB but not in K422 or HBL-1 cells (Figure 2A). In contrast, proliferation determined by CFSE staining was markedly reduced in sensitive K422 and BJAB cells, whereas resistant HBL-1 cells were unaffected (Figure 2B).

We next determined if AZD5363 sensitivity of PTEN-deficient DLBCLs translates into an in vivo setting. To this end, we created xenograft mouse models of the 2 PTEN-deficient GCB DLBCL lines K422 and WSUDLCL2. Treatment of both models with AZD5363 resulted in significantly reduced tumor growth compared with vehicle-treated mice (P = .029 for AZD5363 vs vehicle on day 18 in K422; P = .036 for AZD5363 vs vehicle on day 11 in WSUDLCL2; Figure 2C).

To analyze if AZD5363 activity translates into a setting of relapsed/refractory DLBCL patients, we used a cell line (UPF4D) and a corresponding patient-derived xenograft (PDX) mouse model (WEHA) from a refractory GCB DLBCL patient.²⁶  UPF4D cells showed high levels of phospho-AKT (p-AKT) and complete absence of PTEN (Figure 2D). Subsequent in vitro treatment of these cells showed complete resistance to AZD8835 but high sensitivity to AZD5363 with strong induction of apoptosis (Figure 2E-F), further suggesting that AKT activation promotes tumor growth in PTEN-deficient DLBCLs. Application of AZD5363 to the corresponding PDX mouse model WEHA resulted in significantly decreased tumor growth compared with vehicle-treated mice (P = 7.86 × 10⁻⁷ for AZD5363 vs vehicle on day 11; Figure 2G). The effect was even stronger than in the cell line xenograft models, indicating that AZD5363 activity translates into a relapsed/refractory DLBCL patient setting.

Next, we investigated the molecular mechanisms of AZD5363 activity in different DLBCL models. As an ATP-competitive inhibitor that promotes AKT membrane localization and subsequent AKT phosphorylation, AZD5363 treatment of 6 and 24 hours strongly increased p-AKT levels in the PTEN-deficient lines BJAB, K422, and U2932 (Figure 3A).^22,27  Phosphorylation of the AKT target PRAS40, an mTOR inhibitor in its unphosphorylated state, was completely lost, indicating inhibition of mTOR by AZD5363. These results were confirmed in UPF4D cells (Figure 3B).

To elucidate which biologic processes are affected by AZD5363 treatment, GEP was performed after 6, 12, 18, and 24 hours of AZD5363 incubation in 2 PTEN-deficient and AZD5363-sensitive cell lines (K422 and U2932). A common gene expression signature was created consisting of genes that were consistently down- or upregulated across all time points in both lines (Figure 3C; supplemental Figure 2; supplemental Table 3). This analysis revealed 96 genes that were significantly downregulated (P ≤ 2.5 × 10⁻⁴; false discovery rate [FDR] = .006) and 97 genes that were significantly upregulated (P ≤ 2.5 × 10⁻⁴; FDR = .012). To analyze in an unbiased fashion which biological processes are affected by AZD5363, we performed a gene set enrichment analysis using a database of 16.963 gene expression signatures.^28-32  This approach revealed that 14 previously described MYC target gene signatures were significantly enriched with downregulated genes following AZD5363 treatment, suggesting that AKT inhibition downregulates the transcription factor MYC (Figure 3D; supplemental Tables 4 and 5). In addition, previously described gene sets correlated with metabolism were enriched with downregulated genes (supplemental Table 4). In contrast, only 1 of 11 previously described PI3K/AKT gene signatures was significantly enriched with downregulated genes and only at moderate levels (supplemental Table 6). This signature was created in the PTEN-deficient GCB DLBCL cell line HT following PTEN reexpression, suggesting that PTEN target genes at least partially overlap with genes controlled by AKT signaling.⁹ 

To verify the GEP results, we determined MYC expression following AZD5363 treatment on messenger RNA (mRNA) and protein levels. MYC mRNA expression was not downregulated by AZD5363 (log₂(ratio) = .0682) as shown by our GEP data. In contrast, MYC protein levels were significantly reduced after 6 and 24 hours of AZD5363 treatment in sensitive BJAB, K422, and U2932 cells (Figure 3E). To determine to what extent MYC downregulation contributes to the AZD5363-induced cytotoxicity, we performed an MYC rescue experiment. By transducing either an MYC complementary DNA (cDNA) or an empty vector into the AZD5363-sensitive models BJAB and U2932, we could detect a substantial MYC-induced rescue after AZD5363 treatment, indicating that the observed MYC downregulation contributes at least partially to its toxic effects (Figure 3F). To investigate whether this rescue effect is specific to AZD5363, we determined viability of these MYC-expressing cells after doxorubicin treatment. No resistance against doxorubicin was conferred by MYC cDNA expression, indicating the specificity of our findings (Figure 3F).

In summary, our in vitro and in vivo data indicate that AKT inhibition might represent a promising therapeutic approach in PTEN-deficient DLBCLs independent of their molecular subtype. The cytotoxic phenotype seems to be at least partially caused by downregulation of MYC.

Our pharmacologic screen identified the PI3Kα/δ inhibitor AZD8835 to be highly active predominantly in ABC DLBCLs (Figure 1A,D). In contrast, the PI3Kδ-specific inhibitor idelalisib was highly effective only in 1 DLBCL model, consistent with limited efficacy in DLBCLs in a small clinical trial.¹⁵  This finding led us to investigate if inhibition of both PI3Kα and PI3Kδ is required to induce cytotoxicity in ABC DLBCLs. To this end, PI3Kα was knocked down using a specific shRNA and cells were treated subsequently with either idelalisib or DMSO. Knockdown of PI3Kα in combination with DMSO was not toxic to HBL-1 and OCI-Ly10 cells, whereas addition of idelalisib showed strong cytotoxicity in both lines (Figure 4A). AZD8835-insensitive BJAB cells as a negative control were unaffected by PI3Kα knockdown and idelalisib treatment, confirming the AZD8835 viability assay (Figures 1D and 4A). To validate the specificity of our approach, HBL-1, OCI-Ly10, and BJAB cells carrying either a nontoxic MSMO1 shRNA or a toxic MYC shRNA were treated with either DMSO or idelalisib. As expected, these cells were completely unaffected by idelalisib (supplemental Figure 3). In summary, our results implicate that simultaneous inhibition of both PI3Kα and PI3Kδ is required to induce cytotoxicity in ABC DLBCLs.

To identify the mechanisms responsible for AZD8835-mediated cytotoxicity, induction of apoptosis was analyzed using Annexin-V/PI staining in 2 sensitive (HBL-1 and TMD8) and 1 insensitive (BJAB) model. Predominantly in HBL-1 and less pronounced in TMD8 cells, a significant increase in apoptotic cells was observed, whereas no apoptosis was detectable in BJAB cells (Figure 4B). Analyses of proliferation after AZD8835 treatment using CFSE staining showed strongly decreased proliferation in HBL-1 and TMD8 but not in BJAB cells (Figure 4C). These results suggest that PI3Kα/δ inhibition results in induction of apoptosis and inhibition of proliferation in ABC DLBCLs.

To analyze in vivo activity of AZD8835, we created xenograft mouse models of the 2 ABC DLBCL cell lines OCI-Ly10 and TMD8. AZD8835 can be administered using 2 schedules, 25 mg/kg twice a day continuously and 100 mg/kg twice a day on day 1/day 4 that have been shown to be equally effective.²³  Both schedules significantly reduced tumor growth in our models (P = .0091 for AZD8835 vs vehicle on day 22 in TMD8 [25 mg/kg twice a day continuously]; P = .0003 for AZD8835 vs vehicle on day 11 in OCI-Ly10 [100 mg/kg twice a day on day 1/day 4]; Figure 4D). To verify these in vivo data in a setting of relapsed/refractory DLBCL patients, we used the PDX model KTC, which was derived from a treatment-refractory non-GCB DLBCL patient.²⁶  AZD8835 significantly decreased tumor growth in the KTC model (P = 5.36 × 10⁻⁵ for AZD8835 vs vehicle on day 15 [100 mg/kg twice a day on day 1/day 4]; Figure 4E), suggesting that PI3Kα/δ inhibition is effective in ABC DLBCLs in vivo.

To investigate molecular effects of PI3Kα/δ inhibition in ABC DLBCLs, we analyzed sensitive U2932 and HBL-1 cells after 6 and 24 hours of AZD8835 treatment. Inhibition of PI3Kα/δ induced a strong p-AKT decrease at both time points. PRAS40 phosphorylation was decreased as well, but to a lesser degree compared with AZD5363-treated cells (Figures 3A and 5A).

To investigate which signaling cascades are affected by AZD8835 treatment in ABC DLBCLs, GEP was performed after 6, 12, 18, and 24 hours of PI3Kα/δ inhibition in sensitive U2932, HBL-1, and OCI-Ly10 cells. A common gene expression signature from these lines was created revealing 96 genes that were significantly downregulated (P ≤ 2.5 × 10⁻⁴; FDR = .009) and 94 genes significantly upregulated (P ≤ 2.5 × 10⁻⁴; FDR = .009) after AZD8835 treatment (Figure 5B; supplemental Figure 4; supplemental Table 7). Interestingly, several known NF-κB target genes like JUNB, NFKBIA, NFKBIE, and TNF were detected among the top downregulated genes (Figure 5B), suggesting that PI3Kα/δ inhibition downregulates NF-κB signaling. To verify NF-κB inhibition after AZD8835 treatment, we performed an unbiased gene set enrichment analysis³²  using our signatures database.^28-31  Indeed, 12 previously described NF-κB target gene signatures were detected among the top downregulated gene sets, suggesting a crosstalk between the PI3K/AKT and the NF-κB pathway in ABC DLBCLs (Figure 5C; supplemental Tables 8 and 9). Interestingly, only 1 of the previously described PI3K/AKT gene expression signatures was moderately enriched with downregulated genes (supplemental Table 10). We also detected several MYC gene expression signatures to be significantly enriched with downregulated genes, indicating that PI3Kα/δ inhibition downregulates MYC signaling (supplemental Table 11). MYC protein downregulation following AZD8835 treatment was confirmed in 2 sensitive ABC DLBCL lines (U2932 and HBL-1; supplemental Figure 5A), whereas MYC knockdown was not detectable in resistant GCB DLBCL models potentially explaining their insensitivity to PI3Kα/δ inhibition (supplemental Figure 5B).

In order to confirm the GEP data with respect to NF-κB signaling, NF-κB target gene expression levels were determined by quantitative PCR. For all target genes, a significant downregulation after 24 hours of AZD8835 treatment was detectable in sensitive HBL-1, OCI-Ly10, and TMD8 but not in resistant OCI-Ly3 and BJAB cells. The level of depletion was comparable to effects caused by the positive control inhibitor ibrutinib (Figure 5D). Western blotting in these 5 lines confirmed the knockdown for the NF-κB targets BCL-XL, IRF4, and IκB-ζ on protein levels in sensitive (HBL-1, OCI-Ly10, and TMD8) but not in insensitive models (OCI-Ly3 and BJAB), validating that PI3Kα/δ inhibition downregulates NF-κB signaling in AZD8835-sensitive ABC DLBCLs (Figure 5E).

Finally, to investigate NF-κB activity at the nuclear level, cytosolic and nuclear extracts from HBL-1, TMD8, U2932, OCI-Ly3, and BJAB cells were prepared following AZD8835 treatment of 6 and 24 hours. PI3Kα/δ inhibition significantly decreased nuclear expression levels of the NF-κB subunits RelA and p50 in sensitive (HBL-1, TMD8, and U2932) but not in insensitive models (OCI-Ly3 and BJAB), further confirming NF-κB pathway inhibition (Figure 5F).

An interesting finding of our inhibitor screen was the insensitivity of the ABC DLBCL and NF-κB-dependent model OCI-Ly3 to PI3Kα/δ inhibition. Analyses of NF-κB target genes already indicated a lack of subsequent NF-κB inhibition in OCI-Ly3 (Figure 5D-E). This finding was further confirmed by analyzing p-IκB-α levels as marker of NF-κB activation. Whereas p-IκB-α levels markedly decreased in HBL-1 and TMD8 cells following AZD8835 treatment, no change was observable in OCI-Ly3 cells. In contrast, the IκB kinase inhibitor AFN700³³  significantly decreased p-IκB-α levels in all 3 models (Figure 5G), suggesting that AZD8835 resistance in OCI-Ly3 cells could be caused by lack of subsequent NF-κB inhibition.

OCI-Ly3 cells are characterized by the activating CARD11^L244P mutation leading to constitutive NF-κB signaling and ibrutinib resistance.^4,25,34  To assess if this mutation causes resistance to PI3Kα/δ inhibition, we transduced HBL-1 and TMD8 cells with either a CARD11^L244P cDNA or an empty vector (Figure 5H; supplemental Figure 6A). Subsequently, we treated these cells with AZD8835 or ibrutinib as a control inhibitor. Interestingly, expression of the CARD11^L244P mutation caused complete resistance to AZD8835 and ibrutinib in both models, suggesting that activating CARD11 mutations can induce PI3Kα/δ inhibitor resistance (Figure 5H). Because we also detected an MYC downregulation after AZD8835 treatment (supplemental Table 11; supplemental Figure 5A), we assessed the contribution of MYC knockdown to the AZD8835-induced cytotoxicity. To this end, we transduced HBL-1 and TMD8 cells with either an MYC cDNA or an empty vector and treated these cells with AZD8835 (Figure 5I; supplemental Figure 6B). In HBL-1 cells, no rescue was observed, whereas only a moderate partial rescue was detectable in TMD8 cells (Figure 5I). These results imply that the cytotoxicity induced by PI3Kα/δ inhibition is predominantly caused by inhibition of NF-κB signaling rather than MYC downregulation.

Sensitivity to PI3Kα/δ inhibition seems to be predominantly caused by downregulation of NF-κB signaling. We treated 4 ABC DLBCL cell lines (OCI-Ly10, TMD8, HBL-1, and OCI-Ly3) with a combination of different AZD8835 and ibrutinib concentrations, to determine if simultaneous PI3Kα/δ and BTK inhibition might have synergistic effects on cell viability. Whereas the single agents had only modest effects, their combination resulted in synergistic effects in OCI-Ly10, TMD8, and HBL-1 cells. In contrast, OCI-Ly3 cells were not sensitive to either single drug or the combination (Figure 6A).

These observations were confirmed in vivo using xenograft mouse models of OCI-Ly10, TMD8, and the KTC PDX model. Single ibrutinib treatment only slightly affected tumor growth in all models consistent with previous reports.³⁵  In OCI-Ly10 and TMD8, AZD8835 alone resulted in growth reduction, while administration of both inhibitors almost completely abolished tumor growth (P = .002 for AZD8835 vs AZD8835/ibrutinib on day 11 in OCI-Ly10; P = .001 for AZD8835 vs AZD8835/ibrutinib on day 22 in TMD8; Figure 6B). For KTC, single PI3Kα/δ as well as combined inhibitor treatment resulted in complete tumor depletion without visible synergistic effects. For this model, the synergism was only visible when leaving pretreated mice untreated for several days. We observed that AZD8835 pretreated tumors reengrafted significantly faster compared with AZD8835/ibrutinib pretreated tumors (P = 1.22 × 10⁻⁵ for AZD8835 vs AZD8835/ibrutinib on day 18; Figure 6B). In summary, these results demonstrate that combined PI3Kα/δ and BTK inhibition seems to be highly effective in ABC DLBCL models in vitro and in vivo.

This study identified novel survival addictions to 2 different branches of the PI3K/AKT signaling pathway. AKT signaling is crucial for PTEN-deficient DLBCLs, whereas PI3Kα/δ−induced activation of NF-κB is critical for ABC DLBCLs. Each pathway is activated in an almost exclusive fashion in these subtypes, conferring preferential sensitivity to AKT and PI3Kα/δ inhibitors. Our results were obtained using various in vitro and in vivo models including PDX models that are ideal for drug development due to their strong resemblance to human DLBCL.^36,37  We found that ABC DLBCLs without CARD11 mutations are addicted to PI3Kα/δ signaling, explaining most likely why the PI3Kδ-specific inhibitor idelalisib did not achieve responses in a small trial conducted in relapsed/refractory DLBCL patients.¹⁵Our work using a specific shRNA targeting PI3Kα and the PI3Kδ inhibitor idelalisib highlights that resistance to PI3Kδ inhibition can be overcome by simultaneous PI3Kα inhibition. PI3Kα/δ inhibition leads to depletion of NF-κB signaling, indicating a crosstalk between both pathways. A possible link between these signaling cascades might represent PDK1 that acts downstream of PI3K as suggested by previous work.⁸  Activating CARD11 mutations seem to cause resistance to PI3Kα/δ inhibition, as CARD11-mutated OCI-Ly3 cells were unaffected by PI3Kα/δ inhibition. Likewise, sensitive cells that exogenously express CARD11 mutations became resistant to AZD8835. In contrast, MYC aberrations present in some ABC DLBCLs did not influence sensitivity.

A combined approach of BTK and PI3Kα/δ inhibition using ibrutinib and AZD8835 was highly effective in vitro and in vivo. Ibrutinib as a single agent was tested in a recently completed phase 2 trial.⁴  However, this strategy was only moderately active in ABC DLBCLs, as only ∼40% of relapsed/refractory ABC DLBCL patients responded.⁴  Furthermore, these responses were usually only of short duration. Especially our in vivo PDX model data suggest that combined inhibition of BTK and PI3Kα/δ is substantially more active compared with ibrutinib alone, supporting the results of a previous high-throughput combination screen that showed effectiveness of ibrutinib and PI3K inhibitors in ABC DLBCL models.³⁸  Collectively, these data provide a therapeutic rationale to combine PI3Kα/δ and BTK inhibitors in future clinical trials.

In contrast to PI3Kα/δ dependency in ABC DLBCLs, PTEN-deficient DLBCLs seem to be addicted to AKT signaling, as the AKT-specific inhibitor AZD5363 induced apoptosis and inhibited cell proliferation predominantly in this DLBCL subgroup. MYC aberrations in our models had no impact on inhibitor sensitivity. Interestingly, 2 models with PTEN expression were also sensitive to AZD5363, suggesting that subsets of PTEN-positive DLBCLs might also benefit from AKT inhibition. However, the molecular features determining AKT sensitivity in PTEN-positive DLBCLs are unclear and should be addressed in future work. Sensitivity to AKT inhibition was previously shown in DLBCLs using different AKT inhibitors.^39,40  However, the predominant activity in a molecularly defined subgroup has not yet been described. PTEN deficiency is a hallmark of GCB DLBCLs. Accordingly, the vast majority of sensitive models derive from GCB DLBCL patients.⁹  However, 1 of the sensitive cell lines was the PTEN-deficient ABC DLBCL line U2932. This finding suggests a common dependency in ABC and GCB DLBCLs with PTEN loss and argues in favor of a patient selection in clinical trials based on the PTEN expression status rather than on the cell of origin. Inhibition of AKT significantly downregulated expression of MYC, mirroring effects of PTEN reexpression in PTEN-deficient DLBCLs and suggesting that PTEN and AKT target genes overlap at least partially.⁹  Interestingly, other previously identified PI3K/AKT signatures were not downregulated following AKT inhibition. This finding is surprising and could be related to the fact that these signatures were created in other settings or disease entities.

In summary, our work highlights that a thorough molecular characterization of DLBCLs is necessary to use novel therapeutic compounds in the most effective manner. Based on our data, PI3Kα/δ inhibitors such as AZD8835 should be tested in clinical studies in ABC DLBCLs without CARD11 mutations, whereas AKT inhibition using AZD5363 or other specific AKT inhibitors should be investigated in DLBCL patients with PTEN loss.

This work was supported by research grants from AstraZeneca, the Deutsche Krebshilfe, the Deutsche Forschungsgemeinschaft (DFG), the Else Kröner-Fresenius-Stiftung, the Brigitte und Dr. Konstanze Wegener-Stiftung, and the Swiss National Science Foundation (Sinergia grant) (G.L.). In addition, this work was supported by the DFG EXC 1003 Cells in Motion-Cluster of Excellence, Münster, Germany (G.L.). Support was provided by the Ministry of Health of the Czech Republic grant AZV 15-27757A and the Grant Agency of the Czech Republic grant GACR14-19590S (P.K.).

Contribution: T.E. designed research, performed experiments, analyzed data, and wrote the manuscript; P.K. and J.T.L. performed and analyzed experiments; M. Grau performed bioinformatic and biophysical analyses; P.V., J.M., D.T., K.H., U.M.P., M. Grondine, M.M., B.D., M.P., K.E., and D.S. performed and analyzed experiments; M.Z. performed bioinformatic and biophysical analyses; A.M.S., G.O., W.E.B., B.R.D., F.C., M.T., P.L., and S.T.B. analyzed data; and G.L. designed research, analyzed data, and wrote the manuscript.

Conflict-of-interest disclosure: J.T.L., K.H., U.M.P., M. Grondine, M.M., B.R.D., F.C., and S.T.B. are employees of AstraZeneca. G.L. received research funding from AstraZeneca. The remaining authors declare no competing financial interests.

Correspondence: Georg Lenz, Translational Oncology, University Hospital Münster, Albert-Schweitzer-Campus 1, 48149 Münster, Germany; e-mail: georg.lenz@ukmuenster.de.