PAK Kinase Inhibition Has Therapeutic Activity in Novel Preclinical Models of Adult T-Cell Leukemia/Lymphoma

Adult T-cell Leukemia/Lymphoma (ATLL) is a highly aggressive T-cell neoplasm caused by human T-cell lymphotropic virus (HTLV)-1 (1, 2). Patients diagnosed with the aggressive subtypes carry some of the worst prognosis of any of the non-Hodgkin lymphomas. The 5-year overall survival (OS) rate of patients with ATLL is reported to be only 14% (3). In accordance with the endemic regions of HTLV-1, ATLL is most frequently diagnosed in southwest Japan and the Caribbean islands. Most patients with ATLL diagnosed in North American (NA-ATLL) are of Caribbean descent. Compared with the Japanese patients (J-ATLL), NA-ATLLs tend to have a more aggressive clinical course, higher rates of chemo-refractory disease, and worse prognosis (4–6). Such differences in clinical behavior may be accounted for by distinct somatic mutation patterns between these two patient populations (7). Although there is no standard therapy for ATLL, the initial treatment often employs combination chemotherapy followed by allogeneic transplantation (1). Yet, the high rate of chemo-resistance disease often leads to treatment failure and dismal survival outcomes. Hence, an attractive strategy is to design mechanism-based treatment that could target novel pathways critical in maintaining ATLL phenotype in vivo.

The current model for ATLL pathogenesis depicts the initial infection of CD4 T cells by HTLV-1, followed by downregulation of the HTLV-1 Tax protein, and progressive accumulation of genetic and epigenetic changes in the infected T cells that ultimately lead to oncogenic transformation and disease progression (2). In an integrated genomics and transcriptomics study of a large cohort of 426 J-ATLL patients, many frequently altered genes encoding candidate driver mutations in ATLL transformation and progression were reported (8). These genes are involved in the TCR/NF-κB signaling, T-cell trafficking, and immune surveillance. Among the 35 genes frequently targeted by both somatic alterations and the HTLV-1 onco-protein Tax, p21-activated kinase (PAK) 2 is amplified in 24% of the cases (8). PAKs are a family of serine/threonine kinases composed of six isoforms (PAK1-6). These isoforms are separated into two groups. Group I PAKs (PAK1/2/3) are activated by signals emanating from the small GTPase CDC42 or RAC, whereas group II PAKs (PAK4/5/6) are only activated by CDC42 (9). In solid tumors, PAKs are frequently upregulated and/or hyper-activated either as the consequence of gene amplification (PAK1/4) or more so by changes in upstream regulators (8, 10). PAKs influence cancer cell proliferation and survival through a number of downstream signaling pathways, which often feed into the nuclear transcription program. Among these, the ones most relevant to ATLL biology are TCR/NF-κB, STAT5, ERK, β-catenin, and Raf/BAD (8). PAKs can also regulate cancer cell migration and metastasis through their cytoplasmic substrates involved in cytoskeletal remodeling, for example, PAK1/2 can modulate Rac GTP activity and lamellipodia formation (9).

In the ATLL microenvironment, migrating leukemic cells rely on lamellipodia formation at the leading edge to bind to and spread on stromal cells (11). Previous studies have shown that this process involves the ectopically expressed adhesion molecule CADM1/TSLC1 (11, 12), and is a key step in organ infiltration of ATLL cells in animal models (13). In patients with ATLL, CADM1 expression in combination with CD7 downregulation closely correlates with clonal expansion of leukemic cells (14). Functionally, trans-homophilic interactions of CADM1 activates PI3K (15) and NF-κB (16), whereas trans-heterophilic CADM1 interaction triggers recruitment of the RAC1 GEF, TIAM1, to the CADM1-cytoplasmic domain, leading to RAC1 activation and remodeling of actin-network and lamellipodia formation (11). Although RAC1 is known to be required for CADM1 function in ATLL, PAK kinases supporting CADM1-dependent stroma adhesion of ATLL cells have not been defined.

Because cancers with aberrantly activated PAKs are often addicted to their activity (9) and PAK signaling can drive acquired drug resistance to targeted therapies in several types of cancers (17–20), the PAK family kinases have been explored as therapeutic targets in a number of solid tumors and hematologic malignancies (18, 19, 21, 22). In this study, we evaluated the therapeutic value of PAK inhibition in ATLL preclinical models. We also examined the role of PAK1 and PAK2 in ATLL cell proliferation, survival, and stroma adhesion.

Additional methods and details are described in Supplementary Materials and Methods, available online at the CCR Web site.

Su9T01, ED40515(+), ED40515(−), ED41214(+), ED41214(−), ATL55T(+), ATL43T(+), and ATL43Tb(−) were J-ATLL cell lines. ATL13, ATL18, and ATL21 were long-term CD4⁺ T-cell cultures derived from NA-ATLL patients (Supplementary Fig. S1; ref. 7). ED40515(+), ED41214(+), ATL55T(+), ATL43T(+) cells, and the NA-ATLL cell lines were grown in the presence of 100 unit/mL human IL2. All J-ATLL cell lines, the five HTLV-1-negative human T cell lines (Jurkat, CEM, HuT78, MAC1, OCI-Ly13.2), and the human follicular dendritic cell (FDC) cell line, HK (23), were cultured in RPMI with 10% FBS. Two additional HTLV-1-negative T cell lines HH and FePD1 were maintained in IMDM and DMEM containing 20% FBS, respectively. All NA-ATLL cell lines and short-term cultures were maintained in IMDM containing 20% human serum. All medium were supplemented with 100 unit/mL penicillin and 100 μg/mL streptomycin. All cell cultures has been routinely tested for mycoplasma contamination.

PF-3758309 (PF, Selleckchem; Catalog No. S7094) is an ATP-competitive inhibitor that targets PAK1, PAK2, PAK3, and PAK4. IPA-3 (Tocris; Catalog No. 3622) is a selective non-ATP competitive PAK1 inhibitor that also binds PAK2 and PAK3. The IC₅₀ of PF-3758309 and IPA3 were determined using resazurin (R&D Systems; Catalog No. AR002)-based viability assays in which cells were exposed to various concentrations of the reagents for 48 hours. To evaluate the response to IPA-3, a structurally related inactive compound, PIR3.5 (Tocris; Catalog No. 4212), was also tested.

A total of 7 × 10³ HK cells/well were seeded in a 96-well plate and allowed to attach overnight before the assay. Parental ATLL cells or ATLL cells that were transfected with siRNA oligos 48 hours earlier were fluorescently labeled with the Calcein AM dye according to the manufacturer's instructions (Invitrogen; Catalog No. C3100MP). Labeled parental ATLL cells were then treated with either DMSO or PF at 2.5 μmol/L for 1 hour prior to exposure to HK. Subsequently, ATLL cells were added to the HK feeder-coated 96-well plate in triplicate at 7 × 10⁴ cells/well and allowed to adhere for 180, 150, 120, 90, 30, and 15 minutes. Non-attached cells were washed away four times with 1x PBS. Absorbance reading was performed at 480 to 520 nm with the Fluostar plate reader. Adherence of ATLL cells to HK feeder was calculated as the fraction A/B. A is the absorbance values after washing away unattached cells minus average absorbance values of HK cells alone after washing. B is the absorbance values of cells before washing minus average absorbance values of HK cells alone before washing.

A patient-derived xenograft (PDX) model was established from patient PBMCs (ATL18) in NSG (NOD/SCID/γC-/-) mice by serially passaging subcutaneous tumor mass in female NSG mice. In this model, distal dissemination is a constant feature that always involves lung, liver, and spleen whereas kidney and bone marrow are less frequently infiltrated. A cell line-derived xenograft (CDX) model was similarly developed using ATL43Tb(−) cells but metastasis in this model is largely restricted to the lung. PF treatment was initiated when tumors reached ∼100 mm (3). Animals were randomly assigned to one of the three groups to be treated by vehicle, 6 or 12 mg/kg/day of PF. PF was formulated in 2% carboxymethyl cellulose (24) and administered intraperitoneally. Throughout the experiment, overall health of the animals was closely monitored, body weight and tumor growth were frequently recorded. Tumor sizes were measured using an electronic caliper. Tumor volume (mm³) was calculated using the formula of (A × B²)/2, where A and B are the tumor length and width in mm, respectively. Tumor growth inhibition (TGI) was calculated as TGI (%) = (V_c − V_t)/(V_c − V₀) × 100, where V_c, V_t are the median tumor volume of control and treated groups at the end of the study and V₀ is the median tumor volume at the start of the study. All mice were maintained and treated in compliance with IACUC approved protocols at Albert Einstein College of Medicine.

Copy number variation in a cohort of 426 Japanese ATLL samples was studied using SNP array karyotyping (8). Among the 26 focally amplified region is a band on 3q29, encoding PAK2 (Supplementary Fig. S2A). In all, PAK2 amplification was detected in 24.4% of all ATLL cases with a preferential enrichment in the aggressive subtypes (acute/lymphomatous vs. chronic/smoldering: 27.5% vs. 16.0%; Benjamini–Hochberg correction, q = 0.09). There is a trend for increased PAK2 mRNA expression in 3q29 amplified cases (Supplementary Fig. S2B) Most importantly, PAK2 amplification was significantly associated with a trend for poor overall survival (P = 0.029, Fig. 1A).

Given PAK2 amplification in ATLL, we wanted to examine expression pattern of this kinase in primary samples. We first established mRNA expression profiles for all six PAK family members in this disease. In acute ATLL, a significant fraction of peripheral blood mononuclear cells (PBMC) are leukemic T cells. We therefore performed qRT-PCR assays for PAK1-6 on 6 PBMC samples from patients with ATLL and compared them to four healthy controls (Fig. 1B). PAK2 is expressed at the highest levels among all samples followed by PAK1. Ranking third, PAK4 mRNA level is approximately one log lower than PAK1 but still approximately one log higher than PAK3 and PAK6. PAK5 mRNA is undetectable in most patient samples. The same pattern is observed from the RNA-seq dataset of the Japanese ATLL cohort (Supplementary Fig. S2C).

To measure PAK mRNA expression in pure populations of ATLL cells, we then analyzed eight ATLL cell lines and compared the results to normal resting and activated CD4⁺ T cells, a pooled normal Treg sample, and four cell lines representing other T-cell non-Hodgkin lymphomas (Jurkat and CEM, acute lymphocytic leukemia; HuT78, cutaneous T-cell lymphoma; OCI-Ly13.2, peripheral T-cell lymphoma; Fig. 1C). The pattern for PAK isoforms among ATLL cell lines largely resembles that seen with patient PBMCs, although PAK3 and PAK6 mRNAs exhibit much more intragroup heterogeneity. Compared with normal resting CD4 T cells, ATLL cell lines under-express several PAK isoforms including PAK1, PAK2, PAK4, and PAK6. The same trend exists in another ATLL gene expression dataset (Supplementary Fig. S2D; ref. 25). Of note, when compared with activated CD4 T cells and other types of T-cell malignancies, PAK2 mRNA is also reduced in ATLL cell lines (Fig. 1C). In general, Tregs have the lowest levels of all PAKs among all tested samples except for PAK2, which is expressed at comparable levels in Tregs and ATLL cell lines. Although the cause of this apparent ATLL-associated PAK2 downregulation is currently unknown, the observation does suggest that the level of PAK2 may have become rate-limiting in this disease and thus gene amplification events are positively selected in aggressive ATLLs. This interpretation is in line with observations made on PAK1/2 proteins. Western blot analysis shows significant variability of PAK1 protein across all diagnostic categories of T-cell malignancies, whereas PAK2 protein exhibits less variability across the panel (Fig. 1D). It should be noted that the characteristic ATLL-associated reduction of PAK2 mRNA is not reflected at the protein level. We interpreted this result as a suggestion that posttranscriptional mechanisms such as protein stability may operate to compensate for the reduced PAK2 mRNA expression. In summary, our result indicates that PAK1/2/4 are the most abundantly expressed PAK isoforms in ATLL and that PAK2 is regulated differently from the other PAKs during ATLL development.

Given the PAK family expression patterns in ATLL, we selected two PAK inhibitors for pharmacologic PAK inhibition. PF-3758309 (PF) is the first PAK inhibitor to advance to clinical trial. It is an ATP-competitive inhibitor with documented activity against all PAK family members to various degree (26, 27). IPA-3 is an allosteric inhibitor that has specific activity against group I PAKs but spares group II PAKs (28). In all, seven J-ATLL cell lines, seven North American patient samples, and three NA-ATLL cell lines were tested (Fig. 2A–C; Supplementary Fig. S3). For PF, the 48 hours IC₅₀ values range from 1.8 to 13.4 μmol/L for the seven Japanese cell lines tested. IL2 requirement culture condition did not appear to impact PF sensitivity. For NA-ATLLs, the IC₅₀ values range from 1.2 to 14.4 μmol/L for the seven primary patient samples and 2.9 to 8.2 μmol/L for the three cell lines. Of note, ATL6a/6b/6c and ATL19a/19b are serial samples collected from chemo-refractory patients treated with high-dose chemotherapy in combination with antiviral agents. We also tested IPA-3 responsiveness in six Japanese cell lines and two North American ATLL samples. Relatively low IC₅₀s were observed (6.6–16.9 μmol/L), demonstrating in vitro efficacy of both pharmacologic inhibitors in both Japanese and North American ATLLs (Fig. 2C; Supplementary Fig. S3C and S3D). Interestingly, activated normal CD4 T cells are highly sensitive to PF (IC₅₀ = 2.5 μmol/L), suggesting that PAK kinases play an extremely important role in their proliferation/survival.

To explore the role of individual PAK isoforms in the observed PF activity, we examined changes in total PAK1/2 proteins as well as phospho PAK1/2/4 following 1 hour of PF treatment at 2.0 to 2.5 μmol/L, which are below the majority of 48 hours IC₅₀ values for the selected samples including six J-ATLL cell lines and four NA-ATLL samples (Fig. 2D). Both PAK1 and PAK2 were readily detected in all samples examined except ED40515(+), which has the lowest levels of both PAKs. Lower levels of total PAK1 was also noted for ATL55T(+) and Su9T01. We examined all four types of phosphorylated PAK1/PAK2 and found that all of them exhibited cell line to cell line variability; yet, there is a close correlation between phospho-Ser20-PAK2 and the ability of a cell line to form tight aggregates in suspension culture (not shown). Interestingly, PF treatment reduced PAK2 phosphorylation on two residues (phospho-Ser20 and phospho-Ser141) while enhancing the signal on the third (phospho-Thr402; Fig. 2D). Because binding of PF to the primed ATP-binding site in PAK is expected to lock the enzyme in this partially activated state, the observed changes in PAK2 is consistent with the model for its activation (9). In comparison, under the PF treatment conditions, phosphorylated PAK1 was either not altered (phospho-Ser144-PAK1) or often enhanced (phospho-Ser199- and phospho-Thr423-PAK1) but never reduced. In most samples tested, strong phospho-PAK4 signals were detected. Yet, PF only reduced phospho-PAK4 in two cell line, ED40515(−) and Su9T01, which are the most sensitive samples to PF (Fig. 2C). Two conclusions can be made from these observations. First, at 2.0 to 2.5 μmol/L, PF could effectively inhibit PAK2 kinase activity in all samples containing activated PAK2; whereas all seven tested samples contained activated PAK4, it was only reduced by PF in the two sensitive cell lines. Signals representing activated PAK1 were very weak and not reduced in these experiments. Second, when it comes to correlation with in vitro PF response, the best predictor of sensitivity appears to be low level expression of both PAK1 and PAK2 as well as phospho-PAK4 (Fig. 2D).

Because the resazurin-based in vitro viability test measures both cell proliferation and survival effects, we next evaluated these two cellular responses separately. Six J-ATLL and three NA-ATLL cell lines were treated with the indicated doses of PF and apoptosis was measured using Annexin V/PI staining (Fig. 3A; Supplementary Fig. S4). In all samples tested, more apoptotic cells were detected at 48 hours compared with 24 hours, and 10 μmol/L of PF triggered more pronounced apoptotic cell death than 2.0 μmol/L at 24 hours (ATL13 and ATL21; Supplementary Fig. S4). The two most sensitive Japanese cell lines [ED40515(+) and Su9T01] also exhibited the largest apoptotic fractions (40%–50%). Next, we examined cell cycle changes in PF-treated samples (Fig. 3B; Supplementary Figs. S5 and S6). The most significant phenotype was a prominent G₁–S transition block detected in all samples, which was accompanied by corresponding S phase reductions. Variable extent of G₂–M arrest was also observed. In three cell lines showing notable G₂–M accumulation [ATL55T(+), ED40515(+), and ATL43Tb(−) at 40 hours only], CYCLIN B1 and phospho-Ser10 H3 signals were increased thus confirming the activation of G₂ checkpoint. However, this is not the case in ED40515(−) or Su9T01 (Supplementary Fig. S5B).

To determine the molecular mechanisms of PF response, we analyzed three cell signaling molecules (STAT5 and ERK1/2) known to be regulated by PAKs and important for ATLL biology (9, 29, 30). In addition, c-MYC and three apoptosis regulators (MCL1, BAD, and BAX) were also examined. Of note, c-MYC and BAD have previously been shown to be the kinase substrates of PAK2 and PAK1/2, respectively (10). ATLL cells were treated with DMSO or PF for 2, 4, and 8 hours, and whole cell lysates were analyzed by Western blotting (Fig. 3C). The IL-2-independent cell line, ATL43Tb(−), did not contain activated/phosphorylated STAT5. In all three IL-2-dependent cell lines, PAK inhibition notably reduced PY-STAT5 as early as 2 hours into the treatment. Oncogenic c-Myc levels were reduced after PAK inhibition in all samples. Anti-apoptotic MCL1 was also markedly reduced whereas the level of pro-apoptotic BAX was elevated at early stage of the treatment. Because PAK1 can promote ERK1/2 activation independent of its kinase activity (31), we also examined phospho-ERK1/2. All three J-ATLL cell lines showed no or very little phospho-ERK1/2 signals at the baseline, but the two IL-2-dependent lines [ATL55T(+) and ATL43T(+)] upregulated this signal following PF treatment. The cell line ATL18 contained strong phospho-ERK1/2 at the baseline, which was markedly inhibited by PF at 4 and 8 hours, implying that ERK1/2 activation is controlled by different mechanisms in this sample. In the three Japanese cell lines, neither total nor phospho-Ser112-BAD was altered by PF treatment; yet, both were consistently reduced by PF in ATL18 starting from the 2-hour time point. Most of these early changes could still be detectable at 48 hours (Supplementary Fig. S7). The unexpected increase in phospho-BAD may reflect other kinases that can also act on the Ser112 residue (10). Overall, despite the variability in ERK1/2 and BAD, the observed molecular changes were consistent with and likely contributed to the apoptosis and cell-cycle inhibition phenotypes shown in Fig. 3A and B.

In monotypic suspension culture, primary ATLL cells often form large aggregates, which will dissociate when plated on a feeder layer of HK cells (a human follicular dendritic cell line; ref. 23), mimicking in vivo microenvironment (Fig. 4A). Adherence to HK also markedly improved cell viability with unfractionated patient PBMCs and freshly sorted CD4⁺ leukemic cells (Fig. 4B), suggesting that adhesion-induced cell signaling events regulate both cytoskeletal remodeling and cell death control programs. Because PAKs regulate cancer cell migration and lamellipodia formation by modulating RacGTP activity, and the CADM1–Tiam1–Rac complex has been shown to promote dissemination of ATLL cells in vivo (11, 15), we examined the impact of pharmacologic PAK inhibition on the ability of ATLL cells to adhere to HK (Fig. 4C). Using a Calcein AM-based cell adherence assay, we monitored HK adherence of seven ATLL cell lines over a period of 3 hours, comparing the results obtained with and without PF pretreatment. Among the seven cell lines tested, two were CADM1-negative [Supplementary Fig. S8A, Su9T01, ED40515(−)] and neither showed significant HK adherence in 3 hours (Fig. 4C). All five CADM1-expressing cell lines including the ATL18, demonstrated time-dependent HK adherence with variable kinetics. Most importantly, PF pretreatment significantly reduced HK adherence in all five CADM1-expressing cell lines but did not alter the low, basal HK binding activity of the CADM1-negative line, Su9T01. Flow cytometry analysis revealed that defective binding to HK was not due to PF-induced CADM1 downregulation (Supplementary Fig. S8B). These observations support the notion that PAKs positively regulate stromal adherence by CADM1-expressing ATLL cells (Supplementary Fig. S8C).

To examine involvement of PAK isoform(s) in the HK adherence process, we analyzed PAK1 and PAK2 protein during the first 150 minutes of ATLL-HK contact in three adhesion-competent cell lines (Fig. 4D). Activated/phospho-Ser144-PAK1 was only detected in ATL43T(+) cells, and was modestly reduced by PF pretreatment. Phospho-Ser141-PAK2 was detected in both ATL43T(+) and ATL55T(+) cells and this signal was nearly completely inhibited by PF pretreatment. All cell lines expressed phospho-Ser20-PAK2. This signal as well as total PAK2 were reduced after at 30 and 150 minutes following HK exposure in ATL55T(+) and ED40515(+) cells. In ATL43T(+), however, these PAK2 signals appear to be slightly increased after 30 minutes of HK interaction. In all three cell lines, PF pretreatment decreased phospho-Ser20-PAK2 at the baseline (0 minute) and late stage of HK exposure (150 minutes). In summary, changes in total and/or phospho-Ser20-PAK2 but not any of the PAK1 signals correlated with PF-induced adhesion defect.

To directly evaluate the function of individual PAK isoforms, we turned to siRNA-based knockdown (KD) approach. Both PAK1 and PAK2 were substantially and specifically reduced at 48 hours following siRNA transfection in ATL55T(+) (Fig. 4E). Interestingly, we observed cross-talk between PAK1 and PAK2 because PAK1 protein was increased in PAK2 KD cells. We then subjected these cells to 1 hour pretreatment by PF before analyzing their capability to adhere to the HK feeder. As shown in Fig. 4F, PAK1 KD by itself reduced HK binding capability by 15.3% and 32.6% of the remainder can be further suppressed by PF. In comparison, PAK2 KD lead to 33.3% loss of HK binding capability and the remainder activity was largely resistant to PF (PF-associated binding loss was only 18.1%). Because PAK1 level increased in PAK2 KD cells, the result indicates that the loss of PAK2 function in cell adhesion could not be compensated by PAK1. Combined with the observations in Fig. 4D, these data suggest that PAK2 is the major target of PF activity in the HK adhesion assay whereas PAK1 has a minor contribution. This conclusion is further supported by time course analysis of PAK1 and PAK2 KD cells in adhesion test (Supplementary Fig. S8D).

Using siRNA-mediated KD approach, we also examined the role of PAK1 and PAK2 in cell proliferation and survival in ATL55T(+) and ATL43Tb(−) cells. ATL55T(+) cells underwent moderate and substantial apoptosis following PAK1 and PAK2 KD, respectively (Fig. 5B; Supplementary Fig. S9A); yet, PAK1 or PAK2 KD did not alter survival of ATL43Tb(−) cells (Fig. 5B; Supplementary Fig. S9B). We did not observe notable cell cycle changes in either cell lines (Supplementary Fig. S9C). As to the molecular changes that may account for the apoptosis phenotype following PAK1 and PAK2 KD, we observed a reduction in total- and PY-STAT5, p-ERK1/2, and MCL1 proteins 72 hours following siRNA transfection (Fig. 5C). The pro-apoptosis protein BAD was elevated at 48 hours only. Interestingly, similar changes were also detected in the apoptosis-resistant ATL43Tb(−) cell line except PY-STAT5, which is absent at baseline in this IL-2-independent cell line (Supplementary Fig. S10). This raises an interesting question as to whether PY-STAT5 is a critical mediator of PAK1/2 function in cell survival.

In addition to the NA-ATLL cell lines, we have also established and characterized PDX models using primary samples from our patients. In these models, distal tissue/organ dissemination from subcutaneous engraft site is a prominent feature and the pattern is similar to human ATLL (Fig. 6B). Using a PDX models derived from ATL18 and a CDX model derived from ATL43Tb(−), we evaluated in vivo anti-ATLL activity of PF. In the initial experiment using the CDX model, PF administered at 12 mg/kg twice a week achieved significant TGI in 19 days (TGI = 54.9%; Supplementary Fig. S11A). We then initiated a follow up study using the PDX model, where PF was administered daily over a period of 3 weeks (Fig. 6A). At the 12 mg/kg daily dose, PF nearly completely inhibited the subcutaneous tumor growth (TGI = 86.8% on day 16), whereas a substantial albeit more moderate inhibitory effect was achieved with 6 mg/kg (TGI = 59.0% on day 16). Ki67 staining revealed reduced cell proliferation in the subcutaneous tumor mass in PF-treated animal; however, TUNEL-based apoptosis assay did not detect notable differences due to PF treatment (Supplementary Fig. S11D). Most importantly, PF completely prevented distal dissemination of leukemic cells from the subcutaneous site. This is evident from the histology of spleen, liver, lung, kidney, and bone marrow (Fig. 6B) as well as human CD45-based flow cytometry quantification of infiltrating leukemic cells (Fig. 6C). We also examined CADM1 expression pattern at different sites of disease presentation. In keeping with a requirement for CADM1 in tissue/organ infiltration by ATLL cells, nearly all leukemic cells located in internal organs expressed this adhesion molecule (83%–100%) whereas in the subcutaneous site, CADM1 expression was much more variable (30.3%–88.3%) (Fig. 6D). These results demonstrate that PF has potent anti-ATLL activity in xenograft ATLL models where it can strongly inhibit not only in situ tumor growth but also distal tissue/organ dissemination. Based on the overall health status, body weight change, and heart histology, the PF treatment protocol did not cause measurable, off-site toxicity in the test animals (Supplementary Fig. S11B and S11C).

ATLL is a very aggressive T-cell malignancy with no effective therapy (1, 3, 5). Motivated by the observation that PAK2 is amplified in 24% of Japanese patients with ATLL (8), we examined therapeutic activity of PAK inhibition and characterized roles of PAK family kinases in this disease. We report there that PAK1/2/4 are the three most abundantly expressed PAK isoforms in ATLL. Functional studies demonstrate that the pan-PAK inhibitor PF has strong in vitro and in vivo anti-ATLL activity in a number of preclinical models. The observed activity of PF is likely attributed to its ability to target several PAK isoforms simultaneously because genetic silencing of either PAK1 or PAK2 only produced variable and moderate cytotoxicity. PAK2, but likely not PAK1, is the major mediator of CADM1-mediated stromal interaction, which is an important step in systemic dissemination of the disease. This is consistent with the observation that PAK2 is more frequently amplified in the aggressive subtypes of ATLL.

Among the six PAK isoforms, PAK2 and PAK4 are ubiquitously expressed, whereas the other four are restricted to specific tissue/organs (9, 32). In all ATLL samples tested, PAK2 mRNA is the most abundantly expressed followed by PAK1 and PAK4 (Fig. 1B and C). The protein products of all three of these genes can be readily detected in ATLL samples tested (Fig. 1D). Although PAK1 and PAK2 have virtually identical substrate specificities, they play different and nonredundant roles in development and cancer (32). For example, PAK1 but not PAK2 have been implicated as a direct regulator of nuclear gene expression program by phosphorylating transcription factors CtBP and STAT5 (29, 33, 34). For cytoskeletal remodeling, rapid turnover of focal contacts at the leading edge of migrating breast cancer cells requires PAK1 (35, 36), whereas modulation of RhoA activity during master cell degranulation is controlled by PAK2 but not PAK1 (37, 38). In this study, we have found that ATLLs under-express PAK2 mRNA relative to normal CD4 T cells and other T-cell malignancies (Fig. 1C; Supplementary Fig. S2D); yet, PAK2 protein levels appear to be comparable to those found in other T-cell malignances (Fig. 1D) implying additional posttranscriptional regulation. We also found that PAK2 has a more important contribution than PAK1 to stromal adhesion and cell survival in ATLL cells (Figs. 4F and 5B). It is therefore possible that as a disease category, ATLLs tend to under-express PAK2 mRNA, and this may have rendered PAK2 levels limiting for in vivo disease progression, and hence PAK2-amplification events are positively selected in aggressive ATLLs.

We selected PAK1 and PAK2 for gene-specific functional studies since they are the most abundantly expressed PAKs in ATLL. Knockdown experiments showed that the contribution of these two genes to cell autonomous proliferation and survival is not universal. In ATL43Tb(−) cells grown in suspension, they play negligible roles; in ATL55T(+), PAK2 KD caused significant apoptosis while PAK1 KD-induced cell death was more modest. Given the fact that the PF IC₅₀ dose for ATL55T(+) is nearly twice of that for ATL43Tb(−) (Fig. 2A) and yet ATL55T(+) was more sensitive to PAK1/2 KD, it is possible that targets other than PAK1 and PAK2 also mediated PF toxicity in the in vitro drug sensitivity tests. One such possibility is PAK4. Among the six Japanese cell lines studied by western blotting, the two most sensitive ones have very low phospho-Ser474-PAK4 signals [Su9T01 and ATL43Tb(−)] (Fig. 2B). Therefore, when all of our findings are considered as a whole, there is a formal possibility that both PAK2 and PAK4 mediated PF-induced toxicity in the in vitro toxicity assays. Non-PAK kinases that could be targeted by PF under our experimental conditions could also be involved (24).

Multi-organ involvement is the cause of mortality in ATLL. Therefore, ATLL treatment must effectively block systemic dissemination of leukemic cells. Two experiments in this study addressed this issue. In the adherence assays, PF significantly inhibited CADM1-dependent interaction with HK largely through PAK2 (Fig. 4). Although the importance of CADM1 in ATLL progression and systemic dissemination has been documented in the literature (11, 13, 14), this is the first time that a specific member of PAK is shown to operate downstream of CADM1 in this disease. Because stromal interaction significantly enhanced survival of primary ATLL cells (Fig. 4A and B), and this process is largely PAK2-dependent, PAK2 is expected to function as a major survival factor in ATLL microenvironment as well. It is worth noting here that, in our xenograft study, PF dosed daily at 12 mg/kg potently inhibited subcutaneous tumor growth and completely prevented distal dissemination of leukemic cells (Fig. 6). Combined with the results from our biochemical analysis and PAK1/2 knockdown experiments, such PF activities should be largely attributed to PAK2 with minor contributions from other PAK isoforms, such as PAK1 and PAK4.

In summary, our findings in this study showed that PAK2, a gene frequently amplified in ATLL, facilitates CADM1-mediated stromal interaction and promotes ATLL cell survival. Furthermore, PAK inhibition through an inhibitor such as PF may hold great promise as a targeted therapy for aggressive ATLLs. Although PF was withdrawn from further clinical testing due to inadequate oral availability, additional PAK inhibitors have since been designed (24) and some of them may possess a favorable therapeutic window for a disease associated with frequent PAK2 amplification.

U. Steidl reports receiving other commercial research support from GlaxoSmithKline, Bayer Healthcare, and Aileron Therapeutics; holds ownership interest (including patents) in Stelexis Therapeutics; and is a consultant/advisory board member for Pieris Pharmaceuticals, Aileron Therapeutics, Stelexis Therapeutics, and Bayer Healthcare. M. Janakiram is a consultant/advisory board member for Miragen and Seattle Genetics. No potential conflicts of interest were disclosed by the other authors.

We are grateful to Drs. Thomas Waldmann and Michael Petrus (NIH/NCI) and Prof. Naomichi Arima (Kagoshima University, Japan) for the gift of the Japanese ATLL cell lines; Prof. Naomichi Arima for valuable advice on establishing ATLL long-term cultures; and Drs. John Chan (City of Hope) and Giorgio Inghirami (Weill Cornell Medical College) for the gift of HH, MAC1, and FePD1 cell lines. We also thank Dr. Jonathan Chernoff (Fox Chase Cancer Center, Philadelphia) for insightful discussion and Dr. Robert Stanley (Albert Einstein College of Medicine) for assisting with the design of PAK inhibitor sensitivity tests. Flow cytometry analyses were performed at the Albert Einstein College of Medicine Flow Cytometry facility, which is supported by an NCI Cancer Center Support Grant (P30CA013330). FACS sorting was performed at the Einstein Human Stem Cell FACS and Xenotransplantation Facility, which is supported by a grant from the New York State Department of Health (NYSTEM Program) for the shared facility (C029154). This work was supported by research grants from the Leukemia & Lymphoma Society Translational Research Project (No. 6471-15); Dr. Louis Skarlow Memorial Trust (B.H. Ye); and Aids Malignancy Consortium grant UM1CA121947 (M. Janakiram). E.Y. Chung and Y. Mai are supported by the Harry Eagle Scholarship from the Department of Cell Biology, Albert Einstein College of Medicine, and U.A. Shah was supported by the hematology/oncology fellowship program, Montefiore Medical Center and Jacobi Medical Center.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.