Genome-wide CRISPR screen identifies CDK6 as a therapeutic target in Adult T-cell leukemia/lymphoma.

Adult T-cell leukemia/lymphoma (ATLL) is an aggressive T-cell malignancy with a poor prognosis with current therapy. Here we report genome-wide CRISPR-Cas9 screening of ATLL models, which identified CDK6, CCND2, BATF3, JUNB, STAT3, and IL10RB as genes that are essential for the proliferation and/or survival of ATLL cells. As a single agent, the CDK6 inhibitor palbociclib induced cell cycle arrest and apoptosis in ATLL models with wild type TP53. ATLL models that had inactivated TP53 genetically were relatively resistant to palbociclib owing to compensatory CDK2 activity, and this resistance could be reversed by APR-246, a small molecule activator of mutant TP53. The CRISPR-Cas9 screen further highlighted the dependence of ATLL cells on mTORC1 signaling. Treatment of ATLL cells with palbociclib in combination with mTORC1 inhibitors was synergistically toxic irrespective of the TP53 status. This work defines CDK6 as a novel therapeutic target for ATLL and supports the clinical evaluation of palbociclib in combination with mTORC1 inhibitors in this recalcitrant malignancy.

essential for proliferation and suvival 5 . BATF3 serves as a master regulatory 1 transcription factor that together with IRF4 coordinates superenhancer function 2 throughout the ATLL genome. Importantly, this regulatory module is augmented by 3 the HTLV-1 HBZ protein, which is also essential for ATLL cell viability 5 . 4 This shRNA screening effort utilized a sub-genomic library targeting 1,051 5 genes, raising the possibility that other essential genes in ATLL remain to be discovered. 6 The clustered regularly interspaced short palindromic repeats (CRISPR)-associated 7 endonuclease 9 (Cas9) system is a genome editing technology by which targeted genes 8 can be inactivated efficiently and specifically, thereby enabling genome-wide functional 9 genomic screening [8][9][10][11][12] . In recent studies, CRISPR-Cas9 screens have successfully 10 identified essential genes involved in cancer cell viability 8,9 . In this study, we performed 11 unbiased genome-wide CRISPR-Cas9 screens in multiple ATLL models with the goal 12 of identifying new therapeutic strategies for this aggressive malignancy. 13 14 Downloaded from http://ashpublications.org/blood/article-pdf/doi/10.1182/blood.2021012734/1847836/blood.2021012734.pdf by guest on 02 December 2021 expected, both factors were essential in ATLL as judged by the CRISPR-Cas9 screens, 1 but IRF4 was removed from our ATLL-specific essential gene set due to its essentiality 2 in MCL lines. In follow-up experiments, sgRNAs targeting BATF3 or IRF4 3 (sgBATF3 or sgIRF4) were specifically toxic for all five ATLL cell lines tested ( Figure  4 2A; Supplemental Figure S1A). 5 The BATF3 BZIP transcription factor binds to AP-1 motifs by forming 6 heterodimers with other BZIP family members, but the relevant BATF3-interacting 7 subunit in ATLL is unknown. We noted that three sgRNAs targeting the BZIP protein 8 JUNB were highly toxic for ATLL cell lines in the CRISPR-Cas9 screens while MCL 9 cell lines were unaffected ( Figure 1D). The immunohistochemical analysis of primary 10 biopsy samples revealed that JUNB protein was expressed in all ATLL cases 11 (Supplemental Figure 1B- In normal T cells, JUNB forms heterodimers with BATF3 and acts as an AP-1 7 transcription factor that cooperatively binds with IRF4 to composite AP-1/IRF DNA 8 motifs 16,17 . JUNB can also heterodimerize with the HTLV-I-encoded transcription 9 factor HBZ 17 , an indispensable molecule in ATLL pathobiology 1,5 . An antibody 10 against JUNB efficiently immunoprecipitated both BATF3 and HBZ in ATLL cell lines 11 ( Figure 2G). Together, these findings suggest that JUNB, BATF3, HBZ and IRF4 12 control an essential regulatory network in ATLL, and as such are worthy, while 13 challenging, candidates for the therapeutic targeting. 14 As indicated by the CRISPR-Cas9 screens ( Figure 1D), knockout of STAT3 1 by two sgRNAs (sgSTAT3) was toxic for all ATLL lines tested but not for the MCL 2 lines ( Figure 3A and B). In KK1 ATLL cells, the toxicity of STAT3 knockout was 3 reversed by the ectopic expression of a wild-type STAT3 coding region that was 4 engineered to be resistant to the STAT3 sgRNA ( Figure 3C). This was not the case in 5 ST1 ATLL cells, which harbored a STAT3 D556N mutation, which is a hotspot somatic 6 mutation in the DNA binding domain, which has been observed previously in ATLL 3 , 7 extranodal NK-T cell lymphoma 18 , peripheral T cell lymphoma not otherwise specified 8 (PTCL-NOS) 19 and diffuse large B cell lymphoma 20-23 ( Figure 3D). Given that 9 STAT3 mutations are present in 21.4% of primary ATLL tumors, many of which have 10 been shown to be gain-of-function 3 , we suspected that this may also be true of 11 STAT3 D556N . The ectopic expression of an sgSTAT3-resistant STAT3 coding region 12 bearing the D556N mutation almost completely rescued ST1 cells from 13 sgSTAT3-mediated toxicity ( Figure 3D). In contrast, the ectopic expression of 14 sgSTAT3-resistant cDNAs encoding STAT3 Y640F or STAT3 D661Y , two highly recurrent these results suggest that this line depends on autocrine IL-10 signaling. By contrast, 1 the IL-2-independent lines ST1 and Su9T01 depended on the IL-22 receptor subunits 2 IL22RA1 and IL10RB, as well as STAT3. Notably, ST1 and Su9T01 are the only 3 ATLL cell lines tested that expressed high mRNA levels of IL22, suggesting that these 4 lines acquired a dependence on autocrine IL-22 signaling (Supplemental Figure S2B, 5 right panel). 6 The IL-22 receptor activates STAT3 through JAK1 and TYK2 in normal T 7 cells 25 . In the CRISPR-Cas9 screen, KK1 and Su9T01 cells were strongly dependent 8 on both of these kinases, whereas ST1 cells had only a moderate dependence on JAK1. 9 Although the JAK1/2 inhibitor ruxolitinib was able to kill IL-2-dependent ATLL lines, 10 which depend on JAK1, this agent had little if any effect in IL2-independent ATLL 11 lines, including Su9T01 and ST1 (Supplemental Figure S2C). In Su9T01 cells this 12 could be due to compensatory TYK2 activity whereas in ST1 cells, the gain-of-function 13 STAT3 D556N mutation may have blunted the effect of the drug. 14 Four sgRNAs targeting CDK6 were highly toxic for ATLL cell lines in the 1 CRISPR-Cas9 screens, as were two CCND2 sgRNAs ( Figure 1D). CCND1, CCND2, 2 and CCND3 are G 1 /S-phase-specific cyclins that interact with CDK4 or CDK6 to form 3 an active kinase, which promotes the G1-S phase cell cycle transition by 4 phosphorylating and inactivating the retinoblastoma (Rb) protein, 26,27 . In confirmatory 5 experiments, ATLL cells were transduced with a CDK6-targeting sgRNA (sgCDK6), 6 thereby reducing levels of CDK6 and phosphorylated Rb by immunoblot analysis 7 ( Figure 4A). sgCDK6 caused time-dependent toxicity in all five ATLL cell lines but did 8 not affect control MCL lines ( Figure 4B; Supplemental Figure S3A). By contrast, four 9 of these ATLL lines were unaffected by an sgRNA targeting CDK4 (sgCDK4), whereas 10 MCL lines were CDK4-dependent (Supplemental Figure S3A). ATLL cells were 11 rescued from the toxicity of sgCDK6 by ectopic expression of a CDK6 coding region 12 that was engineered to be sgCDK6-resistant, whereas expression of CDK4 only rescued 13 partially ( Figure 4C; Supplemental Figure 3B and C). As expected, knockout of 14 CDK6 in ATLL cells caused a pronounced G 1 -phase arrest ( Figure 4D F and I). By contrast, ATLL cell lines were unaffected by sgRNAs targeting CCND1 7 or CCND3 (Supplemental Figure S3A). These findings are consistent with the fact 8 that ATLL cell lines had considerably higher levels of CCND2 protein than CCND1 or 9 CCND3 ( Figure 4H). Further, CCND2 mRNA levels were higher in biopsy samples 10 from patients with ATLL than in ALK-positive anaplastic large-cell lymphoma 11 (ALK+ALCL) biopsies, and roughly equivalent to those in PTCL-NOS biopsies ( Figure  12 4I). By contrast, CCND1 and CCND3 mRNA levels were lower in ATLL that in 13 ALK+ ALCL and PTCL-NOS. The toxicity of CCND2 knockout was efficiently 14 reversed by ectopic expression of an sgRNA-resistant CCND2 coding region, but also 15 by the CCND1 and CCND3 coding regions ( Figure 4G; Supplemental Figure S3J and K), indicating that three D-type cyclins have similar functions in these cells. Of note 1 in this regard, knockout of CCND2 in ST1 cells led to an upregulation of CCND3, 2 likely blunting the toxic effect of sgCCND2 in this cell line (Supplemental Figure 3I). 3 4 The CDK4/6 inhibitor palbociclib inhibits ATLL proliferation and survival 5 Given that CDK6/CCND2 controls cell cycle progression in ATLL and that 6 CDK6 knockout induced apoptosis in ATLL cells, we explored the potential of CDK6 7 as a therapeutic target in ATLL. For this, we evaluated palbociclib, an inhibitor of 8 CDK4 and CDK6 that has been FDA-approved for advanced breast cancer and has 9 shown preliminary clinical activity in MCL 28 . Treatment of 11 ATLL cell lines with 10 palbociclib for 4 days substantially reduced the numbers of viable cells, with an average 11 IC 50 (1.776 M) that was comparable to that observed in MCL cell lines (1.770M) 12 ( Figure 5A). Palbociclib treatment of ATLL lines induced a dose-dependent decrease 13 in the phosphorylation of Rb (Figure 5B), a G 1 -phase cell cycle arrest, and a 14 The sensitivity of ATLL cell lines to palbociclib varied over a wide IC 50 range 1 (9-6500 nM), with some lines relatively insensitive to this drug. A previous study 2 reported that TP53 mutation is associated with relative insensitivity to another 3 CDK4/CDK6 inhibitor (abemaciclib) in various cancer types 29 . Given that somatic 4 mutation and genomic deletion in TP53 are frequent events in ATLL 3,30 , we sequenced 5 the TP53 exonic regions in 11 ATLL cell lines, revealing TP53 mutations in 4 lines 6 (MT1, KK1, TL-Om1, and ATL43Tb(-)), wild type TP53 in 4 lines (ED41214C(-), S1T, 7 ST1 and KOB), and a TP53-null phenotype based on the lack of a delectable RT-PCR 8 product (ED40515(-), ATL-55T(+), and Su9T01) (Supplemental Table S3). ATLL 9 cell lines with genetically altered TP53 (mutated and null phenotype) tended to be 10 insensitive to palbociclib compared to cell lines harboring intact TP53 (p=0.058; Figure  11 5F). 12 To functionally evaluate whether TP53 inactivation alters the sensitivity to 13 palbociclib, ATLL cell lines carrying mutated TP53 were treated with APR-246 14 (PRIMA-1MET), a small molecule that can restore transcriptional activation by mutant 15 TP53 isoforms 31 . Indeed, APR-246 treatment rendered TP53-mutated ATLL cells sensitive to palbociclib but had no effect in TP53-null cells ( Figure 5G). To directly 1 link TP53 status to palbociclib sensitivity, we transduced ST1 ATLL cells, which are 2 TP53 wild type, with a lentivirus co-expressing an sgRNA targeting TP53 (sgTP53) and 3 a GFP reporter gene, resulting in a cell pool in which roughly half of the transduced 4 cells were GFP + . Over an 8-day treatment course with palbociclib, the GFP + /TP53 -5 knockout population outgrew the GFPpopulation in a time-and dose-dependent 6 manner ( Figure 5H; Supplemental Figure S4D). This TP53 knockout-mediated 7 resistance to palbociclib was also evident in assays of cell proliferation, cell cycle 8 progression, and apoptosis ( Figure 5I-K). Taken together, these results indicate that 9 loss-of-function TP53 genetic alterations confer insensitivity to palbociclib in ATLL. 10 TP53 prevents the G 1 /S cell cycle transition by inducing expression of p21 11 (CDKN1A), which binds and inhibits the activity of CDK2/Cyclin E 32,33 . Notably, 12 inactivation of CDK6 or palbociclib treatment of ST1 ATLL cells, increased expression 13 of p21 as well as another CDK2 inhibitor p27, but this was not observed in ATLL lines 14 with genetic inactivation of TP53 ( Figure 5L and M), suggesting that the suppression of 15 CDK2 might be necessary for an optimal response of ATLL cells to palbociclib. To test this hypothesis, we transduced TP53-mutated KK1 and TP53-null Su9T01 ATLL 1 cells with a lentivirus co-expressing an sgRNA targeting CDK2 (sgCDK2) with a GFP 2 reporter gene and monitored the ratio of the GFP + /sgCDK2 + cell population to the GFP -3 nontransduced population. The CDK2-knockout population decreased in abundance 4 during palbociclib treatment in a time-dependent manner, but this was not the case in 5 cells transduced with a control sgRNA (sgAAVS1) ( Figure 5N; Supplemental Figure  6 4E and F). In another test of this hypothesis, we stably inactivated TP53 in ST1 cells 7 using a TP53 sgRNA, or transduced the cells with a control sgRNA (sgAAVS1). 8 After transduction of the TP53 knockout ST1 cells with the sgCDK2/GFP expression 9 vector, we exposed the cells to palbociclib and observed a time-dependent decrease in 10 the GFP + /sgCDK2 + population, an effect that did not occur in the control ST1 cells 11 ( Figure 5O). Taken together, these data demonstrate that optimal sensitivity to 12 palbociclib in ATLL requires TP53 to inhibit CDK2 function, suggesting that 13 monotherapy with this drug is likely to work preferentially against ATLL tumors with 14 wild type TP53. 15

mTORC1 is a targetable vulnerability in ATLL 1
To identify additional therapeutic targets in ATLL, we interrogated our 2 whole-genome CRISPR screen data for dependencies of ATLL cells on signaling and 3 regulatory pathways that have been associated with T-cell malignancy. 3,19,34, 35 We 4 noted a dependency on MTOR and other components of the mTORC1 pathway ( Figure  5 6A), which we validated by sgRNA-mediated knockout of MTOR in multiple ATLL 6 lines ( Figure 6B and C; Supplemental Figure S5A). mTORC1 directly phosphorylates 7 4EBP1 and S6 kinase, which in turn phosphorylates ribosomal protein S6, thereby 8 stimulating mRNA translation and mitochondrial activity/biogenesis 36 . By 9 immunoblot analysis, both 4EBP1 and S6 were phosphorylated in ST1 ATLL cells 10 ( Figure 6D). 11 As the combinatorial benefit of CDK and mTORC inhibitors was previously 12 reported in several cancer types 37 , we hypothesized that partial pharmacological 13 inhibition of the mTORC1 pathway might synergize with palbociclib in killing ATLL 14 cells irrespective of TP53 status. We treated TP53 wild type (ST1) and TP53 mutant 15 (KK1) ATLL cells with a fixed concentration of the mTORC1 inhibitor (2.5 M) or DMSO along with a range of concentrations of palbociclib. At all palbociclib 1 concentrations, concurrent inhibition of mTORC1 with everolimus significantly reduced 2 viable cell numbers compared with treatment with palbociclib alone ( Figure 6E; 3 Supplemental Figure S5B). Of note, the two drugs synergized in both TP53 wild type 4 and TP53 mutant ATLL lines, as indicated by a combination index below 1 (see 5 Methods; Figure 6E; Supplemental Figure S5B). As mentioned above, knockout of 6 TP53 in ST1 cells rendered the cells relatively resistant to palbociclib alone but 7 combined treatment with everolimus resulted in synergistic toxicity ( Figure 6F). 8 These synergistic effects were also evident in experiments combining palbociclib and 9 AZD8055, a dual mTORC1 and mTORC2 inhibitor (Supplemental Figure S5C). 10 Combined palbociclib/everolimus treatment decreased phosphorylation of Rb to a 11 greater degree that observed with either agent alone ( Figure 6G), as was also true for the 12 combination of palbociclib and AZD8055 (Supplemental Figure S5D). 13 Phosphorylation of S6 and 4EBP1 was fully suppressed by everolimus or AZD8055, 14 alone or in combination with palbociclib, as expected ( Figure 6H   We next investigated the effect of inhibiting CDK6 and/or mTORC1 in 1 primary ATLL patient samples. All patients had intact TP53 except for the presence 2 of a common single-nucleotide polymorphism (P72R). Consistent with ATLL cell line 3 data, palbociclib treatment inhibited the proliferation of primary ATLL cells in a 4 dose-dependent manner ( Figure 7A and B). Moreover, the combination of palbociclib 5 with everolimus or AZD8055 inhibited cell proliferation to a greater degree treatment 6 with either agent alone in primary ATLL cells from five patients ( Figure 7A and B). 7 Notably, these treatments did not suppress the proliferation of normal bystander CD4 8

T-cells, CD8 T-cells and monocytes except for mTORC1 inhibitor-treated B-cells in 9
PBMNC from ATLL patients (Supplemental Figure 6A). Furthermore, cell proliferation 10 of healthy donor-derived T-cells were also not affected (Supplemental Figure 6B and C). 11 To explore the efficacy and tolerability of this combination in vivo, we utilized a mouse 12 xenograft model created using the ATL43Tb(-) ATLL cell line. Treatment of 13 xenografted mice with palbociclib or everolimus monotherapy inhibited tumor growth 14 compared to vehicle control. Combined treatment with both drugs markedly inhibited 15 tumor growth to a greater degree than either drug alone ( Figure 7C-E). Of note, there was no significant difference in systemic toxicity among the four groups of mice during 1 drug treatment, as judged by body weight ( Figure 7F). These results were also 2 confirmed another xenograft model created using the Su9T01 ATLL cell line ( Figure  3 7G and H; Supplemental Figure 6D). To discover essential genes and potential new therapeutic targets in ATLL, 7 we performed unbiased genome wide CRISPR/Cas9 library screening. The resulting 8 dataset provides a blueprint of ATLL biology, including some known dependencies as 9 well as additional essential pathways that can be attacked using targeted agents that are 10 clinically available. Our in-depth analysis of CDK6 as a molecular target in ATLL led 11 us to discover a therapeutic synergy between the CDK6 inhibitor palbociclib and 12 mTORC1 inhibitors, a combination that can now be tested in clinical trials. 13 Previous work identified BATF3 and IRF4 as essential transcription factors in 14 ATLL 5 , but the present work revealed that ATLL cells depend equally on the B-ZIP 15 factor JUNB, which heterodimerizes with BATF3 to form an AP-1 DNA binding transcription factor. Importantly, JUNB also co-immunoprecipitated with HBZ, an 1 essential, HTLV1-encoded B-ZIP factor that drives expression of BATF3 and its 2 extensive transcriptional network in ATLL 5 , suggesting that JUNB is a key component 3 of a feed-forward regulatory loop that sustains ATLL viability 5 . Although transcription 4 factors have been traditionally challenging as therapeutic targets, they are increasingly 5 accessible as targets using PROTAC technology to target them for proteolytic 6 degradation 38 . 7 The importance of the JAK/STAT signaling pathway in ATLL first became 8 evident with the discovery of recurrent gain-of-function STAT3 mutations in 21.4% of 9 ATLL cases. 3 . Unexpectedly, our CRISPR screens revealed that autocrine signaling 10 by IL-22 or IL-10 may provide additional mechanisms by which STAT3 can be 11 activated in ATLL. Recently, it was reported that transgenic expression of HBZ in a 12 mouse model promotes autocrine IL-10 signaling in normal mouse T cells, which may 13 be potentiated by interaction of HBZ with STAT3 39 . Thus, the essential transcription 14 factor network in ATLL may intersect with JAK/STAT signaling, perhaps explaining 15 the frequent acquisition of STAT3 mutations in ATLL. Therapeutic strategies to target the JAK/STAT pathway in ATLL will have to contend with the mechanistic diversity 1 by which this pathway is engaged in ATLL. For instance, successful therapeutic 2 strategies to attack JAK/STAT signaling in ATLL will need to consider which JAK 3 family kinases are engaged in a particular ATLL tumor as well as the presence of 4 gain-of-function STAT3 mutations, which likely render ATLL cells independent of 5 upstream signals from JAK family kinases. In particular, ATLL lines with autocrine 6 IL-22 signaling can activate both JAK1 and TYK2, potentially explaining why the 7 JAK1/JAK2 inhibitor ruxolitinib was ineffective against these lines.
These 8 observations may also explain why clinical efficacy of ruxolitinib has been limited to 9 the IL-2-dependent phase of ATLL 40,41 . 10 Our study revealed CDK6 as an attractive molecular target in ATLL. Loss 11 of the endogenous CDK4/6 inhibitor CDKN2A (p16) is common in ATLL (24-26%) 3,42 , 12 suggesting that deregulation of the G 1 /S cell cycle transition is deeply implicated in 13 ATLL pathobiology. Pharmacological inhibition of CDK6 by palbociclib suppressed 14 ATLL cell proliferation and survival, although the toxicity was suboptimal in some 15 the loss of TP53 function in a subset of ATLL lines. Somatic mutation in TP53 and 1 genomic deletion of chromosome 17p13.1, harboring the TP53 locus, are frequent 2 genetic events in ATLL, occurring in 17.8% and 26% of patients, respectively 3,30 . 3 Given the above, the TP53 status of ATLL tumors should be considered as a potential 4 biomarker of response in future clinical trials of palbociclib in ATLL. 5 To overcome the insensitivity to palbociclib of ATLL lines with inactive 6 TP53, we developed a rational combination of palbociclib with mTORC1 inhibitors. 7 These agents were synergistically toxic in multiple ATLL models owing, in part, to 8 concerted suppression of Rb phosphorylation. Importantly, this therapeutic 9 combination not only induced cell cycle arrest but also triggered apoptosis in ATLL 10 cells, irrespective of TP53 status. Treatment of mice bearing ATLL xenografts with 11 palbociclib plus everolimus reduced tumor growth significantly and was apparently well

CONFLICT-OF-INTEREST DISCLOSURE
The authors declare no competing financial interests related to this work.