PTEN regulated PI3K-p110 and AKT isoform plasticity controls metastatic prostate cancer progression

PTEN loss, one of the most frequent mutations in prostate cancer (PC), is presumed to drive disease progression through AKT activation. However, two transgenic PC models with Akt activation plus Rb loss exhibited different metastasis development: Pten/RbPE:−/− mice produced systemic metastatic adenocarcinomas with high AKT2 activation, whereas RbPE:−/− mice deficient for the Src-scaffolding protein, Akap12, induced high-grade prostatic intraepithelial neoplasias and indolent lymph node disseminations, correlating with upregulated phosphotyrosyl PI3K-p85α. Using PC cells isogenic for PTEN, we show that PTEN-deficiency correlated with dependence on both p110β and AKT2 for in vitro and in vivo parameters of metastatic growth or motility, and with downregulation of SMAD4, a known PC metastasis suppressor. In contrast, PTEN expression, which dampened these oncogenic behaviors, correlated with greater dependence on p110α plus AKT1. Our data suggest that metastatic PC aggressiveness is controlled by specific PI3K/AKT isoform combinations influenced by divergent Src activation or PTEN-loss pathways.

INTRODUCTION more dependent on AKT2, and in some cases, on AKT3, whereas the re-expression of PTEN, which dampened oncogenic behavior, converted dependency to AKT1. Moreover, inhibition of clonogenic survival and chemotaxis of PTEN-de cient PC cells required the combined use of p110β plus AKT2 inhibitors, whereas for PTEN-positive cells, p110α plus AKT1 inhibitors had the greater effect. Our data strongly suggest that the clinical targeting of the PI3K-AKT axis in PC requires knowledge of PTEN status followed by the inhibition of the appropriate p110/AKT isoforms.
sh/siRNA: The RPCCC Gene Modulation Core provided the PTEN shRNA clones (Dharmacon) V2LHS_192536, V2LHS_92314, and control pGIPZ. Cells were infected with high-titer lentivirus containing the shRNA sequences and selected using puromycin (2 µg/mL). si/shRNA sequences are described in Supplementary Table S2. siRNA transfection: Cells at 50-80% con uence in 6-well tissue culture plates were transfected with siRNA in Lipofectamine 2000 or 3000 (Invitrogen) as per manufacturer's protocol, and then cell lysates were analyzed after 48-72 h.
Proliferation assays: Transfected or treated cells (1-2x10 3 ) were seeded on 96-well plates and allowed to grow for the indicated times. Cells were washed with PBS, xed with ice-cold 100% methanol for 10 min at -20°C, and stained with 0.5% crystal violet. Dried plates were solubilized with 10% acetic acid for 20 minutes at room temperature, and absorbance read at 595 nm.
Methylcellulose and anoikis assays: Cells were grown in 3D by suspending in media containing 1.3% methylcellulose (Sigma, St. Louis, MO) on plates pre-coated with 0.4% agarose, or grown in suspension atop 1.4% agarose-coated plates with complete (+ 10% FBS) or serum-depletion (+ 0.5% FBS) media. CellTiter-Glo 3D Cell Viability Assay (#G9681; Promega, Madison, WI) was used to determine the number of viable cells in 3D cell culture according to the manufacturer's protocol. An opaque-walled 96-well plate compatible with the Veritas Microplate Luminometer (Turner BioSystems, Sunnyvale, CA) was used, and the luminescent signal recorded according to the manufacturer's protocol.
Clonogenic assay: Survival of adherent cells was performed described [42], using 200 cells/well for T402 and 22Rv1, and 400 cells/well for LNCaP in 24-well plates.
RNA-Seq: Prostates were harvested from Akap12 −/− ;Pb-Cre:Rb1 / and Pb-Cre:Pten / :Rb1 / mice by microdissection, snap frozen in liquid nitrogen, and stored at -80°C. RNA was isolated from frozen tissue using TRIzol (Invitrogen) according to the manufacturer's instructions. RNA samples were interrogated by the RPCCC Genomics Shared Resource using TruSeq Stranded Total RNA with the RiboZero Gold library prep kit with 100ng input. Alignment to the mouse genome (mm10 version) was performed by the RPCCC Bioinformatics Shared Resource using RefSeq [43] and the UCSC Genome Browser [44]. Quality control for the raw reads was performed using fastqc [45] and adapter trimming was done using atropos [46]. Spliced alignments of reads to the reference genome was done using TopHat2 [47], allowing a maximum of 1 mismatch/read, and quality control for this alignment was done using RSeQC software [48]. The differential expression report was generated using DESeq2 [49].
Data analysis and statistics: The majority of data analysis and gure generation was performed in GraphPad Prism v7. Signi cance in the Oncomine database (ThermoFisher) was de ned as non-overlap of rst quartile primary prostate tumor samples and third quartile of metastatic prostate samples. Oncoprints and survival curves were generated from cBioPortal [53] for primary and mPC samples.

Page 7/27
AKT activation not su cient to induce prostatic adenocarcinoma. The loss of PTEN is one of the most prevalent changes in primary PC disease [54] (Fig. 1A). PTEN loss is a likely marker of cancer initiation because biallelic deletion of Pten in transgenic mouse models is su cient to induce HG-PIN in less oncogenic backgrounds such as C57BL/6 [55]. The loss of PTEN lipid phosphatase activity is thought to activate oncogenic AKT by allowing it to bind through its intrinsic PH domain to PI3K-generated PIP3 at the plasma membrane [56,57]. Indeed, PTEN loss correlates with increased AKT activation levels in more advanced cases of human PC, based on increased relative levels of AKT poS473 staining [58,59], serving as strong predictor of biochemical recurrence [60]. Progression to prostatic adenocarcinoma and distal metastases requires additional losses in tumor suppressors such as Rb or Smad4 [40,61], mimicking their frequent losses in primary PC (Fig. 1A). The notion that PTEN, RB1, or SMAD4 play important roles in regulating metastatic PC (mPC) is evidenced by gene losses that are more frequent in mPC than in primary lesions (Figs. 1A&B; Supplementary Fig. S1A). Consistent with AKAP12's known metastasis suppressor function [10], AKAP12 loss is 3.5-fold more frequent in mPC than in primary PC (Figs. 1A&B).
Although the loss of AKAP12 is less frequent than PTEN loss in primary PC (10% vs. 31% in the TCGA dataset), PTEN or AKAP12 losses statistically co-occur with RB1 loss (Supplementary Table S1), and these combinations show statistical signi cance in predicting disease-free survival (DFS) using TCGA datasets (Fig. 1C). AKAP12 is thought to attenuate oncogenic Src signaling by scaffolding pools of Src to lipid rafts, away from integrin/FAK/growth factor receptor-rich plasma membrane sites [13]. Indeed, the inducible upregulation of Akap12 in MLL[Tet OFF -SSeCKS/Akap12] PC cells [7] decreased relative phosphotyrosyl-p85α and AKT poS473 levels without affecting total AKT, p85α or PTEN levels (Fig. 1D).
This correlates with a statistical co-occurrence between AKAP12 loss and either increased levels of the SFK member, LYN (copy number gain, transcriptional upregulation) or Src poY416 , a shared marker of SFK activation [62] (Fig. 1E). This is consistent with the notion that LYN promotes PC progression and mPC formation [16, 63-65], whereas FYN, whose levels trend towards mutual exclusivity with AKAP12 loss (Fig. 1E), is thought to promote progression of neuroendocrine PC [66]. Yet, whereas both Pten/Rband Akap12/Rb-null prostate lesions exhibit Akt activation ( Fig. 2A) [3,40], their mPC progression pro les differ (Table 1): Pten/Rb-null mice develop aggressive prostatic adenocarcinomas associated with systemic metastases, whereas Akap12/Rb-null mice develop HG-PIN plus local, indolent lymph node metastases. This suggests that PTEN or AKAP12 control divergent AKT oncogenic progression pathways, with the latter likely depending more on SFK roles. Indeed, the majority of primary PC cases with PTEN loss are distinct from those with SRC or LYN gain ( Supplementary Fig. S1A). Thus, "AKT activation" in the context of RB loss is not su cient for progression to adenocarcinoma.  . S1B), suggesting a more important role for AKT2 in mPC progression. IB analysis of prostate lysates from 12 week-old WT, Akap12/Rb-, or Pten/Rb-null mice indicated that relative Akt poS473 levels were increased in Akap12/Rband Pten/Rb-null lesions compared to levels in WT prostates ( Fig. 2A). Ser473 is phosphorylated by mechanistic target-ofrapamycin complex 2 (mTORC2) [70] and PI3K [71], and is required to potentiate AKT serine/threonine kinase activity [30]. In addition, the relative increase of the pan-AKT substrate, PRAS40 poT246 , suggests similar levels of overall Akt activation in Akap12/Rband Pten/Rb-null compared to WT prostates ( Fig.   2A). In contrast, the relative level of Akt poT308 , a PDK1-mediated phosphorylation required for AKT activity [72], was elevated only in Pten/Rb-null prostates, when normalized to β-actin as a loading control. However, we found no change in relative PDK1 protein or activation levels in WT, Akap12/Rb-, or Pten/Rbnull prostates (Fig. 2B). As we showed previously [19], the loss of Akap12 alone was su cient to induce activated Akt in all four prostatic lobes (Fig. 2C) using an Ab that recognizes poS473/474 shared by AKT1/2. Compared to WT prostates, higher Akt1/2 poS473/474 levels were also detected in Akap12/Rband Pten/Rb-null prostates (Fig. 2C). There was increased nuclear signal in Pten/Rb-null adenocarcinomas ( Fig. 2C; bottom), consistent with a previous report of localization of AKT1 in the cytoplasm and AKT2 in the nucleus of PC-3 cells [39]. Moreover, whereas total Akt1 protein levels were similar in all three prostate genotypes, the relative levels of Akt2 and Akt3 were increased in Pten/Rb-null prostates ( Fig. 2A).
RNA-seq analysis showed no overall changes in Akt1 and Akt3 levels between Akap12/Rband Pten/Rbnull prostates, compared to an upregulation of Akt2 RNA in the Pten/Rb-null prostates (Fig. 2D). In order to assess the relative activation levels of AKT isoforms, AKT isoform proteins were immunoprecipitated using isoform-speci c Abs and the pull-downs probed for Akt poS473 (Fig. 2E). The relative activation level of Akt1 was similar in Akap12/Rband Pten/Rb-null prostates whereas Pten/Rb-null prostates showed increases in Akt2 and Akt3 activation levels, correlating with increased protein levels ( Fig. 2A).
Next, we probed these tissue samples with an Ab that detects AKT canonical substrates based on the shared phosphorylated motif, RXXS po /T po (Fig. 2F). Whereas the Akap12/Rb-null lysates had quantitative differences in the levels of several substrates compared to WT lysates, the Pten/Rb-null lysates also had qualitative differences, suggesting the targeting of unique substrates. This nding is consistent with the notion that different AKT isoforms may predominate in the two Tg PC models.
We then addressed how PTEN status controls PC oncogenic growth by re-expressing PTEN-GFP (vs. GFP alone in controls) in LNCaP or T402 cells ( Table 2, Fig. 3A), the latter derived from a murine Pten/Rb-null adenocarcinoma [40]. As well, we produced an isogenic pair of PTEN-positive 22Rv1 cells expressing shPTEN or scrambled (control) shRNA ( Supplementary Fig. S2D). PTEN re-expression in T402 cells neither changed protein levels of Akt isoforms or Ar (Fig. 3A), the relative expression of an Ar-regulated 19-gene panel (Fig. 3B), nor proliferation in 2D conditions with androgen-containing media (Supplementary Figs. S2A-C). Consistent with its role as a tumor suppressor, PTEN re-expression decreased relative PC invasiveness (Fig. 3C), clonogenicity (Fig. 3E), chemotaxis (Figs. 4A&C) and tumor formation (Fig. 6F), whereas PTEN knockdown in isogenic 22Rv1 cells increased clonogenicity (Fig. 3E). 2A&E), we asked if the knockdown of Akt2 would inhibit in vitro parameters of metastatic growth, using transwell assays for chemotaxis or Matrigel invasion, or by assaying for survival using either clonogenic or anoikis assays. AKT knockdowns in T402 and LNCaP cells were isoform-speci c (Fig. 3D, lower panel; Supplementary Fig. S3E). The knockdown of Akt2, but not Akt1, in the Pten-negative T402 PC cells ( Supplementary Fig. S3E) decreased invasiveness (Fig. 3C). In contrast, the re-expression of PTEN, which decreased the invasiveness of control T402 cells, switched dependence from Akt2 to Akt1. Similarly, the knockdown of either AKT2 or AKT3, but not AKT1, inhibited LNCaP invasiveness (Fig. 3D). PTEN reexpression decreased the invasiveness of LNCaP cells to the limits of detection (< 20 cells/ eld), thereby making it impossible to assess the effects of AKT isoform knockdown. Although AKT3 levels were quite low in LNCaP cells (requiring IP from 0.5mg of lysate protein followed by IB), knockdown caused a decrease in invasiveness (Fig. 3D), strongly suggesting that AKT3 promotes invasiveness in these cells.
Taken together, these data identify critical roles for AKT2 and AKT3 in the invasiveness of PTEN-de cient PC, and that upon PTEN re-expression, reliance on AKT1 increases.
Role of PI3K-p110 and AKT isoforms in PTEN-regulated survival and chemotaxis. Because PTEN-negative tumor cells have been reported to depend more on p110β for oncogenic growth [33,56], we next assessed how PTEN expression affected clonogenic survival of the PC isogenic pairs, and if PTEN affected sensitivity to small molecule inhibitors of PI3K-p110 and AKT isoforms, or after knockdown of p110/AKT isoforms. PTEN re-expression decreased the relative clonogenic survival of LNCaP or T402 cells, whereas PTEN knockdown increased survival of 22Rv1 cells (Fig. 3E). To address the role of PI3K and AKT isoforms in controlling clonogenic survival, we rst identi ed concentrations of p110α, p110β, AKT1 and AKT2 inhibitors that had minimal effect on the 2D proliferation of PC lines but were signi cantly above the IC50's reported for each drug (examples in Supplementary Figs. S2E & S4D). Importantly, we sought to show that the effects of the isoform-speci c inhibitory drugs mimicked what we found with isoformspeci c AKT and/or p110 si/shRNAs. Only the combination of p110βi and Akt2i signi cantly reduced the number of colonies in T402, whereas in T402[PTEN] cells, sensitivity changed to a combination of Akt1i and p110αi (Fig. 3F). Similar results were seen in LNCaP (Fig. 3G) and with other PTEN-positive ornegative human PC cells lines ( Supplementary Fig. S3A&B), or when combining knockdown of p110 ( Supplementary Fig. S3F) and AKT isoforms ( Supplementary Figs. S3C&D). Thus, these data suggest a plasticity with which PTEN directs survival dependency through both p110α and AKT1, whereas PTENde cient cells depend on both p110β and Akt2. Akt3 was not included because of the lack of Akt3speci c inhibitors.
We then analyzed the PC isogenic cell panel for the effects of PTEN on chemotaxis. 22Rv1, which are very poor at chemotaxis even if PTEN is knocked down, were omitted. The reintroduction of PTEN signi cantly reduced chemotaxis in LNCaP and T402 (Fig. 4A). We next asked if the differential PI3K and AKT drug sensitivities observed in the clonogenic assays also affected chemotaxis. Chemotaxis in T402 was inhibited by the combination of AKT2 and p110β inhibitors ( Fig. 4B; Supplementary Fig. 4B). PTEN re-expression abrogated most of the combined effect of AKT2i plus p110βi. As was observed in the clonogenic assays, T402 chemotaxis was not inhibited by p110αi and/or AKT1i, whereas in T402[PTEN], chemotaxis was sensitive to the combination of p110αi and Akt1i. Similar results were observed in LNCaP using isoform-inhibitory drugs (Fig. 4C) or RNAi ( Supplementary Fig. S4A), noting that 0.3 µM AKT2i was insu cient to inhibit LNCaP 2D proliferation ( Supplementary Fig. S2E) but su cient to inhibit chemotaxis (Fig. 4D).
Zhang et al. [37] showed that the dependence of PTEN-de cient BT549 (breast) and PC3 (prostate) cancer cells on p110β was likely due to the selective binding by CRKL to p110β, facilitated by Srcphosphorylated p130Cas. This correlated with increased suppression of PTEN-de cient tumor growth by combining p110β and Src inhibitors. We recapitulated these ndings in LNCaP and T402 cells using both out p110βi and the Src inhibitor, Saracatinib (Supplementary Fig. S4C). Additionally, Crkl RNA levels are roughly 4.2-fold higher in Pten/Rb-null than in Akap12/Rb-null tumors (Supplementary Fig. S4D). These data strengthen the notion that the p110β/AKT2 pathway activated in the absence of PTEN is separate from the Src/p110α/AKT1 pathway activated in the absence of AKAP12 (Fig. 1D).
In a previous study, CA-AKT1 S473D , but not CA-AKT2 S474D , rescued the phosphorylation of the mTORC2dependent substrate, ATP-citrate lyase, in brown adipocytes [41]. The stable expression of AKT1 S473D or AKT2 S474D in T402 cells equally increased the number and abundance of phospho-AKT substrates irrespective of PTEN status ( Supplementary Fig. S4E), validating the notion that they encode CA kinase variants. However, there was no distinction in the ability of CA-AKT isoform to increase chemotaxis in either T402 or T402[PTEN] cells ( Supplementary Fig. S4F). We found similar results using AKT1 or AKT2 constructs fused to an N-terminal myristylation domain known to potentiate associated kinase activity [73,74], namely, no distinction in the ability to induce phospho-AKT substrates and chemotaxis (data not shown). Thus, it is likely that in LNCaP or T402 cells, the CA mutants cannot differentiate AKT1-vs. AKT2-speci c functions.
SMAD4 loss as a marker of p110β/AKT2 dependence in PTEN-de cient PC cells. Previous data showed that Smad4 loss potentiates PC metastasis formation in Pten PE:−/− mice [61]. We analyzed whether Smad4 might serve as a marker of metastatic progression that could differentiate the aggressive Pten/Rb-null PC model from the indolent Akap12/Rb-null HG-PIN model. SMAD4 RNA levels inversely correlated with AKT2, but not AKT1, RNA levels in human PC cell lines (Fig. 5A) and when comparing Akap12/Rbvs. Pten/Rb-null prostate lesions (Fig. 5B). This corresponded to lower Smad4 protein levels in the more metastatic Pten/Rb-null tumors (Fig. 5C) and in human metastatic PC (Supplementary Fig.  S5). Smad4 protein levels were increased 2-to 2.5-fold by the knockdown of Akt2 (Fig. 5D) but not by the knockdown of Akt1 or Akt3 (Fig. 5E). In LNCaP and 22Rv1 cells, the knockdown of SMAD4 using two different siRNAs led to a signi cant increase in chemotaxis (Fig. 5F). Treatment with AKT2i induced SMAD4 expression in LNCaP (Figs. 5G&H) but not in LNCaP[PTEN] cells (Fig. 5H), and this correlated with decreased abundance of po-AKT substrates but not total AKT1 or AKT2 (Fig. 5H), con rming the e cacy of AKT2i. In contrast, concentrations of AKT1i or p110αi that did not inhibit LNCaP 2D proliferation or survival ( Supplementary Figs. S2E & S3A) also failed to decrease LNCaP chemotaxis (Fig. 4C), strengthening the role of AKT2 in controlling chemotactic motility of PTEN-negative PC cells.
We next compared how PTEN re-expression affected AKT signaling in 2D vs. 3D growth. This is because we previously showed that activated Src had a maximal ability to activate AKT in 3D growth conditions [75], and because the effect of AKT2 on LNCaP survival was manifest in 3D, but not in 2D growth [38].
Interestingly, the re-expression of PTEN did not signi cantly reduce activated AKT (relative AKT poSer473 or AKT2 poSer474 levels, the latter using an Ab speci c for activated AKT2) in T402 grown in 2D (Fig. 6A). In contrast, PTEN-mediated reduction in relative AKT poT308 levels were observed in cells grown in 3D (suspension in methylcellulose) (Fig. 6B). We then analyzed how PTEN affected the ability of serum to induce AKT activation in 2D vs. 3D conditions. While the overnight growth in 3D with serum had minimal to no ability to activate AKT, a 30 min treatment of FBS ("3D + FBS Stim.") to serum-starved ("3D + 0.5%FBS") LNCaP or T402 cells induced more relative AKT poSer473/474 and AKT poT308 in PTEN-negative cells than in PTEN re-expressing cells (Figs. 6C&D). We next determined if PTEN controlled survival under anoikis conditions through a greater dependence on AKT1 or AKT2. LNCaP and LNCaP[PTEN] cells were grown in non-adherent conditions (48 h on agarose-coated plates) while being treated with DMSO, AKT1i or AKT2i, followed by quanti cation of cell viability. LNCaP viability was more dependent on AKT2, whereas viability of LNCaP[PTEN] cells was more dependent on AKT1 (Fig. 6E). Taken together, these data strongly suggest that PTEN suppression of AKT activation is potentiated under 3D conditions, exempli ed here by increased survival under anchorage-independent conditions but also by other 3D conditions shown earlier such as invasiveness and chemotaxis.
Therapeutic targeting of PTEN-negative PC requires combining PI3K-p110β and AKT2 inhibitors. We addressed how targeting p110 and/or AKT isoforms affected the progression of primary orthotopic tumors and the establishment of spontaneous metastases in SCID male mice. Consistent with PTEN's tumor suppressor function, the orthotopic injection of T402[PTEN] cells (prostatic anterior lobe) failed to yield growing tumors after 80 days (Fig. 6F). Individual inhibitors for p110α, p110β, AKT1 or AKT2, or for the p110α/AKT1 combination, slightly decreased tumor growth from days 12-18 of drug treatment in comparison to vehicle controls, however, none of these translated to statistically signi cant effects at day 35. In contrast, treatment with the combination of p110β/AKT2 inhibitors showed statistically signi cant tumor suppression over controls. Moreover, the p110β/AKT2 inhibitor combination resulted in a statistically signi cant decrease in the metastatic colonization by LNCaP-C4-2B[luc/GFP] cells, as assessed by Alu-speci c qPCR as described previously [52] (Fig. 6G).

DISCUSSION
The high frequency of PTEN loss in primary PC as well as transgenic prostate models showing that Pten loss is su cient to induce a tumor-prone state for PC initiation and progression has been premised on the assumption that this translates to activation of oncogenic PI3K-AKT signaling [21,57,76]. However, our comparison of two Tg PC models, Pten/Rb-null vs. Akap12/Rb-null, which share similar levels of Akt activation plus Rb loss, yet which have different outcomes in regard to primary PC and mPC progression, strongly suggests that "AKT activation" alone is insu cient to explain these differences. Based on growing evidence showing that the three AKT isoforms play differing roles in the progression of some cancers, including PC [38,39], the differing disease phenotypes in our two transgenic models could be explained by the differential activation and preferential dependence on speci c AKT isoforms. Although previous studies compared the possible roles for AKT isoforms after AR re-expression in PC-3 cells [77] or by comparing AR-positive vs. -negative human PC cell lines [78], ours is the rst study to analyze the effect of PTEN re-expression or knockdown in isogenic cell lines. Our data strongly suggest that the loss of PTEN induces mPC through a PI3K-p110β/AKT2-mediated pathway. In contrast, the loss of AKAP12, which normally scaffolds Src and attenuates Src-induced PI3K/AKT activation, correlates with weaker oncogenic progression signaling through p110α and AKT1 (Fig. 6H).
We show evidence suggesting that PC is marked by at least two relatively non-overlapping driver pathways, both of which impact PI3K/AKT signaling: PTEN loss and Src-family kinase activation. This divergence is underlined by TGCA Firehose data showing that the roughly a third of primary PC cases exhibiting gene ampli cations in the SFK genes, SRC, LYN and FYN, have little overlap with those suffering PTEN deletions (Supplemental Table S1). Indeed, SFK mutations known to induce oncogenic activation are extremely rare in PC (0/491 cases in TCGA Firehose), and thus, our development of the Akap12/Rb-null transgenic PC model was meant to genocopy the oncogenic activation of SFK, given AKAP12's role as a Src scaffolding protein [13] and our nding that AKAP12 loss and LYN gain co-occur in TCGA datasets.
Our data indicate that Pten/Rb-null PC lesions express higher protein and activation levels of Akt2 than HG-PIN lesions from Akap12/Rb-null mice. Although higher Akt3 protein levels were detected in Pten/Rbnull PC lesions, this increase was not manifest at the RNA level. Importantly, 2D proliferation, Matrigel invasiveness, chemotaxis, clonogenic survival and anchorage-free (anoikis) survival in human and mouse PTEN-negative PC cell lines were more dependent on AKT2 or AKT3 (the latter only in the case of LNCaP invasiveness) than on AKT1, consistent with a previous report [38] showing that 3D spheroid growth of PTEN-negative PC cells relies more on AKT2. However, our data show that AKT2 inhibition on multiple measures of in vitro and in vivo metastatic growth/motility of PTEN-de cient PC cells is potentiated by co-inhibiting p110β. In contrast, PTEN re-expression, although decreasing these measures of metastasis, reversed dependency to p110α plus AKT1.
The synergistic effect of co-inhibiting p110 and AKT isoforms may re ect incomplete inhibition of each part of a linear PI3K-AKT pathways, but it is also consistent with previous studies showing AKTindependent functions for PI3K [79] [90] studies show therapeutic e cacy in targeting of AKT or co-targeting AKT and the androgen axis/androgen receptor in castration-recurrent PC, yet ours is the rst to use isogenic PC cell lines to address how PTEN and AKT isoform affect metastatic signaling and progression. Importantly, our data not only validate the notion of compensatory plasticity between p110 and AKT isoforms, as described previously [33,34], they identify a role for PTEN in controlling which p110-AKT isoform pairs serve as drivers. More speci cally, in the context of Rb loss, p110β/AKT2 seem to drive aggressive mPC progression whereas p110α/AKT1 drive a more indolent mPC disease. Indeed, Crkl, which seems to drive the activation of p110β in PTEN-de cient cancer cells [37], was signi cantly increased in PC tumors from Pten/Rb-null vs. Akap12/Rb-null mice. One possibility for the different mPC disease outcomes might be that these pathways have different effects on AR signaling, based on a report of increased AR activation through increased AKT signaling in PTEN-de cient PC tumors [91]. However, PTEN expression did not affect Ar protein levels or the expression levels of 19 Ar-regulated genes in T402 cells.
In sum, our data strongly argue that mPC progression pathways can be therapeutically targeted using the appropriate combination of p110 and AKT isoform inhibitors based on foreknowledge of PTEN, SFK and AKAP12 genomic status.

COMPETING INTERESTS
None.
DATA AVAILABILITY STATEMENT Materials and reagents described in the manuscript, including all relevant raw data, will be freely available to any researcher wishing to use them for non-commercial purposes, without breaching participant con dentiality.