Spectrum of activity and molecular correlates of response to phosphatidylinositol ether lipid analogues, novel lipid-based inhibitors of Akt Free

The serine/threonine kinase Akt is a promising target in cancer. We previously identified five phosphatidylinositol ether lipid analogues (PIA) that inhibited Akt activation and selectively killed lung and breast cancer cells with high levels of Akt activity. To assess the spectrum of activity in other cell types and to compare PIAs with other inhibitors of the phosphatidylinositol 3-kinase (PI3K)/Akt/mammalian target of rapamycin (mTOR) pathway, we compared growth inhibition by PIAs against the PI3K inhibitors LY294002 and wortmannin and the mTOR inhibitor rapamycin in the NCI60 cell line panel. Although each of these compounds inhibited the growth of all the cell lines, distinct patterns were observed. The PIAs were the least potent but the most cytotoxic. The broad spectrum of activity of PIAs was confirmed in vivo in hollow fiber assays. The response to PIAs was significantly correlated with levels of active but not total Akt in the NCI60, as assessed using COMPARE analysis. However, a number of molecular targets were identified whose expression was more highly correlated with sensitivity to PIAs than active Akt. Expression of these molecular targets did not overlap with those that correlated with sensitivity to LY294002, wortmannin, or rapamycin. A COMPARE analysis of the National Cancer Institute chemical screening database revealed that the patterns of activity of PIAs correlated best with patterns of activity of other lipid-based compounds. These studies show that although PIAs are widely active in cancer cells, which correlates with the presence of its intended target, active Akt, PIAs are biologically distinct from other known inhibitors of the PI3K/Akt/mTOR pathway. [Mol Cancer Ther 2006;5(3):713–22]

Akt is a serine/threonine kinase that controls cellular growth, migration, metabolism,⁵

and survival. Akt is an attractive therapeutic target in cancer because it contributes to tumorigenesis and therapeutic resistance. Phosphatidylinositol ether lipid analogues (PIA) are analogues of the products of phosphatidylinositol 3-kinase (PI3K) that were designed to target the pleckstrin homology domain of Akt (1). PIAs inhibit Akt translocation, phosphorylation, and kinase activity and preferentially induce apoptosis in breast and lung cancer cell lines with high levels of active Akt (2). A second Akt-independent activity of PIAs has recently been identified: activation of the proapoptotic stress kinase p38α⁵. This activation is both direct and indirect and is likely based on structural similarities between the pleckstrin homology domain of Akt and p38α. Although p38α is not required for PIA-induced apoptosis, p38α activation does contribute to PIA-induced toxicity, stressing the importance of “off-target” effects that can underlie the efficacy of cancer drugs.

The development of PIAs is one example of the many efforts in academia and industry to develop inhibitors of the PI3K/Akt/mammalian target of rapamycin (mTOR) pathway. Because many components in the pathway have been targeted, including PI3K, PDK-1, Akt, and mTOR, comparison studies will be required to reveal which inhibitors are truly targeted and which are most effective. In that vein, we assessed the spectrum of activity of PIAs in vitro and in vivo and compared responses to PIAs in the NCI60 cell line panel with that of two PI3K inhibitors (LY294002 and wortmannin) and an mTOR inhibitor (rapamycin). We show that cellular responses to PIAs indeed correlate with levels of active Akt. Using COMPARE analysis, we identify novel molecular targets that are highly correlated with response to PIAs and identify structurally related compounds that have similar patterns of activity.

Methods for evaluation of cell growth inhibition in the NCI60 cell line panel were published previously (3, 4). Briefly, PIAs were solubilized in DMSO, diluted into RPMI 1640 + 5% fetal bovine serum, and added to 96-well plates containing cell lines that were previously cultured for 24 hours. After a 48-hour incubation with the PIAs, the media were removed, and the cells were fixed and stained with sulforhodamine B. Unbound dye was removed with five washes of 1% acetic acid, and the plates were allowed to air dry. The dye was resolubilized in Tris buffer, and the absorbance at 515 nm was measured. The concentration that produced 50% growth inhibition compared with a DMSO control (GI₅₀), total growth inhibition, or 0% growth, compared with a DMSO control (TGI), and the concentration that produced death of 50% of the cells present at the start of the experiment (LC₅₀) were determined.

Cells were grown inside polyvinylidene difluoride fibers and inserted into the i.p. and s.c. compartments of nude mice as described previously (5). PIA23 was dissolved in 10% DMSO in saline and given i.p. every day for 4 days, with dosing beginning on the third day after fiber implantation at doses of 25 and 37.5 mg/kg.

Levels of phosphorylated Akt (phospho-Akt) at S473 and T308, as well as total Akt protein were measured previously in the NCI60 cell line panel by immunoblotting, and these data are available through the Developmental Therapeutics Program web site.⁶

The COMPARE algorithm (6) was used to determine Pearson correlation coefficients between the −log(GI₅₀) of the PIAs and the measured levels of total Akt and phospho-Akt in the cell lines.

To date the Developmental Therapeutics Program has assembled over 220,000 molecular target measurements in the NCI60 cell line panel. The COMPARE algorithm and the JMP statistical software package (SAS Institute, Cary, NC) was used to determine Pearson correlation coefficients between the −log(GI₅₀) of the PIAs and the levels of molecular targets that have been measured in the NCI60 cell line panel. Given the uncertainty in an individual measurement from a single microarray, we first identified a subset of genes whose measurement on multiple microarray platforms were well correlated (Pearson correlation coefficient ≥0.5). This was done using microarray data sets available through the Developmental Therapeutics Program web site and included four arrays: a cDNA array (7, 8), an Affymetrix HUM6000 array (from Millenium Pharmaceuticals, Cambridge, MA), an Affymetrix U95A array done in triplicate (Novartis, Cambridge, MA), and an Affymetrix U95A-E array (Gene Logic, Gaithersburg, MD). Using this subset of 5,796 genes (∼20,000 total measurements), we did correlation analyses with the growth inhibition patterns of these compounds. The results were further filtered to include only those genes where measurements on multiple microarrays gave significant correlations with growth inhibition by the PIAs.

As of 2005, the National Cancer Institute has evaluated compounds for anticancer activity, with publicly available data for ∼43,000 compounds. The COMPARE algorithm was used to search for compounds that have been tested in the NCI60 cell line panel that have a similar sensitivity profile to the PIAs.

To assess the spectrum of cancer cell types that are sensitive to PIAs, we screened PIA5, PIA6, PIA23, PIA24, and PIA25 (NSC726850, NSC726851, NSC726852, NSC726853, and NSC726854) in the NCI60 cell line panel. All PIAs were effective in inhibiting the growth of all of the cell lines, with the GI₅₀s for most falling between 1 and 20 μmol/L. The composite dose-response curves were similar for all five PIAs (Fig. 1, left and middle). For the majority of the cell lines, PIAs begin to inhibit growth at 1 μmol/L. Above 10 μmol/L, the growth of all cell lines is sharply inhibited. PIAs are uniformly cytotoxic at 100 μmol/L. To determine if this was characteristic of other inhibitors of the pathway, we compared the sensitivity of the PIAs that inhibit Akt with three other inhibitors of the pathway: LY294002 (NSC697286) and wortmannin (NSC221019), both of which are PI3K inhibitors, and rapamycin (NSC226080), which is an mTOR inhibitor (Fig. 1, right). LY294002 and wortmannin have more gradually shaped dose-response growth curves than the PIAs. Rapamycin, unlike PIAs, LY294002, or wortmannin, is a very potent cytostatic agent (10 nmol/L to 10 μmol/L). At doses between 10 and 100 μmol/L, however, rapamycin is cytotoxic. These studies show that PIAs are less potent than the PI3K inhibitors or rapamycin. At the highest doses tested, PIAs are widely cytotoxic like rapamycin and more cytotoxic than LY294002 or wortmannin.

To determine if differences could be discerned between individual PIAs, we compared the efficacy of each PIA in each cell line (Fig. 2). Overall, the mean graphs of the GI₅₀ concentrations of the five PIAs are similar, except that PIA24 seemed less efficacious. (The TGI and LC₅₀ data for PIAs are shown in Supplementary Figs. S1 and S2, respectively.)⁷

The cell lines RPMI-8226, HOP-92, NCI-H460, U251, TK-10, and PC3 cells were uniformly inhibited by PIAs. The most sensitive cell line to all PIAs was the prostate cancer cell line PC-3. For the other pathway inhibitors, the renal carcinoma cell line CAKI-1 was the most sensitive to wortmannin, the melanoma cell line MALME-3M was most sensitive to LY294002, and the central nervous system cell line SF-295 and the leukemia cell line SR were the most sensitive to rapamycin (Supplementary Figs. S3-S5).⁷ Thus, inhibitors of the PI3K/Akt/mTOR pathway decrease cellular proliferation and viability with cell line specificity.

To assess the spectrum of activity of PIAs in vivo, we assayed an active PIA (PIA23) in hollow fiber assays that use 12 of the 60 cell lines in the NCI60. PIA23 was effective in inhibiting the growth of 8 of 12 of the cell lines grown inside hollow fibers placed in either the i.p. or the s.c. compartments, and cytotoxicity was noted (Table 1). PIA23 inhibited growth in an equal number of fibers located in the s.c. and i.p. compartments. This is notable because PIA23 was given by i.p. injection, which shows that PIA23 was able to reach effective circulating concentrations.

We then compared responsiveness in the hollow fiber assays with levels of Akt activation. The most sensitive cell line in the assay was the cell line MDA-MB-435, formerly believed to be a breast cancer cell line but recently was shown to be virtually identical to the melanoma line M14 (9). MDA-MB-435 cells have been previously shown to contain a high level of endogenous Akt activity (10). Conversely, MDA-MB-231 cells, which did not respond to PIA23, have low levels of active Akt and do not respond to LY294002 (10). The correlation with active Akt was not absolute, because the cell line with the most phospho-Akt on the panel, U251, did not respond to PIA23, whereas a cell line with a relatively low amount of phospho-Akt, NCI-H522, did respond. This shows that response to PIA treatment is not solely dependent on Akt and may rely on expression of other targets.

To further evaluate whether the cytotoxicity of the PIAs is related to modulation of endogenous Akt activity, which was suggested by our earlier studies, the sensitivity to PIAs (GI₅₀) was compared with the levels of active Akt (as defined by phosphorylation of S473 or T308), and total Akt in the NCI60 cell line panel (Table 2). The Pearson correlation coefficients (PCC) that compare response to PIAs with S473 phosphorylation were marginally statistically significant for all PIAs (PCC ≥0.28). The PCC for T308 was marginally statistically significant for all but PIA24. In contrast, the PCCs for total Akt were below zero. This confirms that sensitivity of the PIAs correlates positively with the presence of active Akt but not total Akt.

To determine whether expression of other molecular targets might correlate with sensitivity to PIAs, we did analyses using a subset of the public microarray data. Multiple molecular targets were identified whose expression correlated positively with response to PIAs, indicating that cell lines with higher levels of expression of these genes tended to be more sensitive to the compounds. The 10 highest scoring genes with PCCs ≥ 0.4 are rank listed in Table 3. The highest correlations for each PIA were higher than that observed with active Akt. The highest PCC for any target was 0.67 for the association between the −log(GI₅₀) of PIA24 and syntaxin 1A. Similar analysis was done for molecular targets whose expression negatively correlated with sensitivity to PIAs, and multiple molecular targets were identified whose PCCs were less than or equal to −0.4 (Table 4). Overall, there were more positively correlated targets that were identified than negatively correlated ones. For example, the only negative correlating gene for PIA24 was STAT3. Although no overarching categorization was obvious, several genes that regulate chromatin (HDAC1-positively correlated, SMARCA3-negatively correlated) and transcription (KLF4, NSEP1, c-myc, SOX9, and ETS2-positively correlated, AP-2α-negatively correlated) were identified.

Because the active PIAs are structurally related and kill cancer cells similarly, we sought to identify common targets that correlated with response to PIAs. We compared molecular targets that had PCC ≥ 0.4 and identified those that were common to three or more PIAs (Table 5). Fourteen targets positively correlated with sensitivity to multiple PIAs. Three of 14 positively correlated targets (mesotrypsin, c-myc, and guanylate kinase 1) were common across all five PIAs. Although there were no negatively correlated targets that were in common with PIA24, five targets were common across the other four PIAs (cyclin E, ErbB3, palmitoyl-protein thioesterase 2, preferentially expressed antigen in melanoma, and myosin heavy polypeptide 10; Table 6). Of the group of five PIAs, PIA5, PIA6, and PIA25 shared the most targets in common, whereas PIA24 had the least number of targets in common with any of the other PIAs. These data show that other targets correlate better with responsiveness than active Akt and suggest that PIA24 has different markers for responsiveness than other PIAs.

Because of the apparent distinction between PIAs and other inhibitors of the PI3K/Akt/mTOR pathway, we used COMPARE to search the NCI60 database to determine if other compounds have shown similar sensitivity profiles in the NCI60 cell line panel. When PIAs were compared against each other, PIA5, PIA6, PIA23, and PIA25 exhibited high correlation coefficients (0.74 < PCC < 0.91), indicating their similarity (data not shown). When PIA24 was compared against the other PIAs, the highest PCC was 0.67 for the correlation with PIA6 (data not shown). When PIA5, PIA6, PIA23, or PIA25 was used as a seed against the entire database, structurally similar compounds were identified (Fig. 3A-E). The compound whose GI₅₀ pattern showed the most similarity to the PIAs was NSC643826. This had a coefficient of 0.78 compared with PIA5, 0.70 with PIA6, 0.69 with PIA23, and 0.76 with PIA25. This and many of the other top matching compounds are phospholipids, like the PIAs. Some of these compounds have an amino group attaching the phosphate group with the lipid side chain, whereas the PIAs have an ether linkage attaching the side chain. Unlike the PIAs, which have an inositol head group, all of the phospholipids that were identified contained nitrogen in the head group (many within a choline moiety). Two of the PIAs (PIA5 and PIA6) had similarity to the alkylphospholipids (miltefosine or perifosine) in their top matches. PIA24 did not show similarity to any phospholipids in its top five matches (Fig. 3D). Although PIA24 correlated with two structures that had hydrocarbon tails, these did not contain a phosphate group. Overall, probing the database with PIA24 yielded the most diverse structures. To confirm the validity of this approach, we compared the activity profiles of LY294002, wortmannin, and rapamycin and obtained values of (0.31 < PCC < 0.49) for all combinations, indicating that this type of analysis is capable of associating inhibitors that target the same pathway. These studies show that PIAs have the greatest similarity to other lipid-based molecules and validate the biological distinction between PIAs and other inhibitors of the pathway, such as LY294002, wortmannin, or rapamycin.

Our studies indicate that PIAs can kill a variety of cancer cells in a manner that is dependent upon Akt activity yet is distinct from other inhibitors of the pathway. In the NCI60 cell line panel, PIAs were able to inhibit the growth of all the cell lines within the dose range tested (up to 100 μmol/L), with the GI₅₀ for the majority of the cell lines falling between 1 and 20 μmol/L. Although broadly effective, these compounds showed only moderate potency. A possible reason for this might be the affinity of the lipid side chain of PIAs for serum, a phenomenon that has been observed previously (2). Despite this moderate potency, activity in 8 of 12 cell lines in hollow fiber assays was observed, indicating that effective circulating levels of PIAs can be achieved.

How do PIAs compare to other inhibitors of the pathway, such as LY294002 and rapamycin? The PI3K inhibitors LY294002 and wortmannin caused growth inhibition in a similar dose range as the PIAs; however, PIAs caused more cytotoxicity than LY294002 or wortmannin at the higher doses. Although the sensitivity patterns of rapamycin, LY294002, and wortmannin correlated with each other, a COMPARE analysis of the PIAs with these other inhibitors did not yield any significant correlation. This indicates that the pattern of growth inhibition in the different cell types to PIAs versus these other agents is dissimilar. Therefore, one might predict that clinical activity and/or toxicities of these compounds might also be dissimilar. Clinical trials that assess tolerability and efficacy will determine which inhibitors are the most promising drugs.

Establishing the relationship between target modulation by an agent and anticancer effects is an important step in targeted drug development. Sensitivity to PIAs weakly correlated with levels of phospho-Akt but not total Akt in the NCI60 cell line panel. However, we also found a number of other targets whose expression was more highly correlated to PIA sensitivity than active Akt. Several of these were common across three or more PIAs. It is unknown whether PIAs modulate any of these targets directly, but these targets that positively or negatively correlate with sensitivity to PIAs could potentially serve as predictive biomarkers for responsiveness to PIAs.

The COMPARE algorithm was also used to identify compounds that have been screened in the NCI60 that have similar sensitivity profiles to the PIAs. This yielded phospholipids that differed in the head group and linkage to the lipid side chain, as well as length of the side chain. One group of lipid-based agents that was identified (alkylphospholipids) has diverse functions and is currently under development as anticancer therapies. Interestingly, the alkylphospholipids that correlated with PIA5 and PIA6 (miltefosine and perifosine) also inhibit the translocation and activation of Akt (11, 12).

These studies show a wide spectrum of activity for PIAs, provide evidence of in vivo activity, validate active Akt as a predictive factor for PIAs, and identify new molecular targets that are associated with response to PIAs. Whether levels of expression of these novel targets contribute to the cytotoxic effects of PIAs, or whether they may serve as useful predictors of PIA sensitivity, will be the subjects of future research. Regardless, the wide spectrum of activity, the in vivo efficacy, and the novel combination of biological activities of PIAs that includes inhibition of Akt and activation of p38α highlight the potential of these compounds as cancer therapeutics.

We thank the members of the Dennis lab for helpful discussions.