The WW domain of IQGAP1 binds directly to the p110α catalytic subunit of PI 3-kinase

Recently there has been renewed enthusiasm for targeting protein–protein interactions as a therapeutic strategy for treating cancer and other diseases, and for using peptide-based inhibitors as drugs or drug leads in such approaches [1–11]. Signaling scaffold proteins — large, multidomain proteins that bind to multiple members of one or more signaling pathways — participate in numerous protein–protein interactions that could be targeted to disrupt signaling [7,12–16]. Indeed, studies with human IQGAP1, a scaffold protein for multiple cancer-associated signaling pathways [17–21], have led to the discovery that its 26-residue WW domain, when engineered as a cell-penetrating peptide, shows broad anti-tumor activity with minimal associated toxicity [12,22].

WW domains are one of the smallest independently folding protein interaction domains, and thus have a strong potential as leads in peptidomimetic development. These domains are typically between 25 and 35 residues in length and fold into a three-stranded antiparallel beta-sheet [23–25]. WW domains are characterized as proline recognition domains, and typically bind to proline-rich short linear motifs in their binding partners [26]. A cell-penetrating derivative of the IQGAP1 WW domain was first developed by Jameson et al. [12], who showed that this peptide inhibited the proliferation, migration, viability and tumor-forming ability of breast, colorectal and melanoma tumor cells that contained activating mutations in the RAS/MAPK pathway. Non-transformed or primary cells were unaffected. They further showed that the WW peptide, when delivered systemically, significantly increased the lifespan of mice bearing pancreatic tumors [12]. Choi et al. [22] showed that the WW peptide reduced the viability of multiple additional breast cancer cell lines, but did not adversely affect normal control lines. The anti-tumor activity of the WW peptide has also been confirmed in other studies [27,28].

In view of these exciting results, it is imperative to determine the binding partner(s) of the WW domain. For many years, it was believed that the ERK1 and ERK2 MAP kinases bound to the WW domain of IQGAP1, and were the only proteins to do so [29–31]. Indeed, this belief motivated the Jameson et al. study, as these workers sought to target ERK signaling by using the WW domain as a competitive inhibitor of ERK-IQGAP1 binding. In 2017, however, we showed that ERK1 and ERK2 do not bind to IQGAP1's WW domain, but instead bind to the nearby IQ domain [32]. This finding revived the search for the authentic target of the IQGAP1 WW domain.

A significant clue in this regard came from a landmark 2016 study from the Anderson laboratory [22]. These authors showed that IQGAP1 acts a scaffold protein for the phosphoinositide 3-kinase (PI3K) pathway by binding to several enzymes that modify phosphoinositides, including PI3K itself. PI3Ks are a family of enzymes that phosphorylate inositol lipids in response to signals from growth factors, cytokines, and other cellular and environmental cues. In so doing, they regulate cell proliferation, growth, survival, motility, intracellular trafficking, and metabolism [33–37]. The most common PI3K isoform is a heterodimer of a catalytic subunit designated p110α (gene PIK3CA) and a regulatory subunit designated p85α (gene PIK3R1). Following activation by growth factors, insulin, or other signals, the p110α/p85α holoenzyme phosphorylates the inositol ring of phosphatidylinositol-(4,5)-biphosphate (PIP2) on the 3-position to generate phosphatidylinositol-(3,4,5)-triphosphate (PIP3), a key intracellular second messenger. Both subunits of PI3K, particularly p110α, are recognized oncogenic drivers that are frequently mutated in human cancer and tissue overgrowth syndromes. In addition to being activated by direct mutation, PI3K activity can also be pathologically activated by other means, including mutations in other oncogenes, loss of tumor suppressor genes (especially the PTEN tumor suppressor), and oncovirus-mediated transformation [33–37]. Hence, PI3K activation is a hallmark of cancer [35].

Here, we show that the WW domain of human IGQAP1 binds directly to human p110α protein, the catalytic subunit of the PI3K holoenzyme, but does not bind to p85α, ERK1, ERK2, MEK or MEK2. We also show that the anti-tumor WW peptide has these same binding specificities. In addition, we present a model of this WW domain, and investigate which residues in the domain participate in the interaction with p110α. These findings contribute to a more precise understanding of IQGAP1-mediated scaffolding, and of how IQGAP1-derived therapeutic peptides might inhibit tumorigenesis.

Although IQGAP1 had been shown to act as a scaffold protein for the PI 3-kinase pathway [22] and the RAS/RAF/MEK/ERK pathway [29,32], the precise binding partner of IQGAP1's WW domain was unknown. We took a ‘candidate protein' approach to finding this partner, using components from both of these pathways. First, the WW domain of human IQGAP1 (residues 684–710, Figure 1A) was fused at its N-terminus to Schistosoma japonicum glutathione S-transferase (GST), and the resulting fusion protein (GST-WW) was expressed in bacteria and purified by adsorption to glutathione-Sepharose beads. Next, GST-WW (or GST alone as a negative control) was incubated with full-length human p85α or p110α that had been produced in radiolabeled form by in vitro translation (Figure 1B). We also assessed the ability of GST-WW to bind to in vitro translated human ERK2 protein (Figure 1B). Bead-bound complexes were collected by sedimentation, washed extensively, and analyzed by SDS–PAGE and autoradiography. As shown in Figure 1C, full-length p110α bound efficiently to WW, but p85α and ERK2 did not. Furthermore, the binding of p110α was specific, because only trace precipitation of p110α occurred when GST was used instead of the GST-WW fusion protein. As an additional control, we tested two other small GST fusion protein fragments for binding to p110α, neither of which bound above the background (Supplementary Figure S1).

Initially, we used two different variants of full-length human p110α protein. The first version contained an N-terminal myc-epitope tag and a C-terminal CAAX box [38], and the second version was the unaltered 1086-residue wild-type protein. Quantification of 12–21 independent experiments for each variant is shown in Figure 1D, and demonstrates that the presence or absence of the myc and CAAX tags did not affect the binding of GST-WW to p110α. Furthermore, for both variants, their binding to GST-WW was statistically significantly different from their binding to GST alone, with a very high level of confidence (P < 0.0001). The unaltered wild-type variant was used in the experiments shown in the remainder of the manuscript. Figure 1D also shows that for p85 and ERK2, both of which were assessed repeatedly over multiple different experiments, the trace binding to GST-WW was indistinguishable from the level of binding to GST alone (P > 0.99 for p85α, P > 0.40 for ERK2).

In addition to not binding to ERK2 (Figure 1), GST-WW also did not exhibit detectable binding to human ERK1, MEK1 or MEK2 (Figure 2A,B). These negative results with RAS/MAPK pathway components confirmed our previously published data that used a different experimental approach to show that ERK1, ERK2, MEK1 and MEK2 all bind to the IQ domain of IQGAP1, and not to the WW domain [32]. In addition, these results demonstrated that our GST-WW construct was not non-specifically sticky.

The WW domains of some proteins bind to other regions of the same protein; these in-trans or in-cis interactions mediate homodimerization or autoinhibition [39,40]. Hence, we also asked whether GST-WW was able to bind to full-length IQGAP1 protein, or to IQGAP1 truncations lacking the WW domain. We found no evidence for such self-interaction (Figure 2C,D). However, it should be noted that our binding assay would not be able to detect weak, millimolar-affinity interactions that might nevertheless be significant if driven by local concentration effects.

Preliminary domain-mapping experiments indicated that the N-terminal adapter-binding domain (ABD) of p110α was sufficient for binding to the WW domain of IQGAP1 (data not shown; these data will be published in a forthcoming study). To ask if full-length IQGAP1 protein could bind to p110α, we expressed two different fusions of the ABD to GST, and purified them from bacteria. GST-p110α(6–123) encodes the core of the ABD, whereas GST-p110α(1–190) includes N-terminal residues (residues 1–5) that are not visible in the crystal structure, as well as C-terminal residues (residues 124–190) that the link the ADB to the Ras-binding domain [41]. As shown in Figure 2E, both of these fusion proteins bound to full-length IQGAP1 protein that had been produced by in vitro translation. Hence, the WW domain of IQGAP1, purified from bacteria, binds to full-length p110α (Figures 1 and 2A,B), and the ABD of p110α, purified from bacteria, binds to full-length IQGAP1 (Figure 2E).

To summarize, the experiments described in this section demonstrate that the WW domain of human IQGAP1 binds directly and specifically to the catalytic subunit of human PI 3-kinase.

As mentioned in the Introduction to this paper, a cell-penetrating IQGAP1 WW peptide has shown an ability to selectively kill various neoplastically transformed cells and primary clinical specimens in culture and to inhibit tumorigenesis and prolong survival in mouse models [12,22,27,28]. As originally designed by Jameson et al. [12], this peptide contains, in order from N-terminus to C-terminus, eight D-arginine residues to promote cell penetration, a 10-residue myc-epitope tag to facilitate detection by immunostaining, and residues 680–711 of human IQGAP1. Some workers have raised the concern that the addition of the polyarginine or epitope sequences may have altered the binding specificity of this peptide [42]. To begin to explore this issue, we synthesized the identical peptide, and analyzed its in vitro binding specificity (Figure 3). As seen with GST-WW purified from bacteria (Figures 1 and 2), the synthetic anti-tumor WW peptide bound to p110α, but did not exhibit detectable binding to p85α, ERK1 or ERK2 (Figure 3A,B). Thus, the poly-d-arginine and myc-epitope sequences did not alter the binding specificity of the WW domain with the ligands tested. In particular, they did not endow the WW domain with an ability to bind to ERK1 or ERK2.

We also tested the ability of the WW peptide to act as a competitive inhibitor of p110α binding. As shown in Figures 3C and 4D, the WW peptide was indeed able to effectively compete with GST-WW for binding to p110α.

In cells, the majority of p110α molecules are bound to p85α in the form of p110α/p85α heterodimers, although p85α also has independent functions [43]. To ask if the WW domain of IQGAP1 could bind to p110α/p85α heterodimers, we co-translated p110α and p85α in the same in vitro translation reaction, and then assessed the binding of this co-translated mix to GST-WW; p110α translated alone and p85α translated alone were included for comparison (Figure 4A). As shown in Figure 1 and in Figure 4B, p85α alone did not display detectable binding to GST-WW, whereas p110α bound efficiently. Nevertheless, p85α co-translated with p110α bound to GST-WW (Figure 4B, see also Supplementary Figure S2). Indeed, the results, when quantified and corrected for labeling efficiency, indicated that there was an ∼1 : 1 molar ratio of p85α to p110α in the bound fraction (see Figure 4B legend and Supplementary Figure S2 for quantification). This experiment demonstrates that the WW domain of IQGAP1 is able to bind to p110α/p85α heterodimers as well as to p110α monomers.

In the p110α/p85α heterodimer, the catalytic activity of p110α is repressed by p85α, but this repression is relieved when the two SH2 domains in p85α bind to phosphotyrosine residues on receptor tyrosine kinases or adaptor proteins (see Figure 4C for the domain structure of p85α). This repression can also be relieved by mutations in p110α or p85α [44,45]. For p85α (gene = PIK3R1), such mutations cluster in the iSH2 domain, and are often nonsense or frameshift mutations that result in the expression of a shortened version designated p65α [46–48]. To determine if the WW domain of IQGAP1 could bind to the mutationally activated p110α/p65α heterodimer, we co-translated p110α with a truncated derivative of PIK3R1 encoding p65α (residues 1–571, Figure 4C); p110α translated alone and p65α translated alone were also included for comparison. As shown in Figure 4D, p65α alone did not display detectable binding to GST-WW, whereas p110α once again bound efficiently. However, similar to the results obtained with p85α, p65α co-translated with p110α bound to GST-WW, with an ∼1 : 1 molar ratio of p65α to p110α in the bound fraction (Figure 4D, see also Supplementary Figure S2).

To further demonstrate that p110α/p85α heterodimer can bind to the WW domain of IQGAP1, we obtained recombinant PI3K holoenzyme purified from insect cells, and assessed binding to GST-WW purified from bacteria (Figure 4E). As shown in Figure 4F, both the p110α and p85α subunits of PI3K co-sedimented with GST-WW, confirming that the purified holoenzyme does indeed bind to GST-WW. In addition, this experiment rules out the possibility that other macromolecules (such as other proteins, DNA, or RNAs) present in the rabbit reticulocyte lysate used to produce radiolabeled proteins for this study are necessary for the WW-p110α interaction.

To summarize, the results in this section show that the WW domain of IQGAP1 is able to bind directly to both the unactivated p110α/p85α heterodimer and the mutationally activated p110α/p65α heterodimer.

To determine if the WW-binding domain of IQGAP1 was sufficient to associate with PI3K in cells, a plasmid expressing GST-WW was transiently transfected into the human HEK293 cell line, cell extracts were prepared, and GST-WW protein and associated proteins were isolated by sedimentation. As shown in Figure 5, both p110α and p85α were co-sedimented with GST-WW, but not with the GST alone control (Figure 5, lower panels, lanes 1 and 2).

In the same experiment, we expressed two additional controls. The first of these was GST-AARAA, a mutant variant of IQGAP1's WW domain that contains multiple substitutions in key conserved residues; as will be shown in Figure 7, this variant does not bind to p110α in vitro. This mutant also failed to associate with p110α in cells (Figure 5, lane 3).

The second control we expressed was GST-p85α, which we anticipated would serve as a positive control for p110α binding. This protein was very weakly expressed (data not shown), yet its presence was evident by its ability to pull-down endogenous p110α from the extract (Figure 5 lane 4, lower panels). Endogenous p85α also co-sedimented with GST-p85α; this is presumably attributable to the ability of p85α to homodimerize [49,50]. To summarize, the WW domain of IQGAP1 specifically associated with p110α/p85α in HEK293 cells.

WW domains are found in plants, fungi, protists, and animals. An alignment of the WW domain of IQGAP1 with 93 other well-recognized human WW domains, a tree, sequence logos, and other analyses are presented in Supplementary Figures S3–6. These analyses revealed that the WW domains closest in sequence space to IQGAP1 include the single WW domains from IQGAP3, IQGAP2 and WAC, as well as both of the tandem WW domains from WBP4 (a.k.a. FBP21), PRPF40A (a.k.a. FBP11), PRPF40B, TCERG1 (a.k.a. FBP28) and ARHGAP12. There are no published or unpublished structures of the WW domain from any IQGAP ortholog in the Protein Data Bank. Nevertheless, by comparing the sequence of IQGAP1's WW domain to published structures of WW domains from WBP4, TCERG1 and PRPF40A [51–55], we constructed a schematic model of IQGAP1's WW domain (Figure 6A). Human IQGAP has 21 residues between the two domain-defining tryptophans, a feature shared with over 80% of human WW domains. Thus, henceforth, we will use the numbering scheme shown in Figures 6A and 7A, in which these tryptophans are at positions 1 and 23, to discuss the general sequence features of WW domains. A tryptophan at position 1 and a tryptophan or phenylalanine at position 23 are found in over 95% of human WW domains, including IQGAP1's. Other sequence features that IQGAP1 shares with the majority of human WW domains include a glycine at position 9, aromatic residues at positions 12 and 13, an asparagine or aspartate at position 15, a threonine position 18, and a proline at position 26.

The schematic representation (Figure 6A) is consistent with the structure predicted by Alphafold [56,57], which became available subsequently (Figure 6B,C). In this structure, the WW domain of IQGAP1 is predicted to fold into a canonical WW domain structure: an antiparallel beta sheet with three short strands connected by two loops, with main-chain hydrogen bonds providing stability (Figure 6B). As in other WW domains [52,53,55,58], the side chain of the first domain-defining tryptophan (residue 1 in Figure 6A, W685 in IQGAP1) faces the underside of the sheet, where, together with the tyrosine at position 13 (Y697 in IQGAP1), the asparagine at position 15 (N699 in IQGAP1) and the proline at position 26 (P710 in IQGAP1), it constitutes a hydrophobic core that provides additional stability to the domain (Figure 6C). Alternating side chains in the beta strands (residues 2, 4, 10, 12, 14, 16, 18, 21 and 23) face up relative to the plane of the sheet to form the top side of domain; these residues, as well as residues in one or both of the connecting loops, are frequently observed to participate in ligand binding in other WW domains [53–55,59–62]. Collectively, these features suggest that the WW domain of IQGAP1 adopts a canonical WW domain structure [23–25]. There are, however, several differences between IQGAP1 and the consensus WW domain sequence that may be important (see further below and Discussion).

Previously, we constructed a mutant version of IQGAP1 that contained five different substitution mutations in highly conserved residues in its WW domain [32]. These five changes include (i) alanine substitutions to four residues that constitute a hydrophobic core on the underside of the beta sheet in other WW domains [52,53,55,58], and in the predicted structure shown in Figure 6; (ii) A tyrosine-to-arginine change in a residue (position 12/Y696 in Figure 7A) whose side chain points up into the ligand binding site and helps form the proline-binding groove in other WW domains [23,53,55]. In our previous work, we demonstrated that these five mutations did not influence the binding affinity of IQGAP1 for ERK2, consistent with our finding that the IQ domain of IQGAP1 was necessary and sufficient for binding to ERK2 [32]. Here, we reintroduced those five mutations (W685A, Y696R, Y697A, N699A, P710A) into GST-WW and renamed this variant ‘ARAAA’ (see Figure 7A). We also constructed a second quintuple mutation variant of GST-WW designated ‘AARAA’. This variant contains three of the above changes, plus two more substitutions to residues in the second beta strand. As shown in Figure 7B (top panels), both of these variants were completely defective for binding to p110α. The results of eight independent experiments with these two mutants is quantified in Figure 7C, and indicated that the lack of binding exhibited by these two variants is very highly significant. Hence, combining multiple substitutions in conserved residues in the WW-domain of IQGAP1 abolished its ability to bind to p110α.

To begin to explore residues in IQGAP1's WW domain that may directly contact p110α, we used site-directed mutagenesis to examine the role of five individual residues thought to participate in ligand binding in other WW domains [51–55,59–67]; these residues are represented in ball-and-stick format in Figure 6C. Moving from N-terminal to the C-terminal, the first residue we examined was position 10/Y694. This position, which sits at the beginning of beta strand 2, is an aromatic residue in only four other human WW domains; instead, most human WW domains have an arginine or lysine at this position. As shown in Figure 7B,C, substitution of Y694 with alanine reduced the binding of GST-WW binding to p110α by ∼40%, and substitution with arginine abolished binding.

The next residue we examined is position 12/Y696. This residue is a tyrosine or phenylalanine in 96% of human WW domains, and its mutation to alanine [66] or arginine [64] greatly reduced ligand binding in other WW domains tested. In published structures, it is typically observed to help form the binding groove for one or more prolines in the ligand [23,53,55]. Interestingly, mutation of this residue to alanine had no detectable effect on GST-WW binding to p110α, nor did the more drastic substitution of arginine for the native tyrosine (Figure 7B,C).

In most human WW domains, the position 14 residue is usually either a branched-chain hydrophobic residue (valine, leucine or isoleucine; 63% of human WW domains) or an aromatic residue (30%). In human IQGAP1, position 14 is a histidine. Position 14 has been shown to be involved in ligand binding in multiple WW domains [53,55,59–61], although in none of these examples was the position 14 residue a histidine. As shown in Figure 7C, in the IQGAP1 WW domain, a histidine to arginine substitution at position 14/H698 reduced binding by roughly 50%. Interestingly, substitution of this residue with alanine reproducibly increased binding to p110α, by an average of 250% (Figure 7D).

Position 16 has also been shown to be involved in ligand binding in multiple studies [59–62], although in all these studies the position 16 residue was a histidine, as it is in the majority (57%) of human WW domains. In human IQGAP1, position 16/L700 is a leucine. As shown in Figure 7C, substitution of L700 with either alanine or arginine both reduced binding to p110α by ∼50%.

Finally, position 23/W707, like position 12/Y696, is typically observed to help comprise the proline-binding groove [23,53,55,59]. A tryptophan to alanine substitution at this position in GST-WW reduced binding to p110α by roughly 75% (Figure 6C).

To summarize, mutational changes to four of the five individual residues we examined resulted in decreased binding to p110α, consistent with expectations based on studies of other WW domains. On the other hand, there were some possibly important differences and nuances; see Discussion for further analysis.

Over the course of these studies, we performed over 20 independent, quantitative binding assay experiments between the IQGAP1 WW domain and p110α (Figure 1D). From these data, we were able to obtain an estimate of the 130 µM for the dissociation constant (K_d) of this interaction, with a 95% confidence interval of 105–153 µM (see Supplementary Table S1). Typically, WW domains that have affinities in this range are aided in binding to their ligands by additional interactions between the binding partners that occur outside of the WW domain; see Discussion for further analysis.

IQGAP is a multifunctional scaffold protein that has been implicated in many physiological processes. It has an evolutionarily ancient role in regulating cytoskeletal rearrangements in fungi and animals [68,69]. In mammals, it also functions as a scaffold protein for the RTK/RAS/MAPK pathway and the PI3K pathway — two of the pathways most frequently mutated in cancer [70–72] — and additionally regulates several other cancer-relevant signaling pathways [17–21]. Much work still needs to be done to understand how IQGAP1 and other scaffold proteins coordinate signaling in localized subcellular regions depending on cell type and cell context, how this may go awry in disease, and how it might be exploited for new targeted therapies. Despite the challenges, continued progress in this area may open new opportunities for personalized medicine. A promising direction in this regard is blocking kinase-scaffold interactions as a therapeutic strategy for treating cancer and other diseases.

Here, we searched for a binding partner for the WW domain of IQGAP1, motivated by recent findings that a cell-penetrating version of this domain has potent anti-tumor properties with minimal toxicity to non-transformed cells and tissues (see Introduction for further details). We found that this domain binds directly to p110α, the catalytic subunit of PI 3-kinase, which is one of the most frequently mutated proteins in human cancer. In contrast, this WW domain did not bind to ERK1, ERK2, MEK1, MEK2 or other parts of IQGAP1 (Figures 1 and 2). Moreover, we found the same set of binding specificities for the cell-penetrating anti-tumor WW peptide that contains poly-d-Arginine and myc-epitope tags in addition to the WW domain residues (Figure 3). Furthermore, we found that the WW domain of IQGAP1 did not bind directly to p85α, the regulatory subunit of PI 3-kinase, but could bind to the p110α/p85α holoenzyme, both in vitro (Figure 4) and in human tissue culture cells (Figure 5). IQGAP1's WW domain also bound to the p110α/p65α heterodimer, which contains a truncated, mutationally activated variant of p85α (Figure 4), demonstrating that the WW domain of IQGAP was able to bind to both unactivated and mutationally-activated PI3K. Finally, we built a model of IQGAP1's WW domain (Figure 6), and demonstrated the role of key residues in this domain in p110α binding (Figure 7). Our results provide new insights into IQGAP1-mediated scaffolding, and suggest that cell-penetrating WW peptides may exert their anti-tumor effects by disrupting the interaction of IQGAP1 with the catalytic subunit of PI 3-kinase.

No x-ray or NMR structures of IQGAP1's WW domain have been published. Nevertheless, by comparison to published structures of WW domains that are close in sequence to IQGAP1, we built a model of IQGAP1's WW domain (Figure 6). Using this model as a reference, we then tested the role of selected residues in this WW domain on p110α binding. First, we made multiple substitutions to residues that constitute the hydrophobic core on the underside of the domain, and have been shown in other WW domains to contribute to folding and stability. Consistent with expectations, both multi-substituted variants we tested were severely defective in binding to p110α (Figure 7). We also expressed one of these multi-substituted variants in human cells, and showed that it was defective in interacting with endogenous PI3K (Figure 5).

Next, we tested the role of individual residues in IQGAP1's WW domain that are predicted to be situated on the ligand binding surface (Figure 7). We found that mutations to residues in positions 10, 14, 16 and 23 (see Figure 7A for numbering) all affected the binding of GST-WW to p110α. These results are consistent with the hypothesis that these residues either directly contact p110α, or influence the positioning of residues that do.

The results of mutating position 16 (L700 in IQGAP1) and position 23/W707 were the most straightforward. All tested mutations of these residues reduced binding to p110α by 50% or more. Moreover, there is considerable precedent for a direct role in ligand binding for residues in these positions in other WW domains [23,53,55,59–62].

The results with position 14/H698 were more complicated, in that substitution with alanine actually increased binding to p110α by ∼250%, whereas substitution with arginine reduced binding by over 50%. Interestingly, position 14 is known to be a key residue in determining the specificity of binding in other WW domains [60]. Moreover, substitutions to the position 14 residue in the WW domain of PIN1 have also been shown to increase ligand binding affinity [73]. Modifications of H698 may thus be worth pursuing as a means to develop IQGAP1 WW peptides and peptidomimetics that have higher affinity for p110α, and therefore might display greater anti-tumor potency as well.

In many published structures of WW-ligand complexes, the position 10 residue is not observed to make direct contact with the ligand. In both cases where the position 10 residue does contact the ligand, position 10 is an arginine [65,67]; indeed, the position 10 residue is a basic, positively charged arginine or lysine in most human WW domains. In contrast, position 10 in IQGAP1 is a tyrosine. We found that substituting this residue with arginine (Y694R) virtually eliminated binding to p110α, and that substitution with alanine reduced binding by ∼40% (Figure 7). The simplest interpretation of these data is that Y694 participates directly in binding to p110α.

In most published structures of WW-ligand complexes, the position 12 residue is aromatic, as it is in IQGAP1, and is observed to make direct contact with a proline in the ligand [23,53,55]. Furthermore, mutation of this residue to alanine [66] or arginine [64] was found to substantially compromise binding in other WW domains tested. For IQGAP1's WW domain, however, we found that substituting this residue with either alanine (Y696A) or arginine (Y696R) had no discernable effect on binding to p110α (Figure 7). The simplest interpretation of these data is that the side chain of Y696 does not make critical contacts with p110α. Collectively, the results with position 10/Y694 and position 12/Y696 suggest that there may be some differences between how the WW domain of IQGAP1 binds p110α and the precedent established by the majority of WW domain studies. Another feature which suggests that there may be unique aspects to the manner in which the WW domain of IQGAP1 binds its ligand(s) is the presence of a glycine at position 21. IQGAP3 is the only other human WW domain containing a glycine at this position; instead, 86% of human WW domains have a serine or threonine here, and these side chains have been shown to participate in ligand binding in many instances [53,55,61,62].

WW domains are proline recognition domains that typically bind to proline-rich short linear motifs in their binding partners [26]. Further work will be required to map the binding site(s) on p110α that bind to the WW domain of IQGAP1. In a systematic study of 42 WW domains that focused on identifying ligands, the WW domain of IQGAP1 was found to be unfolded under the conditions studied, and a ligand could not be identified [63]. The lack of ligand identification in this previous study cannot be solely attributed to the lack of folding, since some WW domains are observed to fold in the presence of ligand [63,74,75]. If indeed there is coupled folding and binding in the interaction of IQGAP1's WW domain with p110α, this could help explain the modest affinity of the interaction that we observed. It is a well-established hypothesis that coupled folding and binding may promote low-affinity yet high-specificity molecular recognition [76].

It is difficult to express a PI3K catalytic subunit like p110α in the absence of a regulatory subunit such a p85α, or vis versa, as the expressed protein either forms heterodimers with its endogenous partner or is rapidly degraded [41,43,44]. We were able to overcome this difficulty by using a rabbit reticulocyte lystate in vitro translation system to express p110α and p85α. Reticulocytes do not express endogenous p110α/p85α [77], and we found no evidence of p110α or p85α degradation in this system, even when the proteins were translated separately.

Using this approach, we showed that the WW domain of IQGAP1 binds directly to p110α, yet we found no evidence that this WW domain binds directly to p85α. Anderson and colleagues, however, have shown that both the WW and IQ domains of IQGAP1 are necessary for its binding to the PI3K holoenzyme in cells, and that a fragment of IQGAP1 containing both of these domains binds directly to p85α [22]. A plausible model accounting for all these observations is that both the WW domain and the IQ domain of IQGAP1 co-operate to bind to the p110α/p85α holoenzyme, with the WW directly contacting p110α, and the IQ domain contacting p85α. This cooperation may function to increase the affinity of IQGAP1–PI3K binding, as well as to optimally position PI3K relative to a localized pool of its substrate PIP2. Cooperative ligand binding involving a WW domain and a neighboring domain has been observed in other cases [23]. For example, the WW domain in dystrophin cannot bind its ligand without the help of the adjacent EF hand domain [61]. Furthermore, in proteins that contain more than one WW domain, two such domains can co-operate to bind their target with higher affinity and specificity [78,79].

Figure 8 shows a model of select binding and scaffolding interactions mediated by the WW and IQ domains of IQGAP1, based on pioneering work by other groups [12,22,29,80–83], and extended by the findings presented here and in our previous study [32]. First, multiple components of the RTK/RAS/MAPK cascade — including RTKs, BRAF, MEK1/2, and ERK1/2 — interact with the IQ domain of IQGAP1 (Figure 8A). Second, in the PI3K pathway, IQGAP1 interacts with PI3K using both its WW and IQ domains, the former recognizing p110α and the latter primarily recognizing p85α (Figure 8B). It is also possible that there are further interactions between IQGAP1 and p110α in addition to those mediated by the WW domain. IQGAP1 also interacts with PIPKI, the enzyme immediately upstream of PI3K, via its IQ domain (Figure 8B) [22,83,84]. Third, we posit that cell-penetrating IQGAP1 WW peptides may block the productive engagement of PI3K with IQGAP1 by directly interfering with WW-p110 binding (Figure 8C).

Herein we showed that the p110α catalytic subunit of human phosphoinositide 3-kinase (PI3K) is a direct binding partner of the WW domain of human IQGAP1, and made progress on the structure/function characterization of this interaction. Our findings propound a refined model of IQGAP1-mediated scaffolding, and suggest that blocking the productive engagement of PI3K with IQGAP1 is a key on-target effect of anti-neoplastic IQGAP1 WW-derived peptides.

The mammalian genes used in this study were human IQGAP1 (NCBI accession number NM_003870), bovine PIK3CA (NM_174574), human PIK3CA (NP_006209), human PIK3R1 (NM_181523), human ERK1 (MAPK3; NM_002746), human ERK2 (MAPK1, NP_620407), human MEK1 (MAPK2K1, NM_002755), and human MEK2 (MAP2K2, NM_030662).

GST-WW contains residues 679–719 of human IQGAP1 inserted into plasmid pGEXLB [85]. The region of interest was amplified in a polymerase chain reaction (PCR) using plasmid pCR-BluntII-TOPO-hIQGAP1 [32] as the template and primers hIQGAP1(679-x) and hIQGAP1(x-719) (See Supplementary Table S2 for all PCR primer sequences). The PCR product was cut with BamHI and SalI and cloned into pGEXLB that had been digested with the same enzymes. For the experiment shown in Figure 3 and elsewhere, QuikChange (Agilent) was used for site-directed mutagenesis; all mutations were confirmed by sequencing.

For the experiments shown in Figure 2, the constructs for producing full-length IQGAP1 protein and IQGAP(1-678) by in vitro transcription/translation have been described [32]. To make IQGAP1(743-1657), the region of interest was amplified in a PCR using plasmid pCR-BluntII-TOPO-hIQGAP1 as the template and primers hIQGAP1(743-x) and hIQGAP1(x-1657). The PCR product was cut with XmaIII and SalI and cloned into pGEM3Z that had been digested with the same enzymes. The resulting construct, pGEM3Z-hIQGAP1(743–1657) contains the 743–1657 open reading frame downstream of a T7 promoter.

A plasmid containing the bovine PIK3CA gene with an N-terminal myc-epitope tag and a C-terminal CAAX box, downstream of a T7 promoter, was obtained as a gift from Julian Downward [38]. We converted this to a sequence that encodes human p110α protein by site-directed mutagenesis, as the human and bovine amino acid sequences differ at only two residues. Specifically, we introduced R532K and C535S substitutions into the bovine coding sequence. The resulting plasmid is designated T7-hb-p110α, where ‘hb’ stands for ‘humanized bovine'.

Plasmid pGEM3Z-p110α contains the coding sequence for full-length untagged human p110α protein downstream of a T7 promoter. To construct this plasmid, the human p110α-encoding open reading frame was amplified by PCR using primers p110-1-x and p110-end-down and T7-hb-p110α as the template. This process removed both the myc tag and the CAAX box, and replaced them with wild-type sequence. The resulting PCR product was digested with BamHI and SalI and inserted into the corresponding sites of pGEM3Z.

Plasmid pGEM3Z-p85α contains the coding sequence for full-length human p85α protein downstream of a T7 promoter. To construct this plasmid, the PIK3R1 open reading frame was amplified by PCR using primers p85up and p85down. The template for this reaction was a human PIK3R1 sequence-verified cDNA clone from the mammalian gene collection and sold by Dharmacon (Clone ID 30528412). The resulting PCR product was digested with BamHI and SalI and inserted into the corresponding sites of pGEM3Z.

Plasmid pGEM3Z-p65α (encoding residues 1–571 of p85α protein) was constructed from pGEM3Z-p85α by introducing a stop codon at codon 572 of the PIK3R1 coding sequence by site-directed mutagenesis.

To construct GST-p110α(1–190) GST-p110α(6–123) for use in Figure 2E, the regions of interest were amplified in a polymerase chain reaction (PCR) using plasmid pGEM3Z-p110α as the template and primer p110-1-x or p110-6-x as the upstream primer and p110-x-190 or p110-x-123 as the downstream primer (Supplementary Table S2). The PCR product was cut with BamHI and SalI and cloned into pGEXLB that had been digested with the same enzymes.

The constructs for in vitro transcription/translation of full-length ERK1, ERK2, MEK1 and MEK2 used in Figures 1–3 have been described elsewhere [85]. The MEK1(1–60) and MEK2(1–64) constructs used in Supplementary Figure S1 have also been described elsewhere [86].

The constructs for the cell-based experiments in Figure 5 were made in multiple steps. First, the open reading frames encoding GST, GST-WW and GST-WW-AARAA were amplified in a PCR reaction using primers GST-WW-up and GST-WW-down (Supplementary Table S2) and the corresponding pGEXLB-based plasmid as the template. This strategy placed a unique EcoRI site upstream of the start codon for GST and a unique XbaI site downstream of the stop codon of the open reading frame (ORF). The PCR products were digested with EcoRI and XbaI and inserted into the mammalian expression vector pcDNA-V5-HIS-B (Invitrogen Corp.) that had been cut with the same enzymes. The resulting plasmids (pcDNA-GST, pcDNA-GST-WW, and pcDNA-GST-WW-AARAA) express the GST-fusion constructs driven by the CMV promoter; the V5 and HIS6 tags are not included in the fusion proteins.

pcDNA-GST-p85α was constructed by excising the human p85α ORF from pGEM3Z-p85α as a BamHI-SalI fragment, and inserting into pGEXLB that had been cut with the same enzymes, yielding pGEXLB-p85α, which served as a cloning intermediate. In the next step, the GST-p85α ORF in this plasmid was amplified using primer GST-WW-up and GST-WW-down and subcloned into pcDNA-V5-HIS-B as described above.

Proteins labeled with [³⁵S]-methionine were produced by coupled transcription and translation reactions (T7, Promega). Translation products were partially purified by ammonium sulfate precipitation [87], and resuspended in binding buffer (20 mM Tris–HCl (pH 7.1), 125 mM KOAc, 0.5 mM EDTA, 1 mM DTT, 0.1% (v/v) Tween20, 12.5% (v/v) glycerol) prior to use in binding assays.

GST fusion proteins were expressed in bacteria, purified by affinity chromatography using glutathione-Sepharose (Cytiva Life Sciences) and quantified as described elsewhere [85,88].

Protein binding assays were performed as described previously [85,88]. Quantification of binding was performed on a Typhoon TRIO+ Imager using phosphorimaging mode. Percent binding was determined by comparing the input with the amount that was co-sedimented. This was then corrected for minor differences in loading of the GST fusion protein, which was assessed by quantifying a TIFF image of the Coomassie-blue-stained gel using ImageJ software. Each binding assay presented in this paper was repeated at least four separate times (i.e. four independent experiments), with duplicate points (i.e. technical replicates) in each experiment. Technical replicates in a given experiment are averaged together to obtain a single data point. We define ‘independent experiments’ as experiments performed on different days, with fresh batches of GST-fusion proteins and in vitro translated proteins.

Co-immunoprecipitation experiments (Figure 3) were performed as described previously [89]. PIK3CA/PIK3R1 (p110a/p85a) recombinant human protein for use in the experiment shown in Figure 4F was purchased from Thermo Fisher Scientific (catalog number PV4788).

Human embryonic kidney (HEK) 293T cells were cultured in Dulbecco's Modified Eagle Medium (DMEM; Life Technologies) supplemented with 10% calf serum in a humidified 37°C incubator with 5% CO₂. For the pull-down assays shown in Figure 5, cells were transfected, using X-tremeGene HP DNA Transfection Reagent (Roche), with 10 µg of pcDNA-GST, pcDNA-GST-WW, pcDNA-GST-WW-AARAA or pcDNA-p85α. Transfected cells were incubated for 24 h and two plates of cells per treatment were harvested and flash frozen.

Frozen cells were lysed by gentle pipetting in 200 µl ice cold LPD buffer (20 mM Tris–HCl (pH 7.1), 125 mM KOAc, 2 mM EDTA, 1 mM DTT, 12.5% (v/v) glycerol, 1% (v/v) Nonidet P-40, plus protease inhibitors) followed by rocking for 45 min at 4°C. The extracts were clarified by centrifugation at 13 800×g for 15 min. Total protein concentration was determined by Bradford assay. Binding assay samples contained 1 mg of clarified cell extract plus 20 µl of a 50% slurry of glutathione sepharose beads in a total of 200 µl LPD buffer. Samples were rocked for 1 h at 4°C, then pelleted at low speed in a minifuge. Pellets were washed 3× in ice cold Wash Buffer (LPD buffer in which the Nonidet P-40 was reduced to 0.1%) and then assayed by SDS–PAGE and immunoblotting.

The 50-residue synthetic peptide used in Figure 3 was synthesized by GenScript U.S.A. Inc. Its sequence is:

[D-Arg]₈-EQKLISEEDL-DNNSKWVKHWVKGGYYYYHNLETQEGGWDEPP.

This sequence is identical with that of the peptide that selectively inhibited tumor and cancer cell proliferation and viability in other studies [12,22,27,28].

Protein A/G agarose beads for immunopreciptation (Figure 3) were purchased from Thermo Fisher Scientific, and the antibody was c-Myc monoclonal antibody 9E10 (Life Technologies, catalog number 132500).

The antibodies used in this study for immunoblotting (Figure 5) were (1) GST-Tag monoclonal antibody 26H1 (Cell Signaling Technologies catalog number 2624S); (2) PI3 Kinase p110α rabbit monoclonal antibody (Cell Signaling Technologies catalog number 4249); (3) PI3 Kinase p85, N-SH3, clone AB6 mouse monoclonal antibody (Sigma–Aldrich catalog number 05-212).

Statistical analysis of binding assay results was performed using Welch's unequal variance t-test with two tails or Welch's ANOVA test. Where necessary, P-values were adjusted for multiple hypothesis testing using Dunnett's T3 multiple comparison test. The Graphpad Prism software package was used for statistical analysis.

For the experiment shown in Figure 7, background binding to GST was subtracted from each value, and then all measurements in that experiment were normalized by defining the binding of wild-type GST-WW to p110α as 100. This procedure complicates statistical analysis as the wild-type control (i.e. wild-type GST-WW binding to p110α) has a mean of 100 and a standard deviation of zero. Hence, we assessed significance based upon the distance between the mutant variant mean and the normalized value of 100, measured in units of the 95% confidence interval (CI) of the mean of the variant. We considered a distance of less than 1 CI as not significant (‘ns’ in Figure 7), a distance between 1 and 2 CI as significant (* in Figure 7), a distance between 2 and 3 CI as highly significant (** in Figure 7), and a distance greater than 3 CI as very highly significant (*** in Figure 7). A similar procedure was followed for the analysis of the experiment shown in Figure 3E.

The predicted structure for human IQGAP1 is entry AF-P46940-F1 in the Alphafold database https://alphafold.ebi.ac.uk. Molecular graphics and analyses were performed with UCSF ChimeraX [90], developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from National Institutes of Health R01-GM129325 and the Office of Cyber Infrastructure and Computational Biology, National Institute of Allergy and Infectious Diseases.

The authors declare that the main data supporting the findings of this study are available within the article and its Supplementary Material.

The authors declare that there are no competing interests associated with the manuscript.

This work was supported by (1) the National Cancer Institute of the National Institutes of Health under award number P30CA062203, and the UC Irvine Comprehensive Cancer Center using UCI Anti-Cancer Challenge funds; (2) the National Center for Research Resources and the National Center for Advancing Translational Sciences, National Institutes of Health, through Grant UL1 TR001414; (3) University of California Cancer Research Coordinating Committee award CTR-20-637218. The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies.

Open access for this article was enabled by the participation of University of California in an all-inclusive Read & Publish agreement with Portland Press and the Biochemical Society under a transformative agreement with UC.

Lee Bardwell: Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing — original draft, Project administration, Writing — review and editing. A. Jane Bardwell: Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing — original draft, Project administration, Writing — review and editing. Madhuri Paul: Conceptualization, Formal analysis, Supervision, Validation, Investigation, Visualization, Methodology, Writing — original draft, Writing — review and editing. Kiku C. Yoneda: Investigation. María D. Andrade-Ludeña: Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Investigation, Writing — review and editing. Oanh T. Nguyen: Investigation. David Fruman: Conceptualization, Resources, Formal analysis, Supervision, Funding acquisition, Investigation, Writing — review and editing.

ABD

adapter-binding domain

CI

confidence interval

GST

glutathione S-transferase

ORF

open reading frame

PCR

polymerase chain reaction

PI3K

phosphoinositide 3-kinase

PIP2

phosphatidylinositol-(4,5)-biphosphate

TNT

transcription and translation