A high-avidity biosensor reveals plasma membrane PI(3,4)P2 is predominantly a class I PI3K signaling product

1Class I phosphoinositide 3-OH kinase (PI3K) signaling is central to the control of growth and metabolism in animals. Overactivation of this pathway is the most common event in cancer, yet given its major role in insulin signaling, inhibition of the pathway triggers insulin resistance and type 2 diabetes. Therefore, the ability to selectively manipulate PI3K signaling could have tremendous therapeutic benefit. Efforts to accomplish this goal are a major focus of the biomedical enterprise.

2At the molecular level, PI3K signaling involves the generation of the plasma membrane (PM) second messenger lipids phosphatidylinositol 3,4,5-trisphosphate (PIP3) and phosphatidylinositol 3,4-bisphosphate (PI(3,4)P2) that activate downstream effector proteins like the serine/threonine kinase Akt. PIP3 is the major lipid produced, and most functions of the pathway are attributable to it. PI(3,4)P2 has instead been viewed as either a degradation product or an alternative activator of the pathway. However, selective functions for PI(3,4)P2 have recently been described that are independent of PIP3. These include the formation of lamellipodia and invadopodia, along with clathrin-mediated and clathrin-independent endocytosis. In each case, these functions could conceivably be driven by, or occur independently of, class I PI3K signaling.

3Synthesis of PI(3,4)P2 can proceed via three routes. In the first, class I PI3K directly generates PI(3,4)P2 and PIP3 by 3-OH phosphorylation of the respective PM phosphoinositides PI4P and PI(4,5)P2. Subsequently, the observation that PI(3,4)P2 synthesis lags behind PIP3 in stimulated cells, coupled with the discovery of the PIP3-specific 5-phosphatase enzymes SHIP1 and SHIP2, led to the proposal of a second route: PI(3,4)P2 production by removal of the 5-OH phosphate from PIP3. More recently, a third route has been established, again invoking direct phosphorylation of PI4P, this time by class II PI3K enzymes. However, the relative contributions of these pathways, and how they couple to disparate PI(3,4)P2-dependent cellular functions, remain unclear.

4Resolving how the spatial/temporal dynamics of PI(3,4)P2 signaling couples to different biological functions requires approaches to identify the lipid in intact, living cells. Isolated lipid binding domains fused to fluorescent reporters often make highly selective genetically encoded biosensors for this purpose. The pleckstrin homology (PH) domain on the C terminus of Tandem Ph-domain containing Protein 1 (TAPP1) exhibits specific binding to PI(3,4)P2 in the test tube. As a result, several studies have employed fluorescent protein conjugates of this domain to track PI(3,4)P2 signaling, though the domain fails to detect resting levels or the limited accumulation of the lipid in response to stimuli such as insulin-like growth factor.

5Herein, we developed a higher-avidity tandem trimer of PH-TAPP1. We show PI(3,4)P2 generation is sufficient to recruit the probe, which is exquisitely selective for the lipid over other phosphoinositides. We then demonstrate that the class I PI3K pathway, acting via PIP3 synthesis, dominates PI(3,4)P2 accumulation in cells. Our data also support the recently proposed direct degradation of both PI(3,4)P2 and PIP3 by the lipid phosphatase and tumor suppressor PTEN. Collectively, our data show that the class I PI3K pathway is the most potent driver of PI(3,4)P2-dependent signaling.

6The TAPP1 protein (encoded by the PLEKHA1 gene) contain both N-terminal PH domains and cPHs (Fig. 1 A), the latter of which selectively binds PI(3,4)P2. Previous studies using the isolated TAPP1 cPH as a lipid biosensor found no detectable translocation in response to stimuli that induce modest PI(3,4)P2 accumulation. To improve avidity, tandem dimers of TAPP1 fragments containing cPH have been used. However, these constructs are based on a fragment including the entire C terminus, which carries binding sites for other proteins in its tail, and have been shown to induce dominant negative effects. Therefore, we built tandem dimers and trimers of the isolated cPH (residues 169–329 in human TAPP1) as previously defined and included a nuclear export sequence (Fig. 1 A).

7Expression of EGFP-tagged cPH monomers, dimers, or trimers in COS-7 cells maintained in 10% serum exhibit a predominantly cytosolic localization of the probe when viewed by confocal microscopy (Fig. 1 B). However, comparison with another cytosolic protein (infrared fluorescent protein [iRFP]) revealed enrichment of cPHx2 and cPHx3 at the cell periphery; normalization of the two signals followed by subtracting the iRFP signal from that of EGFP (Materials and methods section) showed striking cPHx2/3 peripheral localization (Fig. 1 B). This signal was abolished by a PM-targeted INPP4B phosphatase that specifically degrades PI(3,4)P2, though not by a catalytically inactive mutant (Fig. 1 B). Quantification of the difference between EGFP and iRFP signals (Fig. 1 C with Kruskal–Wallis statistic 152.5; P < 10−4) yielded significantly increasing signal from cPHx1 to cPHx3; INPP4B-CAAX significantly reduced cPHx3 signal (P < 10−4 in each case; Dunn’s multiple comparison test), whereas the C842A inactive mutant did not (P > 0.99). Therefore, cPHx2 and cPHx3 could detect PI(3,4)P2 at the PM of living cells in the presence of serum.

8A potential caveat to using tandem arrays of lipid binding domains is that the resulting probe may cluster lipids and exhibit aberrant localization or mobility in the membrane. To determine whether this was the case, we performed single-molecule imaging of cPHx1, cPHx2, and cPHx3 mobility in the PM by tagging with photoactivatable mCherry (PAmCherry). As shown in Fig. 1 D, full activation of the probe with violet light led to uniform labeling of the PM in total internal reflection fluorescence (TIRF) microscopy (TIRFM), whereas low activation intensities allowed us to image single molecule trajectories. Analysis of the mean square displacement of these trajectories over time revealed free Brownian motion of all three probes (Fig. 1 E). cPHx1, cPHx2, and cPHx3 diffused with an apparent diffusion coefficient of ∼0.3 µm2/s (Fig. 1 F) and did not substantially differ (F = 0.71, P = 0.50, one-way ANOVA), and all exhibited a relative deviation ratio close to 1 (Fig. 1 G), meaning there was no substantial difference in actual displacement relative to that predicted by free Brownian diffusion. Therefore, the tandem array of domains did not slow the probe’s mobility in the membrane.

9The lifetime distribution of the single molecules on the membrane followed a single exponential decay with a characteristic off-rate constant (koff) that increased linearly with laser power due to increased photobleaching (Fig. 1 H). Extrapolation of this off rate to the intercept at zero illumination power allowed us to estimate the true off rate of the probes. cPHx1 has a lifetime on the membrane of ∼90 ms (off rate = 11 s–1), whereas cPHx2 and cPHx3 both showed lifetimes of ∼140 ms. This change in dissociation rate of the tandem arrays was comparatively modest and certainly not multiplicative as would be expected from multiligand binding. This implies that for the majority of probe molecules at steady state, the PH domains are not all occupied by lipid, most likely because of the low abundance of PI(3,4)P2. The enhanced PM binding observed in Fig. 1 C likely stems from transient dual or triple occupancy of the PH domains slowing the overall off rate of the population (and perhaps enhancing the on rate for cPHx3 vs. cPHx2).

10A potential limitation of biosensors is that when bound to lipid, they will protect the lipid from consumption by enzymes and occlude interaction with endogenous effector proteins, both of which may be needed for local enrichment of the lipid. Therefore, free diffusion of the probe:lipid complex can disrupt local enrichment rather than reporting on it. From the relationship r = (2D/koff)0.5, the distance r that cPHx3 typically diffuses while attached to lipid can be estimated as ∼290 nm, not much larger than the diffraction limit. This implies there will be a limited propensity of the probes’ free diffusion to “smear out” local enrichment of PI(3,4)P2 molecules that are detectable with diffraction-limited optical imaging.

11To determine if PM localization of cPH probes is sensitive to acute depletion of PI(3,4)P2, we used chemically induced dimerization between FK506 binding protein 12 (FKBP) and the FKBP12 Rapamycin Binding (FRB) domain of mTor to recruit INPP4B to the PM (Fig. 1 I). Rapamycin-induced dimerization and thus INPP4B recruitment led to depletion of PM-associated cPH signal in TIRFM (Fig. 1 J; two-way ANOVA of area under the curve before and after rapamycin addition, F(1, 200) = 314.5, P < 10−4; P < 10−4 for cPHx1, cPHx2, and cPHx3, Sidak’s multiple comparisons test). The depletion was progressively greater for cPHx3 > cPHx2 > cPHx1 (comparing area under the curve, F = 75.64; P < 10−4; cPHx2 vs. cPHx3, P = 0.02; cPHx1 vs. cPHx2, P < 10−4), reflecting the increased avidity of the tandem probes for PM PI(3,4)P2. On the other hand, probes for other inositol lipids, including PI4P (P4Mx1), PI(4,5)P2 (PH-PLCD1), PIP3 (PH-ARNO2Gx2), or PI(3,4)P2/PIP3 (PH-AKT), did not show a significant change. Notably, when the data were normalized to emphasize the kinetics of the change, no difference between the rates of cPHx1, cPHx2, or cPHx3 dissociation after INPP4B recruitment was observed (Fig. 1 J, right; Kruskal–Wallis statistic = 1.19, P = 0.55), indicating that the higher-avidity tandem probes did not effectively sequester lipid from the INPP4B enzyme. Furthermore, recruitment of inactive INPP4B mutant C842A did not lead to substantial depletion of the cPH probes from the PM (Fig. 1 K).

12Given the much higher avidity of cPHx3 and cPHx2 for the PM, we wanted to rule out a weak interaction with other lipids that might synergize to influence PM targeting by these probes. To this end, we used chemically induced dimerization again, this time using the INPP5E phosphatase domain (Fig. 1 L), since this domain is active against PI(4,5)P2 and PIP3, generating PI4P and PI(3,4)P2, respectively. INPP5E recruitment did not displace PI4P-binding P4Mx1 or any of the cPH probes from the PM, though it did displace probes that bound to PIP3 or PI(4,5)P2 (Fig. 1 M). Therefore, these two inositol lipids do not influence the PM association of cPH probes.

13Previous studies have implicated a role for PI(3,4)P2 at clathrin-coated structures. We tested for an enrichment for our cPHx3 probe at these structures in 293A cells in which the endogenous clathrin had been labeled with a split GFP approach; mCherry-cPHx3 was not enriched at these structures, though an mCherry-tagged PI3K-C2α construct clearly was (Fig. S1 A). This is not surprising, since (1) another inositol lipid firmly established in clathrin-mediated endocytosis, PI(4,5)P2, is also not seen to be enriched at clathrin-coated structures and (2) careful modeling of the quantities of PI(3,4)P2 required for bud maturation posits that there may be just enough to interact with the endogenous SNX9 effector protein, resulting in undetectable biosensor recruitment.

14A recent strategy found that addition of the clathrin binding module from Auxilin1 generates a coincidence detector for inositol lipids at clathrin-coated structures. With such coincident detector probes, we could measure PI(4,5)P2 enrichment with a PH-PLCδ1 but not with our cPHx3 PI(3,4)P2 probe (Fig. S1 A). This was in contrast with a tandem C-terminal truncation, also from human TAPP1, which does show enrichment in previous reports and in our experiments (Fig. S1 A). What is the cause of this discrepancy? We noted that the C terminus of TAPP1 contains a Leu-Val-Asp-Leu-Asp clathrin binding box that could explain enrichment at clathrin-coated structures. Indeed, a tandem dimer of the C-terminal domain without the PI(3,4)P2-binding PH domain still enriched at clathrin-coated structures when fused to the Auxilin1 clathrin binding module (Fig. S1 A). In support of these findings, we found that the TAPP1 PH-domain + C-terminal tandem dimer fused to the Auxilin1 module did not dissociate from the membrane after INPP4B recruitment, when cPHx3 does (Fig. S1 B). Therefore, enrichment of TAPP1-derived protein fragments at clathrin-coated structures does not depend on PI(3,4)P2.

15Together, these results allowed us to conclude that (1) the cPHx3 probe derived from TAPP1 exhibits more prominent membrane localization than previously used single and tandem dimer versions, (2) the tandem trimer configuration does not disrupt free diffusion in the plane of the membrane nor sequester PI(3,4)P2 to an extent that prevents INPP4B access and (3) the tandem trimer configuration exhibits localization that depends solely on the phosphoinositide PI(3,4)P2. These are crucial criteria in the definition of a reliable genetically encoded lipid biosensor. However, there is another critical criterion that must be met, which we tested next.

16We have argued that a crucial but often ignored criterion for a high-fidelity lipid biosensor is a sole requirement of the lipid to drive biosensor localization. A convenient way to test this criterion is to induce ectopic synthesis of the lipid elsewhere in the cell and test whether this is sufficient to recruit the biosensor. To this end, we turned to the Legionella effector protein LepB, the N terminus of which possesses PI3P 4-OH kinase activity, thus generating PI(3,4)P2 via a noncanonical pathway not used by mammalian cells. We found that expression of LepB led to substantial PI3P depletion from cells and consequent swelling of the endosomal compartment that depends on this lipid for function (unpublished data). Therefore, to induce acute LepB activity, we turned to an optogenetic approach that utilizes genetic code expansion to incorporate an unnatural, caged amino acid into the protein.

17This system relies on mutating the target protein at the desired codon to the infrequently used amber stop codon (UAG). The mutant gene is then transfected into cells along with plasmids encoding an engineered pyrrolysyl-tRNA synthetase/tRNA pair and the caged amino acid is added to the media. Here a hydroxycoumarin-caged lysine (HCK) is transacylated onto the tRNA and thence ribosomally incorporated into the mutated gene in response to the UAG codon. In the context of LepB, the bulky hydroxycoumarin group incorporated onto lysine 39 blocks the active site. However, illumination of cells expressing this system with 405-nm light causes photolysis of the coumarin group, liberating the lysine residue and generating wild-type, active LepB (Fig. 2 A). This provides precise and acute spatiotemporal control over LepB function in live human cells.

18Optogenetic activation of LepB in cells expressing cPHx3 caused synthesis of PI(3,4)P2 on endosomes and striking recruitment of the cPHx3 probe (Fig. 2 B). This optogenetic activation was only observed in cells grown in the presence of exogenous HCK and depended on cotransfection with the LepBK39UAG, ruling out off-target effects on endogenous UAG-containing genes or of light exposure alone. Similarly, we observed clear translocation of PH-AKT to endosomes, which is expected given this PH domain’s binding to both PIP3 and PI(3,4)P2. However, no recruitment of the PIP3-specific PH-ARNO2Gx2 was observed, demonstrating that there was no further phosphorylation of the lipid at the 5-OH. Finally, we observed some depletion of the PI3P biosensor FYVE-EEA1, though the biosensor remained mostly endosome bound, indicating a small fraction of total PI3P was converted to PI(3,4)P2 over the time course of these experiments (Fig. 2 B).

19Collectively, these results demonstrate that generation of PI(3,4)P2 in a cellular membrane is indeed sufficient to recruit cPHx3, which, given its extensive in vitro selectivity, now passes all the key criteria that define a high-fidelity, genetically encoded biosensor for PI(3,4)P2. The fidelity of the probe established, we next turned our attention to deploying this tool to answer some central questions about the lipid’s metabolism and function.

20The canonical view of PI(3,4)P2 synthesis induced by class I PI3K signaling is that PI3K generates PIP3, which is then dephosphorylated at the 5-OH position by inositol polyphosphatase 5-phosphatase (INPP5) family members SHIP1 or SHIP2. Both enzymes are activated by tyrosine phosphorylation cascades initiated by the activated receptor tyrosine kinase. The extent of PI(3,4)P2 accumulation varies depending on receptor activation, with insulin or insulin-like growth factor reported to generate only modest amounts of the lipid, which have not been detected with a single cPH probe.

21We tested the capacity of cPHx3 to detect insulin-induced PI(3,4)P2 production in HeLa cells (Fig. 3 A). To unambiguously detect PIP3 in the same cells, we used a tandem dimer of the PH domain from ARNO, specifically selecting the 2G splice variant with high PIP3 affinity and introducing the I303E mutation that prevents PH domain interaction with Arl GTPases. We selected this domain since it exhibits clear discrimination for PIP3 over PI(3,4)P2. Insulin stimulation of HeLa cells generated a robust, transient increase in PIP3 biosensor at the PM, with a lagging, yet more sustained appearance of the PI(3,4)P2 biosensor (Fig. 3, B and C)—a clear recapitulation of the original biochemical measurements of PI3K product generation, apparently supporting the canonical view.

22As an alternative mechanism to activate class I PI3K, we turned to chemically induced dimerization to recruit the inter-Src homology 2 (iSH2) domain from the p85 regulatory subunit of PI3K (Fig. 3 A). This system recruits endogenous PI3K p110 catalytic subunits to the membrane, inducing PIP3 synthesis. However, in this situation, tyrosine kinase-mediated activation of SHIP phosphatases is not expected. Nevertheless, iSH2 recruitment to the PM of COS-7 cells induced rapid PIP3 synthesis and the same, lagging PI(3,4)P2 accumulation (Fig. 3 D). Use of the selective inhibitor GDC-0941 confirmed this was due to class I-PI3K (Fig. S2 A). We therefore speculated that, in this context at least, PI(3,4)P2 accumulation might be driven by direct phosphorylation of PI4P by p110, given that the enzyme is known to perform this reaction in a test tube.

23To test this speculation, we devised an experiment wherein the INPP5E enzyme would be corecruited to the PM in conjunction with iSH2. Since INPP5E will deplete both PI(4,5)P2 and PIP3, accumulation of PIP3 should be prevented (Fig. 3 E). PIP3 will be degraded into PI(3,4)P2, but synthesis of PI(3,4)P2 will only be sustained if the p110 enzymes indeed directly convert PI4P into PI(3,4)P2. Compared with controls using a catalytically impaired INPP5E D556A mutant, INPP5E completely blocked the accumulation of the PIP3 biosensor at the PM after corecruitment (Fig. 3 F). However, cPHx3 initially recruited to the PM at a similar rate to control, though its synthesis was rapidly cut off when INPP5E was corecruited (Fig. 3 G). Therefore, PI(3,4)P2 synthesis cannot be sustained in the absence of PIP3 generation, even under conditions where SHIP phosphatases are not predicted to become activated. Presumably, the PIP3 5-phosphatase activity responsible comes either from basal SHIP activity through interaction of its C2 domain with acidic PM lipids or else from other INPP5 family members, which are all competent at converting PIP3 to PI(3,4)P2. Identification of the enzymes responsible will be a key question for future work. We also noted that the increase in PI(3,4)P2 after combined iSH2 and INPP5E recruitment was transient. This is most likely due to endogenous INPP4 or PTEN (see Direct hydrolysis of PI(3,4)P2 by PTEN section) degrading PI(3,4)P2. The decline could be prevented by an inhibitor, bpV(phen) (bisperoxovanadium 1,10-phenanthroline), that blocks both of these enzymes, as shown in Fig. S2 B.

24Thus far, these experiments addressed the pool of PI(3,4)P2 generated downstream of class I PI3K signaling. However, activity of class II PI3K has been shown to function in a more constitutive capacity. We therefore wanted to identify the source of PI(3,4)P2 that we observed at the PM in growing cells (Fig. 1). To this end, we employed the potent and class I PI3K-selective inhibitor GDC-0941 at 250 nM to distinguish class I and class II activities (Fig. 3 H). After inhibitor treatment for 20 min, cells were subject to chemically induced dimerization to recruit a coexpressed FKBP-INPP4B to the PM (Fig. 3 I), the goal being to degrade any remaining PI(3,4)P2. As shown in Fig. 3 I, GDC treatment led to the depletion of almost the entire pool of PM-associated PI(3,4)P2 from cells grown in serum. We cannot rule out that the small additional decrease induced in the 5 min after rapamycin addition is not an addition artifact (Fig. 3 I); area under the curve measurements for after rapamycin treatment versus the 5-min period preceding it do not show a significant change (P = 0.43, Sidak’s multiple comparisons test) as compared with the change following DMSO treatment (P < 10−4, Sidak’s multiple comparison test after two-way repeated-measures test, F = 61.9, P < 10−4). However, the data do clearly indicate that the overwhelming majority of PI(3,4)P2 present in the PM of cells is generated by the class I PI3K pathway.

25The decline in PI(3,4)P2 after GDC-0941 treatment (Fig. 3 I) began immediately after treatment with the compound and is to be expected since synthesis of its immediate precursor, PIP3, is being blocked. However, this decline was already apparent within 4 min, whereas no decline (and even a small increase) was apparent after PIP3 was depleted via recruitment of the phosphatase INPP5E in Fig. 1 M. How can this be so? Unlike GDC treatment, INPP5E leads to direct conversion of PIP3 into PI(3,4)P2, inducing an initial burst of PI(3,4)P2 synthesis as all remaining PIP3 is converted. Indeed, repeating these experiments in the presence of our PIP3 biosensor revealed that initial increases in cPHx3 at the PM mirrored the declines in PIP3 (Fig. S2 C, inset). However, imaging for extended period revealed that once PIP3 had been depleted, levels of PI(3,4)P2 began to decline (Fig. S2 C)—much as they did after recruitment of INPP5E in conjunction with iSH2 in Fig. 3 G. We did note that cPHx3 at the PM did not decline under these conditions as much as it did with GDC-0941 treatment. Possible explanations include (1) the continued activity of PIP5K and CI-PI3K after INPP5E recruitment, permitting greatly reduced but not completely blocked PI(4,5)P2, PIP3, and PI(3,4)P2 synthesis, and (2) the lack of PI(4,5)P2 at the PM yielding reduced PTEN activity.

26So far, our data support a canonical view of class I PI3K signaling, which is dominated by conversion of PI(4,5)P2 to PIP3, followed by degradation back to PI(4,5)P2 by PTEN or conversion to PI(3,4)P2 via INPP5 enzymes, most prominently SHIP1 and SHIP2. PTEN and INPP5 thus represent a bifurcation of the pathway: in the former case, toward a simple inactivation, and in the latter, the conversion to a modified but still active signaling state. PI(3,4)P2 is ultimately degraded to PI3P by INPP4A/B, a process intimately linked to endocytosis. However, it has been perplexing that the accumulation of PI(3,4)P2 seen after PI3K activation does not lead to a detectable increase in PI3P levels, implying an alternative route of degradation. Recently, Malek and colleagues have proposed that PTEN in fact directly converts PI(3,4)P2 to PI4P, terminating this signaling as well. The evidence was based on the requirement for PTEN knockout for EGF-stimulated PI(3,4)P2 accumulation, in addition to showing that a PI(3,4)P2 3-phosphatase activity in MCF10 cell cytosol was lost in PTEN nulls. However, direct evidence for hydrolysis of PI(3,4)P2 in intact, living cells is still lacking.

27The principle reason for the ongoing ambiguity over cellular activity of PTEN against PI(3,4)P2 has been the ambiguity when interpreting changes in lipid levels in cells. When PTEN loss induces PI(3,4)P2 accumulation, this can be explained by failure of PTEN to degrade PI(3,4)P2 directly—or alternatively, due to impaired PIP3 degradation, leaving more substrate available for the INPP5 enzymes (Fig. 4 A). Careful mathematical modeling suggested the latter explanation did not explain the accumulation in MCF10 cells. Nonetheless, the evidence is indirect. We therefore devised an experiment using the LepB PI3P 4-OH kinase. Since we showed that this enzyme generates an endosomal pool of PI(3,4)P2 devoid of PIP3 (Fig. 2), activity of PTEN leading to depletion of PI(3,4)P2 can be unambiguously assigned to direct hydrolysis of the lipid in this context (Fig. 4 B).

28Indeed, use of chemically induced dimerization of FKBP-PTEN to FRB-LAMP1–positive endosomes/lysosomes caused a rapid depletion of endosomal PI(3,4)P2 detected with cPHx3 in LepB-expressing COS-7 cells (Fig. 4 C). That this was truly PI(3,4)P2 to PI4P conversion induced by PTEN was further indicated by a clear accumulation of PI4P on these membranes (Fig. 4 D). Therefore, our data provide a direct demonstration of PI(3,4)P2 hydrolysis by PTEN in intact, living cells and confirm the novel role of this enzyme in terminating PI(3,4)P2 signaling in addition to signals driven by PIP3.

29In this study, we report a high-avidity probe for PI(3,4)P2, cPHx3-TAPP1, which satisfies three crucial criteria for a genetically encoded lipid biosensor: (1) binding of the PH domain is exquisitely specific for PI(3,4)P2, (2) PI(3,4)P2 binding is absolutely necessary, and (3) PI(3,4)P2 binding is sufficient for localization in living cells (Figs. 1 and 2). We use this high-fidelity and high-sensitivity probe to develop the first direct evidence for the canonical pathway of PI(3,4)P2 synthesis by class I PI3K (i.e., by PIP3 dephosphorylation; Fig. 3), first proposed nearly three decades ago. We also provide direct evidence for a much more recent proposition: that the tumor suppressor PTEN also terminates PI(3,4)P2 signals in addition to those mediated by PIP3 (Fig. 4). We also generated the first optogenetically activatable lipid kinase, LepB (Fig. 2).

30How PI(3,4)P2 signaling couples to different functions, and their relationship (or not) with the class I PI3K pathway, is still poorly understood. This has largely been due to the difficulty in studying PI(3,4)P2 in isolation from the other, more abundant phosphoinositides such as PI(4,5)P2 and PIP3. However, recent advances in mass spectrometry are allowing sensitive detection of this lipid. We anticipate that new tools for optical detection of PI(3,4)P2 in living cells will complement these approaches and greatly accelerate research in this area.

31While this manuscript was in preparation, another TAPP1-derived PI(3,4)P2 biosensor was reported, which like ours also excluded the C-terminal portion of the protein. This probe reported very similar dynamics for PI(3,4)P2 relative to PIP3 to those reported here, and the study reached the same conclusion as to the role of class I PI3K in synthesis. Although based on a single PH domain rather than the tandem trimer we report, the probe used by could detect PI(3,4)P2 in response to IGFI in NIH-3T3 cells, a feat not accomplished with previous single-domain probes. The difference likely stems from the inclusion of a hydrophobic solvatochromic fluor that penetrates the bilayer, favorably altering the partition coefficient of the probe. This fluor allows ratiometric measurement of membrane association, which is conveniently calibrated against liposomal standards for precise measurement of mole percentage. This improved quantitative capacity, though superior to our sensitive but uncalibrated measurement of membrane localization with cPHx3, is less convenient. The solvatochromic fluor must be covalently attached to recombinant protein and introduced to cells via microinjection (or perhaps electroporation). We therefore believe our simple, transfectable plasmid-based biosensor will greatly benefit efforts to dissect the cellular functions of PI(3,4)P2.

32For now, the clearest demonstration is that most PI(3,4)P2 accumulating in cells is derived from INPP5 activity on class I PI3K-synthesized PIP3. Without stimulation, resting levels of PI(3,4)P2 are extremely low, with immunocytochemical estimates suggesting that perhaps 40% of this is generated by the class II PI3K-C2α. The function of the greatly expanded class I PI3K-driven pool remains an ongoing question. Conversion of PIP3 to PI(3,4)P2 will terminate signals generated by PIP3-selective effectors, maintain signals by more promiscuous effectors (like Akt), and initiate signaling by PI(3,4)P2-selective effectors. What is the purpose of this tiered lipid signaling system? A clue comes from perhaps the best characterized PI(3,4)P2-selective effector protein: the TAPP proteins from which our probe is derived. Mice harboring point mutations in Plekha1 and Plekha2 that disrupt PI(3,4)P2-binding of the TAPP1 and TAPP2 cPHs exhibit augmented PI3K signaling through insulin and B cell receptors, suggesting a major function of the lipid is in negative feedback of class I PI3K signaling. Interestingly, PI(3,4)P2 is also implicated in endocytosis. Therefore, it could be possible that PI(3,4)P2 down-regulates PI3K signals by triggering internalization of activated receptors.

33The extent to which receptor endocytosis effectively down-regulates class I PI3K signaling is currently unclear. PI(3,4)P2 has been reported on early endosomes after PDGF stimulation of NIH-3T3 fibroblasts, where it recruits Akt2, suggesting continued PI3K-mTOR signaling. In contrast, Rab5 was shown to recruit INPP4A and thus remove PI(3,4)P2 from nascent endosomes. Moreover, we failed to observe an endosomal localization of our PI(3,4)P2 probe in serum stimulated 293A cells (Fig. S1 A) or after insulin stimulation of HeLa cells (Fig. 3 B). We also found that a previous report of TAPP1-based probe localization to endocytic structures is likely driven by a clathrin box in the C terminus of the protein (Fig. S1 A). It is of course entirely possible that PI3K signaling capacity from endocytic structures depends on the receptor and cell type. Notably, continued PI3K signaling or termination are both compatible with proposed roles for PI3K-C2α–derived PI(3,4)P2 in clathrin-coated vesicle maturation. Termination of endosomal PI3K signals must occur at some point, however, since continued PI3K-Akt-mTOR activation would conflict with PI3K-C2β–mediated mTOR down-regulation at late endosomes.

34A second major finding is that PI(3,4)P2 is directly hydrolyzed by PTEN. This result is surprising since meticulous work with purified protein found much poorer PTEN activity against PI(3,4)P2 versus PIP3. The discrepancy likely derives from the very different enzyme–substrate interactions in living cells; indeed, turnover numbers for PTEN in living cells are several orders of magnitude higher than in the test tube. It therefore seems likely that PTEN regulates PI(3,4)P2 signaling in addition to PIP3. The recent observations that PTEN reduced signaling at forming clathrin-coated structures and slows their maturation is entirely consistent with a role for PI(3,4)P2 during endocytosis. Finally, since PTEN’s activity on PI(3,4)P2 is the direct reversal of the PI4P conversion mediated by class II PI3K and since PTEN has recently been found to localize to endosomal structures, PTEN seems poised to regulate PI3K-C2β signaling from late endosomes/lysosomes too.

35In summary, we have developed and validated a sensitive genetically encoded lipid biosensor for PI(3,4)P2. We show this lipid accumulates in cells primarily in response to class I PI3K activity, via circuitous dephosphorylation of PIP3. We also provide evidence for direct hydrolysis of PI(3,4)P2 by PTEN. Collectively, these results provide fresh impetus to dissect the physiological and pathological signaling outcomes driven by PI(3,4)P2 and provide a powerful new tool to aid in this endeavor.

36The COS-7 (ATCC CRL-1651) and HeLa (ATCC CCL-2) cells were cultured in DMEM (low glucose; Life Technologies 10567022) supplemented with 10% heat-inactivated fetal bovine serum (Life Technologies 10438–034), 100 units/ml penicillin, 100 µg/ml streptomycin (Life Technologies 15140122), and 1:1,000 chemically defined lipid supplement (Life Technologies 11905031) at 37°C in a humidified atmosphere with 5% CO2. They were passaged twice per week by dissociation in TrpLE (Life Technologies 12604039) and diluting 1 in 5. 293A cells with their endogenous CLTA alleles tagged with split GFP were generated exactly as described using a protocol we have described. Briefly, Platinum Cas9 (Thermo Fisher) was precomplexed with gRNA and electroporated into 293AsfGFP1-10 cells in combination with a single-stranded HDR template (IDT). After recovery, cells were sorted by FACS for GFP-positive cells. They were cultured under identical conditions to the COS-7 and HeLa cells.

37For transfection, cells were seeded in 35-mm tissue culture dishes with 20-mm number 1.5 cover glass apertures (CellVis) coated with 5 µg fibronectin (Life Technologies 33016-015). Between 1 and 24 h after seeding, cells were transfected with 1 µg total plasmid DNA precomplexed with 3 µg lipofectamine2000 (Life Technologies 11668019) in 200 µl Opti-MEM (Life Technologies 51985091) according to the manufacturer’s instructions. Cells were imaged 18–26 h after transfection. For unnatural amino acid incorporation, medium was supplemented with 250 µM hydroxycoumarin lysine in parallel with the transfection.

38The pPAmCherry-C1 (Addgene plasmid 31929) was a kind gift of Vladislav Verkhusha (Albert Einstein College of Medicine, New York, NY; ). The EGFP (Aequorea victoria GFP with F64L and S65T mutations with human codon optimization), mCherry, Rhodopseudomonas palustris bacteriophytochrome BphP2-derived near-iRFP variant iRFP713, and Entacmaea quadricolor GFP-like protein eqFP578 variant pTagBFP2 were used in the Clontech pEGFP-C1 and N1 backbones as described previously. TAPP1 cPH and LepB sequences were obtained as synthetic gblocks (IDT). All plasmids were verified by dideoxy sequencing. Constructs generated in this study are freely distributed through Addgene. Plasmids were constructed using standard restriction cloning or NEB HiFi assembly (New England Biolabs E5520S) or else obtained from the sources indicated in Table 1.

39Rapamycin (Thermo Fisher BP2963-1) was dissolved in DMSO at 1 mM and stored in aliquots at −20°C; the final concentration used in cells was 1 µM. 4 mg/ml Insulin zinc solution (Thermo Fisher 12585014) was stored in aliquots at −20°C. The GDC-0941 (EMD-Millipore 5.09226.0001) was dissolved in 2 mM DMSO and stored in aliquots at −20°C. Aliquots of hydroxycoumarin lysine were stored at −20°C in DMSO at 100 mM.

40Cells were imaged in 1.6 ml FluoroBrite DMEM (Life Technologies A1896702) supplemented with 25 mM Hepes (pH 7.4) 1:1,000 chemically defined lipid supplement with or without 10% heat-inactivated fetal bovine serum. For treatments, 0.4 ml of this medium containing fivefold the final concentration of compound was applied to the dish (or 0.5 ml for a second addition).

41Confocal microscopy was performed on a Nikon TiE inverted stand with an A1R resonant scan head and fiber-coupled four-line excitation (Ex) LU-NV laser combiner equipped with 405-, 488-, 561-, and 640-nm lines. 8 or 16 frame averages were used to improve signal to noise. A 100× 1.45 NA plan-apochromatic oil-immersion objective was used throughout. Blue (405 nm Ex and 425–475 nm emission [Em]) and yellow/orange (561 nm Ex and 570–620 nm Em) channels were imaged concurrently, alternating with concurrent imaging of green (488 nm Ex and 500–550 nm Em), far/infrared (640 nm Ex and 663–737 nm Em), and a transmitted light channel for DIC. The hexagonal confocal pinhole was set to 1.2× Airy disc size of the longest wavelength imaged.

42For TIRFM, we used a second Nikon TiE microscope fitted with a TIRF illuminator arm fiber coupled to an Oxxius L4C laser combined equipped with 405-, 488-, 561-, and 640-nm lasers. A 100× 1.45 NA plan-apochromatic oil-immersion objective was used to deliver the high angle of incidence illuminating beam and acquire the images by epifluorescence. Images were acquired on a Zyla 5.5 sCMOS camera (Andor) with 2 × 2 binning in rolling shutter mode. Blue (405 nm) and yellow/orange (561 nm Ex) channels were imaged through a dual-pass 420- to 480-nm and 570- to 620-nm filter (Chroma), whereas green (488 nm) and far/infrared (640 nm Ex) used a dual-pass 505- to 550-nm and 650- to 850-nm filter (Chroma).

43Optogenetic activation of LepB was performed by confocal microscopy. After acquiring ∼30 s of data with 405 nm illumination set to zero power, transmission was turned up to 20% of the maximum available power from the LU-NV unit.

44For single-molecule imaging, PAmCherry was imaged without pixel binning in global shutter mode with 25-ms exposures and 30% illumination power with 561 nm and 0.8% 405 nm for photoactivation from the 100-mW Oxxius lasers. A 16 × 16-µm region of PM was imaged for tracking.

45Images in Nikon nd2 image format were imported into the open access image analysis package Fiji, using the LOCI BioFormats importer. A custom written macro was used to combine fields into a single enlarged image for the purposes of image analysis (though never for presentation).

46Regions of interest (ROI) were drawn around each cell, and a custom-written macro was used to measure the average pixel intensity in these ROI and then normalize each pixel to this value. The normalized iRFP (cytosolic) image was then subtracted from the normalized EGFP-cPH channel to yield the subtracted image.

47Cells were measured inside ROI. A second image channel (mCherry-Rab5 in Fig. 2 and LAMP1-FRB-iRFP in Fig. 4) was autothresholded and used to generate a mask to measure the normalized pixel intensity of the EGFP channel, as previously described in detail.

48The ROI corresponding to the footprint of individual cells were defined, and after subtracting camera noise, intensity levels over time were measured. Average pixel intensity in each frame t was normalized to the pretreatment average level Ft/Fpre.

49Single molecule trajectories were segmented and tracked using the open-source Fiji implementation of the u-track single particle tracking algorithm. A difference of Gaussians filter was used with a hard threshold and an estimated dimeter of 0.5 µm for the estimated diameter (i.e., an ∼8 × 8-pixel neighborhood) of single molecules. Coordinates of the trajectories were exported as XML files, and mean-square displacement at different time lags was calculated along with trajectory lifetime distributions using custom written code in Python.

50Example images were selected based on having the best signal to noise possible, while also having measured values close to the sampled population median, and are always within the central interquartile range.

51Data presentation and statistics was performed using Graphpad Prism 7. Data were subject to the D’Agostino and Pearson normality test to select for parametric or nonparametric tests.

52Fig. S1 shows that cPHx3 does not label endocytic or endosomal structures. Fig. S2 shows that PM PI(3,4)P2 is derived from PIP3 synthesized via the class I PI3K pathway (additional evidence).