Live-Cell Translocation Experiments to Identify PIP3-Regulated PH Domains
Using confocal imaging, we carried out live-cell localization and translocation studies on 130 YFP-tagged mouse PH domain constructs and two tandem PH domain constructs expressed in NIH 3T3 fibroblasts (; ~40% of predicted mouse PH domains). We stimulated the cells with PDGF to induce robust production of PIP3 at the PM and found that 27 of the single PH domain constructs and both of the tandem PH domain constructs translocated to the PM (see Figure S1
and Table S1
available online). Translocation was assessed by visual inspection and by measuring the receptor-triggered change in cytosolic over PM fluorescence intensity in multiple cells (Table S7
Genome-wide Live-Cell Imaging Identifies a Subset of Mouse PH Domains Regulated by Receptor Stimulation
shows translocation time courses of PH domains from AKT, a well-known PIP3-sensitive growth regulator; from PDK1, which was often thought to be persistently at the PM, but which we now show can translocate in living cells; and from five translocators we identified: the PH domains from the adaptor proteins SH3BP2, CNSKR2, and AKAP13, a RhoGEF FGD6, and a steroid-related protein OSBPL7.
We also found 24 PH domains that were prelocalized to the PM and remained PM localized in response to PDGF stimulation (, top panel). We did not see significant reductions of the PLCδ—or of any other constitutively PM localized—PH domain upon PDGF stimulation, indicating a lower relative activation of PLCγ compared to that of PI3K.
The majority of PH domains were cytosolic or vesicular localized and were not regulated by receptor stimulation (, bottom panel). Basal localization images of all tested mouse PH domains are shown Figure S2
, and phenotypes are summarized in Tables S2
PM Translocation of PH Domains Is Regulated by PI3K
To confirm that the translocation of the PH domains to the PM after PDGF stimulation was due to PI3K activation, we carried out three sets of control experiments. First, we verified that adding the PI3K inhibitor LY294002 dissociated the translocated PH domains from the PM ( shows four examples). Second, we showed that pretreatment of cells with 50 µM LY294002 for 3 min before stimulation prevented PDGF-induced PM translocation of the previously translocating PH domains (see Figures S9C and S9F
). Third, we coexpressed a version of the p85 regulatory subunit of PI3K that acts as a dominant-negative construct (Dhand et al., 1994
) and showed it prevented PDGF-induced PM translocation (see for two examples).
Control Experiments and Characterization of Two Translocators: SH3BP2-PH and FGD6-PH
The results of these three sets of control experiments are consistent with 26 of the translocating PH domains being PIP3 regulated. The one exception was AKAP13-PH, which translocated to the PM as an indirect effect of PIP3-induced peripheral actin polymerization (Figure S3
For all translocating PH domain constructs tested, the half-maximal dissociation time from the PM was ~100 s following LY294002 addition ( shows a time course of the SH3BP2-PH domain). We also confirmed for two of the PIP3-regulated PH domains we identified that translocation of the PH domain translates into translocation of the full-length protein as has been shown previously for AKT and PSCD. shows the receptor-triggered translocation of full-length FGD6 and SH3BP2, which suggests that the translocation of the PH domains can confer their translocation capacity to the full-length proteins.
In Vitro PI(4,5)P2 versus PI(3,4,5)P3 Binding Correlated with PM Localization and Translocation
We used lipid blot assays to compare the live-cell translocation and localization data to in vitro binding to different phosphoinositol lipids (). Table S2
contains a summary of all measured lipid blot binding interactions. Figure S2
contains bar graphs showing selected lipid blot binding values for all mouse PH domain constructs.
In Vitro Binding Selectivity of PH Domains to PI Lipids and Lipid Controls
We applied different clustering approaches to the measured binding, localization, and translocation data. Increased PI(4,5)P2 binding correlated with increased PIP3 binding () and with increased PI(3,4)P2 binding (Figure S10
In the few cases where the PH domain bound significantly more to PIP3 or PI(3,4)P2 than to PI(4,5)P2 in vitro, the PH domain did translocate (for example, AKT3 and DAPP1). Also, the PH domain from PLCδ1, which is widely used as a biosensor for PI(4,5)P2, stands out as a PI(4,5)P2-specific binder. However, other than these cases, there was no apparent correlation between the in vitro lipid binding and in vivo localization studies, and most of the PM-translocating and persistently PM-localized PH domains bound PIP3 and PI(4,5)P2 to a similar degree.
A few of the PH domains, such as the second PH domain in PLEK2 (PLEK2-PH2), showed PI(3,4)P2- over PI(3,4,5)P3-specific binding. However, we did not observe that the PLEK2-PH2 domain translocation time to the PM was significantly slower than that of more PIP3-specific lipid binders as has been reported in lymphocytes (Allam and Marshall, 2005
), suggesting that PI(3,4)P2 and PIP3 serve a similar functional role in inducing translocation for the receptor stimuli and cell type used in our study. As a technical comment, it should also be noted that some of the domains that were reported in the literature to bind PIP3 in vitro or that failed to translocate in our live-cell experiments might be a result of different flanking sequences and other construct properties or may reflect insufficiently strong receptor stimulation to identify weak interactors.
Search for a PIP3-Binding Motif that Predicts the Translocation Data
To find a sequence motif for PIP3 regulation, the PH domain sequences first needed to be correctly aligned. Using ClustalW and other multisequence alignment methods that we tried did not result in an alignment consistent with the known structural data, presumably due to the low sequence homology among PH domains (~7%–30%). We used instead a Hidden Markov Model (HMM) for PH domains (Pfam PF00169) and verified that the resulting alignment was consistent with the locations of the variable loops and ends of β strands by comparison to known structures: PDK1 (Komander et al., 2004
), PLEKHA1 (Thomas et al., 2001
), Akt (Thomas et al., 2002
), BTK (Hyvonen and Saraste, 1997
), PSCD3 (Ferguson et al., 2000
; Lietzke et al., 2000
), and DAPP1 (Ferguson et al., 2000
These known PH domain structures show that the loop region between the first and second β strands forms the core of the interaction with the phosphates of the inositol headgroup. For several of the PH domains, secondary interactions occur in the loop region between β strands 3 and 4, as well as in the loop region between β strands 6 and 7. shows an alignment of the three loop regions for selected PIP3-regulated PH domain constructs identified in our study. We highlighted in red the amino acids that have been shown to directly interact with the phosphates of the inositol headgroups. When aligning the 26 identified PIP3-regulated PH domains, we found that they belonged to 16 different subclasses with little sequence homology between them ().
Applying a Previously Published PIP3-Binding Motif to Explain the Translocation Data
By building on available structural data, a PIP3-binding motif based on eight criteria was proposed (Isakoff et al., 1998
). The main criteria in this motif were the presence of a Lys residue two amino acid positions before the end of the first β strand and in the second β strand a Lys or Arg at the second amino acid position and an Arg in the fourth amino acid position.. When applied to the sequences of the tested mouse PH domains, we found that 10 of the 26 translocating PH domains did not meet the eight criteria (summarized in Table S3
). This prompted us to search for a method that is better suited to predict PIP3 regulation.
Development of a Recursive-Learning Strategy to Predict PIP3 Regulation
We tested the idea that a probabilistic sequence comparison can distinguish between PIP3-regulated and nonregulated PH domains. The approach that we developed was based on separating the aligned sequences of the tested PH domains into two groups: PIP3-regulated PH domains and non-PIP3-regulated PH domains. For the group of PIP3-regulated PH domains, we calculated a two-dimensional sequence profile matrix, PT. This 162 × 20 matrix () contained for each of the 162 positions in the aligned sequences the probabilities to find one of the 20 amino acids. The column values at a position summed to 1, unless there was a gap at that position in one or more of the sequences, in which case the column values summed to less than 1. Analogously, we used the aligned sequences of the non-PIP3-regulated PH domains and derived a second sequence profile matrix, PNT, with the same 162 × 20 dimensions.
Development of a Recursive-Learning Strategy for Predicting PIP3 Regulation
We then made the assumption that each sequence position contributes independently to PIP3 regulation. This allowed us to use logarithms in our calculation of an overall score. We defined a recursive functional classification matrix (RFC-matrix) by ratioing the elements in the PT
matrices and taking the logarithm:
In the resulting RFC-matrix, values > 0 indicate positive correlation with PIP3 regulation and values < 0 indicate negative correlation.
This RFC-matrix facilitates the calculation of a predicted PIP3-regulation score, SRFC, for a particular PH domain. SRFC was calculated by first converting the sequence of the PH domain to be scored into a binary sequence profile matrix, Pseq, with the same dimensions as the RFC-matrix. All the values in each column of Pseq are equal to 0, except for the value corresponding to the amino acid in the sequence, which is equal to 1. The PIP3-regulation score for a particular PH domain can then be calculated by element-by-element multiplication of the two matrices, Pseq and RFC, and then summation (). The resulting RFC score was used to predict PIP3 regulation.
We first used this approach to score just the positions in the three variable-loop and flanking β strand regions that have been shown to interact with PIP3 (). While scoring these three loop regions showed an improved predictive value over the motif above for separating the non-PIP3-regulated (blue) from the PIP3-regulated (red) PH domains, there was still not a clean separation (). We then calculated an overall RFC score by allowing all amino acid positions to contribute to the score. Markedly, the RFC score based on the entire PH domain accurately separated PIP3-regulated PH domains from the non-PIP3-regulated ones ().
Validating the RFC Algorithm by Testing PH Domains from Other Species
While the RFC algorithm accurately predicted the PIP3 responses of the 130 tested mouse PH domains, a validation of the RFC algorithm required applying it to an independent set of untested PH domains. We used the algorithm to score all PH domains from human, C. elegans
, D. melanogaster
, S. pombe
, and S. cerevisiae
, as well as the mouse PH domains we had not tested in our first set of experiments. In a first test of the predictions, we cloned and tested mammalian PH domains that were not homologous to PH domains we had already tested. We successfully cloned 11 human PH domains and two additional mouse domains from the 26 sequences we targeted by PCR. Out of the seven PH domains that had high SRFC
scores and were thus predicted to be PIP3 regulated, all seven translocated ( and Figure S8
). The human PH domains with scores at or below borderline either weakly translocated (Sbf1, SRFC
=8) or did not translocate at all ( and Figure S8
Prediction of PIP3-Regulated PH Domains in Other Species
PI3K signaling has been studied extensively in Drosopohila and C. elegans, where it has been shown to control cell size, cell number, and aging. Based on the prediction of the RFC algorithm, there were five potential PIP3-regulated PH domains in Drosophila, and three in C. elegans. We tested seven Drosophila PH domains. Out of the four that the RFC algorithm predicted to be PIP3 regulated, all four of them translocated (dAkt1, dGrp1, dBtk29a, and dCG14366; ). dCG5004-PH and dCG12467-PH, which had RFC scores in the border region (SRFC = 19 and 8, respectively), and dVap7-PH, which had a very low score (SRFC = −6), all did not translocate.
In C. elegans
, we tested six PH domains. Among the three PH domains predicted to be PIP3 regulated, we identified two translocators (cAkt1-PH and cAkt2-PH; ) but found that cSec7-PH, which had a high RFC score, did not translocate. The failure of the cSec7 to be PIP3 regulated could be due to missing positive charges in the first loop that are present in the other translocating Sec7 homologs and in translocating PH domains from other species. The low-scoring C. elegans
PH domains we tested, cF59A6.5, cObr1, and cLet502, did not translocate. shows examples of low-scoring PH domains that did not translocate, and Table S6
lists all SRFC
scores for the different species.
We then used the RFC algorithm to search the S. cerevisiae
and S. pombe
genomes for PIP3-regulated PH domains. While both yeast species do not have PI3K isoforms, there is a PTEN-related putative PIP3 phosphatase in S. pombe
, and it has been suggested that PIP3 might be generated via other pathways (Mitra et al., 2004
). Almost all the yeast PH domains scored below the cutoff value for PIP3 regulation, but a few had RFC scores in the boundary region (Figure S4
and Table S6
). We cloned and tested 12 of the highest scoring S. pombe
and S. cerevisiae
PH domains but did not find evidence for PIP3-mediated translocation.
These translocation results from other species were then used to recursively update the RFC-matrix. This allowed for an even more accurate prediction of the PI3K-regulated proteome in the different species. show surveys of the predicted and the tested PIP3-regulated PH domains from C. elegans
to mammals (data for S. pombe
, S. cerevisiae
, and Dictyostelium
are shown in Figures S4 and S5
). These plots show the cumulative number of PH domains (y-axis) with values higher than a given RFC score (x-axis). The cumulative representation was chosen because it shows the distribution of the RFC scores, as well as the corresponding number of PH domains present in a particular species. shows the numbers of PH domains predicted for mouse, with tested PH domains marked by colored bars: red for PIP3 regulated, and blue for not regulated. As a control, we compared the distribution of SRFC
scores for scrambled versions of the tested mouse PH domains to the actual scores (Figure S6
). The average SRFC
score of the scrambled sequences was −5 with a standard deviation of 9. SRFC
scores > 22 become trustworthy because they are three standard deviations from the median of this random noise. This scrambled sequence analysis also defined a boundary region where the PIP3 prediction is ambiguous one to three standard deviations from the median corresponding toSRFC
scores between 4 and 22.
Analogously to the survey of the mouse domains in , shows the PH domain predictions for human and zebrafish and shows them for Drosophila and C. elegans. The observed significant match between predicted cutoff scores and experimental results suggests that the algorithm provides a useful framework to predict the PIP3-regulated proteome in different species.
Because PDGF stimulation of NIH 3T3 cells results in robust PIP3 increases at room temperature, most experiments were carried out at room temperature. However, we did confirm that there is no change in translocation phenotype at 37°C (Table S8
PIP3 Regulation Is Predicted by Amino Acids at Positions across the PH Domain, Not Just the PIP3-Binding Pocket
We next determined which positions in the aligned PH domains contribute most to predicting whether PH domains are PIP3 regulated or not. shows a two-color representation of the RFC-matrix used to score the plots in . The values are positive (red) if a particular amino acid is more frequently found at that position in the PIP3-regulated group of sequences, and negative (blue) if it is found more frequently in the non-PIP3-regulated group. The positions of the three PIP3-interacting loop regions (marked by yellow lines) and the flanking β strands (labeled in blue), as well as the sequences of AKT1 and PSCD3, have been added underneath the plot for orientation.
Structural and Evolutionary Insights into PIP3-Regulated Proteomes Derived from the RFC Algorithm
We determined the relevance of specific amino acid positions for PIP3-regulated PH domains by averaging the contribution from each position. This was done by squaring the RFC-matrix elements and adding the values for the 20 possible amino acids. The resulting averages were plotted in a bar graph (bottom of ). The height of each bar is a measure of the relevance of the position for predicting PIP3 responsiveness. The 13 positions that have the maximal effect on PIP3 regulation were marked with stars in and were overlaid on the structure of AKT1 (). These positions are more prominent near the PIP3-binding pocket, but are also scattered inside the structure and at locations far from the interaction site. Together with the lack of sequence homology among the subclasses of PIP3-regulated PH domains (), this argues that PIP3 regulation of different subclasses is based on amino acids located across the PH domain.