Macropinocytosis is an actin-driven process of large-scale and non-specific fluid uptake used for feeding by some cancer cells and the macropinocytosis model organism Dictyostelium discoideum. In Dictyostelium, macropinocytic cups are organized by ‘macropinocytic patches’ in the plasma membrane. These contain activated Ras, Rac and phospholipid PIP3, and direct actin polymerization to their periphery. We show that a Dictyostelium Akt (PkbA) and an SGK (PkbR1) protein kinase act downstream of PIP3 and, together, are nearly essential for fluid uptake. This pathway enables the formation of larger macropinocytic patches and macropinosomes, thereby dramatically increasing fluid uptake. Through phosphoproteomics, we identify a RhoGAP, GacG, as a PkbA and PkbR1 target, and show that it is required for efficient macropinocytosis and expansion of macropinocytic patches. The function of Akt and SGK in cell feeding through control of macropinosome size has implications for cancer cell biology.
Macropinocytosis is an ancient process used by cells to take up large amounts of fluid (King and Kay, 2018; Swanson, 2008). Actin-driven protrusions are extended in a cup-shape and then closed to engulf extracellular medium into an internal vesicle (Buckley and King, 2017). Macropinocytosis occurs in a variety of mammalian cell types and clinically important settings, perhaps most notably in certain cancer cells as a way of obtaining nutrients (Bloomfield and Kay, 2016; Commisso et al., 2013; Kim et al., 2018; Palm et al., 2017).
Macropinocytic and phagocytic cups are formed by a ring of protrusive F-actin under the plasma membrane that is distinct from other large F-actin structures, such as pseudopods. Several cellular components involved in the organization of these cups have been identified in mammalian and Dictyostelium discoideum cells, the most important of which seem to be Ras, Rac1 and the phospholipid PIP3 [phosphatidylinositol (3,4,5)-trisphosphate in mammalian cells, plasmanylinositol (3,4,5)-trisphosphate in Dictyostelium (Clark et al., 2014)] (and, by extension, phosphoinositide 3-kinase, PI3K and PTEN, which regulate PIP3) (Araki et al., 2007; Bar-Sagi and Feramisco, 1986; Fujii et al., 2013; Hoeller et al., 2013; Ridley et al., 1992; Veltman et al., 2016; Yoshida et al., 2009).
The proteins organizing macropinocytic cups are better known as members of both the growth factor signalling cascade and as oncogenes or tumour suppressors. Growth factors signal through receptor tyrosine kinases (RTKs), which recruit PI3Ks to the plasma membrane and activate downstream effectors such as Ras (Hu et al., 1992; Margolis and Skolnik, 1994). Active Ras, in turn, interacts with class 1 PI3Ks through their Ras-binding domain and activates them to produce PIP3, an interaction that is critical for the growth of certain tumours (Castellano et al., 2013; Gupta et al., 2007). PI3K activation leading to macropinocytosis can also occur independently of Ras (Palm et al., 2017). Activating mutations in the RTK/Ras signalling pathway occur in nearly half of cancers, and activating mutations of the PI3K pathway, mostly PI3Ks and PTEN, in a third, although these groupings include proteins not involved in macropinocytosis (Kandoth et al., 2013; Sanchez-Vega et al., 2018). Ras proteins are especially frequently mutated with, strikingly, ∼95% of pancreatic cancers driven by K-Ras (Fernandez-Medarde and Santos, 2011; Kandoth et al., 2013; Prior et al., 2012; Waddell et al., 2015). Furthermore, loss of the RasGAP NF1 leads to increased Ras activation and tumour development (Bollag et al., 1996; Gutmann et al., 2017).
Macropinocytosis in Dictyostelium is highly analogous to macropinocytosis in mammalian cells; however, the situation is simplified as growth factor signalling and RTKs are not present. In Dictyostelium, macropinocytosis is used for nutrient acquisition and accounts for the vast majority of fluid uptake, making accurate quantification simple (Hacker et al., 1997; Williams and Kay, 2018a). Additionally, both forward and reverse genetic approaches can be employed (Bloomfield et al., 2015; Paschke et al., 2018).
The Dictyostelium macropinocytic cup is formed around a template ‘macropinocytic patch’ composed of activated Ras and Rac, PIP3 and F-actin, with a rim of the Arp2/3 activators SCAR/WAVE and WASP (Hoeller et al., 2013; Veltman et al., 2016). Other known components are Coronin, the myosin-I proteins and certain formins (Brzeska et al., 2016; Hacker et al., 1997; Junemann et al., 2016). The PIP3-phosphatase PTEN is excluded from the macropinocytic patch, but is present on the rest of the plasma membrane (Hoeller et al., 2013; Iijima and Devreotes, 2002).
PIP3 is vital for efficient macropinocytosis in both amoebae and mammalian cells (Araki et al., 1996; Hoeller et al., 2013; Williams and Kay, 2018b). PIP3 acts by recruiting PIP3-binding proteins to the plasma membrane, often through a PH domain. A considerable number of these proteins exist but which are important for micropinocytosis, and what function they carry out, is not known (Park et al., 2008; Zhang et al., 2010). The Akt protein kinases are oncoproteins that are major downstream effectors of PIP3 in growth factor signalling and mammalian target of rapamycin complex 1 (mTORC1) activation (Dibble and Cantley, 2015; Staal et al., 1977). We examine the role for Akt protein in Dictyostelium macropinocytosis.
Dictyostelium has a single Akt protein, PkbA, with a PIP3-binding PH domain that recruits it to the plasma membrane (Meili et al., 1999; Tanaka et al., 1999). PkbA phosphorylates target proteins at the Akt consensus sequence (RxRxxS/T) (Alessi et al., 1996; Kamimura et al., 2008; Liao et al., 2010). In addition, there is a variant kinase, PkbR1, which was first described as an Akt kinase and is commonly referred to as one, but is more similar to the highly related SGK kinases (Goldberg et al., 2006). PkbR1 is constitutively targeted to the plasma membrane by lipid modification, phosphorylates an overlapping set of target proteins with PkbA and is activated in the same PIP3-dependent manner by TORC2 and phosphoinositide-dependent kinase-1 (PDK1, also known as PDPK1 in mammals) proteins, similar to other SGK kinases (Alessi et al., 1996, 1997; Jacinto et al., 2006; Kamimura and Devreotes, 2010; Kamimura et al., 2008; Kobayashi and Cohen, 1999; Liao et al., 2010; Meili et al., 2000; Murray et al., 2005; Sarbassov et al., 2005; Stephens et al., 1998; Stokoe et al., 1997).
We show here that PkbA and PkbR1, and their activating protein kinases, are required for efficient macropinocytosis. They act downstream of PIP3 to increase macropinosome size, and hence fluid uptake, by increasing the macropinocytic patch size. We identify some of their targets by phosphoproteomics, including the RhoGAP GacG, which is essential for efficient macropinocytosis and without which cells also have reduced macropinosome and macropinocytic patch size.
PkbA and PkbR1 are together required for efficient fluid uptake by macropinocytosis
Fluid uptake by a Dictyostelium mutant lacking all five Ras-activated PI3Ks (denoted PI3K1-5-) is practically abolished in shaking culture (Hoeller et al., 2013). However, some mutant strains have a conditional fluid uptake defect in these conditions, which is not apparent when cells are attached to a surface (Novak and Titus, 1997). Additionally, fluid uptake by poorly macropinocytosing cells can sometimes be increased by growing them in medium supplemented with 10% fetal bovine serum (FBS) (Bloomfield et al., 2015). We therefore re-assessed fluid uptake by surface-attached PI3K mutants by using flow cytometry to measure uptake of fluorescent dextran. Fluid uptake was nearly abolished in double-mutants for PI3K1 and PI3K2 (PI3K1-2-) and other multiple PI3K mutants (Fig. 1A), and this was not improved by supplementing the medium with 10% FBS (Fig. S1A).
The protein kinases PkbA (an Akt kinase) and PkbR1 (an SGK kinase) can be activated downstream of PIP3 so could be effectors of PIP3 during the process of macropinocytosis. In vegetative cells, PkbR1 is present on the whole membrane, including macropinocytic cups (Movie 1). Similar to Akt in mammalian cells, PkbA is recruited specifically to macropinocytic cups, although this occurs most strongly after the cup is initially formed (Movie 2) (Yoshida et al., 2015).
Single and double knockouts of PkbA and PkbR1 (denoted PkbA-, PkbR1- and PkbA-/PkbR1-, respectively) were made in our standard laboratory strain, Ax2 (Ka) (Fig. S1B) using conditions that avoid the need for macropinocytosis-dependent growth (Paschke et al., 2018). Phosphorylation of Akt substrates detected by western blotting was largely unaffected in the single knockouts and in the PI3K1-5- strain but was nearly abolished in the PkbA-/PkbR1- strain (Fig. S1C). This is likely explained by redundancy between PkbA and PkbR1, and PIP3-independent activation of PkbR1 (Kamimura et al., 2008).
Fluid uptake was modestly reduced in the PkbA- and PkbR1- single mutants, in agreement with previous data for PkbA- cells (Rupper et al., 2001). However, it was almost abolished in the double mutant, showing that PkbA and PkbR1 are together essential for effective fluid uptake (Fig. 1B). This also held true in medium supplemented with 10% FBS (Fig. S1D). The PkbA-/PkbR1- mutant phenocopies the PI3K1-5- strain, consistent with PkbA and PkbR1 functioning redundantly as downstream effectors of PIP3. Phenotypes for both strains are henceforth assessed in HL5 medium containing 10% FBS, to guard against possible artefacts due to nutrient starvation.
Since PkbA and PkbR1 are together essential for efficient fluid uptake, we asked whether their activating kinases, TORC2 and PDK1, are also required. TORC2 has four subunits: Tor, PiaA/Rictor, Rip3/Sin1 and Lst8, all of which, excluding Tor, can be knocked out in Dictyostelium. These mutants have decreased, but not abolished, activation of PkbA and PkbR1 during chemotaxis (Lee et al., 2005). Fluid uptake was decreased in TORC2 subunit mutants (Fig. 1C), although not to the same extent as in PkbA-/PkbR1- cells, and was largely rescued by supplementation of the medium with 10% FBS (Fig. S1E). As these cells can proliferate in HL5 medium, further assessment of these cells was performed using HL5.
The fluid uptake defect in PkbA-/PkbR1- cells can be substantially, but not entirely, overcome by overexpression of PkbA. However, this rescue was abrogated when the TORC2 site (T435) was mutated to an alanine residue, showing the importance of this phosphorylation (Fig. 1D). In contrast, mutation of the equivalent site in PkbR1, T470, to an alanine residue had no effect on rescue, although the rescue was relatively small. This indicates that PkbR1 activity in the context of macropinocytosis is independent of phosphorylation by TORC2.
Our observation of a fluid uptake defect in the TORC2 subunit mutants is in contrast to what was found in previously published work, where no defect was seen (Rosel et al., 2012). Possible explanations for this difference could be the strain background or previous suppression of the phenotype resulting from prolonged growth in axenic conditions. The fluid uptake defect is, however, consistent with the inhibition of fluid uptake by the Tor inhibitor torin 1 (which inhibits TORC1 and TORC2), while rapamycin (which inhibits TORC1 only) does not inhibit fluid uptake (Rosel et al., 2012; Williams and Kay, 2018b).
Dictyostelium has two PDK1 proteins: PdkA and PdkB. PdkA, but not PdkB, is recruited to PIP3-containing membrane regions, such as macropinosomes, in a PI3K-dependent fashion, although binding to PIP3 has not been shown in vitro (Kamimura and Devreotes, 2010; Liao et al., 2010). PdkA could, therefore, be responsible for PIP3-dependent activation of PkbA and PkbR1, like a classical PDK1 protein, with PdkB responsible for PIP3-independent activation of PkbR1 (Alessi et al., 1997; Kamimura et al., 2008; Stokoe et al., 1997). In agreement with this model, fluid uptake by PdkA- mutants was nearly abolished, similar to PI3K1-5- and PkbA-/PkbR1- cells, but was actually increased in PdkB- cells (Fig. 1E). These defects were somewhat ameliorated in medium containing 10% FBS (Fig. S1F). As PdkA- cells can proliferate in non-supplemented HL5 medium, further experiments were performed in this medium.
Mutation of the PDK1 site in PkbA (T278) to an alanine residue resulted in a protein that failed to rescue the fluid uptake defect of PkbA-/PkbR1- cells (Fig. 1F), showing the importance of phosphorylation at this site for macropinocytosis. For PkbR1, in contrast to the seemingly dispensable TORC2 phosphorylation, and consistent with the fluid uptake defect of PdkA- cells, mutation of the PDK1 site (T309) to an alanine residue also prevented rescue (Fig. 1F).
These results show that PkbA and PkbR1 and their activators, TORC2 and PdkA, mediate efficient fluid uptake by macropinocytosis in Dictyostelium. PIP3-independent activation of PkbR1 by PdkB seemingly has no function in macropinocytosis. The defect in PkbA-/PkbR1- cells is as severe as in PI3K1-5- cells, and thus could account for their phenotype, though additional PIP3 effectors are not excluded.
PkbA-/PkbR1- cells only have proliferation defects in axenic conditions
Cells utilize macropinocytosis to obtain nutrients for proliferation in liquid medium. As expected from their fluid uptake defect, PkbA-/PkbR1- cells do not proliferate in HL5 medium, either when surface-attached (Fig. 2A) or in shaking suspension (data not shown), consistent with an earlier report using an independently isolated mutant (Meili et al., 2000). Curiously, PkbA- cells have a larger proliferation defect than their fluid uptake defect would suggest. Further investigation revealed that the rate of macropinocytosis decreases in these cells upon prolonged incubation in HL5 medium (Fig. S2A,B), for unknown reasons. Supplementation of the medium with 10% FBS improved the proliferation rates of the PkbA and PkbR1 single mutants, and allowed proliferation of the double mutant (Fig. 2B).
In contrast, cells grown on bacteria obtain nutrients by phagocytosis and have low rates of fluid uptake by macropinocytosis. On bacteria the PkbA/PkbR1 mutants all proliferate at approximately the same rate as the parental Ax2 strain (Fig. 2C), indicating that cellular processes necessary for proliferation are unaffected. As expected, PkbA-/PkbR1- and PI3K1-5- cells were of a comparable size to Ax2 cells when grown on bacteria (Fig. S2C), although they were slightly smaller when in growth medium (Fig. S2D).
Macropinocytosis is strongly upregulated when cells are transferred from bacteria to growth medium, and potentially a failure in this process could underlie the fluid uptake defect in PkbA-/PkbR1- cells. However, we found that PkbA-/PkbR1- cells upregulate macropinocytosis in a similar, although much reduced, way to what is seen in parental Ax2 cells when transferred from growth on bacteria to HL5 medium plus 10% FBS (Fig. 2D).
The normal proliferation of PkbA-/PkbR1- cells on bacteria and our inability to restore normal rates of fluid uptake and proliferation to them in any growth medium tested indicate that they have a specific defect in macropinocytosis, rather than in general cell physiology.
Macropinocytic patches appear to be normally organized in PkbA-/PkbR1- cells
To investigate the underlying cause of the fluid uptake defect in PkbA-/PkbR1- cells, we examined the organization of their macropinocytic patches. Patches of active Ras and PIP3 were still formed in the mutant (Fig. 3A), and PTEN was excluded from them (Fig. 3B), as expected. F-actin (Fig. 3C) and activated Rac (Fig. 3D) localization, relative to active Ras, was also unperturbed in the mutant. SCAR/WAVE localization was examined using basal patches of active Ras, a surrogate for macropinosomes, as the marker is weak and easily bleached (Veltman et al., 2016) (Fig. 3E). SCAR/WAVE was still enriched at the patch periphery. When this was quantified, no decrease in SCAR/WAVE enrichment at the ring was observed for the PkbA-/PkbR1- cells nor, in this instance, for the PI3K1-5-, contrary to expectations (Veltman et al., 2016).
We could not identify any differences in the organization of macropinocytic patches in the PkbA-/PkbR1- cells, suggesting the reduced fluid uptake is not due to aberrant patch organization.
PkbA and PkbR1 increase the size of macropinocytic patches
We used microscopy to examine the rate at which macropinosomes form and their size. Macropinosomes form at a reduced rate in the PI3K1-5- and PkbA-/PkbR1- strains compared to in the Ax2 strain (Fig. 4A), although this is not statistically significant and cannot account for the entire fluid uptake defect. We ruled out the possibility that these macropinosomes were rapidly recycled to the extracellular medium by using a short uptake time (Fig. S3A).
Instead, we find that PI3K1-5- and PkbA-/PkbR1- mutants form smaller macropinosomes. Macropinosome diameter at internalization was reduced by ∼40% in both cases (Fig. 4B), corresponding to a 3–4-fold decrease in macropinosome volume and thus accounting for a large proportion of the decrease in fluid uptake. A similar reduction in macropinocytic patch size was observed (Fig. 4C), indicating a defect in macropinocytic patch size leads to smaller macropinosomes.
The size of particles that cells can phagocytose correlates with the size of macropinocytic patches formed (Bloomfield et al., 2015; Williams and Kay, 2018b). We measured phagocytosis of various size particles by PkbA-/PkbR1- and PI3K1-5- cells. The PkbA-/PkbR1- and PI3K1-5- mutants were less proficient than Ax2 at uptake of yeast (a comparatively large particle) (Fig. S3B), while phagocytosis of smaller beads was unchanged (Fig. S3C), consistent with the smaller macropinocytic patches made by these mutants.
The TORC2 subunit knockout and PdkA- mutants have similar phenotypes to the PI3K1-5- and PkbA-/PkbR1- cells. Macropinosome formation was reduced, but not to a statistically significant extent, in all the TORC2 subunit mutants, but was significantly reduced for PdkA- cells (Fig. 4D). For all these mutants, macropinosome size was decreased (Fig. 4E). However, although PdkA- mutants had a corresponding decrease in macropinocytic patch size, the TORC2 subunit mutants were largely unaffected in this regard (Fig. 4F). This data is summarized in Fig. 4G.
PI3K1-5- and PkbA-/PkbR1- cells make significantly smaller macropinocytic patches and macropinosomes than their parent, which (due to the cubic relation of linear dimensions and volume) leads to a much larger reduction in macropinosome volume. These results further imply that there is a local positive-feedback loop in which PkbA and PkbR1 increase the activity of upstream macropinocytosis components, such as active Ras, thus enlarging macropinocytic patches.
Identification of PkbA/PkbR1 targets by phosphoproteomics
The postulated feedback loop between upstream macropinocytosis components and PkbA/PkbR1 might operate indirectly and by many different potential routes, but ought to be mediated by targets of both PkbA and PkbR1. We therefore used an unbiased phosphoproteomic screen to identify such targets.
We compared the phosphoproteomes of parental Ax2 cells, PkbA-, PkbR1- and PkbA-/PkbR1- mutants and PkbR1- cells treated with LY294002 to inhibit PI3K and hence PkbA. Phosphopeptides were isolated from these cells as described in the Materials and Methods and the relative amounts of each phosphopeptide in each sample were compared by tandem mass tag mass spectrometry over three biological replicates (Table S1). We selected phosphopeptides that were >4-fold decreased in the PkbA-/PkbR1- mutant and >2-fold decreased in the PkbR1- plus LY294002 cells compared to in the Ax2 cells in at least two of the three repeats. A total of 61 peptides from 45 proteins met these criteria.
The kinases likely to be responsible for specific phosphorylations of the filtered set of peptides were predicted using NetPhos 3.1 (Blom et al., 2004). The 16 proteins likely to be phosphorylated by PkbA/PkbR1 (i.e. those with an Akt consensus motif) are shown in Table 1, while the remainder, which are more likely to be phosphorylated by other kinases are in Table S2. Of the potential PkbA/PkbR1 targets, several are predicted to function in signalling, G-protein regulation, intracellular trafficking and digestion (Fig. 5A). Gene Ontology (GO) enrichment analysis of the identified proteins revealed strong enrichment for terms associated with G-protein function, intracellular transport and phosphoinositide binding (Fig. 5B; Table S3) (Ashburner et al., 2000; The Gene Ontology Consortium, 2017). Notably, the terms that contained more than one protein were largely concerned with G-protein regulation. These predicted functions are consistent with proteins involved in cellular feeding by macropinocytosis. Additionally, several of these proteins have previously been identified as PkbA/PkbR1 targets, validating the approach (Charest et al., 2010; Kamimura et al., 2008; Tang et al., 2011).
To assess the extent to which PkbA and PkbR1 phosphorylate overlapping substrates, we identified phosphopeptides that, compared to Ax2, were found >4-fold decreased in one mutant but not the other in at least two of the three repeats. Only a few specific targets were found: KrsB and Kif6 appear to be PkbA-specific targets while DDB_G0278629 and GefS appear to be PkbR1 specific (Table S4).
Since the PkbA/PkbR1-mediated increase in macropinocytic patch size most likely occurs through signalling events, we focused on the targets involved in G-protein function, protein complex scaffold activity and protein kinases. Pats1 mutant (Pats1-) cells have been previously shown not to have an axenic growth defect, so are unlikely to have a macropinocytosis defect and are not analysed further (van Egmond and van Haastert, 2010). A mutant of the scaffold protein (ScaA-) was obtained from the Dictyostelium stock centre (Fey et al., 2013; Sawai et al., 2008), while knockout mutants were made of GacG, GacH, GefS and KrsB (denoted GacG-, GacH-, GefS- and KrsB-, respectively) (Fig. S4). We were unable to make a GxcX-knockout mutant.
Fluid uptake of these mutants was measured in HL5 medium (Fig. 5C) and HL5 medium containing 10% FBS (Fig. 5D) and found to only be decreased in ScaA- cells (∼50%) and GacG- cells (>90%, similar to PI3K1-5- and PkbA-/PkbR1-). Unsurprisingly mutants of KrsB and GefS, which are downstream of only one of the kinases, did not show a fluid uptake defect. That no defect was found for knockouts of the other proteins may be due to redundancy, a compensatory change in macropinosome formation or size, or a lack of involvement. ScaA is part of the Sca1 complex, which has previously been shown to activate TORC2, but is re-localized following phosphorylation by PkbA/PkbR1, limiting PkbA/PkbR1 activation (Charest et al., 2010). We therefore focused on the RhoGAP, GacG.
GacG- mutants have smaller macropinocytic patches
Consistent with the large fluid uptake defect and a potential role downstream of PkbA and PkbR1, GacG- cells are unable to grow in HL5 medium, despite growing relatively normally on bacteria (Fig. 6A; Fig. S5A). They can however grow slowly in HL5 plus 10% FBS (Fig. 6A).
Underpinning the fluid uptake defect are fewer and smaller macropinosomes (Figs 6B,C; Fig. S5B), similar to what is seen for the PI3K1-5- and PkbA-/PkbR1- mutants. Unlike those mutants, however, GacG- cells were much more motile (Fig. S5C–F) (Nichols et al., 2019). The excessive motility made direct measurement of macropinocytic patch size difficult because the same markers are recruited to pseudopods, albeit more faintly, making it difficult to distinguish the two under the conditions used. As a proxy, we instead assessed yeast uptake, finding that this was reduced in GacG- cells to a similar extent to in PI3K1-5- and PkbA-/PkbR1- cells (Fig. 6D; Fig. S3B). Although there was a slight decrease in phagocytosis of smaller particles, this was not significant (Fig. S5G), indicating GacG- cells are less able to phagocytose larger particles. This suggests that GacG- cells, like PI3K1-5- and PkbA-/PkbR1- cells, are defective at making large macropinocytic patches. While overexpression of untagged GacG could rescue the GacG- fluid uptake defect, albeit poorly (Fig. S5H), GacG tagged N- or C-terminally with GFP failed to rescue the fluid uptake defect, preventing a meaningful analysis of GacG localization.
As GacG is predicted to be a RhoGAP, it is likely to function by inactivating a Rho protein. The principal Rho protein involved in macropinocytosis is Rac1, which must be inactivated for macropinosome internalization in mammalian cells and likely also in Dictyostelium (Dumontier et al., 2000; Fujii et al., 2013). However, there was no sign of stalled macropinosomes failing to close in GacG- mutants, as would be expected if GacG acted on Rac1. Additionally, patches of active Rac completely colocalize with active Ras patches in GacG- cells, as in Ax2 cells (Figs 6E and 3D). Taken together, these data indicate that GacG does not act on Rac1.
GacG- cells take up much less fluid than Ax2 cells due to a decrease in macropinosome formation rate and size because of the smaller macropinocytic patches. These phenotypes are very similar to that of the PkbA-/PkbR1- and PI3K1-5- mutants, suggesting that GacG lies in the same molecular pathways as a required downstream effector.
In this work we show that, together, the protein kinases PkbA and PkbR1 are nearly essential for fluid uptake in Dictyostelium. The defect in fluid uptake by PkbA-/PkbR1- cells is as great as that for cells lacking all Ras-activated PI3Ks. In both mutants, this is largely due to a decrease in macropinosome volume due to a reduction in the size of the macropinocytic patches. As PkbA and PkbR1 act downstream of PIP3, lack of activation of PkbA and PkbR1 could account for the phenotype of the PI3K mutants, although a role for other PIP3-binding proteins is not excluded (Fig. 7).
Macropinocytic cups form spontaneously in the plasma membrane of Dictyostelium cells, without the need for receptor stimulation. Therefore, the role of the protein kinases PkbA and PkbR1 in macropinocytosis is unlikely to be in conventional signal transduction to activate macropinocytosis. The evidence presented here shows that a major function of PkbA and PkbR1 following their recruitment/activation at PIP3 patches is to make these cups larger, and so able to engulf more liquid or, in the case of phagocytosis, take up larger particles. This role, in which PkbA and PkbR1 activation leads to a local increase in activated Ras (as indicated by larger macropinocytic patches) and hence their activator, PIP3, suggests that PkbA and PkbR1 are part of a local positive-feedback loop.
Our search by phosphoproteomics for targets phosphorylated by PkbA and PkbR1 yielded a small list of proteins. Interestingly, several of these are potentially involved in endosome trafficking (microtubule motors, a Rab GTPase and a lysosomal protein were identified), suggesting that PkbA and PkbR1 may have further functions after macropinosome internalization. Focusing instead on proteins potentially involved in a feedback loop to the macropinocytic patch, we showed that mutants of the RhoGAP GacG have extremely poor fluid uptake. As observed for the PkbA-/PkbR1- and PI3K1-5- cells, there was a decrease in macropinosome formation and patch size, suggesting that GacG is an important downstream effector of PkbA and PkbR1 during macropinocytosis.
How GacG affects macropinocytic patch size and formation, and how this is related to the increased motility of the mutant remains to be determined. Although GacG does not appear to inactivate Rac1, other Rho proteins are either present at, or have effects on, the macropinocytic patch (Lee et al., 2003; Rivero et al., 1999; Seastone et al., 1998; Somesh et al., 2006). It is possible that, by switching off one or more of these Rho proteins, a RasGEF could be activated or a RasGAP inhibited, either directly or indirectly, limiting the activation of the Ras-PIP3 pathway (Fig. 7). This would allow cytoskeletal resources to be devoted to forming pseudopods rather than macropinosomes, similar to wild isolates and axenic cells grown on bacteria or entering development, resulting in the observed increase in motility (Fisher et al., 1989; Veltman, 2015; Veltman et al., 2014; Williams and Kay, 2018b). Alternatively, the increase in motility observed in the mutant could be due to an unrelated function of GacG in limiting motility.
In mammalian cells, activating Ras mutations and increased PIP3 production allow macropinocytosis-based feeding, resulting in increased proliferation (Commisso et al., 2013; Kim et al., 2018). Work with Dictyostelium implicates several further proteins of the Ras/PI3K pathway in organizing micropinocytosis; apart from Ras itself, the RasGAP NF1, PI3K, PTEN and now an Akt (PkbA) and an SGK (PkbR1) protein kinase, along with their activating kinases, TORC2 and PDK1, are involved. The present work suggests that activating mutations of Akt, and possibly SGK, could drive macropinocytic feeding in mammalian cells, increasing the size of macropinosomes and hence nutrient acquisition. Consistent with this hypothesis, recent studies have shown Akt to be activated at sites of macropinocytosis, while inhibition of macropinocytosis inhibits Akt activation (Chiasson-MacKenzie et al., 2018; Erami et al., 2017; Yoshida et al., 2018). Akt inhibition does not depress macropinosome formation in macrophages, although these studies did not rule out smaller macropinosomes and reduced fluid uptake (Pacitto et al., 2017; Yoshida et al., 2015). Macropinocytic uptake of collagen by hepatic stellate cells, measured quantitatively by flow cytometry, is reduced by Akt inhibition (Bi et al., 2014). Whether this is due to smaller macropinosomes was not investigated but is a tantalising possibility.
A possible effect on macropinosome size has meaningful implications for the functions of Akt and SGK in tumour growth. The involvement of Akt proteins in the proliferation of cancer cells and tumour development is clear from knockout and inhibitor studies, including in mice with overactivated Ras or PTEN deficiency (Chen et al., 2006; Skeen et al., 2006; Sommer et al., 2013). SGK has been shown to act redundantly with Akt proteins in proliferation of breast cancer cells (Sommer et al., 2013). Decreased proliferation due to decreased Akt activity is usually viewed as being due to decreased mTORC1 activity, but this work suggests a second pathway where Akt and SGK protein kinases increase nutrient acquisition by macropinocytosis, and leads to the prediction that dual inhibition of SGK and Akt kinases would prevent cells feeding by macropinocytosis. A downstream target in mammalian cells similar to GacG would also be a potentially interesting and effective target for a macropinocytosis-specific inhibitor.
MATERIALS AND METHODS
Dictyostelium culture conditions
Dictyostelium discoideum Ax2 (Ka) cells and derivatives were used in this study (see Table S5) and were cultivated on Klebsiella aerogenes bacteria on SM plates at 22°C unless otherwise stated. Nutrient media [HL5 (Formedium, UK), HL5 plus 10% FBS and simple uprgulation medium (SUM) (Williams and Kay, 2018b)] were supplemented with dihydrostreptomycin (100 µg/ml), ampicillin (100 µg/ml) and kanamycin (50 µg/ml). Strains were stored frozen in horse serum plus 7.5% DMSO in liquid nitrogen and scraped out using a 16G hypodermic needle (Becton Dickinson) onto K. aerogenes and SM agar when needed. Typically, cells were passaged once a week for 4–5 weeks before being refreshed. Except where indicated, we used a standardized set of mutants created in our laboratory strain of Ax2, most of which were made using transformation conditions that did not depend on axenic proliferation, reducing the risk of suppressing mutations.
Cell uptake measurements
All cell uptake measurements were performed as described (Williams and Kay, 2018a,b). Briefly, cells were cultivated on SM agar with K. aerogenes bacteria, harvested, washed free of bacteria and inoculated in the indicated nutrient medium to 105 cells/ml, with triplicate 50 μl samples in wells of a 96-well plate. This was incubated for 24 h at 22°C before 50 μl of the same medium containing 1 mg/ml TRITC–Dextran (155 kDa, Sigma-Aldrich) was added for 1 h unless otherwise specified. This was then decanted, the plate washed by submersion in ice-cold KK2 buffer (16.6 mM KH2PO4, 3.8 mM K2HPO4, pH 6.1) and the cells detached using 100 µl 5 mM sodium azide dissolved in KK2MC (KK2 plus 2 mM MgSO4 and 100 µM CaCl2). The fluorescence of each cell in the samples was then measured with an LSR_II flow cytometer (BD Biosciences) and analysed using FlowJo. For the rescue experiments, cells were cultivated in KK2MC containing two optical density at 600 nm (OD600 nm) units of Ka and selection marker in six-well plates, washed three times in situ using KK2MC and inoculated in HL5 medium plus 10% FBS containing selectable marker in a fresh six-well plate. Samples were taken from this plate for up to 3 days after inoculation for fluid uptake measurement.
For measurement of phagocytosis, cells were harvested from bacteria as above. In the case of the 1.5 µm beads (Polysciences) uptake, the cells were diluted into KK2MC and plated as before, and allowed to settle for ∼30 min. 1.5 μm YG beads were washed free of sodium azide and diluted into KK2MC to 2×108 beads/ml. 50 μl of this was added to each well for 20 min, after which cells were washed, detached and internalized fluorescence measured as above and analysed as described previously (Sattler et al., 2013). For yeast uptake, the resuspended cells were placed into 5 ml conical flasks at 5×106 cells/ml and shaken for ∼30 min before addition of 107 sonicated TRITC-conjugated yeast particles/ml. 200 µl samples were taken at 0 and 60 min, mixed with 2 µl 0.4% Trypan Blue (Sigma-Aldrich) by shaking for 3 min to quench the fluorescence of uninternalized yeast, washed twice in KK2 plus 10 mM EDTA and the fluorescence determined on a Perkin Elmer LS 50 B fluorimeter (excitation at 544 nm, emission at 574 nm, each with a 10 nm slit).
Transformation of Dictyostelium
Vectors were made according to standard procedures, propagated using Escherichia coli XL10 and isolated using the ZR miniprep classic kit (Zymogen).
Approximately 106 Dictyostelium cells were harvested from growth zones on bacterial plates into 1 ml H40 buffer, pelleted by centrifugation at max speed (20,000 g) on a benchtop centrifuge for 10 s and resuspended in 100 µl H40 buffer (Paschke et al., 2018). Vector was added to the cells [500 ng overexpression vector (Table S6) and 2 µg linearized knockout vector (Table S7)] and they were chilled on ice with 2 mm electroporation cuvettes (SLS). The cells were transferred to the 2 mm cuvettes then subject to square wave electroporation in a Bio-Rad GenePulser Xcell (2×350 V for 8 ms, with a 1 s gap) and added to 2 ml KK2MC plus 2 OD600 nm K. aerogenes in a 3.5 cm dish to recover for 5 h.
After recovery, for cells transformed with an overexpression vector, the appropriate selection antibiotic (10 µg/ml G418, although this was doubled when working with the PkbA-/PkbR1- cells due to poor marker expression, and 100 µg/ml hygromycin) was added to the dish and the dish swirled to ensure an even distribution. In the case of knockout transformations, two 50 ml falcon tubes containing 30 ml KK2MC plus 2 OD600 nm K. aerogenes and selection antibiotic were prepared, and the dish split between the two tubes (200 μl and 1.8 ml). These were vortexed to mix and 150 µl put into each well of two 96-well plates per tube, and incubated at 22°C. Transformants were typically obtained after 4 days (overexpression) and 6–8 days (knockout).
Screening for mutants
Confluent wells of cells resistant to drug were obtained. To screen these, the medium in the well was pipetted up and down to resuspend the cells and 2 μl from the well was added to 20 μl lysis buffer with 20 μg/ml freshly added proteinase K in a 0.2 ml PCR tube (Paschke et al., 2018). The tube was vortexed to mix, then the proteinase K was inactivated by heating to 95°C for 1 min. 2 µl of DNA was added to a 25 µl PCR reaction to screen for sequence from the resistance cassette to outside the construct, giving a product that could only derive from a mutant (primer sequences are shown in Table S8).
Once wells containing mutants were identified, these were plated clonally onto SM plates with K. aerogenes and individual colonies were screened (colony appearance usually took 4 days), with DNA this time being prepared using a g-DNA miniprep kit (Zymogen).
To make double mutants, the resistance cassette in a single mutant strain was removed by Cre-Lox recombination: 500 ng of pDM1489 (Paschke et al., 2018) was transformed into cells, selected for, and then plated out on SM plates with K. aerogenes to remove the selective pressure. Clonal populations that had lost both resistance markers were identified; sometimes this took more than one passage of cells on bacteria.
Macropinosome formation rate
Cells were harvested from bacterial plates, washed free of bacteria and incubated in 2 ml of the indicated medium for 24 h at 105 cells per well on a six-well plate. The medium was removed, the cells resuspended in 1 ml SUM and transferred to a two-well microscope slide (Nunc) and allowed to settle for ∼1 h. The medium was removed and 0.5 ml SUM containing 2 mg/ml FITC–dextran (70 kDa, Sigma-Aldrich) was added for 1 min, removed and the cells washed twice with KK2MC, fixed using 4% paraformaldehyde for 20 min, washed five times with PBS (pH 5) and stored at 4°C. Z-stacks with 0.1 μm steps were taken using a Zeiss 700 series microscope and the number of dextran-positive vesicles counted manually using FIJI software.
Cells were transformed with overexpression vectors as above (Table S6). Transformants were washed free of bacteria, and 105 cells were incubated in 2 ml medium for 24 h in a six-well plate at 22°C. The medium was removed and replaced with SUM, and the cells transferred to two-well microscope slides for ∼30 min before imaging using a Zeiss 700 series microscope. Movies were typically taken for 5 min with 1 frame per sec. The active Ras marker Raf1-RBD (Hoeller et al., 2013) was used to observe macropinosomes and macropinocytic patches unless otherwise specified. When measuring macropinosome diameters, random cross-sections of cells were used, meaning any individual diameter measured using this technique could be at any section through the macropinosome, but in the aggregate the average tends to halfway between the equator and the poles.
Cells were cultivated on SM agar plates in co-culture with K. aerogenes bacteria. At 24 h before the assay, cells were transferred into 2 ml of either HL5 medium plus 10% FBS or a K. aerogenes/KK2MC suspension (OD600 nm=2) at 2×105 cells/ml and plated into six-well plates. After 24 h, cells were washed five times in KK2MC and resuspended in the same buffer to a final density of 105 cells per ml. The cells were seeded in MatTek 35 mm Petri dishes with a 20 mm 1.5 coverglass and allowed to attach for 20 min. The random motility of the amoebae was observed for 30 min, taking frames every 30 s using a Zeiss 700 series microscope equipped with a 10× air objective. Automatic cell tracking was performed as described previously (Susanto et al., 2017).
Determination of mean generation time
To measure proliferation in shaking axenic culture, cells were cultured to logarithmic phase in shaking axenic culture, diluted to 5×105 cells/ml in fresh medium and counted approximately every 12 h using a haemocytometer.
To measure axenic proliferation of surface attached cells, cells were prepared as above, then 2×104 were plated in 0.5 ml fresh medium in duplicate in multiple 24-well plates that were placed in a 22°C incubator. Approximately every 12 h, the medium was removed from a plate, the wells washed once with buffer and stained with 0.1% Crystal Violet dissolved in 10% ethanol for 20 min. The free Crystal Violet was removed by washing three times with water, after which the Crystal Violet staining the cells was dissolved by addition of 0.9 ml 10% acetic acid for 20 min. The absorbance at 590 nm of the resulting solution was obtained using a Nano Drop 2000 (ThermoFisher Scientific) and the background Crystal Violet staining from a set of wells with no cells was subtracted.
Finally, to measure proliferation with bacteria as the food source, cells were cultivated on SM plates in conjunction with K. aerogenes. 104 cells/ml from growth zones were seeded in a conical flask containing 5 ml KK2MC plus K. aerogenes (at 20 OD600 nm), and the cell density counted approximately every 4 h during the daytime using a haemocytometer.
Cells were harvested from SM plates and washed free of bacteria. 106 cells were inoculated into 10 ml HL5 medium plus 10% FBS in a 9 cm tissue culture plate for 24 h. They were then resuspended by pipetting, pelleted by centrifugation at 370 g for 3 min, the supernatant discarded and washed twice in 50 ml ice-cold KK2MC buffer. Cells were counted and transferred to a 1.5 ml Eppendorf tube, spun down at max speed (20,000 g) in a benchtop centrifuge for 10 s, the supernatant removed by aspiration, resuspended to 2×107 cells/ml in 1× sample buffer (NuPage) containing 2.5% 2-mercaptoethanol and protease and phosphatase inhibitors and incubated for 5 min at 95°C.
10 µl of the cell solutions were run on 4–12% Bis-Tris 10-well gels (Nupage) with See Blue Plus 2 used as a ladder. These were blotted onto immobilon P (Merck) membranes. This was blocked with 2% bovine serum albumin (BSA; ThermoFisher Scientific) in Tris-buffered saline with 0.1% Tween 20 (TBS-T) for 2–3 h, then incubated with primary antibody [against α-tubulin (1:1000, DM1A) and phosphorylated Akt substrates (1:1000, 23C8D2), Cell Signaling Technology] overnight in 5% BSA in TBS-T. Membranes were rinsed twice with TBS-T and washed three times for 10 min before addition of the secondary antibody conjugated to HRP (172-011, BioRad and Ab97051, Abcam; 1:10,000) in 5% BSA for 2–3 h, then washed again. Antibodies were imaged with GE Healthcare detection reagent on a BioRad chemidoc using the chemidoc hi-resolution setting. Blots were stripped using restore PLUS western blot stripping buffer (Thermofisher) as per the manufacturer's instructions before re-probing.
Cell size was determined using an Eclipse flow cytometer (Sony iCyt). Cells were harvested from growth on bacteria, and incubated in a 24-well plate for 24 h in either HL5 medium plus 10% FBS (105 cells/well) or in KK2MC buffer plus 2OD 600 nm Ka (104 cells/well). They were then washed once in situ, resuspended in KK2MC and passed through a 70 µm filter (Sysmex CellTrics). Cells were then run on the Eclipse to measure the median diameter of the cell population.
Sample preparation for mass spectrometry
Approximately 1×108–2×108 vegetative cells for each strain (Ax2, PkbA-, PkbR1-, PkbA-/PkbR1-) were harvested from cultivation on bacterial SM agar plates and washed five times in KK2MC using a benchtop centrifugation at 280 g for 3 min. The harvested cells were resuspended to 2×106 cells per ml in HL5 medium plus 10% FBS, and incubated in a shaking conical flask at 22°C for 24 h.
The cells were then pelleted in a benchtop centrifuge as before, washed twice in KK2MC and resuspended to 107 cells per ml in KK2MC. The PkbR1- cells were split between two conical flasks, one of which had LY294002 (Cayman Chemical) added to 100 µM (to prevent PkbA recruitment to macropinosomes and activation) while the rest of the samples all had 0.2% DMSO vehicle added (Ax2+DMSO=sample 1, PkbA-+DMSO=sample 2, PkbR1-+DMSO=sample 3, PkbR1-+LY294002=sample 4, PkbA-/PkbR1-+DMSO=sample 5). The cells were incubated shaking at 180 rpm at 22°C for 30 min, then mixed in a 1:1 ratio with 10% trichloroacetic acid (TCA), and incubated on ice for at least 30 min to lyse the cells and precipitate the protein.
The protein was pelleted in a benchtop centrifuge at 2400 g for 10 min, then washed twice with 20 ml of ice-cold acetone under the same conditions to remove the TCA. The protein was resuspended in protein solubilization buffer (8 M urea, 20 mM HEPES, pH 8) and the concentration of the samples was measured using a Bradford assay (BioRad), adjusted to 2.5 mg/ml, frozen in dry ice and stored at −80°C.
Following the isolation of protein, 450 µg of each sample was reduced with 5 mM DTT at 56°C for 30 min then alkylated with 10 mM iodoacetamide in the dark at room temperature for 30 min. They were then digested with mass spectrometry grade Lys-C (Promega) at a protein:Lys-C ratio of 150:1 (w/w) for 4 h at 25°C. Next, the samples were diluted from 8 M to 1.5 M urea using 20 mM HEPES (pH 8.5) and digested at 30°C overnight with trypsin (Promega) at a 75:1 (w/w) protein:trypsin ratio. Digestion was stopped by the addition of trifluoroacetic acid (TFA) to a final concentration of 1%. Any precipitates were removed by centrifugation at 13,000 g for 15 min. The supernatants were desalted using homemade C18 stage tips containing 3 M Empore extraction disks (Sigma-Aldrich) and 8 mg of poros R3 (Applied Biosystems) resin. Bound peptides were eluted with 30–80% acetonitrile (MeCN) in 0.1% TFA and lyophilized.
TMT peptide labelling
The lyophilized peptides from each sample were resuspended in 75 µl of 3% MeCN. The peptide concentrations were determined via a Pierce Quantitative Colorimetric Peptide assay (ThermoFisher Scientific) according to the manufacturers' instructions, except the absorbance was measured by Nanodrop Spectrophotometers (ThermoFisher Scientific) at 480 nm. 0.8 mg of each TMT 10plex reagent (ThermoFisher Scientific) was reconstituted in 41 µl anhydrous MeCN. The peptides from each of the samples were labelled with a distinct TMT tag in 170 mM triethylammonium bicarbonate for 1 h at room temperature. The labelling reaction was stopped by incubation with 8 µl 5% hydroxylamine for 15 min. The labelled peptides from each repeat were combined into a single sample and partially dried to remove MeCN in a SpeedVac (ThermoFisher Scientific). After this, the sample was desalted as before and the eluted peptides were lyophilized.
The lyophilized labelled peptides were resuspended in 1.2 ml of 50% MeCN and 2 M lactic acid (loading buffer) in an Eppendorf tube and incubated with 20 mg TiO2 beads (Titansphere, GL Sciences, Japan) at room temperature for 1 h. For the second round of enrichment, the bead slurry was spun down by centrifugation (1657 g for 2 min), and the supernatant was transferred to another Eppendorf tube. The supernatant was incubated with 15 mg of TiO2 beads and the bead slurry was spun down again as before. The beads were transferred to homemade C18 stage tips, washed in the tip twice with the loading buffer and once with 50% MeCN plus 0.1% TFA. Phosphopeptides were eluted sequentially with 50 mM K2HPO4 (pH 10) followed by 50 mM K2HPO4, 50% MeCN (pH 10) and 50% MeCN plus 0.1% TFA. The eluates were partially dried in a SpeedVac after acidification and were desalted as described above.
Basic pH reverse-phase HPLC fractionation
The enriched phosphopeptides were subject to off-line high-performance liquid chromatography (HPLC) fractionation, using an XBridge BEH130 C18, 5 μm, 2.1 mm×150 mm (Waters) column with an XBridge BEH C18 5 μm Van Guard cartridge, connected to an Ultimate 3000 Nano/Capillary LC System (Dionex). Phosphopeptides were resolubilized in solvent A (5% MeCN, 95% 10 mM ammonium bicarbonate, pH 8) and separated with a gradient of 1–90% solvent B (90% MeCN, 10% 10 mM ammonium bicarbonate, pH 8) over 60 min at a flow rate of 250 μl/min. A total of 58 fractions were collected. They were combined into 20 fractions and lyophilized.
The fractionated phosphopeptides were analysed by LC tandem mass spectrometry (MS/MS) using a fully automated Ultimate 3000 RSLC nano System (Thermo Scientific) fitted with a 100 μm×2 cm PepMap100 C18 nano trap column and a 75 μm×25 cm reverse phase C18 nano column (Aclaim PepMap, Thermo Scientific). Lyophilized phosphopeptides were dissolved in solvent A (2% MeCN, 0.1% formic acid) and eluted with a linear gradient from 4 to 50% solvent B (80% MeCN, 0.1% formic acid) over 110 min with a flow rate of 300 nl/min. The outlet of the nano column was directly interfaced via a nanospray ion source to a Q Exactive Plus mass spectrometer (Thermo Scientific). The mass spectrometer was operated in standard data-dependent mode, performing a MS full-scan in the m/z range of 350–1600, with a resolution of 140,000. This was followed by MS2 acquisitions of the 15 most-intense ions with a resolution of 35,000 and normalized collsion energy (NCE) of 33%. MS target values of 3×106 and MS2 target values of 105 were used. The isolation window of precursor was set at 1.2 Da and sequenced peptides were excluded for 40 s.
The acquired MS/MS raw files from the analysis above were processed using Proteome Discoverer (version 2.1, ThermoFisher Scientific). MS/MS spectra were searched against the January 2016 Dictyostelium discoideum UniProt FASTA database, using the Mascot (version 2.4, Matrix Science) search engine. Cysteine carbamidomethylation was set as a fixed modification, while methionine oxidation, N-terminal acetylation (protein), phosphorylation (STY) and TMT6plex (peptide N-terminus and lysine) were selected as variable modifications. Other parameters were set to default values. The abundance values of TMT reporter ions were normalized to total peptide amount. Two output files were generated; one listing the proteins identified and one for peptides. Only proteins and peptides that were identified with a high confidence [false discovery rate (FDR) of <1%] based on a target decoy approach were included in these results. The peptide table was filtered for phosphopeptides with quantitative values and exported as an Excel file.
Quantification and statistical analysis
Statistical analysis was performed using GraphPad prism software, which was used to generate the presented graphs. When there were more than two samples to compare, one-way ANOVA was performed before any further statistical tests. Unless otherwise specified an unpaired t-test was performed (*P≤0.05, **P≤0.01 are marked; P-values of ≤0.1 are also indicated but are not considered to have reached significance). Unless otherwise stated, the mean is plotted with the error bars showing the s.e.m.
The authors thank the Kay laboratory for their input into this project, particularly Gareth Bloomfield and his exhaustive knowledge of the literature, and the LMB microscopy and flow cytometry facilities for excellent scientific and technical support. We are also very grateful to Douwe Veltman for modifying his membrane finder software, Peter Devreotes for providing the PDK1 mutants and Matthew King of Astra Zeneca for support and patience.
Conceptualization: T.D.W., R.R.K.; Methodology: T.D.W., S.-Y.P.-C., P.P.; Validation: T.D.W.; Formal analysis: T.D.W., S.-Y.P.-C., P.P.; Investigation: T.D.W., S.-Y.P.-C., P.P.; Data curation: S.-Y.P.-C.; Writing - original draft: T.D.W.; Writing - review & editing: T.D.W., S.-Y.P.-C., P.P., R.R.K.; Visualization: T.D.W., P.P.; Supervision: R.R.K.; Project administration: R.R.K.; Funding acquisition: R.R.K.
We thank the Medical Research Council UK for core funding (U105115237) to R.R.K. and AstraZeneca for Blue Sky funding.
The authors declare no competing or financial interests.