Coordination of force-generating actin-based modules stabilizes and remodels membranes in vivo

Cells continuously remodel their membranes to drive a broad variety of essential cellular events (Aguet et al., 2016; Basu and Huse, 2017; Haucke and Kozlov, 2018). Membrane remodeling requires the exquisite coordination of different processes that include: (i) modification of the lipid bilayer composition (Inaba et al., 2016; Kameoka et al., 2018), (ii) interaction between membranes and proteins harboring curvature-generating domains (e.g., BAR-domains) (Mim and Unger, 2012; Simunovic et al., 2016), and (iii) application of mechanical forces by cytoplasmic complexes including the cytoskeleton (Bezanilla et al., 2015; Chiaruttini et al., 2015; Crawley et al., 2014; Simunovic and Bassereau, 2014). In particular, the actomyosin cytoskeleton has been implicated in membrane remodeling during processes ranging from cell division and cell migration to membrane trafficking (Dambournet et al., 2011; Kaksonen et al., 2006; Puthenveedu et al., 2010; Ritter et al., 2017; Weiner et al., 2007). Its ability to reshape membranes with different biophysical properties (e.g., plasma membrane domains or vesicular structures) is linked to the flexibility of actin to form structurally and functionally diverse modules, such as branched and linear filaments (Goley and Welch, 2006; Goode and Eck, 2007), and to engage with the various members of the myosin family of motor proteins (Ridley, 2011; Schröder, 2020). F-actin and nonmuscle myosin II (NMII) are perfect examples of this modularity, as they assemble into a range of force-generating structures including contractile rings, sarcomeres, arcs, and lattices (Ebrahim et al., 2013, 2019; Henson et al., 2017; Murugesan et al., 2016).

How the activation and assembly of actomyosin complexes is transduced into the deformation of membranes has been extensively investigated in in vitro model systems (i.e., cell-free systems, cell culture in 2D and 3D, organoids) but few studies have been conducted in vivo. Although reductionist systems allow ready manipulation of experimental conditions, they do not take into account the crucial roles played by the 3D tissue organization and unique cues coming from the vasculature, nervous system, and adjacent cells during the remodeling. The development of intravital subcellular microscopy (ISMic) has opened the door to the visualization of the dynamics of subcellular processes in live animals at a resolution comparable to that achieved in cell culture (Ebrahim and Weigert, 2019; Masedunskas and Weigert, 2008; Scheele et al., 2022; Weigert, 2014; Weigert et al., 2013). This approach has enabled the dissection of mechanisms controlling membrane remodeling in the context of organ physiology (Dunn et al., 2002; Porat-Shliom et al., 2016), cancer (Amornphimoltham et al., 2013; Rompolas et al., 2012), cell motility (Lämmermann et al., 2013; Subramaniam et al., 2000), and membrane trafficking (Masedunskas et al., 2011; Masedunskas and Weigert, 2008; Porat-Shliom et al., 2019; Sandoval et al., 2004). We used ISMic to investigate the role of the actomyosin cytoskeleton in the gradual integration of large exocytic vesicles (termed granules) into the apical plasma membrane (APM) of acinar cells in salivary glands and the pancreas during regulated exocytosis in vivo. This process requires the assembly of a novel actomyosin-based structure composed of two intertwined lattices formed by F-actin and NMII (Ebrahim et al., 2019; Masedunskas et al., 2011; Milberg et al., 2017). We have established that the role of the NMII lattice during granule integration is linked to NMII contractile activity (Ebrahim et al., 2019; Milberg et al., 2017). On the other hand, although we determined that the F-actin lattice is essential for the process, its specific function is not fully understood beyond its contribution to myosin-driven contraction.

Here, we show that immediately after fusion with the APM, the granules become coated with a lattice that is formed by linear actin filaments that are nucleated by mDia1, a member of the formin family of actin nucleators (Courtemanche, 2018; Krebs et al., 2001), that co-assemble with Tpm3.1, a member of the tropomyosin family of actin-associated proteins (Gunning et al., 2015). This step is followed by the assembly of a wave of branched filaments controlled by the Arp2/3 complex (Goley and Welch, 2006; Rouiller et al., 2008), which drives granule integration. Impairment of the lattice assembly through pharmacological inhibition or genetic ablation of mDia1 results in a significant increase in the diameter of the fused granules due to an uncontrolled flow of membranes from the APM and also fusion with adjacent granules (i.e., compound exocytosis [Nemoto et al., 2004]). In contrast, impairment of the Arp2/3 complex or Ezrin, an actin-membrane linker and tension regulator in the ERM family (Ezrin, Radixin, Moesin) (Pelaseyed and Bretscher, 2018; Rouven Brückner et al., 2015; Senju and Tsai, 2022), inhibits formation of the wave and significantly delays membrane integration without affecting the assembly of the F-actin lattice.

Based on our results, we propose a novel mechanism for the remodeling of membranes that is based on the sequential assembly of two distinct F-actin-based modules: a lattice formed by linear actin filaments, which prevents the addition of membranes to the fused granules; and a wave of branched actin filaments, which actively generates forces controlling the integration of the membranes.

To investigate membrane remodeling and F-actin dynamics during regulated exocytosis, we used ISMic imaging of GFP-LF/mTom mice that express the F-actin probe GFP-LifeAct (GFP-LF) (Riedl et al., 2010) and the membrane-targeted peptide mTomato (mTom) (Muzumdar et al., 2007) (Fig. 1 A). As previously shown, the stimulation of exocytosis using isoproterenol (ISOP) leads to the fusion of the secretory granules with the APM (Masedunskas et al., 2011) (Fig. 1, A and B arrows). The diffusion of mTom from the apical canaliculi into the membranes of the secretory granules and the recruitment of GFP-LF make it possible to follow the dynamics of F-actin during the integration process (Fig. 1, A and C) (please note that the terms GFP-LF and F-actin will be used interchangeably) (Ebrahim et al., 2019; Masedunskas et al., 2011; Milberg et al., 2017). F-actin was detected on the membranes 1–2 s after the appearance of mTom (Fig. 1 C) as previously reported (Masedunskas et al., 2011). The diameter of the fused granules did not significantly decrease for the first 6 s (see Materials and methods, N = 55 granules, 24 cells, 5 animals) (Fig. 1, C and D; and Video 1), and during this time F-actin assembled into the previously described lattice (Ebrahim et al., 2019) (Fig. 1 E arrow). Over the following 10–15 sec, the granules decreased in size with a calculated rate of reduction of the surface area of 0.37 ± 0.02 µm²/s (see Materials and methods, N = 55 granules, 24 cells, 5 animals). Notably, as the diameter of the granules began to decrease, the external diameter of the lattice did not significantly change. However, the polymerization of actin continued and manifested as an increase in the thickness of F-actin around the granules (hereafter referred to as the wave) (Fig. 1 C arrows). In addition, we observed the assembly of a denser layer of F-actin at the interface with the granule membranes (Fig. 1 C double arrowheads). High-resolution images acquired in fixed salivary glands confirmed the existence of the dense GFP-LF structure that lies beneath the external lattice and that interfaces with the granule membrane (Fig. 1 F arrowheads) (hereafter referred to as the inner ring). Similar results were obtained in fixed glands excised from mTom mice and stained with phalloidin, thus ruling out any possible artifacts due to the use of the GFP-LF probe (Fig. 1, E and F). Quantitative analysis confirmed that the F-actin thickness inversely correlated with the granule diameter as measured both in GFP-LF/mTom mice imaged by ISMic and in fixed glands excised from mTom mice (Fig. 1, G and H). At around 20 s after fusion, when the diameter of the granules reached 20% of the initial size, the external diameter of the F-actin lattice began to decline steadily until 50–60 s when F-actin was no longer detectable (Fig. 1, C and D). Due to the limitations in the resolution of our instrumentation, we could not perform any reliable dynamic measurement of the granule diameter below 200 nm or visualize the process of F-actin disassembly (Ebrahim et al., 2019). Our analysis of the integration was limited, therefore, to a range of diameters between 1.2 and 0.3 µm. Focus ion beam scanning electron microscopy (FIB-SEM) (Narayan and Subramaniam, 2015) confirmed that during the integration process, the granules maintained roundness and positive curvature (Fig. S1, A–C and see Discussion). Moreover, we observed small membranous profiles bulging from the APM (<300 nm) although we could not distinguish whether they were secretory or endocytic in nature (Fig. S1 A). Finally, we determined that the distribution of the diameters of the fused granules agreed with the measurements performed using light microscopy (Fig. S1 D).

Our data suggest that two distinct F-actin-based populations control membrane integration. To establish whether the lattice and the wave of F-actin are assembled via different mechanisms, we determined which actin nucleators are recruited on the secretory granules after fusion with the APM. We used qPCR (Fig. S2 A) to determine their expression in freshly isolated acinar cells and indirect immunofluorescence in vivo to validate their recruitment on the fused secretory granules (Fig. 2, A and B; and Fig. S2 B). We found that one member of the formin family of linear filament nucleators, mDia1 (Fig. 2 A) (Courtemanche, 2018; Krebs et al., 2001), and components of the actin branching machinery that include two subunits of the Arp2/3 complex (ARPC2 and Arp2, Fig. 2 B and Fig. S2 B, respectively) (Rouiller et al., 2008), cortactin and N-WASP (Fig. S2 B) (Helgeson and Nolen, 2013), were expressed in acinar cells and recruited to the secretory granules after fusion with the APM. Under basal conditions, these molecules localized either at the APM or in the cytoplasm, but not on the unfused granules (Fig. S2 C).

mDia1 initially localized primarily in puncta associated with the secretory granules (Fig. 2 A). The puncta partially overlapped with the F-actin lattice in larger granules (Fig. 2, A ii), beneath the thick F-actin layer in intermediate-size granules (Fig. 2, A iii), and with the inner F-actin ring in the smaller granules (Fig. 2, A iv and v). At all stages, mDia1 preferentially localized at the interface with the membranes (Fig. 2, A ii–v arrows). Of note, we observed that 14 ± 1% (average ± SEM, N = 100 granules, 20 cells, 3 animals) of the mDia-labeled granules were devoid of phalloidin (Fig. 2, A i) and labeled with mTom (Fig. 2, A vi), suggesting that the recruitment of mDia1 precedes F-actin assembly. Similarly, we found that 32 ± 6% (average ± SEM, N = 100 granules, 20 cells, 3 animals) of large granules exhibiting the F-actin lattice were devoid of ARPC2. These granules exhibited a thinner actin coat than those labeled with ARPC2 (0.20 ± 0.02 µm versus 0.32 ± 0.08 µm, average ± SEM, N = 100 granules, 20 cells, 3 animals, t test P < 0.0001). These data suggest that the Arp2/3 complex is recruited and activated after the assembly of the lattice (Fig. 2, B i). Notably, in large granules ARPC2 localized beneath the lattice (Fig. 2, B ii arrows), and as the thickening and the integration progressed, APRC2 appeared to localize in the gap between the lattice and inner F-actin ring and was primarily associated with the latter (Fig. 2, B iii–v).

These data are consistent with the initial recruitment of mDia1 on granule membranes immediately after fusion with the APM to initiate the assembly of linear actin filaments to form the lattice. This is followed by the recruitment of the Arp2/3 complex to initiate the F-actin wave composed of branched actin filaments, possibly in coordination with a second pool of mDia1 (Fig. 2 E). To test this hypothesis, we used ISMic to determine the kinetics of actin nucleator recruitment on the granule membrane with respect to the assembly of F-actin. We cotransfected the acinar cells of the salivary glands of live rats with RFP-Lifeact (RFP-LF) and either mDia1 or Arp2 tagged with m-Emerald (Fig. 2, C and D). Rats were chosen because (i) they can be transfected using an established procedure that doesn’t affect the structure/function of the acini (Sramkova et al., 2009), and (ii) regulation of large granule exocytosis in salivary glands is similar to that of mice (Masedunskas et al., 2011). We showed that both mDia1 and ARPC2 are present on secretory granules following ISOP stimulation in rats (Fig. S2 D). Note, that due to the increased imaging depth in the rat salivary glands, we used laser scanning confocal rather than spinning disk confocal microscopy with a partial loss in spatiotemporal resolution. As expected, both mDia1 and Arp2 were recruited to the fused granule membranes upon ISOP stimulation. mDia1 appeared on the granules ∼2–3 s before the RFP-LF and its levels reached a peak after ∼10 s (Fig. 2 C left panels and Video 2). This pool of transfected mDia1 localized primarily at the edge of the granules and overlapped with the F-actin lattice. We also observed a second pool of mDia1 that was recruited as the wave initiated inside the F-actin lattice (Fig. 2 C left panels, arrow and Video 2). On the other hand, Arp2 appeared on the membranes ∼5 s after RFP-LF, and its level peaked at around 20 s, mirroring the thickening of F-actin (Fig. 2 D left panels and Video 3). Arp2 localized only in the thickened LF layer juxtaposed to the external lattice, confirming the findings obtained by immunofluorescence (Fig. 2 B) and suggesting that the second pool of mDia1 could be required to nucleate linear filaments needed to initiate the branching actin network.

Importantly, our demonstration that two structurally distinct actin populations are sequentially assembled on the secretory granules was validated by the finding that only the F-actin forming on the external lattice colocalized with tropomyosin Tpm3.1 (Fig. 3, A–C and Video 4). Tropomyosins form copolymers with linear actin filaments and we have shown that Tpm3.1 is recruited to the granule membrane with identical kinetics to F-actin during the first 5 s of filament formation (Gunning et al., 2015; Masedunskas et al., 2018). Similarly, NMIIA was also recruited to the granule membrane surface, as shown using both ISMic and indirect immunofluorescence (Fig. 3, D–F and Video 5). As exocytosis progressed, Tpm3.1 and NMIIA remained enriched at the periphery and did not associate with the F-actin pool beneath the lattice, indicating that Tpm3.1 and NMIIA are not major components of the inner wave of F-actin (Fig. 3, B and D).

Our findings suggest that the fusion of the granule with the APM triggers the initial recruitment of mDia1 to the granule membrane and the assembly of actin/Tpm3.1 linear filaments, followed by the recruitment of the Arp2/3 complex and the generation of branched filaments (Fig. 3 G). To determine the role of these two F-actin populations in the integration process, we used a Cre-lox approach to deplete the levels of either mDia1 or ARPC2, a key component of the Arp2/3 complex. Mice floxed for mDia1 (mDia1^fl/fl) (Deguchi et al., 2016) or ARPC2 (ARPC2^fl/fl) (Rotty et al., 2015) were crossed with a strain expressing a tamoxifen (TMX)-inducible Cre recombinase under the control of the Mist1 promoter, which is expressed in salivary glands (Aure et al., 2015). These mice also harbor the mT/mGFP reporter to distinguish Cre-expressing cells (GFP positive) and to visualize the secretory granules after their fusion with the APM (Milberg et al., 2017; Muzumdar et al., 2007) (Fig. S3, A–C; mDia1^{fl/fl Mist1-Cre, mT/mG}, ARPC2^{fl/fl Mist1-Cre, mT/mG}). Mice expressing the Mist1-Cre module and the mT/mGFP reporter (Ctrl^{Mist1-Cre, mT/mG}) were used as controls.

mDia1^{fl/fl Mist1-Cre, mT/mG} and Ctrl^{Mist1-Cre, mT/mG} mice were treated with TMX and imaged after 3–4 wk. At this time, the cellular levels of mDia1 were reduced to 62 ± 4% with respect to the control levels (Fig. S3 B). Any attempt to further deplete mDia1 severely affected the endomembrane system of the acini (Fig. S4 A). Under these conditions, in mDia1^−/− acinar cells, 37% ± 7%. (N = 102 granules, 29 cells, 8 animals) of the secretory granules began to expand immediately after ISOP-induced fusion with the APM and did not integrate for at least 90 s (Fig. 4, A and B left panels; Fig. 4 C and Video 6). 20% ± 6.4% of the granules underwent delayed integration, whereas, the remaining 43% integrated similarly to the controls (Fig. 4 C). TMX administration or Cre expression did not induce this phenotype since the morphology and kinetics of integration of the secretory granules did not change in mDia1^+/+ acinar cells in either mDia1^{fl/fl Mist1-Cre, mT/mG} (Fig. S4 B) or Ctrl^{Mist1-Cre, mT/mG} (Fig. 4, A and B right panels; Fig. 4 C and Video 6) mice treated with TMX. Moreover, mDia1 depletion did not result in defects in the assembly of F-actin on the APM or changes in its morphology (Fig. S4 C), suggesting a direct impact of the linear F-actin associated with the granule on membrane integration. We investigated whether the heterogeneity in the integration phenotype in mDia1^−/− cells was due to differences in the levels of F-actin assembly on the membrane of the granules because of either variability in the extent of mDia1 depletion or compensation from other formins expressed in the acinar cells. We found that F-actin was consistently detected on granules with diameters below 1.5 µm (Fig. S4 D lower panels), whereas, larger granules appeared to be either uncoated or partially coated with F-actin (Fig. S4 D upper and center panels). We confirmed that mDia1 was not detected on the fused granules (Fig. S3 B). To further substantiate these data and visualize the dynamics of F-actin recruitment when formin activity is impaired, we treated the GFP-LifeAct/mTom mice with SMIFH2, a drug that targets the FH2 domains shared by multiple members of the formin family (Rizvi et al., 2009). SMIFH2 treatment phenocopied the effects of mDia1 depletion on granule integration and F-actin assembly in mTom (Fig. 4 E, N = 103 granules in 8 animals), GFP-LifeAct/mTom (Fig. 4 F), and WT strains (Fig. S4 E). We observed that (i) F-actin was not recruited on the population of expanded granules that did not integrate (Fig. 4 F upper panels), and (ii) partial recruitment of F-actin occurred at later time points on a subset of expanded granules, coinciding with the onset of the integration process (Fig. 4 F center panels). Consistently, mDia1 and Arpc2 were recruited on these granules and most likely they were partially functional since we detected enlarged lattice-like structures (Fig. 4 G), and we observed the F-actin wave only in 45% of the granules. (Fig. 4 F center panels, 45 granules, 28 cells, 8 animals). The fact that the remaining delayed granules did not form an F-actin wave suggests that a limited amount of active formins still promotes the assembly of the lattice but does not support the assembly of the Arp2/3-mediated branched filaments. However, although SMIFH2 phenocopied the mDia1 depletion, we cannot rule out the involvement of other formins in this process.

Under these conditions, NMIIA localized on the F-actin-coated large granules (Fig. 4 G) and possibly contributed to driving the integration to completion, as previously reported (Milberg et al., 2017). Finally, as expected, the remaining granules recruited F-actin, did not expand, and underwent normal integration (Fig. 4 F lower panels).

Next, we addressed the mechanism responsible for the expansion of granules in the absence of the F-actin lattice immediately after their fusion with the APM. We envisioned two non-mutually exclusive mechanisms. The first is based on the fact that as the fusion pore forms, the membrane is driven from the APM into the granule by a gradient of membrane tension, as previously shown during granule exocytosis in mast cells and model membranes (Chizmadzhev et al., 1999; Monck et al., 1990). Notably, in the first few seconds after fusion, the surface area of granules increased at a similar rate under both control conditions and when formin was impaired (Fig. 4 H).

On the other hand, when F-actin recruitment was impaired, the granule surface area steadily increased over time with absolute rates similar to those measured during the integration process (Fig. 4 H). Although we cannot measure the overall surface area of the APM due to the high number of villi and folds present in its lumen (Segawa et al., 1998), it is unlikely that a three to fourfold increase in granule surface area (Fig. 4 H) can be attributable solely to a flow of lipid from the APM. Indeed, we occasionally observed membranes being transferred from adjacent granules undergoing compound exocytosis (Kasai et al., 2006; Nemoto et al., 2004) (Fig. 4 D).

Together, these data support the hypothesis that the lattice formed by mDia1-dependent linear filaments provides a scaffold to stabilize the membrane of the granules after their fusion with the APM and to prevent their expansion due to membrane transfer from either the APM or via compound exocytosis (Fig. 4 I).

We next addressed the role of the F-actin wave associated with the Arp2/3 complex and its integration with the function of the lattice. ARPC2^{fl/fl Mist1-Cre, mT/mG} mice were treated with TMX such that after 3–4 wk the cellular levels of ARPC2 were reduced to 71 ± 5% with respect to the control levels (Fig. S3 C). In ISOP-stimulated ARPC2^−/− acinar cells, the granules fused with the APM and did not increase in size; however, their integration was significantly delayed (68 ± 15 s, N = 44 granules, 18 cells, 4 animals) versus (16 ±1 s, N = 61 granules, 24 cells, 5 animals) (Fig. 5, A–C and Video 7), and the diameter of the APM was not affected (Fig. S4 F). Strikingly, the F-actin wave and inner ring were not observed although the structure of both the F-actin and NMIIA lattices were not affected (Fig. 5 D). Accumulation of granules was occasionally observed in unstimulated ARPC2^−/− cells (Fig. S4 G).

The effects of the ARPC2 depletion were validated by acute inhibition of branched actin assembly following administration of the Arp2/3 inhibitor CK666 (Fig. 5, E–G, [Yang et al., 2012]). The integration of the granules was slowed without any effect on their initial diameter (Fig. 5 E and Video 8), and the F-actin thickening was largely abolished (Fig. 5 F), as it was observed only in 9% of the granules (N = 44 granules, 17 cells, 6 animals). Notably, consistent with this phenotype, the recruitment of ARPC2 was affected by CK666 treatment (Fig. 5 G); whereas, the recruitment of mDia1 and NMIIA were not (Fig. 5 G). These data suggest that the Arp2/3-complex-dependent branched network facilitates the integration process but is not essential (see Discussion). This raises the question: how does the branched actin filament network control the integration of the secretory granules?

Actin filaments are known to exert forces on cellular membranes and control membrane tension through proteins such as the members of the ERM (Ezrin-Radixin-Moesin) family of linkers (Pelaseyed and Bretscher, 2018). We found that Ezrin and Radixin, which are expressed in epithelial cells (Fehon et al., 2010), were localized at the APM, while Moesin exhibited a weak plasma membrane staining (Fig. S5 A). However, only Ezrin and Radixin were recruited to fused secretory granules upon ISOP stimulation (Fig. 6 A and Fig. S5 B). In large granules, Ezrin localized beneath the F-actin lattice and became associated primarily with the inner ring at the membrane interface as the integration progressed (Fig. 6 B). On the other hand, Radixin localized only on the larger granules (Fig. S5 B). Ezrin and Radixin are recruited to membranes via their FERM domains (Fehon et al., 2010) and bind to F-actin via their C-terminal domains, which become available upon phosphorylation of conserved T567 and T654 residues, respectively (Paige et al., 2014; Suda et al., 2011). We could only confirm that Ezrin was phosphorylated on the fused secretory granules (Fig. 6, C and D). To decouple F-actin from the membranes, we used NSC 668394 (NSC), an inhibitor of Ezrin phosphorylation (Paige et al., 2014). Immunofluorescence confirmed that phosphorylation of Ezrin was significantly diminished in mice treated with NSC (Fig. S5, D and E). Notably, treatment of GFP-LF/mTom mice with NSC did not inhibit the F-actin wave but prevented the assembly of the inner actin ring around the granules and delayed their integration with a reduced rate of surface area reduction (Fig. 6, E and F; and Video 9). Under these conditions, Ezrin associated with the granules (Fig. 6 G), both the F-actin and NMIIA lattices were not affected (Fig. 6 H left panels), and there were no significant effects on the initial increase in surface area, suggesting that neither Ezrin nor Radixin play a role at earlier time points (Fig. 6 F). Finally, NSC did not affect the recruitment of the ARPC2 complex on the granules (Fig. 6 H right panels). Overall, these findings suggest that the interaction between the branched filaments forming the inner ring and the membrane of the granules is mediated by Ezrin and is required to control granule integration (Fig. 6 I).

We propose a mechanism for the remodeling of membranes during exocytosis in living animals that is regulated by the coordinated action of two distinct force-generating F-actin-based modules. These modules operate in coordination with NMIIA and NMIIB, whose role during exocytosis has been previously described (Ebrahim et al., 2019; Milberg et al., 2017). The first module is initiated by the rapid recruitment of the formin mDia1 on discrete foci of the membranes of the fused granules. mDia1 nucleates a population of linear actin filaments that copolymerizes with Tpm3.1 (Meiring et al., 2018) and assembles around the granules into lattice-like structures (Fig. 7). This lattice is intimately associated with an NMII lattice (Ebrahim et al., 2019) (Fig. 3). The second module is initiated by the recruitment of the Arp2/3 complex that assembles a population of filaments composed of branched actin that interact with the membranes via Ezrin, a member of the ERM family of actin-membrane linkers (Bosk et al., 2011; Tsukita et al., 1994) (Fig. 7). We predict that the F-actin lattice cannot serve as a seed to initiate branched filaments since tropomyosins inhibit the ability of the Arp2/3 machinery to bind to actin (Blanchoin et al., 2001). Although Arp2/3 has been shown to de novo initiate actin filaments (Wagner et al., 2013), it is also possible that the second pool of mDia1 recruited on the membranes beneath the lattice provides the template mother filaments for the Arp2/3 complex (Fig. 2). Nonetheless, we cannot formally rule out the involvement of another member of the formin family.

How do these modules control the remodeling of the membranes of the granules after their fusion with the APM? We envision that in vivo the integration step is driven by the progressive transfer of membranes from the secretory granules to the APM. Impairment of mDia1, which inhibits the early assembly of the F-actin/Tpm3.1 lattice without affecting the NMIIA lattice, results in the expansion of the granules right after fusion. This finding is consistent with both theoretical models and experimental data in mast cells where exocytic vesicles first transiently fuse with the plasma membrane to release quanta of cargo molecules and next are released into the cytoplasm (kiss and run). During this process, which does not involve F-actin coating of the fused granules, membranes are transferred from the plasma membrane to the exocytic vesicles through convective flow via the fusion pore (Chizmadzhev et al., 1999; Monck et al., 1990). Alternatively, the expansion of the granules could be also due, in part, to changes in pH and composition of their content when exposed to the environment of the canaliculi. Therefore, we propose that the primary role of the F-actin lattice is to prevent the expansion of the granules (Fig. 7, “stabilization”). This function could require the lattice being linked to the membrane by Ezrin and/or Radixin, at least at the very early stages, although the NSC treatment did not induce any expansion of the granules and it did not affect the lattice (Fig. 6). It is also possible that under these conditions, the stability of the granules is maintained by the NMIIA lattice or by NMIIB, which also controls granule stability (Milberg et al., 2017). This is consistent with the fact that (i) the F-actin lattice is still preserved when physically dissociated from the granule membrane at a later stage of integration, and (ii) ablation of NMIIB or inhibition of NMII contractile activity also induces an expansion of the granules although to a smaller extent (Ebrahim et al., 2019; Milberg et al., 2017). Moreover, in the absence of F-actin, we observed a two to threefold increase in the granule diameter at later time points, suggesting that this large excess of membrane is provided by fusion with adjacent granules, similar to what we previously found in animals treated with the F-actin disrupting agents cytochalasin D and Latrunculin A, which completely block de novo F-actin assembly (Ebrahim et al., 2019; Masedunskas et al., 2011; Milberg et al., 2017). This also suggests that the lattice provides a barrier to prevent or tune compound exocytosis once the signaling cascade eliciting exocytosis is triggered.

Interfering with either the Arp2/3-dependent machinery or the closely associated Ezrin significantly delays the integration of the granules without affecting their size or the F-actin lattice. Notably, impairment of the Arp2/3 function completely blocks the wave of branched actin polymerization, whereas, impairing Ezrin doesn’t affect the wave but prevents the formation of the F-actin ring, thus suggesting that the wave of branched actin is stabilized by Ezrin at the interface with the granule membrane. These data indicate that the Arp2/3-dependent step facilitates the integration, but it is not essential. We envision that these actin filaments populations work in coordination with forces generated by the contractile activity of NMIIA (Ebrahim et al., 2019; Milberg et al., 2017) (Fig. 7, “integration”), whose assembly on the granules is not affected by interference with Arp2/3 (Fig. 5). However, we also proposed that NMIIA may control the dynamic of the fusion pore since its ablation resulted not only in a delay of the integration but in arresting the integration of the granules to 40–50% of the original size (Milberg et al., 2017).

We cannot rule out a role for actin disassembly during the integration process. Indeed, consistent with what was shown by others during the exocytosis of lamellar bodies in primary alveolar type II cells (Miklavc et al., 2015), we have previously shown Cofilin I is recruited on the fused secretory granules (Milberg et al., 2017).

How then does the Arp2/3 complex control granule integration? In several biological models, it has been shown that arrays of branched actin form expanding gels that generate forces capable of pushing membranes (Jin et al., 2022; Parekh et al., 2005; Picco et al., 2018; Schmidt et al., 2010). We could not determine whether the actin wave expands centripetally or centrifugally. In the former case, we expect that the force generated by the branched actin is directly transmitted to the membranes via the Ezrin machinery. In the latter case, we envision that the integration of the granules could be controlled by the third law of motion. The branched actin expands and generates action forces that are transmitted to the F-actin and the contractile NMIIA lattices. This in turn produces reaction forces transmitted back through the branched network to the membranes via the Ezrin machinery (Fig. 7). Regardless, a key role in this process is played by the Ezrin receptors. We ruled out CD44, one of the best-described receptors for Ezrin (Tsukita et al., 1994), which is not localized at the APM in the salivary glands (not shown), and we will explore other molecules such as PI(4,5)P₂. Alternatively, it is possible that the branched filaments/Ezrin module modulates a tension-driven membrane between the granules and the APM. Moreover, we cannot exclude that the integration is also partially driven by the modulation of cortical actin tension as shown in other systems (Miklavc and Frick, 2020; Schietroma et al., 2007; Yu and Bement, 2007).

Both formins and the Arp2/3 complex have been shown to control different steps of exocytosis in different model systems. For example, in explanted drosophila salivary glands, ablating the Arp2/3 complex (Tran et al., 2015) or mDia (Rousso et al., 2016) results in the expansion of the granules and delays in the integration, respectively. Moreover, the exocytosis of the granules has been proposed to be based on a process termed crumpling that involves the folding of the granule membrane coupled with clathrin-mediated endocytosis (Biton et al., 2023; Kamalesh et al., 2021). As illustrated by high-resolution intravital imaging (Fig. 1) and FIB-SEM (Fig. S1), we did not see any evidence for crumpling, at least in the range of integration studied here (i.e., from 1.2 to 0.3 µm in diameter). At this stage, we don’t know whether these differences reflect the fact that the exocytosis in the salivary glands is a terminal event, or the different geometry of the granule/apical plasma membrane interface, or it is simply due to explanting the glands.

To our knowledge, this is the first time that the strict coordination between these two modules is reported and, strikingly, in live animals under bona fide physiological conditions. We propose that it is very likely that these F-actin-based modules modulate the biophysical properties of granule membranes, including membrane tension, by exerting forces at a nanoscale level that could be transmitted via Ezrin to favor the exchange of membranes between granules and the APM.

In conclusion, these two coordinated force-generating modules that we describe in the salivary glands could potentially operate in other secretory systems where large micron-size vesicles that release their contents in tubular compartments are coated with an actomyosin complex after fusion (e.g., pancreas, lungs, lacrimal glands, and endothelium) (Miklavc et al., 2012; Nightingale et al., 2011; Porat-Shliom et al., 2012). However, since their role is to remodel membranes that are topologically and compositionally an extension of the plasma membrane, it is likely that this machinery may play a role in other processes such as endocytosis or cell migration where membrane tension needs to be dynamically modulated.

All experiments using animals were performed in accordance with the guidelines provided by (1) the National Cancer Institute (National Institutes of Health, Bethesda, MD, USA) Animal Care and Use Committee and were compliant with all relevant ethical regulations regarding animal research; and (2) the NSW Animal Research Act (1985) and Australian National Health and Medical Research Council (NHMRC) “Code” eighth edition (2013). All experiments were approved by the UNSW Sydney Animal Care and Ethics Committee under application 17/100B. Male or female mice (age 8–24 wk) and male and female rats (age 6–10 wk) were used in this study. Wistar/Sprague Dawley rats (150 g) were purchased from the Animal Resources Centre. Standard laboratory chow and water were provided ad libitum. The mTom/mGFP Cre reporter mouse (Muzumdar et al., 2007) was purchased from Jackson Laboratory. Hemizygous Lifeact-GFP and Lifeact-RFP transgenic mice were a gift from Roland Wedlich-Soldner (Riedl et al., 2010). GFP-NMIIA knock-in (KI) mice were generated as described (Zhang et al., 2012) and crossed with the RFP-Lifeact strain. The mouse line expressing Tpm3.1 C-terminally tagged with mNeonGreen (NG) at the endogenous locus (B6-Tpm3tm5[mNeonGreen]Hrd) was generated as described (Masedunskas et al., 2018). Conditional mDia1 (mDia1^fl/fl) (Deguchi et al., 2016) or ARPC2 (ARPC2^fl/fl) (Rotty et al., 2015) flox mice were crossed with mice harboring (TMX)-inducible Cre recombinase under the control of the Mist1 promoter (Aure et al., 2015) and the mT/mGFP reporter (Muzumdar et al., 2007) (Fig. S3, A–C; mDia1^{fl/fl Mist1-Cre, mT/mG}, ARPC2^{fl/fl Mist1-Cre, mT/mG}). Cre recombinase was induced by two intraperitoneal injections of Tamoxifen (75 mg/kg body weight) 2 days apart.

Mice were imaged 3–4 wk after the first injection. Cre-expressing cells (GFP positive) and non-Cre-expressing cells (mTom) could be differentiated by their fluorescence and their secretory granules visualized after their fusion with the APM (Muzumdar et al., 2007; Milberg et al., 2017).

Mice (18–34 g) were anesthetized by i.p. injection of 100 mg/kg ketamine/15 mg/kg xylazine. Salivary glands were surgically externalized and prepared for intravital subcellular microscopy as described (Masedunskas et al., 2011). Salivary glands were stabilized on a coverslip on the microscope stage with vasculature and innervation functionally intact.

For high-resolution 3D imaging, a Nikon Ti2 spinning disk microscope was utilized using a CSU-X1 spinning-disc head, an Orca-fusionBT digital camera, and Apo TIRF 100X 1.49 NA (Fig. 4, A, B, D, and G; Fig. 5, A, B, D, and G; and Fig. S4) and SR HP Plan Apo Lambda 100XC Silicone oil 1.35 NA (Fig. 1; Fig. 2, A and B; Fig. 3, A, B, D, and E; Fig. 6; Fig. S1, E and F; and Figs. S2, S3, and S5) objectives. For live imaging, the acquisition speed was set at 80–300 ms per frame, with Z-stacks acquired at 0.05–0.3 μm apart. Nikon Elements Software was used for post-acquisition image alignment to correct for image drift, denoising, and 3D deconvolution using the “Automatic” option, which used the Richardson–Lucy algorithm (Holmes and Liu, 1989) with (i) an additional step for pre-noise estimation and (ii) automatic determination of the number of iterations. For image acquisition from fixed tissue sections, the sampling frequency at the camera sensor was 45 nm, which was enabled by a ×1.5 tube lens. Although this frequency was well above the required Niquist criterion, the oversampling allowed for an extended dynamic range (when combined with the 16-bit analog-to-digital converter and longer exposure times) as has been used previously (Ebrahim et al., 2019), and, together with Nikon denoising software (Application Note, 2019), allowed an improvement in highlighting high-frequency features in the images.

Intravital imaging was also performed using a Nikon A1 inverted laser scanning confocal microscope fitted with a CFI Plan Apochromat lambda series 60×/1.27NA water immersion objective (Fig. 2, C and D; Fig. 3, C and F; Fig. 4 F; and Fig. 5 F), an Okolab humidified temperature-controlled microscope enclosure, objective heater, and a custom-made stage insert. Animal body and salivary gland temperatures were monitored with a MicroTherma T2 thermometer equipped with rectal and thin implantable probes (Braintree Scientific) and heating adjusted to maintain 37°C body and SG temperature. GFP and mNG were excited with a 488-nm laser, and RFP and tdTomato were excited with a 561-nm laser. For time-lapse imaging, frames were acquired in sequential mode at 2.019 s/frame or 241 ms/frame at a spatial sampling of 100 nm per pixel. An imaging plane with a visible canaliculus or apical membrane was selected as close to the surface of the organ as possible—at a depth of 10–20 μm for rat salivary glands and 8–12 μm for mouse salivary glands. During imaging, any noticeable drift was manually corrected in X, Y, or Z dimension. For some experiments, drugs (200 μM CK666 or 200 μM SMIFH2; Sigma-Aldrich; 250 μM NSC668394; EMD Millipore) or 1.5% DMSO solutions were prepared using prewarmed 37°C saline and perfused over the surface of the gland at 10 μl/min during imaging using a PHD Ultra Nanomite programmable syringe pump (Harvard Apparatus). Exocytosis of secretory granules was stimulated by subcutaneous injection of isoproterenol (ISOP) at 0.025 mg/kg (rats) or 0.01 mg/kg (mice).

Images were recorded using NIS Elements software and either a Nikon A1 confocal or Nikon Ti2 spinning disk microscope. Image processing and data extraction were performed using ImageJ/Fiji (Schindelin et al., 2012), NIS Elements, and Imaris. Drift and motion correction on time-lapse image stacks was carried out with the StackReg ImageJ plugin (Thevenaz et al., 1998). Nikon Elements Software was used for post-acquisition image alignment and depending on the signal-to-noise ratio the Nikon Denoise.ai function was used followed by 3D deconvolution using the Richardson–Lucy algorithm. The Richardson–Lucy/maximum likelihood image restoration algorithm was used for fluorescence microscopy with (i) an additional step for pre-noise estimation and (ii) automatic determination of the number of iterations.

The diameters of the granules (Fig. 1 D; Fig. 4, C and E; Fig. 5, C and E; and Fig. 6 F) were measured in Z-stacks acquired in animals expressing the mTom/mGFP probes. For each granule, the optical slice where the granule exhibited the largest diameter was used. Using the line scan function in Fiji, a line was drawn across the center of the granule, as shown in Fig. S1 E. The fluorescence intensity plot generated exhibited two Gaussian intensity profiles and the peak-to-peak value (d) was used as a measure of the diameter (Fig. S1 E).

As for the diameters of the F-actin lattice (Fig. 1 D), a similar strategy was used. However, as shown in Fig. S5 B, during thickening, the diameters of the lattice were calculated as the distance D between the sides of the peaks at the half maximum intensity. The actin thickness (Fig. 1 G and Fig. S5) was calculated as the average of the half maximum widths of the peaks (t1 and t2) (Fig. S1 F).

Diameters of granules, lattices, and F-actin thicknesses were plotted using GraphPad Prism and expressed as average ± SD. The data were plotted using GraphPad Prism. Significant decreases in diameter values were determined using the one-way ANOVA test (package in Prism). The actin diameter was plotted against the actin thickness (Fig. 1, G and H), and linear regression was performed using Prism.

mDia1^{fl/fl Mist1-Cre, mT/mG}, ARPC2^{fl/fl Mist1-Cre, mT/mG}, and Ctrl^{Mist1-Cre, mT/mG} mice were treated with tamoxifen as described in the text, euthanized, and the salivary glands were processed for cryosections. One cryosection from a control and one from a floxed mouse were placed side by side on the same coverslip and labeled in parallel with the same reagents. Z-stacks of randomly selected areas of the slides were acquired, and maximum projections were created using Fiji. Cell outlines were tracked with the free hand selection tool and the mean intensity for each cell was collected using the measure function. The fluorescence intensities of mGFP- and the mTom-positive cells were reported in Fig. S3, B and C as averages ± the SD for each animal for both the control and the knockout samples. The percentage of depletion was derived by calculating the percentage of reduction in the average fluorescence intensity between the GFP and mTom cells in the floxed mice within the same animal and then averaging the averages for all the animals.

Surface area (SA) was calculated by approximating the granule to a sphere and using the formula $A=4*π*(diam/2)2$⁠. The rate of surface area change was calculated as the difference in SA between two-time points in the time-lapse images divided by the time interval. Statistical analysis for the initial diameter (Fig. 4, C and E) was performed using the one-way ANOVA test (Prism package).

Granule exocytosis events from three to eight animals were quantified, as specified in figure legends. Images for the figures were adjusted for contrast and brightness in Imaris or ImageJ to have an optimal display range for features of interest, such as secretory granules, and then converted to RGB images.

For cardiac perfusion, the left ventricle of the heart was punctured in anesthetized mice, and PBS was injected to remove blood from the right atrium, followed by 25 ml fixative (4% formaldehyde in PBS, pH 7.3). Glands were then excised and placed in fixative for 30 min at RT. After fixation, salivary glands were put through a sucrose gradient (10% → 20% → 30%→ 30%:OCT [1:1]) for cryoprotection. Once the tissue sank in the 30% sucrose:OCT [1:1] solution, it was placed in a mold in OCT and frozen on dry ice. Cryosections (10 μm) were cut and adhered to Superfrost Plus slides (Electron Microscopy Sciences).

Cryosections were washed 3 × 5 min in PBS, permeabilized in 0.5% Triton X-100 in PBS for 20 min, and blocked in 10% NGS in PBS for 1 h at RT. Sections were then incubated with the respective primary antibody (see Table S1) overnight at 4°C. Sections were rinsed 3× with PBS and stained with respective secondary antibodies (see Table S1). Sections were then rinsed 3× with PBS and mounted using Fluormount-G TM (Invitrogen) on a glass slide with no. 1.5 coverslips.

Mouse salivary glands were surgically exposed and bathed in 200 μM CK666, 200 μM SMIFH2, or 250 μM NSC in saline at 37°C, 10 min. Dimethylsulfoxide (1.5%) was used as a control. Mice were then injected subcutaneously with 0.01 mg/kg ISOP to stimulate exocytosis. After 30 min, glands were covered with Parafilm, and cardiac fixation was performed as described above.

DNA was extracted from bacterial overnight cultures using Qiagen EndoFree Plasmid Kit (Qiagen). Plasmid DNA was stored in TE-Buffer and diluted to 1 μg/μl aliquots. Plasmid constructs mDia1-Emerald (#54157; Addgene plasmid) and Arp2-Emerald (#53992; Addgene plasmid) were a gift from Michael Davidson (Magnetic Field Lab, Tallahassee, FL, USA). Lifeact-RFP construct was a gift from Roland Wedlich-Soldner (Riedl et al., 2010).

Rats (150–225 g) were obtained from the Animal Resources Centre, Perth, WA, Australia and allowed to acclimate for 1 wk. Rats were anesthetized and salivary glands transfected as previously described (Masedunskas et al., 2013) with the following modifications: 24 μg of plasmid DNA was mixed with Lipofectamine 3000 according to the manufacturer’s instructions.

Submandibular glands were isolated and dissociated with 1 mg/ml collagenase IV (Sigma-Aldrich) for 20 min at 37°C (shaking at 800 rpm). Samples were filtered through 70-μm cell strainers to obtain single-cell suspensions. The acinar glands were resuspended in a final volume of 2 ml FACS wash buffer (2% HI-FCS, 2 mM EDTA, 0.02% sodium azide in 1× PBS). A total of 1 × 10⁶ GFP-expressing acinar cells and non-expressing cells were sorted using BD InfluxTM flow sorter (BD Biosciences). Sorted cells were then centrifuged (300 × g, 5 min) before RNA isolation using the RNeasy Mini Kit (Qiagen), according to the manufacturer’s instructions. 1.5 μg of RNA was reverse transcribed to cDNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems; Thermo Fisher Scientific) according to the manufacturer’s instructions. qRT-PCR was carried out using primers specific for each actin nucleator. The primers were predesigned and synthesized by Sigma-Aldrich (KiCqStart SYBR Green Primers). 20 ng cDNA was added to each well of a 96-well PCR plate with 1 μM forward and reverse primer for each nucleator. 40 cycles were performed, with denaturing temperature at 95°C for 15 s, annealing at 55°C for 30 s, and extension at 72°C for 30 s. The amount of the PCR product was determined in real-time by using SYBR Green and detected in a BIO-RAD CFX96 real-time system (Bio-Rad Laboratories). The amount of gene expression was calculated by using the ΔΔCT (cycle threshold) method, where ΔCT = CT (gene of interest)—CT (housekeeping gene) describes the difference in CT values and ΔΔCT = ΔCT (treated sample)—ΔCT (untreated sample) describes the difference between the control and the experimental sample. The data was normalized to the expression of the housekeeping genes β2-microglobulin (β2M) and ribosomal protein L13A (RPL13A) and with a universal positive control. Melting curve analysis was done for confirmation of product specificity (list of primers in Table S2).

Salivary glands were fixed in in 2% glutaraldehyde for 2 h at RT, followed by washes in 0.1 M sodium cacodylate buffer. Glands were stained in 1% osmium tetroxide, washed again in cacodylate buffer and distilled water, and then stained with 0.5% uranyl acetate in 0.05 M sodium maleate buffer. Glands were dehydrated through an ethanol gradient and embedded in PELCO Eponate 12. Once sample blocks were cured, the block surface was milled flat using an ultramicrotome to expose the glands on the block surface. Sample blocks were mounted on imaging stubs using conductive glue and prepared for microscope loading.

FIB-SEM images of salivary acini were typically acquired at 5 nm pixel sampling and 15 nm FIB mill step size, and the resulting stacks of eight-bit .tiff images were aligned, inverted, and binned to generate .mrc volumetric reconstructions at 15-nm cubic voxel sampling using established protocols (Baena et al., 2021). Sub-volumes containing canaliculi were then excised computationally and segmented out using 3D Slicer (https://www.slicer.org); a combination of threshold-based segmentation and manual clean-up captured both the ducts as well as fusing salivary granules. Visual inspection revealed several fusing granules that were then excised at the necks to yield label maps of every instance of fusing granules captured in a given volume EM reconstruction. These were converted into 3D meshes for downstream calculations.

The 3D Slicer quantification module was used to calculate the volume (V), surface area (SA), and roundness (deviation from sphericity; perfect sphere = roundness of 1) of each fusing vesicle. Granule diameters were calculated mathematically from V and SA assuming sphericity; these were cross-checked with manual representative measurements at the widest points of the granule cross-sections parallel to the axis of the canaliculus. The measurements were very close except for the smallest vesicles, where there was significant deviation from sphericity, so manual measurements were used throughout.

To test the curvature of these granules, we exported the meshes and used the Trimesh library in Python (https://pypi.org/project/trimesh/) to generate discrete mean curvature measures of each granule. This function returns a mean curvature calculated from the dihedral angles between all planes emanating from each vertex of the mesh. An LUT was applied to the mean curvature value at each vertex from +1 (convex, red) through 0 (planar, white) to −1 (concave, blue) and mapped to the original mesh vertices. This shows visually where and how each granule is curved. The mean of all the mean curvatures per granule was calculated and plotted.

Fig. S1 shows the ultrastructure of the secretory granules indicating that during the integration the membranes of the granules exhibit positive curvature. In addition, the figure shows the method for quantifying the diameter of the granules and the thickness of the F-actin coat. Fig. S2 shows the identification of the actin nucleators in mouse and rat salivary glands. Fig. S3 describes the tamoxifen-inducible knockdown of mDia1 and ARPC2, and the quantification of their cellular level. Fig. S4 shows the effect of mDia1 and ARPC2 depletion/inhibition in mouse salivary glands. Fig. S5 shows the expression, and localization of the ERM proteins and the effects of the inhibitor NSC. Table S1 provides a full list of the antibodies used in this study. Table S2 provides the list of the primers used for qRT-PCR. Video 1 shows the time-lapse of secretory granule integration in a mouse expressing GFP-LF and mTom. Video 2 shows the time-lapse of secretory granule integration in a rat transiently transfected with RFP-LF and mDia1-Emerald. Video 3 shows the time-lapse of secretory granule integration in a rat transiently transfected with RFP-LF and Arp2-Emerald. Video 4 shows the time-lapse of secretory granule integration in a mouse expressing Tpm3.1-NG and RFP-LF. Video 5 shows the time-lapse of secretory granule integration in a mouse expressing GFP-NMIIA and RFP-LF. Video 6 shows the time-lapse of secretory granule integration in mDia1 floxed mice. Video 7 shows the time-lapse of secretory granule integration in ARPC2 floxed mice. Video 8 shows the time-lapse of secretory granule integration in GFP-LF/mTom mice treated with CK666. Video 9 shows the time-lapse of secretory granule integration in GFP-LF/mTom mice treated with NSC668394. Data S1 provides the data underlying all the graphs presented in this study.

The data underlying all the figures are available in the published article and its online supplemental material (Data S1).

We thank Dr. Renee Whan and members of the Katharina Gauss Light Microscope Facility at UNSW Sydney for support with live rodent imaging, and Dr. Maté Biro for his help with an earlier version of data quantification. We thank Drs. Leonid Chernomordik (NICHD, National Institutes of Health [NIH]) and John Hammer III (NHLBI, NIH) for their critical reading of the manuscript and insightful comments.

M. Heydecker was supported by a Faculty of Medicine and Health Tuition Fee Scholarship and a Completion Scholarship from UNSW Sydney, and the Graduate Partnerships Program in the Office of Intramural Training & Education of the NIH. R. Weigert was supported by the NIH, NCI, Center for Cancer Research Intramural Research Program (ZIA BC 011682). K. Narayan was funded in whole or in part from the National Cancer Institute, National Institutes of Health, under contract no. 75N91019D00024. E.C. Hardeman and P.W. Gunning were supported by grants from the Australian Research Council (DP160101623) and the Australian National Health and Medical Research Council (APP1100202, APP1079866).

Author contributions: M. Heydecker: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing—original draft, Writing—review and editing, A Shitara: Data curation, Formal analysis, Investigation, Methodology, Writing—review and editing, D. Chen: Data curation, Resources, D.T. Tran: Investigation, Writing—review and editing, A. Masedunskas: Conceptualization, Investigation, Methodology, M.S. Tora: Investigation, Methodology, Writing—review and editing, S. Ebrahim: Conceptualization, Investigation, Writing—review and editing, M.A. Appaduray: Data curation, Investigation, Methodology, J.L. Galeano Nino: Investigation, A. Bhardwaj: Data curation, Software, Validation, Visualization, K. Narayan: Formal analysis, Investigation, Methodology, Supervision, Visualization, Writing—original draft, E.C. Hardeman: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing—review & editing, P. Gunning: Conceptualization, Funding acquisition, Project administration, Supervision, Writing—review & editing, R. Weigert: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing—original draft, Writing—review & editing.

The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

All the cartoons were generated using Biorender.