Teasing out function from morphology: Similarities between primary cilia and immune synapses

Cytotoxic T lymphocytes (CTLs) play a critical role in adaptive immunity, recognizing and eliminating cells infected by intracellular pathogens as well as cancerous cells. Upon T cell receptor (TCR)–mediated recognition, CTLs form a zone of tight interaction with their target cells termed the immune synapse. CTLs undergo rapid rearrangements of both actin and microtubule cytoskeletons as the centrosome (the microtubule organizing center) polarizes toward the synapse, where it docks at the plasma membrane. Secretory granules loaded with perforin and granzyme move along the microtubules toward the docked centrosome that provides a focal point of secretion within the synapse. This exceptionally precise mechanism of polarized secretion enables CTLs to eliminate their targets without damaging surrounding healthy tissue (Golstein and Griffiths, 2018).

The primary cilium is a small microtubule-based organelle assembled by most eukaryotic cells in response to various developmental cues (Sorokin, 1962). Although historically regarded as a vestigial structure, it has become clear that primary cilia are essential signaling hubs, acting as “sensing antennae” to detect and integrate multiple extracellular cues (Pedersen et al., 2016). Dysfunctions of primary cilia are associated with a wide range of human inherited developmental degenerative diseases termed ciliopathies (Reiter and Leroux, 2017). The characterization and study of ciliopathy-associated phenotypes has improved our understanding of the cilium and the mechanisms regulating this enigmatic structure. Although lymphocytes are one of very few cell types that do not appear to form primary cilia (Hildebrandt and Otto, 2005; Wheatley, 1995), subsequent studies have revealed that immune cells appear to form “frustrated cilia” with centrosome docking but no cilia formation (Griffiths et al., 2010).

Striking parallels between the immune synapse and primary cilia were first noticed at the ultrastructural level, when images of centrosome polarization at the CTL immune synapse were compared with images of basal body docking during ciliogenesis (Stinchcombe et al., 2006; Stinchcombe and Griffiths, 2007). In both ciliogenesis and immune synapse formation, it was clear that docking of the centrosome at the plasma membrane also resulted in polarization of the Golgi apparatus and endocytic compartments toward the site of centrosome attachment, creating a focal point for endocytosis and exocytosis at a defined point on the plasma membrane (Griffiths et al., 2010; Stinchcombe and Griffiths, 2007). Many more molecular parallels between the immune synapse and primary cilia have emerged since these initial observations, and it is now clear that there are multiple molecular and functional similarities between these two seemingly disparate biological structures (Table 1).

The centrosome is composed of two centrioles, the mother centriole with distal and subdistal appendages and the daughter centriole that lacks appendages, surrounded by the pericentriolar material. Primary cilia formation is well described (Reiter and Leroux, 2017); in brief, it involves a series of well-orchestrated steps, initiated by the docking of the mother centriole, via its distal appendages, to what will become the ciliary membrane (Anderson, 1972; Mirvis et al., 2018; Sorokin, 1962). Microtubules extend from the docked mother centriole to form the ciliary axoneme composed of nine doublet microtubules, giving rise to the classical “antenna-like” cilium. In some cellular contexts, vesicles originating from the Golgi and recycling endosomes accumulate around the mother centriole and form a cap-like structure, termed the ciliary vesicle or cap, to which the mother centriole docks. This becomes the ciliary membrane following fusion with the surface (Lu et al., 2015; Reiter et al., 2012; Wang and Dynlacht, 2018; Yee and Reiter, 2015).

A combination of proteomic and super resolution microscopy approaches has led to the identification of several distal appendage proteins essential for ciliogenesis, including CEP164, CEP83, CEP89, SCLT1, C2CD3, and LRRC45 (Graser et al., 2007; Joo et al., 2013; Kurtulmus et al., 2018; Schmidt et al., 2012; Sillibourne et al., 2013; Tanos et al., 2013; Ye et al., 2014). Interestingly, the recruitment of distal appendage proteins to the mother centriole seems to be hierarchical. Use of direct stochastic optical reconstitution microscopy revealed that CEP83 is located at the root of the appendages, while CEP164 accumulates at their periphery (Bowler et al., 2019; Yang et al., 2018). This supports earlier observations that removal of CEP83 impairs the recruitment of CEP164, CEP89, and SCL1 to distal appendages and effectively disrupts their assembly (Tanos et al., 2013). Deletion of CEP83 through RNA interference prevented docking of mother centrioles at the plasma membrane (Tanos et al., 2013), with a subsequent study using CRISPR further demonstrating that CEP83 ablation impairs ciliary vesicle formation (Lo et al., 2019). The finding that mutations in CEP83 are associated with infantile nephronophthisis (NPHP18), a ciliopathy characterized by chronic renal failure (Failler et al., 2014), reinforces a critical role of CEP83 in ciliogenesis. Interestingly, the removal of distal appendages by CEP83 deletion affected subdistal appendage stability and localization, demonstrating a structural coupling of the distal and subdistal appendages (Chong et al., 2020). Of note, it was recently reported that CEP83 is involved in centrosome membrane anchoring in radial glial progenitor cells, indicating an additional role for CEP83 distinct from ciliogenesis (Shao et al., 2020).

An important role for mother centriole distal appendages has also been described in immune cells. Transmission EM tomography revealed that the distal appendages of the CTL mother centriole associate with the plasma membrane upon killing, anchoring the centrosome at the immune synapse (Stinchcombe and Griffiths, 2014; Stinchcombe et al., 2015). The docking step organizes microtubules emanating from the subdistal appendages of the mother centriole to run underneath the synapse and thus guides microtubule-mediated granule transport toward the site of exocytosis (Stinchcombe et al., 2011, 2015; Fig. 1). Intriguingly, CEP83 seems to play an important part in this process in CTLs, as silencing it using RNA interference reduced granule secretion in CTLs (Stinchcombe et al., 2015).

Tight polarization of the centrosome is an early hallmark of both ciliogenesis and immune synapse formation. However, subsequent events diverge, as centrosome docking at the synapse is transient, lasting only minutes, while primary cilia are stable structures persisting for several hours. Two proteins that regulate the stable association of the centrosome with the membrane during ciliogenesis are CP110 and its partner, CEP97, which dissociate from the distal end of the mother centriole, allowing centrosomal docking and axoneme growth. Loss of Cep97 or CP110 promotes primary cilia formation, while ectopic expression of CP110 thwarts it, suggesting the complex suppresses ciliogenesis (Spektor et al., 2007). This is thought to be mediated by the serine/threonine kinase Tau-tubulin kinase-2, which phosphorylates CEP83 to promote CP110 removal from the mother centriole (Bernatik et al., 2020; Goetz et al., 2012; Lo et al., 2019). In contrast, in CTLs the CP110–CEP97 complex is not lost from the mother centriole upon docking, and cilia formation does not occur (Stinchcombe et al., 2015). Interestingly, RNA interference of CP110 in Jurkat T cells was able to induce some primary cilia formation upon serum starvation, indicating immune cells have the capacity to make cilia when CP110 is removed (Prosser and Morrison, 2015). Together, these findings suggest that the retention of CP110–CEP97 might explain why a cilium does not form at the immune synapse (Fig. 1).

To explore the role of centrosome homeostasis and assess the contribution of centrosome docking during CTL killing, one study used a Sas4/p53 deletion model (Tamzalit et al., 2020). As CTLs have been shown to contain multiple pairs of centrioles (Stinchcombe et al., 2015), this is a challenging problem to approach. While Sas4/p53 knockout lymphocytes displayed reduced killing, polarized secretion was unscathed probably due to the maintenance of a rudimentary microtubule organizing center, albeit with a less well-organized microtubule cytoskeleton. However, the Sas4/p53 CTLs were also disrupted in other ways, including alterations in actin remodeling and granule content, both of which would have contributed to the defect in killing (Tamzalit et al., 2020). A recent study showing that pericentriolar material can lead to cilia formation in the absence of centrioles in Caenorhabditis elegans raises the intriguing possibility that centrosome function may also be retained in T cells in the absence of centrioles (Garbrecht et al., 2021).

In addition to changes in the microtubule organization, the formation of both the immune synapse and primary cilium involves rearrangement of the actin cytoskeleton.

Early interaction between a CTL and its target is mediated by actin-rich lamellipodial protrusions that form an interdigitated contact site with the target cell and lead to a rapid accumulation of actin across the contact site (Jenkins et al., 2014; Ritter et al., 2015; Sanderson and Glauert, 1979). The initial actin accumulation at the synapse is quickly followed by its depletion across the structure, as the membrane reorganizes to form a tight contact during killing (Fig. 2). Actin then returns to the contact site before the CTL dissociates and moves on to further targets (Ritter et al., 2015; Stinchcombe et al., 2001). Loss of cortical actin is required for granule delivery, while its recovery drives secretion termination (Ritter et al., 2017). The rapid changes in actin distribution at the synapse are controlled by changes in the plasma membrane phospholipid composition in response to TCR signaling (Gawden-Bone et al., 2018; Ritter et al., 2015). In brief, TCR ligation activates PLCγ, triggering the depletion of phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2), a lipid known to interact with actin-regulating proteins (Rohatgi et al., 2000). Consequently the TCR-induced loss of PI(4,5)P2 reduces actin recruitment where TCR signaling occurs (Gawden-Bone et al., 2018). The conversion of PI(4,5)P2 to DAG also alters the membrane charge required for the electrostatic interactions for phosphatidylinositol 4-phosphate 5-kinase type I (PIP5K) membrane association, excluding PIP5K and ensuring actin depletion from the point of centrosome docking at the synapse (Gawden-Bone et al., 2018; Fig. 3).

The dynamic changes at the immune synapse produce a distinctive membrane specialization during centrosome docking. This resembles the membrane specialization described across the primary cilium, arising from changes driven by inositol polyphosphate 5-phosphatase (INPP5E), a phosphatase that dephosphorylates PI(4,5)P2 to generate phosphatidylinositol 4-phosphate (Dyson et al., 2012). Studies using lipid-specific antibodies (Chávez et al., 2015) or expression of phosphoinositide-binding domains fused to fluorescent proteins (Garcia-Gonzalo et al., 2015) revealed that phosphatidylinositol 4-phosphate is enriched within the ciliary membrane, whereas PI(4,5)P2 and PI(3,4,5)P3 are localized at the base of the cilium, providing a very similar membrane architecture to that across the synapse, with actin depleted in both cases (Fig. 3). INPP5E, required for Hedgehog (Hh) signaling, is found along the cilia and marshals this compartmentalization (Chávez et al., 2015; Dyson et al., 2017; Garcia-Gonzalo et al., 2015; Nakatsu, 2015). Intriguingly, while TCR signaling displaces PIP5K from the immune synapse, growth signals displace INPP5E from the cilium, driving PI(4,5)P2 accumulation at the tip of the cilium, allowing intraciliary actin polarization, which plays a part in cilia shedding and disassembly (Phua et al., 2017). Thus, both the immune synapse and the primary cilium undergo dynamic remodeling of their membrane phospholipid composition, driving actin rearrangement and supporting their respective functions.

Studies in CD4⁺ T cells have shed some light on the roles of actin regulatory proteins in immune synapse function. Briefly, T cell stimulation leads to the activation of actin regulators WAVE2 and WASp that act in concert with Arp2/3 complexes to promote actin nucleation. While WAVE2 localizes at the lamellipodial protrusions, participating in their formation (Nolz et al., 2006; Pauker et al., 2014), WASp modulates the formation of TCR-rich actin foci (Comrie and Burkhardt, 2016; Kumari et al., 2015). Several studies have shown that CTLs lacking the Arp2/3 subunit ARPC1B are unable to form actin-rich lamellipodial protrusions, impairing synapse formation (Brigida et al., 2018; Randzavola et al., 2019; Somech et al., 2017; Volpi et al., 2019). However, in CTLs, loss of ARPC1B also disrupted retromer-mediated recycling of receptors, including TCR and the glucose transporter GLUT-1, impairing signaling and glucose uptake required for cell proliferation (Randzavola et al., 2019). Intriguingly, it has been proposed that centrosomes act as actin organizing centers by nucleating actin filaments in a process mediated by the nucleation promoting factor Wiskott-Aldrich syndrome protein and SCAR homologue (WASH) in combination with the Arp2/3 complex (Farina et al., 2016). Actin reorganization at the centrosomal area seems to contribute to polarized secretion of multivesicular bodies from Jurkat T cells (Bello-Gamboa et al., 2020).

Actin rearrangements also participate in primary cilia formation (Copeland, 2020), and a number of proteomic screens have identified actin-binding proteins in the cilium (Gupta et al., 2015; Ishikawa et al., 2012; Kohli et al., 2017; Liu et al., 2007). Arp2/3 plays a role in primary cilia, with deletion of the Arp3 subunit leading to increased ciliogenesis and ciliary length (Kim et al., 2010; Yan and Zhu, 2013) as well as increased Hh signaling (Drummond et al., 2018), suggesting that F-actin nucleation inhibits primary cilia formation (Kim et al., 2010; Yan and Zhu, 2013). Moreover, actin dynamics at the centrosome also play a part in ciliogenesis, as preciliary vesicles seem to be transported from the pericentrosomal region to the distal appendages of the mother centriole to form the ciliary vesicle via an Arp2/3 branched actin network (Wu et al., 2018; Fig. 1).

Both primary cilia and immune synapses can be viewed as specialized signaling hubs, with each relying on the correct spatial and temporal recruitment of proteins for signal transduction. Over the last decade, a large body of evidence has accumulated which suggests the cilium and synapse share common mechanisms to ensure polarized trafficking (Fig. 4). The best characterized of these are the intraflagellar transport (IFT) proteins. In cilia, IFT proteins associated with their cargo are transported bidirectionally along axonemal microtubules by the molecular motors kinesin 2 (base to tip) and dynein 2 (tip to base; Taschner and Lorentzen, 2016). IFTs participate in the active sorting and transport of cytosolic and membrane proteins destined for the cilium past the impermeable transition zone to the axoneme (Pedersen and Rosenbaum, 2008). Mutations in IFTs give rise to a number of human ciliopathies, supporting their critical role in ciliogenesis and cilium maintenance (Reiter and Leroux, 2017).

Interestingly, a number of IFT components have also been found to be expressed in hematopoietic cells and shown to participate in T lymphocyte activation (Finetti et al., 2009; Finetti et al., 2014). For instance, IFT20 is recruited to the immune synapse formed between a T lymphocyte and an antigen-presenting cell via association with the centrosome as well as the Golgi and post-Golgi compartments (Finetti et al., 2009). This echoes observations that IFT20 associates with the Golgi apparatus at the base of the cilium and functions in the sorting of proteins destined for the ciliary membrane (Follit et al., 2008; Follit et al., 2006). In addition, a study using CD4⁺ T lymphocytes derived from mice with a conditional defect in IFT20 highlighted a role for IFT20 in the recruitment to the immune synapse of linker of activated T cells (LAT), a key adaptor involved in T cell activation to the immune synapse (Vivar et al., 2016).

A proteomic analysis of the interactome of IFT20 in Jurkat cells using quantitative mass spectrometry identified binding partners for the protein (Galgano et al., 2017). These included IFT54 and the Golgin GMAP-210, which were previously described to associate with IFT20 in ciliated cells (Follit et al., 2008; Follit et al., 2009). IFT54 participates in IFT20-mediated TCR recycling, possibly through its role as a negative regulator of microtubule stability (Bizet et al., 2015). Although GMAP-210 was not required for TCR recruitment to the synapse (Galgano et al., 2017), it was found to mediate delivery of LAT-enriched vesicles. Furthermore, ectopic expression of LAT in ciliated cells led to its recruitment to the cilium in a GMAP-210–dependent manner, supporting the idea that similar molecular mechanisms are involved (Zucchetti et al., 2019). It is important to note that the IFTs and particularly IFT20 seem to have general Golgi-related trafficking functions in several cellular contexts that are not restricted to the primary cilia or immune synapse. For example, IFT20 is essential for procollagen trafficking from the endoplasmic reticulum to the Golgi apparatus in central nervous system–derived osteoblasts (Noda et al., 2016).

Endosomal trafficking at the immune synapse and primary cilia share additional molecular effectors (Fig. 4). Indeed, IFTs recruit small Rab GTPases that orchestrate receptor recycling at both structures. Rab8, an essential player in cilium assembly and trafficking (Nachury et al., 2007; Westlake et al., 2011), colocalizes with IFT20 in Rab11⁺ endosomes in activated lymphocytes, and this interaction has been proposed to contribute to vesicle-mediated TCR targeting to the immune synapse (Finetti et al., 2015). The same group also identified Rab29 as an interactor of IFT20, Rab8, and Rab11 (Onnis et al., 2015).

In addition to Rab GTPases, the small GTPase ARL3 and its guanine exchange factor, ARL13B, appear to regulate membrane composition and recruitment of signaling molecules in both the cilium and synapse (Powell et al., 2021). The pivotal role of ARL3 in cilia is clear from the ciliary phenotypes associated with its mutations, including Joubert syndrome and dominant retinitis pigmentosa (Alkanderi et al., 2018; Holtan et al., 2019; Strom et al., 2016). Briefly, ARL3 and ARL13B were previously shown to play a role in mediating solubility of prenylated and myristoylated proteins, thereby controlling recruitment of several proteins to the axoneme (Powell et al., 2021). Intriguingly, the ARL3/ARL13B machinery has been shown to mediate transport of the downstream signaling effector protein Lck to the immune synapse of T cells, revealing an alternative use for this ciliary mechanism for targeting the delivery of signaling proteins in immune cells (Stephen et al., 2018).

Together with the IFTs and small Rab GTPases, ciliated and immune cells share trafficking pathway adaptors, including SNARE family proteins involved in the priming and fusion of vesicles to the plasma membrane. The SNARE VAMP7 is essential for cilium biogenesis in kidney cells (Szalinski et al., 2014) and has also been shown to be required for the recruitment of LAT-containing vesicles to TCR activation sites and consequent cell activation in CD4⁺ T lymphocytes (Larghi et al., 2013). In addition, VAMP7 localizes to cytolytic granules and was proposed to play a role in granule fusion at the synapse (Chitirala et al., 2019). Moreover, the myelin and lymphocyte (MAL) protein participates in primary cilia formation as well as modulating the sorting of Lck and LAT at the immune synapse (Antón et al., 2011; Takiar et al., 2012).

Together, these observations indicated that multiple components of the trafficking machinery, including IFTs, GTPases, and SNAREs, act in both ciliated and immune cells to ensure proper recruitment of signaling components to the axoneme and synapse, as well as their recycling, which is essential for signal termination (Fig. 4).

As part of its role as the “small antenna” of the cell, the ciliary membrane contains a high concentration of transduction receptors. One of the best-described signaling pathways in the cilium is the Sonic Hh pathway (Goetz and Anderson, 2010). Briefly, upon binding of the transmembrane receptor Patched (PTCH) to one of the three Hh ligands (Sonic Hh, Desert Hh, or Indian Hh), Smo is transported to the axoneme by IFTs, where it can disrupt the repression of Gli1 at the tip of the cilia. Gli1 is then activated and free to translocate to the nucleus, where it acts as a transcription factor for Hh genes (Corbit et al., 2005; Rohatgi et al., 2007).

Interestingly, the study of knockout and transgenic models has established a role for Hh signaling in the development of T lymphocytes in the thymus as well as the differentiation of CD4⁺ T lymphocytes into T helper 2 cells (Crompton et al., 2007; Furmanski et al., 2013). Furthermore, Hh signaling has also been shown to be triggered upon TCR activation and contribute to the cytotoxic synapse as detected by Gli1 activation, while deletion or inhibition of Smo disrupted actin rearrangements and decreased CTL killing (de la Roche et al., 2013). These findings suggest a model where the Hh pathway modulates actin rearrangements, controlling CTL secretion (de la Roche et al., 2013).

A growing body of evidence suggests that the primary cilium is a site of extracellular vesicle (EV) release. The studies of the unicellular alga Chlamydomonas reinhardtii and the nematode C. elegans have demonstrated that ectosomes can be released from flagella and primary cilia and mediate extracellular communication. In C. reinhardtii, early electron microscopy showed shedding of membrane microvesicles from the flagella tip (Bergman et al., 1975). Subsequent studies found that these 50- to 200-nm ectosomes contain proteases required for degradation of extracellular matrix and digestion of the mother cell wall required for daughter cell hatching (Wood et al., 2013; Wood and Rosenbaum, 2015). Vesicle fluorescence labeling and live and electron microscopy revealed that ciliated sensory neurons of C. elegans can release microvesicles into their environment (Wang et al., 2014). This process relies on several ciliary proteins and IFT components and participates in animal communication and mating behaviors. Interestingly, in this context, observations of microvesicle shedding have been made both at the base and the tip of the cilia, suggesting two distinct ciliary secretion sites (Wang and Barr, 2018). An alternative model suggests that EVs could be shed at the base of the cilium then transported along the ciliary membrane by the IFT to be released at the tip of the axoneme (Maguire et al., 2015). Exosome-like vesicles have been shown to accumulate around the cilium in the cells of patients suffering from autosomal dominant polycystic kidney disease, as well as in a mouse model recapitulating this human ciliopathy (Hogan et al., 2009). Subsequent studies have found that actin mediates vesicle shedding and ciliary disassembly, termed ectocytosis and decapitation, respectively (Nager et al., 2017; Phua et al., 2017). In one study, the shed vesicles were shown to carry ciliary G protein–coupled receptors (GPCRs), thereby modulating signaling (Nager et al., 2017), while another report focused on a role for these EVs in the disassembly of the cilium (Phua et al., 2017).

A number of studies now suggest that lymphocytes also release exosome-like vesicles at the synapse, in addition to secreted proteins, although the origins of these vesicles is not always clear. Early observations suggested T cells shed vesicles containing the activation-induced cell death mediators FasL and APO2L (Alonso et al., 2011; Martínez-Lorenzo et al., 1999) and TCR–CD3–ζ complexes (Blanchard et al., 2002), as well as mRNA (Mittelbrunn et al., 2011). A more recent study has shown that FasL is localized on the intra luminal vesicles within cytolytic granules, revealing the origins of vesicle released FasL (Lee et al., 2018). Work using a TCR-triggering planar lipid bilayer synapse showed that microvesicles containing TCR were secreted in an endosomal sorting complexes required for transport (ESCRT)–dependent manner, and these could in turn activate surrounding cells (Choudhuri et al., 2014). In keeping with this, exosome-like vesicles isolated from CTLs stimulated with high-affinity ligands could activate bystander CTLs in the absence of antigen (Wu et al., 2019). However, whether these supernatant-isolated exosomes originated from the synapse remains to be determined.

The initial observations of morphological similarities in transmission EM micrographs of centrosome polarization at the immune synapse and primary cilium (Stinchcombe et al., 2006; Stinchcombe and Griffiths, 2007; Stinchcombe and Griffiths, 2014; Stinchcombe et al., 2011) paved the way for what is now a large body of research, extending the range of structural, functional, and molecular similarities and so highlighting the relationship between the cilium and synapse.

One reason the similarity between the synapse and cilium appeared so intriguing in the early studies is that immune cells were known to lack primary cilia despite having a functional centrosome. While not an absolute rule, there seems to be a correlation between the presence of a centriole and the ability to form a cilium in most organisms. Organisms such as Dictyostelium discoideum and Saccharomyces cerevisiae, which lack cilia altogether, are devoid of centrosomes (Debec et al., 2010).

The observation that the CTL centrosome polarizes to the synapse and docks with the membrane via the mother centriole distal appendages upon target contact but fails to form a cilium was therefore particularly striking and indicated that centrosomal docking leads to different outcomes in ciliated and immune cells. This difference is likely critical for the distinct functional requirements of the transient immune synapse versus the stable cilium in ciliated tissues. The similarities in the docking process, nevertheless, suggested immune cells such as CTL might have “commandeered” mechanisms evolved for cilia formation to provide a fast mechanism for repolarizing the cell in response to target encounter (Stinchcombe et al., 2006). The growing number of key molecular players in the range of critical cellular processes that are shared between the synapse and the cilium, from microtubule and actin cytoskeletons rearrangements to membrane trafficking (e.g., IFTs, SNARE components, and GTPases) to signaling pathways (e.g., Hh), as well as common mechanisms (such as the release of EVs), supports the idea that lymphocytes have coopted ciliary machinery to fulfil their cytotoxic functions and hints at an evolutionary link between these structures. Much of our understanding of ciliogenesis came from the study of ciliopathies, a subset of human inherited developmental degenerative diseases (Reiter and Leroux, 2017). Indeed, mutations in genes involved in ciliary function result in a variety of symptoms ranging from polydactyly to blindness to kidney diseases or neurodevelopmental defects (Waters and Beales, 2011). Whether some ciliopathies are accompanied by immune symptoms remains unclear and warrants further investigation. Comparisons between the immune synapse and ciliogenesis have yielded a multitude of molecular similarities between these seemingly disparate biological structures, illustrating how it is possible to tease out function from morphology.

We would like to thank Carla Vazquez-Amos for critical reading of the manuscript.

This work was funded by Wellcome Trust grant 217100. T. Douanne is an awardee of the Foundation Bettencourt Schueller Young Researcher Prize.

G.M. Griffiths belongs to scientific advisory boards for Adaptate and Biotheus.

This research was funded in whole or in part by Wellcome Trust. For the purpose of Open Access, the author has applied a CC-BY public copyright license to any Author Accepted Manuscript (AAM) version arising from this submission.