BBSome trains remove activated GPCRs from cilia by enabling passage through the transition zone

Diffusion barriers establish the identity of the apical membrane in polarized epithelial cells, of the axon in neurons, of the daughter cell in budding yeast, and of cilia by impeding the lateral movement of membrane proteins (Trimble and Grinstein, 2015). The compartmentalization of cilia enables dynamic changes in ciliary composition through regulated trafficking. Upon Hedgehog pathway activation, the seven-transmembrane-protein Smoothened accumulates in cilia, and ciliary exit of the G protein–coupled receptor (GPCR) GPR161 ensures the appropriate transduction of Hedgehog signals (Bangs and Anderson, 2017; Nager et al., 2017). Although trafficking across the tight junction, the axon initial segment, and the yeast bud neck involves a vesicular carrier intermediate, the mechanisms of ciliary barrier crossing remain undetermined. A ciliary diffusion barrier has been localized to the transition zone, an ultrastructural specialization between the transition fibers of the basal body and the cilium shaft (Garcia-Gonzalo and Reiter, 2012; Gonçalves and Pelletier, 2017). Three hypotheses have been advanced for crossing the transition zone (Nachury et al., 2010; Jensen and Leroux, 2017). First, the detection of vesicles inside the transition zone indicates that a vesicular carrier may transport cargoes across this barrier (Jensen et al., 2004; Chuang et al., 2015). Second, indirect evidence for lateral transport between plasma and ciliary membranes (Hunnicutt et al., 1990; Milenkovic et al., 2009) suggests that membranous cargoes laterally traverse the transition zone by active transport. Third, the regulated opening of a gate inside the transition zone may let selective cargoes move through this membranous barrier (Dyson et al., 2017).

The active transport of proteins inside cilia termed the intraflagellar transport (IFT) is powered by microtubule motors moving along axonemal microtubules. It is now clear that axonemal precursors such as α/β-tubulin are delivered to the tips of cilia by anterograde IFT trains (Lechtreck, 2015; Kubo et al., 2016). In contrast, it is not known where, when, or how membrane proteins are selected for ciliary exit, and prior studies of ciliary signaling receptors dynamics by single-molecule imaging failed to uncover extended IFT movements (Ye et al., 2013; Milenkovic et al., 2015).

To characterize transition zone crossing by membranous cargoes, we sought a system where membrane proteins move across the transition zone in a synchronized manner. GPR161 and the prototypical ciliary GPCR somatostatin receptor 3 (SSTR3) both undergo retrieval from the cilium and back into the cell upon activation (Mukhopadhyay et al., 2013; Green et al., 2015). GPR161, a core component of the Hedgehog pathway that couples to Gαs, exits cilia when Smoothened is activated either indirectly by Hedgehog or directly by Smoothened agonist (SAG; Fig. 1 A; Pal et al., 2016). Meanwhile, SSTR3 is a well-characterized Gαi-coupled receptor that undergoes agonist-dependent retrieval (Fig. 1 A; Green et al., 2015; Nager et al., 2017). Consistent with previous research on SSTR3 in hippocampal neurons (Green et al., 2015), most of SSTR3 immunofluorescence was lost from neuronal cilia after 6 h treatment with the ligand somatostatin-14 (sst) or the SSTR3-specific agonist L796,778 (Figs. 1 B and S1, A–C). Similar to endogenous SSTR3, ciliary exit of endogenous GPR161 proceeds over the course of several hours (Mukhopadhyay et al., 2013). Signal-dependent retrieval is thus a considerably slower process than signal-dependent endocytosis.

To dissect ciliary exit, we expressed GPCRs in mouse inner medullar collecting duct 3 (IMCD3) kidney cells, a widely used cell line for ciliary trafficking studies. GPCRs were tagged on the intracellular C terminus with a fluorescent protein (GFP or NeonGreen [NG]; Shaner et al., 2013), whereas a biotinylation acceptor peptide (AP) on the extracellular N terminus combined with coexpression of the biotin ligase BirA enabled pulse-chase studies with fluorescently labeled monovalent streptavidin (mSA; Howarth and Ting, 2008). When ^APSSTR3^GFP under the control of the EF1α promoter was stably expressed in IMCD3 cells by single integration at the FlpIn locus, agonist-dependent exit of SSTR3 from cilia was undetectable. Molecular counting of GFP and comparison of immunofluorescence intensities revealed that pEF1α-driven expression resulted in SSTR3 levels that were an order of magnitude greater than in neurons (Fig. 1 C, top). Congruently, pEF1α-driven ^APSSTR3^GFP expression resulted in a near doubling of cilia length, likely caused by protein overload driving ciliary membrane expansion and compensatory axoneme growth (Fig. 1 C, bottom; Guadiana et al., 2013). To express SSTR3 at levels closer to those found in neurons, we tested a variety of weak promoters and found that an EF1α promoter lacking the TATA box (pEF1αΔ) produced ciliary amounts of SSTR3 similar to those found in neurons (Fig. 1 C, top). pEF1αΔ-driven expression of SSTR3 did not alter ciliary length (Fig. 1 C, bottom). IMCD3-[pEF1αΔ-^APSSTR3^NG] cells recapitulated SSTR3 exit from cilia upon sst addition with nearly identical kinetics as in hippocampal neurons (Figs. 1 B and S1 D). Similarly, the Hedgehog signaling–dependent exit of GPR161 was recapitulated by expressing ^APGPR161^NG3 from the δ-crystallin promoter (Fig. 1 D). pEF1αΔ-driven expression of NPY2R^NG and NG-tagged melanin-concentrating hormone receptor 1 (MCHR1; MCHR1^NG) yielded low ciliary levels (Fig. S1 E) with similar exit kinetics (Nager et al., 2017), thus demonstrating the broad applicability of low-expression promoters for studying the dynamics of ciliary GPCRs.

Calibration of the NG signal with recombinant proteins spotted on glass slides (Fig. 1 E) allowed the measurement of absolute levels of GPCRs per cilia (Fig. S1, F–H) and, together with pulse-chase labeling with mSA647, enabled a specific quantitation of signal-dependent exit rates at close to 500 molecules per hour (Fig. 1, F and G; and Video 1). In support of the precision of our absolute quantitation, using pEF1αΔ-driven SSTR3^NG or SSTR3^GFP and independent calibrators yielded very similar numbers of SSTR3 molecules per cilium (Fig. 1 C).

Further highlighting the power of AP- and NG-tagged GPCRs, the increased signal/noise ratio afforded by the direct labeling with NG or mSA647 compared with immunofluorescence (Fig. S1 I) made it possible to detect very low-abundance proteins (Fig. 1 C) whose presence in cilia escaped detection by traditional immunostaining techniques (Fig. S1 J). The decreased threshold of detection when pulse-labeling with mSA647 ensured a more faithful visualization of exit kinetics than when exit was monitored by immunostaining (Fig. 1, B and F).

In the absence of an agonist, the NG fluorescence of ^APSSTR3^NG increased over time, whereas the signal from pulse labeling with mSA647 remained constant for 6 h (Fig. S1 K). In the presence of an agonist, the exit kinetics of ^APSSTR3^NG were slower when monitored by direct visualization of the NG tag than by pulse-labeling with mSA647 (Fig. S1 K). Because pulse labeling only reports on SSTR3 exit, whereas the NG signal measures the total ciliary levels, these results indicate that the newly synthesized ^APSSTR3^NG continues to enter cilia during the course of the experiment. We surmise that our previous attempts to assay signal-dependent exit of SSTR3 using strong promoters failed because the entry of newly synthesized GPCRs outpaced the slow exit kinetics.

The low-expression systems made it possible to validate the sorting complexes that carry out ciliary entry and exit. IFT-A is a complex of six proteins with structural elements that suggest a common ancestry with coat complexes (Jékely and Arendt, 2006; van Dam et al., 2013), and IFT-A is recruited to membranes by the PI(4,5)P2-binding protein Tulp3 (Mukhopadhyay et al., 2010). Although IFT-A is often described as the central mediator of retrograde transport (Lechtreck, 2015), IFT-A and TULP3 are required for the import of many GPCRs into cilia (Mukhopadhyay et al., 2010, 2013; Loktev and Jackson, 2013; Fu et al., 2016; Badgandi et al., 2017). We confirmed that Tulp3 is required for ciliary entry of SSTR3, NPY2R, and MCHR1 (Figs. 2 A and S2, A–C) and refined indirect interaction data by showing that the ciliary targeting signal of SSTR3 encoded within the third intracellular loop (i3; Berbari et al., 2008a) was specifically and directly recognized by the purified IFT-A complex (Fig. 2 B) or by the IFT-A subunit IFT140 overexpressed in HEK cells (Fig. S2 D). We conclude that IFT-A/Tulp3 functions as a coat adapter complex that mediates GPCR entry into cilia by directly recognizing sorting signals (Fig. 2 C).

Consistent with the requirement for the GPCR activation sensor β-arrestin 2 in GPR161 retrieval (Pal et al., 2016), signal-dependent retrieval of SSTR3 required β-arrestin 2 (Fig. 2, D and E; and Fig. S2, E and F; Green et al., 2015). The BBSome, a complex of eight Bardet-Biedl syndrome (BBS) proteins, resembles coat adapters at the structural level and polymerizes into a planar coat upon recruitment to membranes by the GTP-bound form of the small GTPase Arl6/BBS3 (Jin et al., 2010). The function of the BBSome in entry versus exit remains controversial. Although BBSome mutants have decreased ciliary levels of the GPCRs SSTR3, MCHR1, and NPY2R and of the polycystic kidney disease protein PKD1 (Berbari et al., 2008b; Loktev and Jackson, 2013; Su et al., 2014), GPR161, Smoothened, and D1R fail to exit cilia in BBSome or Arl6 mutants (Zhang et al., 2011; Eguether et al., 2014; Liew et al., 2014; Yee et al., 2015; Nager et al., 2017). Finally, systematic studies find that some proteins accumulate whereas others are depleted from Bbs mutant cilia (Lechtreck et al., 2013; Datta et al., 2015; Mick et al., 2015). In our near-endogenous expression systems, deletion of Arl6, of the candidate Arl6 activator Ift27/BBS19, or of β-arrestin 2 did not reduce the steady-state ciliary levels of SSTR3 (Figs. 2 A and S2 F). Instead, the BBSome, Arl6, and Ift27 were required for the signal-dependent retrieval of SSTR3 and GPR161 (Fig. 2, D and E; and Fig. S2, G and H). The carboxy-terminal tail of GPR161 (GPR161^Ct) and the i3 of SSTR3 (SSTR3ⁱ³) directly interacted with purified BBSome (Fig. 2 F; Jin et al., 2010), suggesting that BBSome coats sort signaling receptors through the direct recognition of cytoplasmic determinants. Because SSTR3 and GPR161 are recognized by β-arrestin 2 in a signal-dependent manner (Roth et al., 1997; Pal et al., 2016), we conclude that the signal-dependent retrieval of GPR161 and SSTR3 is jointly and directly mediated by β-arrestin 2 and the BBSome (Fig. 2 G).

To characterize the mechanisms of transition zone crossing by exiting GPCRs, we first sought to determine how BBSome coats facilitate signal-dependent retrieval. The movement of trains consisting of the IFT complex B (IFT-B) can be visualized by imaging foci of the IFT-B subunit IFT88 tagged with NG traveling in the anterograde and retrograde direction inside cilia (Video 2). Imaging of nematode, Chlamydomonas rheinhardtii, and mammalian cilia has shown that BBSome foci frequently comove with IFT-B foci (Ou et al., 2005; Lechtreck et al., 2009; Liew et al., 2014; Williams et al., 2014), suggesting coupling between the two complexes. To follow the dynamics of the BBSome and of IFT-B during SSTR3 and GPR161 signal-dependent retrieval, we expressed ^NG3BBS5 or ^NG3IFT88 at near-endogenous levels (Figs. 3 A and S2 I, and Video 3). In the absence of signaling, the BBSome and the IFT-B complex localized in a punctate pattern along the cilium (Fig. 3, B and C; and Fig. S2, J–L). Unexpectedly, SSTR3 activation led to a fourfold enrichment of the BBSome and a twofold enrichment of IFT-B at the tip (Fig. 3, B and C; and Fig. S2, J–L). Likewise, activation of the Hedgehog pathway resulted in BBSome accumulation at the ciliary tip (Fig. 3 C and Video 4). Similar to ^NG3BBS5, endogenous BBS9 became enriched at the tip upon activation of the Hedgehog pathway or of SSTR3 (Fig. 3 D). The BBSome thus joins a select group of Hedgehog factors that localize to the tip in a signal-dependent manner consisting of Gli2, Gli3, suppressor of fused homolog (SuFu), and Kif7.

To determine whether tip enrichment of the BBSome represents a necessary step in GPCR retrieval, we sought to define and manipulate the molecular mechanisms of signal-dependent tip accumulation. Because SSTR3 and Smoothened are known to couple with Gαi and reduce cAMP production through Gαi-mediated inhibition of adenylate cyclases 5 and 6 (AC5/6; Yasuda et al., 1992; Shen et al., 2013), we tested for the role of signaling downstream of Gαi in promoting BBSome tip enrichment (Fig. 3 E). Pharmacological inhibition of Gαi by Pertussis toxin (PTX) blocked SAG- and sst-induced BBSome tip accumulation (Fig. 3 C). Meanwhile, inhibition of AC6 (Fig. S2 M) or inhibition of the cAMP-dependent protein kinase (PKA) by Rp-cAMPs or by a cell-permeable PKA inhibitory peptide (PKI; myr-PKI) led to BBSome tip accumulation in the absence of GPCR activation (Fig. 3 C). Furthermore, cilia-targeted PKI (Mick et al., 2015) was sufficient to redistribute BBSome to the tip (Fig. 3 F). Together with the findings of cilia-localized AC5 and AC6 (Masyuk et al., 2008; Kwon et al., 2010; Mick et al., 2015), these results suggest that activation of Smoothened and SSTR3 reduce the tonically high levels of ciliary cAMP (Moore et al., 2016) through Gαi-mediated inhibition of AC5/6 within cilia. Because PKA was recently shown to reside and function inside cilia (Mick et al., 2015; Moore et al., 2016), we propose that PKA antagonizes the recruitment of BBSome to the tip of cilia in unstimulated cells and that activation of Gαi-coupled GPCRs promotes BBSome tip recruitment by reducing the activity of ciliary PKA (Fig. 3 E).

Importantly, pharmacological alterations of the ciliary Gαi–PKA axis concordantly affected signal-dependent redistribution of BBSome to the tip of cilia and signal-dependent GPCR retrieval as the rates of signal-dependent retrieval of GPR161 and SSTR3 were greatly reduced by PTX (Fig. 4, A and B) and significantly accelerated by myr-PKI (Fig. 4, C and D). We note that PKA inhibition was not sufficient to trigger retrieval of GPR161 or SSTR3 in the absence of receptor stimulation (Fig. 4, C and D), suggesting the existence of mechanisms that act nonredundantly with the Gαi–PKA–BBSome axis (e.g., β-arrestin 2). In support of a Gαi–PKA–BBSome axis promoting GPCR exit, SSTR3 activation was sufficient to elicit the retrieval of GPR161 with identical kinetics to Smoothened activation (Fig. 4 E). In contrast, Smoothened activation was not sufficient to promote SSTR3 retrieval (Fig. 4 F). As the mechanisms that underlie the activation of GPR161 remain unknown, it is conceivable that signaling downstream of SSTR3 (through Gαi or Gβγ) triggers activation of GPR161 and that the subsequent engagement of β-arrestin 2 onto GPR161 cooperates with the Gαi–PKA–BBSome axis to promote GPR161 retrieval. In contrast, SSTR3 can only be activated by specific ligands.

To further establish that BBSome tipping represents a necessary intermediate in GPCR retrieval, we sought to identify molecules that recruit the BBSome to the tip of cilia in a signal-dependent manner. The plus end–directed microtubule motor Kif7 represented a candidate tip recruitment factor because Kif7 accumulates at the tip of cilia upon Hedgehog pathway activation (Endoh-Yamagami et al., 2009; Liem et al., 2009) and is necessary and sufficient to promote tip accumulation of the Hedgehog signaling factors Gli2 and Gli3 (He et al., 2014). Furthermore, KIF7 is a genetic modifier of BBS in human patients (Putoux et al., 2011). To follow the behavior of Kif7 and the BBSome in live IMCD3 cells, we stably coexpressed ^NG3BBS5 and Kif7 fused to the red fluorescent protein mScarlet. Smoothened activation led to the correlated coaccumulation of BBS5 and Kif7 at the tip of cilia (Fig. 5, A and B). Furthermore, in the rare instances where a second spot of Kif7 was found along cilia, possibly because part of the axoneme terminates before the tip, a similarly intense second spot of BBS5 was observed at the same location (Fig. 5 C). Because Kif7 depletion abolished tip accumulation of the BBSome (Fig. 5, D and E), these data suggest that Kif7 directly mediates the signal-dependent recruitment of BBSomes to the tip of cilia. Alternatively, it is conceivable that structural defects in cilia of Kif7-depeleted cells indirectly affect the recruitment of BBSomes to the tip. Congruent with a Kif7-mediated recruitment of BBSome to tips of cilia, Kif7 was found to coimmunoprecipitate with several BBSome subunits (Fig. S2 N). Given that dephosphorylation of Kif7 leads to the accumulation of Kif7 at the tip of cilia (Liu et al., 2014), we considered that phosphorylation by PKA may directly antagonize Kif7 tip accumulation and the Kif7–BBSome interaction. Concordantly, PKI led to the correlated coaccumulation of BBS5 and Kif7 at the tip of cilia to the same extent as SAG (Fig. 5, A and B), and elevated cAMP levels decreased the Kif7–BBSome interaction (Fig. 5 F). Finally, Kif7 was required for signal-dependent retrieval of SSTR3 (Fig. 5 G). Together, these results suggest that signaling downstream of Smoothened and SSTR3 leads to Kif7 dephosphorylation and that ensuing recruitment of BBSome to the tip initiates GPCR retrieval.

Analysis of ^NG3BBS5 and ^NG3IFT88 kymographs showed that anterograde BBSome trains, anterograde IFT trains, and retrograde IFT trains moved processively along the length of the cilium regardless of the signaling status (Figs. 6 A and S3, A and B). Yet in unstimulated cells, BBSome trains occasionally detached from retrograde IFT trains before reaching the base (Fig. 6 B), and overall, 90% of BBSome trains failed to reach the base of cilia (Fig. 6, A and C; and Fig. S3 B). Activation of Smoothened or SSTR3 or inhibition of PKA all doubled the number of BBSomes per retrograde train from 10 to 20 (Fig. 6 D) and led to a significant increase in the processivity of retrograde BBSome trains (Figs. 6 C and S3 B). Neither the frequencies nor the velocities of BBSome trains were affected by these treatments (Fig. 6, E and F). Meanwhile, SAG increased the number of IFT-B particles per retrograde train from 62 to 78 (Fig. 6 G). The addition of 106 BBSomes and 206 IFT-Bs to the tip upon SSTR3 activation (Figs. 3 C and S2 L) suggests that the signal-dependent accumulation of BBSomes and IFT-Bs at the tip drives the growth of retrograde trains by increasing the concentration of precursors at the site of assembly. Similar to the BBSome and IFT-B, Arl6 underwent signal-dependent tip accumulation (Fig. 6 H). As Arl6 was required for IFT88 tipping but not BBSome tipping (Fig. S2 L; Nager et al., 2017), Arl6 may recruit IFT-B particles to the tip by increasing the affinity of the BBSome for IFT-B.

PKI was sufficient to trigger tip accumulation of BBSome and Arl6 as well as the formation of large processive retrograde BBSome trains (Fig. 6, D and H; and Fig. S3 B), suggesting that recruitment of Arl6 and BBSomes to the tip may be sufficient to initiate the assembly of large processive retrograde IFT/BBSome trains. Because Arl6 and its candidate activator Ift27 were required for the signal-dependent formation of large processive BBSome trains (Fig. 6, C and D; and Fig. S3 C), we propose that BBSome coats polymerized upon Arl6-GTP binding become stably coupled to IFT-B trains to generate the large, signal-dependent, and processive retrograde BBSome/IFT trains.

Tip redistribution upon GPCR activation was not limited to the BBSome and IFT-B as GPR161 (Fig. 7, A and B) also underwent tip redistribution. Photobleaching the cilium exclusive of the tip revealed an enrichment of GPR161 at the tip upon Hedgehog pathway activation (Fig. 7, A and B). Recovery kinetics show that GPR161 at the tip became less dynamic after SAG treatment (Fig. 7 C). In support of an association between BBSomes and activated GPCRs at the tip of cilia, we found that WGA-mediated immobilization of membrane proteins increased the amount of BBSomes at the tip, most likely because BBSome–GPCR complexes were unable to leave the tip (Fig. 7 D). Furthermore, WGA treatment increased the amount of BBSomes throughout the length of cilia (Fig. 7 D), indicating that BBSome trains become trapped by immobilized GPCRs.

The bright and processive retrograde BBSome tracks observed upon sst addition frequently overlapped with faint tracks of SSTR3 (Fig. 7 E and Video 5). No comovement was observed between the faint BBSome trains and SSTR3 in the absence of Arl6 (Figs. 7 E and S3 D). These observations suggest that BBSome/Arl6 coats capture cargoes and move them from the tip to the base of cilia upon coupling to retrograde IFT trains.

In bulk fluorescence imaging, the few fluorescent GPCRs that were being trafficked tended to be obscured by the fluorescent GPCRs that remain inside cilia. We measured that the largest retrograde trains contained close to 50 BBSomes. Assuming a 1:1 stoichiometry between BBSomes and cargo, this indicates that the brightest retrograde cargo tracks can carry at most 50/3,129 = 1.6% of the total ciliary SSTR3. It is therefore expected that such tracks will be extremely faint by bulk imaging (Fig. 7 E). To overcome the limitations inherent to ensemble imaging, we set out to visualize the molecular events that underlie GPCR retrieval at single-molecule resolution. Combining site-specific biotinylation of AP-tagged GPCRs (Ye et al., 2013) with Streptavidin (SA)-coupled quantum dots (Qdots [also referred to as QD]; ^SAQdot) enabled imaging at 2 Hz for >20 min (compared with 30-60 s when using mSA647; Fig. S3 E). By blocking most surface-exposed biotin groups with mSA before labeling with ^SAQdot, the vast majority of cilia bore no ^SAQdot, and cilia bearing one ^SAQdot were expected to possess a single biotinylated ^APGPCR (Fig. 8 A).

Tracking of ^SAQdot-labeled GPCRs (^QDGPCRs) demonstrated that their dynamics were unperturbed by ^SAQdot labeling, consistent with a single valence of ^SAQdot per GPCR. First, consistent with prior single-molecule tracking studies (Ye et al., 2013; Milenkovic et al., 2015), ^QDSSTR3 displayed a diffusive behavior in the absence of agonist (Figs. 8 D and S4 A; Video 6), and instant velocities of diffusing GPCRs were nearly identical when labeled by mSA647 or ^SAQD655 (Fig. 8 B). Second, congruency of exit rates between bulk imaging and ^QDSSTR3 imaging indicated that ^SAQdot labeling did not impair exit (Fig. 8 C). Mapping the centroid of the Qdot enabled a 7-nm measurement precision (Fig. S3 F) that resolved the lateral displacement of ^QDSSTR3 around the 250-nm diameter of the ciliary membrane during diffusive events (Fig. 8 E, i). Addition of sst led to ^QDSSTR3 undergoing frequent confinements at the base and tip as well as processive retrograde movements (Fig. 8, D and E, ii–iv; Fig. S4 B; and Video 7). Although we occasionally observed apparent directional movement during brief periods of time for ^QDSSTR3 in control-treated cells (Figs. 8 D and S4 A), persistent retrograde movements of ^QDSSTR3 lasting ≥6 s were strictly dependent on SSTR3 activation (Figs. 8 F and S3 G), confirming that signaling drives the formation of cargo-laden IFT/BBSome retrograde trains. This signal-dependent increase in long processive movements was not observed for anterograde movements (Fig. 8 F). Processive transport events did not exhibit lateral displacement (Fig. 8 E, ii–iv), strongly suggesting that each cargo-laden IFT/BBSome train moves along a single axonemal microtubule. GPCRs occasionally resided for several seconds at either the tip or the base, and the frequency of confinement events, defined as residence >15 s, significantly increased in the presence of sst (Figs. 8 G and S3 H). Because processive retrograde transport events were frequently preceded by tip confinement (Fig. 8 E, ii and iv), it is likely that tip confinement of GPCRs reflects a step of cargo capture by BBSome coats assembling at the tip. Nevertheless, ^QDSSTR3 exhibited retrograde movements starting at any point along the cilium (Fig. 8 E, iii; and Fig. S4 B), indicative of activated GPCRs hopping onto retrograde BBSome trains.

Similar to ^QDSSTR3, ^QDGPR161 underwent mostly diffusive behavior in unstimulated cells, and addition of SAG led to frequent retrograde processive transport events (Fig. S4, C and D). Consistent with IFT/BBSome trains powering the retrograde transport of activated GPCRs, long processive retrograde movements of ^QDGPR161 were absent when Arl6 was deleted (Fig. S4 E). The mean velocity of the processive retrograde movements of ^QDSSTR3 was similar to the velocities of IFT and BBSome trains (Fig. 6 F), suggesting that signaling promotes coupling between cargoes and retrograde IFT trains. In support of this hypothesis, coimaging of ^NG3BBS5 and ^QDSSTR3 uncovered instances of comovement between the retrograde BBSome train and ^QDSSTR3 (Fig. 8 H). We conclude that activated GPR161 and SSTR3 are transported in the retrograde direction by the large processive BBSome trains that couple to retrograde IFT trains upon GPCR activation.

The expression of ^APGPCRs at near-endogenous levels enabled the ^SAQdot-mediated visualization of exit from the ciliary compartment at single-molecule resolution. In combined 21 h of single-molecule imaging, three exit events of ^QDGPR161 were observed. In all three events, ^QDGPR161 exit followed a stereotypical sequence of processive retrograde transport, base confinement, and diffusion away from the ciliary compartment (Fig. 9, A and B, iii). Surprisingly, many ^QDGPCRs that underwent retrograde transport and base confinement returned into the shaft of the cilium seemingly by processive anterograde transport (Fig. 9, A and B, i and ii; and Fig. 8, D and E, iii). During two out of the three preexit confinement events we observed, ^QDGPR161 first diffused within an area of <360-nm diameter before near-complete immobilization before exit (Figs. 9 C and S5 A; Videos 8 and 9). We did not observe a clear correlation between successful exit and either base residence time (Fig. 9 D) or area explored before exit (Fig. 9 E), suggesting that completion of exit is a stochastic process. Because ^QDGPR161 diffused within the 0.4-µm-deep focal plane after exit from the ciliary compartment, it is most likely that the GPCR moved into the plasma membrane after exit (Figs. 9 C and S5 A; Videos 8 and 9). Exit events were observed at the expected frequency based on measurements of bulk exit rates (Fig. S5 B), and this confirms that ^SAQdot labeling did not interfere with exit.

Unexpectedly, the position of ^QDGPR161 during signal-dependent base confinement events was often separate from the bulk fluorescence of GPR161^NG3 (Fig. 10 A). Using the profile of the NG channel as a common reference (Fig. 10, A–C; and Fig. S5 C), we mapped the location of the most base-proximal position explored by ^QDGPR161 relative to the transition zone marker Cep290 and the transition fiber marker Cep164 and found that activated ^QDGPR161 moved into the 100-nm space between these two markers (Fig. 10, A–C). We termed the region between the 10th and 50th percentile of GPR161^NG3 fluorescence the intermediate compartment as it defined the area visited by ^QDGPR161 before exit was completed. Systematic analysis of ^QDGPR161 position demonstrated that GPR161 enters the intermediate compartment in a signal- and Arl6-dependent manner (Fig. 10 C), thereby demonstrating that large, processive BBSome trains ferry GPR161 through the transition zone and deliver it to the intermediate compartment.

Tracking ^QDGPR161 over a combined 70-min period of imaging in the presence of SAG revealed that the probability of ^QDGPR161 entering the intermediate compartment is 0.66 during a 1-min interval (Fig. S5 D). Meanwhile, the absolute exit rate (Fig. 1 G) of 0.0043 molecules/min is equivalent to the probability of a single molecule of GPR161 experiencing exit during a 1-min interval. Thus, a comparison of intermediate compartment entry frequency and ciliary exit rates revealed that <1% of intermediate compartment visits productively lead to exit. Meanwhile, 99.3% of intermediate compartment visits are resolved by the GPCR returning into the ciliary compartment (Fig. 9, A and B, i and ii). Thus, although crossing the first diffusion barrier at the transition zone appears to be a prerequisite for ciliary exit, it is not sufficient to commit GPCRs for exit because a periciliary barrier blocks lateral diffusion between the intermediate compartment and the plasma membrane (Fig. 10 F).

Examining the lateral displacement of ^QDGPR161 to the center of the axoneme when ^QDGPR161 reached its most base-proximal position in unstimulated cells confirmed that GPR161 stayed within the expected 250-nm diameter of the ciliary membrane during rapid turnaround events in unstimulated cells (Fig. 10, D and E). The mean and maximal lateral displacements of ^QDGPR161 doubled when ^QDGPR161 reached its most base-proximal position in SAG-treated cells (Fig. 10, D and E). Because ^QDGPR161 visits its most base-proximal location in SAG-treated cells while residing within the intermediate compartment, these data indicate that the diameter of the intermediate compartment is close to 550 nm, similar to the 450-nm diameter defined by the tip of the transition fibers (Lau et al., 2012; Yang et al., 2015, 2017; Kanie et al., 2017). The periciliary barrier thus appears to be located near the point where the transition fibers attach to the ciliary membrane.

Our pharmacological and live-cell imaging manipulations indicate that downstream of SSTR3 and Smoothened, a Gαi-mediated decrease in PKA activity promotes BBSome tip redistribution and subsequent retrieval of SSTR3 and GPR161. Intriguingly, the Gαs-coupled GPCR dopamine receptor 1 (DRD1) becomes enriched in cilia of amygdala neurons when BBSome function is compromised (Domire et al., 2011). It will be important for future studies to determine how Gα-coupled GPCRs are retrieved from cilia and whether BBSome tipping is induced by DRD1 activation.

We note that in olfactory receptor neurons, BBS4 is present in all IFT88-marked trains (Williams et al., 2014). In contrast, IFT tracks in unstimulated IMCD3 cells are often devoid of BBSomes, and we observed frequent uncoupling of BBSomes from IFT trains (Fig. 6 B). This suggests that olfactory receptor neurons resemble IMCD3 cells under signaling conditions with respect to BBSome transport. In this context, it is notable that BBS2 and BBS4 accumulate at the tip of olfactory cilia (Williams et al., 2014), and we propose that BBSome transport is highly active in olfactory receptor neurons. According to our absolute quantitation, an average retrograde train contains 10 BBSomes and 62 IFT-B complexes in untreated cells (Fig. 6, D and G). The BBSome/IFT stoichiometry (6.2:1) we measured by quantitative imaging is in close concordance with that measured using quantitative mass spectrometry in Chlamydomonas (6.5:1; Lechtreck et al., 2009), thus suggesting a low basal rate of BBSome-mediated transport in Chlamydomonas under vegetative conditions.

We found that base confinement events are prerequisites to exit. In contrast to GPR161 and SSTR3 whose activation increased their base confinement frequency, Smoothened activation decreases its base confinement frequency (Milenkovic et al., 2015). As Smoothened accumulates in cilia of unstimulated Bbs mutant cells (Zhang et al., 2011; Eguether et al., 2014; Goetz et al., 2017), this suggests that Smoothened is constitutively retrieved from cilia by the BBSome and that a reduction in Smoothened retrieval underlies the signal-dependent accumulation of Smoothened in cilia.

Although the idea of a periciliary diffusion barrier was initially considered (Nachury et al., 2010), the transition zone has come to be viewed as the sole diffusion barrier of the cilium in recent years (Garcia-Gonzalo and Reiter, 2012; Gonçalves and Pelletier, 2017; Jensen and Leroux, 2017). Our finding that the ciliary membrane is individualized from the plasma membrane by two successive diffusion barriers suggests the existence of an intermediate compartment located between the transition zone and a nearly impassable periciliary barrier. Crossing of the second barrier is extremely infrequent and is often preceded by a near-complete immobilization before exit. In all the exit events we imaged, ^QDGPR161 diffused within the plane of imaging after exit, suggesting that the ^QDGPR161 stayed in the plasma membrane. However, we cannot rule out that this mobility reflects that of an endosome or that endocytosis mediates crossing of the second barrier.

The periciliary barrier is likely to correspond with the recently described distal appendage matrix (DAM) because depletion of the DAM component FBF1 results in the leakage of GPCRs from cilia (Yang et al., 2017). Remarkably, when the receptors for insulin-like growth factor 1 (IGF1) or for transforming growth factor β (TGF-β) undergo signal-dependent exit from cilia, they transiently localize to a zone at the base of cilia that does not overlap with axonemal markers and may correspond to the intermediate compartment. Furthermore, residence of activated IGF1R and TGF-βR at the base of cilia appears to organize downstream signaling for these two pathways (Clement et al., 2013; Yeh et al., 2013). The intermediate compartment may harbor specific lipids as both PI(4,5)P2 and PI(3,4,5)P3 are dynamically enriched at a zone at the base of cilia that is clearly nonoverlapping from axonemal markers in mammalian cells (Dyson et al., 2017). The intermediate compartment may therefore constitute a privileged signaling locale. Finally, the existence of a second barrier explains why transition zone mutants have only mild mislocalization phenotypes and can still assemble cilia.

The ultrastructural location of the diffusion barrier within the transition zone is beyond the resolution of our imaging study. However, considering that ^QDGPR161 explores a compartment that is ∼220 nm long during base confinement events (Fig. 9 C), and given that the distance between the tip of the distal appendages (marked by Cep164) and the proximal end of the transition zone (marked by Cep290) is ∼100 nm (Yang et al., 2015), it is likely that the intermediate compartment encompasses part of the transition zone and that the diffusion barrier is located within the most distal part of the transition zone (Fig. 10 F).

Furthermore, because activated GPR161 only crosses the transition zone when BBSome/Arl6 coat assembly is permitted, the hypothesis that Hedgehog signaling loosens the diffusion barrier of the transition zone (Dyson et al., 2017) cannot account for GPR161 exit. Instead, our data suggest that BBSome/Arl6 coats bound to retrograde IFT trains on the axoneme-facing side and to cargoes on the membrane-facing side facilitate lateral transport through the transition zone. Thus, in contrast to all other known diffusion barriers, the transition zone is a porous barrier that allows the selective permeation of GPCRs bound to BBSome/Arl6 coats. Physical and genetic interactions between the BBSome and transition zone proteins such as Cep290 (Zhang et al., 2014; Barbelanne et al., 2015; Yee et al., 2015; Goetz et al., 2017) are in agreement with a general model where BBSome/Arl6 coats contact the transition zone during crossing. Taking our data into account, we propose that Arl6-GTP increases the coupling between cargoes, BBSomes, and IFT trains to facilitate lateral transport through the transition zone. However, the intimate details of transition zone crossing remain to be determined. In particular, the lack of precedent for selective permeation though a membrane diffusion barrier points to distinguishing biophysical features of the transition zone whose definition promises to enrich the concepts underlying diffusion barriers.

For all experiments, a mouse IMCD3-FlpIn cell line was used (gift from P.K. Jackson, Stanford University, Stanford, CA). IMCD3-FlpIn cells were cultured in DMEM/F-12 (11330-057; Gibco) supplemented with 5% FBS, 100 U/ml penicillin-streptomycin, and 2 mM l-glutamine.

Cell lines expressing SSTR3, GPR161, BBS1, BBS5, and IFT88 were generated using the FlpIn System (Thermo Fisher Scientific). Plasmids with multiple expression cassettes driven by low-expression promoters were built by modification of the pEF5B/FRT plasmid. The second cassette was inserted into an NsiI site that had been introduced in the pEF5B/FRT plasmid between the ampicillin cassette and an EF1α promoter. Subsequent cloning into the NsiI site was used to incorporate NsiI-pPGK-BirA-ER or NsiI-pEF1αΔ-^APSSTR3-IRES-BirA-ER followed by a herpes simplex virus polyadenylation signal (Nager et al., 2017). Coding sequences were amplified from plasmids encoding human BBS1 and BBS5 (gifts from V. Sheffield, University of Iowa, Iowa City, IA), BirA-ER (a gift from A. Ting, Stanford University, Stanford, CA; 20856; Addgene; Howarth and Ting, 2008), mouse GPR161 (BC028163; Mammalian Gene Collection [MGC]; GE Healthcare), human NPY2R (BC075052; MGC; GE Healthcare), mouse IFT88 (IOM20300; UltimateORF; Invitrogen), human MCHR1 (BC001736; MGC; GE Healthcare), PACT (pericentrin and AKAP450 centrosome-targeting domain; gift from S. Munro, Medical Research Council, Cambridge, UK; Gillingham and Munro, 2000), mouse Kif7 (gift from K. Anderson, Memorial Sloan Kettering Cancer Center, New York, NY; He et al., 2014), mouse SSTR3 (gift from K. Mykytyn, Ohio State University, Columbus, OH), and IFT43. BBS1, BBS5, and IFT88 were expressed by the EF1α promoter, SSTR3, NPY2R, and MCHR1 were expressed by the EF1αΔ promoter, and GPR161 was expressed by the Crys promoter. GFP, NG (Shaner et al., 2013), mScarlet (gift from D. Gadella, University of Amsterdam, Amsterdam, Netherlands; 85042; Addgene; Bindels et al., 2017), tdTomato (gift from M. Davidson, Florida State University, Tallahassee, FL; 54653; Addgene; Shaner et al., 2004), and an acceptor peptide (GLNDIFEAQKIEWHE) for the biotin ligase BirA (AP) were used in fusion proteins. Kif7 cDNA was subcloned into a modified pCMV-based plasmid (pmScarlet-C; 85042; Addgene) that was transfected into IMCD3 cells using Lipofectamine 2000, and clones were selected using neomycin resistance. The expression level of ^NG3BBS5 and ^NG3IFT88 relative to endogenous BBS5 and IFT88 were determined by Western blotting (Figs. 3 A and S2 I), and the expression levels of exogenous SSTR3 or GPR161 compared with endogenous GPCRs were determined by measuring the intensity of ciliary fluorescence after immunostaining (Figs. 1 C and S1 B).

CRISPR-based genome editing was conducted by transiently expressing Cas9 and guide RNA from plasmids derivative of pX330 (gift from F. Zhang, Massachusetts Institute of Technology, Cambridge, MA; 42230; Addgene; Cong et al., 2013). The guide sequences were: 5′-AAGCCGCGATATGGGCTTGC-3′ for Arl6 (Liew et al., 2014), 5′-GGAAATGGGTCCCGTCGCTG-3′ for Ift27 (Liew et al., 2014), 5′-ACTCACCCACGGGGTCCACG-3′ for Arrb1 (Nager et al., 2017), 5′-TCTAGGCAAACTTACCCACA-3′ for Arrb2 (Nager et al., 2017), and 5′-ACGTCGCTGCGAGGCATCTG-3′ for Tulp3 (this study).

CRISPR-modified clones were isolated by limited dilution and verified by Western blotting for the disrupted gene product. To confirm gene editing, the modified genomic locus was isolated using the CloneJET PCR cloning kit (K1231; Thermo Fisher Scientific). The genotypes were Arl6, NM_001347244.1:c.-10_25del; c.3_6del; Ift27, the genotype was not characterized; Tulp3, NM_011657.2:c.142_143del; c.142_143insA; Arrb1, NM_177231.2:c.108del;c.108del; and Arrb2, NM_001271358.1:c.112del;c112_113del.

To build constructs with low-expression promoters, an NsiI restriction cloning site was inserted by site-directed mutagenesis before the EF1α promoter in pEF5B/FRT (Nager et al., 2017). The EF1α promoter was then excised by NsiI and SpeI and replaced by either the ubiquitin C promoter (UbC), thymidine kinase (TK), EF1αΔ, CMVΔ, or a minimal chicken lens δ-crystallin promoter (Kamachi and Kondoh, 1993; Ferreira et al., 2011; Morita et al., 2012). The UbC promoter was cloned from pLenti6/UbC/V5-Dest (V49910; Thermo Fisher Scientific). The TK promoter was from pRL TK (E2241; Promega). EF1αΔ consists of a TATA-less EF1α promoter from pEF5/FRT/V5-Dest (V602020; Thermo Fisher Scientific), wherein the TATA box sequence TATAA was mutated to TCCCC. CMVΔ was designed after the CMV(Δ6) promoter (Morita et al., 2012) and was cloned from pcDNA3.1 (V79020; Thermo Fisher Scientific). The chicken lens δ-crystallin promoter was cloned from a pGL3-8×Gli–firefly luciferase plasmid (gift from P. Beachy, Stanford University, Stanford, CA).

Rat hippocampal neurons were dissected from P0 or P1 rat pups (gifts from M. Lin, Stanford University, Stanford, CA) and plated on poly-d-lysine–coated 12-mm #0 cover glass (633009; Carolina Biological Supply). Neurons were cultured in neurobasal medium with serum-free B27 (21103049; Thermo Fisher Scientific) and Gibco GlutaMAX (35050061; Thermo Fisher Scientific). Neurons were identified by nuclear NeuN staining in immunofluorescence experiments.

Plasmids were reverse-transfected by Lipofectamine 2000 (11668027; Thermo Fisher Scientific). In brief, detached cells were plated with the transfection reagent and plasmid in OptiMEM (31985070; Thermo Fisher Scientific). The transfection reagent was replaced by fresh DMEM/F-12 medium after 4-h siRNAs were reverse-transfected by Lipofectamine RNAiMAX (13778030; Thermo Fisher Scientific). In brief, detached cells were plated with the transfection reagent and the siRNA duplex in OptiMEM. Cells were then grown for 48 h before 24 h starvation and subsequent imaging. Control (D-001210-04-05), BBS1 (D-019180-03), BBS2 (D-010080-02), and BBS4 (D-054691-03) were from GE Healthcare, and the Kif7 siRNA (GS16576) and matched control (1027280) were from QIAGEN. siRNA targeting Arl6 (5′-CTTTAGGACTTGAGACATT-3′) was previously validated (Jin et al., 2010).

Small molecules were added at the following concentrations and for the indicated pretreatment times unless otherwise indicated: ACQ090 (20 µM; Novartis; Bänziger et al., 2003), 3-isobutyl-1-methylxanthine (IBMX; 500 µM; 30 min preincubation; I5879; Sigma-Aldrich), L796,778 (10 µM; Merck; Rohrer et al., 1998), PKI (50 µM; 40 min preincubation; BML-P210-0500; Enzo Life Sciences), PTX (10 ng/ml; 16 h preincubation; 180; List Biological Laboratories), Rp-cAMPS (10 µM; 2 h preincubation; sc-24010; Santa Cruz Biotechnology, Inc.), SAG (200 nM; ALX-270-426-M001; Enzo Life Sciences), sst (10 µM; ASR-003; Alomone Labs), and SQ22536 (500 µM; 40 min preincubation; S153; Sigma-Aldrich).

The following antibodies were used: AC3 (rabbit; sc-588; Santa Cruz Biotechnology, Inc.), acetylated tubulin (mouse; clone 6-11B-1; Sigma-Aldrich), Arl6 (rabbit; Jin et al., 2010), β-arrestin 1 (rabbit; 15361-1-AP; ProteinTech), β-arrestin 2 (rabbit; 10171-1-AP; ProteinTech), BBS5 (rabbit; 14569-1-AP; ProteinTech; used for immunoblotting), BBS5 (mouse monoclonal; used for immunoprecipitation; gift from C. Smith, University of Florida, Gainesville, FL; Smith et al., 2013), BBS9 (rabbit; HPA021289; Sigma-Aldrich), centrin (mouse; clone 20H5; EMD Millipore), Cep164 (rabbit; gift from T. Stearns, Stanford University, Stanford, CA), Cep290 (rabbit; gift from S. Saunier, Institut Imagine, Paris, France), c-Myc (mouse; sc-40; Santa Cruz Biotechnology, Inc.), GFP (rabbit; A11122; Invitrogen), GPR161 (rabbit; gift from S. Mukhopadhyay, University of Texas Southwestern, Dallas, TX), IFT139 (rabbit; gift from P. Tran, University of Kansas, Kansas City, KS; Tran et al., 2008), IFT140 (rabbit; 17460-1-AP; ProteinTech), NeuN (mouse; MAB377; EMD Millipore), ninein (rabbit; gift from M. Bornens, Institut Curie, Paris, France; Mogensen et al., 2000), phospho-p44/42 MAPK Erk1/2 Thr202/Tyr204 (rabbit; 4370; Cell Signaling Technologies), SSTR3 (goat; sc-11617; Santa Cruz Biotechnology, Inc.), and TULP3 (rabbit; gift from J. Eggenschwiler, University of Georgia, Athens, GA; Norman et al., 2009). Biotinylated SSTR3 and GPR161 were detected using Alexa Fluor 647–labeled mSA (Ye et al., 2013).

The IFT-A complex was purified from IMCD3 cells stably expressing IFT43 with an N-terminal LAP tag. 3 × 108 cells were harvested and lysed in 6 ml LAP200N buffer (50 mM Hepes, pH 7.4, 200 mM KCl, 1 mM EGTA, 1 mM MgCl₂, 10% glycerol, and 0.05% NP-40) with protease inhibitors. The IFT-A complex was captured on 100 µl Affiprep protein A beads (Bio-Rad Laboratories) coupled to 100 µg GFP antibody, and complexes were eluted in 250 µl LAP200N containing 3 µg TEV protease (Nachury, 2008). The BBSome was purified to near-homogeneity from bovine retinal extract by affinity chromatography on GST-Arl6^GTP followed by MonoS (Jin et al., 2010). GST and GST-SSTRⁱ³ fusion proteins were purified from Escherichia coli and used for capture of pure BBSome or pure IFT-A samples (Jin et al., 2010).

The interaction of ^GFPKif7 with ^MycBBS1 was assayed by cotransfection/coimmunoprecipitation. HEK293 cells were forward transfected in a 6-cm plate with plasmids expressing ^GFPKif7 and either ^MycRFP or a Myc-tagged BBSome subunit. After 2 d, cells were washed and treated with media containing either IBMX or DMSO for 30 min. Cells were then trypsinized, pelleted, and cleaned in a flacon tube and then lysed for 10 min with cold coimmunoprecipitation buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100, 1 mM DTT, 1 mM 4-(2-aminoethyl)benzenesulfonyl fluoride, 800 nM aprotonin, 15 µM E-64, 10 mg/ml leupeptin, 10 mg/ml pepstatin A, and 10 mg/ml bestatin). The resulting lysates were then centrifuged for 15 min, concentration matched, and then added to anti-GFP antibody–coupled beads in coimmunoprecipitation buffer. The beads were rotated for 20 min at 4°C, pelleted, washed with coimmunoprecipitation buffer four times, and eluted by boiling in SDS-PAGE loading buffer.

All imaging was conducted on an DeltaVision Elite (GE Healthcare) equipped with a Plan Apochromat 60× 1.40 NA oil objective lens and a Plan Apochromat 60× 1.49 NA total internal reflection microscopy (TIRF) oil objective lens (Olympus) and a 488-nm laser from DeltaVision Quantifiable Laser Module, and images were captured with a pco.edge sCMOS camera with near-perfect linearity across its 15-bit dynamic range. The pixel size of the sCMOS camera was 0.1077 µm. The entire microscope was housed within a temperature-controlled environmental chamber. All image acquisition was done in SoftWoRx (6.0; GE Healthcare). Files were imported into ImageJ (National Institutes of Health) and analyzed as described in the following sections. No gamma adjustment was applied during figure preparation.

In a 24-well plate, 50,000–100,000 cells were seeded on acid-washed cover glass (12 mm #1.5; 12-545-81; Thermo Fisher Scientific). Cells were grown for 24 h and then starved for 16–24 h in 0.2% FBS media before experimental treatment. After treatment, cells were fixed with room-temperature 4% paraformaldehyde in PBS for 5 min, extracted in −20°C methanol for 5 min, permeabilized in PBS containing 0.1% Triton X-100, 5% normal donkey serum (017-000-121; Jackson ImmunoResearch Laboratories, Inc.), and 3% bovine serum albumin (BP1605-100; Thermo Fisher Scientific) for 30 min, and subsequently immunolabeled for imaging. In brief, cells were incubated with primary antibodies for 1 h, washed three times with PBS, incubated with dye-coupled secondary antibodies (Jackson ImmunoResearch Laboratories, Inc.) for 30 min, washed two times with PBS, stained with Hoechst DNA dye, washed twice more with PBS, and mounted on slides using fluoromount-G (17984-25; Electron Microscopy Sciences). For nearly every experiment, cilia closest to the coverslip were imaged (ventral cilia) as these cilia often lay perpendicular to the objective and within a single focal plane. In a few select cases (Figs. 3 F and 6 H), cilia pointing away from the coverslip (dorsal cilia) were imaged to reduce background fluorescence. To do so, a z stack of images with 0.3-µm separation was collected and deconvolved using SoftWoRx 6.0 based on iterative-constrained algorithms and subsequently projected by maximum-intensity projections.

400,000 cells were seeded on acid-washed 25-mm cover glass (72223; Electron Microscopy Sciences) in a 35-mm dish. After 24 h of growth, cells were starved for 16 h and transferred to the DeltaVision stage for imaging at 37°C. Cells were imaged in DMEM/F-12 media with Hepes, no phenol red, and 0.2% FBS (11039-021; Gibco). For all >1-h imaging experiments, the imaging chamber was overlaid with a Petri dish containing a moist towel to maintain the imaging volume and pH.

In Fig. 1 C, the number of ^APSSTR3^GFP molecules per cilium resulting from expression driven by the EF1α, UbC, TK, EF1αΔ, and CMVΔ promoters was estimated by comparison with a viral particle containing exactly 120 GFP molecules (Breslow et al., 2013). As ciliary GFP was not detectable by direct imaging of the IMCD3–pCrys-^APSSTR3^GFP line, immunostaining for GFP was used to amplify the fluorescent signal and compare with immunostained GFP in IMCD3-[pEF1αΔ-SSTR3^GFP] cells. To estimate the number of SSTR3 molecules per hippocampal neuron, hippocampal neurons and IMCD3-[^APSSTR3^GFP] cells using the aforementioned promoters were compared by SSTR3 immunostaining. The epitope recognized by the SSTR3 antibody (sc-11617; Santa Cruz Biotechnology, Inc.) is identical between rat and mouse (Santa Cruz Biotechnology, Inc.; personal communication with K. Griffin).

For quantitation of NG-tagged proteins including ^APSSTR3^NG, ^APGPR161^NG3, ^NG3BBS5, and ^NG3IFT88, ciliary molecules were quantified using single NG trimers (NG3) as a calibrator (Prevo et al., 2015). NG3 protein was recombinantly expressed in E. coli and purified by nickel affinity. The purified protein was then sparsely immobilized on a 18 × 18–mm coverslip for imaging on a DeltaVision microscope. Imaging was done in live-cell imaging solution (A14291DJ; Invitrogen). To confirm that fluorescent foci originate from a single NG3, we confirmed that fluorescence was lost by photobleaching in three discrete steps (Figs. 1 E and S1 H). The mean fluorescent intensity from >1,257 molecules was then used to estimate the number of NG-tagged molecules in cilia.

To quantify the absolute number of ciliary proteins, serum-starved live IMCD3 cells were imaged in live-cell imaging solution with the same exposure setting as used in single NG3 fluorescent quantitation. Therefore, N_cilia = (F_{cilia_NG} − F_background)/((F_NG3/n) * R_expression) where N_cilia is the absolute number of ciliary protein, F_{cilia_NG} is the total ciliary fluorescence detected from NG- or NG3-labeled proteins, F_background is the background fluorescence measured in the adjacent area, F_NG3 is the fluorescent intensity of a single NG3 protein, and n determined by the single or triple NG tag was used. For example, n = 3 for ^APSSTR3^NG, and n = 1 for GPR161^NG3. R_expression is the abundance ratio between the NG-tagged form of a given protein and the total amount of that protein in cilia. For pEF1α-^NG3BBS5 and pEF1α-^NG3IFT88, R_expression = 0.55 and 0.51, respectively, which were measured by Western blotting (Figs. 3 A and S2 I). For pEF1αΔ-^APSSTR3^NG, R_expression = 1 because IMCD3 cells do not express SSTR3. For ^APGPR161^NG3, R_expression could not be directly measured because tagging of GPR161 at the C terminus (e.g., in GPR161^NG3) interferes with recognition by the anti-GPR161 antibody developed by Mukhopadhyay et al. (2013). We thus used ^APGPR161 as an intermediate calibrator between ^APGPR161^NG3 and endogenous GPR161. A plasmid encoding pCrys-^APGPR161 was transfected into IMCD3 cells, and the relative expression levels of transiently expressed ^APGPR161 and stably expressed ^APGPR161^NG3 were measured by mSA647 pulse-labeling. We used the equation R_{AP.GPR161.NG3:AP.GPR161} = (F_{mSA.Gpr161.NG3})/(F_mSA.GPR161) where R_{AP.GPR161.NG3:AP.GPR161} is the ratio between stably expressed ^APGPR161^NG3 and transiently expressed ^APGPR161 in cilia, F_mSA.GPR161 is the ciliary fluorescence signal measured by mSA647 labeling of transiently expressed ^APGPR161, and F_{mSA.Gpr161.NG3} is the ciliary fluorescence signal measured by mSA647 labeling of stably expressed ^APGPR161.

The relative ciliary expression levels of endogenous GPR161 and ^APGPR161 were measured by anti-GPR161 antibody. We used the equation R_{endoGPR161:AP.GPR161} = F_Ab/(F_Ab + ^APGPR161 − F_Ab) where R_{endoGPR161:AP.GPR161} is the ratio between endogenous GPR161 and transiently expressed ^APGPR161 in cilia, F_Ab is the ciliary fluorescence signal measured by immunofluorescence of untransfected cells with the anti-GPR161 antibody, and F_Ab + ^APGPR161 is the ciliary fluorescence signal measured by immunofluorescence of cells transiently expressing with ^APGPR161 with the anti-GPR161 antibody.

Therefore, N_{AP.GPR161.NG3} = (F_cilia.NG − F_background)/F_NG3; R_{endoGPR161:AP.GPR161.NG3} = R_{endoGPR161:AP.GPR161}/R_{AP.GPR161.NG3:AP.GPR161}; N_endoGPR161 = N_{AP.GPR161.NG3}/R_{endoGPR161:AP.GPR161.NG3}; and R_expression = N_{AP.GPR161.NG3}/(N_{AP.GPR161.NG3} + N_endoGPR161), where R_{endoGPR161:AP.GPR161.NG3} is the ciliary abundance ratio between endogenous GPR161 and stably expressed ^APGPR161^NG3; N_endoGPR161 and N_{AP.GPR161.NG3} are the absolute numbers of endogenous GPR161 and stably expressed ^APGPR161^NG3 in cilia; F_{cilia_NG} is the total ciliary fluorescence detected in the green channel from ^APGPR161^NG3; F_background is the background fluorescence measured in the adjacent area; F_NG3 is the fluorescent intensity of a single NG3 protein; and R_expression is the ciliary abundance ratio between stably expressed ^APGPR161^NG3 and total GPR161. Using this strategy, we determined that N_{AP.GPR161.NG3} = 1,226, N_endoGPR161 = 949, and R_expression = 0.56.

To measure SSTR3 exit, cells expressing ^APSSTR3^NG were first washed three times with PBS containing 5 mM MgCl₂ (PBS-Mg) and then pulse-labeled with Alexa Fluor 647–labeled mSA (mSA647) for 5–10 min. To remove the unbound mSA647, cells were washed three times with PBS containing 5 mM MgCl₂. Cells were then imaged after addition of sst (Nager et al., 2017). For each time point, the integrated Alexa Fluor 647 fluorescence density was measured using ImageJ. The cilia-adjacent fluorescence was subtracted as the background, and a mathematical photobleaching correction was applied: F_cilia = (F_{mSA647_measured}/F_{mSA647_1}) + ((1 – e-λ) * (n − 1)), where λ is the photobleaching decay constant, n is the number of images taken, F_{mSA647_measured} is the integrated mSA647 fluorescence measured for image n, F_{mSA657_1} is the measurement for the first time point, and F_cilia is the reported fluorescence. In this equation, F_cilia is reported in relative fluorescence units (RFUs).

^APGPR161^NG3 was assayed similarly except that quantitations were done by NG fluorescence intensity using the following equation: GPR161_cilia = (F_{NG3_n}/F_{NG3_1}) * (µF_{_NG3_xilia} * µF_{_NG3}), where GPR161_cilia is the number of GPR161 molecules in the cilium, F_{NG3_n} is the photobleaching-corrected integrated NG fluorescence measured for image n, F_{NG3_1} is the measurement for the first time point, µF_{_NG3_cilia} is the mean integrated NG fluorescence from a population of ^APGPR161^NG3 cilia, and µF_{_NG3} is the mean integrated green fluorescence from individual NG3 molecules.

To measure the exit rate of SSTR3 and GPR161 in the first 2 h after adding agonist (Fig. 1, F and G), R_exit/h = N_cilia * (F_{cilia_1} − F_{cilia_2h})/(2 * F_{cilia_1}), where R_exit/h is the number of SSTR3 or GPR161 that exited cilia per hour, F_{cilia_1} is the background-corrected ciliary fluorescent intensity before adding agonist, F_{cilia_2h} is the background-corrected ciliary fluorescent intensity after 2 h of agonist treatment, and N_cilia is the absolute number of SSTR3 or GPR161 molecules in cilia as described in previous section.

To assess significant differences in GPCR removal, we used multiple regression. First, raw data were linearly fitted (F_cilia = m * time + c). Conditions were then compared using a z statistic: z = (m₁ – m₂)/(se_1,2) where m₁ and m₂ are the fitted slopes for two experiments, and se_1,2 is the propagated standard error of the slopes: se_1,2 = √(s_m1² + s_m2²) where s_m1 and s_m2 are the SDs for m₁ and m₂, respectively. Z statistics were converted to p-values for statistical interpretation.

To analyze intraciliary trafficking, IFT88 and BBS5 proteins were genetically labeled with NG3 and rapidly imaged with TIRF (4 Hz) for short time periods (30–60 s). TIRF illumination reduced nonciliary (background) fluorescence, permitting visualization of dim trains. The resulting videos were analyzed by ImageJ for generating kymographs. KymographClear and KymographDirect (Mangeol et al., 2016) were used to deconvolve anterograde from retrograde trains for measuring relative intensities for BBSome trains, whereas Multi Kymograph was used for unaltered presentation (e.g., Fig. 6 A).

The processivity of IFT-B or BBSome trains along axonemal microtubules was defined as the duration for which a fluorescent focus of IFT88 or BBS5 unidirectionally moved along the cilium at a rate expected for kinesin- or dynein-mediated transport (0.3–0.6 µm/s). Only events occurring for >3 s were considered. Measurements were made from kymographs of ^NG3IFT88, ^NG3BBS5, or ^tdTomatoIFT88 and ^NG3BBS5. In brief, kymographs were visualized in ImageJ, and lines were drawn along the long axis of the cilium (“the leg”) and the processive movement (“the hypotenuse”). The angle between the leg and the hypotenuse was measured (“the included angle”). Comparison of the leg and the hypotenuse was used to quantify the processivity of an IFT- or BBS-containing train. Comparison of the hypotenuse and the included angle was used to measure the velocity of IFT and BBS trains.

To estimate the number of IFT88 or BBS5 molecules per train, the total NG3 fluorescence along the ciliary axoneme was first divided by the number of trains: T_{RFU_mean} = µ_{RFU_NG3_axoneme}/N_train, where T_{RFU_mean} is the estimated ^NG3IFT88 or ^NG3BBS5 fluorescence of a single train, µ_{RFU_NG3_axoneme} is the mean integrated fluorescence along the axonemes of several cilia (excluding the base and the tip), and N_train is the number of trains counted by kymograph analysis in the cilia when µ_{RFU_NG3_axoneme} was measured. To increase measurement precision for calculating µ_{RFU_NG3_axoneme}, cilia were imaged by epifluorescence rather than TIRF illumination as the TIRF field did not reproducibly illuminate cilia from different cells. T_{RFU_mean} was then used to calculate the absolute number of IFT88 or BBS5 molecules per train by using the NG3 standard described in the Absolute quantitation of cilia-localized markers section: T_{labeled_abs} = T_{RFU_average}/µ_{F_NG3}, where T_{labeled_abs} is the mean number of labeled molecules per train, and µF_{_NG3} is the previously measured mean fluorescence of a single NG3. As T_{labeled_abs} does not account for unlabeled IFT88 or BBS5 within a train, a correction must be applied that relates the ratio of unlabeled to labeled molecules in the cell: T_abs = T_{labeled_abs} * ((W_NG3 + W_endogenous)/W_NG3), where T_abs is the number of IFT88 or BBS5 molecules per train, and W_NG3 and W_endogenous are the integrated intensities of Western blot bands for NG3-tagged and endogenous molecules using antibodies against mouse IFT88 or BBS5 (Figs. 3 A and S2 I). T_abs was used to calculate the number of molecules in an anterograde (T_{abs_anterograde}) versus retrograde (T_{abs_anterograde}) train: T_abs = (f_anterograde * T_{abs_anterograde} + f_retrograde * T_{abs_retrograde})/(f_anterograde + f_retrograde), where f_anterograde and f_retrograde are the frequency of either anterograde or retrograde trains. Using KymographClear analysis on ^NG3IFT88, we measured f_anterograde = 21.4 trains/min and f_retrograde = 18.5 trains/min. T_{abs_anterograde} and T_{abs_retrograde} can now be solved using the size ratio (R_T) of anterograde versus retrograde trains: R_T = _{µRFU_T_anterograde}/µ_{RFU_T_retrograde}, where µ_{RFU_T_anterograde} and µ_{RFU_T_retrograde} are the mean NG3 intensity of anterograde or retrograde trains. By KymographClear analyses, R_T = 1.69 for ^NG3IFT88, and R_T = 1.78 for ^NG3BBS5. Relating T_{abs_anterograde} to T_{abs_retrograde}: T_{abs_anterograde} = T_{abs_retrograde} * R_T.

One can calculate the number of IFT88/BBS5 particles in anterograde and retrograde trains: T_{abs_anterograde} = (f_anterograde + f_retrograde) * T_abs/(f_anterograde + fr_etrograde/R_T); T_{abs_retrograde} = (f_anterograde + f_retrograde) * T_abs/(f_anterograde * R_T + f_retrograde).

The longitudinal fluorescence intensities of BBS5, IFT88, SSTR3, and GPR161 were measured in ImageJ by a plot profile of a 5-pixel-wide line along the long axis of the cilium. To average data from multiple cilia, pixel intensities were assigned a length percentage, with 0% referring to the base and 100% referring to the tip. Values were then grouped into 5% bins and averaged. Bin means were then averaged across multiple cilia and plotted (e.g., Fig. 3 B).

To quantify ^NG3BBS5 tipping, a 5 × 7–pixel box was centered at the cilium tip, and the integrated fluorescence intensity within the box was measured by ImageJ. The resulting values were background corrected by subtracting the fluorescence immediately adjacent to the cilium. The values from multiple ciliary tips were averaged and then converted to absolute numbers of BBS5 particles using the NG3 calibration.

Molecules of SSTR3 or GPR161 present on the surface of IMCD3-[pEF1αΔ-^APSSTR3^NG] and IMCD3-[pCrys-^APGPR161^NG3] cells were labeled with ^SAQdot through the affinity of SA for the extracellular biotinylated AP tag. Both cell lines express an ER-targeted BirA ligase that biotinylates the N-terminal AP tag of each GPCR (Howarth and Ting, 2008; Nager et al., 2017). Single GPCR labeling was achieved through a blocking strategy. First, to block surface-exposed biotinylated receptors, cells were washed three times with PBS and then incubated with 10 nM mSA for 15 min. Cells were washed three times with PBS and incubated with 1 nM ^SAQdot (Q10123MP; Thermo Fisher Scientific) solution for 15 min. As the ^SAQdot only binds to receptors newly arrived at the surface during the 15 min of labeling, this approach labels very few receptors. Lastly, cells were washed three times with PBS and transferred to imaging media (DMEM/F-12, Hepes, and no phenol red media; 11039-021; Gibco) with 0.2% FBS and 1 µM biotin to block unbound SA on ^SAQdot.

Qdot-labeled SSTR3 and GPR161 molecules exited the cilium at rates congruent with bulk imaging (Fig. 1, F and G). By counting the number of ^QDSSTR3 per cilia after 2 h treatment with vehicle or sst (Fig. 8 C), we determined that sst treatment resulted in the loss of 25% of cilia-localized Qdots, consistent with the SSTR3 exit rate measured by bulk fluorescence (Fig. 1 F). As a second test, the frequencies of ^QDGPR161 exit events captured by live imaging matched the frequencies predicted from imaging the decrease in bulk fluorescence of GPR161^NG3 (Fig. S5 B).

To monitor single receptor trafficking, cilia bearing one Qdot were imaged on the DeltaVision microscope at 2 Hz for 5–20 min. Cilia were identified by the NG signal from the C-terminal NG and NG3 tags. The centroid of ^SAQdot-labeled single SSTR3 and GPR161 were tracked with dynamic programming method. To achieve subpixel accuracy, each detected spot position was finally replaced by the center of gravity of a small neighborhood (3 * 3) in a postprocessing step (Sage et al., 2005). The localization precision of Qdot was measured by mapping the centroid of an immobile Qdot, where localization precision σ_xy = (σ_x² + σ_y²)^1/2 (Deschout et al., 2014).

Processivity, residence, and confinement frequencies were based on kymograph analyses (examples shown in Fig. S4). Processive ^QDGPCR movements were defined as consecutive frames where a Qdot moved longitudinally along the ciliary axoneme. To distinguish motor-driven from diffusive events, only processive motions that extend for at least six consecutive frames (Fig. S3 G) or 10 consecutive frames (Fig. 8 F) were plotted. Residence at the base and tip was operationally defined as residing at the basal or distal 10% length of the cilium for >2.5 s (Fig. 8 G). As freely diffusing receptors occasionally remain at the cilia base or tip for >2.5 s, confinement was defined by a more stringent criterion of events lasting >15 s (Fig. 8 G).

The most base-proximal position of ^QDGPR161 was mapped on a coordinate system established by the profile of ciliary GPR161^NG3 (Fig. 10, A–C). The same approach was applied to map the immunofluorescence intensity center of Cep290 and Cep164 (Fig. 10, A–C). Intermediate compartment visits were defined as events where the Qdot centroid crossed the 50th percentile of NG intensity in the GPR161^NG longitudinal scan profile (Fig. 10, C and F).

All box plots display the second, third, and fourth quartiles along with whiskers that represent values within the 1.5× the interquartile range. Outliers exceeding the whiskers are plotted as points.

Fig. S1 shows that low-expression promoters recapitulate the physiological exit kinetics of SSTR3. Fig. S2 identifies IFT-A/Tulp3 as the importer of GPCRs into cilia and BBSome/Arl6 as the exporter. Fig. S3 presents BBSome and IFT train processivity and single-molecule quantitation. Fig. S4 shows kymographs of single molecules of ^QDSSTR3 and ^QDGPR161. Fig. S5 shows visualization of ^QDGPR161 exit at single-molecule resolution. Video 1 shows sst-dependent removal of SSTR3 from cilia. Video 2 shows IFT-B foci movements in cilia. Video 3 shows comovement of BBSome and IFT-B. Video 4 shows BBSome foci movement in cilia. Video 5 shows dynamics of BBSome-mediated retrieval. Video 6 shows behavior of Qdot-labeled SSTR3 in control-treated cells. Video 7 shows behavior of Qdot-labeled SSTR3 in sst-treated cells. Videos 8 and 9 show direct observations of ciliary exit of Qdot-labeled GPR161.

We thank A. Kopp and J. Goldstein for technical assistance, and we thank K. Hill and A. Benmerah as well as the Nachury laboratory for comments on the manuscript.

This work was supported by National Institutes of Health funding (GM089933) to M.V. Nachury and a Damon Runyon Cancer Research Foundation Fayez Sarofim Fellowship to A.R. Nager (DRG 2160-13).

The authors declare no competing financial interests.

Author contributions: M.V. Nachury conceived and coordinated the project and wrote the paper with contributions from all authors. F. Ye and A.R. Nager jointly developed the cell lines used for imaging of GPCRs and BBSomes. F. Ye conducted all single-molecule imaging and the quantitative analysis of BBSome trains and exit.