Δ9-Tetrahydrocannabinol inhibits Hedgehog-dependent patterning during development

Congenital malformations affect approximately 8 million newborns worldwide each year and are a leading cause of death for infants and children of all ages (Christianson et al., 2006; Krauss and Hong, 2016; Wallingford, 2019). In some cases, mutations in single genes or exposure to individual teratogens is sufficient to cause a developmental disorder in most or all those affected (Amberger et al., 2019; Gilbert-Barness, 2010; Webber et al., 2015). For many of the most common structural birth defects, however, a single causative factor cannot be identified, and the underlying etiology for such disorders is poorly understood. In these cases, it is likely that genetic and environmental risk factors interact to elevate the chance of a defect occurring in specific developmental processes (Beames and Lipinski, 2020; Fraser, 1980; Krauss and Hong, 2016; Lovely et al., 2017). Genome sequencing has led to identification of numerous birth defect-associated variants, many of which appear to predispose individuals to a given anomaly and presumably act with additional factors (Webber et al., 2015). Identification of subthreshold environmental risk factors by epidemiology is more difficult.

Cannabis is the illicit drug most commonly used during pregnancy and, with expanded legalization and decreased perception of risk, use is increasing (Volkow et al., 2017; Young-Wolff et al., 2017). Meta-analysis of studies through 2014 concluded that maternal cannabis use is associated with low birth weight and increased likelihood of requirement for neonatal intensive care (Gunn et al., 2016). Recently, in Colorado, a correlation was reported between: (1) increased cannabis usage during pregnancy; (2) increased fetal phytocannabinoid exposure levels; and (3) an increase in major structural developmental defects (Reece and Hulse, 2019). Use of other drugs and tobacco remained static or fell in Colorado during the reporting period. A similar pattern was observed with rising incidence of atrial septal defects in multiple US states and Australia (Reece and Hulse, 2020). These correlations suggest that cannabinoids might be teratogenic, but they do not demonstrate causality.

A common birth defect that serves as a model for gene-environment interactions and multifactorial etiology is holoprosencephaly (HPE) (Beames and Lipinski, 2020; Hong and Krauss, 2018; Roessler et al., 2018). HPE is caused by failure to define the midline of the forebrain and/or midface. HPE comprises a phenotypic continuum ranging from complete failure to partition the forebrain into hemispheres with accompanying cyclopia, through to mild midfacial midline deficiency (Cohen, 2006; Muenke and Beachy, 2001). The Hedgehog (HH) signaling pathway is a key regulator of many developmental processes, including patterning of the forebrain and facial midline, limbs and digits, and ventral neural tube (VNT) (Petryk et al., 2015; Sagner and Briscoe, 2019; Tickle and Towers, 2017). HPE is associated with heterozygous mutations in the HH pathway (Dubourg et al., 2018; Roessler et al., 2018). Clinical presentation of HPE is highly variable, and many mutation carriers are unaffected, even in pedigrees. These observations have led to a multifactorial, ‘mutation plus modifier’ model, in which heterozygous mutations may be insufficient for severe phenotypes and their penetrance and expressivity are graded by additional genetic variants and/or environmental exposures (Dubourg et al., 2018; Hong and Krauss, 2018; Roessler et al., 2012).

HH ligands activate a conserved signaling pathway (Kong et al., 2019; Lee et al., 2016; Petrov et al., 2017). In the absence of HH, the primary receptor patched 1 (PTCH1) acts to inhibit the activity of a second membrane protein, smoothened (SMO). Binding of HH to PTCH1 relieves inhibition of SMO. SMO then signals to activate pathway target genes via GLI transcription factors. SMO is a class F, G protein-coupled receptor (GPCR). PTCH1 appears to function as a transporter to restrict accessibility of SMO to its activating ligands, namely cholesterol and oxysterols. HH ligands block PTCH1 function, allowing SMO access to cholesterol and oxysterols, thus activating SMO signaling (Qi and Li, 2020; Radhakrishnan et al., 2020). These events occur in the primary cilium, an organelle in which HH pathway components are trafficked and concentrated for signaling (Bangs and Anderson, 2017; Gigante and Caspary, 2020). Consistent with the notion that SMO is itself a ligand-regulated receptor, many small molecules that act as SMO agonists or antagonists have been identified (Sharpe et al., 2015). Although PTCH1 function is sufficient for SMO inhibition, HH signal reception also requires at least one of three co-receptors (CDON, BOC, GAS1). CDON, BOC and GAS1 have overlapping roles and are collectively required for HH signaling (Allen et al., 2011; Izzi et al., 2011; Wierbowski et al., 2020; Zhang et al., 2011). Mice with targeted mutations in any one of these co-receptors have a selective and partial loss of HH pathway function.

In a search for endogenous lipids that act as SMO antagonists, Eaton and colleagues identified endocannabinoids as inhibitors of HH signaling (Khaliullina et al., 2015). Endocannabinoids were effective as HH pathway inhibitors in both developing Drosophila wing disks and cultured mammalian cells. Endocannabinoids are fatty acids/alcohols linked to polar head groups that signal through the GPCRs cannabinoid receptor-type 1 (CB1R; CNR1) and -type 2 (CB2R; CNR2) (Lu and Mackie, 2020; Maccarrone et al., 2015). Phytocannabinoids are bioactive ingredients of cannabis and include Δ9-tetrahydrocannabinol (THC) and cannabidiol (CBD), the former being the major psychoactive component. These compounds exert their effects via CB1R, CB2R, and/or other receptors, with CB1R responsible for mediating the major neurobehavioral effects of THC (Lu and Mackie, 2020; Schurman et al., 2020). Importantly, THC and CBD inhibit HH signaling in a similar manner to endocannabinoids, whereas structurally unrelated CB1R and CB2R agonists/antagonists do not (Khaliullina et al., 2015). Cannabinoids have been proposed to inhibit HH signaling at the level of SMO, although this mechanism was not uniform among those analyzed and has been questioned by others (Sever et al., 2016).

These findings raise the possibility that in utero exposure to phytocannabinoids might be teratogenic, owing to an ability to inhibit HH signaling at crucial points during development. We report here that THC is teratogenic to genetically sensitized mice harboring a subthreshold deficit in HH pathway signaling strength. THC dose-dependently induced HH loss-of-function phenotypes, including HPE, in these mice but not in wild-type mice. THC therefore acted as a ‘conditional teratogen’, dependent on a complementary but insufficient genetic insult. Furthermore, in in vitro assays, THC displayed properties similar to the bone fide SMO inhibitor SANT1. Together, these results raise the possibility that human cannabis consumption during early pregnancy may expose embryos to HH inhibition, presenting an environmental risk factor for birth defects.

Cdon^−/− mice on a 129S6 background have a subthreshold defect in HH signaling and are sensitive to induction of HPE by both genetic and environmental modifiers (Hong and Krauss, 2018). They therefore model human HPE and are an ideal model in which to assess THC teratogenicity. 129S6 Cdon^+/− mice were intercrossed and pregnant females received a single dose of THC at 5, 10 or 15 mg/kg, administered intraperitoneally at embryonic day (E) 7.5. The E7.5 time point was used because it was most effective for induction of HPE in wild-type mice by the potent SMO inhibitor vismodegib (Heyne et al., 2015). We measured blood concentrations of THC and metabolites in pregnant female mice, and the peak levels achieved at the highest THC dose used were similar to those achieved by humans after inhalation of 34 mg cannabis (180-200 ng/ml) (Grotenhermen, 2003) (Fig. S1).

Embryos were harvested at E14 and scored as positive for HPE if they displayed a fused upper lip, an unambiguous phenotype without gradation that arises as a consequence of defective craniofacial midline patterning (Hong and Krauss, 2012). Vehicle- and THC-treated Cdon^+/+ and Cdon^+/− embryos did not have HPE, nor did vehicle-treated Cdon^−/− embryos. In contrast, THC-exposed Cdon^−/− embryos displayed mid-facial HPE phenotypes in a dose-responsive manner (Fig. 1A,B; Table S1). Hematoxylin and Eosin (H&E)-stained sections revealed that THC-treated Cdon^−/− embryos had fully partitioned forebrains, but displayed substantial diminution of midfacial structures, including close-set and rudimentary vomeronasal organs, and reduced width of the nasal septum cartilage (Fig. 1A-A‴; Fig. S2). These results classify THC-induced HPE in Cdon^−/− mice as the relatively mild, lobar category (Krauss, 2007). Whole-mount in situ hybridization at E10 revealed reduction in expression of two direct SHH target genes, Gli1 and Nkx2-1, in the rostroventral midline of THC-treated Cdon^−/− embryos (Fig. 1C).

To investigate the effects of THC on gene expression more quantitatively, RNA was extracted from dissected forebrains of E9.0 embryos, and quantitative RT-PCR (qRT-PCR) was performed for several SHH target genes. Although there was some variability, possibly related to partial penetrance, mRNA levels of Gli1, Nkx2-1 and Shh were significantly reduced in THC-treated Cdon^−/− embryos (Fig. 1D). Ptch1 mRNA levels trended lower in THC-treated Cdon^−/− embryos, although this was not statistically significant owing to variability between the samples (P=0.08). In contrast, levels of Bmp4 mRNA, which controls CNS patterning dorsally and medially and is not known to be directly regulated by HH signaling, was similar between controls and THC-treated Cdon mutants (Fig. 1D). Additionally, these results may underestimate the reduction in expression of SHH target genes in the most affected region of THC-treated Cdon^−/− embryos because non-midline structures were by necessity present in the dissected tissue. Taken together, these results show that Cdon mutation and THC synergized to induce HPE in mice.

We next sought to assess the effects of in utero exposure to THC on a second HH-dependent patterning process. Sonic HH (SHH) produced by the notochord and floor plate (FP) forms a ventral-to-dorsal gradient of pathway activity in the developing neural tube. In response to distinct levels of SHH pathway activity, expression of specific transcription factors is induced in specific progenitor zones of the VNT. These proteins include: FOXA2 (in the FP), NKX2-2 (in p3 progenitors) and OLIG2 (in pMN motor neuron progenitors). Pregnant dams were treated with a single dose of THC (15 mg/kg) at E8.0 and embryos were analyzed at E9.5, by immunofluorescence (IF) analysis of forelimb-level sections. The E8.0 time point was used because it was effective for inhibition of SHH-dependent VNT patterning in wild-type embryos by the SMO inhibitor cyclopamine (Ribes et al., 2010). THC reduced the number of FOXA2⁺ FP cells by >50% and of NKX2-2⁺ p3 progenitors by >35% in Cdon^−/− embryos (Fig. 2). The more dorsally positioned OLIG2⁺ pMN progenitors, which require a lower level of HH signal to be induced, were not affected (Fig. 2). These results are very similar to those obtained by removal of one copy of Shh from Cdon^−/− mice (Tenzen et al., 2006). Taken together, our findings demonstrate that THC is teratogenic in genetically sensitized mice, producing two well-established, SHH loss-of-function phenotypes: HPE and defective VNT patterning.

Khaliullina et al. reported that THC inhibited a HH-dependent reporter construct in NIH3T3 cells, a well-established cell culture system for studying HH signaling and the mechanisms of pathway inhibitors (Khaliullina et al., 2015; Taipale et al., 2000). We confirmed this observation. Treatment of NIH3T3 cells with recombinant SHH induced an ∼10-fold increase in expression of the direct, endogenous target gene Gli1 as measured by qRT-PCR (Fig. 3A). THC dose-dependently inhibited Gli1 induction in response to SHH, with an IC₅₀ of ∼1 µM. Similar results were obtained when these cells were stimulated with the direct SMO agonist SAG (Fig. S3A), and with SHH treatment of a second cell system, freshly prepared mouse embryo fibroblasts (MEFs) (Fig. S3B).

It has been reported that cannabinoids inhibit HH signaling at the level of SMO, although differences were seen between the cannabinoids examined, and THC was not tested (Khaliullina et al., 2015). PTCH1 and SUFU are negative regulators of HH signaling. MEFs that are null for either gene display constitutive pathway activity, with PTCH1 functioning upstream of SMO, and SUFU acting downstream of SMO (Kong et al., 2019; Lee et al., 2016; Petrov et al., 2017). To identify the position within the HH pathway at which THC acts, we first tested its ability to inhibit constitutive expression of Gli1 in Ptch1^−/− and Sufu^−/− MEFs. THC attenuated Gli1 expression in Ptch1^−/− MEFs, but not in Sufu^−/− MEFs, in a dose-dependent manner (Fig. 3B). Therefore, THC modulates SHH signaling downstream of PTCH1 and upstream of SUFU. Although these data place the likely point of action of THC at the level of SMO, this could occur via direct SMO inhibition or indirectly, e.g. via perturbation of primary cilia, the cellular site of signaling to GLI.

SMO activation is a two-step process. Ciliary transport of SMO (step 1) is followed by SMO activation within cilia (step 2) (Fig. 3C) (Rohatgi et al., 2009; Wilson et al., 2009). A low level of SMO trafficks through primary cilia constitutively. Some direct SMO inhibitors, such as SANT1, induce a SMO conformation that inhibits trafficking and prevents ciliary accumulation in response to activators, such as SHH and SAG. Another class of SMO inhibitor, exemplified by cyclopamine, induces a SMO conformation that does not inhibit trafficking, but which resists activation within cilia. Therefore, cyclopamine-type inhibitors trap SMO in primary cilia even in the absence of SHH or SAG, yet block activation of SMO within cilia, even in the presence of these pathway activators (Rohatgi et al., 2009; Wilson et al., 2009). We assessed the effect of THC on ciliary translocation of endogenous SMO in NIH3T3 cells treated with various HH pathway-regulating compounds (Fig. 3D,E; Table 1). The percentage of cells with ciliary SMO and fluorescence intensity of ciliary protein expression were determined. THC alone did not trigger SMO translocation to cilia, but it blocked SMO accumulation in cilia in response to SHH, cyclopamine or SAG. THC also inhibited constitutive ciliary localization of SMO in Ptch1^−/− MEFs. THC did not affect cilium length or ciliary levels of the primary cilia marker ARL13B (Fig. S4A,B). Therefore, THC acts similarly to SANT1, preventing SMO translocation into primary cilia.

The phytocannabinoid CBD is structurally related to THC and has also been reported to inhibit HH signaling (Khaliullina et al., 2015). In contrast to our results with THC, Khaliullina et al. found that CBD did not prevent localization of exogenously expressed SMO to primary cilia in response to SAG (Khaliullina et al., 2015). Because these results and our own findings with THC were dichotomous, we tested CBD as well. CBD inhibited SHH-induced Gli1 expression in NIH3T3 cells with an IC₅₀ of ∼1.2 µM, a value similar to that of THC (Fig. 4A). Like THC, CBD alone did not promote translocation of SMO to primary cilia (Fig. 4B,C). We found that CBD did reduce endogenous SMO ciliary translocation in response to SHH. However, despite having similar IC50 values, CBD was significantly less effective than THC in reducing: (1) the percentage of cells with SMO⁺ cilia; and (2) the steady-state amount of SMO within SMO⁺ cilia, in SHH-treated cells (Fig. 4B,C; Table 1). CBD did not affect cilium length or ciliary levels of the primary cilia marker ARL13B (Fig. 4C). CBD may therefore inhibit HH signaling by a mechanism somewhat distinct from either SANT1-type or cyclopamine-type SMO inhibitors.

To identify whether THC may be a direct inhibitor of SMO, we measured its ability to compete with a fluorescent derivative of cyclopamine (Bodipy-CPA) for binding to SMO (Chen et al., 2002). We transfected HEK293 cells with an expression vector encoding Myc epitope-tagged SMO or with an empty vector as a control. SMO-Myc-expressing cells, but not control cells, displayed strong Bodipy-CPA binding (Fig. 5). SANT1 and KAAD-cyclopamine (a potent analog of cyclopamine) each competed with Bodipy-CPA for SMO binding (Fig. 5). THC at 10 µM, but not 1 µM, also inhibited Bodipy-CPA binding (Fig. 5). In contrast, 10 µM CBD did not significantly diminish Bodipy-CPA binding to SMO-expressing cells (Fig. 5). Although 10 µM is a high concentration of THC, cholesterol (an endogenous SMO ligand) competed with Bodipy-CPA in a similar concentration range (Huang et al., 2016). Furthermore, the fact that CBD, a close structural congener of THC, was not effective at this dose argues that the ability of THC to compete with Bodipy-CPA is specific.

The psychoactive effects of THC are exerted via CB1R, encoded by the gene Cnr1 (Lu and Mackie, 2020). It has been proposed that cannabinoids inhibit HH signaling via CB1R-SMO heterodimers (Fish et al., 2019). We sought to explore this possibility further. First, if CB1R is required for cannabinoids to inhibit SMO, CB1R must be expressed in cells that display a cannabinoid-sensitive HH response. NIH3T3 cells and MEFs are HH responsive, and THC inhibits this response (Fig. 3A; Figs. S3A,B). However, we did not detect Cnr1 mRNA or CB1R protein in either cell type, whereas they were easily detected in mouse whole-brain lysate (Fig. 6A,B).

We next performed IF analysis for CB1R on VNT sections from embryos collected at E8.5 (both neural fold and early NT stages), E9.5, E10.5 and E11.5. CB1R was detected in some cells of the ventricular zone at the neural fold stage, consistent with previous reports (Gilbert et al., 2015; Psychoyos et al., 2012). However, little to no specific CB1R IF signal was observed in the developing VNT at E8.5 early neural tube stage, E9.5 or E10.5, with the exception of a few cells at the ventricular edge of the floor plate that reproducibly exhibited a strong signal (Fig. 6C). Robust CB1R expression was observed at E11.5, but this was localized to the HH-non-responsive mantle domain, which contains more differentiated neurons (Fig. 6C). These results demonstrate that CB1R levels are very low in VNT cells at developmental stages when they are undergoing THC-sensitive, SHH-dependent neuronal patterning. These findings are consistent with mRNA in situ hybridization data showing a lack of Cnr1 expression at similar stages in the chick VNT (Watson et al., 2008), patterning of which closely resembles the mouse VNT (Dessaud et al., 2008). In general, Cnr1 expression follows neuronal differentiation (Watson et al., 2008), consistent with our IF results in the VNT. Our IF with CB1R antibodies was further validated by staining sections of E14 cerebral cortex (Fig. 6C), which reproduced localization patterns similar to those previously reported (Mulder et al., 2008).

We next attempted to co-immunoprecipitate CB1R and SMO. Extracts of E14 and adult mouse brains were immunoprecipitated with antibodies to CB1R and then blotted for CB1R and SMO. CB1R and SMO were detected in whole lysates by western blot. However, although CB1R was efficiently immunoprecipitated, SMO was not detectable in these precipitates (Fig. 6D).

Finally, to test for effects of CB1R in THC-mediated inhibition of HH signaling, we established NIH3T3 cells that expressed CB1R in a doxycycline-inducible manner. We reasoned that if CB1R levels were rate-limiting for the inhibition of HH signaling by THC, exogenous expression of CB1R in NIH3T3 cells would sensitize them to THC-mediated attenuation of SHH-induced expression of Gli1. Treatment of these cells with 0.05 µg/ml doxycycline induced CB1R expression to a level similar to that seen in Neuro-2a cells, which display CB1R-dependent signaling in response to cannabinoids (Graham et al., 2006; He et al., 2005) (Fig. S5A). CB1R expression did not alter the ability of THC to dose-dependently inhibit Gli1 mRNA induction in response to SHH stimulation (Fig. 6E). Furthermore, CB1R was not present in primary cilia, in control, THC-, SHH- or THC+SHH-treated cells (Fig. S5B). Taken together, we conclude that CB1R is not required for THC-mediated inhibition of HH signaling.

We report here that THC is teratogenic in genetically sensitized mice (129S6 Cdon^−/− mice). A single in utero dose of THC resulted in two hallmark HH loss-of-function phenotypes: HPE and VNT patterning defects. In vitro experiments argue that THC exerted these effects as a direct, albeit relatively weak, inhibitor of the essential HH pathway signal transducer SMO. As the effects of THC were dependent on a complementary, but insufficient, genetic deficiency, we categorize it as a ‘conditional teratogen’. Cannabis consumption during pregnancy may therefore result in partial inhibition of a major morphogenetic pathway in embryos, thereby contributing to the complex combination of genetic and environmental risk factors that are thought to underlie many common developmental disorders.

Eaton and colleagues first reported that endocannabinoids and phytocannabinoids inhibit HH signaling, but did not investigate cannabinoid effects in a vertebrate model in vivo, and did not focus on THC (Khaliullina et al., 2015). Our results are consistent with the conclusion that THC is a direct inhibitor of SMO. These findings include: (1) the inhibitory activity of THC mapped downstream of PTCH1 and upstream of SUFU; (2) THC prevented translocation of SMO to primary cilia upon pathway activation, without obviously altering cilia themselves; and (3) THC competed with a cyclopamine derivative for binding to SMO expressed on cells. THC therefore acted similarly to the well-studied SMO inhibitor SANT1, although it is much less potent than SANT1. The ability of THC to induce HH loss-of-function phenotypes in mice with a subthreshold HH signaling defect is fully consistent with its function as a relatively weak SMO inhibitor. In fact, THC treatment and Shh heterozygosity acted similarly in Cdon^−/− mice, in that each enhanced the effects of Cdon mutation on facial midline and VNT patterning, but neither was sufficient to perturb patterning on their own (Tenzen et al., 2006).

The THC structural analog CBD inhibited HH signaling in vitro with a similar IC50 as THC, but was much weaker at affecting SMO translocation to cilia and did not effectively compete for binding to SMO. CBD is therefore likely to inhibit HH signaling by a molecular mechanism somewhat distinct from that of THC, even if SMO might be its target. The SMO transmembrane domain has several potential binding modalities for small molecules (Kowatsch et al., 2019; Qi and Li, 2020), and it is possible that CBD binds in a deeper portion of the transmembrane pocket than cyclopamine, and so does not displace it. Such binding may induce a SMO conformation that does not efficiently block ciliary trafficking but inhibits adoption of an active conformation. Alternatively, CBD could bind to the amino-terminal cysteine-rich domain, displacing allosteric regulators (Huang et al., 2018). THC and CBD exert their effects outside the HH pathway via different receptors, so, despite being closely related, these compounds have important structural features that distinguish their activities (Dos Santos et al., 2021). The psychoactive properties of THC are the major reason people consume cannabis; these properties are not shared by CBD. However, CBD is available over the counter and viewed by many as generally beneficial, or at least innocuous, so it will be important to test CBD for conditional teratogenicity also.

While this work was in progress, Fish et al. reported studies on the effects of several phytocannabinoids and synthetic cannabinoids, including THC, in C57BL/6J mice (Fish et al., 2019). Consistent with the work reported here, THC induced ‘minor alterations of pupil shape and size’ in these mice, but exacerbated effects on eye morphology induced by fetal alcohol exposure (Fish et al., 2019).

It has been suggested that cannabinoids inhibit HH signaling via CB1R-SMO heterodimers (Fish et al., 2019). Evidence for this conclusion included co-immunoprecipitation of CB1R with SMO (Fish et al., 2019). However, our findings showed that CB1R is not required for THC to inhibit the HH pathway and argue against CB1R-SMO interaction. There are several lines of evidence for these conclusions. First, NIH3T3 cells and MEFs responded to HH signaling in a THC-inhibitable manner, yet they do not express detectable Cnr1 mRNA or CB1R protein. Furthermore, THC inhibited VNT patterning, but VNT cells expressed little CB1R from E8.5 to E10.5, a period when they are acquiring neuronal identity in response to SHH signaling. This is consistent with mRNA in situ hybridization for Cnr1 expression at a similar stage of the developing chick VNT (Watson et al., 2008). Second, we attempted to co-immunoprecipitate CB1R and SMO without success. The previously reported, co-precipitating band identified as CB1R in anti-SMO immunoprecipitates migrated at ∼100 kDa (Fish et al., 2019). The calculated molecular weight of CB1R is 53 kDa, approximately the size of the band we detected in our experiments. A recent study used multiple CB1R antibodies against brain lysates from control and Cnr1 null mice (Esteban et al., 2020). All antibodies recognized a band of ∼53 kDa, that was specifically lost in the knockout mice. It is likely, therefore, that the ∼100 kDa band that co-immunoprecipitated with SMO was not CB1R. Third, exogenous expression of CB1R in NIH3T3 cells did not alter the dose dependency of THC-mediated inhibition of SHH signaling. Furthermore, CB1R did not localize to primary cilia in the presence or absence of THC, plus or minus SHH ligand. It was reported that the CB1R inhibitor SR141716A rescued eye phenotypes associated with in utero exposure of C57BL/6 mice to the synthetic cannabinoid CP55,940, or to CP55,940 plus alcohol (Fish et al., 2019). However, the eye phenotypes in this system are not distinctive to HH loss of function, and CB1R agonism may play a role in their genesis. Finally, although the role of endocannabinoids as inhibitors of vertebrate HH signaling is unknown, it is very unlikely that any such role would require CB1R or CB2R, as Cnr1;Cnr2 double mutant mice are viable and do not display any reported HH gain-of-function phenotypes (Rowley et al., 2017; Sophocleous et al., 2017). This is in contrast to the strong gain-of-function phenotypes seen in embryos upon loss of established negative regulators of the HH pathway (e.g. Ptch1, Sufu, Gnas, Gpr161, Ankmy2) (Goodrich et al., 1997; Mukhopadhyay et al., 2013; Regard et al., 2013; Somatilaka et al., 2020; Svard et al., 2006).

Taken together, we conclude that CB1R is not required for THC to inhibit HH signaling. Although our findings do not conclusively rule out formation of CB1R-SMO complexes, neither our work nor published studies offer compelling evidence that such complexes exist.

The causes of common developmental disorders are often unknown. Many cases likely arise from a combination of subthreshold genetic and environmental insults, i.e. multifactorial etiology (Beames and Lipinski, 2020; Krauss and Hong, 2016). Whole-genome sequencing should eventually reveal a great majority of genetic contributions to such defects; subthreshold environmental risk factors, however, are more difficult to identify. The HH signaling pathway plays a fundamental role during development and is involved in growth and morphogenesis of a wide variety of body structures, including limbs, brain, heart and craniofacial structures (Petryk et al., 2015; Sagner and Briscoe, 2019; Tickle and Towers, 2017). Mutations in HH pathway components are involved in several developmental disorders, including Greig cephalopolysyndactyly syndrome, Gorlin syndrome, Brachydactyly A1 and HPE (Nieuwenhuis and Hui, 2005). Environmental agents that perturb HH signaling during specific times during embryogenesis may therefore be risk factors for developmental disorders, perhaps working in concert with predisposing genetic variants.

Cannabis is the most frequently used illicit drug during pregnancy and in young women of childbearing age; as legalization progresses, use will presumably increase. Furthermore, the THC concentration of both medicinal and recreational cannabis is currently very high (∼20% THC) (Cash et al., 2020). Identification of a potential role of cannabis consumption in the etiology of structural birth defects is significant, as it would represent an avoidable risk factor. Cannabinoids are HH pathway inhibitors but they have not been linked to human HPE, the most common outcome of genetic deficiency in HH signaling (Croen et al., 2000; Linn et al., 1983; Miller et al., 2010; van Gelder et al., 2009, 2010). HPE occurs in ∼1:250 conceptions, with >97% succumbing to intrauterine lethality (Leoncini et al., 2008; Shiota, 2021; Shiota and Yamada, 2010). Ordinarily, epidemiology would be used to assess whether specific environmental agents are risk factors, but epidemiological studies on possible teratogenic effects of cannabis have been neither designed nor powered to detect a link with HPE. There are several reasons for this. First, the >97% in utero lethality rate results in low case numbers that can be studied (Muenke and Beachy, 2001). Second, the window of sensitivity to teratogen-induced HPE in model systems is narrow and equivalent to the third week of human pregnancy (Heyne et al., 2015), a time when many women do not yet know they are pregnant. Such women may use cannabis at this stage, stop when they know they are pregnant, and self-report as not using the drug (Linn et al., 1983). Finally, complex etiologies that involve subthreshold insults, as can occur with HPE, are often difficult to parse with epidemiology. Nevertheless, recent findings that correlate increased cannabis usage with specific structural birth defects (Reece and Hulse, 2019, 2020), combined with our identification of THC as a conditional teratogen, indicates that further research into this question is important and of significant public health relevance.

All animal procedures were conducted in accordance with institutional guidelines for the care and use of laboratory animals as approved by the Institutional Animal Care and Use Committee (IACUC) of the Icahn School of Medicine at Mount Sinai. Two- to three-month-old Cdon^+/− mice on a 129S6/SvEvTac (129S6) background (Cole and Krauss, 2003; Hong and Krauss, 2012) were mated for 1-2 h and plugged females were collected. The time of the plug was designated as E0.0. THC was injected intraperitoneally at selected times and various doses with 0.3% Tween-80 in saline as a vehicle. Offspring were examined as whole embryos at E14.0 for external signs of HPE. In other experiments, embryos were collected at various time points for analysis and staged for comparison by number of somites (see figure legends). Injected females were also assessed for plasma concentration of THC and metabolites over 60 h using a BIOO Scientific MaxSignal THC ELISA kit, according to the manufacturer's instructions. THC did not have obvious effects on embryo size at any stage.

THC and CBD were obtained from the NIDA Drug Supply Program. Δ9-THC (50 mg/ml in ethanol solution) was evaporated under nitrogen gas, dissolved in 0.9% NaCl (saline) containing 0.3% Tween 80 to a concentration of 0.75 mg/ml (DiNieri and Hurd, 2012) for administration to mice. THC and CBD were diluted to 0.75 mg/ml in ethanol solution for treatment of cells in vitro.

Whole-mount RNA in situ hybridization was performed according to standard protocols (Hong and Krauss, 2012). Briefly, E10 embryos were dissected out and fixed in 4% paraformaldehyde in PBS, dehydrated through a graded methanol series, and stored at −20°C. Rehydrated embryos were treated with 10 μg/ml proteinase K (Qiagen) in PBS, 0.1% Tween-20 (PBST) according to stage. Embryos were rinsed with PBST, post-fixed and hybridized with digoxygenin (DIG)-labeled probe in hybridization mix (50% formamide, 1.3× SSC pH 5.0, 5 mM EDTA, 50 μg/ml yeast tRNA, 100 μg/ml heparin, 0.2% Tween-20, 0.5% CHAPS) overnight at 65°C. They were washed, blocked with 2% Roche blocking reagent (Roche, 11096176001), 20% heat-inactivated lamb serum in Tris-buffered saline with 0.1% Tween-20 (TBST) and incubated with alkaline phosphate-conjugated anti-DIG antibody (Roche, 11093274910, 1:2000) in blocking buffer overnight at 4°C. After washing with TBST and NTMT (100 mM NaCl, 100 mM Tris pH 9.5, 50 mM MgCl₂ and 0.1% Tween-20), embryos were stained with BM Purple AP substrate (Roche) in the dark. Stained embryos were cleared in 80% glycerol and photographed with a Jenoptik ProgRes C3 camera attached to Nikon SMZ 1500 stereomicroscope. Captured images were assembled by Helicon Focus software (Helicon Soft).

Embryos were fixed in 4% paraformaldehyde and embedded in paraffin for preparation of 7-μm-thick sections for H&E staining. Sections were then dehydrated through graded ethanol and xylene and mounted with Permount (Fisher Scientific). For VNT staining, embryos were infiltrated with sucrose and then embedded in O.C.T frozen medium. Cryosections (12 μm thick) of the forelimb region were collected and transferred to Superfrost Plus slides. After post-fixation in 4% paraformaldehyde, sections were washed three times in PBX (0.2% Triton X-100 in PBS). Primary and secondary antibodies were diluted in blocking buffer (1× PBX plus 3% BSA). Sections were incubated in primary antibodies overnight at 4°C. After washing with PBX, sections were incubated in secondary antibodies for 2 h at room temperature. The primary antibodies used were: OLIG2 (Millipore Sigma, AB9610, 1:1000), CB1R (Cayman Chemical, 10006590, 1:200), FOXA2 (Developmental Studies Hybridoma Bank, 4C7, 1:20), NKX2-2 (Developmental Studies Hybridoma Bank, 74.5A5, 1:20), ARL13B (NeuroMab, 75-287, 1:500), β-Tubulin III (Tuj1) (Millipore Sigma, T8660, 1:200). Secondary antibodies were: anti-rabbit Alexa 568 (Thermo Fisher Scientific, A-11011, 1:500) or Alexa 488 (Thermo Fisher Scientific, A-11008, 1:500); anti-mouse IgG1 Alexa 647 (Thermo Fisher Scientific, A-21240, 1:500); anti-mouse IgG2b Alexa 488 (Thermo Fisher Scientific, A-21141, 1:500); anti-mouse IgG2a Alexa 594 (Thermo Fisher Scientific, A-21135, 1:500). VECTASHIELD antifade mounting medium (Vector Laboratories, H-1000) was used to prevent signals from fading. Cryosections were imaged using a Leica SP5 DMI confocal microscope with a 20× objective. For VNT staining, five embryos of each group were analyzed. Between three and six individual sections per mouse were quantified for DAPI-, FOXA2-, NKX2-2- and OLIG2-positive cells. Data were processed using Fiji ImageJ software.

A mouse Cnr1 cDNA fragment (Addgene, plasmid #13391) was cloned into an all-in-one tetracycline-inducible plasmid (pAS4.1w.Ppuro-aON, Academia Sinica, Taiwan). Recombinant lentivirus was produced by transfecting HEK293T cells with Lipofectamine 2000 (Invitrogen, 11668019) or TransIT-2020 transfection reagent (Mirus, MIR5404) and a third generation lentiviral package system, including pMD2.G, pRSV-Rev and pMDLg/RRE packaging plasmids (Dull et al., 1998). Recombinant lentivirus was used to infect NIH3T3 cells and stably expressing cells selected with 5 µg/ml puromycin (Thermo Fisher, A1113803). Conditional expression of CB1R was induced by the addition of doxycycline (Millipore Sigma, D9891).

NIH3T3 cells (obtained from Dr Stuart Aaronson, Icahn School of Medicine at Mount Sinai, New York, USA), HEK293T (obtained from ATCC), MEFs (freshly prepared), Neuro-2a cells (obtained from Dr Lakshmi Devi, Icahn School of Medicine at Mount Sinai, New York, USA), and Ptch1^−/− and Sufu^−/− MEFs (obtained from Dr Rune Toftgard, Karolinska Institute, Stockholm, Sweden) were grown in DMEM supplemented with 10% fetal bovine serum, penicillin and streptomycin (Thermo Fisher Scientific, 15140-122). All cells were free of contamination. For qRT-PCR, 80% confluent NIH3T3 cells or MEFs were cultured overnight in DMEM plus 2% FBS, after which they were incubated for another 24 h in fresh DMEM plus 2% fetal bovine serum supplemented with 5 ng/ml SHH (recombinant human amino terminal fragment of SHH, StemRD), SAG (Millipore Sigma, 566660) or cyclopamine (LC Laboratories, C-8770) in the presence or absence of antagonist. After incubation, total RNA was extracted using the RNeasy mini kit (Qiagen). For mouse E9 forebrain and E14 whole brain analyses, total RNA was extracted with the Trizol/RNeasy mini kit. Reverse transcription and cDNA production were performed with the Superscript III first strand synthesis system (Invitrogen). qPCR was performed using iQ SyBR Green Supermix (Bio-Rad) on an iCycler iQ5 (Bio-Rad). Gene expression levels were normalized to Gapdh. The primer sequences were as follows: Gapdh, AACGACCCCTTCATTGAC (forward primer) and TCCACGACATACTCAGCAC (reverse primer) (Dong et al., 2008); Gli1, CCAAGCCAACTTTATGTCAGGG (forward primer) and AGCCCGCTTCTTTGTTAATTTGA (reverse primer) (Harvard primer, 6754002a1); Cnr1, CTGGTTCTGATCCTGGTGGT (forward primer) and TGTCTCAGGTCCTTGCTCCT (reverse primer) (Ludányi et al., 2008); Nkx2-1, 5′-TCCAGCCTATCCCATCTGAACT-3′ (forward primer) and 5′- CAAGCGCATCTCACGTCTCA-3′ (reverse primer) (Zhang et al., 2007); Ptch1, 5′- AAAGAACTGCGGCAAGTTTTTG-3′ (forward primer) and 5′-CTTCTCCTATCTTCTGACGGGT-3′ (reverse primer) (Harvard primer, 6679519a1); Shh, 5′-CTGGCCAGATGTTTTCTGGT-3′ (forward primer) and 5′-GATGTCGGGGTTGTAATTGG-3′ (reverse primer) (Ochoa et al., 2010); Bmp4, 5′-GCTTCTGCAGGAACCAATGGA-3′ (forward primer) and 5′-TCCCGGTCTCAGGTATCAAACTAG-3′ (reverse primer) (Du et al., 2010).

Cells were plated on glass slides coated with rat tail collagen (Sigma-Aldrich, 11179179001). After incubation, cells were fixed in 4% paraformaldehyde at room temperature for 10 min and then in 100% methanol at −20°C for 4 min. Fixed cells were permeabilized in 0.1% Triton X-100 and blocked in 1× PBS plus 3% bovine serum albumin for 30 min. Cells were incubated in primary antibodies overnight at 4°C and then secondary antibodies for 2 h at room temperature. The primary antibodies used were: SMO (Santa Cruz Biotechnology, sc-166685, 1:50), ARL13B (Proteintech, 17711-1-AP, 1:500), γ-tubulin (Abcam, ab-11316, 1:200) and acetylated tubulin (Sigma-Aldrich, T7451, 1:500). Secondary antibodies used were: anti-rabbit Alexa 488; anti-mouse IgG1 Alexa 647; anti-mouse IgG2a Alexa 594 (Thermo Fisher, 1:500). Images were captured on a Leica SP5 DMI confocal microscope with a 63× oil objective. A shifted overlay of SMO and ARL13B signals was used for better visualization. For quantification of ciliary SMO, first a mask was determined by using the ARL13B image, and then the mask was applied to the corresponding SMO image where the fluorescence intensity was measured. An identical area nearby but outside the cilia was also measured to determine background fluorescence. Background subtraction was applied on each primary cilium. The percentage of ciliary SMO was determined by fluorescence intensity. Compared with the mean value of positive controls, when treatments decreased SMO signal by more than 50%, cilia were considered negative. The lengths of primary cilia were measured by the ARL13B signal. Between 100 and 140 cilia were analyzed per condition. Data were quantified using Fiji ImageJ software.

To detect the interaction of CB1R and SMO, mouse brain lysates were harvested in lysis buffer containing 50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 2 mM EDTA, 10% glycerol, 0.5% Nonidet P40, 5 mM NaF, 1 mM Na₃VO₄, and a protease inhibitor cocktail (Sigma-Aldrich). Whole brain lysate was precleared with Protein G agarose (Roche, 11243233001) and the supernatant was immunoprecipitated with CB1R antibody overnight at 4°C. The lysate was then incubated with Protein G agarose for 3 h at 4°C. Material pulled down was washed with lysis buffer three times and eluted by boiling. Eluates were separated by SDS-PAGE and transferred to PVDF membranes, which were immunoblotted with primary antibodies against CB1R and SMO overnight at 4°C. Secondary antibodies used for western blotting were: anti-mouse (Cell Signaling Technology, 7076, 1:5000), HRP-conjugated anti-rabbit (Cell Signaling Technology, 7074, 1:5000) or HRP-conjugated anti-rabbit secondary antibodies, light chain specific (Genetex, GTX221666, 1:2000; Jackson ImmunoResearch, 211-032-171, 1:5000), which do not detect IgG heavy chain.

BODIPY-CPA assays were carried out as detailed by Chen et al.(2002) with minor modifications. Briefly, SMO tagged at its C terminus with Myc epitope was expressed in 293T cells by transient transfection. Empty vector was transfected as a control. Twenty-four hours later, cells were plated on rat tail collagen-coated glass slides, and 24 h after that were incubated for 4 h in OPTI-MEM with 5 nM BODIPY-CPA (BioVision, 2160-50), in the presence or absence of competitor drug SANT1 (Millipore Sigma, 559303) or KAAD-cyclopamine (Calbiochem, 239804) at 37°C. The cells were washed with PBS and fixed in 4% paraformaldehyde for 15 min, followed by permeabilization with PBS containing 0.1% Triton X-100. SMO was detected with Myc antibody (Santa Cruz Biotechnology, sc-40, 1:500) and Alexa 568-conjugated mouse secondary antibody (Thermo Fisher, A-11011, 1:500). The cells were then imaged on a Leica DM5500 B upright microscope, capturing ten fields of view from each of three experiments, with the mean intensity from the ten fields/experiment plotted as data points in Fig. 5B. Representative fields are shown. Cells not expressing SMO-Myc or not incubated with BODIPY-CPA were used for background subtraction for both BODIPY and Alexa 568 channels. Data are presented as background-subtracted BODIPY intensity and Alexa 568 intensity for SMO-Myc-expressing cells.

Experiments were performed at least three independent times. Data were analyzed with GraphPad Prism 8. Graphs show mean values and error bars indicate s.d. Statistical analysis of the data was performed using two-tailed unpaired Student's t-test, ordinary one-way ANOVA with Tukey's multiple comparison test, or two-tailed Fischer's exact test as noted. Statistical significance was classified as *P<0.05, **P<0.01 or ***P<0.001. A P-value of <0.05 was considered statistically significant.

The authors dedicate this paper to the memory of Suzanne Eaton. We thank Rune Toftgard for providing Ptch1^−/− and Sufu^−/− MEFs, Lakshmi Devi for providing Neuro-2a cells, Hungwen Chen for expression vectors, and Paul Wassarman and Phil Soriano for critical comments on the manuscript.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199585