A functional genetic screen identifies aurora kinase b as an essential regulator of Sox9-positive mouse embryonic lung progenitor cells

Unravelling the mechanisms that govern embryonic lung stem cell self-renewal and differentiation is crucial for understanding the developmental processes that can go awry in lung development, and could reveal potential targets for the acceleration of lung maturation in pre-term infants. Lung development involves a series of coordinated proliferative and differentiation processes mediated by epithelium-mesenchyme interactions. Mouse lung development starts with the elongation of the tree-like branching network from the primordial lung buds [pseudoglandular stage: embryonic day (E) 9.5-E15.5; canalicular stage: E15.5-E16.5], followed by differentiation of the cells at the tip of the branches into saccules that will form the mature alveolar units where gas exchange occurs [saccular stage: E16.5 to perinatal; alveolar stage: perinatal to postnatal day (P) 30]. These four stages are conserved in mouse and human, although there are variations in the timing and localisation of these processes (Nikolić et al., 2018).

Formation of the branching network in the embryonic lung is driven by signalling from the mesenchyme to the endoderm, notably through Fgf10 and Wnt (Bellusci et al., 1997; Volckaert and De Langhe, 2015; Yuan et al., 2018), as well as mechanical forces applied by lung liquid and fetal breathing movements (Tang et al., 2018). Lung stem cells present at the distal tip of the lung embryonic endoderm receive these signals and drive branching morphogenesis (Chang et al., 2013; Rawlins et al., 2009a). They express the transcription factors Sox9 (sex-determining region Y-related high mobility group-box 9) and Id2 (inhibitor of DNA binding 2). As the branching tree grows distally, proximal cells in the airways downregulate Sox9 and upregulate Sox2, specifying them towards the airway lineage (club, ciliated and secretory cells). The appearance of Sox2 is the first indicator of bronchiolar cell fate specification (Alanis et al., 2014). Once elongation of the branches ends, Sox9⁺ cells downregulate Sox9 and differentiate into alveolar type 1 (AT1) and alveolar type 2 (AT2) cells to form the mature alveolar sacs (Alanis et al., 2014; Rawlins et al., 2009a). Conditional deletion of Sox9 in the lung endoderm using Shh-cre leads to defects in lung branching morphogenesis and death at birth due to breathing difficulties (Chang et al., 2013; Rockich et al., 2013). The dividing airway tree is reduced in these animals, although some branching occurs, indicating that branching is not driven exclusively by a transcriptional programme regulated by Sox9. More studies are required to identify the cell-intrinsic factors that regulate the multipotent and proliferative state of Sox9⁺ cells to drive the formation of the branching tree before lineage specification. This gap of knowledge is likely due to the difficulties in systematically interrogating gene functions in a high-throughput assay that mimics the stem cell state of Sox9⁺ cells present in the early lung primordia.

Functional genomic screens have been extensively used in many different organs to identify regulators of tissue-specific stem cells (Deneault et al., 2009; Kinkel et al., 2015; Sheridan et al., 2015). These large-scale functional screens require the combination of tools that can efficiently alter large numbers of genes together with an assay that can specifically evaluate the effect of these alterations. Short hairpin RNAs (shRNAs) are common tools used to perturb the expression of large numbers of genes in pooled genetic screens, presenting the advantage of inducing knockdown of target genes for the identification of genes playing a crucial role in developmental processes and stem cell regulation (Chen et al., 2012; Deneault et al., 2009; Sheridan et al., 2015). The second key requirement for a functional screen is the development of a robust cellular assay that closely mimics in vivo cellular properties. In vitro sphere assays have been used in many systems, including the lung, to maintain progenitor cell capacity and reduce cell differentiation (Chimenti et al., 2016; Reynolds and Weiss, 1996; Sheridan et al., 2015). They also constitute an ideal system in which to study early lung progenitors that are difficult to manipulate in vivo. Here, we established in vitro lung pneumosphere cultures that recapitulated the phenotype of early Sox9⁺ lung embryonic progenitor cells. We demonstrated that these cells were amenable to genetic and chemical manipulations and performed a boutique shRNA library screen targeting 130 genes encoding factors regulating transcriptional regulation and proliferation. We identified Aurkb (aurora kinase b), a kinase involved in multiple mitotic processes (Hindriksen et al., 2017), as a key regulator of Sox9⁺ cell proliferation. Conditional knockout of Aurkb in the lung endoderm using Shh-cre reporter mice and in Sox9⁺ cells using the Sox9-cre^ERT2 mouse line confirmed that Aurkb was redundant for the proliferation and development of some embryonic tissues but required for lung branching morphogenesis. Interestingly, loss of Aurkb did not affect proximal airway lineage specification and regulation of Sox2⁺ cell differentiation, but did affect distal lung progenitor cells, without altering cell polarity.

We first established a culture system in which lung progenitor cells were maintained and amenable to retroviral infection. The pneumosphere culture was developed based on 3D culture of stem and progenitor cells assays, such as neurospheres and mammospheres (Reynolds and Weiss, 1996; Sheridan et al., 2015) (Fig. 1A). Epithelial cells isolated from E11.5 lungs cultured in vitro formed non-clonal spheroid structures (Fig. S1A) that were highly proliferative and could be maintained in culture for 10 days without detection of cleaved-caspase 3-positive apoptotic cells at each passage (Fig. 1B). Immunohistochemical analysis of the pneumospheres revealed that they were enriched for early lung progenitor cells and maintained this phenotype over time, as demonstrated by the expression of Sox9 throughout the 10 days culture (Fig. 1C). Expression of Sox2, a marker of airway progenitor cells, was observed early in the culture but was rapidly downregulated (Fig. 1C). To characterise the pneumospheres further, we performed RNA sequencing (RNAseq) on cells harvested at day 3, day 7 and day 10. A multidimensional scaling (MDS) plot revealed that samples clustered together by the number of days in culture rather than by experimental replicates (Fig. S1B). The expression of embryonic lung progenitor markers, such as Nkx2-1, Sox9, Id2, Tgfb2, Bmp4 and Etv5, was stable across different time points (Fig. 1D), indicating that this 3D culture condition maintains the progenitor phenotype of the cells (Morrisey and Hogan, 2010). Consistent with the immunostaining results, Sox2 was downregulated from day 7 (Fig. 1E). Although no cell death was detected in the spheres, it is not clear whether Sox2⁺ cells died in culture and we failed to capture them by immunostaining, or if they differentiated to a Sox9⁺ phenotype. We then evaluated the expression of known markers of lung cell differentiation and observed low expression levels of markers of ciliated (Foxj1), basal (Trp63), neuroendocrine (Ascl1), goblet (Spdef), bronchial (Scgb3a2) or AT2 (Sftpc) cells (Fig. 1D), further showing that cells in pneumospheres are maintained in an undifferentiated state. Recently, Nichane et al. described a similar in vitro culture system of Sox9⁺ progenitor cells (Nichane et al., 2017). To confirm that our culture conditions enriched for Sox9⁺ progenitor cells, we compared the gene expression profile of cells in pneumospheres with the gene signature of Sox9⁺ progenitor cells previously described by Nichane et al. (2017). Gene set enrichment analysis revealed that cells harvested at each time point have a gene expression profile that strongly correlated with the gene signature of Sox9⁺ progenitor cells from the Nichane study (P-values: 1×10⁻¹⁹ for day 3, 1×10⁻¹⁵ for day 7, 2×10⁻¹³ for day 10; Fig. S1C). In addition, when cells in pneumospheres were cultured in conditions to induce cell differentiation, we detected cells of the alveolar lineages (type 1 and type 2) as well as some keratin 5-positive basal cells, indicating that Sox9⁺ cells in pneumospheres maintained their differentiation capacity (Fig. S1D). Together, these results indicate that pneumosphere cultures of E11.5 lungs are enriched with Sox9⁺ progenitor cells.

To evaluate whether the pneumosphere culture was a reliable method to study regulators of Sox9⁺ progenitor cells, we assessed the effect of known regulators of lung developmental pathways on proliferation and survival. Fibroblast growth factor 10 (Fgf10) is essential during lung development, particularly in branching morphogenesis where Fgf10, expressed in the lung mesoderm, signals to its receptors, Fgfr1/2, in the endoderm to promote growth of Sox9⁺ progenitor cells (Bellusci et al., 1997; Yuan et al., 2018). BGJ398 is a pan-FGFR inhibitor (Guagnano et al., 2011) currently in clinical trials for the treatment of FGFR1-amplified breast and lung cancer (Nogova et al., 2017; Weeden et al., 2015). Treatment of pneumospheres with increasing concentrations of BGJ398 in vitro dramatically reduced cell proliferation (Fig. 1F). To evaluate whether genetic modification of the cells in pneumospheres would mimic in vivo genetic alterations, we used shRNA to knock down the expression of Yy1, a known regulator of early lung morphogenesis (Boucherat et al., 2015). We knocked down Yy1 using two validated Yy1 shRNAs (Majewski et al., 2010) then measured how well these or non-silencing (nonS) control cells competed with the untransduced cells over 10 days in culture. As the virus contained a blue fluorescent protein (BFP) cassette, this was easily achieved by measuring the percentage of BFP⁺ cells during the 10-day culture period. Yy1 knockdown did not affect cell survival but resulted in reduced competitiveness of the Sox9⁺ progenitor cells (Fig. 1G), as expected for a gene for which depletion would alter the fitness of lung epithelial cells (Boucherat et al., 2015). Conversely, loss of the tumour suppressor gene Trp53 (Donehower et al., 1992) resulted in increased competitiveness of Sox9⁺ progenitor cells in pneumospheres (Fig. S1E). Overall, these results demonstrate that the pneumosphere culture system enriches for Sox9⁺ progenitor cells and that competition assays in this system can be used to identify genes that affect proliferation and survival of Sox9⁺ progenitor cells.

Having established a robust culture method for early lung epithelial progenitor cells, we conducted a pooled shRNA screen to identify novel regulators of lung development, in triplicate harvests of Sox9⁺ progenitors. A boutique library of 1057 shRNAs housed within a retroviral backbone, targeting 130 genes that might play a role in transcriptional regulation, proliferation and cell fate specification, was used (Keniry et al., 2016; MacPherson et al., 2019) (Table S1, Fig. S2A). We included 15 negative control nonsilencing shRNAs and 23 positive control shRNAs targeting known genes controlling lung development: Hdac1, Hdac3 and Carm1 (O'Brien et al., 2010; Wang et al., 2016, 2013). The retroviral backbone contained a puromycin resistance cassette along with the BFP mentioned earlier. We transduced the Sox9⁺ progenitors on the day of harvest at a low multiplicity of infection, then selected transduced cells with puromycin. Cells were harvested at day 3, day 7 and day 10 post-transduction. shRNA integrants were amplified in each sample by barcoded PCR and amplicon sequencing was used to determine the change in representation of each individual shRNA over time (Fig. 2A). We performed a differential representation analysis over the time course to obtain a list of shRNAs ranked by P-values, where the slope was used to indicate the hairpins that showed the greatest change between day 3 and day 10 harvests (Table S2). We filtered the list to identify the top-ranked shRNAs that had an absolute logFC greater than 0.2 (Fig. 2B, Table S2), which revealed 37 shRNAs targeting 33 genes. This approach identified hairpins against our three positive control genes, Hdac1 (two shRNAs) (Wang et al., 2013), Hdac3 (Wang et al., 2016) and Carm1 (O'Brien et al., 2010), demonstrating the validity of our hit identification. None of the 15 non-silencing shRNAs used as negative controls appeared in the top-ranked hairpin list. Known regulators of lung development were excluded from further analyses, as were shRNAs targeting genes that had already been shown to have little effect on lung development based on mouse knockout studies (nine shRNAs, eight genes: Prdm9, Mlh3, Chd6, Alkbh2, Alkbh3, Jmjd1c, Jmjd4, Prdm2) (Hayashi et al., 2005; Lathrop et al., 2010; Lipkin et al., 2002; Ringvoll et al., 2006; Steele-Perkins et al., 2001; Yoo et al., 2016; Zhu et al., 2016). Genes that were expressed at a low level in E11.5 RNAseq data were also removed from future validation experiments (B230120H23Rik/Map3k20, Chd3, Nup153), reducing the list of genes of interest to 19 genes targeted by 21 hairpins. All 19 genes were expressed at high levels in E11.5 lung epithelial cells and in pneumospheres (Fig. 2C) and their expression did not change during culture (Fig. S2B).

We validated individual shRNAs used in the screen by assessing knockdown efficiency of each of the remaining 21 hairpins in mouse embryonic fibroblasts (MEFs). Seven shRNAs, targeting six genes – Aurkb, Sae1, Rnf20, Mettl8, Smc2 and Ruvbl1 – induced a significant knockdown of their target (Fig. 2D), whereas 14 did not induce significant knockdown (Fig. S2C). To determine whether knockdown of these six genes of interest would affect cell proliferation in a competitive pneumosphere assay, we set up competition assays exactly as performed for Yy1 and Trp53 knockdowns previously. Of the seven hairpins that significantly knocked down their target gene, all except the shRNA against Mettl8 had a significant effect on the competitive behaviour, recapitulating the results of the screen (Fig. 2E).

In summary, from the 130 genes targeted in the pooled shRNA screen, we identified five genes (Aurkb, Sae1, Rnf20, Smc2 and Ruvbl1) that are candidate regulators of early lung Sox9⁺ progenitor cells.

All five hits represent exciting potential regulators of lung progenitors. We decided to begin by studying the expression and role of Aurkb in the embryonic lung and after birth. Analysis of publicly available gene expression profiles of mouse embryonic and postnatal lung (Beauchemin et al., 2016) revealed that Aurkb was highly expressed from E9.5 until the saccular stage of lung development (E16.5) when its expression started to decrease until birth (Fig. 3A). Another wave of Aurkb expression occurs postnatally until approximately P9 when expression decreased again and remained at lower levels until adulthood (Fig. 3A). Interestingly, the expression pattern of Aurkb closely mimicked the expression of the proliferation marker Ki67 (Mki67), consistent with the role of Aurkb in proliferation and mitosis (Hindriksen et al., 2017). The expression of Aurka also followed a similar pattern whereas expression of Aurkc was low at all time points (Fig. S3A). Aurka immunohistochemistry confirmed high levels of expression in the embryonic lung at E13.5 and E14.5 in both the endoderm and the mesoderm, and lower levels of expression at E15.5 (Fig. S3B). To evaluate which cells in the embryonic lung expressed Aurkb, we performed immunohistochemistry analysis in lung tissue at E13.5, E15.5 and E17.5. Similar to the RNA expression data, we observed high expression of Aurkb at E13.5 and E15.5 in both the mesenchyme and the endoderm (Fig. 3B). At E17.5, the expression of Aurkb appeared dramatically reduced and only a small number of E-cadherin (cadherin 1)-positive epithelial cells were stained with Aurkb (Fig. 3B). At E13.5 and E15.5, staining of adjacent sections with Sox9 revealed that Aurkb was expressed in Sox9-positive cells, albeit not exclusively (Fig. 3B). Expression of Aurkb in Sox9⁺ cells was high at E13.5 and E15.5 and not detected at E17.5 when Sox9⁺ cells have committed to an alveolar phenotype. These results suggest that Aurkb may play a role in regulating Sox9⁺ embryonic lung early progenitor cells.

The RNA expression analysis of Aurkb indicated it may play a role during alveoli differentiation and/or airway elongation postnatally. We therefore generated Sftpc-cre^ERT2;Aurkb^fl/fl and Scgb1a1-cre^ERT2; Aurkb^fl/fl mice and evaluated the effect of Aurkb deletion in the first weeks after birth. In both cases, we did not observe any overt lung defects, suggesting that loss of Aurkb may be redundant during postnatal lung growth and differentiation (Fig. S4A,B). We also used these models to evaluate the role of Aurkb in adult lung proliferation and repair following injury. Scgb1a1-cre^ERT2;Aurkb^fl/fl mice did not display any defect in airway repair following naphthalene injury (Fig. S4C). Expression of the club cell marker CC10 (Scgb1a1) was similar in both genotypes at day 6 post-injury, indicating that repair processes are in place even in the absence of Aurkb (Fig. S4D,E). Analysis of the expression of Aurkb in naphthalene-injured lung further showed that Aurkb was not expressed in proliferating progenitor cells activated following injury (Fig. S4F). In contrast, Aurka was detected in Ki67⁺ proliferating cells, suggestive of a possible role for Aurka, but not Aurkb, in adult lung progenitor cell proliferation (Fig. S4G). Overall, these analyses reveal that Aurkb does not appear to regulate lung postnatal growth and adult lung progenitor cell proliferation in response to naphthalene injury.

Aurkb knockout mice implant normally but die in early post-implantation stages (Fernández-Miranda et al., 2011). Aurkc likely compensates for loss of Aurkb pre-implantation but downregulation of Aurkc after implantation leads to lethality (Fernández-Miranda et al., 2011). Use of conditional knockout mice is therefore necessary to examine the role of Aurkb in organ-specific development. We generated Shh-cre;Aurkb^fl/fl mice (Aurkb^Shh-cre hereafter) and compared them with Aurkb^fl/fl or Shh-cre;Aurkb^fl/+ mice. Given that Aurkb^fl/fl or Shh-cre;Aurkb^fl/+ were phenotypically identical, we used either as our control animals to reduce animal use (Aurkb^Ct hereafter). Shh is expressed in the ventral foregut endoderm at E9.5 and expression of cre driven from this locus causes excision of exons 2-6 in the Aurkb gene in the endoderm of the primary lung bud (Fernández-Miranda et al., 2011). Shh is also expressed in the embryonic notochord, floor plate and limb, indicating that use of this cre reporter mouse model could lead to additional defects outside the lung (Echelard et al., 1993). Aurkb^Shh-cre animals did not survive at birth. We noticed that the mutant pups were smaller than Aurkb^Ct pups and displayed a shorter trunk, and a hypomorphic tail was observed from E13.5 (Fig. 4A). This phenotype prompted us to investigate the skeletal structure in Aurkb^Shh-cre embryos. Staining of bones and cartilage at E17.5 demonstrated absence of tail bones and a reduction in the number of lumbar vertebrae (Fig. S5A). This vertebral defect was not accompanied by a neural tube closure defect, but implies a possible role for Aurkb in somite formation that will be investigated in future studies. However, other highly proliferative organs in which Aurkb is excised with the Shh-cre reporter, such as the limbs (Echelard et al., 1993), were not affected by loss of Aurkb (Fig. S5B,C), indicating that Aurkb is not required for the normal development of all embryonic organs.

Strikingly, no lungs were detected by histological analysis of Aurkb^Shh-cre embryos at E16.5 and E13.5 (Fig. 4B). Additional analysis of E18.5 whole embryos by micro-CT imaging confirmed the absence of lungs in Aurkb^Shh-cre embryos at this time point (Fig. 4C, Fig. S5D). We then performed histology on E11.5 embryos and could detect only a rudimentary lung bud in the Aurkb^Shh-cre embryos (Fig. 4B). Nkx2-1 was detected in the remnant of the trachea at E11.5 (Fig. 4D), indicating that separation of the trachea from the oesophagus and lung specification occurred. To characterise the hypomorphic lung further, we performed 3D imaging using light-sheet microscopy. Staining of E11.5 dissected lungs with the lung epithelial markers Nkx2-1 and E-cadherin revealed the presence of primitive epithelial structure in Aurkb^Shh-cre lungs compared with the branched structure already present in Aurkb^Ct lung (Fig. 4E,F). In wild-type mice, Sox9⁺ progenitor cells are present at the tip of the branching lung buds and detectable from E11.5 (Alanis et al., 2014). We therefore analysed the expression of Sox9 in Aurkb^Ct and Aurkb^Shh-cre lungs. Sox9⁺ cells were clearly observed at the tip of the branches in control animals, but were completely absent in Aurkb^Shh-cre lungs (Fig. 4E), suggesting a role for Aurkb in the specification of these lung progenitor cells that are essential for lung branching morphogenesis (Chang et al., 2013). Expression of Sox9 was observed in the precursors of the cartilage of the trachea in both control and conditional knockout mice (Fig. 4E), consistent with the known role for Sox9 in tracheal cartilage formation (Turcatel et al., 2013). Ex vivo culture of E11.5 lungs in an air-liquid interface further demonstrated absence of branching in Aurkb^Shh-cre lungs cultured for 2 days, whereas Aurkb^Ct lungs formed an extensive branching tree (Fig. 4G). Altogether, these results indicate that loss of Aurkb in the early lung endoderm completely abrogated lung development, indicative of an essential role for Aurkb in lung formation.

To investigate more specifically the role of Aurkb in Sox9⁺ lung progenitor cells in vivo, we generated Sox9^creERT2;Aurkb^fl/fl mice (Aurkb^Sox9-creER hereafter). Expression of Sox9 is detected in the nascent lung buds in the anterior foregut from E10.5 (Alanis et al., 2014). Pregnant females were therefore injected with tamoxifen at E7.5 and E9.5 to induce cre-mediated recombination and Aurkb deletion as soon as Sox9 expression is induced in the lung primordia. Macroscopic evaluation revealed that Aurkb^Sox9-creER embryos were smaller than their wild-type or heterozygous littermates, and presented with a cleft palate and cranial abnormalities (Fig. S6A-C). Analysis of the skeleton of the embryos showed absence of bone mineralisation in the spine and defects in rib formation (Fig. S6C,D), reminiscent of the chondrodysplasia observed in the Col2a1-cre;Sox9^fl/fl mice (Akiyama et al., 2002). These data suggest that Aurkb is required for the normal development of many Sox9-expressing cell types.

To confirm cre-mediated excision in the lung buds, we crossed Aurkb^Sox9-creER mice with mTmG^fl/+ mice and evaluated GFP expression as a marker of cre recombination. Co-staining of E14.5 lungs with GFP and Sox9 showed colocalisation of the two markers with GFP expression in the tip of the primary lung buds (Fig. S6E, Movie 1), as well as the precursors of the cartilage rings of the trachea as previously described (Park et al., 2009). Aurkb^Sox9-creER E14.5 lungs were smaller than Aurkb^Ct lungs (Fig. 5A) and analysis of Haematoxylin & Eosin (H&E)-stained sections showed a reduced number of branches with dilated airways (Fig. 5B). Expression of Aurka was not altered in Aurkb^Sox9-creER lungs (Fig. S6F). In normal lung morphogenesis, Sox9⁺ cells drive branching and proliferation at the distal tip of the branches between E10.5 and E15.5, whereas Sox2 expression is turned on in proximal cells and induces a commitment to an airway cell phenotype (Alanis et al., 2014). To understand whether loss of Aurkb was affecting Sox9⁺ cells at the tip of the branches or interfering with Sox2 expression and airway cell specification, we performed whole-mount staining of E14.5 control and mutant lungs and imaged them with 3D light-sheet microscopy (Fig. 5C, Movie 2). The epithelial marker E-cadherin delineated the epithelium in both Aurkb^Ct and Aurkb^Sox9-creER lungs, although the number of cells and intensity of the staining was reduced in Aurkb^Sox9-creER lungs. Expression of Sox9 was detected at the tip of the branches in all genotypes, but the Sox9-positive region in Aurkb^Sox9-creER lungs was enlarged and cells were organised in a multi-layered structure instead of the normal pseudostratified epithelium observed in control lungs (Fig. 5C, optical sections). Staining with both markers demonstrated a clear defect in branching morphogenesis, where the airway tree was minimal in Aurkb^Sox9-creER lungs compared with their wild-type counterparts. Aurkb and the chromosomal passenger complex (CPC) are involved in apical-basal cell polarisation during cytokinesis and lumen formation in epithelial organs (Capalbo et al., 2019; Overeem et al., 2015). However, cell polarity was not affected by the loss of Aurkb, as evidenced by PKCζ (a marker of apical polarity) staining (Fig. S6G). Formation of a lumen was maintained in the distal lung buds of Aurkb^Sox9-creER embryos, indicating that Aurkb loss in Sox9⁺ cells does not affect lumen formation, but rather alters airway bifurcation. Sox2 staining showed that proximal branches of Aurkb^Sox9-creER lungs expressed Sox2, demonstrating that airway lineage differentiation occurred, and primary and secondary bronchi had formed, but that formation of tertiary bronchi and bronchioles was drastically reduced. To confirm the defect in branching morphogenesis, we isolated E12.5 lungs from Aurkb^Ct and Aurkb^Sox9-creER embryos from pregnant females injected with tamoxifen at E7.5 and E9.5 and cultured them ex vivo (Fig. 5D). Quantification of branching tips showed a reduced number of branches in Aurkb^Sox9-creER lungs compared with control lungs (Fig. 5E). Immunofluorescence staining of ex vivo lung culture showed similar data to light-sheet 3D imaging; enlarged Sox9⁺ airway sacs were observed at the tip of the branches and Sox2⁺ cell differentiation occurred normally in the proximal airways (Fig. 5F). Of note, formation of the cartilaginous ring around the tracheas was also perturbed (Fig. 5C), indicative of a role for Aurkb in Sox9⁺ cartilaginous precursor cells.

Overall, these results suggest that loss of Aurkb in Sox9⁺ cells impairs branching bifurcation, resulting in enlarged distal tips and an incomplete branching network that is required for normal lung morphogenesis.

Aurkb is part of the chromosomal passenger complex along with inner centromere protein (Incenp), borealin (Ccda8) and surviving (Birc5), where it regulates key mitotic events (Carmena et al., 2012). This known role of Aurkb in mitosis and cytokinesis prompted us to evaluate how disruption of Aurkb function affected Sox9⁺ cell proliferation. Treatment of Sox9⁺ cells in pneumospheres with the Aurkb-specific inhibitor AZD1152 (de Groot et al., 2015) resulted in a reduced number of cells (Fig. S7A). Treatment of E11.5 lungs in an ex vivo air-liquid interface culture with AZD1152 (de Groot et al., 2015) completely abrogated lung branching, without reducing the expression of Ki67 (Fig. 6A). Given that Ki67 marks all proliferating cells, this result suggested that cells may be arrested in the cell cycle. We therefore performed a more detailed analysis of the cell cycle in Sox9⁺ cells and evaluated how loss or inhibition of Aurkb affected the transition of pneumosphere cells through the cell cycle. Hairpin knockdown of Aurkb in pneumosphere culture resulted in more than 80% knockdown of Aurkb expression (Fig. S7B) and induced an accumulation of cells in the G2/M phase of the cell cycle (Fig. 6B,C). Similarly, inhibition of the kinase activity of Aurkb with AZD1152 showed a dose-dependent increase of cells in G2/M, also associated with an increased percentage of apoptotic cells (Fig. 6D,E).

Overall, these results indicate that Aurkb mediates progression of Sox9⁺ early lung progenitor cells through the cell cycle, and this activity is required for branching morphogenesis.

Sox9⁺ cells present at the tip of the embryonic lung buds play a crucial role in normal lung development as early progenitor cells responsible for branch bifurcation and elongation from E10.5 to E15.5. In the saccular phase of lung development (E16.5-P3), Sox9⁺ cells differentiate to form the alveolar unit necessary for the first breath at birth. However, little is known about the cell-intrinsic mechanisms maintaining progenitor activity in Sox9⁺ cells and the switch to alveolar lineage commitment at the end of embryonic lung morphogenesis, likely due to the technical challenges isolating Sox9⁺ cells from early lung buds. Here, we performed a functional genetic screen to identify intrinsic regulators of early progenitor activity in Sox9⁺ cells. We identified Aurkb as a crucial regulator of Sox9⁺ cells, required for branch bifurcation in the earliest phases of lung development.

Our genetic screen targeting 130 genes involved in transcriptional regulation and proliferation identified eight genes with a possible role in lung development. Three of these genes had previously been described as important for lung morphogenesis: Hdac1, Hdac3 and Carm1. Histone deacetylase 1 (Hdac1) has been shown to play a role in early lung progenitor cells; loss of Hdac1 hindered normal branching morphogenesis and blocked Sox2 expression necessary for proximal airway specification, leading to the formation of aberrant non-functional lungs (Wang et al., 2013). Loss of Hdac3, by contrast, did not affect the early lung endoderm cells, but rather altered AT1 cell specification in late embryonic lung development (Wang et al., 2016). Coactivator-associated arginine methyltransferase 1 (Carm1), also known as protein arginine methyl transferase 4 (PRMT4), was shown to play a role in the late phase of alveolar cell maturation, with Carm1 knockout lungs characterised by hyperproliferative AT2 cells and lack of AT1 cells, indicative of a role for Carm1 in Sox9⁺ cell maturation in the distal lung (O'Brien et al., 2010). Identification of these genes indicate that our screen could not only identify regulators of early progenitor activity in Sox9⁺ cells, but also genes involved in the regulation of late progenitor cells specified to the alveolar lineages (Frank et al., 2019). Conversely, Ezh2, a known regulator of lung development (Galvis et al., 2015; Snitow et al., 2015), was not identified in this screen, consistent with the role of Ezh2 in airway lineage differentiation but not in Sox9⁺ cells (Galvis et al., 2015; Snitow et al., 2015). These results demonstrate the specificity of our assay in identifying genes involved in regulating Sox9⁺ cells, either in early lung branching morphogenesis or during alveolar cell specification in late embryogenesis. As with any genetic screen, it is, however, possible that we missed the identification of other master regulators of Sox9⁺ cells, given that the knockdown efficiency of each individual shRNA present in the shRNA library was not validated and some shRNAs may not have induced efficient knockdown. Nevertheless, this shRNA screen in pneumospheres led to the identification of five putative novel regulators of early lung progenitor cells.

Of all shRNAs inducing a reduction in cell representation in the pneumosphere assay, shRNA targeting Aurkb had the most profound effect, reducing cell numbers by more than 70%. We took advantage of four different conditional knockout mouse models to evaluate the role of Aurkb in early lung development, postnatal growth and repair in response to injury during adulthood. Our data show that Aurkb plays a role in the embryonic phase of lung morphogenesis but did not participate in lung growth or repair after birth, indicating that Aurkb is redundant in mediating proliferation of adult lung progenitor cells, at least in the context of naphthalene-induced lung damage. This result contrasts with what has been described in another branching organ, the salivary gland, where Aurkb is required for regeneration of the adult gland and acini formation after injury (Shaalan and Proctor, 2019). Both Aurka and Aurkb are upregulated in mitotic cells, whereas Aurkc plays a role exclusively during meiosis in germ cells. However, there is little evidence for a complete compensatory mechanism by Aurka for Aurkb loss owing to their distinct localisation on the mitotic spindle (Hochegger et al., 2013; Li et al., 2015; Willems et al., 2018). Aurka may compensate for Aurkb catalytic activity during mitosis but not for its role in cytokinesis (Rannou et al., 2008). Our results suggest that Aurka is likely to play an important role in the regulation of adult lung progenitor cell proliferation, but further studies in lung progenitor cells will be required to understand this fully.

The use of two distinct embryonic cre-recombinase mouse models demonstrated the key role played by Aurkb in the lung primordia. Loss of Aurkb in the primary lung endoderm using Shh-cre completely aborted lung formation, indicating that Aurkb is instrumental for the differentiation and proliferation of progenitor cells present at the tip of the lung bud. Expression of Sox9 in these progenitor cells is crucial for lung branching morphogenesis (Chang et al., 2013; Rockich et al., 2013), and Sox9 expression was abrogated after loss of Aurkb in Aurkb^Shh-cre embryos. Use of Aurkb^Sox9-creER mice was therefore necessary to investigate specifically the role of Aurkb in Sox9⁺ cells present at the tip of the lung bud. Tamoxifen activation of the cre reporter facilitated the temporal control of cre-mediated recombination and induction of Aurkb excision after Sox9⁺ cells had formed. The expression of Sox9 was not altered in Aurkb-deficient cells, yet lung branching morphogenesis was dramatically reduced, indicating that Aurkb is necessary for the proper function of Sox9⁺ cells in the embryonic lung. Both Aurkb^Shh-cre and Aurkb^Sox9-creER mice displayed skeletal defects. Shh is expressed in the embryonic notochord, floor plate and limbs (Echelard et al., 1993). Sox9 is expressed in embryonic precursor cells in multiple organs, including chondroprogenitors and differentiated chondrocytes (Akiyama et al., 2002) and plays a crucial role in craniofacial development and chondrogenesis (Lee and Saint-Jeannet, 2011). Although the role of Aurkb has not been explored in skeletal development, our results are indicative of a role for Aurkb in both skeletal patterning and bone formation, but not limb development. This suggests that Aurkb can have tissue-specific roles in embryonic development.

Aurkb is involved in many different processes throughout mitosis. In somatic cells, Aurkb functions primarily as part of the CPC. As a cell enters mitosis, the C-terminal domain of Aurkb regulates its localisation, targeting Aurkb to the centromeres at prophase where it forms a complex with the scaffold protein inner centromere protein, borealin and survivin to form the CPC (Hindriksen et al., 2017). Aurkb is the key catalytic component of the CPC, which is involved in chromosomal condensation, cohesin removal from chromosome arms during prophase, detection of erroneous kinetochore to microtubule attachments, regulation of the mitotic checkpoint, shortening of segregated chromosomes, and cytokinesis (Carmena et al., 2012). Cytokinesis mediates the physical separation of dividing cells, and in 3D epithelia participates in the positioning of the apical surface of cells for lumen formation and cell-fate decisions (Capalbo et al., 2019; Dionne et al., 2015; Jaffe et al., 2008; Overeem et al., 2015). Aurkb^Sox9-creER lungs formed a lumen at the tip of the terminal buds, indicating that apical-basal polarity is not perturbed in Aurkb-deficient lungs. However, branch bifurcation was dramatically impaired. FGF signalling is an essential regulator of branching morphogenesis (Bellusci et al., 1997; Jones et al., 2019), yet treatment of Aurkb-deficient lung ex vivo with Fgf10 did not rescue the branching pattern (Fig. S8), indicating that other mechanisms are at play in Aurkb-deficient Sox9⁺ cells. We observed that upon chemical inhibition or downregulation of Aurkb, Sox9⁺ cells were blocked in G2/M phase of the cell cycle. Cell cycle arrest has previously been linked to reduced branching morphogenesis (Yao et al., 2017) and is likely the mechanism at play here. Given the important role of cytokinesis in asymmetric cell division and in the architecture of epithelial tissue (Herszterg et al., 2014), it is also possible that loss of Aurkb perturbed cytokinesis in Sox9⁺ cells leading to disrupted cellular bifurcation (Andrew and Ewald, 2010; Wang et al., 2017). Sox9⁺ cells are also present at the tip of branches in other branching organs during development, such as the kidney (Reginensi et al., 2011) and the salivary gland (Chatzeli et al., 2017). In these developing tissues, loss of Sox9 disrupts branching morphogenesis (Chatzeli et al., 2017; Reginensi et al., 2011). It would be interesting to evaluate how loss of Aurkb affects branching morphogenesis in these organs.

Functional genetic screens are a powerful tool for elucidation of the genetic regulators of biological processes. To be successful, they rely on a strong model system and a robust genetic targeting library. The present shRNA screen in pneumospheres identified Aurkb as a potent regulator of early lung progenitor cells, as evidenced by conditional knockout studies. This study paves the way for the identification of other genetic regulators of Sox9⁺ progenitor cells identified in this screen, or for future similar strategies.

Shh-cre mice (Harfe et al., 2004) were purchased from The Jackson Laboratory. Sox9-cre^ERT2 (Soeda et al., 2010) mice were obtained from Dr Chen (University of Texas, MD Anderson Cancer Center, USA). Aukb^fl/fl mice (Fernández-Miranda et al., 2011) were obtained from the CNB Mouse Embryo Cryopreservation Facility (Spain) and Scgb1a1-cre^ERT2 mice (Rawlins et al., 2009b) from Professor Hogan (Duke University, NC, USA). All animal experiments were conducted according to the Walter and Eliza Hall Institute of Medical Research Animal Ethics Committee guidelines (AEC 2013.028; 2016.024). Male and female mice were used. Tail tips and embryonic tissue was digested for at least 2 h at 55°C in DNA lysis buffer with 200 μg/ml proteinase K (Roche). The PCR reaction was performed using 1X GoTaq Green Mix (Promega) and 0.5 µM per forward and reverse primers (Table S3). PCR cycle conditions for the Aurkb^fl allele were 94°C for 5 min; 35 cycles of 94°C for 30 s, 60°C for 30 s and 72°C for 1 min; 72°C for 10 min. PCR cycle conditions for the cre allele were 96°C for 5 min; 30 cycles of 94°C for 30 s, 55°C for 30 s and 72°C for 1 min; 72°C for 4 min.

Tamoxifen was delivered at a dose of 100 mg/kg for adult mice (Sigma-Aldrich). For Aurkb^Sox9-Cre mice, tamoxifen was injected to the pregnant dam at E7.5 and E9.5. For Aurkb^Sftpc-crER mice, pregnant dams received an intraperitoneal tamoxifen injection at E16 and E18 and pups were delivered by caesarean section. Foster mums were used before pup collection at P21. For postnatal growth studies, Aurkb^{Scgb1a1-CreER} pups received an intragastric injection of tamoxifen (50 µg per pup) at P1 and P3 before collection at P7 and P21. For naphthalene injury, three doses of tamoxifen were injected in adult male Aurkb^{Scgb1a1-CreER} or Aurkb^Ct mice (day 0, 2, 4) and naphthalene injection was delivered on day 14 post-tamoxifen. Naphthalene was dissolved in corn oil and injected intraperitoneally at 250 mg/kg; mice were collected at defined time points post-injury.

Lungs were dissected from E11.5 embryos and dissociated in 0.05% Trypsin-EDTA (Gibco). Cells were filtered and resuspended at 200,000 cells/ml/well and plated in pneumosphere media [PSM; DMEM-Ham's F-12+Glutamax (Gibco), 2% B27 supplement (Life Technologies), 1×ITS (Gibco), 20 ng/ml bFGF (Becton-Dickinson), Heparin (Sigma-Aldrich), 10 ng/ml EGF (Becton-Dickinson)] for 1 h in a 6-well regular tissue culture plate to allow for attachment of mesenchymal cells. Supernatant-enriched epithelial cells were harvested and cells were then resuspended at 50,000 cells/ml/well in PSM in a 24-well ultra-low attachment plate (Corning) in 10% CO₂, 5% O₂ at 37°C. Cells were split at day 3 and day 7 and re-plated at 50,000 cells/ml/well. For competition assays, epithelial-enriched E11.5 cells were transduced with BFP-shRNA-expressing retroviruses and plated in an ultra-low attachment plate in pneumosphere media without selection pressure. Each experiment included non-silencing (nonS) controls and test shRNAs. BFP expression was monitored by flow cytometry at day 3, 7 and 10. BFP expression was normalised to day 3 expression and compared with the nonS control. Cells were analysed on a BD LSRFortessa 1 or BD LSRFortessa X20.

Pneumosphere differentiation was based on previously published protocols (Nichane et al., 2017; Barkauskas et al., 2013). Briefly, E11.5 lungs were processed as per the pneumosphere protocol. Epithelial-enriched progenitor cells were mixed at a 2:5 ratio with adult EpCAM⁻ sorted mesenchymal cells in MTEC/Plus media [DMEM-Ham's F-12+Glutamax (Gibco), 15 mM HEPES (Gibco), 10 µg/ml ITS (Corning), 0.1 µg/ml cholera toxin (Sigma-Aldrich), 25 ng/ml epidermal growth factor (Becton-Dickinson), 30 µg/ml bovine pituitary extract (Corning), 0.01 M retinoic acid (Sigma-Aldrich), Rock inhibitor (STEMCELL Technologies)]. Cells were mixed at a 1:1 ratio with Matrigel (BD Biosciences) and plated in a 24-well 0.4 µM Transwell insert (Corning). MTEC/Plus media was added to lower chamber and media was changed every 2 days. Cells were grown for 16 days.

293T cells were cultured in DMEM (Life Technologies) supplemented with 10% fetal bovine serum (FBS) and grown at 37°C in a humidified atmosphere with 10% (v/v) CO₂. Retrovirus was produced using calcium phosphate-mediated transient transfection of 293T cells. 293T cells at 80% confluency were transfected with VSVg envelope vector, MD1-gag-pol structural vector and BFP-shRNA retroviral construct at a ratio of 24:8:1. For pooled shRNA screen virus, constructs were pooled in an equimolar fashion and this DNA pool was used to make virus. Otherwise, individual constructs were used. Media was refreshed 16 h post-transfection. Supernatant was harvested 48 h post-transfection, filtered (0.45 µM) and concentrated using Poly(ethylene glycol) (PEG). MEFs and pneumospheres were transduced with concentrated retroviral supernatant. MEFs were cultured in DMEM (Gibco) with 10% (v/v) FBS (Life Technologies) and viral supernatant added at 50-80% MEF confluence. For pneumospheres, viral supernatant was added to E11.5 lung cells plated on ultra-low attachment plates. Media was changed 24 h after addition of retroviruses. As required, cells were selected with 2 µg/ml of puromycin (Sigma-Aldrich).

E11.5 lungs were processed as per the pneumosphere protocol. Epithelial-enriched E11.5 cells were plated at 10,000 cells/well in 384-well, clear, flat-bottomed plates and treated with BGJ398 or AZD1152 (Selleckchem). Cell proliferation was determined 3 days later using a CellTiter 96 Aq_ueous One Solution Cell Proliferation Assay (Promega) and light absorbance read at 450 nM on the Wallac EnVision (Perkin-Elmer). For the bromodeoxyuridine (BrdU) incorporation assay, BrdU (1 mM/ml, BD Biosciences) was added to the culture media and left for 2 h. Cells were then processed using the APC BrdU Flow kit (BD Biosciences). Cells were analysed on a BD LSR Fortessa.

E11.5 lungs were dissected and placed at the air-liquid interface on a 0.8 µM membrane in a 12-well tissue culture plate in DMEM-F12 with Glutamax (Gibco), 100 U/ml penicillin and 100 μg/ml streptomycin. Lungs were treated for 3-4 days with DMSO, BGJ398 or AZD1152 (Selleckchem) at the indicated concentration. For staining, lungs were fixed with 4% paraformaldehyde (PFA) (Sigma-Aldrich), permeabilised in 0.5% Triton X-100 (Sigma-Aldrich) and blocked in 10% goat or donkey serum (Abcam) for 1 h at room temperature. Primary antibodies (Table S4) were incubated overnight at 4°C followed by secondary antibodies (Table S4) for 2 h at room temperature in the dark. DAPI (Thermo Fisher Scientific) was added at 1/1000 for 10 min. Membranes with lung were transferred to slides and mounted with Fluoromount-G (SouthernBiotech). Samples were imaged using a Zeiss LSM780 or Zeiss LSM880 NLO confocal microscope.

Slides were dewaxed using standard histology protocols. Antigen retrieval was performed using 10 mM pH 6 citrate buffer (0.1 M citric acid, 0.1 M sodium citrate, H₂O) in a DakoCytomation Pascal pressure chamber. Sections were blocked in 10% goat or donkey serum for 1 h and then incubated with antibodies overnight at 4°C followed by fluorophore-conjugated antibodies for immunofluorescence or horseradish peroxidase-conjugated secondary antibodies (Table S4) (Vector Laboratories) for immunohistochemistry for 1 h. DAPI was added for 10 min and slides were mounted with Fluoromount-G (SouthernBiotech) for immunofluorescence imaging. For immunohistochemistry, Vectastain ABC (Vector Laboratories) was added for 30 min at room temperature and developed with diaminobenzidine. Immunohistochemical slides were imaged with a Nikon Eclipse 50i inverted microscope or a Zeiss Axioplan inverted microscope. Immunofluorescence slides were imaged using a Zeiss LSM780 confocal microscope or a Zeiss LSM880 NLO confocal microscope. Quantification of CC10⁺ cells was performed in an automated manner using QuPath. Airways regions were drawn manually and cells from this region were detected with StarDist, a machine-learning object detection tool. The cut-off value was based on the distribution of mean intensities of the marker into the cytoplasm.

For light-sheet imaging, Aurkb^Shh-cre and Aurkb^Sox9-creER or control lungs were dissected but left attached to the head for easier tissue processing. Experimental tissue was fixed in 4% PFA (Sigma-Aldrich) for 1 h. Samples were permeabilised in PBS containing 0.5% Triton X-100 (Sigma-Aldrich) for 1 h at room temperature, and blocked by adding 10% goat or donkey serum for 1 h. Primary antibodies were added and incubated for 2 days at 4°C on a plate rocker. Samples were washed in PBS for 6 h, then secondary antibodies were added and incubated for 2 days at 4°C on a plate rocker. Primary antibodies and secondary antibodies are detailed in Table S3. After 6 h of washing in PBS, DAPI was added at 1/1000 for 30 min and the samples were re-fixed in PFA for 1 h. Samples were cleared in ScaleA2 clearing solution (4 M urea, 10% v/v glycerol, 0.1% v/v Triton X-100) (Hama et al., 2011) for 1-2 weeks, with solution exchanges every 3-4 days. Samples were mounted in 2% low melting point agarose in 1.5 mm glass capillaries and re-immersed in ScaleA2 solution for refractive index matching. Lungs were imaged using a Zeiss Z.1 Light-sheet microscope equipped with a 5×/0.16 objective. 3D reconstruction was performed in Imaris 9.3.1 (Bitplane).

Micro-CT tissue processing on Aurkb^Shh-cre and Aurkb^Ct E18.5 embryos was performed using the STABILITY protocol (Hsu et al., 2016; Wong et al., 2013). Briefly, E18.5 embryos were harvested and placed in 4% PFA (Sigma-Aldrich) for 3 days at 4°C. Embryos were then transferred to 20 ml of STABILITY buffer [4% w/v PFA, 4% w/v acrylamide (Bio-Rad), 0.05% bis-acrylamide (Bio-Rad), 0.25% w/v VA044 initiator (Wako Chemicals), 0.05% w/v Saponin (Sigma-Aldrich), in 1× PBS] for 3 days at 4°C and placed in a desiccator with nitrogen gas. Samples were then incubated at 37°C for 3 h to initiate crosslinking reaction. Once the hydrogel had formed, extra gel was removed and samples were placed into 0.1 N v/v iodine solution (Sigma-Aldrich) for 5 days, iodine solution was changed every 2 days. Samples were mounted in 10 ml tubes, imaged via the Skyscan 1276 Micro-CT scanner (Bruker) and reconstructed in Bruker Skyscan nRecon v1.7.3.1. Lungs were segmented and visualised in Imaris 8.1.

E11.5 lung epithelial cells were infected with pooled screen virus at a MOI of 2 to obtain 50-80% transduction efficiency, prior to plating at 50,000 cells/ml/well in an ultra-low attachment plate. Cells were left for 16 h and then media was replaced with fresh PSM. Cells were allowed to recover for 6 h before puromycin selection (2 µg/ml puromycin). Cells were collected at day 3, day 7 and day 10 for DNA extraction using the Blood and Tissue DNA extraction kit (Qiagen). shRNA libraries were prepared from 0.25 µg of DNA from each sample in a 50 µl total volume PCR reaction. The shRNA proviral integrants were amplified with primers annealing to common regions of the shRNAs and a unique barcode and Illumina sequencing adaptors. PCR reactions contained 0.25 µg DNA, 1× Phusion High Fidelity Buffer (NEB), 2 µl 10 mM dNTPs, 2 µl P7R individual primer, 2 µl P5F common primer (Keniry et al., 2016; MacPherson et al., 2019), 1 U Phusion DNA polymerase (NEB). Pooled libraries were sequenced on the Illumina Hi-Seq platform using 100 bp, single-end reads. Samples were quantified on a TapeStation using the D1K ScreenTape. Samples were pooled (50 ng each) for a total concentration of 200 ng and run on a 1.5% agarose-TAE gel. DNA (380 bp band) was extracted from the gel using the Gel Extraction Kit (GE Healthcare). shRNAseq library sample was submitted to the Australian Genome Research Facility for sequencing on the Illumina Hi-Seq platform using 100 bp, single-end reads at a depth of 35,000 reads/hairpin/sample from the 162 million total number of reads. Importantly, these samples were sequenced pooled with RNAseq libraries to ensure diversity in the sample.

Screen analysis was performed using the edgeR package (version 3.14.0) (Robinson et al., 2009). Raw sequencing data was analysed using the ‘processAmplicons’ function (Dai et al., 2014), and shRNAs with a count per million (CPM) >5 in at least three samples were retained for further analysis. A general linear model analysis was performed on the counts from each hairpin, which included effects for replicate number and day (3, 7 or 10) as a numeric covariate (slope). Using edgeR's quasi-likelihood method (Lun et al., 2016) with ‘robust=TRUE’ (Phipson et al., 2016) to test for changes slope over time, hairpins were ranked by their adjusted P-values and filtered further based on their log-fold changes over the time course (Benjamini and Hochberg, 1995).

E11.5 lungs were harvested and dissociated in 0.05% Trypsin-EDTA (Gibco) and CD31⁻CD45⁻EpCAM⁺ cells sorted on FACS ARIA (Becton Dickinson). RNA from pneumospheres and sorted E11.5 cells was extracted using RNeasy Minikit (Qiagen). Libraries were prepared using the Illumina TruSeq RNA kit from 1 µg total RNA, and 200-400 bp sized products were selected and cleaned up using AMPure XP magnetic beads. Final cDNA libraries were quantified using a Qubit dsDNA Assay Kit (Life Technologies) and sequenced on the Illumina NextSeq platform using 100 bp, single-end reads. RNAseq count data were examined using log counts per million (logCPM) values calculated by edgeR software (McCarthy et al., 2012; Robinson and Oshlack, 2010), which uses a default prior count of 2. An unsupervised clustering MDS plot, using the default of 500 most variable genes, was then created on the logCPM values using limma software (Ritchie et al., 2015). Limma's gene set enrichment barcode plots were used to compare our expression data with that of Nichane et al. (2017). The gene set from Nichane et al. (cluster IA in Fig. 3) contained 1085 genes that had matching gene symbols between the datasets. For each of day 3, 7 and 10 samples, the mean logCPM value was used to rank genes from lowest to highest expression, where logCPM values were calculated within limma using the voom function (Law et al., 2014).

RNA was extracted using the RNeasy Minikit (Qiagen). cDNA was synthesised using 2 µg of RNA, 500 ng Oligo(dT)₁₅ (Promega) and 200 U Superscript III Reverse Transcriptase (Life Technologies). PCR reaction mix contained 0.2 µM forward and reverse primers (Table S3), 0.1 µM UPL probe (Roche) and 1× LightCycler 480 Probes Master Mix (Roche). PCR was run on LightCycler 480 Real-Time PCR instrument with cycle conditions of 95°C for 10 min; 45 cycles of 95°C for 10 s, 60°C for 30 s, 40°C for 30 s. Cycle thresholds (Ct) were calculated using the standard curve method, using Hmbs mRNA expression as a control for variation in cDNA concentration between samples (Larionov et al., 2005).

We thank Tamara Beck, Andrew Keniry and Lachlan Whitehead for advice and technical expertise. We thank WEHI Bioservices, Histology, Flow cytometry and Genomics facilities and the Centre for Dynamic Imaging for their expert technical support.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.199543